You are on page 1of 15

IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO.

1, MARCH 2020 131

Robust LMI-LQR Control for Dual-Active-Bridge


DC–DC Converters With High
Parameter Uncertainties
Peizhou Xia, Haochen Shi , Student Member, IEEE, Huiqing Wen , Senior Member, IEEE,
Qinglei Bu , Student Member, IEEE, Yihua Hu , Senior Member, IEEE,
and Yong Yang , Member, IEEE

Abstract— This article presents an improved linear-quadratic


regulator (LQR) control based on linear matrix inequali-
ties (LMIs) to optimize the key parameters for the closed-loop
control of dual-active-bridge (DAB) converters with high system
uncertainty. First, the polytopic model of an uncertain DAB
converter is introduced according to the simplified DAB equiv-
alent circuit. LMIs are then used in the improved LQR control
to derive the optimized control parameters under the given
constraints. An improved LMI-LQR hybrid closed-loop control
is adopted with the output current introduced in the control
loop to enhance the dynamic performance. The performance
of the proposed LMI-LQR is compared with the conventional
LQR in terms of transient responses under various load and line
disturbances. Both the simulation and experimental results are
provided to validate the advantages of the proposed control.
Fig. 1. System diagram of the automotive dc DPS.
Index Terms— Dual-active-bridge (DAB) dc–dc converter,
dynamics, linear-quadratic regulator (LQR) control, robustness. devices (SSDs) such as batteries and supercapacitors to guar-
antee the normal operation of automotive dc DPS for the
I. I NTRODUCTION generated power fluctuation and the load changes. So far, many

I SOLATED bidirectional dc–dc converters (IBDCs) are


widely used as the intermediate energy transfer control
units between the dc bus and loads in the distributed power
IBDC topologies have been discussed, including two-device
topologies such as dual-flyback converter, dual-Cuk converter,
and dual-Zeta converter [3], [4], four-device topologies such
system (DPS), such as the automotive power system [1]. as forward-flyback converter, dual-push-pull converter, push-
In the dc DPS, IBDCs are becoming essential for the dc-bus pull-forward converter, push-forward-flyback converter, and
voltage stabilization, system stability, and efficiency due to dual-half-bridge converter [5]–[7], and eight-device topologies
their merits of reduced power conversion stages and easy such as dual-active-bridge (DAB) converter [8]. Among all
power flow regulation [2]. A system diagram of the auto- these IBDCs, DAB converters have the largest power transfer
motive power system structure is shown in Fig. 1, where capacity and the lowest filter volume. Besides, DAB converters
IBDCs will regulate the charge/discharge of energy storage exhibit other advantages, such as inherent soft switching,
modular design, high power density, and symmetric structure.
Manuscript received October 28, 2019; revised January 10, 2020; accepted
February 14, 2020. Date of publication February 20, 2020; date of current Thus, they are becoming preferred topology for many applica-
version March 20, 2020. This work was supported in part by the Research tions, such as solid-state transformer [9], electric vehicle [10],
Development Fund of Xi’an Jiaotong-Liverpool University (XJTLU) under and microgrid [11].
Grant RDF-16-01-10 and Grant RDF-17-01-28, in part by the Research
Enhancement Fund of XJTLU under Grant REF-17-01-02, in part by the Dynamics of DAB converters under complicated operating
Suzhou Prospective Application Programme under Grant SYG201723, and conditions, such as the input voltage variation, load change,
the XJTLU Key Programme Special Fund under Grant KSF-A-08, Grant start-up process, no load, and the output voltage or power
KSF-E-13, and Grant KSF-T-04. (Corresponding author: Huiqing Wen.)
Peizhou Xia, Haochen Shi, Huiqing Wen, and Qinglei Bu are reference change, become the most challenging design issues.
with the Department of Electrical and Electronic Engineering, Xi’an Regarding the control of DAB converters, the classical con-
Jiaotong-Liverpool University, Suzhou 215123, China (e-mail: peizhou. trol strategies, such as single-phase-shift (SPS) control [12],
xia14@alumni.xjtlu.edu.cn; haochen.shi@xjtlu.edu.cn; huiqing.wen@xjtlu.
edu.cn; qinglei.bu@xjtlu.edu.cn). dual-phase-shift control [13], trapezoidal control, extended-
Yihua Hu is with the Electronics Engineering Department, University of phase-shift (EPS) control [14], and triple-phase-shift (TPS)
York, York YO10 5DD, U.K. (e-mail: yihua.hu@york.ac.uk). control [15], are originated from the linear averaged model at
Yong Yang is with the School of Rail Transportation, Soochow University,
Suzhou 215131, China (e-mail: yangy1981@suda.edu.cn). a certain operation point by neglecting the high-frequency rip-
Digital Object Identifier 10.1109/TTE.2020.2975313 ples. However, strong nonlinear characteristics are commonly
2332-7782 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
132 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO. 1, MARCH 2020

observed in the practical dynamic performance of DAB con- complexity compared with the buck converter since totally
verters considering the devices’ voltage drop, switching tran- eight power devices should be controlled for DAB converters.
sients, dead time, saturation of the duty cycle, minimum phase Thus, the design of LQR controller for DAB converters
shift, and parasitics. becomes difficult. Furthermore, the classical LQR must be
Thus, two-category methods can be adapted to address the improved, especially its dynamic performance considering
nonlinear characteristics. One way is to improve the converter the applications for complicated isolated bidirectional power
model accuracy with the classical linear-average-model-based converters such as DAB converters with complicated circuit
controls. Another way is to propose the optimal control structure and high system uncertainty.
strategies to breakthrough the limitations of the classical The LQR controller is normally derived by a pole placement
control strategies. The former category covers many improved method, solving algebraic Riccati equations, or advanced
modeling methods. For instance, a bilinear discrete-time DAB intelligent optimization techniques. In [32], the weighting
model considering the parasitics and digital control delay was matrices of LQR controller are obtained by the pole placement
built to improve the stability in [16]. A new reduced-order method. In [33], the particle swarm optimization is used
average-value modeling for DAB converters was discussed to search for the best weighting factors in the quadratic
in [17] by considering the conduction losses and transformer cost function of LQR. Besides, the intelligent optimization
power losses. In [18], a discrete-time DAB model was pre- techniques, such as genetic algorithms (GAs), differential
sented by considering the input filter, output filter, and time evolution (DE), imperialist competitive algorithm (ICA), and
delays. The latter category methods include many optimal flower pollination algorithm (FDA), have been discussed for
controls that are able to get rid of the limitation of the classical the LQR controller [34]–[36]. However, these methods show
control strategies. For instance, a reference modification model obvious limitations in the determination key parameters of
with the feedforward control was presented to improve the LQR controller; the pole placement method is based on
transient response of DAB converters in [19]. A feedforward the trial-and-error, which cannot guarantee the optimal per-
with the coefficient optimization was discussed in [20] to formance for different operation conditions. The intelligent
achieve fast regenerative braking for DAB-based ac motor optimization techniques are quite complicated in the opti-
drive system. An active disturbance rejection control (ADRC) mization of weighting matrices. Furthermore, the performance
was discussed in [21], which is regarded as an advanced con- of the conventional LQR controller must be improved by
trol in addressing the disturbances and uncertainty. The sliding considering the uncertain parameter such as the input voltage
mode control in [22] is capable of improving the dynamic or load resistance. Otherwise, the uncertain parameters may
performance and attaining tight output voltage regulation. deteriorate the robustness of the closed-loop system when the
However, the computational burden of these methods is rela- load or input voltage is dramatically changed.
tive high. The predictive current mode controller is proposed Here, an improved LQR control based on linear matrix
to reduce settling time and dc offset current in [23], which is inequalities (LMIs) is presented in this article to address
more targeted on the load variation condition. Combined with the abovementioned concern and optimize the key control
SPS, the virtual current control in [24] and virtual direct power parameters for the DAB converters. In this article, the proposed
control were discussed in [25]. Similarly, combined with TPS, LMI-LQR control can extend to multiple plants by addressing
a power-balancing control was proposed in [26] to improve the the system uncertainty at different operation points and achieve
dynamic performance. Originated from the natural switching robust stability. Besides, the proposed control can cover dif-
surface (NSS) principle, a boundary control was proposed ferent design requirements such as pole placement restrictions
in [27] to achieve fast dynamic response for different transient and enhance the overall control performance. Finally, an LMI-
conditions. Furthermore, the NSS-based boundary control can LQR-based hybrid closed-loop control is implemented with
address the constant power load instability concern for the the output current introduced in the control loop to further
electric vehicle applications [27]. However, these strategies enhance the dynamic performance. The main simulation and
may result in unstable behavior or deteriorated performance experimental results are presented to validate the effectiveness
for large perturbations in the line and load. of the proposed LMI-LQR control.
As one optimal control, linear-quadratic regulator (LQR)
provides good robustness by minimizing a quadratic cost
II. P OLYTOPIC M ODELING OF U NCERTAIN
function or performance indexes such as the settling time,
DAB C ONVERTERS
overshoot during transient, and steady-state error in output
voltage of the buck converter with physical constraints such The operations of DAB converters are analyzed here, and
as the phase and gain margins and bandwidth [28]. Similar the polytopic model of an uncertain DAB converter is intro-
research works can be found and their results show that the duced according to the simplified equivalent circuit. The
LQR method can improve the dynamic response of dc–dc resistive load will be discussed here, the same as the research
converters with less response time, smaller overshot voltage, in [14] and [28]. For other types of loads such as inductive
and reduced transient power losses [28]–[31]. However, these load, a similar analysis can be made accordingly, which will
research works are mainly focused on nonisolated dc–dc be not discussed due to the limited space. The diagram
converters with one power device, such as the single-ended of DAB converters is shown in Fig. 2, which consists of
primary inductor converter (SEPIC), buck converter, or boost two symmetrical switching bridges connected through the
converter. However, the DAB converter shows higher control auxiliary inductor L and transformer. The typical operation

