7th Power Electronics, Drive Systems & Technologies Conference (PEDSTC 2016)

16-18 Feb. 2016, Iran University of Science and Technology, Tehran, Iran

Model Predictive Control of Classic Bidirectional

DC-DC Converter for Battery Applications

A. Pirooz, R. Noroozian
Department of Electrical Engineering
University of Zanjan
Zanjan, Iran
[email protected], [email protected]

Abstract— In this paper, controlling of a DC-DC bidirectional Controlling bidirectional DC-DC converters is another
converter is investigated using model predictive control (MPC) challenge that has been the subject of many researches in
technique. The converter is supposed to connect a battery bank literature. Controlling method must provide fast dynamic
to the DC bus. The main purpose of MPC method is to maintain response and follow the reference signal in both power flow
the DC bus voltage in a predefined rated range, which means directions. Pulse width modulation (PWM) and phase-shift
injecting active power to the DC bus to increase bus voltage and control [7], hysteresis current control (HCC) [8] and fuzzy
discharging the battery, and on the contrary absorbing active control [9] are different controlling techniques that could be
power from the DC bus to decrease bus voltage and charging the applied to bidirectional DC-DC converters.
battery. The control strategy also disconnects the battery bank
while DC bus voltage is in the predefined rated range. Model predictive control (MPC) is a derivative of
Simulations have been done in PSCAD/MATLAB software predictive control family which uses a predefined cost
interfacing and results show the feasibility of the proposed function in order to make system variables follow their
controlling strategy to conduct active power on both directions. reference values. An accurate mathematical discrete-time
model of the system is needed to predict future behavior of the
Keywords—bettery bank; model predictive control; bidirectional variables. On each time step, subject to the predicted values
DC-DC converter; DC voltage control and the cost function, an optimization problem is solved
online over a finite prediction horizon which leads to optimal
I. INTRODUCTION controlling actions. Basic idea of MPC is easy to understand
Nowadays distributed generation (DG) units are highly and it could be applied to every system with known
integrated into DC and AC distribution networks. Renewable mathematical models. A large number of calculations needs to
energy sources as photovoltaic and wind power units cannot be done online, but with powerful microprocessors available
supply the needs of power continuously in a distribution today, this cannot be a challenge for implementing MPC into
network due to their dependencies to environmental conditions power electronic converters which operate with fast dynamic
[1]. A battery bank is used to compensate the power difference behavior [10]. This research applies MPC to a classic
between loads and generation units [2]. Obviously batteries bidirectional DC-DC converter. It takes advantage of
are not infinite sources of power and discharge in a rather independent buck and boost operational modes of the
short period. In these cases also surplus power in the system is converter. Low voltage (LV) and high voltage (HV) sides of
used to charge the battery banks. As a result, the converter the converter are connected to a battery bank and a DC
connecting the battery bank to the system must allow power distribution network bus respectively. A predefined rated
flow between two buses on both directions. Bidirectional range is provided for the DC bus voltage which has a lower
converters are also used in DC uninterruptable power supplies and upper limit. A controlling strategy is proposed based on a
(UPS), electrical vehicles and ships [3-5]. voltage feedback signal from DC bus to discharge the battery
while DC bus voltage value is below the predefined rated
Basically, unidirectional DC converters as buck and boost, range, to charge the battery while DC bus voltage value
doesn't allow reverse current conductance since they have exceeds the predefined rated range and to disconnect the
diodes in their topology. By replacing diodes with controllable battery (idle mode) while DC bus voltage value is in the
switches they could be turned into bidirectional converters [6]. predefined rated range. The battery itself is modeled by its
The classic bidirectional converter which is shown in Fig. 1 is internal voltage VBAT and resistance RBAT. Accurate modeling
a combination of buck and boost converter which has the same of the battery bank is out of the subject of this research.
structure in both cases. The inductor is on the low voltage
(LV) side which the battery is connected to. Existence of the The rest of this paper is organized as follows. Section II
inductor on LV side decreases current ripple which is includes modeling of the converter. Predictive control of the
advantageous for charging and discharging the battery and converter is discussed in section III. Section IV includes
makes this converter suitable for battery applications. proposed controlling strategy for charging/discharging the
battery. Finally, simulation results using PSCAD/MATLAB
software interfacings are demonstrated in section V.

