1720 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 38, NO.

2, FEBRUARY 2023

Adaptive Model Predictive Control of an Interleaved

Boost Converter Using Real-Time Updated Model
Hongyu Zhang , Student Member, IEEE, Yuren Li, Renyou Xie , Student Member, IEEE,
Jian Song , Student Member, IEEE, Bo Liang , and Yigeng Huangfu , Senior Member, IEEE

Abstract—To optimize the voltage regulation performance and in industrial applications, due to its optimized control effects
enhance the robustness, an adaptive model predictive controller is with a flexible and straightforward design process [5], [6].
proposed in this article for an interleaved dc–dc boost converter. Restricted by the computational burden, MPC was originally
The predictive model is constructed by linearizing the nonlinear
equations of the converter at the current operating point. A novel applied to the process control with slow time-varying dynamics
parameter update mechanism is developed based on the static [7]. With the development of computing hardware and opti-
model, which enables fast and accurate identification of the model mization method, MPC has been successfully applied in power
parameters, thus the control law can guarantee the desired control electronics, such as neutral-point-clamped inverters [8], [9],
performance in the wide operating range. Based on the adaptive servo motors [10], and power converters/rectifiers [11], [12].
model, a disturbance observer is constructed to correct the model
predictions and mitigate parameter uncertainties. Then an explicit Moreover, literature [13] implements an MPC method with
control law is derived by solving the optimization problem offline, a long horizon on a field-programmable gate array (FPGA),
and the stability analysis is carried out using the linear control which further promotes the application of MPC in fast dynamic
theory. Finally, the effectiveness of the proposed method is verified systems.
by both simulation and experimental results. As a model-based method, the performance of MPC greatly
Index Terms—Disturbance observer (DOB), interleaved boost depends on the accuracy of the system model. The insufficient
converter, model predictive control (MPC), offset-free tracking. knowledge of model parameters and other unmodeled dynamics
will cause prediction errors and guide the control system along
I. INTRODUCTION a different trajectory from the reference one. For this reason, the
integral action was integrated into MPC to mitigate steady-state
APID development in renewable energy has promoted the
R utilization of power converters. Since the output voltage of
renewable energy sources, such as fuel cells and photovoltaics is
errors [14]. Note that the integral action also interacts with other
control performances, such as transient behaviors, tracking, and
robustness. Thus, applying integral action to mitigate the offset
low and unregulated, power converters are required as the inter-
may deteriorate the other properties of the closed-loop system.
faces for energy sources and loads. Among them, the interleaved
Numerous efforts have been made to remove the control offsets
boost converter is a promising candidate due to its superiorities
of MPC, which motivates the design of adaptive MPC (AMPC).
in power density, efficiency, and fault tolerance [1]. However,
In general, the realization of AMPC can be divided into model
the nonminimum phase behaviors and random load variations of
approximation, adaptation techniques, parameter identification,
the converter bring great difficulties to its accurate control. Thus,
and disturbance estimation methods.
many advanced control strategies have been developed such
For model approximation methods, the system dynamics are
as backstepping control [2], sliding mode control [3], flatness
approximated by a set of submodels. Therefore, the parameter
control [4], etc. The existing methods have improved the control
uncertainties could be captured in the wide operating range. In
performance from different aspects.
[15], a parameter-varying model consisting of a set of linearized
Among different control methods, model predictive control
models at different operating points was applied to describe the
(MPC) is known as one of the most promising approaches
converter dynamics. In [16], the authors developed an AMPC
for modular multilevel converters, and a fuzzy approximation
Manuscript received 28 May 2022; revised 6 September 2022; accepted 17
October 2022. Date of publication 25 October 2022; date of current version system with adaptive weight vectors was applied to estimate the
18 November 2022. This work was supported in part by the Key Research unknown nonlinear system dynamics based on the input/output
and Development Program of Shaanxi under Grant 2021GY-292, in part by the data. However, in [15] and [16], the prediction accuracy is
Aviation Science Foundation under Grant ASFC-20200019053004, and in part
by the National Natural Science Foundation of China under Grant 61873343. affected by the number of submodels, and the model parameters
Recommended for publication by Associate Editor J. Popovic-Gerber. cannot exceed the predefined minimum/maximum ranges. The
(Corresponding author: Yuren Li.) AMPC with adaptation techniques normally utilizes adaptive
The authors are with the School of Automation, Northwestern Polytechnical
University, Xi’an 710129, China (e-mail: [email protected]; updating laws to estimate the sensitive parameters of the system.
[email protected]; [email protected]; [email protected]. In [17], an AMPC was proposed for discrete-time linear systems,
edu.cn; [email protected]; [email protected]). and the adaptive law for model parameters was developed based
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TPEL.2022.3216600. on the gradient descent method. Some other adaptation laws
Digital Object Identifier 10.1109/TPEL.2022.3216600 were designed and integrated into AMPCs for servo motors, such

0885-8993 © 2022 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: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: ADAPTIVE MPC OF AN INTERLEAVED BOOST CONVERTER USING REAL-TIME UPDATED MODEL 1721