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
XIA et al.: ROBUST LMI-LQR CONTROL FOR DAB DC–DC CONVERTERS WITH HIGH PARAMETER UNCERTAINTIES 133

Fig. 4. Simplified DAB model.


Fig. 2. Schematic of DAB converters.

derived by combining (2) and (3):


2N V0 D − N V0 + V1
I L (t0 ) = − . (4)
4L f s
Combining (1)–(3), the general expression of the inductor
current within the half cycle (Ts /2) can be obtained as
⎧V + NV 2N V0 D − N V0 + V1


1 0
(t − t0 ) −

⎪ L 4L fs



⎪ (t < t < t )
⎨ 0 2
iL = (5)

⎪ V1 − N V0 2N V0 D + N V0 − V1



⎪ (t − t0 ) +

⎪ L 4L fs

(t2 < t < t4 ).
Thus, the input power of primaryside can be expressed as
2V1 t4
Pi = V1 · I1Avg = i L (t)dt. (6)
Fig. 3. Typical waveforms of DAB converters with the phase-shift control. Ts t0
According to (6), it can be seen that the input power Pi is
related to the input voltage V1 and average value of input
waveforms of DAB are shown in Fig. 3, which demonstrate current I1Avg . Assume that there are no power losses in
that the primary and secondary ac voltages VAB and v CD  are switching devices and transformers, and the output power has
square wave with 0.5 duty cycle, while the phase difference to be equal with the input power, Po = Pi . Then, based
between v AB and v CD is defined as D. It can be seen that the on inductor current equation in (5) and input power in (6),
operation of DAB converter can be further divided into four the output power can be expressed as
substructures, which cover the period from t0 → t2 , t2 → t4 , 
2V1 t4
t4 → t6 , and t6 → t8 . Po = Pi = i L (t)dt
The DAB converter can be analyzed for different substruc- Ts t0
tures according to Figs. 2 and 3. From t0 to t2 , the inductor N V1 V0
= · D(1 − D). (7)
current linearly increased, thus 2L f s
V1 + N V0
i L (t) = I L (t0 ) + (t − t0 ) . (1) Thus, the output current i 2 becomes
L
The phase shift is defined by the time between t0 and t2 . N V1
i2 = · D(1 − D). (8)
Therefore, the inductor current at time t2 is expressed by 2L f s
V1 + N V0 DTs According to the expression of the output current as shown
I L (t2 ) = I L (t0 ) + · (2) in (8), a current-source model is built. Specifically, the current
L 2
where D is the phase-shift ratio between primary and source will inject power to the output capacitor and load, and
secondary sides, Ts represents the switching period, and Fig. 4 shows the simplified model for DAB converters.
t2 − t0 = DTs /2. According to the Kirchhoff voltage law, the dynamic aver-
Applying a similar analysis to the period between t2 and t4 age model of DAB converters can be expressed by
and considering that t4 − t2 = (1 − D)Ts /2, the inductor dv 0 1 N V1
current at time t4 can be obtained as C0 = − v0 + · D(1 − D). (9)
dt R 2L fs
V1 − N V0 Since all these dynamic variables contain the dc components
I L (t4 ) = I L (t2 ) + · (t4 − t2 )
L and ac small-signal components, the average output voltage
V1 − N V0 Ts and duty cycle can be expressed as
= I L (t2 ) + · (1 − D) . (3)
L 2
Considering that the symmetry of the inductor current, v 0 = Ve + ṽ 0
I L (t0 ) = −I L (t4 ), the following expression of I L (t0 ) can be D = De + d̃ (10)

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
134 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO. 1, MARCH 2020

where Ve and De represent the dc component of the reference where G k is a known matrix, composed by [ Ak , Bk ], and λk
output voltage and phase-shift duty cycle, respectively, and is the uncertain parameter.
ṽ 0 and d̃ represent the ac component of the output volt- In the DAB converter, two major uncertain elements are
age and phase shift, respectively. Combining (9) with (10), considered, which are the input voltage V1 and load R L .
the small-signal model of DAB converters can be obtained as Other elements in the DAB converter are assumed to be
d(ṽ 0 ) ṽ 0 N V1 N V1 2 clarified. Thus, n p = 2 and parameter vector p = [1/R, V1 ].
=− + (1 + 2De )d̃ − d̃ . (11) Furthermore, these two parameters are limited within the
dt RC0 2LC0 f s 2LC0 f s
following boundaries:
Since the ac component of phase shift d̃ is much smaller
1/R ∈ [1/Rmax , 1/Rmin ] , V1 ∈ [V1 min , V1 max ] . (17)
than the reference phase shift De , (1 + 2De )d̃ is much larger
than d̃ 2 . Thus, the square of ac component d̃ 2 can be omitted The defined polytopic model has NT = 2n p vertices that
so that the small-signal model can be simplified as contain all possible values of uncertain matrices, and the
d(ṽ 0 ) ṽ 0 N V1 vertices (boundaries) can be described as follows:
=− + (1 + 2De ) d̃. (12) ⎡ ⎤ ⎡ ⎤
dt RC0 2LC0 fs 1 N V1 min
− 0 (1 + 2D )
In order to achieve the zero steady-state error, a state A1 = ⎣ Rmax C0 ⎦ B1 = ⎣ 2LC0 f s e ⎦
 −1
˜ = − v˜0 dt is introduced, which stands for the
variable x(t) 0 0
⎡ ⎤ ⎡ ⎤
integral of output voltage error. Thus 1 N V1 max
− 0⎦ (1 + 2De )⎦
  A2 = ⎣ Rmin C0 B2 = ⎣ 2LC0 f s
x 1 (t) ṽ 0 −1
x(t) = = . (13) 0 0
x 2 (t) − ṽ 0 dt
A 3 = A 2 B3 = B1
Combining the output voltage expression (12) and the small A 4 = A 1 B4 = B1 . (18)
signal of DAB (9), the following differential equation can be
obtained as: This model will be used in Section III to build an LQR
 controller for DAB converters and limit the upper or lower
ẋ 1 (t)
ẋ(t) = = Ax(t) + Bd(t) (14) band of LQR controller. If all the uncertainty was bounded
ẋ 2 (t)
within the vertices of this polytopic model, the system stability
where A and B are the state-space matrices that can be can be guaranteed with the LQR controller.
expressed by
⎡ ⎤ ⎡ ⎤
1 N V1 III. P ROPOSED LMI-LQR C ONTROL
− 0 (1 + 2D )
A = ⎣ RC0 ⎦ B = ⎣ 2LC0 f s e ⎦
. (15) In this section, the principle of LMI is introduced, and
−1 0 0 then, the LQR controller is formulated into the form of LMI.
Because the state-space matrices A and B affect the Finally, an optimal LMI-LQR algorithm is presented for the
dynamic response of DAB converters, it is important to DAB converter.
analyze the characteristics of A and B in the design of
the closed-loop controller. For a designed DAB prototype, A. Quadratic Stability of an Uncertain Plant
normally, these parameters, such as L, C0 , f s , and N, can be Given a linear time-invariant system with the expression as
regarded constant, while the load resistance R and the input
voltage V1 are uncertain, which will change with respect to ẋ = Ax + Bu. (19)
the transmission power and the voltage conversion ratio.
By introducing a state-feedback gain K with the expression
Thus, the state-space matrices A and B will be influenced by
uncertain variables, including the load or input disturbances. of “u = −K x” into the transfer function (19), the state
Here, a vector p is introduced by grouping all these uncertain function of the closed-loop control system can be expressed as
terms altogether, p = ( p1, p2 , . . . , pn p ). Each uncertain term ẋ = Ax + Bu = (A − B K )x. (20)
pi will be constrained between the minimum value pmin and
the maximum value pmax . According to the Lyapunov theory, there exists a matrix P
Generally, the possible values within vector p are bounded with the quadratic function as
within a hyperrectangular with NT = 2n p vertices in the V (x) = x T Px > 0 ∀x = 0. (21)
parameter space Rn p . The system matrices A and B for
each vertex in this hyperrectangular correspond to the set Furthermore, the necessary and sufficient condition that
{G 1 , . . . , G k }, which is a polytopic model, and this model can assure the system quadratically stable is to satisfy the
will be convexly combined by matrices finite vertices with inequality as
their values determined by the matrix polytopic model. Thus
V̇ (x) = x T ((A − B K )x)T Px + x T P((A − B K )x))
[ A( p), B( p)] ∈ Co{G 1 , . . . , G n } = x T ((A − B K )T P + P(A − B K )x) > 0 ∀x = 0
NT
 NT
 (22)
:= λi G i , λi ≥ 0, λi = 1 (16)
i=1 i=1 P > 0 indicates that the matrix P is positive definite.