978-1-5090-0375-4/16/$31.00 ©2016 IEEE 517

Authorized licensed use limited to: Suranaree University of Technology provided by UniNet. Downloaded on December 10,2024 at 07:56:36 UTC from IEEE Xplore. Restrictions apply.
 +
S1
iL L RBAT
S1 VDC +
iL L S2 VBAT
EBAT
VDC _ _
S2 VBAT DC BUS Battery Bank
(a)
S1 L iL
+ _
Fig. 1. Classic bidirectional DC-DC converter topology. VL +

VDC
+
_ C VBAT
II. CONVERTER MODELING _

The bidirectional DC-DC converter shown in Fig. 1 is a

combination of buck and boost converters. While S1 is (b)
switching and S2 is constantly off, the converter operates in L iL
independent buck mode, DC bus voltage level is decreased to +
VL
_

+
battery voltage level, and power flows from the DC bus to
charge the battery. While S2 is switching and S1 is constantly C VBAT
off, the converter operates in independent boost mode, battery _

voltage level is increased to DC bus voltage level, and power

injected to the DC bus discharges the battery. Due to the (c)
specific application and battery behavior, the converter is
supposed to operate only in continuous current mode (CCM). Fig. 2. Buck mode operation: a) buck mode topology b) switch on S1=1 c)
switch off S1=0.
A. Buck Mode
To obtain an accurate equation for prediction block,
discrete-time model of the system is needed due to inherent
Buck mode topology is shown in Fig. 2 (a). There are two procedure of MPC. By using Euler forward method, discrete-
semiconductors, controllable switch S1 and a diode. The time model of first order systems could be obtained by
inductor L is used to store and release electrical energy approximation of derivatives as below:
depending on switch S1 state. When S1 is on (S1=1), energy is
stored in the inductor L through DC bus as shown in Fig. 2 (b).
When S1 is off (S1=0), energy is released from the inductor L dx x(k + 1) − x(k )
to the battery as shown in Fig. 2 (c). Consequently, = (3)
dt Ts
continuous-time state space equations of the converter
becomes as below [11]:
where Ts is the sampling time. However, for higher order
systems this method introduces significant error, thus exact
dx(t ) discretization must be done [10]. By using (3), discretization
= ( A1 + A2u (t )) x(t ) + Bu(t )VDC of (1) is resulted as:
dt (1)
y (t ) = Cu(t ) x(t )
­E x(k) + FVDC , y(k) = Gx(k) for S1 = 1
where state variables are inductor current iL (t ) and battery x(k + 1) = ® 1 (4)
¯ E2 x(k) , y(k) = 0 for S1 = 0
voltage v BAT (t ) :
the matrices E1, E2, F and G are also defined as below:
x (t ) = [i L (t ) v BAT (t )] T
(2)
E1 = 1 + ( A1 + A2 )Ts , E2 = 1 + A1Ts
system matrices A1, A2, B and C are defined as below. u(t) is F = BTs , G = C
status of the switch S1. If S1 is on, u(t)=1 and if S1 is off,
u(t)=0. y(t) is also the output of the system. note that 1 is the identity matrix.
ª 1º
ª0 0 º «0 − ª1º B. Boost Mode
A1 = «0 − 1 », A2 = « L », B = « », C = [0 1].
1 » L Boost mode topology is shown in Fig. 3 (a). There are two
« R BAT C »¼ « 0 » «0»
¬ ¬ ¼ semiconductors, controllable switch S2 and a diode. The
¬C ¼
inductor L is used to store and release electrical energy

