Model Reference Adaptive Control Design for a Shunt Active Power Filter System

*Kuo-Kai Shyu, Ming-Ji Yang, Yen-Mo Chen and Yi-Fei Lin

Department of Electrical Engineering

National Central University
Chung-Li City, Taoyuan County 320, Taiwan R.O.C.
*E-mail [email protected]

Abstract – In this paper, model reference adaptive control significant switching losses. Consequently, the shunt APF
(MRAC) is proposed for a single-phase shunt active power is considered to be the most basic structure for APF
filter (APF) to improve line power factor and to reduce line [5]-[7].
current harmonics. The proposed active power filter In the past, most active power filter uses proportional
controller forces the supply current to be sinusoidal with low
plus integral and differential (PID) control in the
current harmonics and in phase with the line voltage. The
advantages of using MRAC over conventional proportional- current/voltage control loop [7]-[9]. When using
integral (PI) control are more flexibility, adaptive and fixed-gains PID controllers, it is necessary to retune them
robustness; moreover, MRAC can self-tune the controller for different operation regions, not to mention change in
gains to assure the system stability. Since the APF is a different loading power. Furthermore, recent applications
bilinear system, it is hard to design the controller. This paper need faster transient response, minimum power dissipation,
will solve the stability problem when linearization method is and robustness, which the PID fails to satisfy.
used to approach the nonlinearity of the system. Besides, In view of the aforementioned problems of PID
using Lyapunov’s stability theory and Barbalat’s Lemma, an controlled APF, this study adopts model reference adaptive
adaptive law is designed to guarantee asymptotic output
control (MRAC) [10]. MRAC is a well-established method
tracking for the system. To verify the proposed APF system,
a digital signal controller is adopted to implement the that has demonstrated its capabilities in many researches;
algorithm of MRAC and an 1-kVA laboratory prototype is for example [11]-[14]. In MRAC systems, the desired
built to test the feasibility. performance of the plant is expressed via a reference
model, which gives the desired response to a command
Index Terms

Active power filter, model reference

signal. Hence, considerable flexibility is granted to the

adaptive control, power factor, harmonics. designer to alter the goals by modifying the reference
model.
I. INTRODUCTION This paper presents an MRAC APF based to cancel the
harmonic/reactive components in the line current so that
With the increasing of nonlinear loads in utility line, the current flow into and from the gird is sinusoidal and in
harmonic problem has been concerned ever more than phase with the gird voltage. Since the APF is a bilinear
before. Those nonlinear loads, such as diode rectifiers, system, it is difficult to apply MRAC directly. This paper
thyristor converters and some electronic circuits, on will solve the stability problem when linearization method
industrial, commercial and residential equipments, is used to approach the nonlinearity of the system.
drawing non-sinusoidal currents, pollute the utility line due Parameter update laws for tuning the control gains are
to the current harmonics generated by them. They have derived by the Lyapunov function. Furthermore, the
brought about many problems in utility power, such as: stability of MRAC is guaranteed based on the Lyapunov
low power factor, low energy efficiency, interference by stability and Barbalat’s Lemma. MRAC developed in
EMI and distortion of the line voltage, etc. [1]. Therefore, theory so far is mostly based on a higher order algorithm
the standard regulations and recommendations such as IEC that requires much more complicated mathematical
61000-3-2 [2] and IEEE 519 [3], enforce to limit the above processing and far exceeds the capability of real-time
problems. For the problems of harmonic pollution, passive processing of the previous microprocessors. On the other
and active power filters are typical approaches used to hand, analog circuits can only keep the fixed parameters
improve the power factor and to eliminate harmonics. In under invariant environmental condition. In other words,
the past passive LC filters are generally used to reduce they are subject to variation of controlled parameters in the
these problems but they have many demerits such as bulky, presence of environmental changes, temperature rise,
heavy, resonance, tuning problem, fixed compensation, change in humidity, etc. Fortunately, the present-day
noise, increased losses, etc. [4]. On the contrary, the active development in microprocessors and their ability enough
power filter can solve above problems and is often used to to operate in conjunction with power electronic devices
compensating current harmonics and low power factor has opened up an era of high power quality. Consequently,
caused by nonlinear load. In active power filter (APF) we adopted a digital signal controller (dsPIC30F4012) [16]
connection, it was roughly classified in series (series APF) to process operation of MRAC for APF in this paper. The
and in parallel (shunt APF). Typical series APF is in the main advantage of using digital signal controller is more
cascade path of the power, compare with shunt APF, it flexibility to modify parameters of the reference model and
processes all the power and thus requires more high then enhance performance of the APF. Besides, the
current and voltage rating of devices and involves complexity of the hardware circuit is greatly reduced by