as the Lyapunov method [18], and the Popov hyperstability the-

ory [19]. But for this kind of AMPC, the algorithm complexity
will increase with the number of sensitive model parameters.
Some online parameter identification laws are also used to
identify the model parameters in the prediction model. These
algorithms do not require prior knowledge of the control plants.
In [20], a predictive control was proposed for power converters,
and the pseudo-partial derivative was updated by a data-driven
iterative learning method. In [21], an AMPC with neuro-fuzzy
parameter estimation was developed for the grid-forming con-
Fig. 1. Topology of TIBC.
verters. The recursive least squares (RLS) algorithm was used to
identify the predictive model in [22]. The main disadvantage of
[20], [21], [22] is the utilization of complex and computationally
intensive algorithms and high data storage requirements. To 2) The proposed method takes the circuit fault conditions
address this problem, an adaptive reference MPC was developed into account, thus generating the optimized control per-
in [23] to reduce the computation costs. It models the virtual formance over a more general operating range.
system using the linear fitting method, and a virtual reference 3) A DOB is integrated into the AMPC, which allows a
is generated to mitigate mismatches, which has better transient comprehensive consideration of all the parameter uncer-
performance than MPC with integrator. Except for these AMPCs tainties.
mentioned above, disturbance estimation is also an efficient The rest of this article is arranged as follows. The converter
method to deal with model mismatches, and some AMPCs analysis and modeling are presented in Section II. The proposed
are implemented by applying disturbance observers (DOB). In adaptive model predictive controller with the parameter update
[24], the Luenberger observer was adopted to estimate the input mechanism and DOB is presented in Section III. To validate the
voltage and inductor resistance. The Luenberger observer was effectiveness of the proposed MPC method, the simulation and
also applied in [25] to estimate the load resistance, thus the experimental results are presented in Section IV and Section V.
developed AMPC method exhibited strong robustness against The conclusion is drawn in Section VI.
load variations.
The AMPC works mentioned above can enhance the control II. CONVERTER TOPOLOGY AND MODELING
performance and robustness to certain mismatched parameters.
A two-phase interleaved boost converter (TIBC) is applied to
However, for boost converters, the consideration of parameter
verify the feasibility and effectiveness of the proposed method,
mismatches is not comprehensive, such as the negative effects
as shown in Fig. 1. L1 , L2 are the inductors, C is the capacitor,
caused by the nonminimum phase behaviors and variations in
Q1 , Q2 are the power switches, D1 , D2 are the diodes, RL is the
inductance. Besides, it should be noted that the AMPC works
equivalent resistor, and δ1 , δ2 are the gate signals generated by
described above mainly focus on dealing with the parameter
the pulsewidth modulation (PWM) blocks. Besides, vin is the
uncertainties under normal operating conditions, and few works
input voltage, vo is the output voltage, iL1 , iL2 are the inductor
have been extended to suppress parameter uncertainties and
currents, and io is the load current.
guarantee the optimized control performance under circuit fault
For modularity design, the circuit parameters of two modules
conditions.
are considered to be identical
To address the aforementioned problems, this article proposes
a composite AMPC law that combines parameter identification L1 = L2 = L, d1 = d2 = d (1)
and an observer for interleaved boost converters. The adaptive
predictive model consists of two units, the linear model and the where d1 , d2 are the duty ratios for the power switches Q1 , Q2 .
lumped disturbance term. A novel parameter update mechanism The dynamic model of TIBC is obtained as
is proposed to identify the model parameters. Different from the
diLi
existing identification laws, the parameter update mechanism L = vin − (1 − di ) vo , i = 1, 2 (2.a)
is developed based on the static model, which enables fast and dt
accurate parameter identification with low computational costs. dvo vo
C = (1 − d1 ) iL1 + (1 − d2 ) iL2 − . (2.b)
Based on the adaptive model, a DOB is adopted to compensate dt RL
for the lumped disturbances, including the non-minimum phase
Ignoring the parasitic parameters and assuming the converter
behaviors, linearization errors, and other parasitic parameters.
operates in continuous conduction mode. Let the derivatives in
Finally, an explicit closed-loop solution is obtained by solving
(2) be zero, the static model of TIBC is
the receding horizon optimization problem offline.
The main contributions of this article include the following: 
Vo = Vin / (1 − D)
1) A novel parameter update mechanism is developed based (3)
Vo = (IL1 + IL2 ) (1 − D) RL
on the static model, which enables fast and accurate pa-
rameter identification with low computational costs in the where Vo , Vin , D, IL1 , IL2 represent the steady-state values of
presence of large external disturbances. the variables vo , vin , d, iL1 , iL2 , respectively.

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
1722 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 38, NO. 2, FEBRUARY 2023

2Vin
where a = RL2 C , b = CV o
, and c = RLLV o
CVin .
Denoting u = ILref , y = vo , and the equivalent control plant
for the voltage loop is reformulated as
ẏ = −ay + bu + w (8)
where w is defined as the lumped disturbance, and the DOB is
designed later to estimate its value.
2) Parameter Update Mechanism: Note that the control plant
(8) is obtained by the linearization of the nonlinear system (4),
it can only describe the converter dynamics near the equilibrium
point (5). The values of a and b are influenced by the operating
point, and some external disturbances can lead the converter
Fig. 2. Proposed control scheme for TIBC.
to operate far from the nominal operating point. But for the
typical DOB-based AMPC method, the nominal values of a, b
are utilized and all the system uncertainties are estimated by
III. OVERALL CONTROL STRATEGY the observer. In this case, if we ignore the observer dynamics,
A. Model Conversion and Configuration the unconstrained solution of the optimization problem derived
from (8) will be a time-invariant state feedback control law, and
1) Cascade Control Scheme: The cascade control scheme for the corresponding control gains are constant values calculated
boost converters is shown in Fig. 2. The outer loop regulates the from the weighting factors and nominal model parameters a, b.
output voltage and generates the reference inductor current for Although the DOBs can provide integral actions to remove
inner loop. The inner loop regulates the inductor current to the the prediction errors, the fixed control gains are only locally
reference value and generates duty ratios for PWM blocks. This optimized near the nominal operation point [26], and the control
article focuses on the voltage controller, and the PI algorithm is performance may be degraded at other operating points.
applied to the inner loop. Since the inner loop generally has a To accurately describe the converter dynamics and guarantee
larger bandwidth than the outer loop, the tracking performance the optimized control performance in the wide operating range,
of the inner loop is assumed to be ideal. Then according to (2), a parameter update mechanism is developed based on the static
the input-output relationship between the output voltage and model (3), which utilizes the inductor currents and duty ratios
inductor current is to estimate the actual values of model parameters a and b. The
dvo core idea of the parameter update mechanism is to continuously
Cvo = (iL1 + iL2 ) (vin − L · dILref /dt) − vo2 /RL . (4) linearize the voltage dynamics around its actual operating point.
dt
Since the changes in inductance and capacitance are small, only
Note that the voltage dynamics (4) is the nonlinear form, thus the influence of load variations and open-circuit fault on model
the nonlinearity and unstable behavior requires an efficient and parameters is considered, and the adverse effects caused by
fast stabilizing control. However, the nonlinear model will lead the other parasitic parameters will be estimated by the DOB
to a nonlinear MPC, hence the optimization problem requires to designed in Section III-B.
be solved online, which will increase the computational burden According to (7), the parameter a is inversely proportional to
of the microprocessor. And the computation time for finding the the equivalent load resistor RL . During the operation of TIBC,
control input will also restrict the sampling rate, thus affecting the actual value of a varies with the load power in a wide range.
the transient performance of the converter. According to the static model (3), replacing the load resistor RL
To develop a linear MPC method to reduce the computational with measured values of output voltage and inductor current, a
burden, the voltage dynamics (4) is first linearized by the small- can be approximated by
signal modeling method. Let the derivative in (4) be zero, the 2 2 (IL1 + IL2 ) (1 − D)
equilibrium point is â = =
RL C CVo
IL1,2 = ILref = vo2 / (2vin RL ) . (5) 2iL1 (1 − d1 ) + 2iL2 (1 − d2 )
≈ . (9)
Cvo
Linearizing (4) around the equilibrium point (5) gives
  The load variation will also affect parameter b. The value of
dṽo dĩLref 2Vo ṽo b is proportional to vin , which is mainly affected by the load
CVo = 2 ILref ṽin − LILref + Vin ĩLref −
dt dt RL power in most renewable energy sources. Besides, suppose that
(6) a single-phase open-circuit fault occurs, the equilibrium points
where ṽo , ṽin , and ĩL are the small-signal disturbance values. of TIBC in (5) will switch to
According to (6), the control-to-output transfer function of
iLi = 0, iLj = ILref = vo2 / (vin RL ) , i, j = 1, 2, i = j.
the voltage loop is
(10)
Vo (s) b − cs In this case, the nominal value of b in (7) can be obtained as
Gvi (s) = | ṽin = 0 = (7)
ILref (s) s+a b = Vin /(CVo ), which is half of the original value. Based on