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
XIA et al.: ROBUST LMI-LQR CONTROL FOR DAB DC–DC CONVERTERS WITH HIGH PARAMETER UNCERTAINTIES 135

Assume that there is a matrix P > 0 that satisfies the such that
following expression:
AP + PA T + BY + Y T B T + x˜0 x˜0 T < 0. (28)
V̇ (x) = x ((Ai + Bi K ) P + P(Ai + Bi K ))x > 0
T T
Now, the inequality in (28) is homogeneous. Furthermore,
∀x = 0 ∀i = 1, 2, . . . , N (23) K = Y P−1 will not change with u. Therefore, if matrix A x̃ 0
where Ai is the vertex of the polytopic model as analyzed in is controllable, LMI can be expressed as
Section II. Therefore, expression (23) can be used to judge AP + PA T + BY + Y T B T + I < 0. (29)
whether the system within this uncertain range are quadrat-
ically stable or not. In another way, an appropriate matrix Thus, when u > 0, for arbitrary matrices P and Y that
variable P should be determined in order to guarantee the satisfy (28), both uP and uY can automatically fulfill (29).
quadratic stability of the system. Since the variables discussed Here, a second auxiliary variable X is introduced so that the
nonlinear term Tr(Rw Y P−1 Y T Rw ) can be replaced, while
1/2 1/2
in this article are linearly correlated, they can be solved by the
convex optimization. Schur’s complement can be used to simplify the inequality
constraint. Then, one target of performance index can be
rewritten as
B. LMI Derivation min X
For the system described by (14), an LMI formulation X

solution for LQR problem can be adopted. In order to evaluate such that
⎡ ⎤
the quadratic stability of proposed linear system with the 1 1
1
2
X Rw Y ⎦
closed-loop control (20), a quadratic cost function can be X > Rw Y P−1 Y T Rw ↔ ⎣
2 2
1 > 0. (30)
expressed as Y T Rw2 P
 ∞
As a result, the complete LMI formulation of this LQR
J= (x̃ T (Q w + K T Rw K )x̃)dt (24)
0 problem can be expressed as
where Q w is a symmetric semidefinite matrix, Rw is a min Tr (Q w P) + Tr (X)
symmetric definite matrix, and K is state-feedback gain of P,Y,X
closed control system. Besides, [ A Bu ] must be controllable. subject to
Thus, LQR can be considered as the weighted minimization
AP + PA T + ⎡
BY + Y T B T +⎤I < 0
process for a linear combination of the state variable x̃ and 1
input variable ũ. The weighted matrix Q w describes which X R 2
w Y⎦
⎣ 1 > 0, P > 0. (31)
state variable has to be controlled more strictly than others, T 2
Y Rw P
and this matrix can be a diagonal matrix with each weighted
value decided by the importance of matching physical state Once the solution is obtained from this minimizing process,
value [28]. The matrix Rw determines the actual control the optimal LQR controller can determine K from K = Y P −1 .
behavior, which is mainly dependent on the state deviation. Thus, the LQR control can be formulated as the process in
In order to find the state-feedback gain K in (24), there solving a convex optimal problem.
should be a matrix P that can ensure the following condition: According to the polytopic model of DAB converters and
d T combining (18) with (31), the complete LMI formulation of
(x Px) = −x̃ T (Q w + K T Rw K )x̃. (25) this LQR controller can be obtained as
dt
min Tr(Q w P) + Tr(X)
P,K ,X
Now, a trace operator Tr(·) is introduced, which is the
sum of all elements on the main diagonal of an n × n subject to
matrix. This trace operator satisfies a T Xb = Tr(Xba)T , A1 P + PA1T + B1 Y + Y T B1T + I < 0
and then, the performance index with closed-loop control is
equivalent to A2 P + PA2T + B2 Y + Y T B2T + I < 0
A3 P + PA3T + B3 Y + Y T B3T + I < 0
J = Tr((Q w + K T Rw K )P) (26)
∞ A4 P + PA4T + B4 Y + Y T B4T + I < 0
⎡ ⎤
where matrix P = 0 (x̃ x̃ T )dt is positive definite, and 1
2
satisfies ⎣ X R w Y ⎦ > 0, P > 0.
1 (32)
T Y  Rw2 P
(A + B K )P + P(A + B K ) + x˜0 x˜0 = 0
T
(27)
According to the expressions of the LMI solution for LQR
where x˜0 is the initial value of state variable x˜i . In order to controller shown in (31) and (32), the optimal feedback gain
eliminate the multiplication term of K P, the positively defined K for DAB converters must be determined by considering
matrix Y = K P is introduced. Then, the optimal feedback gain high system uncertainty, especially the input voltage V1 and
K can be found by minimizing the following expression: the load resistance R. Thus, in order to enhance the robustness
 1 1
performance, the parameters’ uncertainty, including the input
min Tr(Q w P) + Tr Rw2 Y P−1 Y  Rw2
P,Y voltage V1 and the load resistance R, must be restricted by

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
136 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO. 1, MARCH 2020

Fig. 5. Diagram of the conventional single-PI LMI-LQR control for DAB


converters.

the polytopic model through A1 – A4 and B1 –B4 . Section III-C Fig. 6. Diagram of the dual-PI LMI-LQR closed-loop control for DAB
demonstrates three controller designs based on parameters converters.
proposed by LMI-LQR in order to find the most sufficient
control. The detailed calculation of K will be discussed in
Section III-D.

