3.state of Charge Estimation
Uploaded by
cary.avery6023.state of Charge Estimation
Uploaded by
cary.avery6023.1.1: Welcome to the course!
Welcome to the course!
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Welcome to Battery State of Charge (SOC) Estimation!
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
This course is the third in a specialization that investigates
the proper management and control of battery packs, usually comprising many cells
BMS must estimate nonmeasurable quantities that describe battery pack condition
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 4
3.1.1: Welcome to the course!
What topics will we study in this course?
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
In this course, you will learn:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Preparatory definitions, math concepts loop once each measurement interval while pack is active
How to derive the steps of the linear Kalman filter (KF)
How to think about and implement the linear Kalman filter
How to derive and implement the extended Kalman filter (EKF)
How to derive and implement the sigma-point Kalman filter
(a.k.a., unscented Kalman filter, SPKF/UKF)
How to solve real-world issues via information provided by
Kalman filter (honors)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 4
3.1.1: Welcome to the course!
What skills will you gain in this course?
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
After completing the course, you’ll be able to:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Implement simple voltage- and current- loop once each measurement interval while pack is active
based SOC estimators and understand their limitations
Explain purpose of each sequential-probabilistic-inference step
Execute provided Octave script for linear KF, evaluate results
Execute provided Octave script for SOC estimation using EKF
on lab-test data, evaluate results
Execute provided Octave script for SOC estimation using SPKF
on lab-test data, evaluate results
Implement method to detect and discard faulty voltage-sensor
measurements (honors)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 4
3.1.1: Welcome to the course!
Prerequisites; for further study
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Prerequisites:
loop once each measurement interval while pack is active
Course 1: Intro. to Battery Management Systems
Basic background in BMS requirements, sensing
Course 2: Equivalent Circuit Cell Model Simulation
Specific background in ESC cell model
Probability theory (esp. Gaussian random variables)
Resource:
We’ll study Ch. 3 of Battery Management Systems,
Vol. 2, Equivalent-Circuit Methods, Artech House
For further study, you can confer this optional resource
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 4 of 4
3.1.2: What is the importance of a good SOC estimator?
States versus parameters
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
A BMS must be able to estimate two fundamentally
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
different types of nonmeasurable battery-pack quantity
States change quickly (e.g., state-of-charge, diffusion voltage, hysteresis voltage)
Parameters change slowly (e.g., cell capacities, resistances, aging effects)
SOC estimate for all cells is important input to balancing, energy, power calculations
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 6
3.1.2: What is the importance of a good SOC estimator?
Simple versus better
key off: store data
key on: initialize
While we might be interested in estimating entire meas. voltage
current
temperature
estimate
state of
charge (SOC)
estimate
state of
health (SOH)
balance
cells
compute
power
limits
battery-model state, we first focus on estimating SOC only loop once each measurement interval while pack is active
You will learn some simple methods that lack robustness
Then, you will learn methods that estimate entire
battery-model state, enabling more advanced applications
Recall, SOC is something like a dashboard fuel gauge that
reports a value from “Empty” (0 %) to “Full” (100 %)
But, while there exist sensors to measure a gasoline level in
a tank accurately, there is (presently) no sensor available to
measure SOC
And, accurate SOC estimates produce additional benefits. . .
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 6
3.1.2: What is the importance of a good SOC estimator?
Additional benefits (1)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Longevity: If a gas tank is over-filled or emptied, tank is fine
current state of state of power
temperature charge (SOC) health (SOH) cells limits
However, over-charging or over-discharging a battery cell
loop once each measurement interval while pack is active
may cause permanent damage and result in reduced lifetime
An accurate SOC estimate may be used to avoid harming cells by not permitting
current to be passed that would cause damage
Performance: Without good SOC estimator, one must be overly conservative when
using battery pack to avoid over/undercharge due to trusting poor estimate
With a good estimate, especially one with known error bounds, one can
aggressively use the entire pack capacity
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 6
3.1.2: What is the importance of a good SOC estimator?
Additional benefits (2)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Reliability: Poor estimators behave differently for different
current state of state of power
temperature charge (SOC) health (SOH) cells limits
usage profiles (e.g., different vehicle drive cycles)
loop once each measurement interval while pack is active
A good SOC estimator is consistent and dependable for any driving profile,
enhancing overall power-system reliability
Density: Accurate battery state estimates allow battery pack to be used aggressively
within design limits, so pack doesn’t need to be over-engineered
This allows smaller, lighter battery packs
Economy: Smaller battery systems cost less. Warranty service on reliable systems
costs less
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 4 of 6
3.1.2: What is the importance of a good SOC estimator?
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
BMS must estimate battery-cell states and parameters
current state of state of power
temperature charge (SOC) health (SOH) cells limits
This course focuses on estimating battery-cell states
loop once each measurement interval while pack is active
There are simple state estimators and better state estimators
You will learn how to implement some of each category
“Better” state estimators can improve battery-system longevity, performance,
reliability, density, and economy
It is often worth investing some effort to implement a “better” estimator
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 5 of 6
3.1.2: What is the importance of a good SOC estimator?
Credits
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Credits for photos in this lesson
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Gauge on slide 2: Pixabay license (https://pixabay.com/en/service/license/),
https://pixabay.com/en/fuel-gauge-gasoline-measure-meter-40193/
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 6 of 6
3.1.3: How do we define SOC carefully?
A careful definition of state of charge
key off: store data
key on: initialize
Course 1 introduced an electrochemical definition of SOC
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Defined lithium concentration stoichiometry: D cs;avg =cs;max ❝s✱♠ ①
loop once each measurement interval while pack is active
Stoichiometry is intended to remain between 0% and 100% ✦✶✁✂✄
Then, cell SOC is computed as: ´k D .k 0% /=.100% 0% /
Issue: there is (presently) no direct way to measure
concentrations that allow calculating SOC
So, we must infer or estimate the SOC using measurements of
only cell terminal voltage and cell current
We’ve already noticed that while cell OCV is closely related to ☎✵✆
SOC, the terminal voltage is a poor predictor of OCV unless the
cell is in electrochemical equilibrium (and hysteresis is negligible)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 9
3.1.3: How do we define SOC carefully?
“Fully charged”
key off: store data
key on: initialize
So, how can we know true cell SOC to evaluate estimators?
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
How can we know true SOC for any other purpose? loop once each measurement interval while pack is active
KEY POINT : Some definitions can calibrate our lab tests
DEFINITION : A cell is fully charged when its OCV equals vh .T /,
a manufacturer-specified voltage (may be function of T )
e.g., vh .25 C/ D 4:2 V for LMO; vh .25 C/ D 3:6 V for LFP
To bring cell to fully charged state: execute CC charge until
terminal voltage reaches vh .T /, followed by CV until charging
current becomes infinitesimal
We define the SOC of a fully charged cell to be 100 %
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 9
3.1.3: How do we define SOC carefully?
“Fully discharged”
key off: store data
key on: initialize
Cell is fully discharged when OCV equals vl .T /,
meas. voltage estimate estimate compute
DEFINITION :
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
a manufacturer specified voltage (may be function of T ) loop once each measurement interval while pack is active
e.g., vl .25 C/ D 3:0 V for LMO; vl .25 C/ D 2:0 V for LFP
To bring cell to fully discharged state: execute CC discharge
until terminal voltage equals vl .T /, followed by CV until
discharge current becomes infinitesimal
We define the SOC of a fully discharged cell to be 0 %
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 9
3.1.3: How do we define SOC carefully?
“Total capacity”
key off: store data
key on: initialize
Cell total capacity Q is quantity of charge
meas. voltage estimate estimate compute
DEFINITION :
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
removed as cell is brought from fully charged state loop once each measurement interval while pack is active
to fully discharged state
SI unit for charge is coulombs (C), but more common to
use units of ampere hours (Ah) or milliampere hours
(mAh) to measure the total capacity of a battery cell
The total capacity of a cell is not a fixed quantity: it
generally decays slowly over time as the cell degrades
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 4 of 9
3.1.3: How do we define SOC carefully?
“Discharge capacity”
key off: store data
key on: initialize
meas. voltage estimate estimate compute
Discharge capacity Qrate is quantity of charge
balance
DEFINITION :
current state of state of power
temperature charge (SOC) health (SOH) cells limits
removed as cell discharged at constant rate from fully
loop once each measurement interval while pack is active
charged state until terminal voltage v.t/ reaches vl .T /
Since based on loaded terminal voltage rather Discharge curves at different C rates
1C rate
than OCV, is strongly dependent on cell’s internal
4
2C rate
Terminal voltage (V)
4C rate
resistance, and so also rate and temperature
3.8 8C rate
Unless i.t/ ! 0, discharge capacity is less than
3.6
3.4
total capacity
3.2
Likewise, cell SOC is nonzero when terminal
3
voltage reaches vl .T / when i.t/ > 0 0 1 2 3 4
Discharge capacity (Ah)
5 6
The discharge capacity of a cell at a particular rate and temperature is not a fixed
quantity: it also generally decays slowly over time as the cell degrades
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 5 of 9
3.1.3: How do we define SOC carefully?
“Nominal capacity”
key off: store data
key on: initialize
Cell nominal capacity Qnom is manufacturer-
meas. voltage estimate estimate compute
DEFINITION :
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
specified quantity intended to be representative of 1C-rate loop once each measurement interval while pack is active
discharge capacity Q1C of a particular manufactured lot of cells
at room temperature, 25 C
The nominal capacity is a constant value
Since nominal capacity is representative of a lot of cells and
discharge capacity is representative of a single individual cell,
Qnom ¤ Q1C in general, even at beginning of life
Also, since Qnom is representative of a discharge capacity and
not a total capacity, Qnom ¤ Q
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 6 of 9
3.1.3: How do we define SOC carefully?
“Residual capacity” and “state of charge”
key off: store data
key on: initialize
meas. voltage estimate estimate compute
DEFINITION : Cell residual capacity is quantity of charge that
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
would be removed from cell if it were brought from its loop once each measurement interval while pack is active
present state to a fully discharged state
DEFINITION : Cell state-of-charge is ratio of residual capacity
to total capacity
These definitions are consistent with the relationships
Z
1 t
´.t/ D ´.0/ .t/i.t/ dt , and ´k C1 D ´k k ik t=Q
Q 0
that you have already seen
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 7 of 9
3.1.3: How do we define SOC carefully?
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Careful definitions of important quantities important for
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
remainder of the specialization
Have defined what is meant by “fully charged”, “fully discharged”, “total capacity”,
“discharge capacity”, “nominal capacity”, “residual capacity”, and “state of charge”
In particular, definition of state of charge is consistent with the equations you
learned in course 2 of the specialization
Can now build on these definitions to build estimators for any of the above
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 8 of 9
3.1.3: How do we define SOC carefully?
Credits
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Credits for photos in this lesson
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Gauge on slides 2–3: Pixabay license (https://pixabay.com/en/service/license/),
modified from
https://pixabay.com/en/fuel-gauge-gasoline-measure-meter-40193/
Coffee on slide 4: Pixabay license (https://pixabay.com/en/service/license/),
modified from https://pixabay.com/en/coffee-espresso-cup-drink-3129995/
VHBW cell on slide 6: Raimond Spekking (own work) [CC BY-SA 4.0
(http://creativecommons.org/licenses/by-sa/4.0)], via Wikimedia Commons,
https://commons.wikimedia.org/wiki/File:
VHBW_for_HB5V1HV_Replacement_Li-Ion_Battery-7119.jpg
Glass on slide 7: Pixabay license (https://pixabay.com/en/service/license/),
modified from https://pixabay.com/en/glass-cognac-smoke-full-empty-300558/
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 9 of 9
3.1.4: What are some approaches to estimating battery cell SOC?
Poor, voltage-based method to estimate SOC
key off: store data
key on: initialize
meas. voltage estimate estimate compute
We begin to consider the question, “how should we
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
estimate battery cell state of charge?”
loop once each measurement interval while pack is active
One method would be to measure cell terminal 100
True SOC and voltage-based estimate
voltage under load, v.t/, and look up on “SOC
SOC estimate
True SOC
SOC and estimate (%)
80
versus OCV” curve 60
SOCest = SOCfromOCVtemp(v,T,model);
Ignores effects of i.t/ R0 losses, diffusion
40
20
voltages, and hysteresis on v.t/
0
Wide flat areas of OCV curve dilute accuracy of 0 100 200
Time (min)
300 400 500
estimate — very noisy
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 7
3.1.4: What are some approaches to estimating battery cell SOC?
Poor, voltage-based method to estimate SOC
key off: store data
key on: initialize
meas. voltage estimate estimate compute
A modification of this method assumes a cell model
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
v.t/ D OCV.´.t// i.t/R0 and then looks up
loop once each measurement interval while pack is active
v.t/ C i.t/R0 on “SOC versus OCV” curve 100
True SOC and voltage-based estimate
SOCest = SOCfromOCVtemp(v+i*R0,T,model); 80 SOC and estimate (%)
SOC estimate
True SOC
Better, but still ignores effects diffusion voltages, 60
hysteresis and so is still noisy
40
Filtering helps but adds delay, which must be 20
accounted for
0
Hysteresis is another complicating factor 0 100 200
Time (min)
300 400 500
Even though its estimates are noisy, we’ll find an application for this modified
method in the next course
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 7
3.1.4: What are some approaches to estimating battery cell SOC?
Poor, current-based method to estimate SOC
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Coulomb counting keeps track of charge in, out of cells via
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Z
loop once each measurement interval while pack is active
1 t
Ó .t/ D Ó .0/ b . /imeas. / d
Q 0
imeas .t/ D itrue .t/ C inoise .t/ C ibias .t/ C inonlin .t/ C isd .t/ C ileakage .t/
Okay for short periods of operation when initial conditions are known or can be
frequently “reset”
Subject to drift due to current sensor’s fluctuations, current-sensor bias, incorrect
capacity estimate, other losses
Uncertainty/error bounds grow (without limit) over time until estimate is “reset”
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 7
3.1.4: What are some approaches to estimating battery cell SOC?
Model-based state estimation
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
An alternative to a voltage-only method or a current-only
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
method is somehow to combine the approaches
Model-based estimators implement Process Noise Sensor Noise
True System
algorithms that use sensed Input
State
Measured Output
measurements to infer internal hidden System Model
state of dynamic system State Est.
Predicted Output
Mathematical model of system is assumed known (e.g., ESC cell model)
Same input propagated through true system, model, measured and predicted
outputs compared; error used to update model’s state estimate
Output error due to: state, measurement, model errors;
Update must be done carefully to account for all of these.
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 4 of 7
3.1.4: What are some approaches to estimating battery cell SOC?
Linear Kalman filter
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Under specific conditions, Kalman filter (KF, special case of
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
sequential probabilistic inference) gives optimal state estimate
We study linear KF and some of its variants throughout remainder of course
We start by assuming a general, possibly nonlinear, state-space model
xk D f .xk 1; uk 1; wk 1/
yk D h.xk ; uk ; vk /,
where uk is known (measured) input signal, wk is a process-noise random input,
and vk is a sensor-noise random input (ESC cell model fits this framework!)
Functions f ./ and h./ may be time-varying, but we generally omit time
dependency from notation to avoid clutter
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 5 of 7
3.1.4: What are some approaches to estimating battery cell SOC?
Sequential probabilistic inference
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF is special case of sequential probabilistic inference (SPI):
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Estimate present state xk of dynamic system using all
loop once each measurement interval while pack is active
measurements Yk D fy0 ; y1 ; ; yk g
②✝✞✟ ✠✡☛☞ ✌✍
❖❜s❡r✈✦❞ ❢❨❥❳ ✳✎✏ ✑ ✒ ✓ ✴
❯♥♦✧★✩✪✫✬✭
①❦ ✷ ✁ ✂✄✶ ☎✆
✔ ✕✖✗ ✘✙ ✚ ✛✜ ✢✣✤ ✥
The observations allow us to “peek” at what is happening in true system
Based on observations and model, we estimate state
Process- and sensor-noise randomness always cause imperfect estimates
So, to understand SPI solution, must study vector random processes
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 6 of 7
3.1.4: What are some approaches to estimating battery cell SOC?