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: ADAPTIVE MPC OF AN INTERLEAVED BOOST CONVERTER USING REAL-TIME UPDATED MODEL 1723

the static model (3), the parameter b is approximated by where Γ1 (τ ) = 1, Γ2 (τ ) = [τ, 0.5τ 2 ] , Y1 (t) = y(t), and
2Vin 2 (1 − D) (1 − d1 ) + (1 − d2 ) Y2 (t) = [ẏ(t), ÿ(t)]T .
b̂ = ≈ ≈ . (11) In order to enhance the prediction accuracy and eliminate
CVo C C
steady-state errors, the disturbance estimations ŵ are utilized to
As the saturation on the duty ratio is [0, 1], (1 − d1 ) or correct the output predictions. Hence the derivatives in (16) can
(1 − d2 ) will be equal to zero when a single-phase open-circuit be expressed in the matrix form as
fault occurs. The estimation of â in (9) is still valid for the
fault mode, since the inductor current of the faultless circuit Y2 (t) = Yw (t) + Wa ū (t) (17)
will rise to meet the load demand. In this process, the value of
T
â can be adapted to the converter dynamics automatically, thus where Yw (t) = Y (t) + Wb d(t), Y (t) = [−ây(t), â2 y(t)]
guaranteeing the relative accuracy of the system model.
b̂ 0 1 0
Wa = , Wb = , d (t) = [ŵ, 0]T .
B. Design of the Disturbance Observer −âb̂ b̂ −â 1
According to the analysis in Section III-A, the impact of load Define the reference values corresponding to Y1 (t) and Y2 (t)
variations and open-circuit faults on the system dynamics is not as Yc1 (t) and Yc2 (t) respectively. The future reference voltage
infinite. Assume the derivatives of lumped disturbances in (8) yr (t + τ ) within the prediction time can be obtained in a similar
are bounded and satisfy [27]
way as (16). And define ūr (t) = [ur (t), u̇r (t)]T as the desired
ẇ < ∞, lim ẇ = 0. (12) control sequence, by substituting the desired condition Y2 (t) =
t→∞
Yc2 (t) into (17), the ūr (t) is obtained as
Substituting the estimated values of â and b̂ into (8), and the
DOB is given as ūr (t) = Wa−1 [Yc2 (t) − Yw (t)] . (18)
 
ŵ˙ = −L0 ŵ − ây + b̂u + L0 ẏ (13) The future control input uc (t) and the desired control input
ur (t) in (15) is expressed as
where ŵ is the estimated value of w, and L0 is the observer gain.
Define the estimation error as er = w − ŵ, and combine (8) uc (t) = Γr (τ ) ū (t) , ur (t) = Γr (τ ) ūr (t) (19)
and (13), the derivative of er is
where Γr (τ ) = [1, τ ]T .
ėr = −L0 er + ẇ. (14) Combining (16) and (19), the performance index in (15) can
Therefore, the estimation error er is bounded and its bounds be formulated as
are related to ẇ. With the assumption that ẇ = 0 as t → ∞, the   ΓT1
Ts
estimation error er is asymptotically stable. J= 1
(Y1 − Yc1 )T , (Y2 − Yc2 )T
2 0
ΓT2
Y1 − Yc1
C. Adaptive Predictive Controller Design × Q [Γ1 , Γ2 ] + (ū − ūr )T ΓTr RΓr (ū − ūr ) dτ
Y2 − Yc2
In this article, the closed-form MPC developed in [28] is used = (Y1 − Yc1 )T ξ1 (Y1 − Yc1 ) + 2(Y1 − Yc1 )T ξ2 (Y2 − Yc2 )
to obtain the optimal control law. For power converters, the cost
+(Y2 − Yc2 )T ξ3 (Y2 − Yc2 ) + (ū − ūr )T ξ4 (ū − ūr )
function can be defined as
 Ts  (20)
1 where
J= Q[yr (t+τ)−yp (t+τ)]2 +R[uc (t)−ur (t)]2 dτ
2 0  Ts
1 Q
(15) ξ1 = ΓT1 QΓ1 dτ = Ts ,
where yref and yp (t + τ ) are the reference value and predicted 2 0 2
value of the output voltage, respectively. u(t) and ur (t) are the 
1 Ts
Q  Ts2 Ts3

future control input and the desired control input, respectively. ξ2 = ΓT1 QΓ2 dτ =
2 0 2 2 6
Ts is the prediction time, and Q, R are the weighting factors.
 
In (15), the predicted voltage yp (t + τ ) should be expanded 1 Ts
Q
Ts3 Ts4
into the (ρ + r)th order Taylor series expansion. ρ is the input ξ3 = ΓT2 QΓ2 dτ = 3
Ts4
8
Ts5
2 0 2
relative degree and r is the control order. The control order r 8 20
is selected to describe the variation of the control input within  
Ts Ts2
1 R Ts
the prediction time and ensure the closed-loop stability [28]. ξ4 = ΓTr RΓr dτ = Ts2
2
Ts3
.
Since the input relative degree of the plant (8) is one, the control 2 0 2
2 3
order is selected as one in this article. Then, the control sequence
within the prediction time is defined as ū (t) = [u(t), u̇(t)]T , The derivative of the cost function with respect to ū gives
and the predicted output voltage is expressed as  
∂J/∂ ū = 2 WaT ξ3 Wa + ξ4 ū + 2WaT ξ2T (Y1 − Yc1 )
Y1 (t)
yp (t + τ ) = [Γ1 (τ ) , Γ2 (τ )] (16) − 2ξ4T ūr + 2WaT ξ3T [Yw − Yc2 ] . (21)
Y2 (t)

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
1724 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 38, NO. 2, FEBRUARY 2023