C. Proposed LMI-LQR Closed-Loop Control


In order to show the advantages of the proposed LMI-LQR
closed-loop control, three LMI-LQR closed-loop control algo-
rithms will be discussed here, and then, a comparative inves-
tigation will be conducted in the following sections.
Single-PI LMI-LQR Closed-Loop Control: The
proportional-plus-integral controller (PI) controllers are Fig. 7. Diagram of the proposed LMI-LQR based closed-loop control for
DAB converters.
widely utilized due to their simplicity in implementation. The
adoption of LQR is aimed at the optimization of PI controlling
parameters to enhance the performances, such as transient
voltage v˜0 and closed-loop controller K 1 and K 2 . The actual
response, tracking, and stability. However, the PI controller
phase-shift duty cycle is obtained by: D = De + d̃.
using parameters from LQR still shows the limitations in
Dual-PI LMI-LQR Closed-Loop Control: Considering the
terms of the control precision, which may lost stability under
possible advantages of dual-PI controller [37], for a compara-
system uncertainty [21]. Considering the advantage of the
tive analysis, the dual-PI LMI-LQR control is presented, where
proposed LMI-LQR, as indicated in the previous analysis for
the integrated control elements are used for both the output
the polytopic model and LMI derivation, here the single-PI
voltage and output current control. Then, the feedback gain is
based LMI-LQR control is first considered.
expressed as K = [K 1 K 2 K 3 K 4 ]. The corresponding control
Since the system matrix for DAB converter in (14) contains
 diagram is shown in Fig. 6.
a proportional element x and an integrated element x,
The overall closed-loop control for the dual-PI LMI-LQR
the optimal state-feedback gain matrix K based on (31) can
closed-loop control can be expressed as
be expressed as

K = [K 1 K 2 ] (33)
d̃ = u = K 1 (v e − V0 ) + K 2 (v e − V0 )
where K 1 is the closed-loop gain for proportional element x 
and K 2 is the closed-loop gain for integral element x. Thus, +K 3 (i e − I0 ) + K 4 (i e − I0 ) (35).
the diagram of the conventional single-PI LMI-LQR control
for DAB converters is shown in Fig. 5.
In Fig. 5, De represents the dc component of the phase-shift Proposed LMI-LQR Closed-Loop Control: In order to
duty cycle, which is calculated by the reference voltage Ve and realize the rated output voltage with zero steady error
reference current Ie . The specific expression in Fig. 5 for the and improved dynamic performance for DAB converters,
derivation of De is shown as follows: an improved LMI-LQR closed-loop control is implemented;
the LMI-LQR control is adopted to guarantee the stability
N V1 Ve
Ie2 = De (1 − De ). (34) of the whole system, and the load current disturbance is
2RL fs introduced to enhance the load step-change response. The
The reference of phase shift De is derived from (7). When proposed LMI-LQR closed-loop control diagram is shown
the power flow direction is set from the primary side to in Fig. 7.
secondary side, De has lower and upper limit at 0 and 1. In Fig. 7, an additional proportional feedback gain K 3
A similar analysis can be conducted when the power flow is used for the added current loop controller in the pro-
direction is opposite, and the difference is that De , in that posed LMI-LQR hybrid control system that can enhance the
case, will be bounded within −1 and 0. The ac component of load step-change response. The optimal feedback matrix is
the phase shift d̃ is calculated by the ac component of output expressed by K = [K 1 K 2 K 3 ].

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
XIA et al.: ROBUST LMI-LQR CONTROL FOR DAB DC–DC CONVERTERS WITH HIGH PARAMETER UNCERTAINTIES 137

TABLE I
DAB C ONVERTER PARAMETERS

Fig. 8. Control diagram.

D. Controller Design for the Proposed LMI LQR


For different LMI-LQR control strategies, the robustness of
closed-loop control is determined by the state-feedback gain of
integral and proportional V0 component. The proportional and
integral gain of output current I0 is mainly used to improve the
dynamic response under load changing condition. The main
challenge in the design of the proposed LMI-LQR control for
DAB converters is to guarantee the robust stability with respect
to the defined performance indexes, such as the steady-state
error, transient oscillations, settling time, and robustness for
a wide range of perturbations. The main parameters of DAB
converters are summarized in Table I. During the optimization,
the phase margin should be no less than 60◦ and the gain
margin should be no less than 10 dB while maintaining the
overshoot less than 25% of nominal output value.
The weighted variable −3matrix is set as
1e 0
Qw = Rw = 1. (36)
0 1e7
By solving the LMI derivation result of LQR provided
at (32), the feedback gains are calculated as K 1 = −0.1153
and K 2 = 16.5567 according to the “mincx” solver. Finally,
the optimal closed-loop control parameters are expressed as
K LMI−LQR = [−0.1153 16.5567]. (37)

For the conventional LQR controller, the gain K LQR can be


obtained by directly using “lqr” command in MATLAB as
K LQR = [−0.22 33]. (38) Fig. 9. (a) Root locus for a fixed K 1 . (b) Root locus for a fixed K 2 .

Once the control parameters are determined, the root locus


to 16.5567 can keep the system stable. Fig. 9(b) shows that
method can be used to robustness and stability performance of
K 1 = 0.11 is appropriate when K 2 is fixed at 16.5567.
the controller. The DAB converter is a classical single–input-
The “step” function is used to measure the settling time and
single–output (SISO) system, and using the Laplace transform,
the overshoot of the whole closed-loop system with the set
the small-signal model of (12) can be expressed as
parameters. The measured results include: the settling time
1 N V1 ts = 0.0195 s and overshoot is 0, which verify the correctness
sV(s) = − V(s) + X (s) (39) of the parameters.
RC0 2LC0 f s
where X (s) represents the phase-shift duty cycle d̃(t) in (12), IV. S IMULATION E VALUATION
which is the input variable of this SISO system: Then,
the transfer function is derived as In order to validate the effectiveness of the proposed control,
a DAB converter simulation model was built. For the simu-
V (s) N V1 1 lation, the input voltage range is changing from 50 to 100 V
H (s) = = (1 + 2De ) and the load resistance is changing from 20 to 60 . The dead
X (s) 2LC0 f s s + RC
1
0
time for two devices in the same bridge of DAB converters
75
= . (40) is 200 ns in order to avoid the shoot-through fault. Although
2.2 × 103 s + 1 the DAB converter has the bidirectional power transmission
Thus, the control diagram can be shown in Fig. 8. capability, here only the forward power transmission with the
The root locus method is used to analyze the effect of power from the input port V1 to the output port V0 is presented
two parameters K 1 and K 2 . Fig. 9(a) shows the root locus considering the symmetrical structure of the DAB converters.
occurs when K 1 is fixed at 0.1153. It shows that K 2 equal With the forward operation mode, the phase shift D is set

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
138 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO. 1, MARCH 2020

Fig. 11. Various closed-loop control scheme with LMI-LQR controller under
load step change from 20 to 60  at t = 0.01 s, and back to 20  at t = 0.06 s:
the proposed LMI-LQR closed-loop control (red line), the single-PI LMI-LQR
closed-loop control (green line), and the dual-PI LMI-LQR closed-loop control
(blue line).

