Professional Documents
Culture Documents
Abstract—This paper presents a multi-objective model predic- inductance has been used to form a boost converter for
tive control (MPC) approach for a three-phase, split converter-fed integrated AC and DC charging of the batteries. The use of
switched reluctance motor (SRM) integrated battery charger for asymmetric half-bridges improves the system’s fault tolerance
Electric Vehicle (EV) application. The proposed control is based
on finite control set (FCS) model predictive control which is capability, however, the charging of batteries becomes chal-
suitable for online prediction with a unit prediction horizon. With lenging due to the high number of control objectives involved.
the help of a tailored multi-objective cost function, the proposed These objectives include power factor correction, input current
MPC achieves unity input power factor while also equalizing the reference tracking and SoC equalization so as to avoid unequal
batteries’ state of charge (SoC) at the end of the charging cycle or over charging of batteries. The work in [11] has utilized
regardless their initial SoC. The performance of the proposed
MPC has been verified using MATLAB/Simulink. a multi-stage hysteresis controller to achieve power factor
Index Terms—Electric Vehicle, Plug-in Hybrid Electric Vehicle, coreection (PFC) and SoC equalization.
Charger, Switched reluctance motor, State of Charge, Model In such a multi-objective control application, model pre-
Predictive Control, Finite Control Set. dictive control makes an excellent candidate because of its
systematic approach and ability to deal with multi-input multi-
I. I NTRODUCTION output (MIMO) systems [12], [13]. MPC of battery chargers
Module-1 Module-4
Sxu Dxu S1u D1u S4u D4u
Bx Cx L3y
B1 C1 L1a L2a B4 C4 L3a L4a
L4y
Sxd D1d S1d D4d S4d
Dxd
x’
L2y
(a) Module-2 Module-5
Module-6
S2u D2u S5u D5u
Module-3
Module-5
B2 C2 L1b L2b B5 C5 L3b L4b
Module-2
Module-1 Module-4
D2d S2d D5d S5d
BATTERIES
SoC
[ B5 B2
SoCB2
B5
2 contains six power stages and appears quite complex with SoC
[ B6 B3
SoCB3
B6
the modeling perspective. However, with some simplifications,
the circuit can be redrawn as a multi-switch boost converter. q4 q5 q6 iRef PLL q1 q2 q3
sinθ
The simplified converter diagram along with the proposed
3 iL Vi iL,ref 3
control structure is shown in Fig. 3. It can be seen that in
the positive half-cycle, the inductor L is charged using S1d , (VBj, zBj) MPC (VBj, zBj)
jϵ[4,5,6] jϵ[1,2,3]
S2d and S3d through an equivalent diode Dp , while in the
negative half-cycle, the same is achieved using S4d , S5d and Fig. 3: Simplified circuit diagram of the integrated charger along with MPC
control structure.
S6d through Dn . It is important to note here that the equivalent
circuit shown in Fig. 3 has been drawn for the convenience of
modeling and the exact circuit might look different. Moreover, discharging) and κ is the capacity of the battery. Moreover,
all the switches and diodes have been considered to be ideal i(t) represents the instantaneous current of the battery (i(t) is
for model simplification purposes. positive in discharging and negative in charging).
1) Inductor current: In the positive half-cycle of the input In the considered boost converter configuration shown in
AC source Vi , the diode Dp is forward-biased and the current Fig. 3, the charging current of a battery Bj will only flow
flows through the inductor L from the switches S1d , S2d and when its corresponding switch function qj is zero. Moreover,
S3d . In this case, the voltage across the inductor VL is given the amplitude of this current will be equal to the inductor
by current iL . Considering this, one can rewrite (6) for all the
VL (t) = Vi (t) − (1 − qj )VBj (2) batteries as a function of the inductor current iL and the switch
j=1,2,3 function of the corresponding module qj as
where VBj represents the terminal voltage of the correspond- dzBj (t) η(1 − qj )iL (t)
= (7)
ing battery Bj and the switch function qj is given by dt κBj
1 if Sjd is ON. where j ∈[1,2,3,4,5,6], and zBj and κBj denote the SoC and
qj = (3)
0 if Sjd is OFF. the capacity of the battery Bj , respectively.