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Direct lookup of terminal voltage in OCV versus SOC table
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
gives very poor estimate of cell SOC
Modifying lookup to account for ohmic resistance helps, but not enough
Coulomb counting also has many problems we would like to avoid
Model-based state estimators combine voltage and current measurements, using a
cell model to do so, to yield better state estimates
Sequential probabilistic inference is the general framework that describes
model-based state estimators of interest
We will study SPI this week, as preparation for developing KF solution
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 7 of 7
3.1.5: Understanding uncertainty via mean and covariance
Review of probability
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Sequential probabilistic inference (and hence Kalman filters)
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
seek to find best state estimate in the presence of process
and sensor noises on measurements
By definition, noise is not deterministic—it is random in some sense
So, to discuss the impact of noise on the system dynamics, we must understand
(vector) “random variables” (RVs)
Can’t predict exactly what we will get each time we measure the RV, but
Can characterize likelihood of different possible measurements by RV’s
“probability density function” (pdf)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 6
3.1.5: Understanding uncertainty via mean and covariance
Review of random vectors
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
As review, define random vector X , sample vector x0
current state of state of power
temperature charge (SOC) health (SOH) cells limits
2 3 2 3
loop once each measurement interval while pack is active
X1 x1
6
X2 7 6
x2 7
where X1 through Xn are themselves
XD 6
6 ::
7
7, x0 D 6
6 ::
7
7,
scalar RVs, and x1 through xn are scalar
4 : 5 4 : 5
Xn xn constant values these RVs can take on
Random vector X described by (scalar function) joint pdf fX .x/ of vector X
fX .x0 / means fX .X1 D x1 ; X2 D x2 ; ; Xn D xn /
fX .x0 / dx1 dx2 dxn is probability that X is between x0 and x0 C dx
fX .x0 / is scaled probability or “likelihood” of measuring sample vector x0
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 6
3.1.5: Understanding uncertainty via mean and covariance
Key properties of joint pdf of random vector
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
fZX .x/ 0 Z8 x
current state of state of power
cells
1.
temperature charge (SOC) health (SOH) limits
Z 1 1 1
loop once each measurement interval while pack is active
2. fX .x/ dx1 dx2 dxn D 1
1 1 1 Z
Z 1 1 Z 1
3. xN D EX D xfX .x/ dx1 dx2 dxn
1 1 1
4. Correlation matrix (note outer product, not inner product):
Z 1Z 1 Z 1
X D EXX T D ::: xx T fX .x/ dx1 dx2 dxn
1 1 1
D X xN . Then,
5. Covariance matrix: Define X
e
1Z 1 Z 1 Z
X D E.X x/.X
e N N D
x/ .x x/.x
N N T fX .x/ dx1 dx2 dxn
x/
T
1 1 1
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 6
3.1.5: Understanding uncertainty via mean and covariance
Properties of correlation and covariance
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
e
current state of state of power
X is symmetric and positive-semi-definite (psd). This means
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Xy 0 8 y
y eT
Notice that correlation D covariance for zero-mean random vectors
The covariance entries have specific meaning:
X /i i D X
.e 2
i
X /ij D ij X X D .e
.e i X /j i
j
The diagonal entries are the variances of each vector component
Correlation coefficient ij is a measure of linear dependence between Xi and
Xj ; jij j 1
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 4 of 6
3.1.5: Understanding uncertainty via mean and covariance
The multivariable Gaussian pdf
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
There are infinite variety in pdfs; however, we assume only
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
(multivariable) Gaussian pdf in KF
All noises and the state vector itself are assumed to be Gaussian random vectors
Gaussian or normal pdf is (we say “X N .x; N eX /”)
1 1
fX .x/ D exp N T eX1.x x/
.x x/ N
.2/n=2 e 1=2 2
X
jeX j D det.eX /; eX1 requires positive-definite e
X
Contours of constant fX .x/ are hyper-ellipsoids, centered at xN , rotated via e
X
Good news. . . We won’t need to work directly with this equation very much!
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 5 of 6
3.1.5: Understanding uncertainty via mean and covariance
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
To develop sequential-probabilistic-inference solution, must
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
have background understanding of random variables (RVs)
RVs are described by probability density functions (pdfs)
These pdfs have important properties, which we have reviewed
In particular, we will be making use of mean and covariance a lot
The pdf we will assume for all RVs is multivariable Gaussian (or normal) distribution
While this seems complicated at first, it turns out to simplify the math a lot later on
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 6 of 6
3.1.6: Understanding joint uncertainty of two unknown quantities
Independence and uncorrelation
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
INDEPENDENCE : Iff jointly-distributed RVs are independent, then
current state of state of power
temperature charge (SOC) health (SOH) cells limits
fX .x1 ; x2 ; : : : ; xn / D fX1 .x1 /fX2 .x2 / fXn .xn /
loop once each measurement interval while pack is active
Means that joint pdf factors into product of marginal pdfs (not generally true)
“The particular value of the random variable X1 has no impact on what value we
would obtain for the random variable X2 ”. . . no nonlinear or linear relationship
UNCORRELATED : Two jointly-distributed RVs X1 and X2 are uncorrelated iff
cov.X1 ; X2 / D E.X1 xN 1 /.X2 xN 2 /T D 0
Can show that this implies that EX1 X2T D EX1 EX2T : Means that expectation
of a product factors into product of expectations (not generally true)
Implies 12 D 0; uncorrelated means there is no linear relationship between RVs
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 7
3.1.6: Understanding joint uncertainty of two unknown quantities
Independence versus uncorrelation
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
In general, condition for RVs to be uncorrelated is much
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
weaker than for them to be independent
If jointly-distributed RVs X1 and X2 are independent they must also be
uncorrelated: independence implies uncorrelation
However, uncorrelated RVs are not necessarily independent
If joint normally (Gaussian) distributed RVs are uncorrelated, are also independent
This is a (very, very) special case
This is one of the reasons we assume that all pdfs in the sequential-probabilistic-
inference solution are Gaussian
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 7
3.1.6: Understanding joint uncertainty of two unknown quantities
Conditional probability
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We can define a conditional pdf
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
fX jY .x jy/ D fX;Y .x; y/=fY .y/
as the likelihood that X D x given that Y D y has happened
Note: Marginal pdf fY .y/ may be calculated as
Z1
fY .y/ D fX;Y .x; y/ dx
1
For each y , integrate out effect of X
Conditional pdf = joint pdf / marginal pdf (of RV being conditioned on)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 7
3.1.6: Understanding joint uncertainty of two unknown quantities
Bayes’ rule
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
As a direct extension of the definition of a conditional pdf,
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
fX;Y .x; y/ D fX jY .x jy/fY .y/
D fY jX .y jx/fX .x/
Therefore, we can solve for one conditional pdf in terms of other
fY jX .y jx/fX .x/
fX jY .x jy/ D
fY .y/
This is known as Bayes’ rule
Relates posterior probability to prior probability, forms a key step in KF derivation
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 4 of 7
3.1.6: Understanding joint uncertainty of two unknown quantities
Conditional expectation
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Can define conditional expectation as what we expect the
current state of state of power
temperature charge (SOC) health (SOH) cells limits
value of X to be given that Y D y has happened
Z
loop once each measurement interval while pack is active
1
EX D x jY D y D EX jY D xfX jY .x jY / dx
1
Conditional expectation is critical
KF is an algorithm to compute Exk j Yk . . . expected value of model state vector
given complete set of measurements made
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 5 of 7
3.1.6: Understanding joint uncertainty of two unknown quantities
Central limit theorem
P
key off: store data
key on: initialize
D X , and X are independent, identically distributed
meas. voltage estimate estimate balance compute
If Y
current state of state of power
temperature charge (SOC) health (SOH) cells limits
i i i loop once each measurement interval while pack is active
(IID), and have finite mean and variance: then Y is approx.
normally distributed: approximation improves as more RVs are summed
Special case: linear combination of Gaussian RVs results in a Gaussian RV
Since state of dynamic system adds up effects of lots of independent random
inputs, reasonable to assume that distribution of state tends to normal distribution
Leads to key assumptions for derivation of KF, as we will see:
Assume state xk , process noise wk , sensor noise vk are normally distributed RVs
Assume wk and vk are uncorrelated with each other
Even when these assumptions are broken in practice, KF works quite well
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 6 of 7
3.1.6: Understanding joint uncertainty of two unknown quantities
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
If two RVs are independent, joint pdf equals product of
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
marginal pdfs and knowing value of one RV cannot be used to
help predict value of other in any way
If two RVs are uncorrelated, expected value of product equals product of expected
values and knowing value of one RV cannot be used with linear equation to help
predict value of other
In most cases, RVs we will work with are correlated, so conditional expectation will
help us to predict value of one RV given values of other(s)
Central limit theorem justifies assumption that RVs are Gaussian
Uncorrelated Gaussian RVs are also independent: a very special case
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 7 of 7
3.1.7: Understanding time-varying uncertain quantities
Overview of vector random (stochastic) processes
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
A stochastic or random process is a family of random vectors
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
indexed by a parameter set (“time” in our case)
For example, we might refer to a random process Xk for generic k
Value of random process at any specific time k D m is a random variable Xm
Usually assume stationarity
The statistics (i.e., pdf) of the RV are time-shift invariant
Therefore, EXk D xN for all k and EXk1 XkT2 D RX .k1 k2 /
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 7
3.1.7: Understanding time-varying uncertain quantities
Properties and important points
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
D
current state of state of power
cells
EXk1 XkT2 .
temperature charge (SOC) health (SOH) limits
1. Autocorrelation: RX .k1 ; k2 / If stationary, loop once each measurement interval while pack is active
RX . / D EXk XkTC
Provides a measure of correlation between elements of the process having
time displacement
RX .0/ D X2 for zero-mean X
RX .0/ is always the maximum value of RX . /
2. Autocovariance: CX .k1 ; k2 / D E.Xk1 EXk1 /.Xk2 EXk2 /T . If stationary,
CX . / D E.Xk x/.X
N kC x/ N T
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 7
3.1.7: Understanding time-varying uncertain quantities
White noise
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
White noise: Some processes have a unique autocorrelation:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
1. Zero mean
2. RX . / D EXk XkTC D SX . / where . / is the Dirac delta. . / D 0 8 ¤ 0
Therefore, the process is uncorrelated in time
Clearly an abstraction, but proves to be a very useful one
White noise Correlated noise
4 0.2
0.15
2
0.1
0.05
Value
Value
0
0
−0.05
−2
−0.1
−4 −0.15
0 200 400 600 800 1000 0 200 400 600 800 1000
Time Time
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 7
3.1.7: Understanding time-varying uncertain quantities
Shaping filters: Idea
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Will assume noise inputs to dynamic systems are white
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Limiting assumption, but one that can be easily fixed
loop once each measurement interval while pack is active
à Can use second linear system to “shape” the noise as desired.
Pr✻✈✼✽✉✾ ✿❁❝❂❃❄❅ ◆❆❇ ❈❉❊❋■❏❑
❲❤✐t❡ ❙✂❛♣✄❞
✢ ✣✤ ✥ ✙✚ ● ✳③ ✴ ✛✜
♥♦ s✁ ✇❦ ☎✆✝✞✟ ✠✶❀✡
✭✮✯✰
☛☞✌✍✎
❍ ✪✫ ✬ ✦ ✧★ ✩ ②✘
✏✑✒✓✔ ✕✷✖✗
✱✲✵✸✹✺
Can drive our linear system with noise that has desired characteristics by
introducing shaping filter H.´/ that itself is driven by white noise
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 4 of 7
3.1.7: Understanding time-varying uncertain quantities
Shaping filters: Model
key off: store data
key on: initialize
meas. voltage estimate estimate compute
Combined system GH.´/ looks exactly the same as before,
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
but G.´/ is not driven by pure white noise any more
Analysis augments original system model Shaping filter with white input and
with filter states. Original system has desired output statistics has
xk C1 D Axk C Bw w1;k xs;k C1 D As xs;k C Bs w2;k
yk D C xk w1;k D Cs xs;k
Combine into larger-order augmented system driven by white noise:
x k C1 A B w Cs xk
xs;k C1
D 0 As xs;k
C 0
Bs
w2;k
xk
yk D C 0 xs;k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 5 of 7
3.1.7: Understanding time-varying uncertain quantities
Gaussian processes
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We will work with Gaussian noises to a large extent
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Uniquely defined by the first- and second central moments
loop once each measurement interval while pack is active
of the statistics à Gaussian assumption not essential
Our filters will always track only the first two moments.
NOTATION : Until now, we have always used capital letters for random variables
State of system driven by random process is an RV, so we could call it Xk
More common to retain standard notation xk and understand from context that we
are discussing an RV
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 6 of 7
3.1.7: Understanding time-varying uncertain quantities
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Random process is a family of RVs indexed by time
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Will assume our random processes are stationary
Autocorrelation and autocovariance measure self-predictability of a signal at
different time offsets
White noise is zero mean signal, completely uncorrelated with self (“completely
random”)
An abstraction, but a very useful one
If noises in a system of interest are not white, can filter white noise to create same
general characteristics
From now on, unless stated otherwise, all noise signals will be white and Gaussian
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 7 of 7
3.1.8: Where from here?
Summary of this week
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
This week, you learned:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Advantages of a good state estimator compared with a
loop once each measurement interval while pack is active
(mathematically or computationally) simple state estimator
A careful definition of SOC that is compatible with a physical understanding
Limitations of simple voltage-based and current-based (coulomb counting)
methods for SOC estimation
The promise of better estimates from model-based estimation
Accordingly, we reviewed probability concepts to enable future studies
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 1 of 3
3.1.8: Where from here?
Where from here?
key off: store data
key on: initialize
meas. voltage estimate estimate compute
Next week, you will learn about “Gaussian sequential
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
probabilistic inference” state estimation
loop once each measurement interval while pack is active
4/23/18, 3)08 PM
Under certain assumptions, this gives optimal (minimum
mean-squared error) estimate of system state even when
measurements contain noise
Will see that this solution has two primary steps, each having
three sub-steps (six steps total)
Will apply these steps to derive linear Kalman filter
In following weeks, will apply these steps to derive nonlinear
extended Kalman filter and sigma-point Kalman filter as well
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 2 of 3
file:///Users/glp/OneDrive%20-%20University%20of%20Colorado%20C…ING/MOOC-BMS/COURSE3/FIGURES/FIGURES-3-1-8/ancient-1300292.svg Page 1 of 1
3.1.8: Where from here?
Credits
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Credits for photos in this lesson
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
“Map” on slide 2: Pixabay license
(https://pixabay.com/en/service/license/), from
https://pixabay.com/en/ancient-historic-map-old-parchment-1300292/
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | The importance of a good SOC estimator j 3 of 3
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Vector notation
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
This week, we derive the Kalman filter algorithm, which is a
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
special case of sequential probabilistic inference
We begin by defining some math notation that we will use from now on
Superscript “ ” indicates a predicted quantity based only on past measurements
Superscript “C” indicates an estimated quantity based on both past and present
measurements
Symbol “ O ” indicates a predicted or estimated quantity: xO C or xO
Symbol “ Q ” indicates an error: the difference between a true and predicted or
estimated quantity: xQ D x xO
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 1 of 8
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Matrix notation
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Symbol “” is used to denote correlation between the two
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
arguments in its subscript (autocorrelation if only one is given)
xy D Exy T and x D Exx T
Furthermore, if the arguments are zero mean (as they often are in the quantities we
talk about), then this represents covariance
xQ yQ D ExQ yQ T D E.xQ Q yQ
Ex/. Q T
Ey/
for zero-mean xQ and yQ
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 2 of 8
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Cost function to minimize (optimize)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Desire state estimate that minimizes “mean-squared error”
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
O kMMSE.Yk / D arg min E xk xO kC 2 j Yk
2
x
xO k
D arg min E .xk xO kC /T .xk xO kC / j Yk
xO k
D arg min E xkT xk 2xkT xO kC C .xO kC /T xO kC j Yk
xO k
C
Solve for xO k by differentiating cost function and setting result to zero
d T T C
0 D E xk xk 2xk xk O C .xO kC/T xO kC j Yk
dxO C k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 3 of 8
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Preliminary solution to state estimate
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
To do so, note the following identities from vector calculus,
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
d T d T d T
dX
Y X D Y; dX
X Y D Y; and
dX
X AX D .A C AT /X
Then,
d T T C
0 D CE xk xk 2xk xk O C .xO kC/T xO kC j Yk
dxO k
D E 2.xk xO kC/ j Yk D 2xO kC
0 2E xk j Yk
O kC D E xk j Yk
x
Desire to find algorithm for computing for Exk j Yk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 4 of 8
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Prediction error and innovation
key off: store data
key on: initialize
Qk D xk xO k
meas. voltage estimate estimate balance compute
Define prediction
current state of state of power
error where
cells
x
temperature charge (SOC) health (SOH) limits
O D E xk j Yk
loop once each measurement interval while pack is active
xk 1
Error is always “truth minus prediction” or “truth minus estimate”
We can’t compute error in practice, since truth value is not known
But, we can prove statistical results using this definition that give an algorithm for
estimating the truth using measurable values
Also, define the measurement innovation (what is new or unexpected
in the
measurement) as yQk D yk O
yk where yOk D E yk j Yk 1
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 5 of 8
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Prediction error and innovation are zero mean
key off: store data
key on: initialize
Both xQ k and yQk can be shown to be zero mean using
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
method of iterated expectation: EY EX jY X j Y D EX X
loop once each measurement interval while pack is active
Q D E xk E E xk j Yk 1 D E xk E xk D 0
E xk
E yQk D E yk E E yk j Yk 1 D E yk E yk D 0
Note also that xQ k is uncorrelated with past measurements as they have already
been incorporated into xO k
E xQ k j Yk 1 D E xk E xk j Yk 1 j Yk 1 D 0 D E xQ k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 6 of 8
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Predict/correct solution
key off: store data
key on: initialize
Q j Yk
meas. voltage estimate estimate balance compute
Consider now expanding the relationship E
current state of state of power
cells
xk
temperature charge (SOC) health (SOH) limits
loop once each measurement interval while pack is active
E Q j Yk D E xkj Yk
xk E xO k j Yk
xO kC xO k
This is true because xO k D Exk j Yk 1 is a constant vector, and further
conditioning on Yk has no additional effect
We can also expand this relationship a different way
Q j Yk D E xQ k j Yk 1; yk D E xQ k j yk
E xk
Combining both expansions, we have xO kC D xO k C E xQ k j yk , which is a
predict/correct sequence of steps, as promised
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 7 of 8
3.2.1: Predict/correct mechanism of sequential probabilistic inference
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
You learned meaning of notation we use, including
current state of state of power
temperature charge (SOC) health (SOH) cells limits
superscripts “ ” and “C”, symbols “ O ” and “ Q ”, and symbol “”
loop once each measurement interval while pack is active
We are solving a minimum mean-squared-error
problem to find a state estimate
Initial solution was xO kC D Exk j Yk
Later refined this to be xO kC D xO k C E xQ k j yk
A predict/correct structure
But, what is ExQ k j yk ? That’s what we look at next
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 8 of 8
3.2.2: The Kalman-filter gain factor
Pathway to finding Kalman gain
key off: store data
key on: initialize
O C D xO C ExQ j y
meas. voltage estimate estimate balance compute
So far, we have derived
current state of state of power
state estimate
cells
x
temperature charge (SOC) health (SOH) limits
k k k k
But, what is E xQ k j yk ?