The results verify the complete operations of the DAB con-


verter under the proposed LMI-LQR controller.
Before comparing the LMI-LQR controller with conven-
tional LQR controller, three controlling schemes discussed
in Section III have to be evaluated under the same transient
condition. Specifically, the load is changing from 20 to 60 
at t = 0.01 s, and back to 20  at t = 0.06 s. In Fig. 10(b),
the duty ratios of three controllers under load changes are sim-
ulated. Dual-PI controller exhibits the best dynamic response
among three schemes since it requires the lowest transient
Fig. 10. When input voltage is fixed at 50 V, the output voltage waveform time to reach steady state [indicated by the black dotted line
v 0 and the corresponding transient duty ratio d of DAB converter under
(a) simulated steady-state voltage at primary side v AB , secondary side v CD , in Fig. 10(b)]. Moreover, the proposed controller has lower
and the inductor current i L . (b) Transient phase-shift ratio D under three transient time than the single-PI controller. At t = 0.15 s,
controlling topologies. the duty ratio of system under single-PI controller even cannot
reach steady-state value of De = 0.4. Thus, in terms of the
dynamic response, the proposed scheme is better than the
within [0, 1]. For the backward mode, a similar analysis can be single-PI scheme while slightly less than the dual-PI scheme.
applied and the only difference is that the phase shift D should Fig. 11 shows that the transient time and the peak voltage
be set within [−1, 0]. In the dynamic performance evaluation, during the load-step change process by using the single-PI
the scenarios of the step change in load R and input voltage LMI-LQR with the diagram shown in Fig. 5 are larger
V1 are considered. Both the proposed LMI-LQR controller than those by using the proposed LMI-LQR-based hybrid
and the conventional LQR controller are implemented and closed-loop control. It proves that the additional proportional
their performance can be compared. For the conventional LQR current feedback gain in the proposed closed-loop control can
controller, the gain K LQR can be obtained by directly using ensure better dynamic response under load changing condition.
“lqr” command in MATLAB as Fig. 11 also shows that the steady-state error of the proposed
K LQR = [−0.22 33 0.12]. (41) LMI-LQR control is smaller than that of the dual-PI LMI-LQR
control although the proposed LMI-LQR control shows a sim-
The third parameter K 3 = 0.12 is the proportional gain ilar transient time and relatively larger voltage peak. Further-
from the current loop. According to Section 7, an extra more, the steady-state voltage oscillations are much smaller
proportional control of current could enhance the dynamic by using the proposed LMI-LQR closed-loop control, which
performance. The determination of K 3 can refer back to the is beneficial to the system stability. For the dual-PI LMI-LQR
single-PI controller. Through simulation tests, K 3 is chosen control, although the additional current PI controller enhances
as 0.12. Furthermore, based on derivation from Section III-D, the dynamic response, it will also introduce additional insta-
the controlling parameters of LMI-LQR controller will be bility issue that affects the steady-state operation of DAB
set as converter. According to the simulation results, the conclusion
K LMI−LQR = [−0.1153 16.5567 0.12]. (42) can be drawn that the proposed LMI-LQR hybrid control
can achieve a good balance between the dynamics and the
To begin with, the proposed LMI-LQR is implemented steady-state performance.
in MATLAB/Simulink and its effectiveness is evaluated. Fig. 12 shows the main simulation results by using the two
Fig. 10(a) shows the simulated steady-state waveforms, includ- LQR controllers under the conditions of the step change in
ing the voltage of v AB and v CD , and the inductor current i L . load R or input voltage V1 . Both the simulated output voltage

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
XIA et al.: ROBUST LMI-LQR CONTROL FOR DAB DC–DC CONVERTERS WITH HIGH PARAMETER UNCERTAINTIES 139

represent the results under the conventional LQR controller.


Fig. 12(a) shows the differences in terms of the dynamics by
using the two controllers when the load step change from
20 to 40  at t = 0.01 s. Obvious oscillations can be
observed in the simulated waveforms of v 0 and i 0 by using
the conventional LQR controller. The settling time under the
conventional LQR control is longer than that of the proposed
LMI-LQR control. For the load step change from 40 to 20 
at t = 0.06 s, the observed performance with two controls
is similar in terms of the settling time. For a larger step
change in the load between 20 and 60 , as illustrated by
the blue curves of v 0 and i 0 in Fig. 12(b), the oscillations of
DAB converter under the conventional LQR control becomes
more serious. It maintains continuous oscillation before the
load returns to 20  at t = 0.06 s. However, as indicated
by the red curves in Fig. 12(b), the proposed LMI-LQR
controller still works correctly with fast dynamics. Thus,
the conclusion can be drawn that the proposed control can
guarantee the stability, robustness, and satisfied performance
of DAB converter even for large perturbations in the load.
Fig. 12(c) shows the simulated results when the input voltage
experiences a step change of 50% at t = 0.01 s, namely from
V1 = 50 V to V1 = 75 V. It indicates obvious oscillations
by using the conventional LQR control and the settling time
is close to 0.04 s, while the proposed LMI-LQR can make
the transient dynamic process much faster, which is less than
0.02 s. At t = 0.06 s, the input voltage returns back to
V1 = 50 V, and the dynamic performances of two controls
are quite similar. For a larger step change in the input voltage
between 50 and 100 V, as illustrated by the blue curve of v 0
in Fig. 12(d), the DAB converter under the conventional LQR
control shows significant oscillations during the whole period
from t = 0.01 s to t = 0.06 s with a higher input voltage of
100 V. However, the red curve of v 0 in Fig. 12(d) indicates that
the proposed LMI-LQR can regulate the transient dynamics
process well and the transient time is observed less than
0.02 s. At t = 0.06 s, the input voltage changes from
V1 = 100 V to V1 = 50 V, and the proposed LMI-LQR
also exhibits a satisfied transient performance for this large
step-down change condition. Small oscillations can still be
observed with the proposed LMI-LQR from the simulated
output voltage curve. Thus, with larger perturbations in the
input voltage, the advantages of the proposed control in terms
of the robustness and dynamics improvement become more
obvious.
Fig. 13 shows the stability and transience performances of
DAB converters with different controls when the reference
voltage changes from 50 to 40 V. Fig. 13(a) and (b) shows the
Fig. 12. Comparison between the proposed LMI-LQR controller (red line) simulated waveforms of the output voltage v 0 , which indicates
and the conventional LQR controller (blue line) for different transients. that both controllers can achieve a fast response of DAB con-
(a) Load step-change from 20 to 40  at t = 0.01 s, and back to 20 
at t = 0.06 s. (b) Load step change from 20 to 60  at t = 0.01 s and back
verters with respect to the reference voltage change and remain
to 20  at t = 0.06 s. (c) Input voltage step change of 50% at t = 0.01 s, stable at the new reference voltage. However, the proposed
returns to nominal state at t = 0.06 s. (d) Input voltage step change of 100% LMI-LQR controller can realize the transients smoothly with
at t = 0.01 s returns to the nominal state at t = 0.06 s. fewer oscillations. Fig. 13(a) and (b) shows the simulated
waveforms of the phase-shift ratio D, which is calculated as
v 0 and the load current i 0 are provided for the comparison. 0.24 according to (8). As shown in Fig. 13(c) and (d), both
The red curves represent the outputs of DAB converters under controllers show the same phase-shift ratios for the steady
the proposed LMI-LQR controller, whereas the blue curves states, including before and after the step-change instant.

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
140 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO. 1, MARCH 2020

Fig. 14. Experimental platform.

TABLE II
C OMPARISON OF C ALCULATION B URDEN

DAB converter that has the main parameters listed in Table I.


The current and voltage sensor has been integrated on the
prototype of DAB converter to measure the output current
and voltage in real time. The proposed control algorithm
was implemented by controller DSP28335 where the related
closed-loop control parameters K 1 –K 3 were introduced in
code of DSP for adapting different control schemes. In order to
evaluate the performance of the traditional LQR controller, we
proposed LMI-LQR control with uncertain parameters such as
the input voltage or the load condition in Table I. Specifically,
the input voltage range for the experiments is set from 50 to
100 V and the load resistance is changing from 20 to 60 .
The programmable electronic load, IT8514C+, was used for
the load change experiments, whereas programmable power
supply IT6514C was used for the input voltage variation.

