Scribd
Upload
56 views

Model Predictive Control of Industrial Processes

This document discusses model predictive control (MPC) applied to an industrial power plant process. It first describes the process and existing PID control loops. Then it covers identifying linear models of the process using neural networks and polynomial models. An MPC controller is designed using a 4-state space model, with constraints on the gas flow input. Simulation results show MPC improves control quality and reduces gas consumption compared to the original PID loops. Future plans include further validation and minimizing raw material use with MPC.

Uploaded by

Energya Corporacion
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
56 views

Model Predictive Control of Industrial Processes

This document discusses model predictive control (MPC) applied to an industrial power plant process. It first describes the process and existing PID control loops. Then it covers identifying linear models of the process using neural networks and polynomial models. An MPC controller is designed using a 4-state space model, with constraints on the gas flow input. Simulation results show MPC improves control quality and reduces gas consumption compared to the original PID loops. Future plans include further validation and minimizing raw material use with MPC.

Uploaded by

Energya Corporacion
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 33

Model predictive control of industrial

processes
Vitali Vansovitš
Contents

Industrial process (Iru Power Plant)


Neural network identification
Process identification – linear model
Model predictive controller
Results
Future plans

2 (33)
Industrial process (Iru Power Plant)

Three similar boilers in


parallel
Each boiler can work in:
basic mode (constant
load)
control mode (variable
load)
Control mode is used to
control plant output
temperature

3 (33)
Industrial process control
Two cascade control loop
Cascade control loop
Control loop

SP3+ SP2+ SP1+ Output


PID 3 PID 2 PID 1 Process 1 Process 2 Process 3
– – –

fast
slow
slowest

4 (33)
Industrial process control (Iru Power Plant)

C3: kp = 0,25; ti = 20
C2: kp = 0,9; ti = 240
C1: kp = 0,5; ti = 180

5 (33)
Industrial process (Iru Power Plant)
84
83
82
P. output SP
81
80 P. output ME
79
78
0:00 1:12 2:24 3:36 4:48 6:00 7:12

130
125
120
Setpoint
115
110 Measurement
105
0:00 1:12 2:24 3:36 4:48 6:00 7:12

3000
2800
2600
2400 Water flow

2200
2000
0:00 1:12 2:24 3:36 4:48 6:00 7:12

6 (33)
Neural network model
Recursive prediction

y – output temperature prediction


u – vector of water, air and gas flows
7 (33)
Neural network 5 min prediction

126

124

122
Temperature, deg

120

118

116

Temperature, deg
114 Predicted temperature, deg

112
0 100 200 300 400 500 600 700 800 900
Time, min

8 (33)
Neural network first results
5 minutes predictions were good
Simulation was not possible
After 10 cycles model output was not close enough to real
process output
After 100 cycles model output was already somewhere in
outer space
So, it was decided to start from simple things – linear
model identification with conventional methods.

9 (33)
Process identification
Process is simple, but nonlinear:

Q r m fuel r m fuel
T
Q c mwater T c mwater

As water flow is more or less constant during normal


operation, so mainly it can be seen as linear.
Matlab Identification Toolbox was used for
identification.
Most suitable results were acquired from polynomial
and state space models.

10 (33)
Polynomial model identification

11 (33)
State-space model

12 (33)
Process model
State space model with 4 states (ss2) was selected
as:
it is simpler than oe221 (6 states in state space representation)
shows better results then ss1 (1 state)

4 state model showed better fit among all state


space models (2, 3, 5 etc. states)

13 (33)
MPC (Model predictive controller)
Model identification
To avoid perfect “theoretical results” model for MPC was
identified from the data acquired one year later at the
time when boiler load was different (~1/8 of previous
example) and nonconstant.

14 (33)
MPC structure

15 (33)
MPC observer
xˆ k | k xˆ k | k 1 L y k yˆ k | k 1
xˆ k 1 | k Axˆ k | k Bu k
yˆ k | k 1 Cxˆ k | k 1

xˆ k 1 | k AI L C xˆ k | k 1 Bu k AL y k
ek xk xˆ k | k 1 ek 1 A LC e k

Observer is stable - estimation error converges to zero

16 (33)
MPC design (1)
2 2
There is a cost function: V k Z k T k Q
U k R

where
Z(k) – predicted process output
T(k) – reference trajectory
U(k) – process input changes
Q, R – weight matrices
2
a A
a T Aa

Z k xk uk 1 U k

k k xk uk 1

Tracking error Free response with U = 0


17 (33)
MPC design (2)
From this function optimal input changes for
unconstrained case are found:
1 1
U k H G
2

where G 2 T
QE k
T
H Q R

18 (33)
MPC constraints
Constraints are in the form:

U k U k Z k
E 0, F 0, G 0
1 1 1

Which can be transformed to: W U k w


So, we solve constrained optimization problem:

T
minimize U k H U k G T U k
subject to inequality constraint.
Quadratic programming problem

19 (33)
MPC for water boiler
1 manipulated process input (gas flow)
2 measured process disturbances (water & air flows)
1 measured process output

Prediction horizon 10 steps


Control horizon 2 steps

Gas flow constraints (m3/h):


min = 0 max = 16000
max down rate = 2000 max up rate = 2000

Gas flow rate weight = 0,01


Process output weight = 10

20 (33)
MPC for water boiler (Simulink)

21 (33)
Outputs comparison

22 (33)
Control quality

23 (33)
Gas consumption

24 (33)
Earn some money
Same control quality, but less raw materials

25 (33)
Setpoint decreased 0,6 (1)
Output comparison

26 (33)
Setpoint decreased 0,6 (2)
Gas consumption difference

~8000 m3

27 (33)
Setpoint decreased 0,6 (3)
Sum of errors

28 (33)
Very rough estimation
Gas price 400 € / 1000 m3
Difference 8000 m3
Period 16 hours
Season length 50 days

Economy
= 50 * (24/16) * 8000 * 400 / 1000
= 240000 € / season

29 (33)
Some results

With MPC control became better:


Sum of errors of output are twice lower
MPC with bad model is better than PI cascade
Raw material consumption is on the same level
If we change setpoint then we can decrease raw
material consumption with the same control quality
as PI loops have.

30 (33)
Future plans

Compare MPC with simulated PI loop


Minimize raw material consumption with MPC (not
setpoint change)
Find better way of identification
Identify all parts of the process (including actuators)
Make model based control closer to reality
Try controller on real plant some day
Etc.

31 (33)
References
Oliver Nelles, Nonlinear System Identification,
Springer, Berlin, 2001
J.M.Maciejowski, Predictive Control with Constraints,
pdf book, 2000
Real life

32 (33)
Thank you!

Questions???

You might also like