loop once each measurement interval while pack is active
To evaluate this expression, we consider very generically the problem of finding
f .x j y/ if x and y are jointly Gaussian vectors
We combine x and y into an augmented vector X where
D yNN
xQ xQ yQ
X D x
y
; E X
x
; e
X D yQ xQ yQ
We will then apply this generic result for f .x j y/ to our specific problem of
determining f .xQ k j yk / by setting x D xQ k and y D yk
We can then use this conditional pdf to find E xQ k j yk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 1 of 9
3.2.2: The Kalman-filter gain factor
Expanding the conditional pdf
key off: store data
key on: initialize
To proceed, we first write f .x j y/ D f .x; y/=f .y/
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Note that this is proportional to
T !
exp 1 x xN 1
x N
x
f .x; y/
2 y yN e
X y N
y
/ ,
f .y/ exp 1
2 .y N T
y/ yQ 1 .y y/N
where the constant of proportionality is not important to subsequent operations
Recall that exp.A/= exp.B / D exp.A B /. . . which we take advantage of next
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 2 of 9
3.2.2: The Kalman-filter gain factor
Condensing the exponential term
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
The combined exponential argument becomes
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
T
1 x xN 1 x x N C 21 .y y/
N T yQ 1 .y y/
N
2 y yN
e
X y y N
To condense notation somewhat, define xQ D x xN and yQ D y yN
Then, the terms in the exponent become,
T
1 xQ 1 xQ C 1 yQ T 1yQ
yQ yQ yQ
2 e
X 2
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 3 of 9
3.2.2: The Kalman-filter gain factor
Inverting covariance (step 1)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
To proceed, we must invert e
current state of state of power
X . To do so, we substitute the
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
following transformation (which you can verify)
xQ xQ yQ xQ yQ yQ 1 xQ xQ yQ yQ 1 yQ xQ
yQ xQ yQ
D I
0 I 0
0
yQ
I
yQ 1 yQ xQ
0
I
Then,
1 1 1 1
xQ xQ yQ xQ xQ yQ yQ 1 yQ xQ xQ yQ yQ 1
yQ xQ yQ
D I
1
yQ yQ xQ
0
I 0
0
yQ
I
0 I
.xQ xQ yQ yQ 1 yQ xQ / 1 0 xQ yQ yQ 1
D I
yQ 1 yQ xQ
0
I 0 yQ 1
I
0 I
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 4 of 9
3.2.2: The Kalman-filter gain factor
Inverting covariance (step 2)
key off: store data
key on: initialize
D xQ
meas. voltage estimate estimate balance compute
Defining Schur complement of : M xQ yQ yQ 1 yQ xQ ,
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
1
exponent D 2
Q T 1
x M x Q xQ T M 1xQ yQ yQ 1yQ yQ T yQ 1yQxQ M 1xQ
C yQ T yQ 1yQxQ M 1xQ yQ yQ 1yQ C yQ T yQ 1yQ yQ T yQ 1yQ
Last two terms cancel; remaining terms can be grouped as
1
exponent D 2
Q
x
1 T
xQ yQ yQ y M
1
x Q Q 1
xQ yQ yQ y Q
In terms of only the original variables, exponent is
T 1
1
2
x N C xQ yQ yQ
x
1
.y N
y/ xQ xQ yQ yQ yQ xQ
1
x x N C xQ yQ yQ 1
.y N
y/
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 5 of 9
3.2.2: The Kalman-filter gain factor
A solution to the generic conditional Gaussian
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Comparing this exponent to Gaussian standard form
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
T 1
1
2
x N C xQ yQ yQ 1.y y/
x N xQ
1
xQ yQ yQ yQ xQ x x N C xQ yQ yQ 1.y y/
N
Infer that mean of f .x j y/ must be xN C xQ yQ yQ 1 .y y/
N
Infer that covariance of f .x j y/ must be xQ xQ yQ yQ 1 yQ xQ
So, we conclude that
f .xj y/ N .xN C xQ yQ yQ 1.y y/;
N xQ 1
xQ yQ yQ yQ xQ /
E x j y D E x C xQ yQ yQ y E y
1
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 6 of 9
3.2.2: The Kalman-filter gain factor
State update equation
key off: store data
key on: initialize
D yQk C yOk , we get
meas. voltage estimate estimate balance compute
Applying this to our problem, when yk
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
E Q j yk D Q C
xk E xk xQ y;k
Q Q
1
y;k yk E yk
D Q C E xk xQ y;k
Q y;k
1
Q yk Q C yOk E yQk C yOk
D C 0 xQ y;k
Q y;k
1
Q yk yk Q C O .0 C yOk / D xQ y;k Q yQk
1
Q y;k
Lk
Putting all of the pieces together, we get the general update equation:
xO C D xO C L yQ k k k k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 7 of 9
3.2.2: The Kalman-filter gain factor
Uncertainty of state estimate
key off: store data
key on: initialize
State-estimate covariance (uncertainty) C
meas. voltage estimate estimate balance compute
current state of state of power
xQ ;k is
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
C T
xQ ;k D E .xk xO kC/.xk xO kC/T D E .xk xO k / Lk yQk .xk xO k / Lk yk Q
D E xQ k Lk yQk xQ k Lk yQk T
T
D xQ ;k Lk E yQk
.x k yQk Lk C Lk y;k
Q k /T E xQ T
Q Lk
T
y;k
Q LT
k
Lk y;k
Q
D xQ ;k Q Lk
Lk y;k
T
Final output of Gaussian sequential probabilistic inference has two parts:
1. State estimate. Best guess of present state value, xO kC
2. Covariance estimate. Uncertainty of xO kC , C
xQ ;k can yield error bounds
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 8 of 9
3.2.2: The Kalman-filter gain factor
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Generic Gaussian sequential probabilistic inference recursion:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
O kC D xO k C Lk yk yOk D xO k C Lk yQk
x
C
xQ ;k D xQ ;k
T
Q Lk
Lk y;k
where
O D E xk j Yk 1
xk xQ ;k D E .xk xO k /.xk xO k /T D E .xQ k /.xQ k /T
O kC D E xk j Yk
x
C
xQ ;k D E .xk O kC/.xk xO kC/T D E .xQ kC/.xQ kC/T
x
Ó k D E ´k j Yk 1 Q D E .yk yOk /.yk yOk / D E .yQk /.yQk /
y;k
T T
Lk D E .xk O k /.yk yOk / y;k Q D xQ y;k
T 1 1
x Q y;k
Q
Note that this is a linear recursion, even if the system is nonlinear(!)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 9 of 9
3.2.3: Summarizing the six steps of generic sequential probabilistic inference
Steps 1a–1b
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Generic Gaussian probabilistic inference solution can be
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
divided into two main steps, each having three sub-steps
General step 1a: State prediction time update
Each time step, compute an updated prediction of the present value of xk , based
on prior information and the system model
O D E xk j Yk
xk 1 D E f .xk 1 ; uk 1 ; wk 1 / j Yk 1
General step 1b: Error covariance time update
Determine the predicted state-estimate error covariance matrix xQ ;k based on
prior information and the system model
We compute xQ ;k D E .xQ k /.xQ k /T , where xQ k D xk xO k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 1 of 5
3.2.3: Summarizing the six steps of generic sequential probabilistic inference
Step 1c
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
General step 1c: Predict system output yk
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Predict the system’s output using prior information
loop once each measurement interval while pack is active
O D E yk j Yk
yk 1 D E h.xk ; uk ; vk / j Yk 1
Summarizing general step 1
Step 1a: State prediction time update
Prediction
Step 1b: Error covariance time update
Step 1c: Predict system output
Step 2a: Estimator gain matrix
Correction
Dr. Gregory L. Plett j University of Colorado Colorado Springs
Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 2 of 5
Step 2b: State estimate measurement update
3.2.3: Summarizing the six steps of generic sequential probabilistic inference
Step 2c: Error covariance measurement update
Step 2
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
General step 2a: Estimator gain matrix Lk
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Compute the estimator gain matrix Lk D xQ y;k 1
loop once each measurement interval while pack is active
Q y;kQ
General step 2b: State estimate measurement update
Compute the posterior state estimate by updating the prediction using the Lk and
the innovation yk O
yk
O C D xO k C Lk .yk
xk O
yk /
General step 2c: Error covariance measurement update
Compute the posterior error covariance matrix
C C
xQ ;k D E .x Q k /.xQ kC/T
D xQ ;k Lk y;k Q Lk
T
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 3 of 5
Step 1a: State prediction time update
Prediction
Step 1b: Error covariance time update
3.2.3: Summarizing the six steps of generic sequential probabilistic inference
Summarizing step 2
key off: store data
key on: initialize
Step 1c: Predict system output meas. voltage estimate estimate balance compute
Summarizing general step 2:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Step 2a: Estimator gain matrix
Correction
Step 2b: State estimate measurement update
Step 2c: Error covariance measurement update
KEY POINT : Estimator outputs state estimate xO kC , error covariance estimate C
xQ ;k q
That is, we have high confidence that the truth lies within xO kC 3 diag.C
xQ ;k /
Estimator then waits until next sample interval, updates k , proceeds to step 1a
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 4 of 5
3.2.3: Summarizing the six steps of generic sequential probabilistic inference
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
current state of state of power
Step 1a: State prediction time update temperature charge (SOC) health (SOH) cells limits
✂ ✝ ✠ ✚
①❖❦ ❉ ❊ ✁ ✄ ❥ ❨☎✆✶ ✞ ✟ ❢ ✳✡ ☛☞✌ ❀ ✉ ✍✎✏ ✑ ✇✒✓✔ ✴ ✕ ✖✗✘✙ ✛
loop once each measurement interval while pack is active
Prediction
Step 1b: Error covariance time update
✧ ✷ ✺
✻✢✜◗✣✤ ✥ ✦ ★✪✩✬✫ ✭✮✰✯✲✱ ✵❚ ✸ ✹ ✼✽ ✾ ✿ ❂❁❄❃ ❅❆❇ ❈ ❋ ❍●❏■ ❑▲ ❪▼
Step 1c: Predict system output
❯ ❴ ❜ ③
②◆P ❘ ❙ ❱❲ ❳ ❩❬❭❫ ❵ ❛ ❤❝❞ ❡ ❣ ✐ ❧ ♠ ✈ ♥ ♦ ♣ qrst ④
Step 2a: Estimator gain matrix
⑤ ⑥ ⑦ ⑧❶⑨⑩❸❷ ❹❺ ❻ ❼❽
❿❾ ➀➁ ➂
Correction
Step 2b: State estimate measurement update
➄➃➆➅ ➇ ➉➈➋➊ ➌ ➍ ➎ ➏➐➑ ➒ ➔➓→ ➣↔
Step 2c: Error covariance measurement update
↕➜➙➛➝➞ ➟ ➠➤➡➢➥➦ ➧ ➨ ➩ ➫ ➯➭ ➲➳ ➵ ➺➸ ➻
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 5 of 5
3.2.4: Deriving the three Kalman-filter prediction steps
Application to linear systems: Kalman filter
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We now apply general Gaussian sequential probabilistic
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
inference solution to specific case of a linear system
Linear systems have desirable property that all pdfs remain Gaussian if stochastic
inputs are Gaussian, so the assumptions made in deriving filter steps hold exactly
The linear Kalman filter assumes that the system being modeled can be
represented in the “state-space” form
xk D Ak 1xk 1 C Bk 1uk 1 C wk 1
yk D Ck xk C Dk uk C vk
Input is uk , output is yk , state is xk ; process noise is wk , sensor noise is vk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 1 of 7
3.2.4: Deriving the three Kalman-filter prediction steps
Assumptions on noises
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We assume that wk and vk are mutually uncorrelated white
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Gaussian random processes, with zero mean and covariance
matrices having known value:
( (
Ewn wkT D
e
w; n D kI Evn vkT D
vQ ; n D kI
0; n ¤ k: 0; n ¤ k,
and Ewk x0T D 0 for all k
The assumptions on the noise processes wk and vk and on the linearity of system
dynamics are rarely (never) met in practice, but the consensus of the literature and
practice is that the method still works very well
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 2 of 7
3.2.4: Deriving the three Kalman-filter prediction steps
Deriving the linear Kalman filter (1a)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF step 1a: State prediction time update.
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Here, we compute the predicted state
loop once each measurement interval while pack is active
O D E f .xk 1; uk 1; wk 1/ j Yk 1
xk
D E Ak 1xk 1 C Bk 1uk 1 C wk 1 j Yk 1
D E Ak 1xk 1 j Yk 1 C E Bk 1uk 1 j Yk 1 C E wk 1 j Yk 1
D Ak 1xO kC 1 C Bk 1uk 1,
by the linearity of expectation, noting that wk 1 is zero-mean
INTUITION: When predicting the present state given only past measurements, the
best we can do is to use the most recent state estimate and system model,
propagating the mean forward in time
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 3 of 7
3.2.4: Deriving the three Kalman-filter prediction steps
Deriving the linear Kalman filter (1b)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF step 1b: Error covariance time update
current state of state of power
temperature charge (SOC) health (SOH) cells limits
First, we note that the prediction error is xQ k D xk xO k , so
loop once each measurement interval while pack is active
Q D xk xO k D Ak 1xk 1 C Bk
xk 1 uk 1 C wk 1 O kC 1
Ak 1 x Bk 1 uk 1
D Ak 1xQ kC 1 C wk 1
Therefore, the covariance of the prediction error is
xQ D E .x
k
Q k /.xQ k /T D E .Ak 1xQ kC 1 C wk 1/.Ak 1xQ kC 1 C wk 1/T
D E Ak 1xQ kC 1.xQ kC 1/T ATk 1 C wk 1.xQ kC 1/T ATk 1 C Ak 1xQ kC 1wkT 1 C wk 1wkT 1
D Ak 1CxQ ;k 1ATk 1 C ew
Therefore cross terms drop out of the final result since the white process noise
wk 1 is not correlated with the state at time k 1 and is zero mean
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 4 of 7
3.2.4: Deriving the three Kalman-filter prediction steps
Intuition for step 1b
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
current state of state of power
INTUITION : When estimating state-prediction error covariance temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
xQ D Ak C T
C ew
k 1 xQ ;k 1 Ak 1
Best to use most recent covariance estimate and propagate forward in time
For stable systems, Ak 1 C T
xQ ;k 1 Ak 1 is contractive: covariance gets “smaller”
State of stable systems always decays toward zero in absence of input, or
toward a known trajectory if uk ¤0
Term tells us that we tend to get increasingly certain of state estimate over time
On the other hand, e
w adds to the covariance
Unmeasured inputs add uncertainty to our estimate because they perturb the
trajectory away from the known trajectory based only on uk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 5 of 7
3.2.4: Deriving the three Kalman-filter prediction steps
Deriving the linear Kalman filter (1c)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF step 1c: Predict system output
current state of state of power
temperature charge (SOC) health (SOH) cells limits
We predict the system output as
loop once each measurement interval while pack is active
O D E h.xk ; uk ; vk / j Yk 1 D E Ck xk C Dk uk C vk j Yk
yk 1
D E Ck xk j Yk 1 C E Dk uk j Yk 1 C E vk j Yk 1
D Ck xO k C Dk uk ,
since vk is zero-mean
O is best guess of system output, given only past measurements.