B. Executing Time Comparison


Executing time for different closed-loop controls is shown
in Table II. The simplest single-PI controller for output voltage
has lowest executing time that is equal to 7.16 μs. Because the
conventional LMI LQR and the proposed LMI-LQR control
Fig. 13. Simulated results when reference changes from 50 to 40 V under
(a) conventional LQR controller and (b) proposed LMI-LQR controller. The are used the same hybrid optimized closed-loop control but
phase-shift ratio of DAB converters with the changing reference voltage under different state-feedback gains. Therefore, it has same cal-
(c) conventional LQR controller and (d) proposed LMI-LQR controller. culation burden as 7.21 μs, which is slightly larger than
closed-loop control due to additional proportional current gain.
However, a much faster transient performance is exhibited by The dual closed-loop control needs the largest executing time
using the proposed LMI-LQR control. Furthermore, with the as 7.42 μs. For all closed-loop control, the executing time
proposed LMI-LQR control, the oscillations in the phase-shift of these controllers is quite similar. All these schemes can
ratio are reduced. Fig. 13 verifies that the proposed control be easily implemented in DSP28335 that exhibits the highest
can ensure the robustness and improve the dynamic perfor- computational frequency at 150 MHz.
mance for the closed-loop control with the reference voltage
uncertainty.
C. Load Change Transient
The experimental results by using different LMI-LQR con-
V. E XPERIMENTAL R ESULTS
trollers for a large load perturbation are shown in Fig. 15. The
A. Experimental Setup control diagram of three closed-loop controls for DAB con-
In Fig. 14, the experimental platform was built to verify verters, including the conventional single-PI LMI-LQR con-
the results derived in Section III. It consists of a 200-W trol, dual-PI LMI-LQR control, and the proposed LMI-LQR

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
XIA et al.: ROBUST LMI-LQR CONTROL FOR DAB DC–DC CONVERTERS WITH HIGH PARAMETER UNCERTAINTIES 141

Fig. 16. Comparison of experimental results under load perturbation changes


from 20 to 40  and returns to 20 . (a) Single-PI LMI-LQR control.
(b) Proposed LMI-LQR control.

LMI-LQR controller and the proposed LMI-LQR controller


are found effective in realizing the stable operation of the pro-
totype for the load change. However, as shown in Fig. 16(a),
the system under the conventional LQR controller shows
poor damping performance and requires more time to reach
the steady state. The measured settling time by using the
conventional LQR control is close to 0.03 s. With the proposed
Fig. 15. Experimental comparison of three LMI-LQR controls under load LMI-LQR control, the measured settling time becomes much
perturbation changes from 20 to 60  and returns to 20 . (a) Single-PI
LMI-LQR control. (b) Dual-PI LMI-LQR control. (c) Proposed LMI-LQR shorter, which is around 0.02 s. Figs. 15 and 16 show that the
control. transient time with the proposed LMI-LQR control remains
similar regardless of the load perturbation range, which verifies
the unique merit of the proposed LMI-LQR control in achiev-
control, is shown in Figs. 5–7, respectively. During the test,
ing the robustness and stability of DAB converters. Fig. 17
the load resistance was jumped from 20 to 60  and finally
shows the start-up process of DAB converter by using the
returns to 20 . With the conventional single-PI LMI-LQR
conventional LQR control and the proposed LMI-LQR control.
control, big oscillations can be observed for the whole period
It can be seen that the dynamic performance of two controllers
from t = 0.01 s to t = 0.05 s with a higher load resistance
is almost the same at the start-up process.
of 60 , as shown in Fig. 15(a). However, the proposed
LMI-LQR control can ensure the stability of the system within
0.02 s, as shown in Fig. 15(c). The experimental results D. Input Voltage Change Transient
match the simulation results well. The dual-PI closed-loop Fig. 18(a) and (b) shows the experimental results of two
control shows less transient time around 0.01 s than the controllers when the input voltage experiences a step change
proposed control, as shown in Fig. 15(b). However, it also of 50%, specifically from 50 to 75 V, as shown by the purple
has a larger voltage fluctuation and obvious steady-state error. curves. Due to the limitation of dc power supply used in the
It proves that the addition PI controller for the output current experiment, it is unable to achieve the negative step changes
may introduce additional instability issue, which validates of voltage. Therefore, only positive step-change condition is
the theoretical analysis. The results show that the proposed tested in the experiment compared with the simulation results.
LMI-LQR has the most balanced performance among the three In addition, the input voltage cannot instantly change from
methods. 50 to 75 V due to the limited dv/dt output capability of
Fig. 16(a) and (b) shows the experimental results for a small dc power supply. Fig. 18(a) shows that the overshoot of the
load perturbation. Specifically, the load resistance was jumped conventional LQR control is around 5 V. However, obvious
from 20 to 40  and finally returns to 20 . Both the single-PI oscillations with an amplitude of 3 V in the output voltage

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
142 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO. 1, MARCH 2020

Fig. 17. Start-up process of DAB converters under (a) conventional LQR
controller and (b) proposed LMI-LQR controller.

can be observed, which results in the final settling time that


is around 0.024 s. Compared with the conventional LQR
control, the proposed LMI-LQR control can quickly suppress
the oscillation and ensure the stability of the output voltage
within 0.01 s. Therefore, the proposed LMI-LQR control
shows good dynamic performance and less transient time than
the conventional LQR control. When the input voltage expe-
riences a larger step, specifically from 50 to 100 V as shown
in Fig. 18(c) and (d), more serious oscillations are observed in
the output voltage waveform by using the conventional LQR
control, which requires at least 0.035 s to suppress the voltage
ripple. Although the settling time of the proposed LMI-LQR
control is also increased with larger step changes of the input
voltage which is around 0.017 s, it still exhibits better dynamic
performance than the conventional LQR control. Thus, it is
clear that the proposed LMI-LQR control can successfully
ensure less transient time and oscillation period than the
conventional single-PI LMI-LQR control, especially in the
larger input voltage variation condition.
Experimental results of two controllers under step-down Fig. 18. Main experimental results under (a) and (c)conventional LQR control
change of the input voltage from 75 to 50 V are shown when the input step changes from 50 to 75 V and (b) and (d) proposed
LMI-LQR control when the input step changes from 50 to 75 V.
in Fig. 19. Fig. 19(a) shows that the output voltage of the con-
ventional LQR control is dropped around 5 V, and the settling
time is around 0.012 s. The proposed LMI-LQR control can
ensure the stability of the output voltage within 0.006 s. There- experiences a step-down change from 50 to 40 V. As shown
fore, the proposed LMI-LQR control shows good dynamic in Fig. 20, both the conventional LQR control and the proposed
performance and less transient time than the conventional LQR LMI-LQR control achieve fast tracking performance with
control. respect to the reference voltage change and retain stable
at the new reference voltage. The transient time by using
the conventional LQR control is 0.015 s, and the multiple
E. Output Reference Change Transient oscillations are observed during the transient moment. Then,
Fig. 20 compares the experimental results by using two the experimental waveform of the output voltage by using
LMI-LQR controls when the output reference voltage Ve the proposed LMI-LQR control exhibits smooth transient with

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
XIA et al.: ROBUST LMI-LQR CONTROL FOR DAB DC–DC CONVERTERS WITH HIGH PARAMETER UNCERTAINTIES 143

VI. C ONCLUSION
A novel approach for controlling bidirectional isolated DAB
converters digitally using state-feedback techniques and LMI
technique has been proposed. According to the simplified DAB
equivalent circuit, the polytopic model is introduced, which
can consider the parameter uncertainty, whereas the LMI tech-
nique can guarantee the system robust stability. Consequently,
an LMI-LQR-based hybrid control is derived with the output
current introduced in the closed-loop control to enhance the
dynamic performance. Three LMI-LQR closed-loop controls
for DAB converters have been discussed, including the conven-
tional single-PI LMI-LQR control, dual-PI LMI-LQR control,
and the proposed LMI-LQR control. Different dynamic scenar-
ios, including the load step change, input-voltage step change,
start-up process, and the output reference voltage change, are
considered to evaluate the performance improvement of the
proposed LMI-LQR control. Both simulation and experimental
results under various scenarios are presented to validate the
advantages of the proposed control in terms of the fast tran-
sient response and system robustness stability improvement
under various load and line disturbances.
As an advanced mathematical tool that recently emerged
Fig. 19. Main experimental results under (a) conventional LQR control when as an effective method in convex optimization, with the
the input step changes from 75 to 50 V and (b) proposed LMI-LQR control introduction of LMI in the conventional linear control LQR,
when the input step changes from 75 to 50 V.
the proposed algorithm shows obvious advantages that can be
summarized as: 1) improved system stability and robustness
with fewer oscillations; 2) easy implementation for system per-
formance optimization by incorporating several performance
indices with some constraints through LMI formula; and
3) high capability in dealing with system uncertainty through
building polytopic model of uncertainty circuit. Although
this article is mainly focused on the SPS-based LMI-LQR
control optimization, the proposed LMI-LQR closed-loop
control is universal and can be easily extended to the
multiple-phase-shift (MPS) control in order to improve the
power conversion efficiency and simultaneously the dynamic
performance.