Letting ∂J/∂ ū = 0 and combining (18), the solution of the Applying Laplace transform into (13), the transfer function
optimization problem (20) is of the DOB is
 −1
ū = − ξ3 Wa +Wa−T ξ4 ξ2T (Y1 −Yc1)−Wa−1 (Yw −Yc2) . ŵ = G1 (s) y (t) − G2 (s) û (t) (26)
(22)
Taking the first row of ū∗ , the unconstrained optimal control where G1 (s) = (s+a 0 )L0
s+L0
b0 L 0
, G2 (s) = s+L 0
.
law û(t) is obtained as Combining (24)–(26), the open-loop and closed-loop transfer
functions of the control system are
û (t) = −k1 (y (t) − yref ) + k2 (ẏref + ây (t) − ŵ) (23)
kb b0 k1 (s + L0 )
−1 Go (s) = (27)
where k1 is the first row of (ξ3 Wa + Wa−T ξ4 ) ξ2T , k2 = 1/b̂ s2 + [(ka − kb ) a0 + L0 kb ] s
Ts b̂Q[12Qb̂2 Ts2 −160RâTs +240R] y (s)
k1 = . Gc (s) =
(3Q2 b̂4 +48QRâ2 b̂2 )Ts4 −96QRâb̂2 Ts3 +104QRb̂2 Ts2 +240R2 yr (s)
Define the maximum allowable inductor current is ILmax , the kb b0 k1 (s + L0 )
optimal control law can be obtained by introducing the posterior =
s2 + [(ka − kb ) a0 + L0 kb + kb b0 k1 ] s + kb b0 k1 L0
constraint on (23). (28)
The overall control block diagram of AMPC cascaded with
PI control is shown in Fig. 2. When load changes or open-circuit where kb = b1 /b0 , ka = a1 /a0 .
faults occur, although they cannot be predicted in advance, To simplify the analysis, the mismatches of model parame-
the parameter update mechanism (9)–(11) can still identify the ter a1 is analyzed first. Letting kb = 1 and the characteristic
real values of model parameters at the new operating point equation is reformulated with respect to ka as
through the changing inductor current and duty ratio. Then, the a0 s
estimated values â, b̂ will be transmitted to the DOB (13) and 1 + ka = 0. (29)
s2 + (L0 − a0 + b0 k1 ) s + b0 k1 L0
control law (23). Based on the adaptive model, the observer
will calculate the total error between the control plant and the For illustration purposes, letting a0 = 400, b0 = 2000,
practical system. Then, the control law will calculate the optimal L0 = 500, k1 = 0.25. Then, the Bode plot of the open-loop
control gains k1 , k2 for the current operating point and generate transfer function (27) with different values of ka is shown in
the unconstrained control input. After a posterior constraint, the Fig. 3(a). Besides, the root locus of (29) is shown in Fig. 3(b),
reference inductor current is sent to the inner loop PI control to where ka changes from 0.25 to 4.
generate the duty ratios. It can be seen from Fig. 3(a) that with the increase of ka , the
system bandwidth decreases, which leads to a slower dynamic
D. Stability Analysis response. However, according to Fig. 3(b), when ka ≤ 1, two
poles are complex roots, which means that there is an overshoot
The stability of the proposed AMPC law is established by the in the step response. Hence, making ka = 1 can avoid overshoot
following theorem. and obtain a relatively high dynamic response.
Remark 1: For the prediction time Ts ≤ 3/(2â) , the control Then, the model mismatches caused by model parameter b are
gain k1 > 0 always holds; for the prediction time Ts > 3/(2â), discussed. Letting ka = 1 and the characteristic equation with
the condition for k1 > 0 is R/Q < (3b̂2 Ts2 )/(40âTs − 60). respect to kb is reformulated as
Theorem 1: Suppose that the lumped disturbance satisfies the
boundary conditions provided in (12), and the weighting factors (L0 − a0 + b0 k1 ) s + b0 k1 L0
1 + kb = 0. (30)
are selected as Remark 1, then the closed-loop control system is s 2 + a0 s
asymptotic stable. Making a0 = 400, b0 = 2000, L0 = 500, k1 = 0.25 and the
The proof of Remark 1 and Theorem 1 is given in Appendix. Bode plot the open-loop transfer function of (27) with different
Moreover, to better illustrate the impact of parameter update values of kb is shown in Fig. 4(a). Besides, the root locus of
mechanism on the control performance, assuming the converter (30) is shown in Fig. 4(b), where kb changes from 0.25 to 2.
operates at a nonrated point, the plant (8) is rewritten as The parameter b is determined by the circuit structure, which
ẏ = −a1 y + b1 u + w (24) may decrease due to circuit fault. It can be seen from Fig. 4(a)
that smaller kb leads to smaller bandwidth. Besides, according
where a1 , b1 are obtained based on the current operating point. to Fig. 4(b), when 0.44 ≤ kb ≤ 1, there is an overshoot in the
Considering the mismatch effect of a1 , b1 , assume the model step response. Therefore, making kb ≥ 1 could obtain a better
parameters used to formulate the control law are a0 , b0 , then the control performance. However, it should be noted that the outer
optimal control law in (23) can be reformulated as loop bandwidth cannot be too large to avoid measurement noise.
û (t) = −k1 (y (t) − yref ) + (a0 y (t) − ŵ) /b0 (25) Besides, it is generally lower than the observer bandwidth.
For the proposed AMPC method, since a parameter update
where the time derivative of the reference value ẏref is omitted mechanism is utilized to estimate the actual values of a, b in (8)
since it is not always available, and k1 is of the same form as in in real-time, the condition ka ≈ 1, kb ≈ 1 always holds, hence it
(23) but calculated from a0 , b0 . can guarantee the desired control performance in a wide range.

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: ADAPTIVE MPC OF AN INTERLEAVED BOOST CONVERTER USING REAL-TIME UPDATED MODEL 1725

Fig. 4. Influence of the variation of kb on control performance.

Fig. 3. Influence of the variation of ka on control performance.

Moreover, substitute ka = 1, kb = 1 into (28) and replace

b1 with its estimated value b̂, the closed-loop transfer function
is

y (s) b̂k1
Gc (s) = = . (31)
yr (s) s + b̂k1
To simplify the analysis, suppose the weight ratio is tuned as
R/Q = 0, and we only focus on eliminating the tracking errors.
Then a simpler version of k1 in (23) is derived as k1 = 4/(b̂Ts ).
This indicates that the system bandwidth b̂k1 is a constant value,
which is only determined by the prediction time Ts , and changes
in model parameters a, b will not affect the system dynamics. Fig. 5. Influence of L0 on the antidisturbance performance.
In addition, the transfer function from disturbance to system
output can be obtained from (24) to (26) as
disturbance of TIBC is generally low frequency, but the upper
y (s) s limit of L0 is limited by the sampling rate and sensor noise.
Gc (s) = =   . (32)
w (s) s2 + L0 + b̂k1 s + b̂k1 L0
IV. SIMULATION RESULTS
For illustration purposes, letting b̂ = 2000, k1 = 0.25. Then, To validate the theoretical analysis, the converter model with
the Bode plot of (32) with different L0 is given in Fig. 5. It can the proposed method is built and tested in MATLAB\Simulink.
be seen that a larger L0 leads to stronger robustness, due to the The circuit parameters of the converter are given in Table I. The

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
1726 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 38, NO. 2, FEBRUARY 2023

TABLE I
CIRCUIT PARAMETERS OF TIBC

Fig. 7. Simulation results of the converter under different R/Q.