INTUITION : yk
The best we can do is to predict the output given the output equation of the
system model, and our best guess of the system state at the present time
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 6 of 7
3.2.4: Deriving the three Kalman-filter prediction steps
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We have started to specialize the general Gaussian
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
sequential-probabilistic-inference steps to the special case of
a linear system, to develop the Kalman-filter equations
The general steps are not implementable as computer programs, since they involve
statistical operations (expected values)
However, the specialized Kalman-filter steps we have developed so far are
completely valid program operations
By examining the equations, we can gain insight into the operation of a Kalman filter
and have confidence that it is performing reasonable actions
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 7 of 7
3.2.5: Deriving the three Kalman-filter correction steps
Finding the Kalman gain matrix
key off: store data
key on: initialize
KF step 2a: Estimator (Kalman) gain matrix Lk D xQ y;k
meas. voltage estimate estimate balance compute
1
current state of state of power
cells
Q y;k
Q
temperature charge (SOC) health (SOH) limits
To compute the Kalman gain, we must first compute several
loop once each measurement interval while pack is active
covariance matrices: we first find y;k
Q
Q D yk yOk D Ck xk C Dk uk C vk Ck xO k Dk uk D Ck xQ k C vk
yk
Q D E .Ck xQ k C vk /.Ck xQ k C vk /
T
y;k
D E Ck xQ k .xQ k /T CkT C vk .xQ k /T CkT C Ck xQ k vkT C vk vkT
D Ck xQ ;k CkT C vQ .
Cross terms are zero since vk is zero mean and is uncorrelated with xQ k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 1 of 8
3.2.5: Deriving the three Kalman-filter correction steps
Finding the Kalman gain matrix
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We now find xQ y;k
current state of state of power
Q :
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
ExQ k yQkT D E Q
xk .Ck xk Q C vk /T D E xQ k .xQ k /T CkT C xQ k vkT
D T
xQ ;k Ck
Combining y;k
Q and xQ y;k
Q in Lk D xQ y;k 1
Q ,
Q y;k
Lk D xQ ;k CkT Ck xQ ;k CkT C vQ 1
INTUITION : Note that the computation of Lk is the most critical aspect of Kalman
filtering that distinguishes it from a number of other estimation methods
The whole reason for calculating covariance matrices is to be able to update Lk
Lk is time-varying: It adapts to give the best update to the state estimate based
on present conditions
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 2 of 8
3.2.5: Deriving the three Kalman-filter correction steps
Finding the Kalman gain matrix
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
current state of state of power
(continued): Recall that we use Lk in the equation
cells
INTUITION
temperature charge (SOC) health (SOH) limits
O C D xO k C Lk .yk O
loop once each measurement interval while pack is active
xk yk /
Q , indicates relative need for correction to xO k and
The first component to Lk , xQ y;k
how well individual states within xO k are coupled to the measurements
Q D xQ ;k Ck
We see this clearly in xQ y;k T
xQ ;k tells us about state uncertainty at the present time, which we hope to reduce
as much as possible
A large entry in xQ ;k means that the corresponding state is very uncertain and
therefore would benefit from a large update
A small entry in xQ ;k means that the corresponding state is very well known
already and does not need as large an update
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 3 of 8
3.2.5: Deriving the three Kalman-filter correction steps
Finding the Kalman gain matrix
key off: store data
key on: initialize
Q D
meas. voltage estimate estimate balance compute
xQ ;k CkT
current state of state of power
INTUITION (continued): Continuing to look at xQ y;k
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
The CkT term gives the coupling between state and output
Entries that are zero indicate that a particular state has no direct influence on a
particular output and therefore an output prediction error should not directly
update that state
Entries that are large indicate that a particular state is highly coupled to an
output so has a large contribution to any measured output prediction error;
therefore, that state would benefit from a large update
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 4 of 8
3.2.5: Deriving the three Kalman-filter correction steps
Finding the Kalman gain matrix
key off: store data
key on: initialize
(continued): Still looking at Lk D xQ y;k
meas. voltage estimate estimate balance compute
1
current state of state of power
Q ,
cells
INTUITION Q y;k
temperature charge (SOC) health (SOH) limits
yQ tells us how certain we are that measurement is reliable
loop once each measurement interval while pack is active
If yQ is “large,” we want small, slow updates
If yQ is “small,” we want big updates
This explains why we divide the Kalman gain matrix by yQ
The form of yQ D Ck xQ ;k CkT C vQ can also be explained
Ck xQ CkT indicates how error in state contributes to error in output estimate
vQ term indicates uncertainty in sensor reading due to sensor noise
Since sensor noise is assumed independent of the state, uncertainty in
Q D yk
yk ykO adds the uncertainty in yk to the uncertainty in yOk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 5 of 8
3.2.5: Deriving the three Kalman-filter correction steps
Finding the Kalman gain matrix
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF step 2b: State estimate measurement update
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Computes a posteriori state estimate by updating a priori
loop once each measurement interval while pack is active
estimate using estimator gain and output prediction error yk ykO
OC
x
k D xO k C Lk .yk O
yk /
INTUITION : yk O
is predicted measurement, based on present state prediction
Therefore, yk yOk is what is unexpected or new in the measurement
We call this term the innovation. The innovation can be due to a bad system
model, state error, or sensor noise
So, we want to use this new information to update the state, but must be careful
to weight it according to the value of the information it contains
Lk is the optimal blending factor, as we have already discussed
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 6 of 8
3.2.5: Deriving the three Kalman-filter correction steps
Finding the Kalman gain matrix
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF step 2c: Error covariance measurement update
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Finally, we update error covariance matrix
loop once each measurement interval while pack is active
C T
xQ ;k D xQ ;k Q Lk D xQ ;k Lk y;k
T T
Lk y;k Q y;k
Q xQ y;k
Q
D xQ ;k Lk Ck xQ ;k
D .I Lk Ck /xQ ;k
INTUITION :
A covariance matrix is positive semi-definite, and Lk y;kQ LTk is also a
positive-semi-definite form, and we are subtracting this from the predicted-state
covariance matrix; therefore, the resulting covariance is “lower” than the
pre-measurement covariance
Measurement update has decreased our uncertainty in state estimate
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 7 of 8
3.2.5: Deriving the three Kalman-filter correction steps
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Have now derived the entire Kalman filter algorithm
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Next week, you will learn how to implement and visualize KF
KEY POINT: Repeating from before, recall that the estimator output comprises the
state estimate xO kC and error covariance estimate C xQ ;k q
That is, we have high confidence that the truth lies within xO kC 3 diag.C
xQ ;k /
COMMENT: If a measurement is missed for some reason, then skip steps 2a–c for
that iteration. That is, set Lk D 0 and xO kC D xO k and C xQ ;k D xQ ;k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 8 of 8
3.2.6: Where from here?
Summary of this week
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
This week, you learned:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
That the generic MMSE solution is Exk j Yk
loop once each measurement interval while pack is active
That generic Gaussian probabilistic inference solution uses predict/correct
recursion to find this solution
Predict and correct steps both have three sub-steps
With certain assumptions, can specialize these steps to linear Kalman filter
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 1 of 3
3.2.6: Where from here?
Where from here?
key off: store data
key on: initialize
Will apply linear KF to linearized battery-cell model to help
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
visualize the steps involved
loop once each measurement interval while pack is active
See how to simulate random systems as preparation for
evaluating KF via simulation
Learn how to implement KF in Octave/MATLAB and evaluate
results computed by the KF
Some “improvements” to KF to handle real-world numeric
round-off issues
How KF also provides helpful ways to detect sensor faults
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 2 of 3
3.2.6: Where from here?
Credits
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Credits for photos in this lesson
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
“Map” on slide 2: Pixabay license
(https://pixabay.com/en/service/license/), from
https://pixabay.com/en/map-treasure-pirate-309928/
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Introducing the linear Kalman filter as a state estimator j 3 of 3
3.3.1: Visualizing the Kalman filter with a linearized cell model
Visualizing the linear Kalman filter
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Linear Kalman-filter equations naturally form a recursion:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Initialization
①❖✵❈ ✱ ✻✁◗ ❀✂ ➎➏➐➑ ➒➓➔→ ➣↔↕➙➛➜➝ ➞➟➠➡➢➤➥➦➧ ➨
1a 2c “Simple” to implement on
☎✄❦✆ ❉ ❆✝✞✶ ✠✟☛☞✌
✡ ✍ ❇✎✏✑ ✉ ✒✓✔ ⑩❸❶❷❹❺ ❻ ❼❽ ❾ ❿ ➀ ➁➂ ➃➄➇➅➆➈➉ a digital computer
1b 2b However, note that our
✕✘✖✗✙✚ ✛ ✜✢✣✤ ✥★✦✧✩✪✫✬ ✭✮✯✰
❚ ✲✳✇❡
Meas. ❤❣❥✐ ❧ ♥♠♣♦ q r s t③④ ⑤ ⑦⑥⑧ ⑨ cell models are nonlinear,
Meas.
1c
➌➍
2a ➊➋ so we cannot apply
②✴✷ ✸ ✹✺ ✽✼✿✾ ❁ ❂❃ ❄ ❅ ▲ ❊ ❋●❏❍■❑▼ ◆❘P ❬❙❯ ❱❨❲❳❩❭ ❪❴❫ ❵❛✈❜ ❝❞❢ (linear) Kalman filter to
them directly
Prediction Correction
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 1 of 8
3.3.1: Visualizing the Kalman filter with a linearized cell model
Setting up a demonstration (1)
key off: store data
key on: initialize
meas. voltage estimate estimate compute
To demonstrate the KF steps, we’ll develop
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
and use a crude cell model
OCV versus SOC for four cells at 25°C
1
´ k C1 D 1 ´ k ik
3600 Q
4
Open-circuit voltage (V)
voltk D 3:5 C 0:7 ´k R0 ik 3.5
Notice that we have:
3
Linearized the OCV relationship
Omitted diffusion voltages 2.5
Omitted hysteresis voltages 0 20 40 60 80 100
State of charge (%)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 2 of 8
3.3.1: Visualizing the Kalman filter with a linearized cell model
Setting up a demonstration (2)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Model still isn’t linear because of “3.5” in output equation
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Debias measurement via yk D voltk 3:5 and use
loop once each measurement interval while pack is active
D 1 ´k 36001 Q ik
OCV versus SOC for four cells at 25°C
´ k C1
4
Open-circuit voltage (V)
yk D 0:7 ´k R0 ik
Define state xk ´k and input uk ik
3.5
For the sake of example, we will use 3
Q D 10000=3600 and R0 D 0:01
Yields state-space description with A D 1, 2.5
B D
1 10 4 , C D 0:7, and D D 0:01 0 20 40 60 80 100
w D 10 , and vQ D 0:1
State of charge (%)
We also model e 5
We assume no initial uncertainty so xO 0C D 0:5 and C
xQ ;0 D 0
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 3 of 8
3.3.1: Visualizing the Kalman filter with a linearized cell model
Iteration 1
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We look at the first iteration of the linear KF, assuming
current state of state of power
temperature charge (SOC) health (SOH) cells limits
D 1, i1 D 0:5 and v1 D 3:85
loop once each measurement interval while pack is active
i0
①❖❦ ❉ ❆✁✂✶ ☎✄✆✝✞❈ ✟ ❇✠✡☛ ✉ ☞✌✍ ✏✎✒✑ ✓ ✔ ✕ ✵✿✺ ✖ ✗✘✙✹ ✚ ✛ ✜ ✢✣✤✾✥✦
✻★✧◗❀✩ ✪ ✫✬✭✮ ✯✲✰✱✳✴✷✸ ✼✽❁❂ ❚ ❃ ❄✇❡ ❅●❊❋❍■ ❏ ❑ ▲ ▼ ◆ P ❘ ❙❯❱❲ ❳ ❨❩❬❭
②❪❫ ❴ ❵❛ ❝❜❢❞ ❣ ❤✐ ❥ ❧ ♥♠♦ ♣ qrs t ✈③④⑤⑥⑦ ⑧ ⑨⑩❶❷ ❸ ❹❺❻ ❼ ❽❾❿➀➁➂➃
➄ ➅ ➆ ➇➊➈➉➋➌ ➍➏➎ ➐➑➒ ➓➣➔→↔↕ ➙➜➛ ➝➞➠➟ ➡➢➤ ➥ ➦ ➧ ➨➩➫➭ ➯ ➲➳➵➸➺➻➼➽ ➾ ➚➪➶➹ ➘ ➴➷➬➮➱✃
❐ ❒❮❰ÏÐÑÒÓÔÕÖ×
ÙØÛÚ Ü ÞÝàß á â ã äåæ ç éèê ë ÿþ✶❈ ❉ ✵✿✹✾ ✁ ✂ ✻✄☎✆✝✞✟✠✡☛☞✺ ✳✌✍✸✎ ✏ ✑✒✓✔✕✖✗✴
ìíîïðñ òó ô õö÷ø ù úûüý ✘ ✙✚✛✜✢✣✤✥✦
✧ ①★ ✱
◗ ❀❦ ✩ ✪■ ✫ ▲ ✬ ✭ ✮ ✯✰✷✼✽
✲ ❁ ❄❂ ❯❱
❃ ❅❆ ❇ ❊❋ ● ❍❏❑▼◆❖P❘❙❚ ❲ ❳❨❩❬ ❭ ❪❫
❴❵
❧♠
p ❛ ❜❝❞❡❢❣❤✐❥
Output: Ó D 0:4999 3 9:9995 10 6 D 0:4999 0:0094866
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 4 of 8
3.3.1: Visualizing the Kalman filter with a linearized cell model
Iteration 2
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
For the second iteration of the linear KF, let
current state of state of power
temperature charge (SOC) health (SOH) cells limits
D 0:5, i2 D 0:25, and v2 D 3:84
loop once each measurement interval while pack is active
i1
❈ ✟ ❇✠✡☛ ✉ ☞✌✍
①❖❦ ❉ ❆✁✂✶ ☎✄✆✝✞ ✏✎✒✑ ✓ ✵✿✹✾✔✕✖✗✘ ✙ ✚✛✜✢ ✣ ✤✥✺ ✦ ✧★✩✪✫✽✬
✻✮✭◗❀✯ ✰ ✱✲✳✴ ✷❁✸✼❂❃❄❅ ❊❋●❍
❚ ■ ❏✇❡ ❑ ▲ ❙ ❯❱❲❳❨❩❬❭❪❫❴ ❵ ❛❜❝❞ ❢ ❣❤✐❥❧♠♥♦♣qrs
◆▼ P❘
⑧
②t✈ ③ ④⑤ ⑦⑥⑨ ⑩ ❶❷ ❸ ❹ ❻❺❼ ❽ ❾❿➀➁➂➃➄➅➆➇➈ ➉ ➊➋➌➍ ➎ ➏➐➑➒ ➓ ➔→➣↔↕➙➛➜
➡ ➨ ➳
➝ ➞ ➟ ➠➤➢ ➥➦ ➧➩ ➫➭➯ ➲➸➵ ➺➻ ➼➾ ➚➪➹➶ ➘ ➬ ➮ ➱ ✃❐❒❮❰ÏÐÑÒÓÔÕ Ö ×ØÙÚÛÜÝÞßàáâãäåæ ç èéêë ì íîïðñò
➽ ➴➷
ó ôõö÷øùúûüý
ÿþ❦❈ ❉ ①❖✁ ✂ ▲ ✄ ✳②☎ ✆ ✞✝✟ ✴ ✑✏✷✒ ✓ ✵✔✕✾✖✗✘ ✙ ✚✛✜✢✣✶✤✥✦✧★✩✪✫✬ ✯ ✰✱✲✻✼❀❁❂❃
✭✇❤❡r✠ ✡☛ ☞ ✸✿✽✹ ✌ ✍✎✺✮ ❄ ❅❆❇❊❋●❍■
❏▼❑◗◆P ❘ ❙❚ ❯ ❱ ❲ ❳❨ ❩❬❫❭❪❴❵ ❛❞❜❝❢❣ ✐ ❥❧ ♠ ♥♦♣qst✉✈③④ ⑤ ⑥⑦⑧⑨ ⑩ ❶❷❸❹❺❻❼❽❾❿➀➁
➂ ➃➄➅➆➇➈➉➊➋➌➍➎
p
Output: Ó D 0:4998 3 1:99976 10 5 D 0:4998 0:013416
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 5 of 8
3.3.1: Visualizing the Kalman filter with a linearized cell model
Covariance
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Note that covariance (uncertainty) converges over time
current state of state of power
temperature charge (SOC) health (SOH) cells limits
✆✟✝✞✠✷ ✡ ☛✿✾☞✌✍✎✏✑✒✓✔ ✕✘✖✗✙✸ ✚ ✛✜✢✣✤✼✥✦✧★✩✪
loop once each measurement interval while pack is active
✻①◗ ❀✶ ❉ ✁✂✄✵☎✺
✫✭❈✬✮✯ ✰ ✱ ✲✹✳✴✽❁ ❂ ❃❄❅❆❇❊❋●❍■❏❑ ▲❖▼◆P❘ ❙ ❚❯❱❲❳❨❩❬❭❪❫❴ ❵❝❛❜❞❡ ❢ ❣❤✐❥❦❧♠♥♦♣qr
Covariance increases during propagation, is then reduced by each measurement
Covariance still oscillates in steady state between xQ ;ss and C
xQ ;ss
q C
Estimation error bounds are 3 xQ ;k for 99 % assurance of estimate’s accuracy
C
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 6 of 8
3.3.1: Visualizing the Kalman filter with a linearized cell model
Results
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
The plots below show a sample of the Kalman filter operating
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
We’ll soon look at how to write code to evaluate this example
▼❛❧♠◆♥ ❖Pt❡r ✐◗ ❘❝❯❱♦❲ ❊r❂♦❃ ❝❄✈❛❅✐❆♥❇❡ ❉①❋ ▲
◗ ●❦ ❍❏❞ ❑◆▼ ❖P
★✩✪ ✢✣✤✥
✥✦✧
❁
✢✣✤ ❀
✿ ✙✚✛✜
×
❀✿ ✚✛✜ ❚❁❂❃ ✷✵
✴✷ ❊❄❅❆❇❈❉❋ ✴
❙ ✗✘✙ ●❍❏❑▲ ✲✳✱
✰✯ ✕✖✗✘
✔✕✖ ✮
❈ ❘❯❙
❚
✵
❱❨❲
❳
✲✑✒✓ ✑✒✓✔
✸ ✁ ✹✂✄ ✺☎✆ ✻✝✞ ✼✟✠ ✽✡☛ ✾☞✌ ✶✍✎✏ ✸ ✁ ✹✂✄ ✺☎✆ ✻✝✞ ✼✟✠ ✽✡☛ ✾☞✌ ✶✍✎✏
■✫✬✭✮✯✰✱✳ ■✦✧★✩✪✫✬✭
Note that Kalman filter does not perform especially well since vQ is quite large
However, these are best-possible results, since KF is the optimum MMSE estimator
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 7 of 8
3.3.1: Visualizing the Kalman filter with a linearized cell model
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF implements an MSEE-optimal state estimator for linear
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
systems if assumptions regarding system noises are met
KF equations naturally form a recursive algorithm for estimating the state
While the KF works only for linear systems, we can linearize system dynamics for
an approximate result (will learn better ways later)
Example with a simplified battery model demonstrates the kinds of results you can
expect from a KF
The KF will provide a state estimate. . .