R EFERENCES
[1] T. LaBella, W. Yu, J.-S. Lai, M. Senesky, and D. Anderson,
“A bidirectional-switch-based wide-input range high-efficiency isolated
resonant converter for photovoltaic applications,” IEEE Trans. Power
Electron., vol. 29, no. 7, pp. 3473–3484, Jul. 2014.
[2] H. Wen, W. Xiao, and B. Su, “Nonactive power loss minimization in a
bidirectional isolated DC–DC converter for distributed power systems,”
IEEE Trans. Ind. Electron., vol. 61, no. 12, pp. 6822–6831, Dec. 2014.
[3] J.-W. Yang and H.-L. Do, “Soft-switching dual-flyback DC–DC con-
verter with improved efficiency and reduced output ripple current,” IEEE
Trans. Ind. Electron., vol. 64, no. 5, pp. 3587–3594, May 2017.
[4] D. Murthy-Bellur and M. K. Kazimierczuk, “Isolated two-transistor zeta
converter with reduced transistor voltage stress,” IEEE Trans. Circuits
Fig. 20. Experimental comparison when the output reference voltage Ve Syst. II, Exp. Briefs, vol. 58, no. 1, pp. 41–45, Jan. 2011.
changes from 50 to 40 V. (a) Conventional LQR controller. (b) Proposed [5] F. Zhang and Y. Yan, “Novel forward–flyback hybrid bidirectional
LMI-LQR controller. DC–DC converter,” IEEE Trans. Ind. Electron., vol. 56, no. 5,
pp. 1578–1584, May 2009.
[6] Z. Zhang, O. C. Thomsen, and M. A. E. Andersen, “Optimal design of a
push-pull-forward half-bridge (PPFHB) bidirectional DC–DC converter
with variable input voltage,” IEEE Trans. Ind. Electron., vol. 59, no. 7,
fewer oscillations within 0.01 s. It indicates that the proposed pp. 2761–2771, Jul. 2012.
[7] B. Han, C. Bai, J. S. Lee, and M. Kim, “Repetitive controller of
LMI-LQR controller has better dynamic performance than capacitor-less current-fed dual-half-bridge converter for grid-connected
the conventional LQR controller under the reference voltage fuel cell system,” IEEE Trans. Ind. Electron., vol. 65, no. 10,
step-down change condition. pp. 7841–7855, Oct. 2018.

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
144 IEEE TRANSACTIONS ON TRANSPORTATION ELECTRIFICATION, VOL. 6, NO. 1, MARCH 2020

[8] H. Wen, B. Su, and W. Xiao, “Design and performance evaluation of a [28] C. Olalla, R. Leyva, A. El Aroudi, and I. Queinnec, “Robust LQR control
bidirectional isolated DC–DC converter with extended dual-phase-shift for PWM converters: An LMI approach,” IEEE Trans. Ind. Electron.,
scheme,” IET Power Electron., vol. 6, no. 5, pp. 914–924, May 2013. vol. 56, no. 7, pp. 2548–2558, Jul. 2009.
[9] H. Shi et al., “Minimum-backflow-power scheme of DAB-based solid- [29] A. Sel, U. Gunes, O. Elbir, and C. Kasnakoglu, “Comparative analysis
state transformer with extended-phase-shift control,” IEEE Trans. Ind. of performance of the SEPIC converter using LQR and PID controllers,”
Appl., vol. 54, no. 4, pp. 3483–3496, Jul. 2018. in Proc. 21st Int. Conf. Syst. Theory, Control Comput. (ICSTCC),
[10] H. van Hoek, M. Neubert, and R. W. De Doncker, “Enhanced modulation Oct. 2017, pp. 839–844.
strategy for a three-phase dual active bridge—Boosting efficiency of an [30] A. Deihimi and M. E. S. Mahmoodieh, “Analysis and control of battery-
electric vehicle converter,” IEEE Trans. Power Electron., vol. 28, no. 12, integrated DC/DC converters for renewable energy applications,” IET
pp. 5499–5507, Dec. 2013. Power Electron., vol. 10, no. 14, pp. 1819–1831, Nov. 2017.
[11] Q. Ye, R. Mo, and H. Li, “Low-frequency resonance suppression of [31] D. O. Neacsu and A. Sirbu, “Energy savings with LQR control of
a dual-active-bridge DC/DC converter enabled DC microgrid,” IEEE DC/DC converters,” in Proc. 44th Annu. Conf. IEEE Ind. Electron. Soc.
J. Emerg. Sel. Topics Power Electron., vol. 5, no. 3, pp. 982–994, (IECON), Oct. 2018, pp. 1198–1203.
Sep. 2017. [32] S. A. Lindiya, K. Vijayarekha, and S. Palani, “Deterministic LQR
[12] A. Rodriguez, A. Vazquez, D. G. Lamar, M. M. Hernando, and controller for DC–DC buck converter,” in Proc. Biennial Int. Conf.
J. Sebastian, “Different purpose design strategies and techniques to Power Energy Syst., Towards Sustain. Energy (PESTSE), Jan. 2016,
improve the performance of a dual active bridge with phase-shift pp. 1–6.
control,” IEEE Trans. Power Electron., vol. 30, no. 2, pp. 790–804, [33] B. Ufnalski, A. Kaszewski, and L. M. Grzesiak, “Particle swarm
Feb. 2015. optimization of the multioscillatory LQR for a three-phase four-wire
voltage-source inverter with an LC output filter,” IEEE Trans. Ind.
[13] X. Liu et al., “Novel dual-phase-shift control with bidirectional inner
Electron., vol. 62, no. 1, pp. 484–493, Jan. 2015.
phase shifts for a dual-active-bridge converter having low surge current
[34] J. V. Fonseca, I. S. Abreu, P. H. M. Rego, M. D. P. M. Wolff,
and stable power control,” IEEE Trans. Power Electron., vol. 32, no. 5,
and O. F. Silva, “A genetic algorithm convergence and models for
pp. 4095–4106, May 2017.
eigenstructure assignment via linear quadratic regulator (LQR),” IEEE
[14] B. Zhao, Q. Yu, and W. Sun, “Extended-phase-shift control of isolated Latin Amer. Trans., vol. 6, no. 1, pp. 1–9, Mar. 2008.
bidirectional DC–DC converter for power distribution in microgrid,” [35] H. Asadi, S. Mohamed, C. P. Lim, and S. Nahavandi, “Robust optimal
IEEE Trans. Power Electron., vol. 27, no. 11, pp. 4667–4680, Nov. 2012. motion cueing algorithm based on the linear quadratic regulator method
[15] K. Wu, C. W. de Silva, and W. G. Dunford, “Stability analysis of and a genetic algorithm,” IEEE Trans. Syst., Man, Cybern., Syst., vol. 47,
isolated bidirectional dual active full-bridge DC–DC converter with no. 2, pp. 238–254, Feb. 2017.
triple phase-shift control,” IEEE Trans. Power Electron., vol. 27, no. 4, [36] T. Tarczewski, L. J. Niewiara, and L. M. Grzesiak, “An application of
pp. 2007–2017, Sep. 2012. flower pollination algorithm to auto-tuning of linear-quadratic regulator
[16] L. Shi, W. Lei, Z. Li, J. Huang, Y. Cui, and Y. Wang, “Bilinear discrete- for dc-dc power converter,” in Proc. 20th Eur. Conf. Power Electron.
time modeling and stability analysis of the digitally controlled dual Appl. (EPE ECCE Europe), Sep. 2018, pp. P.1–P.8.
active bridge converter,” IEEE Trans. Power Electron., vol. 32, no. 11, [37] A. Özdemir and Z. Erdem, “Double-loop PI controller design of the
pp. 8787–8799, Nov. 2017. DC–DC boost converter with a proposed approach for calculation of
[17] K. Zhang, Z. Shan, and J. Jatskevich, “Large- and small-signal the controller parameters,” Proc. Inst. Mech. Eng., I, J. Syst. Control
average-value modeling of dual-active-bridge DC–DC converter con- Eng., vol. 232, no. 2, pp. 137–148, Nov. 2017.
sidering power losses,” IEEE Trans. Power Electron., vol. 32, no. 3,
pp. 1964–1974, Mar. 2017.
[18] F. Krismer and J. W. Kolar, “Accurate small-signal model for the digital
control of an automotive bidirectional dual active bridge,” IEEE Trans.
Power Electron., vol. 24, no. 12, pp. 2756–2768, Dec. 2009.
[19] F. Kurokawa, A. Yamanishi, and S. Hirotaki, “A reference modifi-
cation model digitally controlled DC–DC converter for improvement
of transient response,” IEEE Trans. Power Electron., vol. 31, no. 1,
pp. 871–883, Jan. 2016.
Peizhou Xia was born in Chengdu, China, in 1996.
[20] D. Sha, J. Zhang, X. Wang, and W. Yuan, “Dynamic response He received the B.E. degree in electrical engineering
improvements of parallel-connected bidirectional DC–DC converters for from the Department of Electrical and Electronic
electrical drive powered by low-voltage battery employing optimized Engineering, Xi’an Jiaotong-Liverpool University,
feedforward control,” IEEE Trans. Power Electron., vol. 32, no. 10, Suzhou, China, in 2018, and the M.S. degree in
pp. 7783–7794, Oct. 2017. electrical power engineering from the Department
[21] J. Yang, H. Cui, S. Li, and A. Zolotas, “Optimized active disturbance of Engineering, University of Edinburgh, Edinburgh,
rejection control for DC–DC buck converters with uncertainties using a U.K., in 2019.
reduced-order GPI observer,” IEEE Trans. Circuits Syst. I, Reg. Papers, His current research interests include bidirectional
vol. 65, no. 2, pp. 832–841, Feb. 2018. dc–dc converters, power electronics, and renewable
[22] S. Zou, S. Zheng, and M. Chinthavali, “Design, analyses and validation power conversion systems.
of sliding mode control for a DAB DC–DC converter,” in Proc. IEEE
Transp. Electrific. Conf. Expo (ITEC), Jun. 2019, pp. 1–6.
[23] S. Dutta, S. Hazra, and S. Bhattacharya, “A digital predictive current-
mode controller for a single-phase high-frequency transformer-isolated
dual-active bridge DC-to-DC converter,” IEEE Trans. Ind. Electron.,
vol. 63, no. 9, pp. 5943–5952, Sep. 2016.
[24] X. Gao, L. Fu, F. Ji, and Y. Wu, “Virtual current based direct power
control strategy of dual-active-bridge DC–DC converter,” in Proc. IEEE
Int. Conf. Mechatronics Autom. (ICMA), Aug. 2019, pp. 1947–1952. Haochen Shi (Student Member, IEEE) was born
[25] W. Song, N. Hou, and M. Wu, “Virtual direct power control scheme of in Hubei, China, in 1992. He received the B.S.
dual active bridge DC–DC converters for fast dynamic response,” IEEE degree from the Department of Electrical Engi-
Trans. Power Electron., vol. 33, no. 2, pp. 1750–1759, Feb. 2018. neering, China Three Gorge University, Yichang,
[26] N. Hou, W. Song, Y. Li, Y. Zhu, and Y. Zhu, “A comprehensive China, in 2014, and the M.Eng. degree from the
optimization control of dual-active-bridge DC–DC converters based on Department of Electrical Engineering, University of
unified-phase-shift and power-balancing scheme,” IEEE Trans. Power Leicester, Leicester, U.K., in 2015. He is currently
Electron., vol. 34, no. 1, pp. 826–839, Jan. 2019. pursuing the Ph.D. degree with the University of
[27] G. G. Oggier, M. Ordonez, J. M. Galvez, and F. Luchino, “Fast transient Liverpool, U.K.
boundary control and steady-state operation of the dual active bridge His current research interests include bidirectional
converter using the natural switching surface,” IEEE Trans. Power dc–dc converter, electrical vehicles, and renewable
Electron., vol. 29, no. 2, pp. 946–957, Feb. 2014. power conversion systems.