518
Authorized licensed use limited to: Suranaree University of Technology provided by UniNet. Downloaded on December 10,2024 at 07:56:36 UTC from IEEE Xplore. Restrictions apply.
 depending on switch S2 status. When S2 is on (S2=1), energy is ­ E x(k ) + FVBAT for S = 1
stored in the inductor L through DC bus as shown in Fig. 3 (b). x(k + 1) = ® 1
¯ E2 x(k ) + FVBAT for S = 0 (7)
When S2 is off (S2=0), energy is released from the inductor L
to the battery as shown in Fig. 3 (c). Consequently, y (k ) = Gx(k )
continuous-time state space equations of the converter
becomes as below [11]: the matrices E1, E2, F and G are also defined as below:
E1 = 1 + ( A1 + A2 )Ts , E 2 = 1 + A1Ts
dx (t ) F = BTs , G = C
= ( A1 + A2 u (t )) x(t ) + BVBAT
dt (5) note that 1 is the identity matrix.
y (t ) = Cx(t )
III. MODEL PREDICTIVE CONTROL OF THE CONVERTER
where state variables are inductor current iL (t ) and DC bus
MPC scheme of buck and boost converters are presented in
voltage v DC (t ) : this section. Discrete-time modeling of the converter has been
achieved on previous section. MPC is generally classified into
discrete-time modeling, prediction and optimization blocks.
x (t ) = [i L (t ) v DC (t )]T (6) Measurements are imported to the algorithm on prediction
stage, based on previous modelings, all possible future
system matrices A1, A2, B and C are defined as below. u(t) is behavior of controlled variables are predicted as below:
status of the switch S2. If S2 is on, u(t)=1 and if S2 is off,
x pi (t k +1 ) = f p {x(t k ), S i }
u(t)=0. y(t) is also the output of the system. (8)
i = 1,..., n
ª 1 º ª 1º
«0 − » 0 ª1º future predicted values of control variables xpi(tk+1) are a
A1 = « 1 L ,A =« L », B = « », C = [0 1]
1 » « » L function of current values of control variables x(tk) subject to
« »
2
1 «0»
− «− 0» ¬ ¼ switching state of the converter Si. i is number of possible
¬« C R DC C ¼» ¬ C ¼ states for switchings. For example, there are two possible
switchings (on or off) for boost mode as there is only one
controllable switch. Equations (4) and (7) are used as
prediction function for buck and boost modes respectively.
The output of prediction stages is a two dimensional array for
iL L RBAT each operation mode. All predictions are forwarded to the
VDC optimization stage. MPC defines the output control actions
EBAT subject to a predefined cost function:
VBAT
g i = f g {x ∗ (t k +1 ), x pi (t k +1 )} (9)

gi is a function of predicted values imported from prediction

iL L io block xpi(tk+1) and future values of reference signal x*(tk+1).
Since the reference signal is a DC value, it can be supposed
that it remains constant during sampling time Ts:
VBAT S2 C VDC RDC
x ∗ (t k +1 ) ≈ x ∗ (t k ) (10)
gi is usually defined as error between variable values and their
reference value. Consequently the cost function needs to be
iL L Diode io minimized by MPC scheme. On each sampling interval Ts , the
switching state that minimizes gi is chosen by MPC scheme to
VBAT be applied on next sampling interval. The cost function for
C VDC RDC
buck mode is defined as below:
­° | V ∗ (k + 1) − VBAT ,S1 =1 (k + 1) | for S1 = 1
J buck = ® BAT ∗
(11)
°̄| VBAT (k + 1) − VBAT , S1 =0 (k + 1) | for S1 = 0
Fig. 3. Boost mode operation: a) boost mode topology b) switch on S2=1 c)
switch off S2=0. and also for boost mode:

where RDC is the equivalent resistance seen by the DC bus. By ­° | V ∗ (k + 1) − VDC,S2 =1 (k + 1) | for S 2 = 1
J boost = ® DC (12)
using (3), discretization of (5) is resulted as: ∗
°̄| VDC (k + 1) − VDC,S2 =0 (k + 1) | for S 2 = 0

519
Authorized licensed use limited to: Suranaree University of Technology provided by UniNet. Downloaded on December 10,2024 at 07:56:36 UTC from IEEE Xplore. Restrictions apply.
 ∗ ∗
V DC ( k ), V BAT ( k ), iL ( k ), vbuck V DC ( k ), V BAT ( k ), iL ( k ), vboost

x S1 =1 ( k + 1) = E1 x ( k ) + FV DC xS2 =1 (k + 1) = E1x(k ) + FVBAT

x S1 = 0 ( k + 1) = E 2 x ( k ) xS2 =0 (k + 1) = E2 x(k ) + FVBAT

∗ ∗
J buck,S1=1 =| VBAT (k + 1) − VBAT ,S1=1 (k + 1) | J boost,S2 =1 =| VDC (k + 1) − VDC,S2 =1 (k + 1) |
∗ ∗
J buck,S1=0 =| VBAT (k + 1) − VBAT ,S1=0 (k + 1) | J boost,S2 =0 =| VDC (k + 1) − VDC,S2 =0 (k + 1) |

Jbuck,S1=0 < Jbuck,S1=1 Jboost,S2=0 < Jboost,S2=1

S1 = 0 S1 = 1 S2 = 0 S2 = 1

S1 S2

Fig. 4. Simplified flowcharts of MPC scheme a) buck operation b) boost operation.