And uncertainty bounds (error bounds) on that estimate
We will find this very helpful for SOC estimation
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 8 of 8
3.3.2: Introducing Octave code to generate correlated random vectors
Generating correlated random vectors
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Straightforward to implement KF steps in Octave
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
To exercise KF code, however, also must simulate system
whose state is being estimated
Produces system input/output data as input to the KF
Provides “true” state value at every point in time
But, to simulate true system, must be able to create nonzero mean Gaussian noise
with covariance e Y.
That is, we want Y N .y; N eY / but randn.m returns X N .0; I /.
How to convert X to Y ?
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 1 of 5
3.3.2: Introducing Octave code to generate correlated random vectors
A needed transformation
key off: store data
key on: initialize
Suppose that we can find matrix A such that AT A D e
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Y loop once each measurement interval while pack is active
Then, we generate samples x of random vector
X N .0; I / and compute samples y of random vector Y D yN C AT X
Since X is zero-mean, EY D EyN C AT X D yN , as desired
The covariance of Y is
E.Y N
y/.Y N T
y/ D E.AT X /.AT X /T
D E.AT X /.X T A/
D AT EXX
T
A D eY;
I
also as desired
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 2 of 5
3.3.2: Introducing Octave code to generate correlated random vectors
Important matrix factorizations
key off: store data
key on: initialize
So, if we can find A such that AT A D e
meas. voltage estimate estimate balance compute
current state of state of power
Y , we can generate
temperature charge (SOC) health (SOH) cells limits
samples x of X N .0; I / and compute y D yN C AT x
loop once each measurement interval while pack is active
Three ways to generate A
Y computes A such that A A D e
Cholesky factorization of e T
Y as long as eY is
positive definite (all eigenvalues are strictly greater than zero)
Y produces L, D such that LDL D e
LDL factorization of e T
Ypand requires
only that eY be positive semi definite: Can then compute A D DLT
Y produces L, U such that LU D e
LU factorization of e Yand requires
only that p
Y be
e positive semi definite: Can then compute A D diag diag.U / LT
Cholesky and LU are built into Octave; all are built into MATLAB
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 3 of 5
3.3.2: Introducing Octave code to generate correlated random vectors
Example Octave code
key off: store data
key on: initialize
meas. voltage estimate estimate compute
Suppose we wish to generate a single
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
sample of y using Cholesky method:
loop once each measurement interval while pack is active
ybar = [1; 2];
5000 samples with mean [1;2] and
covar = [1 , 0.5; 0.5 , 1];
covar [1,0.5; 0.5,1]
A = chol ( covar ) ; 5
x = randn ([2 , 1]) ;
y = ybar + A '* x ; 4
Suppose we wish to generate 5000
y coordinate
3
samples of y using the LU method: 2
ybar = [1; 2];
covar = [1 , 0.5; 0.5 , 1]; 1
[L , U ] = lu ( covar ) ;
0
A = diag ( sqrt ( diag ( U ) ) ) *L ';
x = randn ([2 ,5000]) ; −1
y = ybar (: , ones ([1 5000]) ) +A '* x ; −4 −2 0 2 4 6
x coordinate
plot ( y (1 ,:) ,y (2 ,:) , '. ') ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 4 of 5
3.3.2: Introducing Octave code to generate correlated random vectors
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
In order to test KF code via simulation, must also simulate
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
system whose state is being estimated
This requires ability to simulate (possibly) correlated (possibly nonzero mean)
noises
Octave natively produces samples of zero-mean uncorrelated Gaussians only
But, can transform these to have desired mean and correlation using a matrix
determined using either the Cholesky or LU matrix decompositions, both of which
are built into Octave
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 5 of 5
3.3.3: Introducing Octave code to implement KF for linearized cell model
Introduction to lesson
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
In this lesson, you will learn how to implement KF in Octave
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Recommend you have KF equations nearby as reference
loop once each measurement interval while pack is active
w D vQ D 1 and
Code will be demonstrated for simulated system where e
xk D xk 1 C uk 1 C wk 1
yk D xk C vk
Can be modified in straightforward ways to simulate other systems having other
noise sources
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 1 of 7
3.3.3: Introducing Octave code to implement KF for linearized cell model
Initializing the simulation
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
The code begins with some definitions, initializations
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
% Initialize simulation variables
SigmaW = 1; % Process noise covariance
SigmaV = 1; % Sensor noise covariance
A = 1; B = 1; C = 1; D = 0; % Plant definition matrices
maxIter = 40; % Number of iterations to execute simulation
xtrue = 0; % Initialize true system initial state
xhat = 0; % Initialize Kalman filter initial estimate
SigmaX = 0; % Initialize Kalman filter covariance
u = 0; % Unknown initial driving input : assume zero
% Reserve storage for variables we want to plot / evaluate
xstore = zeros ( length ( xtrue ) , maxIter +1) ; xstore (: ,1) = xtrue ;
xhatstore = zeros ( length ( xhat ) , maxIter ) ;
SigmaXstore = zeros ( length ( xhat ) ^2 , maxIter ) ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 2 of 7
3.3.3: Introducing Octave code to implement KF for linearized cell model
KF step 1
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Code continues with main KF loop, step 1
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
for k = 1: maxIter ,
% KF Step 1 a : State estimate time update
xhat = A * xhat + B * u ; % use prior value of " u "
% KF Step 1 b : Error covariance time update
SigmaX = A * SigmaX *A ' + SigmaW ;
% [ Implied operation of system in background , with
% input signal u , and output signal z ]
u = 0.5* randn (1) + cos ( k / pi ) ; % for example ...
w = chol ( SigmaW ) '* randn ( length ( xtrue ) ) ;
v = chol ( SigmaV ) '* randn ( length ( C * xtrue ) ) ;
ytrue = C * xtrue + D * u + v ; % z is based on present x and u
xtrue = A * xtrue + B * u + w ; % future x is based on present u
% KF Step 1 c : Estimate system output
yhat = C * xhat + D * u ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 3 of 7
3.3.3: Introducing Octave code to implement KF for linearized cell model
KF step 2
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Code continues with main KF loop, step 2
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
% KF Step 2 a : Compute Kalman gain matrix
SigmaY = C * SigmaX *C ' + SigmaV ;
L = SigmaX *C '/ SigmaY ;
% KF Step 2 b : State estimate measurement update
xhat = xhat + L *( ytrue - yhat ) ;
% KF Step 2 c : Error covariance measurement update
SigmaX = SigmaX - L * SigmaY *L ';
% [ Store information for evaluation / plotting purposes ]
xstore (: , k +1) = xtrue ; xhatstore (: , k ) = xhat ;
SigmaXstore (: , k ) = SigmaX (:) ;
end
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 4 of 7
3.3.3: Introducing Octave code to implement KF for linearized cell model
Plot results
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Finally, there is code to plot results
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
figure (1) ; clf ;
plot (0: maxIter -1 , xstore (1: maxIter ) ', 'k - ' ,...
0: maxIter -1 , xhatstore ' , 'b - - ' , ...
0: maxIter -1 , xhatstore '+3* sqrt ( SigmaXstore ) ', 'm -. ' ,...
0: maxIter -1 , xhatstore ' -3* sqrt ( SigmaXstore ) ', 'm -. ') ; grid ;
legend ( ' true ' , ' estimate ' , ' bounds ') ;
title ( ' Kalman filter in action ') ;
xlabel ( ' Iteration ') ; ylabel ( ' State ') ;
figure (2) ; clf ;
plot (0: maxIter -1 , xstore (1: maxIter ) '- xhatstore ' , 'b - ' ,...
0: maxIter -1 ,3* sqrt ( SigmaXstore ) ', 'm - - ' ,...
0: maxIter -1 , -3* sqrt ( SigmaXstore ) ', 'm - - ') ; grid ;
legend ( ' Error ' , ' bounds ' ,0) ; title ( ' Error with bounds ') ;
xlabel ( ' Iteration ') ; ylabel ( ' Estimation Error ') ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 5 of 7
3.3.3: Introducing Octave code to implement KF for linearized cell model
Sample results
key off: store data
key on: initialize
w D vQ D 1, and
meas. voltage estimate estimate balance compute
Plots show sample output from KF for e
current state of state of power
temperature charge (SOC) health (SOH) cells limits
D xk 1Cuk 1Cwk
loop once each measurement interval while pack is active
xk 1 ❿➀➁➂➃ ➄➅➆➇ ➈➉➊➋➌➍
D xk C vk
❑✯❧♠✰✱ ✳✴✻✼✽ ✾✿ ❀❝❁❂❃❄
✒✓ ❭
yk ✏✑
t✔✕✖
❡✗✘✙✚✛✜✢
❜✣✤✥✦✧
❾
❽ ❬
❼
❻
❺ ❩
Error never converges ✭✮
✬✫
✎ ❹
❷❸❶ ❨
⑩
to zero, but stays ❙ ✍
⑨⑧ ❲❳ ❪❫❴❵❞
⑦ ❢❣❤❥❦♣
⑥
within predicted ☞✌
⑤ ❯❱
bounds (both as ✲✡☛
✵ ✺ ✶ ✁✂ ✷✄ ☎✆ ✸✝ ✞✟ ✹✠
❘❚
❅ ❆ ❇❈ ❉❊ ❋● ❍❏ ▲▼ ◆❖ P◗
expected)
■★✩r❛✪✐♦♥ qs✉✈✇①②③④
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 6 of 7
3.3.3: Introducing Octave code to implement KF for linearized cell model
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
You have now learned how to
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Implement KF code in Octave
loop once each measurement interval while pack is active
Cosimulate a system of interest with the KF to validate KF performance
Code can be used to estimate state of any linear system obeying KF assumptions
Sample results give intuitive sense for what to expect from KF
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 7 of 7
3.3.5: Can we automatically detect bad measurements with a Kalman filter?
Real-world issue: Voltage-sensor faults
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Sometimes systems for which we would like a state estimate
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
have sensors with intermittent faults
Would like to detect faulty measurements and discard them
Time update steps of the KF still implemented
Measurement update steps are skipped (Lk D 0)
KF provides elegant theoretical means to accomplish this goal. Background:
Predicted measurement is yOk D Ck xO k C Dk uk
Prediction covariance (uncertainty) matrix is y;k
Q DC k xQ ;k CkT C Q v
The innovation is yQk D yk yOk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 1 of 7
3.3.5: Can we automatically detect bad measurements with a Kalman filter?
Measurement validation gating
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Can place “measurement validation gate” on measurement
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
using normalized estimation error squared (NEES)
ek2 D yQkT y;k
Q yQk
1
NEES ek2 has Chi-squared distribution with m degrees of freedom, where yk 2R m
If ek2 is outside of bounding value for Chi-squared distribution for a desired
confidence level, then measurement is discarded
Note: If a many measurements are discarded in short time interval, sensor may
truly have failed, or state estimate and covariance may have gotten “lost”
Is sometimes helpful to “bump up” covariance Q ;k , which simulates additional
x
process noise, to help Kalman filter to reacquire
Both done in practice to aid robustness of a real implementation
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 2 of 7
3.3.5: Can we automatically detect bad measurements with a Kalman filter?
NEES is chi-squared
key off: store data
key on: initialize
To prove NEES is chi-squared, define ´k D Mk yQk
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Mean of ´k is E´k D EMk yQk D 0
loop once each measurement interval while pack is active
Covariance of ´k is Q́ ;k D EMk yQk yQkT MkT D Mk y;k
Q MkT
´k is Gaussian (since it is a linear combination of Gaussians)
If we define Mk such that MkT Mk D y;k 1
Q , then
Mk is the lower-triangular Cholesky factor of y;k
Q
1
Also, ´k N .0; I / since Q́ D M M M M D M M M M D I ;k k
T
k k
1 T
k k k
1
k
T T
k
NEES e D ´ ´ D yQ Q yQ is the sum of squares of independent N .0; 1/ RVs
2
k
T
k k
T
k
1
y;k k
So, e is chi-square with m degrees of freedom, where m is the dimension of yQ
2
k k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 3 of 7
3.3.5: Can we automatically detect bad measurements with a Kalman filter?
What does this really mean?
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
ek2 never negative (sum of squares); pdf also asymmetric
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
pdf of chi-square RV X having m degrees of freedom is
1
fX .x/ D x .m=2 e
✜
1/ m=2 ✛ ❞✐str✢❜✉✣✦♦♥ ✇✧★❤ ✩ ✪❡❣✫✬✭✮ ✯✰ ✱✲✶✹✺✻♠
2 m=2
.m=2/ ✓✔✕✖
✏✑✒
Tricky, but don’t need to evaluate in real time ✴
①
✳ ☞✌✍✎ ✥✗✘✙
Instead, use value precomputed from pdf ❢
❋
❉ ✠✡☛
For 1 confidence of a valid measurement, P
✤
❯
✷
need to find 2U such that there is area above
✆✝✞✟
2U (figure drawn for D 0:05)
✵
✸ ✼ ✁✂✄☎
Discard measurement if NEES greater than 2U ✚
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 4 of 7
3.3.5: Can we automatically detect bad measurements with a Kalman filter?
Computer calculation of U
2
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
In MATLAB (Statistics and Machine Learning Toolbox) can
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
find 2U where inverse CDF is equal to 1
X2U = chi2inv (1 -0.01 ,2) % Upper critical value X2U = 9.2103
Function “chi2inv” is built in to Octave
Note that 2U needs to be computed once only, offline
Based only on m and , so doesn’t need to be recalculated as KF runs
For hand calculations a 2 -table is available on next page
If ek2 > 2U , then measurement is discarded (Lk D 0); else, measurement kept
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 5 of 7
3.3.5: Can we automatically detect bad measurements with a Kalman filter?
Manual table-lookup of U
2
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
For chi-squared distribution with m degrees of freedom, table
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
entries list values of 2U . ; m/ for specified upper tail area
Degrees of Upper tail areas
freedom m 0.25 0.10 0.05 0.025 0.01 0.005
1 1.323 2.706 3.841 5.024 6.635 7.879
2 2.773 4.605 5.991 7.378 9.210 10.597
3 4.108 6.251 7.815 9.348 11.345 12.838
4 5.385 7.779 9.488 11.143 13.277 14.860
5 6.626 9.236 11.070 12.833 15.086 16.750
6 7.841 10.645 12.592 14.449 16.812 18.548
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 6 of 7
3.3.5: Can we automatically detect bad measurements with a Kalman filter?
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF has built-in mechanism that enables detecting
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
voltage-sensor errors
Once only, off-line, precompute 2U . ; m/ for yk 2 Rm and desired
As KF executes, every time sample, compute ek2 D yQkT y;k
Q yQk
1
If ek2 > 2U . ; m/, then discard measurement (set Lk D 0)
Otherwise, apply measurement update as usual
If many sequential measurements discarded, consider “bumping up” covariance as
C C
Q ;k D QxQ ;k where Q > 1
x
If problems persist, likely a permanent sensor fault
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 7 of 7
3.3.6: How do I initialize and tune a Kalman filter?
Real-world issue: Initialization
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
If battery load is “off” for a “long” time, just assume that cell
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
voltage is equivalent to OCV:
Reset SOC estimate based on OCV
Set diffusion voltages to zero
Keep prior value of hysteresis state
If load has been off for a “short” period of time
Set up and execute simple time/measurement update (simple KF) equations for
SOC and diffusion voltages
Hysteresis voltages do not change
Run a single-step Kalman filter to update state estimate based on total time off
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 1 of 7
3.3.6: How do I initialize and tune a Kalman filter?