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.
XIA et al.: ROBUST LMI-LQR CONTROL FOR DAB DC–DC CONVERTERS WITH HIGH PARAMETER UNCERTAINTIES 145

Huiqing Wen (Senior Member, IEEE) received the Yihua Hu (Senior Member, IEEE) received the B.S.
B.S. and M.S. degrees in electrical engineering from degree in electrical engineering and the Ph.D. degree
Zhejiang University, Hangzhou, China, in 2002 and in power electronics and drives from the China Uni-
2006, respectively, and the Ph.D. degree in electrical versity of Mining and Technology, Xuzhou, China,
engineering from the Chinese Academy of Sciences, in 2003 and 2011, respectively.
Beijing, China, in 2009. From 2011 to 2013, he was with the Col-
From 2009 to 2010, he was an Electrical Engineer lege of Electrical Engineering, Zhejiang University,
with GE (China) Research and Development Center Hangzhou, China, as a Post-Doctoral Fellow. From
Company, Ltd., Shanghai, China. From 2010 to 2013 to 2015, he worked as a Research Asso-
2011, he was an Engineer with the China Coal ciate with the Power Electronics and Motor Drive
Research Institute, Beijing. From 2011 to 2012, Group, University of Strathclyde, Glasgow, U.K.
he was a Post-Doctoral Fellow with the Masdar Institute of Science and From 2016 to 2019, he was a Lecturer with the Department of Electrical
Technology, Abu Dhabi, United Arab Emirates. In 2013, he joined the Engineering and Electronics, University of Liverpool (UoL), Liverpool, U.K.
Electrical and Electronic Engineering Department, Xi’an Jiaotong-Liverpool He is currently a Reader with the Electronics Engineering Department, Uni-
University (XJTLU), Suzhou, China. He is currently an Associate Professor versity of York (UoY), York, U.K. He has published 100 articles in the IEEE
with XJTLU. He has published more than 50 peer-reviewed technical articles T RANSACTIONS journals. His research interests include renewable generation,
in leading journals. His research interests include renewable energy, electric power electronics converters and control, electric vehicle, more electric
vehicle, power electronics, microgrid, and power semiconductor devices. ship/aircraft, smart energy systems, and nondestructive test technology.
Dr. Wen is also an Associate Editor of IEEE A CCESS , the International Dr. Hu is also an Associate Editor of the IEEE T RANSACTIONS ON I NDUS -
Journal of Photoenergy, and Journal of Power Electronics. TRIAL E LECTRONICS , IET Renewable Power Generation, IET Intelligent
Transport Systems, and Power Electronics and Drives.

Yong Yang (Member, IEEE) received the B.S.


degree in automation from Xiangtan University,
Xiangtan, China, in 2003, the M.S. degree in electri-
cal engineering from Guizhou University, Guiyang,
China, in 2006, and the Ph.D. degree in electrical
Qinglei Bu (Student Member, IEEE) was born engineering from Shanghai University, Shanghai,
in Anhui, China, in 1995. He received the B.S. China, in 2010.
degree in electrical engineering from Xi’an He is currently an Associate Professor with
Jiaotong-Liverpool University, Suzhou, China, the School of Rail Transportation, Soochow Uni-
in 2017. He is currently pursuing the Ph.D. degree versity, Suzhou, China. From December 2017 to
with the University of Liverpool, Liverpool, U.K. December 2018, he was a Visiting Scholar with the
His research interests include bidirectional dc–dc Center for High Performance Power Electronics (CHPPE), The Ohio State
converter, renewable power conversion systems, University, Columbus, OH, USA. He has coauthored more than 60 journal
electrical vehicles, and GaN power electronics. and conference papers. His current research interests include model predictive
control in power electronic converters, distributed energy resource interfacing,
and high-performance motor drive control.

Authorized licensed use limited to: Northwestern University. Downloaded on May 03,2020 at 13:05:41 UTC from IEEE Xplore. Restrictions apply.

You might also like