Simplified flowcharts of MPC scheme for buck and boost minimizes cost function is selected and sent to the converter to
operational modes of bidirectional DC-DC converter are be applied on the next sampling interval. For the buck
shown in Fig. 4. operation, DC bus voltage VDC(k), inductor current iL(k) and
battery voltage VBAT(k) are imported to the prediction stage. A
IV. PROPOSED CONTROL STRATEGY two dimensional array VBAT(k+1) is sent to the optimization
block from prediction stage. This array contains two predicted
The converter aims to regulate DC bus voltage VDC and battery voltage values of the next sampling interval tk+1=tk+Ts
keep it in the predefined rated range. Upper and lower limits subject to switching states. Reference value
are defined for the control scheme as shown in Fig. 5. As DC ∗ ∗
bus voltage VDC drops below the lower limit, bidirectional VBAT (k ) ≈ VBAT (k + 1) is generated from an external loop as
converter switches to boost mode, injects power to the DC bus below:
by discharging the battery in order to keep VDC on lower limit. ∗
On the contrary, as the DC bus voltage VDC exceeds the upper ∗ iDC (k ). VBuck
VBAT (k ) =
limit, bidirectional converter switches to buck mode, absorbs iL ( k )
power from the DC bus by charging the battery in order to (13)
keep VDC on upper limit and also switches to idle mode and
∗
disconnects the battery from the DC bus as DC bus voltage where iDC(k) is the current passing through DC bus and VBuck is
goes into the allowed range. Fig. 6 demonstrates the whole upper limit of the predefined rated voltage range of DC bus.
control scheme. Bidirectional converter is placed between By using (13) for calculating reference value of buck
battery bank bus and DC bus. The main control idea is based operation, the converter controls its output voltage VBAT to
on a voltage feedback signal VDC coming from the DC bus. It keep its input voltage VDC on the upper limit of allowed range.
is passed through a low pass filter to attenuate voltage ripple.
Buck and boost predictive controllers operate independently. V DC
For the boost operation, DC bus voltage VDC(k), inductor
current iL(k) and battery voltage VBAT(k) are imported to the ∗
(VBuck )
prediction stage. A two dimensional array VDC(k+1) is sent to
the optimization block from prediction stage. This array
∗
contains two predicted DC bus voltage values of the next (VDC )
sampling interval tk+1=tk+Ts subject to switching states.
∗ ∗
Reference value VDC (k ) ≈ VDC (k + 1) is also imported to the
∗
optimization block. VDC is lower limit of the predefined rated I DC,min IDC I DC,max
voltage range of DC bus. Cost function (12) is evaluated for
each predicted value and the switching state SBoost that
Fig. 5. Three operational modes based on DC bus voltage droop.

520

Authorized licensed use limited to: Suranaree University of Technology provided by UniNet. Downloaded on December 10,2024 at 07:56:36 UTC from IEEE Xplore. Restrictions apply.
 RBAT
EBAT
VBAT C Vc VDC

S1 S2 VDC (k )
i DC (k )
0,1

0,1

VDC (k )

0,1

iL (k )
S Buck S Boost

VBAT (k +1)
VBAT (k )
∗
V BAT (k )

VDC (k +1)

∗
V DC (k )
∗
V Buck
VDC (k )
i DC (k )

Fig. 6. Block diagram of the proposed control strategy.

Cost function (11) is evaluated for each predicted value

and the switching state SBuck that minimizes cost function is TABLE II. SYSTEM PARAMETERS
selected and sent to the converter to be applied on the next Parameter Value
sampling interval. The outputs of predictive controllers SBoost
Voltage Range Lower Limit ( VDC
∗ )
100 V
and SBuck are headed toward the bidirectional converters
∗
switches. But three range comparators (RC) decide which Voltage Range Upper Limit ( VBuck ) 150 V
operational mode to be activated. DC bus voltage VDC will Battery Rated Voltage 50 V
enter into three range comparators after passing low pass Internal Battery Resistance ( R BAT ) 0.01 Ÿ
filter. The output of each comparator is 0 or 1. Table I shows Converter Inductance ( L ) 8 mH
how range comparators control operational modes. RC1 Converter Capacitance ( C ) 1000 μF
controls the idle mode, RC2 controls the boost mode and RC3 Sampling Time ( Ts ) 33 μs
controls the buck mode.

TABLE I. OPERATIONAL MODE CONTROL

Range Range Range
Mode/RC
Comparator 1 Comparator 2 Comparator 3
Idle 0 0 0
Buck/Charging 1 0 1
Boost/Discharging 1 1 0

V. SIMULATION RESULTS
Steady state buck and boost operations and transition from
boost to buck operation through idle mode by increasing DC
bus voltage have been implemented in PSCAD/MATLAB
interfacing. Fig. 7 and Fig. 8 show the steady state operations
of the converter in buck and boost modes respectively. System
parameters are also demonstrated in Table II. Fig. 7. DC bus voltage and current in buck mode.