According to (7), when the load current is 3.5 A, a =

310, b = 2000. Then, the inequality Ts ≤ 3/(2â) in Remark
1 holds and the system stability is independent of the weight
ratio. As can be seen in Fig. 7(a), all AMPCs with different
weight ratios can regulate the output voltage to the reference
value. Besides, larger R/Q leads to smaller current overshoot
Fig. 6. Simulation results of the converter under different k. but larger voltage drops and longer recovery time. Thus, the
weight ratio is chosen as 4 to balance control variations and
dynamic response.
sampling rates of the inductor current and output voltage are 50 For illustration purposes, the prediction time is tuned as 0.01s
and 2.5 kHz, respectively. The control parameters for the inner so that Ts > 3/(2â). According to Remark 1, the upper limit of
loop PI algorithm are tuned as kp1 = 0.05, ki1 = 30. the weight ratio is derived as 18.7 when the load current is 3.5A.
For the proposed controller, the average filters are utilized to The dynamic response of the converter is shown in Fig. 7(b),
smooth the estimated values â and b̂. All the inductor currents where the weight ratio changes from 4 to 20 at 1s.
and duty ratios used in the parameter update mechanism (9) and As can be observed from Fig. 7(b), when the weight ratio is
(11) take an average value at every 20 sampled values. And the 4, the output voltage can converge to the reference value. When
maximum reference inductor current ILmax is 15 A. the weight ratio changes to 20, the system becomes unstable
and the output voltage drops to 24V, which is consistent with
A. System Verification the analysis in Remark 1 and Theorem 1. The negative control
gain k1 makes the control input negative, which in turn makes
The selection of prediction time Ts and weight ratio R/Q will
the duty ratios of the system reach the lower limit of 0.
be discussed in this section.
First, the prediction time Ts is designed, and the weight ratio
B. Simulation Comparison
is set at zero for simplicity. According to (31), the bandwidth of
the closed-loop system is ωc = 4/Ts . Thus, a smaller prediction To evaluate the dynamic performance of the AMPC method, it
time is helpful to improve dynamic response, but decreases is compared with the adaptive model based-MPC (AM-MPC) for
the noise immunity. The lower limit of Ts is restricted by the the case of load variations and open-circuit faults. The AM-MPC
sampling period Tf , since AMPC is implemented based on the is implemented according to [29], which has a DOB and an op-
prediction of future converter behavior. Define k = Ts /Tf , and timized voltage change rate b̃ calculated by the steepest descent
the dynamic responses with different values of k are presented algorithm. The control parameters of the AM-MPC method are
in Fig. 6, where the load current steps from 3.5 to 7 A at 0.2 s. set as a = 450, Ts = 3 ms, R/Q = 4, and the optimized
As can be seen, when k = 1, too short prediction time leads coefficients are λ = 0.25, μ = 1.
to large static errors, and when the load current steps, there are Fig. 8 shows the simulation results of TIBC under step load
severe oscillations. When k varies from 5 to 20, larger k leads disturbance. The load current steps from 3.5 to 7 A at 0.25 s
to longer recovery time, which is consistent with the analysis and steps back to 3.5 A at 0.35 s. It can be observed from Fig. 8
above. Thus, the prediction time is chosen to be 10 times of the that both the AMPC and AM-MPC controllers can remove the
sampling time in this article. offset caused by load changes. For the AM-MPC controller, the
The weight ratio is used to penalize the control variation and model parameter b varies with the load current. For the AMPC
guarantee system stability. When the prediction time is tuned as controller, a smaller voltage drop/rise and shorter recovery time
4 ms, the dynamic response of the system under different weight are obtained. As the load current changes, the model parameter
ratios is shown in Fig. 7(a), where the load current steps from a steps from 328 to 656 and steps back to 328 at 0.35 s. This
3.5 to 7 A at 0.2 s. is consistent with the analysis in Section III. Since the constant

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: ADAPTIVE MPC OF AN INTERLEAVED BOOST CONVERTER USING REAL-TIME UPDATED MODEL 1727

Fig. 8. Simulation results of AM-MPC and AMPC methods under step load
disturbance. Fig. 10. Simulation results of AM-MPC and AMPC methods under open-
circuit fault.

Fig. 11. Experimental setup.

Fig. 9. Simulation results of AM-MPC and AMPC methods under sinusoidal TABLE II
load disturbance. PARAMETERS OF THREE CONTROLLERS

voltage source is applied in the simulation, the model parameter

b remains unchanged after the fluctuation.
The simulation results of two controllers under time-varying
load disturbance are shown in Fig. 9. The load current is set to
io = 6 + 2sin(20πt)A. It can be observed from Fig. 9 that the
AMPC controller still shows a better dynamic performance, the
voltage fluctuations of AMPC are smaller than the AM-MPC. V. EXPERIMENTAL RESULTS
For the AM-MPC controller, parameter b changes with load The experimental setup is presented in Fig. 11. It consists of a
current. For the AMPC controller, when the load current varies dc power supply, a TIBC, a dSPACE DS1007, an FPGA board,
sinusoidally, the model parameter a varies from 372 to 750, and and an electronic load. The controller is implemented in dSPACE
b remains unchanged, which is consistent with the theoretical platform to generate the control signals, and the FPGA board
analysis in Section III. is used to realize the phase shift. The circuit parameters and
Assume the second phase circuit of TIBC has an open-circuit sampling rates of the converter are the same as that in Section IV.
fault, and the simulation results are presented in Fig. 10. It can In experimental tests, to further evaluate the effectiveness of
be observed from Fig. 10 that when the open-circuit fault occurs the proposed AMPC method, the AM-MPC method and MPC
at 0.2 s, the inductor current of the fault phase drops to zero, and with integrator (IMPC) are chosen as benchmarks in subsequent
the inductor current of the healthy phase gradually rises to meet tests. The AM-MPC is realized according to [29], which utilizes
the load power. For the AMPC controller, the model parameter a DOB to estimate unknown terms and uses the steepest descent
a becomes larger after the fluctuation, this is because when the method to update the optimized change rate b̃. The IMPC uses
open-circuit fault occurs, the system efficiency is reduced and an integral variable to account for disturbance terms.
a larger inductor current is required to match the load. Besides, Since this article mainly focuses on the robustness to the large
the model parameter b gradually decreases from 2100 to 1160, external disturbance, to avoid the influence of parameter tuning
which is consistent with the analysis in Section III. For the AM- and enable a fair comparison, the three controllers are tuned
MPC controller, the model parameter b also changes from 2100 to perform similar tracking performance. The parameters of
to 1560. However, as can be observed, in the transient process, the three controllers are given in Table II, and their tracking
the AMPC controller still has a smaller voltage drop and shorter performances are shown in Fig. 12, where the reference voltage
recovery time than AM-MPC. steps from 48 to 56 V and then steps back to 48 V. It can be