Considering (2), one can write the differential equation for B. Discrete-Time Model
the inductor current iL in the positive half-cycle as
The proposed model predictive control algorithm finds an
diL (t) VL (t) Vi (t) VBj
= = − (1 − qj ) . (4) optimal control input using a discrete-time model of the plant.
dt L L j=1,2,3
L Therefore, it is necessary to convert the continuous-time model
into a discrete-time model using a sampling time Ts . Using
Similarly, the differential equation of the inductor current in
forward Euler approximation, the discrete-time model of the
the negative half-cycle is given by
plant defined by (4), (5) and (7) is given by
diL (t) Vi (t) VBj
=− − (1 − qj ) . (5) Ts
dt L L iLk+1 = iLk + V ik − (1 − qjk )VBjk (8)
j=4,5,6 L j=1,2,3
2) State of charge (SoC): In order to regulate the state of
Ts
charge of different batteries in the charging process, it is also iLk+1 = iLk + − V ik − (1 − qjk )VBjk (9)
L
necessary to consider its dynamics in the model of the plant. j=4,5,6
The dynamics of the state of charge of a battery is defined by η(1 − qjk )iLk Ts
zBjk+1 = zBjk + (10)
[19] κBj
dz(t) ηi(t) where k and k + 1 represent the discrete-time instants kTs
= (6)
dt κ and (k + 1)Ts , respectively. Moreover, (8) and (9) represent
where z(t) denotes the SoC of a battery at time t, η is the the discrete-time equations for the inductor current in positive
Coulumbic efficiency (η ≤ 1 in charging and η = 1 in and negative half-cycles of the input AC voltage, respectively.
IV. MPC F ORMULATION AND C ONTROL A LGORITHM START
Having derived a discrete-time model for the plant in
Acquire iL(k), VB1(k), …,
Section III-B, a model predictive control problem can now VB6(k), zB1(k), …, zB6(k)
be formulated. In the model given by (8)-(10), iL and zBj
represent the states of the system, while qj (j ∈[1,2,3,4,5,6]) iL,ref(k), Vi(k)
are the control inputs. Moreover, it can be seen that both iL No Yes
No Yes
and zBj are monotonically varying functions of the control Vi(k) > 0
input qj . Therefore, a reasonable control performance can
be achieved using a short prediction horizon [20]. For this Mean(zB4,zB5,zB6)|k Mean(zB1,zB2,zB3)|k
application, a prediction horizon of one has been chosen in <0.99 <0.99
No
No No
No
order to keep the computational requirement to a minimum.
Reduce iL,ref (k) Yes
Yes Yes
Yes Reduce iL,ref (k)
A. Cost Function and Control Problem n=1 n=1
In model predictive control, the optimal control strategy is
devised by minimizing a cost function. Therefore, it is very
J(k)=ρ1J1(k)+ρ2J2(k) J(k)=ρ1J1(k)+ρ2J2(k)
important to appropriately define the cost function in order jϵ[4,5,6], iL(k+1) from (9) jϵ[1,2,3], iL(k+1) from (8)
to achieve all the control objectives. For the present control
problem, the control objectives can be listed as follows: No
No No
No
n = n+1 n=8 n=8 n = n+1
1) Power factor correction
Yes
Yes Yes
Yes
2) Input current reference tracking
Un* = argmin(J(k)) Up* = argmin(J(k))
3) Equalization of SoC of different batteries in the charging
process
[q4*, q5*, q6*] = [q1*, q2*, q3*] =
Considering these objectives, the cost function can be divided [Un*(1), Un*(2), Un*(3)] [Up*(1), Up*(2), Up*(3)]
into two parts: one part to achieve the power factor correction
END
and current reference tracking, and the second to equalize the
SoC of batteries in the charging process. Fig. 4: Flowchart of the proposed finite control set MPC algorithm.