1-4244-0136-4/06/$20.00 '2006 IEEE 73

 iS io
iC iR L iC iR L
iL
iL iL
vS vS vx RL vS vx RL

Q1 Q3 iC iRL
(a) (b)
vL vx C vC RL

Q2 Q4
ip

iL
Fig. 1 A shunt APF in parallel with nonlinear load.
− ip

using digital controller. vS

Finally, an 1-kVA prototype was developed to vx −vS
demonstrate the performance of the proposed APF with the vL
MRAC approach. The experimental results show this vS − vx
system can both improve the power factor and greatly −vS
reduce the total harmonic distortion (THD). vS >0 vS <0

II. OPERATION PRINCIPLE OF THE SHUNT APF (c)

The power stage of a shunt power filter needs to pass Fig. 2 Equivalent circuit, inductor current and voltage waveforms
bidirectional current and it is typically composed of a full in each switching cycle for the APF. (a) Equivalent circuit during
bridge or half bridge with an energy storage capacitor at 0 t DTS, (b) Equivalent circuit during DTS t TS, (c)
   

the dc side. In this paper, H-bridge is adopted for the Inductor current and voltage waveforms.
power stage structure of the APF as shown in Fig. 1. From
Fig. 1, the H-bridge is in parallel with a nonlinear load. waveforms of the inductor current and voltage in each
The H-bridge circuit is the same as a single-phase of switching cycle.
voltage-source inverter (VSI) while working as APF (or as Observing the equivalent circuits shown in Fig. 2, one
a PWM rectifier). The APF system connected in parallel has the inductor voltage and current during one switching
with the load could cancel the harmonic/reactive cycle when vS > 0 , are given by
components in the line current (is) so that the current flow
into and from the power line is sinusoidal and in phase v L (t ) = vS + vC
with the power line voltage. In other words, the (2)
1 DT S for 0 ≤ t ≤ DTS ,
compensating current (iL) is injected into the line to force iL (t ) = iL (0) + (v S + vC )dt
L 0
the line current (is) approach to sinusoid and
unity-power-factor can be achieved by APF system. The vL (t ) = vS − vC
currents of the APF system can be expressed as (3)
1 TS for DTS ≤ t ≤ TS .
iL (t ) = iL ( DTS ) + (vS − vC ) dt
(1) L DTS
is = io + i L
Similar results can be obtained when vS<0. Because the
where io is the nonlinear load current. switching frequency is high enough, the load current is
In this paper, the H-bridge operates in bipolar PWM almost unchanged at one switching cycle. Therefore, the
mode. In the H-bridge, the two switch legs are initial and peak values of the inductor current equal those
complementary switching states, respectively. In addition, of the next switching interval. According to the
the operation of the APF can be divided into two modes volt-second balance for inductor at one switching cycle,
and its four switches have the switching frequency of fs. In the average of the inductor voltage is given by
Mode 1, Q2 and Q3 are turned on, while Q1 and Q4 are
turned off during 0<t<DTs, where Ts=1/fs is the switching (vS + vC ) DTS + (vS − vC )(1 − D )TS = 0 . (4)
period and D=Ton/Ts is the duty ratio. In this mode, the
inductor current iL increases in a positive direction and the It implies that
magnetic energy is stored in the inductor L. At the same vS .
time the energy stored in the capacitor C is transferred to vC = (5)
1 − 2D
the line source and the assumptive load RL. In Mode 2, the
switching states of four switches in Mode 1 are reversed Single-phase APF has two reactive elements due to its
during DTs<t<Ts. Conversely, the energy stored in L is output voltage higher than input voltage; otherwise it is not
transferred to C, RL and the line source, and C is charged possible to return energy from capacitor to the load.
accordingly. Fig. 2 shows their equivalent circuits, related Consequently, in bipolar operation mode H-bridge APF

74
represents absolute boost converter with possibility of In the averaged model, the duty ratio “u” can take any
returning energy. value between 0 to 1. The dynamic behavior of the APF
can be described by the following state-space model as
III. DYNAMIC MODEL OF THE APF x = Fx + Gxu + Ew (16)
To design an APF system, its dynamical model must be −1
0 2
given firstly. This section develops a model for the where x = [iL vC ]T , L
, 0 ,
F= 1 G= L
voltage-source inverter (VSI) used as the APF. Figs. 3(a) −1 −2
0
and 3(b) show the equivalent circuits of the VSI at one C RL C C
switching cycle. Define the switching function uS of each T