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
1728 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 38, NO. 2, FEBRUARY 2023

Fig. 12. Tracking performance of three controllers.

Fig. 14. Experimental results of converter under sinusoidal load disturbance.

It can be seen from Fig. 13 that when the load current steps,
the output voltage will deviate from the reference value. All
the three controllers can regulate the output voltage to the
reference value after load changes. But among these three
controllers, the proposed AMPC controller shows the smallest
voltage drop/rise and the shortest recovery time. Therefore, the
AMPC controller shows the strongest robustness against step
load disturbance.

B. Case II: Time-Varying Load Disturbance

To test the robustness of controllers against the time-varying
disturbance, the load current is set to io = 6 + 2sin(20πt)A.
The experimental results of TIBC are shown in Fig. 14.
It can be seen from Fig. 14 that when the load current changes
Fig. 13. Experimental results of TIBC under step load changes. sinusoidally, the output voltage also varies with the load current.
For the IMPC controller, the output voltage varies from 46.6 to
49.8 V. For the AM-MPC controller, the output voltage varies
from 47.1 to 49.4 V. However, for the AMPC controller, the
seen from Fig. 12 that all three controllers exhibit similar and output voltage only varies from 47.4 to 49 V. The proposed
satisfactory tracking performance. Then, the robustness of the method exhibits the strongest robustness against time-varying
three controllers will be compared under external disturbances. load disturbance.

A. Case I: Step Load Disturbance C. Case III: Hardware Open-Circuit Fault

The robustness of controllers against step load disturbance is The robustness of controllers against hardware open-circuit
tested in this part. The load current is set to change between 7 fault is tested. Assume the second phase circuit of the converter
and 3.5 A periodically. The experimental results of TIBC under has an open-circuit fault, and the experimental results of TIBC
different controllers are shown in Fig. 13. are shown in Fig. 15.

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: ADAPTIVE MPC OF AN INTERLEAVED BOOST CONVERTER USING REAL-TIME UPDATED MODEL 1729

TABLE III
QUANTITATIVE PERFORMANCE COMPARISON OF THREE CONTROLLERS

TABLE IV
EXECUTION TIME OF DIFFERENT ADAPTIVE MPC METHODS

To evaluate the computational burden of the AMPC method,

the execution time is compared with IMPC, AM-MPC, and the
RLSs-based AMPC (RLS-AMPC) in [22]. All four controllers
are implemented for the TIBC on DSP28335 (150 MHz). The
execution time of the parameter update module and predictive
control law is given in Table IV.
Fig. 15. Experimental results of TIBC under open-circuit fault.
As can be observed from Table IV, the execution time of
the proposed parameter update mechanism is shorter than those
of the AM-MPC and RLS-AMPC. Since an additional DOB
As can be observed from Fig. 15, when an open-circuit fault
is utilized in AMPC and AM-MPC, the execution time of the
occurs, the inductor current of the fault phase iL2 drops to zero,
predictive control law is larger than that of RLS-AMPC. The
and the inductor current of the healthy phase iL1 gradually rises
total execution time of AMPC is shorter than that of AM-MPC
to match the load power. The healthy phase inductance takes
and RLS-MPC, but larger than that of IMPC since IMPC has
on all the load power, leading to a larger current ripple. Since
a simple structure. However, the increase in execution time is
AM-MPC also contains an optimized change rate b̃, it has a
acceptable considering the improved control performance.
similar control performance to AMPC, but AMPC shows the
smallest voltage drop and the shortest recovery time among these
three controllers. This indicates the proposed AMPC method D. Stability Verification
exhibits the strongest robustness against open-circuit faults.
To quantitatively compare the control performances of three As can be observed from the previous experiments in cases
controllers, the integral of time multiplied by the absolute value I– III, when the prediction time is selected as Ts = 4 ms, the
of error (ITAE) and root-mean-square error (RMSE) [30] are system can converge to the reference value since the inequality
used as performance indexes, calculated as Ts ≤ 3/(2â) in Remark 1 holds.
In this part, the load resistance is set to 13.7Ω, and according
 to (7), a = 310, b = 2000. The prediction time is selected as
 t  t
1 0.01s to satisfy Ts > 3/(2â). According to Remark 1, the upper
ITAE = t |yr − y| dτ , RMSE = (yr − y)2 dτ .
0 0 t limit of the weight ratio is derived as 18.7. The experimental
(33) results of the proposed method with Ts = 0.01s are presented
The performance indexes of three controllers for the above in Fig. 16, where the weight ratio changes from 4 to 20.
three test cases are given in Table III. It can be observed from As can be observed from Fig. 16, when the weight ratio is 4,
Table III that the proposed AMPC method has an obvious the output voltage can be regulated to 48 V, and the duty ratio is
performance improvement for power converters. Specifically, slightly higher than 0.5 due to the effect of parasitic parameters.
for Case I, compared with AM-MPC, the ITAE and RMSE of When the weight ratio R/Q changes to 20, the system becomes
AMPC are reduced by 36.2% and 52.7%. For case II, compared unstable, the output voltage drops to 24V, and the duty ratios of
with AM-MPC, the ITAE and RMSE of AMPC are reduced by PWM blocks drop to 0.
35.7% and 55.9%. For case III, compared with AM-MPC, the Therefore, the experimental results agree with the simulation
ITAE and RMSE of AMPC are reduced by 27.4% and 42.2%. results, which demonstrates the stability analysis in Section III.