Real-world issue: Tuning x;0 Q
key off: store data
key on: initialize
meas. voltage estimate estimate compute
KF “tuning” accomplished via changing xQ ;0 , e
balance
w , vQ
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Ideally, xQ ;0 D E.x0 xO 0C /.x0 xO 0C /T
loop once each measurement interval while pack is active
Since we don’t know x0 exactly, not possible to initialize xO 0C
to exactly correct value: Uncertainty in xO 0C captured by xQ ;0
If we assume model states uncorrelated, xQ ;0 is diagonal
We often make this assumption simply because we have
no better information regarding state correlation
Diagonal elements in xQ ;0 are variances of initial
estimation errors of the corresponding state elements in x
Set diagonal elements of xQ ;0 such that you expect true state
p
to lie between xO 0C 3 diag.xQ ;0 /
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 2 of 7
3.3.6: How do I initialize and tune a Kalman filter?
Real-world issue: Tuning ew
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Assuming model of cell is perfect, e
current state of state of power
w is covariance matrix of
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
“process noise” wk driving cell
C D Axk C B uk C wk
xk 1
Process noise is any unmeasured (zero-mean, white) input that affects state vector
If wk chosen to model current-sensor noise, then e
w can be determined
statistically via experimentation
But, as no model is perfect, w also attempts to capture—in some way—state-
e
equation inaccuracies, so should have larger uncertainty than simply representing
current-sensor noise alone
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 3 of 7
3.3.6: How do I initialize and tune a Kalman filter?
Real-world issue: Tuning v Q
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Assuming model of cell is perfect, vQ is the covariance matrix
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
of “measurement noise” vk
yk D C xk C Duk C vk
Measurement noise is any unmeasured (zero-mean, white) input that doesn’t affect
state vector, but which does corrupt measurements
If vk chosen to model voltage-sensor noise, then vQ can be determined statistically
via experimentation (or data sheet)
But, as no model is perfect, vQ also attempts to capture—in some way—output-
equation inaccuracies, so should have larger uncertainty than simply representing
voltage-sensor noise alone
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 4 of 7
3.3.6: How do I initialize and tune a Kalman filter?
Rate of convergence
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Filter convergence rates are determined by e
current state of state of power
w and vQ
temperature charge (SOC) health (SOH) cells limits
Large e
loop once each measurement interval while pack is active
w says “trust sensor more than model” and makes
state error bounds large
Large vQ says “trust model more than sensor” and converges slowly (pseudo
open loop)
Since model inaccuracies are difficult to quantify, some trial-and-error “tuning” of
e
w and vQ is common
In some cases (noisy sensors or bad model), desired state-estimate accuracy and
convergence rates are simply impossible
In general, it is not possible to have arbitrarily fast convergence to arbitrarily narrow
error bounds
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 5 of 7
3.3.6: How do I initialize and tune a Kalman filter?
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
You have learned how to set xQ ;0 during initialization
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
You have seen tips on tuning e
w , which represents process-
noise (e.g., current-sensor noise and inaccuracies in state equation) uncertainty
You have seen some tips on tuning vQ , which represents sensor-noise (e.g.,
voltage-sensor noise and inaccuracies in voltage equation) uncertainty
Since model inaccuracies are difficult to quantify, some trial-and-error “tuning” of
e
w and vQ is common
e
Since convergence rates depend on w and vQ , it is not possible in general to
have arbitrarily fast convergence to arbitrarily narrow error bounds
Still, KF is optimal MMSE estimator for the assumptions made during derivation
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 6 of 7
3.3.6: How do I initialize and tune a Kalman filter?
Credits
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Credits for photos in this lesson
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Tuning fork on slide 2: By Helihark (own work) [CC BY-SA 3.0
(http://creativecommons.org/licenses/by-sa/3.0)], via Wikimedia Commons,
https://commons.wikimedia.org/wiki/File:Tuning-fork.jpg
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 7 of 7
3.3.7: Where from here?
Summary of this week
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
This week, you learned:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
How to apply linear KF to linearized battery-cell model to
loop once each measurement interval while pack is active
help visualize the steps involved
How to simulate random systems as preparation for evaluating KF via simulation
How to implement Kalman filter in Octave/MATLAB and evaluate results
computed by the filter
Some “improvements” to KF to handle real-world numeric round-off issues
How KF also provides helpful ways to detect sensor faults
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 1 of 3
3.3.7: Where from here?
Where from here?
key off: store data
key on: initialize
How to generalize state-state model to describe nonlinear
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
systems
loop once each measurement interval while pack is active
How to apply two assumptions to derive the extended
Kalman filter from the generic Gaussian
sequential-probabilistic-inference solution
How to write Octave/MATLAB code to implement EKF for a
simple example and for the battery-cell state-estimation
problem
See some examples of applying EKF to estimate cell SOC
using lab-test data
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 2 of 3
3.3.7: Where from here?
Credits
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Credits for photos in this lesson
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
“Map” on slide 2: Pixabay license
(https://pixabay.com/en/service/license/), from
https://pixabay.com/en/map-treasure-map-treasure-pirates-162047/
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Coming to understand the linear Kalman filter j 3 of 3
3.4.1: Introducing nonlinear variations to Kalman filters
Models for nonlinear Kalman filters
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
KF assumes a cell model of the form
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
xk D Ak 1 xk 1 C Bk 1 uk 1 C wk 1
yk D Ck x k C Dk uk C vk
But, cell models are nonlinear, so the standard KF recursion doesn’t apply directly
We now generalize to the nonlinear case, with system dynamics described as
xk D f .xk 1 ; uk 1 ; wk 1 /
yk D h.xk ; uk ; vk /
Functions f ./ and h./ may be time-varying, but we generally omit the time
dependency from the notation for ease of understanding
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 5
3.4.1: Introducing nonlinear variations to Kalman filters
Kinds of nonlinear Kalman filters
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Three basic KF generalizations for nonlinear systems
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Extended Kalman filter (EKF): Analytic linearization of model
at each point in time: problematic, but still popular
Sigma-point (Unscented) Kalman filter (SPKF/UKF): Statistical/empirical
linearization of model at each point in time: can be much better than EKF at
same computational complexity
Particle filters: Most precise, but often thousands of times more computations
required than either EKF/SPKF
This week, we study EKF; next week SPKF
Particle filters are beyond the scope of this course
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 5
3.4.1: Introducing nonlinear variations to Kalman filters
Prediction steps of nonlinear Kalman filters
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF and SPKF both follow same set of general steps as KF
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
Recall general step 1
Step 1a: State prediction time update
Prediction
Step 1b: Error covariance time update
Step 1c: Predict system output
Step 2a: Estimator gain matrix
EKF and SPKF simply have different expressions from KF for evaluating the
Correction
expectation
Stepoperations (because
2b: State estimate of different system
measurement update model)
Dr. Gregory L. Plett j
Step 2c: Error covariance measurement
University of Colorado Colorado Springs
update
Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 5
Step 1a: State prediction time update
3.4.1: Introducing nonlinear variations to Kalman filters
Prediction
Step 1b: Error covariance time update
Update steps of nonlinear Kalman filters
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF andStep
current state of state of power
SPKF
1c: both follow same set of general steps as KF temperature charge (SOC) health (SOH) cells limits
Predict system output loop once each measurement interval while pack is active
Recall general step 2
Step 2a: Estimator gain matrix
Correction
Step 2b: State estimate measurement update
Step 2c: Error covariance measurement update
Again, EKF and SPKF simply have different expressions from KF for evaluating the
expectation operations implied in Step 2a (because of different system model)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 4 of 5
3.4.1: Introducing nonlinear variations to Kalman filters
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Battery cells are nonlinear systems, therefore need nonlinear
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
models and nonlinear Kalman filters
We will study two different approaches to generalizing KF equations for nonlinear
systems, one leading to EKF and the other to SPKF/UKF1
EKF and SPKF (and variants) still obey predict/correct sequence of six steps
So, derivation will involve evaluating those six steps under different sets of
assumptions
1
cf. http://www.ieeeghn.org/wiki/index.php/First-Hand:The_Unscented_Transform
re. name “unscented”
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 5 of 5
3.4.2: Deriving the three extended-Kalman-filter prediction steps
The extended Kalman filter (EKF)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF makes two simplifying assumptions when adapting
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
general sequential inference equations to a nonlinear system:
When computing estimates of the output of a nonlinear function, EKF assumes
Efn.x/ fn.Ex/, which is not true in general
When computing covariance estimates, EKF uses Taylor-series expansion to
linearize the system equations around the present operating point
In this lesson, you will learn how to apply these approximations and assumptions to
derive the EKF prediction steps
In the next lesson, you will learn how to derive the EKF update steps
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 6
3.4.2: Deriving the three extended-Kalman-filter prediction steps
EKF step 1a: State prediction time update
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
The state prediction step is approximated as
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
O D Ef .x ; u ; w
xk k 1 k 1 k 1/ jY k 1
f .xO C ; u ; wN
k 1 k 1 k 1/ ,
where wN k 1 D Ewk 1 k 1 N D 0.)
. (Often, w
That is, we approximate the expected value of the new state by assuming that it is
reasonable to simply propagate xO kC 1 and wN k 1 through the state equation.
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 6
3.4.2: Deriving the three extended-Kalman-filter prediction steps
EKF step 1b: Error covariance time update (1)
key off: store data
key on: initialize
We start with an expression for xQ k :
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Q Dx O D f .x OC N
loop once each measurement interval while pack is active
xk k xk k 1; uk 1; wk 1/ f .xk 1
; uk 1; wk 1/
Approx. first term via Taylor series around prior operating point fxO kC 1 ; uk 1; wkN 1 g
xk f .xO C k 1
; uk 1; N
wk 1/ C df .x k 1;
dxk
uk 1; wk 1/
C
.xk 1 OC
xk 1
/
1 xk 1 DxO k 1
Defined as AOk 1
C df .x k 1;
dwk
uk 1; wk 1/
.wk 1 N
wk 1/
1
wk 1 DwN k 1
Defined as BO k 1
This gives xQ k AO k 1 xk Q C C BO
1 k 1w
ek 1
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 6
3.4.2: Deriving the three extended-Kalman-filter prediction steps
EKF step 1b: Error covariance time update (2)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Substituting this to find the prediction-error covariance:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Q k /.xQ k /T AOk 1CxQ ;k 1AOTk 1 C BOk 1ew BOkT
xQ ;k D E.x
loop once each measurement interval while pack is active
1
Note, by the chain rule of total differentials,
df .xk 1; uk 1; wk 1/ D @f .x k 1;
@xk
uk 1; wk 1/
dxk 1 C @f .x k 1;
@uk
uk 1; wk 1/
duk 1
1 1
C @f .x k 1; uk
@wk
1; wk 1/
dwk 1
1 0
df .xk 1; uk 1; wk 1/ @f .xk 1; uk 1; wk 1/ @f .xk 1; uk 1; wk 1/ duk
dxk
D @xk
C @uk dxk
1
1 1 1 1
@f .xk 1; uk 1; wk 1/ dwk @f .xk 1; uk 1; wk 1/
+
@wk dxk
1
D @xk
1 1 1
0
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 4 of 6
3.4.2: Deriving the three extended-Kalman-filter prediction steps
EKF step 1c: Output estimate
key off: store data
key on: initialize
Similarly, df .xk d1w;uk 1 ;wk 1 / D @f .xk 1@w
meas. voltage estimate estimate balance compute
current state of state of power
;uk 1 ;wk 1 / temperature charge (SOC) health (SOH) cells limits
k 1 k 1 loop once each measurement interval while pack is active
Distinction between total and partial differential not yet critical
Will be in fourth course in specialization
System output is estimated to be
O D Eh.x ; u ; v / j Y
yk k k k k 1
h.xO ; u ; vN /,
k k k
where vN k D Ev k
That is, it is assumed that propagating xO k and the mean sensor noise is the best
approximation to estimating the output
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 5 of 6
3.4.2: Deriving the three extended-Kalman-filter prediction steps
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF makes two fundamental assumptions to generalize KF
current state of state of power
temperature charge (SOC) health (SOH) cells limits
1. Assume Efn.x/ fn.Ex/, which is not true in general
loop once each measurement interval while pack is active
2. Assume Taylor-series expansion to linearize system equations for covariances
Applying these to generic-probabilistic-inference solution yields first three EKF
steps (final three in next lesson)
In general must be careful to distinguish between total and partial differentials
(although not critical here)
First half of EKF is now derived
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 6 of 6
3.4.3: Deriving the three extended-Kalman-filter correction steps
EKF step 2a: Estimator gain matrix (1)
key off: store data
key on: initialize
Recall that the KF gain vector is Lk D xQ y;k
meas. voltage estimate estimate compute
1
balance
current state of state of power
Q y;k
Q
temperature charge (SOC) health (SOH) cells limits
Output prediction error may be approximated
loop once each measurement interval while pack is active
Q D yk
yk O D h.xk ; uk ; vk /
yk O
h.xk ; uk ; vk / N
using Taylor-series expansion on first term
yk h.xO k ; uk ; vNk / C dh.xkd;xuk ; vk / .xk O C dh.xkd;vuk ; vk /
xk / .vk N
vk /
k xk DxO k k vk DNvk
Defined as CO k Ok
Defined as D
Note, much like we saw in Step 1b,
dh.xk ; uk ; vk / dh.xk ; uk ; vk /
dxk
D @h.xk@x; uk ; vk / and
dvk
D @h.xk@v; uk ; vk /
k k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 4
3.4.3: Deriving the three extended-Kalman-filter correction steps
EKF step 2a: Estimator gain matrix (2)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
From these results, we can compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Q CO k xQ ;k CO k C DO k vQ DO k ;
loop once each measurement interval while pack is active
T T
y;k
Q E.xQ k /.CO k xQ k C DO k vQ k /
T
xQ y;k
D xQ ;k COkT
These terms may be combined to get the Kalman gain
D xQ ;k COkT COk xQ ;k COkT C DO k vQ DO kT
1
Lk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 4
3.4.3: Deriving the three extended-Kalman-filter correction steps
EKF step 2b/c: State/covariance meas. update
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF step 2b: State estimate measurement update
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Computes posterior state estimate by updating prediction loop once each measurement interval while pack is active
using estimator gain and innovation yk yOk
O C D xO k C Lk .yk
xk O
yk /
EKF step 2c: Error covariance measurement update
Finally, the updated covariance is computed as
C
xQ ;k D xQ ;k
T
Q Lk
Lk y;k
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 4
3.4.3: Deriving the three extended-Kalman-filter correction steps
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF makes two fundamental assumptions to generalize KF
current state of state of power
temperature charge (SOC) health (SOH) cells limits
1. Assume Efn.x/ fn.Ex/, which is not true in general
loop once each measurement interval while pack is active
2. Assume Taylor-series expansion to linearize system equations for covariances
Applying these to generic-probabilistic-inference solution yields final three EKF
steps
EKF algorithm is now derived!
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 4 of 4
3.4.4: Introducing a simple EKF example, with Octave code
An EKF example
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
We will look at two examples of implementing EKF
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
1. A simple example, with fairly straightforward math
2. The battery-cell example
In this lesson, we implement EKF for model having following dynamics:
p
xk C1 D f .xk ; uk ; wk / D 5 C xk C wk
yk D h.xk ; uk ; vk / D xk C vk
3
with e
w D 1 and vQ D 2
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 9
3.4.4: Introducing a simple EKF example, with Octave code
Computing the derivatives
key off: store data
To implement EKF, we must determine AOk , BO k , CO k , and DO k
key on: initialize
meas. voltage estimate estimate balance compute
current state of state of power
cells
p5 C x C w
temperature charge (SOC) health (SOH) limits
loop once each measurement interval while pack is active
AOk D D D
@f .xk ; uk ; wk / @ k k 1
q
xk DxO kC 2 5CxO kC
@xk @xk
xk DxO kC
p5 C x C w
BOk D @f .xk ; uk ; wk /
D @ k k
D1
@w k @w wk DwN k k
wk DwN k
O D @h.xk ; uk ; vk /
Ck D @ xk3 vk C
D 3.xO k /2
@x k xk DxO k @xk
xk DxO k
Dk O D @h.xk ; uk ; vk / D @ xk3 C
vk
D1
@v k vk DNvk @vk
vk DNvk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 9
3.4.4: Introducing a simple EKF example, with Octave code
EKF initialization code
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Code to implement EKF starts below
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Define simulation constants; reserve storage
loop once each measurement interval while pack is active
% Initialize simulation variables
SigmaW = 1; % Process noise covariance
SigmaV = 2; % Sensor noise covariance
maxIter = 40;
xtrue = 2 + randn (1) ; % Initialize true system initial state
xhat = 2; % Initialize Kalman filter initial estimate
SigmaX = 1; % Initialize Kalman filter covariance
u = 0; % Unknown initial driving input : assume zero
% Reserve storage for variables we might want to plot / evaluate
xstore = zeros ( maxIter +1 , length ( xtrue ) ) ; xstore (1 ,:) = xtrue ;
xhatstore = zeros ( maxIter , length ( xhat ) ) ;
SigmaXstore = zeros ( maxIter , length ( xhat ) ^2) ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 9
3.4.4: Introducing a simple EKF example, with Octave code
EKF steps 1a through 1b
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Main EKF loop starts below
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Also co-simulating system dynamics for sensor inputs
loop once each measurement interval while pack is active
for k = 1: maxIter ,
% EKF Step 1 a : State estimate time update
% ( First compute Ahat , Bhat : Specifics depend on model !)