E=
1
0 and w = vS .
VSI switch to be
L
1 when Q S is turned on Now consider the bilinear state equation (16). If x = x0
uS = (6)
0 when Q S is turned off and u = u0 suffice to f ( x 0 , u 0 ) = Fx0 + Gx 0 u 0 + Ew(t ) = 0
, ( x0 , u 0 ) is called its equilibrium point or operating point.
where S=1-4 is denoted the switch number of the VSI. For
the bipolar PWM scheme, the relationship between the According to the differential equations (14) and (15), the
high and low side switches in the two switch legs can be equilibrium values of the inductor current and the
expressed by capacitor voltage can be obtained as

u1 + u 2 = 1 vS , (17)
(7) x 02 =
u 3 + u 4 = 1. (1 − 2u 0 )

Given the definition of uS for each switch, the relation x 02 (18)

x 01 =
between vx and vC can be expressed as R L (1 − 2 u 0 )

v x = [ u 1 − u 2 ]v C . (8) where x02 and x01 are the equilibrium values of vC and
Accordingly, the capacitor current can be related to the iL , respectively. The average value of the duty ratio u0
inductor current by can be obtained from (17) as
iC + iR = [u1 − u 2 ]iL . (9) 1 v
u0 = 1− S . (19)
2 x02
Therefore, the state equations of the inductor current
and the capacitor voltage are written by Kirchhoff’s laws:
To simplifying a nonlinear system model, a common
1 approach is the linearization method [15]. The following
iL = [ v S − uv C ], (10) will show how the bilinear APF system is linearized. To
L
linearize a bilinear state equation (16) about a constant
1 v (11) equilibrium point ( x0 , u0 ) , let
vC = [ ui L − C ]
C RL
0 = f ( x0 , u 0 , w, t ) = Fx0 + Gx0 u0 + Ew(t ) . (20)
where u=(u1 u2). Dynamical equations (10) and (11) are


hard to be directly controlled due to the control input u is a One can expand the right-hand side of (20) into Taylor
two state (ON-OFF) function. This drawback could be series about ( x0 , u 0 ) and then neglect the high-order
removed if the dynamical equations are converted into the terms.
average model during a switching cycle. Define
∂f ∂f (21)
1 t + TS
(12) x ≈ f ( x0 , u 0 ) + ( x − x0 ) + (u − u0 )
x1 ( t ) = i L (τ )d τ , ∂x ux==ux0 ∂u ux==ux0
TS t 0 0

1 t +TS
(13) Moreover, since the trajectories near ( x0 , u 0 ) are
x 2 (t ) = v C (τ )d τ
TS t
interested, one defines

where x1 (t) and x 2 (t ) represent the average values of the xδ = x − x 0 , u δ = u − u 0 . (22)

inductor current and the capacitor voltage respectively
over a switching period. The averaged state-space model The bilinear state equation (16) is approximately described
of the converter can be written as [17] by a linear state equation of the form

dx1 ( 2u − 1) x2 vS (14) xδ = ( F + Gu 0 ) xδ + (Gx 0 )u δ ≡ A p x δ + B p u δ (23)

= + ,
dt L L
where Ap = ( F + Gu0 ) and B p = Gx0 are the linear fits to
dx2 (1 − 2u ) x1 x (15)
= − 2 . the nonlinear function f ( x, u, w, t ) at x0 and u0.
dt C RLC