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
1730 IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 38, NO. 2, FEBRUARY 2023

Substitute T0 into (A.1), it can be rewritten as

Nu = 12Qb̂2 T02 + (240 − 160âT0 ) R. (A.5)
According to (A.4)-(A.5), for T0 ≤ 3/(2â), k1 > 0 always
holds; for T0 > 3/(2â), the necessary and sufficient condition
for k1 > 0 is
3b̂2 T02
R/Q < . (A.6)
20 (2âT0 − 3)
Fig. 16. Experimental results of the AMPC with Ts = 0.01 s.
B. Theorem 1
VI. CONCLUSION Define the voltage tracking error as ev = y(t) − yref , and
To mitigate model mismatches and enhance the robustness to combine (8), (14), and (23), the closed-loop error system is
disturbances, an AMPC method is developed for an interleaved Ėv,r = Aε Ev,r + Bε ẇ (A.7)
boost converter. The adaptive predictive model is updated at  
each sampling period using a parameter update mechanism where Ev,r = [ev , er ]T , Aε = −b̂k0
1 1
−L0
, Bε = [0, 1]T .
and a DOB. Hence, the proposed method is adaptive to the Since Aε is Hurwitz stable, the error system (A.7) is bounded
current operating point. An explicit control law is derived by and its bound depends on ẇ. With the assumption that ẇ = 0
solving the optimization problem offline, which helps reduce as t → ∞, the error system is asymptotically stable.
the computational burden of the microprocessor.
The simulation and experimental studies are conducted in the REFERENCES
laboratory. The results indicate that compared with the IMPC
[1] O. Hegazy, J. V. Mierlo, and P. Lataire, “Analysis, modeling, and im-
and AM-MPC methods, the control performance of the AMPC plementation of a multidevice interleaved DC/DC converter for fuel cell
method has been improved significantly. The ITAE and RMSE hybrid electric vehicles,” IEEE Trans. Power Electron., vol. 27, no. 11,
of AMPC are reduced by 35.7% and 52.7% under load changes, pp. 4445–4458, Nov. 2012.
[2] Q. Xu, W. Jiang, F. Blaabjerg, C. Zhang, X. Zhang, and T. Fernando,
and the ITAE and RMSE of AMPC are reduced by 27.4% and “Backstepping control for large signal stability of high boost ratio inter-
42.2% when an open-circuit fault occurs. Besides, the execution leaved converter interfaced DC microgrids with constant power loads,”
time of the AMPC method on DSP28335 (150 MHz) is 8.55μs, IEEE Trans. Power Electron., vol. 35, no. 5, pp. 5397–5407, May 2020.
[3] A. Cid-Pastor, R. Giral, J. Calvente, V. I. Utkin, and L. Martinez-Salamero,
which is acceptable for the studied converter with a sampling “Interleaved converters based on sliding-mode control in a ring con-
period of 0.4 ms. figuration,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 58, no. 10,
pp. 2566–2577, Oct. 2011.
[4] Y. Huangfu, Q. Li, L. Xu, R. Ma, and F. Gao, “Extended state ob-
APPENDIX server based flatness control for fuel cell output series interleaved boost
converter,” IEEE Trans. Ind. Appl., vol. 55, no. 6, pp. 6427–6437,
A. Remark 1 Nov./Dec. 2019.
[5] Q. Xu, N. Vafamand, L. Chen, T. Dragičević, L. Xie, and F. Blaabjerg,
For k1 , the sign of its numerator is determined by “Review on advanced control technologies for bidirectional DC/DC con-
verters in DC microgrids,” IEEE J. Emerg. Sel. Topics Power Electron.,
Nu = 12Qb̂2 Ts2 − 160RâTs + 240R. (A.1) vol. 9, no. 2, pp. 1205–1221, Apr. 2021.
[6] Y. Xie, R. Ghaemi, J. Sun, and J. S. Freudenberg, “Model predictive control
In this case, Nu can be considered as a quadratic function of for a full bridge DC/DC converter,” IEEE Trans. Control Syst. Technol.,
prediction time Ts , and the condition for Nu > 0 with any Ts is vol. 20, no. 1, pp. 164–172, Jan. 2012.
obtained as [7] S. J. Qin and T. A. Badgwell, “A survey of industrial model predictive
  control technology,” Control Eng. Pract., vol. 11, no. 7, pp. 733–764,
R/Q < 9b̂2 / 20â2 . (A.2) Jul. 2003.
[8] T. Geyer and D. E. Quevedo, “Performance of multistep finite control
set model predictive control for power electronics,” IEEE Trans. Power
For simplicity, define aT = â Ts , bT = b̂ Ts . The denomina- Electron., vol. 30, no. 3, pp. 1633–1644, Mar. 2015.
tor of k1 can be expressed as [9] J. Scoltock, T. Geyer, and U. K. Madawala, “Model predictive direct power
control for grid-connected NPC converters,” IEEE Trans. Ind. Electron.,
De = 3Q2 b4T + 48QRa2T b2T − 96QRaT b2T vol. 62, no. 9, pp. 5319–5328, Sep. 2015.
[10] M. M. Ismail, W. Xu, J. Ge, Y. Tang, A. K. Junejo, and M. G. Hussien,
+ 104QRb2T + 240R2 . (A.3) “Adaptive linear predictive model of an improved predictive control of
permanent magnet synchronous motor over different speed regions,” IEEE
Then, the following inequality holds Trans. Power Electron., vol. 37, no. 12, pp. 15338–15355, Dec. 2022.
  [11] R. Errouissi, A. Al-Durra, and S. M. Muyeen, “A robust continuous-time
De > f (aT ) = QRb2T 48a2T − 96aT + 104 . (A.4) MPC of a DC–DC boost converter interfaced with a grid-connected
photovoltaic system,” IEEE J. Photovolt., vol. 6, no. 6, pp. 1619–1629,
Hence, f (aT ) is a quadratic function of aT , and its minimum Nov. 2016.
value is always greater than zero. Therefore, when the condition [12] S. Kwak, U. -C. Moon, and J.-C. Park, “Predictive-control-based direct
power control with an adaptive parameter identification technique for
in (A.3) is satisfied, the control gain k1 is always positive. improved AFE performance,” IEEE Trans. Power Electron., vol. 29, no. 11,
It is worth noting that (A.2) is a sufficient condition for k1 > 0, pp. 6178–6187, Nov. 2014.
which can ensure the control gain k1 is positive for any prediction [13] T. Dorfling, H. du Toit Mouton, T. Geyer, and P. Karamanakos, “Long-
horizon finite-control-set model predictive control with nonrecursive
time Ts . However, for a given prediction time T0 , the range of sphere decoding on an FPGA,” IEEE Trans. Power Electron., vol. 35,
R/Q for k1 > 0 will be larger than (A.2). no. 7, pp. 7520–7531, Jul. 2020.
Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.
ZHANG et al.: ADAPTIVE MPC OF AN INTERLEAVED BOOST CONVERTER USING REAL-TIME UPDATED MODEL 1731