In this application, the first part of the cost function has been
defined as the deviation of the measured inductor current iL TABLE I: Parameters used for FCS-MPC
from its reference iL,ref , which is an ideal sinusoid with an
Parameter Name Symbol Value
amplitude iRef . The expression of this cost function, J1 is Input AC Voltage (RMS) Vi 220 V
given by Phase-A Inductance La 3.962 mH
Phase-B Inductance Lb 1.870 mH
J1k = (iL,refk+1 − iLk+1 )2 . (11) Phase-C Inductance Lc 1.411 mH
Battery Nominal Voltage VBj 110 Vdc
J1k represents the cost at time instant k. Here, it is important Battery Capacity κBj 120 Ah
to note that for the positive and negative half-cycles of the Sampling Time Ts 25μ s
input AC voltage, the inductor current is predicted using (8)
and (9), respectively. In the second part of the cost function,
the deviation of battery SoC from the average SoC of the B. Finite Control Set MPC Algorithm
batteries is penalized. This cost function, J2 is given by
In finite control set model predictive control (FCS-MPC),
J 2k = (zBjk+1 − zavgk )2 (12) the dynamic behaviour of the discrete-time plant is predicted
j with every possible control input and the optimal control input
is chosen in order to achieve minimum cost. In the system
where j ∈[1,2,3] for positive half-cycle and j ∈[4,5,6] for
under consideration, there are six active control inputs, namely
negative half-cycle. Similarly, zavg represents the average SoC
q1 , q2 , q3 , q4 , q5 and q6 . Considering each qx can be either 0 or
of the respective batteries in positive and negative half-cycles.
1, intuitively there should be 26 = 64 combinations to evaluate
The final cost function J(k) can be acquired using (11) and
in FCS-MPC. However, this is not the case. As mentioned in
(12) as
Section II-B, only three switches decide the path of the current
Jk = ρ1 J1k + ρ2 J2k (13) in one half cycle i.e., q1 , q2 and q3 in positive and q4 , q5 and
q6 in the negative half cycles, respectively. This effectively
where ρ1 and ρ2 represent the unique weighting coefficients
reduces the possible combinations to 23 = 8 in each half-
for different control objectives in the cost function. These are
cycle.
non-negative scalars and are often tuned heuristically in order
Fig. 4 shows the flowchart of the proposed finite control
to achieve the best dynamic performance. The optimal control
set MPC algorithm. At time instant k, the inductor current
input U ∗ (k) is acquired by minimizing (13), and is given by
iL , the terminal voltages, VB1 , .., VB6 , and SoC, zB1 , .., zB6
Uk∗ = arg min Jk . (14) of the batteries are measured from the plant and the positive
40 100 30
iL
I B1 [A]
30 THD = 1.70 % 20 zB1 : 38%
20 iL,ref 90 10
i L [A], Vi [V]
0 (a)
10 Vi/10
0 80 0 2 4 6 8 10 12 14 16 18 20
I B2 [A]
-20 20 zB2 : 52%
-30 10
60 0 (b)
-40
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time [ms] 50
30
I B3 [A]
40 20 zB3 : 60%
10
Input Current [A]
26 0 (c)
30
0 2 4 6 8 10 12 14 16 18 20
24 20
1
q1 , q 2 , q 3
q1 q2 q3
10 0.5
22 Envelopes 0 (d)
0 0 2 4 6 8 10 12 14 16 18 20
3.5 4 4.5 5 5.5 6 6.5 0 1000 2000 3000 4000 5000
Time [ms]
Time [ms] (a) Frequency (Hz) (b)
Fig. 6: The battery currents in positive half-cycle (a) IB1 . (b) IB2 . (c) IB3 .
Fig. 5: (a) Inductor current reference tracking. (b) Fourier spectrum of the
(d) switching of S1d , S2d , and S3d .
AC line current.
or negative half-cycles is identified by the input AC voltage. other batteries. This ultimately translates into SoC equalization
Depending on the polarity of the AC voltage, the average SoC of the batteries by the end of the charging cycle. The control
of the respective batteries zavg is calculated. If zavg is less inputs q1 ,q2 , and q3 have been shown in 6(d). In the negative
than 99%, the cost function is evaluated for each of the eight half-cycle, the waveforms are fairly similar to Fig. 6 except
switching combinations using the appropriate discrete time for the variation in distribution of current according to the
equations given by (8)-(10). Once all the switching states have respective SoC of the batteries B4 , B5 and B6 .