75
 IV. ADAPTIVE CONTROLLER DESIGN are positive-definite symmetric matrices. Because Ar is a
Hurwitz stable matrix, there exists a unique
In the following, the MRAC scheme with adaptive
control for APF is presented to improve the power factor positive-definite symmetric matrix P ∈ R 2× 2 satisfying
performance subject the change of nonlinear load. The [15]
control objective is either to regulate the error states to A rT P + PA r = − Q (31)
zero or drive the system states to track the states of a
reference model. The adaptive control is to adjust the where Q = QT ∈ R 2×2 > 0 is positive-definite.
controller parameters on-line, based on measured system The derivative of V (t ) along the trajectories of (29) is
response. Thus, one defines a linear time-invariant (LTI)
given by
reference model as
1 T (32)
V =− e Qe + Z T Φ T B Tp Pe + tr Φ Γ −1Φ T
xr = Ar xr + Br r (24) 2
where r ∈ R1 is the reference input, xr ∈ R 2×1 is the state or
vector of the reference model, Ar ∈ R 2×2 and Br ∈ R 2×1 1 T
V =− e Qe + trB Tp PeZ T Φ T + tr Φ Γ −1Φ T . (33)
are the reference model system matrix. Let the tracking 2
error be Accordingly, let
e = xr − xδ . (25)
(34)
Φ = − BTp PeZ T Γ
The objective here is to find an adaptive law so that the
i.e.
tracking error e approaches zero. To achieve the control
objective, the output of the adaptive controller is inferred φ = − B Tp Per T Γ1
by the system model (23) as (35)
ϕ = − B Tp Pex δT Γ2 .
r
[
uδ = θ f θ b ]⋅
xδ
≡ Θ⋅ Z (26)
Substituting (34) into (33) yields
1
where θ f (t ) and θb (t) = [θb1(t) θb2 (t)] are the feedforward V (t ) = − eT Qe ≤ 0 . (36)
2
and feedback gains of the closed-loop system, respectively.
The above equation only ensures that V (t ) is always
Substituting (26) into (23), one has
negative and for all initial conditions, e is bounded.
( )
xδ = Ap + B pθb xδ + ( B pθ f )r . (27) Since V (t ) is monotonically decreasing and bounded,
then integration of V (t ) yields e ∈ L2 . That is
If the feedforward gain θ f (t ) and feedback gain θ b (t )
are converged the optimal parameters θ *f and θ b* ∞
V (t )dt = V (∞) − V (0) (37)
0
respectively by the adaptive controller so that the plant
model match with the reference model, satisfying and from (36)
1
eQeT dt ≤ V (0) , (38)
∞

B pθ *f = Br (28) 2 0

Ap + B pθ b* = Ar . since e is bounded, so that V (0) ∈ L∞ . According to

Barbalat’s lemma [15], e → 0 asymptotically as t → ∞ .
Differentiating (25) with respect to time and combining
Therefore, we can conclude that the system is
with (24) (28), one has
asymptotically stable.


( ) (
e = Ar e + B pθ b* − B pθ b xδ + B pθ *f − B pθ f r ) (29)
However, adaptive laws (35) are not implementable in
practice, because they are not directly related to the control
≡ Ar e + B p Φ Z parameters of (26). To make adaptive laws be practical,
taking the derivative of φ = θ *f − θ f (t) and ϕ = θb* − θb (t )
where
become
Φ = [φ ϕ ] , φ = θ *f − θ f (t) and ϕ = θ b* − θ b ( t ) are
defined the parameter errors. φ = −θ f
(39)
Consider the following Lyapunov function candidate for ϕ = −θ b .
the system (29) as
Comparing (35) with (39), one has the parameter adaptive
V =
1 T
2
[
e Pe + tr ΦΓ −1Φ T ] (30) laws as follows:
θ f = B Tp Per T Γ1
where (40)
Γ1 0 θ b = B Tp Pex δT Γ2 .
Γ= , Γ1 = Γ1T ∈ R1×1 > 0 and Γ2 = Γ2T ∈ R2×2 > 0
Fig. 3 shows the block diagram of the proposed
0 Γ2 MRAC-based APF system.

76
 xr
xr = Ar xr + Br r

Σ
uδ xδ
θf Σ xδ = Ap xδ + Bpuδ

θb

xδ
B Tp PexδT Γ2 dt

B Tp Per T Γ1dt

Fig. 3 Control block of the proposed MRAC approach.

TABLE I
EXPERIMENTAL PARAMETERS AND CONDITIONS

Parameters Values
Power line v S = AC 110Vrms / 60Hz 1φ Fig. 4 Overall hardware structure of the MRAC-APF.
Output voltage vC = 250V
Prototype power rating 1-kVA
PWM switching frequency f S = 20 kHz
Input inductance L = 6 mH
Output capacitance C = 1000 F
Output resistance R L = 10 k