521

Authorized licensed use limited to: Suranaree University of Technology provided by UniNet. Downloaded on December 10,2024 at 07:56:36 UTC from IEEE Xplore. Restrictions apply.
 150 V, converter starts to charge the battery in buck mode
operation to prevent the voltage from exceeding voltage upper
limit.
As shown in Fig. 10, during battery discharge, about 4
amperes of current flows to the DC bus from the battery. On
idle mode, no current is exchanged between battery and DC
bus, consequently during charging mode, current direction is
reversed and battery absorbs 6 amperes of current.

VI. CONCLUSION
This research applies model predictive control to a
Fig. 8. DC bus voltage and current in boost mode.
bidirectional DC-DC converter. Mode activation control
strategy is proposed to control the battery charge, discharge
and idle modes. Each of the three modes operate
independently due to the converters structure. The main
control strategy is based on a voltage feedback signal from DC
bus. A predefined rated range is considered for the bus voltage
value thus the controller defines which mode to be activated.
Simulation results show the feasibility of MPC and proposed
controlling strategy to be used in battery applications due to
proper reference tracking and fast dynamic response during
transitions between modes.

References
[1] Lin, C. C., Yang, L. S., & Wu, G. W. (2013). Study of a non-isolated
bidirectional DC–DC converter. IET Power Electronics, 6(1), 30-37.
[2] Noroozian, R., Abedi, M., Gharehpetian, G. B., & Hosseini, S. H.
(2009). Combined operation of DC isolated distribution and PV systems
for supplying unbalanced AC loads. Renewable Energy, 34(3), 899-908.
[3] Zhang, Z., Thomsen, O. C., Andersen, M. A., Schmidt, J. D., & Nielsen,
H. R. (2009, February). Analysis and design of bi-directional DC-DC
converter in extended run time DC UPS system based on fuel cell and
Fig. 9. Transition between three operational modes. supercapacitor. In Applied Power Electronics Conference and
Exposition, 2009. APEC 2009. Twenty-Fourth Annual IEEE (pp. 714-
719). IEEE.
[4] Abedi, M., Song, B. M., & Kim, R. Y. (2011, September). Nonlinear-
model predictive control based bidirectional converter for V2G battery
charger applications. In Vehicle Power and Propulsion Conference
(VPPC), 2011 IEEE (pp. 1-6). IEEE.
[5] Zahedi, B., Nebb, O. C., & Norum, L. E. (2012, June). An isolated
bidirectional converter modeling for hybrid electric ship simulations. In
Transportation Electrification Conference and Expo (ITEC), 2012 IEEE
(pp. 1-6). IEEE.
[6] Karshenas, H. R., Bakhshai, A., Safaee, A., Daneshpajooh, H., & Jain,
P. (2011). Bidirectional dc-dc converters for energy storage systems.
INTECH Open Access Publisher.
[7] Xu, D., Zhao, C., & Fan, H. (2004). A PWM plus phase-shift control
bidirectional DC-DC converter. Power Electronics, IEEE Transactions
on, 19(3), 666-675.
[8] Cernat, M., Scortaru, P., Tanase, A., & Iov, F. (2007, September).
Hysteretic current controlled ZVS dc/dc converter for automobiles. In
Fig. 10. Battery voltage and current during boost, idle and buck mode. Power Electronics and Applications, 2007 European Conference on (pp.
1-10). IEEE.
Transition from boost to buck mode is shown in Fig. 9. [9] Hazil, O., Bououden, S., Chadli, M., & Filali, S. (2014). Fuzzy model
During 0.23 < t < 0.25 the battery is discharging and converter predictive control of dc-dc converters. In AETA 2013: Recent Advances
in Electrical Engineering and Related Sciences (pp. 423-432). Springer
operates in boost mode and bus voltage is kept on 100 V Berlin Heidelberg.
which is the lower limit of the predefined allowed range. Then [10] Rodriguez, J., & Cortes, P. (2012). Predictive control of power
during 0.25 < t < 0.32 DC bus voltage is increased to the converters and electrical drives (Vol. 40). John Wiley & Sons.
allowed range and all switchings are off since the converter is [11] Erickson, R. W., & Maksimovic, D. (2007). Fundamentals of power
in idle mode. As the voltage reaches its predefined upper limit electronics. Springer Science & Business Media.

522

Authorized licensed use limited to: Suranaree University of Technology provided by UniNet. Downloaded on December 10,2024 at 07:56:36 UTC from IEEE Xplore. Restrictions apply.