been evaluated, the combination with the least value of the The state of charge of different batteries over a complete
cost function is chosen as the optimal switching combination. charging cycle has been shown in Fig. 7 for a current reference
During the charging operation, when zavg reaches 99%, the of 60 A. It can be seen that in the positive half-cycle, the SoC
inductor current reference iL,ref is gradually reduced as zavg of batteries B1 , B2 and B3 converges to a coherent line by
rises from 99% to 100% in order to prevent over charging of the end of the charging cycle. The same can be noticed in
the batteries. the case of B4 , B5 and B6 in the negative half-cycle. This
signifies that regardless the initial SoC of the batteries, the
V. R ESULTS proposed MPC algorithm will equalize their SoC by the end
In order to verify the effectiveness of the proposed control of the charging cycle. This aspect of the proposed control
algorithm, it has been applied to a virtual hardware developed algorithm is particularly useful when some batteries have
in MATLAB/Simulink. The parameters used for the simulation different capacities as compared to the others in the battery
are given in Table-I. A peak AC current reference of 25 A bank.
(iRef = 25 A) has been used in all the simulations unless Another case has been simulated for zB1 = 38%, zB2 = 52%,
stated otherwise. zB3 = 60% and an equal initial SoC of 80% for B4 , B5 , B6 ,
Fig. 5(a) shows the reference tracking of the inductor current respectively. It can be seen from Fig. 8(b) that the controller
iL with the proposed algorithm. The designed controller suc- supplies approximately the same average current to B4 , B5 and
cessfully tracks the reference current iL,ref . Moreover, as the B6 so as to ensure the equalization of SoC during charging,
reference current signal is phase-locked with the input voltage while the average current of B1 , B2 and B3 is distributed in
Vi , a successful current reference tracking also warrants unity accordance with their respective SoC.
input power factor. The lower graph in Fig. 5(a) shows the
magnified view of the inductor current. It can be seen that the VI. C ONCLUSION
inductor current tracks its reference with less than 5% ripple. In this paper, a model predictive control approach has been
Fig. 5(b) shows the fourier spectrum of the AC current ranging developed for a motor drive integrated charger considering a
from 0-5 kHz. The current exhibits 1.7% total harmonic three-phase split converter fed 12/8 switched reluctance motor.
distortion, which is below the maximum allowable standard The operation of the circuit has been described. The dynamic
values [21]. equations of the motor drive integrated charger and batteries
Fig. 6 shows the charging current of the batteries in positive have been developed. An FCS-MPC control strategy has
half-cycle with the proposed control scheme. It can be noticed been developed. The proposed control algorithm successfully
that the battery with the least SoC is given the maximum achieves the current reference tracking while maintaining the
average current so as to charge it faster as compared to the input power factor at unity. The proposed control also achieves
120 [4] O. Ellabban and H. Abu-Rub, “Switched reluctance motor converter
zB1 zB2 zB3 zB4 zB5 zB6 topologies: A review,” in Industrial Technology (ICIT), 2014 IEEE
110 International Conference on, pp. 840–846, IEEE, 2014.
[5] Y. Hu, X. Song, W. Cao, and B. Ji, “New sr drive with integrated
100 charging capacity for plug-in hybrid electric vehicles (phevs),” IEEE
Battery State of Charge (z) [%]
6
on Industrial Electronics, vol. 62, no. 10, pp. 6085–6095, 2015.
4 I B1 : z B1 38% [12] S. Chai, L. Wang, and E. Rogers, “A cascade mpc control structure for a
I B2 : z B2 52% pmsm with speed ripple minimization,” IEEE Transactions on Industrial
2 I B3 : z B3 60% Electronics, vol. 60, no. 8, pp. 2978–2987, 2013.
[13] H. Borhan, A. Vahidi, A. M. Phillips, M. L. Kuang, I. V. Kolmanovsky,
0 and S. Di Cairano, “Mpc-based energy management of a power-split
0 20 40 60 80 100 120 140 16 180 200
(a)
hybrid electric vehicle,” IEEE Transactions on Control Systems Tech-
Time [ms]
nology, vol. 20, no. 3, pp. 593–603, 2012.
8 [14] G. Kujundžić, Š. Ileš, J. Matuško, and M. Vašak, “Optimal charging of
valve-regulated lead-acid batteries based on model predictive control,”
Average Battery