% Note : For this example , x ( k +1) = sqrt (5+ x ( k ) ) + w ( k )
% Note : You need to insert your system ' s f (...) equation here
Ahat = 0.5/ sqrt (5+ xhat ) ; Bhat = 1;
xhat = sqrt (5+ xhat ) ;
% EKF Step 1 b : Error covariance time update
SigmaX = Ahat * SigmaX * Ahat ' + Bhat * SigmaW * Bhat ';
% [ Co - simulate system , with input signal u , and output signal y ]
w = chol ( SigmaW ) '* randn (1) ;
v = chol ( SigmaV ) '* randn (1) ;
ytrue = xtrue ^3 + v ; % y is based on present x and u
xtrue = sqrt (5+ xtrue ) + w ; % future x is based on present u
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 4 of 9
3.4.4: Introducing a simple EKF example, with Octave code
EKF steps 1c through 2b
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Main EKF loop continues below
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Notice the “extra” robustness code at end
loop once each measurement interval while pack is active
% EKF Step 1 c : Estimate system output
% ( First compute Ahat , Bhat : Specifics depend on model !)
% Note : For this example , y ( k ) = x ( k ) ^3
% Note : You need to insert your system ' s h (...) equation here
Chat = 3* xhat ^2; Dhat = 1;
yhat = xhat ^3;
% EKF Step 2 a : Compute Kalman gain matrix
SigmaY = Chat * SigmaX * Chat ' + Dhat * SigmaV * Dhat ';
L = SigmaX * Chat '/ SigmaY ;
% EKF Step 2 b : State estimate measurement update
xhat = xhat + L *( ytrue - yhat ) ;
xhat = max ( -5 , xhat ) ; % don ' t get square root of negative xhat !
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 5 of 9
3.4.4: Introducing a simple EKF example, with Octave code
EKF step 2c
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Main EKF loop concludes below
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Includes code to force SigmaX to be PSD
loop once each measurement interval while pack is active
% EKF Step 2 c : Error covariance measurement update
SigmaX = SigmaX - L * SigmaY *L ';
[~ ,S , V ] = svd ( SigmaX ) ;
HH = V * S *V ';
SigmaX = ( SigmaX + SigmaX ' + HH + HH ') /4; % Help to keep robust
% [ Store information for evaluation / plotting purposes ]
xstore ( k +1 ,:) = xtrue ; xhatstore (k ,:) = xhat ;
SigmaXstore (k ,:) = ( SigmaX (:) ) ';
end ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 6 of 9
3.4.4: Introducing a simple EKF example, with Octave code
EKF plotting code
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
This is an example showing how to plot the results from this
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
EKF code in two different ways
subplot (1 ,2 ,1) ;
t = 0: maxIter -1;
plot (t , xstore (1: maxIter ) , 'k - ' ,t , xhatstore , 'b - - ' , ...
t , xhatstore +3* sqrt ( SigmaXstore ) , 'm -. ' ,...
t , xhatstore -3* sqrt ( SigmaXstore ) , 'm -. ') ; grid ;
legend ( ' true ' , ' estimate ' , ' bounds ') ;
xlabel ( ' Iteration ') ; ylabel ( ' State ') ;
title ( ' Extended Kalman filter in action ') ;
subplot (1 ,2 ,2) ;
plot (t , xstore (1: maxIter ) - xhatstore , 'b - ' ,t , ...
3* sqrt ( SigmaXstore ) , 'm - - ' ,t , -3* sqrt ( SigmaXstore ) , 'm - - ') ;
grid ; legend ( ' Error ' , ' bounds ' ,0) ;
title ( ' EKF Error with bounds ') ;
xlabel ( ' Iteration ') ; ylabel ( ' Estimation error ') ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 7 of 9
3.4.4: Introducing a simple EKF example, with Octave code
Representative results
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Figures below show representative results
current state of state of power
temperature charge (SOC) health (SOH) cells limits
EKF works well for small states, when system is fairly linear
loop once each measurement interval while pack is active
EKF struggles for larger states, when system is more nonlinear
Extended Kalman filter in action EKF error with bounds
7 0.5
Truth
6 EKF estimate
EKF bounds 0
5
Estimation error
4 −0.5
State
3
−1
2
−1.5
1 EKF error
EKF bounds
0 −2
0 10 20 30 40 0 10 20 30 40
Iteration Iteration
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 8 of 9
3.4.4: Introducing a simple EKF example, with Octave code
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Have now seen code to implement EKF on relatively simple
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
nonlinear state-space model
Finding derivatives was most difficult part to do correctly (but, not too bad for this
simple model)
Actual code was straightforward implementation of steps seen earlier this week
Results show that EKF provides good estimates and bounds only for operating
regimes where the model is nearly linear (as expected)
Estimates and bounds are poorer when the model is being operated far away from
a linear region
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 9 of 9
3.4.5: Preparing to implement EKF on an ECM
Derivatives for SOC state
key off: store data
key on: initialize
To implement EKF using ESC model, need AOk , BO k , CO k , D
Ok
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
We first examine components of state equation to find AOk , BO k
loop once each measurement interval while pack is active
Model true cell current as ik C wk , but measure ik only (wk is current-sensor error)
Assume k D 1, leverage adaptivity of EKF overcome this inaccuracy
Then, state-of-charge equation can be written
t
Q
.ik C wk / ´ k C1 D ´k
The two derivatives that we need for this term are (with Q having units [A s]):
@´k C1 @´k C1 t
@´k C
D 1, and
@wk
D Q
´k DÓ k wk DwN
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 8
3.4.5: Preparing to implement EKF on an ECM
Derivatives for R–C states
key off: store data
key on: initialize
D exp.
meas. voltage estimate estimate balance compute
If j
current state of state of power
t=.Rj Cj //, then resistor-currents state temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
equation can be written as
2 3 2 3
1 0 1 1
iR;k C1 D6
4 0 2 7
5 iR;k C6
4 1 2 7
5 .ik C wk /
:: ::: ::
: :
ARC BRC
The two derivatives can be found to be
@iR;k C1 @iR;k C1
D ARC , and D BRC
@iR;k iR;k DiOR;k
C @wk wk DwN
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 8
3.4.5: Preparing to implement EKF on an ECM
Derivatives for dynamic hysteresis state (1)
key off: store data
key on: initialize
.ik Cwk / t
D exp
meas. voltage estimate estimate balance compute
If AH;k
current state of state of power
, then hysteresis state temperature charge (SOC) health (SOH) cells limits
Q loop once each measurement interval while pack is active
hk C1 D AH;k hk C .AH;k 1/sgn .ik C wk /
Taking partial with respect to the state and evaluating at the setpoint (noting that
wk D wN is a member of the setpoint),
.ik C wN k / t
@hk C1
@hk hkDhO C
D exp Q
D ANH;k
wk DwNk
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 8
3.4.5: Preparing to implement EKF on an ECM
Derivatives for dynamic hysteresis state (2)
key off: store data
key on: initialize
Problem: @hk C1 =@wk does not exist at ik C wk D0
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
If we assume that ik C wk > 0,
loop once each measurement interval while pack is active
@hk C1 t t
@wk
D Q
exp
Q
j.ik C wk /j .1 C hk /
If we assume that ik C wk < 0,
@hk C1 t t
@wk
D Q
exp
Q
j.ik C wk /j .1 hk /
Overall, evaluating at Taylor-series linearization setpoint,
@hk C1 t N
D AH;k 1 C sgn.ik C w/
N hO C
@wk hk h D OC Q k
wk w
k
DN
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 4 of 8
3.4.5: Preparing to implement EKF on an ECM
Derivatives for instantaneous hysteresis state (2)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
The zero-state hysteresis equation is defined as
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
(
sgn.ik C wk /; jik C wk j > 0;
sk C1 D
sk ; else
If we consider ik C wk D 0 to be a zero-probability event, then
@sk C1 @sk C1
@sk
D 0, and
@wk
D0
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 5 of 8
3.4.5: Preparing to implement EKF on an ECM
Derivatives for output equation
key off: store data
key on: initialize
We now look at the components that determine CO k and DO k
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
The ESC-model output equation is
X
yk D OCV.´k / C M0sk C M hk Rj iRj ;k R0 ik C vk
j
no longer considering ik to have wk noise added to it (this would add correlation
between process noise and sensor noise).
We have
@yk @yk @yk @yk
@sk
D M0 , @hk
D M, @iRj ;k
D Rj ,
@vk
D1
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 6 of 8
3.4.5: Preparing to implement EKF on an ECM
Derivative of OCV function
key off: store data
key on: initialize
meas. voltage estimate estimate compute
Can use (evenly spaced) OCV data to approximate
balance
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
@yk @OCV.´k /
@´k
D @´k
5
Empirical OCV derivative for six cells
´k DÓ k ´k DÓ k 4
∂ OCV/ ∂ SOC (V)
The figure shows empirical OCV derivative 3
relationships for six different lithium-ion cells 2
There is a little noise, which could be filtered (with a 1
zero-phase filter!) 0
0 10 20 30 40 50 60 70 80 90 100
Filtering not really necessary State of charge (%)
% Find dOCV / dz at SOC = z from { SOC , OCV } data
function dOCVz = dOCVfromSOC ( SOC , OCV , z )
dZ = SOC (2) - SOC (1) ; % Find spacing of SOC vector
dUdZ = diff ( OCV ) / dZ ; % Scaled forward finite difference
dOCV = ([ dUdZ (1) dUdZ ] + [ dUdZ dUdZ ( end ) ]) /2; % Avg of fwd / bkwd diffs
dOCVz = interp1 ( SOC , dOCV , z ) ; % Could make more efficient than this ...
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 7 of 8
3.4.5: Preparing to implement EKF on an ECM
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
To implement EKF on ESC model, need noise assumptions
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Assumed process noise is additive to electrical current
loop once each measurement interval while pack is active
Assumed sensor noise is additive to voltage measurement
Then, must find all partial derivates for AOk , BO k , CO k , DO k
This was a little complicated for the hysteresis state, since the update equation is
not differentiable at all points
However, by a limiting argument, we found all matrices we need
We are ready to implement EKF on ESC model!
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 8 of 8
3.4.6: Introducing Octave code to initialize and control EKF for SOC estimation
Implementing EKF on ESC model
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Refactor EKF code: represent real BMS implementation better
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Initialization routine (initEKF.m), called once at startup
loop once each measurement interval while pack is active
Update routine (iterEKF.m), called every sample interval
“Wrapper” code, coordinates the entire simulation process
Wrapper code begins (load data, store in local variables):
% Load cell model , cell - test data ( incl . variable " DYNData " of which the field
% " script1 " is of interest ) . It has sub - fields time , current , voltage , soc .
load CellModel % loads " model " of cell
load ( ' Cell_DYN_P25 ') ; % loads data from cell test
T = 25; % Test temperature
time = DYNData . script1 . time (:) ; deltat = time (2) - time (1) ;
time = time - time (1) ; % start time at 0
current = DYNData . script1 . current (:) ; % discharge > 0; charge < 0.
voltage = DYNData . script1 . voltage (:) ;
soc = DYNData . script1 . soc (:) ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 6
3.4.6: Introducing Octave code to initialize and control EKF for SOC estimation
Wrapper code (2)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Wrapper code continues by reserving storage, initializing
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
covariances, calling EKF initialization routine
% Reserve storage for computed results , for plotting
sochat = zeros ( size ( soc ) ) ;
socbound = zeros ( size ( soc ) ) ;
% Covariance values
SigmaX0 = diag ([1 e -3 1e -3 1e -2]) ; % uncertainty of initial state
SigmaV = 2e -1; % uncertainty of voltage sensor , output equation
SigmaW = 1 e1 ; % uncertainty of current sensor , state equation
% Create ekfData structure and initialize variables using first
% voltage measurement and first temperature measurement
ekfData = initEKF ( voltage (1) ,T , SigmaX0 , SigmaV , SigmaW , model ) ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 6
3.4.6: Introducing Octave code to initialize and control EKF for SOC estimation
Wrapper code (3)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Wrapper code continues by entering main simulation loop
current state of state of power
temperature charge (SOC) health (SOH) cells limits
“Measure” sensor readings; update EKF
loop once each measurement interval while pack is active
% Now , enter loop for remainder of time , where we update the EKF
% once per sample interval
hwait = waitbar (0 , ' Computing ... ') ;
for k = 1: length ( voltage ) ,
vk = voltage ( k ) ; % " measure " voltage
ik = current ( k ) ; % " measure " current
Tk = T ; % " measure " temperature
% Update SOC ( and other model states )
[ sochat ( k ) , socbound ( k ) , ekfData ] = iterEKF ( vk , ik , Tk , deltat , ekfData ) ;
% update waitbar periodically , but not too often ( slow procedure )
if mod (k ,1000) ==0 , waitbar ( k / length ( current ) , hwait ) ; end ;
end
close ( hwait ) ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 6
3.4.6: Introducing Octave code to initialize and control EKF for SOC estimation
Wrapper code (4)
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Wrapper code continues with plotting/analysis code
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
figure ; plot ( time /60 ,100* sochat , time /60 ,100* soc ) ; hold on
plot ([ time /60; NaN ; time /60] ,...
[100*( sochat + socbound ) ; NaN ; 100*( sochat - socbound ) ]) ;
title ( ' SOC estimation using EKF ') ; xlabel ( ' Time ( min ) ') ; ylabel ( ' SOC (%) ') ;
legend ( ' Estimate ' , ' Truth ' , ' Bounds ') ; grid on
fprintf ( ' RMS SOC estimation error = % g %%\ n ' , sqrt ( mean ((100*( soc - sochat ) ) .^2) ) ) ;
figure ; plot ( time /60 ,100*( soc - sochat ) ) ; hold on
plot ([ time /60; NaN ; time /60] ,[100* socbound ; NaN ; -100* socbound ]) ;
title ( ' SOC estimation errors using EKF ') ;
xlabel ( ' Time ( min ) ') ; ylabel ( ' SOC error (%) ') ; ylim ([ -4 4]) ;
legend ( ' Estimation error ' , ' Bounds ') ; grid on
ind = find ( abs ( soc - sochat ) > socbound ) ;
fprintf ( ' Percent of time error outside bounds = % g %%\ n ' ,...
length ( ind ) / length ( soc ) *100) ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 4 of 6
3.4.6: Introducing Octave code to initialize and control EKF for SOC estimation
EKF initialization code
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
The EKF initialization code is:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
function ekfData = initEKF ( v0 , T0 , SigmaX0 , SigmaV , SigmaW , model )
% Initial state description
ir0 = 0; ekfData . irInd = 1;
hk0 = 0; ekfData . hkInd = 2;
SOC0 = SOCfromOCVtem p ( v0 , T0 , model ) ; ekfData . zkInd = 3;
ekfData . xhat = [ ir0 hk0 SOC0 ] '; % initial state
% Covariance values
ekfData . SigmaX = SigmaX0 ; ekfData . SigmaV = SigmaV ;
ekfData . SigmaW = SigmaW ; ekfData . Qbump = 5;
% previous value of current ( and its sign )
ekfData . priorI = 0; ekfData . signIk = 0;
% store model data structure too
ekfData . model = model ;
end
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 5 of 6
3.4.6: Introducing Octave code to initialize and control EKF for SOC estimation
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Implementation of EKF on ESC model refactors code
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Initialization routine (initEKF.m), called once at startup
loop once each measurement interval while pack is active
Update routine (iterEKF.m), called every sample interval
“Wrapper” code, coordinates the entire simulation process
You have now seen the details of the initialization routine and the wrapper code
Next lesson will present the update routine
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 6 of 6
3.4.7: Introducing Octave code to update EKF for SOC estimation
EKF iteration code, load model
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Implementation of EKF on ESC model refactors code
current state of state of power
temperature charge (SOC) health (SOH) cells limits
“Wrapper” code, coordinates the entire simulation process
loop once each measurement interval while pack is active
Initialization routine (initEKF.m), called once at startup
Update routine (iterEKF.m), called every sample interval, which starts with:
function [ zk , zkbnd , ekfData ] = iterEKF ( vk , ik , Tk , deltat , ekfData )
model = ekfData . model ;
% Load the cell model parameters
Q = getParamESC ( ' QParam ' ,Tk , model ) ;
G = getParamESC ( ' GParam ' ,Tk , model ) ;
M = getParamESC ( ' MParam ' ,Tk , model ) ;
M0 = getParamESC ( ' M0Param ' ,Tk , model ) ;
RC = exp ( - deltat ./ abs ( getParamESC ( ' RCParam ' ,Tk , model ) ) ) ';
R = getParamESC ( ' RParam ' ,Tk , model ) ';
R0 = getParamESC ( ' R0Param ' ,Tk , model ) ;
eta = getParamESC ( ' etaParam ' ,Tk , model ) ;
if ik <0 , ik = ik * eta ; end ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 8
3.4.7: Introducing Octave code to update EKF for SOC estimation
EKF iteration code, load covariances/states
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF iteration code continues
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Load covariances, states from prior iteration
loop once each measurement interval while pack is active
% Get data stored in ekfData structure
I = ekfData . priorI ;
SigmaX = ekfData . SigmaX ;
SigmaV = ekfData . SigmaV ;
SigmaW = ekfData . SigmaW ;
xhat = ekfData . xhat ;
irInd = ekfData . irInd ;
hkInd = ekfData . hkInd ;
zkInd = ekfData . zkInd ;
if abs ( ik ) >Q /100 , ekfData . signIk = sign ( ik ) ; end ;
signIk = ekfData . signIk ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 8
3.4.7: Introducing Octave code to update EKF for SOC estimation
EKF iteration code, steps 1a–1b
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF iteration code continues
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Compute b bk 1, and implement steps 1a and 1b loop once each measurement interval while pack is active
Ak 1 , B
% Step 1 a : Compute Ahat [k -1] , Bhat [k -1]; Then , state estimate time update
nx = length ( xhat ) ; Ahat = zeros ( nx , nx ) ; Bhat = zeros ( nx ,1) ;
Ahat ( zkInd , zkInd ) = 1; Bhat ( zkInd ) = - deltat /(3600* Q ) ;
Ahat ( irInd , irInd ) = diag ( RC ) ; Bhat ( irInd ) = 1 - RC (:) ;
Ah = exp ( - abs ( I * G * deltat /(3600* Q ) ) ) ; % hysteresis factor
Ahat ( hkInd , hkInd ) = Ah ;
B = [ Bhat , 0* Bhat ];
Bhat ( hkInd ) = - abs ( G * deltat /(3600* Q ) ) * Ah *(1+ sign ( I ) * xhat ( hkInd ) ) ;
B ( hkInd ,2) = Ah -1;
xhat = Ahat * xhat + B *[ I ; sign ( I ) ];
% Step 1 b : Error covariance time update
% sigmaminus ( k ) = Ahat (k -1) * sigmaplus (k -1) * Ahat (k -1) ' + ...