[14] R. O. Ramírez et al., “Finite-state model predictive control with integral Yuren Li received the master’s degree in electrical en-
action applied to a single-phase Z-source inverter,” IEEE J. Emerg. Sel. gineering in 1989 from Northwestern Polytechnical
Topics Power Electron., vol. 7, no. 1, pp. 228–239, Mar. 2019. University (NPU), Xi’an, China, where he received
[15] M. E. Albira and M. A. Zohdy, “Adaptive model predictive control for the Ph.D. degree in detection technology and automa-
DC-DC power converters with parameters’ uncertainties,” IEEE Access, tion device, in 2006.
vol. 9, pp. 135121–135131, 2021. He is currently a Full Professor with the School of
[16] X. Liu, L. Qiu, Y. Fang, K. Wang, Y. Li, and J. Rodríguez, “A fuzzy Automation with NPU. His main research interests
approximation for FCS-MPC in power converters,” IEEE Trans. Power include aircraft brake control, power system, and
Electron., vol. 37, no. 8, pp. 9153–9163, Aug. 2022. automation technology.
[17] B. Zhu, Z. Zheng, and X. Xia, “Constrained adaptive model-predictive
control for a class of discrete-time linear systems with parametric un-
certainties,” IEEE Trans. Autom. Control, vol. 65, no. 5, pp. 2223–2229,
May 2020.
[18] K. Yin, L. Gao, R. Chen, Z. Feng, and S. Liu, “Adaptive deadbeat predictive
current control for PMSM with feed forward method,” IEEE Access, vol. 9, Renyou Xie (Student Member, IEEE) received the
pp. 101300–101310, 2021. master’s degree in electrical engineering in 2021 from
[19] X. An, G. Liu, Q. Chen, W. Zhao, and X. Song, “Adjustable model Northwestern Polytechnical University, Xi’an, China,
predictive control for IPMSM drives based on online stator inductance where he received the Ph.D. degree in electrical en-
identification,” IEEE Trans. Ind. Electron., vol. 69, no. 4, pp. 3368–3381, gineering.
Apr. 2022. His current research interests include fuel cells,
[20] W. Wu et al., “Data-driven iterative learning predictive control for machine learning, and distributed optimization.
power converters,” IEEE Trans. Power Electron., vol. 37, no. 12,
pp. 14028–14033, Dec. 2022.
[21] O. Babayomi et al., “Adaptive predictive control with neuro-fuzzy pa-
rameter estimation for microgrid grid-forming converters,” Sustainability,
vol. 13, no. 13, 2021.
[22] L. Wang, Y. Wang, C. Liu, D. Yang, and Z. Chen, “A power distribution
strategy for hybrid energy storage system using adaptive model predictive
control,” IEEE Trans. Power Electron., vol. 35, no. 6, pp. 5897–5906, Jian Song (Student Member, IEEE) received the
Jun. 2020. bachelor’s and master’s degrees in electrical engineer-
[23] Y. Yang, S. -C. Tan, and S. Y. R. Hui, “Adaptive reference model predictive ing in 2021 from Northwestern Polytechnical Uni-
control with improved performance for voltage-source inverters,” IEEE versity, Xi’an, China, where he is currently working
Trans. Control Syst. Technol., vol. 26, no. 2, pp. 724–731, Mar. 2018. toward the master’s degree in electrical engineering.
[24] J. Zhidong et al., “An adaptive model predictive control for DC-DC boost His current research interests include energy man-
converters,” Proc. CSEE, vol. 38, no. 19, pp. 5838–5941, Oct. 2018. agement strategies and renewable energy technolo-
[25] L. Po, L. Ruiyu, S. Tianying, Z. Jingrui, and F. Zheng, “Composite adap- gies.
tive model predictive control for DC–DC boost converters,” IET Power
Electron., vol. 11, no. 10, pp. 1706–1717, Aug. 2018.
[26] H. Makhamreh, M. Trabelsi, O. Kükrer, and H. Abu-Rub, “A Lyapunov-
based model predictive control design with reduced sensors for a PUC7
rectifier,” IEEE Trans. Ind. Electron., vol. 68, no. 2, pp. 1139–1147,
Feb. 2021.
[27] J. Yang, W. X. Zheng, S. Li, B. Wu, and M. Cheng, “Design of a Bo Liang received the Ph.D. degree in detection
prediction-accuracy-enhanced continuous-time MPC for disturbed sys- technology and automation device from Northwest-
tems via a disturbance observer,” IEEE Trans. Ind. Electron., vol. 62, no. 9, ern Polytechnical University (NPU), Xi’an, China, in
pp. 5807–5816, Sep. 2015. 2014.
[28] W.-H. Chen, D. J. Ballance, and P. J. Gawthrop, “Optimal control of He is currently an Associate Research Fellow with
nonlinear systems: A predictive control approach,” Automatica, vol. 39, the School of Automation, NPU. His main research
no. 4, pp. 633–641, 2003. interests include measurement and control technol-
[29] F. Wang, L. He, J. Kang, R. Kennel, and J. Rodriguez, “Adaptive model ogy, electric power system, and power electronics.
predictive current control for PMLSM drive system,” IEEE Trans. Ind.
Electron., to be published, doi: 10.1109/TIE.2022.3179550.
[30] J. Wang and L. Yu, “Adaptive resonant EIDO based optimized position pre-
cision control for magnetic levitation system,” IEEE Trans. Ind. Electron.,
2022, to be published, doi: 10.1109/TIE.2022.3186348.

Yigeng Huangfu (Senior Member, IEEE) received

the M.S. and Ph.D. degrees in electrical engineering
from Northwestern Polytechnical University (NPU),
Xi’an, China, in 2007 and 2009, and the Ph.D. de-
Hongyu Zhang (Student Member, IEEE) received
gree from the University of Technology of Belfort-
the master’s degree in electrical engineering in 2021 Montbéliard, Belfort, France, in 2010.
from Northwestern Polytechnical University, Xi’an,
He has been a Full Professor with NPU. He is
China, where he is currently working toward the Ph.D.
also the Chair of IEEE IES Technical Committee
degree in electrical engineering.
on Transportation Electrification. His main research
His current research interests include power sys- interests include power electrical conversion, trans-
tem, optimal control, and power electronics.
portation electrification, intelligence control of new
energy conversion and renewable energy generation technology.
Dr. Huangfu is an Associate Editor for IEEE INDUSTRIAL ELECTRONICS
APPLICATION and IET Power Electronics.

Authorized licensed use limited to: UNIVERSIDAD DE VIGO. Downloaded on August 21,2023 at 09:19:57 UTC from IEEE Xplore. Restrictions apply.