% Bhat (k -1) * sigmawtilde * Bhat (k -1) '
SigmaX = Ahat * SigmaX * Ahat ' + Bhat * SigmaW * Bhat ';
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 8
3.4.7: Introducing Octave code to update EKF for SOC estimation
EKF iteration code, steps 1c–2a
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF iteration code continues
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Output estimate; estimator gain matrix
loop once each measurement interval while pack is active
% Step 1 c : Output estimate
yhat = OCVfromSOCtemp ( xhat ( zkInd ) ,Tk , model ) + M0 * signIk + ...
M * xhat ( hkInd ) - R * xhat ( irInd ) - R0 * ik ;
% Step 2 a : Estimator gain matrix
Chat = zeros (1 , nx ) ;
Chat ( zkInd ) = dOCVfromS OC temp ( xhat ( zkInd ) ,Tk , model ) ;
Chat ( hkInd ) = M ;
Chat ( irInd ) = -R ;
Dhat = 1;
SigmaY = Chat * SigmaX * Chat ' + Dhat * SigmaV * Dhat ';
L = SigmaX * Chat '/ SigmaY ;
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 4 of 8
3.4.7: Introducing Octave code to update EKF for SOC estimation
EKF iteration code, steps 2b–2c
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF iteration code continues
current state of state of power
temperature charge (SOC) health (SOH) cells limits
State and covariance measurement updates
loop once each measurement interval while pack is active
% Step 2 b : State estimate measurement update
r = vk - yhat ; % residual . Use to check for sensor errors ...
if r ^2 > 100* SigmaY , L (:) =0.0; end
xhat = xhat + L * r ;
xhat ( hkInd ) = min (1 , max ( -1 , xhat ( hkInd ) ) ) ; % Help maintain robustness
xhat ( zkInd ) = min (1.05 , max ( -0.05 , xhat ( zkInd ) ) ) ;
% Step 2 c : Error covariance measurement update
SigmaX = SigmaX - L * SigmaY *L ';
if r ^2 > 4* SigmaY , % bad voltage estimate by 2 std . devs , bump SigmaX
fprintf ( ' Bumping SigmaX \ n ') ;
SigmaX ( zkInd , zkInd ) = SigmaX ( zkInd , zkInd ) * ekfData . Qbump ;
end
[~ ,S , V ] = svd ( SigmaX ) ;
HH = V * S *V ';
SigmaX = ( SigmaX + SigmaX ' + HH + HH ') /4; % Help maintain robustness
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 5 of 8
3.4.7: Introducing Octave code to update EKF for SOC estimation
EKF iteration code, cleanup
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
EKF iteration code continues
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Store results for next time through loop
loop once each measurement interval while pack is active
% Save data in ekfData structure for next time ...
ekfData . priorI = ik ;
ekfData . SigmaX = SigmaX ;
ekfData . xhat = xhat ;
zk = xhat ( zkInd ) ;
zkbnd = 3* sqrt ( SigmaX ( zkInd , zkInd ) ) ;
end
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 6 of 8
3.4.7: Introducing Octave code to update EKF for SOC estimation
Example EKF on ESC results
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
In following example, EKF was executed for a test having
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
dynamic profiles from 100 % SOC down to around 10 % SOC
RMS SOC estimation error = 0:46 %
Percent of time error outside bounds = 0 %
SOC estimation using EKF SOC estimation errors using EKF
120 4
Estimate Estimation error
100 Truth Bounds
Bounds 2
SOC error (%)
80
SOC (%)
60 0
40
−2
20
0 −4
0 100 200 300 400 500 600 0 100 200 300 400 500 600
Time (min) Time (min)
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 7 of 8
3.4.7: Introducing Octave code to update EKF for SOC estimation
Summary
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Implementation of EKF on ESC model refactors code
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Initialization routine (initEKF.m), called once at startup
loop once each measurement interval while pack is active
Update routine (iterEKF.m), called every sample interval
“Wrapper” code, coordinates the entire simulation process
You have now seen the details of the entire codeset plus some example results
EKF works quite well as SOC estimator using ESC model
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 8 of 8
3.4.8: Where from here?
Summary of this week
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
This week, you learned:
current state of state of power
temperature charge (SOC) health (SOH) cells limits
How to generalize linear state-state model to describe
loop once each measurement interval while pack is active
nonlinear systems
How to apply two assumptions to derive the extended Kalman filter from the
generic Gaussian sequential-probabilistic-inference solution
How to write Octave/MATLAB code to implement EKF for a simple example and
for the battery-cell state-estimation problem
See some examples of applying EKF to estimate cell SOC using lab-test data
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 1 of 3
3.4.8: Where from here?
Where from here?
key off: store data
key on: initialize
Next week, you will learn
meas. voltage estimate estimate balance compute
current state of state of power
temperature charge (SOC) health (SOH) cells limits
Sigma-point approach to approximating statistics of a RV loop once each measurement interval while pack is active
passed through a nonlinear function
How to apply sigma-point approach to general Gaussian
sequential-probabilistic-inference solution to derive the
sigma-point Kalman filter
How to write Octave code to implement SPKF for a simple
example and for battery-cell state-estimation problem
See some examples of applying SPKF to estimate cell
SOC using lab-test data
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 2 of 3
3.4.8: Where from here?
Credits
key off: store data
key on: initialize
meas. voltage estimate estimate balance compute
Credits for photos in this lesson
current state of state of power
temperature charge (SOC) health (SOH) cells limits
loop once each measurement interval while pack is active
“Map” on slide 2: Pixabay license
(https://pixabay.com/en/service/license/), cropped from
https://pixabay.com/en/compass-navigation-map-direction-390054/
Dr. Gregory L. Plett j University of Colorado Colorado Springs Battery State-of-Charge (SOC) Estimation | Cell SOC estimation using an extended Kalman filter j 3 of 3
You might also like
- APU Industrial Placement Handbook Version 60% (2)APU Industrial Placement Handbook Version 651 pages
- An Approach to Estimate Lithium-Ion Battery State of Charge Based on Adaptive Lyapunov Super Twisting ObserverNo ratings yetAn Approach to Estimate Lithium-Ion Battery State of Charge Based on Adaptive Lyapunov Super Twisting Observer11 pages
- What Really Is State-Of-Charge (SOC) ?: Negative Electrode Positive ElectrodeNo ratings yetWhat Really Is State-Of-Charge (SOC) ?: Negative Electrode Positive Electrode2 pages
- Hu Et Al. - 2012 - Robustness Analysis of State-of-Charge Estimation Methods For Two Types of Li-Ion BatteriesNo ratings yetHu Et Al. - 2012 - Robustness Analysis of State-of-Charge Estimation Methods For Two Types of Li-Ion Batteries11 pages
- Robust and Adaptive Estimation of State of Charge For Lithium-Ion BatteriesNo ratings yetRobust and Adaptive Estimation of State of Charge For Lithium-Ion Batteries10 pages
- Implement Real-Time Soc Estimation For Ev BatteriesNo ratings yetImplement Real-Time Soc Estimation For Ev Batteries18 pages
- A Universal State-of-Charge Algorithm For BatteriesNo ratings yetA Universal State-of-Charge Algorithm For Batteries6 pages
- Support Vector Based Battery State of Charge Estimator: Terry Hansen, Chia-Jiu WangNo ratings yetSupport Vector Based Battery State of Charge Estimator: Terry Hansen, Chia-Jiu Wang8 pages
- State_of_Charge_Estimation_of_Lithium-Ion_Batteries_in_Electric_Drive_Vehicles_Using_Extended_Kalman_FilteringNo ratings yetState_of_Charge_Estimation_of_Lithium-Ion_Batteries_in_Electric_Drive_Vehicles_Using_Extended_Kalman_Filtering11 pages
- E-Prime - Advances in Electrical Engineering Electronics and Energy0% (1)E-Prime - Advances in Electrical Engineering Electronics and Energy14 pages
- A novel method for SOC estimation of Li-ion batteries using a hybNo ratings yetA novel method for SOC estimation of Li-ion batteries using a hyb15 pages
- Enhanced Soc Estimation of Lithium Ion Batteries With Realtime Data Using Machine Learning AlgorithmsNo ratings yetEnhanced Soc Estimation of Lithium Ion Batteries With Realtime Data Using Machine Learning Algorithms17 pages
- Research Article: State-of-Charge Estimation of Lithium-Ion Battery Pack Based On Improved RBF Neural NetworksNo ratings yetResearch Article: State-of-Charge Estimation of Lithium-Ion Battery Pack Based On Improved RBF Neural Networks10 pages
- Real time SOC estimation for Li-ion batteries in Electric vehicles using UKBF with online parameter identificationNo ratings yetReal time SOC estimation for Li-ion batteries in Electric vehicles using UKBF with online parameter identification19 pages
- Adaptive State of Charge Estimation of Lithium-Ion Batteries With Parameter and Thermal UncertaintiesNo ratings yetAdaptive State of Charge Estimation of Lithium-Ion Batteries With Parameter and Thermal Uncertainties8 pages
- Modeling and Simulation of Battery Management System BMS For Electric VehiclesNo ratings yetModeling and Simulation of Battery Management System BMS For Electric Vehicles11 pages
- Dubarry 2015 J. Electrochem. Soc. 162 A877No ratings yetDubarry 2015 J. Electrochem. Soc. 162 A8779 pages
- 5.3 The State of Charge Estimating Methods For Battery:: 5.3.1 Definition and Classification of SOC EstimationNo ratings yet5.3 The State of Charge Estimating Methods For Battery:: 5.3.1 Definition and Classification of SOC Estimation1 page
- 20.state of Charge Prediction of EV Li Ion Batteries Using EIS A Machine Learning ApproachNo ratings yet20.state of Charge Prediction of EV Li Ion Batteries Using EIS A Machine Learning Approach9 pages
- Coulomb Counting Approach For State of Charge EstimationNo ratings yetCoulomb Counting Approach For State of Charge Estimation33 pages
- State of Charge Estimation For Electric Vehicle Batteries Under An Adaptive Filtering FrameworkNo ratings yetState of Charge Estimation For Electric Vehicle Batteries Under An Adaptive Filtering Framework5 pages
- On-Board State Estimation in Electrical Vehicles Achieving Accuracy and Computational Efficiency Through An Electrochemical Model PDFNo ratings yetOn-Board State Estimation in Electrical Vehicles Achieving Accuracy and Computational Efficiency Through An Electrochemical Model PDF12 pages
- Robust Battery Management System for Electric VehiclesNo ratings yetRobust Battery Management System for Electric Vehicles6 pages
- A Novel Method For Lithium-Ion Battery State of Energy and State of Power Estimation Based On Multi-Time-Scale FilterNo ratings yetA Novel Method For Lithium-Ion Battery State of Energy and State of Power Estimation Based On Multi-Time-Scale Filter10 pages
- Journal of Power Sources: Yujie Wang, Duo Yang, Xu Zhang, Zonghai ChenNo ratings yetJournal of Power Sources: Yujie Wang, Duo Yang, Xu Zhang, Zonghai Chen10 pages
- A Comparison of Baseline Models and a Transformer Network for SOC Prediction in Lithium-Ion BatteriesNo ratings yetA Comparison of Baseline Models and a Transformer Network for SOC Prediction in Lithium-Ion Batteries15 pages
- Estimating The State of Charge of Lithium-Ion Batteries Using Different Noise InputsNo ratings yetEstimating The State of Charge of Lithium-Ion Batteries Using Different Noise Inputs11 pages
- Implementation of Online Battery State-Of-Power and State-Of-Function Estimation in Electric Vehicle ANo ratings yetImplementation of Online Battery State-Of-Power and State-Of-Function Estimation in Electric Vehicle A8 pages
- Joint Estimation of State of Charge and State of Energy of Lithium-Ion Batteries Based On Optimized Bidirectional Gated Recurrent Neural NetworkNo ratings yetJoint Estimation of State of Charge and State of Energy of Lithium-Ion Batteries Based On Optimized Bidirectional Gated Recurrent Neural Network12 pages
- State_of_Charge_Estimation_of_the_Lithium-ion_Power_Battery_based_on_a_Multi-Time-Scale_Improved_Adaptive_Unscented_Kalman_FilterNo ratings yetState_of_Charge_Estimation_of_the_Lithium-ion_Power_Battery_based_on_a_Multi-Time-Scale_Improved_Adaptive_Unscented_Kalman_Filter13 pages
- Energies: Robust State of Charge Estimation For Hybrid Electric Vehicles: Framework and AlgorithmsNo ratings yetEnergies: Robust State of Charge Estimation For Hybrid Electric Vehicles: Framework and Algorithms19 pages
- Spectral method for fatigue damage estimation with non-zero mean stressFrom EverandSpectral method for fatigue damage estimation with non-zero mean stressNo ratings yet
- Somatosensory Teachers Manual: for Somatosensory Science FactsFrom EverandSomatosensory Teachers Manual: for Somatosensory Science FactsNo ratings yet
- Battery Management Systems: Design and implementation for EVsFrom EverandBattery Management Systems: Design and implementation for EVsNo ratings yet
- Mineral Casting As Material For Machine Base Frames of Precision MachinesNo ratings yetMineral Casting As Material For Machine Base Frames of Precision Machines4 pages
- Model Specification Price Paper Straw Production SolutionNo ratings yetModel Specification Price Paper Straw Production Solution3 pages
- MS Mécatronique: Sabani Anass Sabani Anass A4 XXX XNo ratings yetMS Mécatronique: Sabani Anass Sabani Anass A4 XXX X1 page
- Pelletbox 750: The Complete Pelletising System For Dried Poultry Manure That Comes Ready To Be ConnectedNo ratings yetPelletbox 750: The Complete Pelletising System For Dried Poultry Manure That Comes Ready To Be Connected4 pages
- Price Schedule - Low Voltage ComponentsNo ratings yetPrice Schedule - Low Voltage Components94 pages
- Coupling A Supercapacitor With A Small Energy-Harvesting SourceNo ratings yetCoupling A Supercapacitor With A Small Energy-Harvesting Source4 pages
- Operation of A Single Extraction Steam Turbine Driven Centrifugal CompressorNo ratings yetOperation of A Single Extraction Steam Turbine Driven Centrifugal Compressor4 pages
- Manufacture of Miniature Cone Winding MachineNo ratings yetManufacture of Miniature Cone Winding Machine15 pages
- (Hensey 1993) ESSENTIAL TOOLS OF TOTAL QUALITY MANAGEMENTNo ratings yet(Hensey 1993) ESSENTIAL TOOLS OF TOTAL QUALITY MANAGEMENT11 pages
- Tip Clearance Noise of Axial Ow Fans Operating at Design and Off-Design ConditionNo ratings yetTip Clearance Noise of Axial Ow Fans Operating at Design and Off-Design Condition24 pages
- Burra V L Narasimha: Seeking Assignments inNo ratings yetBurra V L Narasimha: Seeking Assignments in2 pages
- Introduction To 500 MW: 8 January 2013 PMI Revision 00 1No ratings yetIntroduction To 500 MW: 8 January 2013 PMI Revision 00 118 pages