V. EXPERIMENTAL VERIFICATIONS

The overall hardware structure of the MRAC-APF is

shown in Fig. 4. Parameters for experiments are given in
Table I. To actually demonstrate the performance of the
proposed APF with the MRAC approach, an 1-kVA
prototype is developed. System parameters are taken as
follows:
0 − 73.3 83333.3 − 72 −18 1 (a)
Ap = , Bp = , Ar = , Br = .
440 − 0.1 − 800 18 0 0
The positive-definite matrices P and Q are
2 .4533 0 .0833 1 0 Weighting scalar
P= , Q= .
0 .0833 2 .4474 0 1
matrices for the adaptation laws are given as
0.000183 0 .
Γ1 = 0.000125 , Γ2 =
0 0.000262
To verify the effective experiments of the same system
controller by the PI controller is given for comparison. Fig.
5 shows the test results of the PI-APF and the proposed
MRAC-APF compensating a diode rectifier with a RC
load. The load power is about 600W. The line current
THD of the PI-APF is dropped from about 85% to about
12% and the PF is improved from 0.65 to 0.98. The same
experiment is performed except that the proposed MRAC (b)
controller is used instead of the PI one. The measured
current THD is dropped from about 85% to about 6.7% Fig. 5 Measured voltage and current waveforms. (a)
and the PF is improved from 0.65 to 0.99. Comparing Fig. Experimental results of the PI-APF. (b) Experimental results of
5(a) with Fig. 5(b), it can be seen that the THD and PF the proposed MRAC-APF. Ch1: Line voltage vS. (400V/div,
controlled by PI are much improved by the proposed 5ms/div) Ch2: Load current io. Ch3: APF current iL. Ch4: Line
MRAC. current iS. (20A/div, 5ms/div)

77
 harmonic components of the load current, such that the
line current are near sinusoidal and the power factor is
increased.

VII. ACKNOWLEDGMENT

This work is supported by the National Science Council

of Taiwan under contract NSC 94-2213-E- 008-011.

VIII. REFERENCE
[1] N. Mohan, T. M. Undeland and W. P. Robbins, Power
Electronics: Converters, Applications, and Design. 2nd
Edition, John Wiley & Sons, Inc., 1996.
[2] CEI/IEC, IEC 61000-3-2 International Standard,
Fig. 6 Experimental waveforms of line voltage vS and line current International Electrotechnical Commission, 1998, Geneve,
iS for a step change in the nonlinear load. (From 200W to 600W.) Switzerland.
(vS: 50V/div, 10ms/div  iS: 10A/div, 10ms/div.) [3] IEEE Recommended Practices and Requirements for
Harmonic Control in Electrical Power Systems, IEEE Std.
519-1992, 1993.
[4] B. Singh, K. Al-Haddad, and A. Chandra, "A review of
active filters for power quality improvement," IEEE Trans.
Ind. Electron., vol. 46, pp. 960–971, Oct. 1999.
[5] W. M. Grady, M. J. Samotyj, and A. H. Noyola, "Survey
of active power line conditioning methodologies," IEEE
Trans. Power Del., vol. 5, pp. 1536–1542, July 1990.
[6] D. A. Torrey and A. M. Al-Zamel, "Single-phase active
power filters for multiple nonlinear loads," IEEE Trans.
Power Electron., vol. 10, no. 3, pp. 263–272, May 1995.
[7] S. Bhattacharya, T. M. Frank, D. M. Divan and B.
Banerjee, "Active filter system implementation," IEEE
Ind. Appl. Mag., vol. 4, pp. 47-63, Sep.-Oct. 1998.
[8] T. E. Nunez-Zuniga and J. A. Pomilio, "Shunt active
Fig. 7 Measured harmonics for the line current which complies power filter synthesizing resistive loads," IEEE Trans.
with the EN61000-3-2 Class D regulation. (The load power is Power Electron., vol. 17, no. 2, pp. 273–278, Mar. 2002.
about 600W.) [9] M. Basu, S. P. Das and G. K. Dubey, "Parallel converter
scheme for high-power active power filters," IEE Proc.
Fig. 6 shows the results of transient test under step Electr. Power Appl., vol. 151, no. 4, pp. 460–466, July
change in the nonlinear load current. From Fig. 6, one can 2004.
observe the test results without overshoot or undershoot at [10] S. Sastry and M. Bodson, Adaptive Control Stability,
Convergence and Robustness. Englewood Cliffs, NJ:
the transient. The robustness of the MRAC-APF can be
Prentice-Hall, 1989.
demonstrated by the results of transient test. Furthermore, [11] G. Ciccarella and P. Marietti, "Model reference adaptive
it is significant that the power devices avoid a danger control of a thermostatic chamber," IEEE Trans. Ind.
which is burned by the exceeding surge current. Moreover, Electron., vol. 36, no. 1, pp. 88–93, Feb. 1989.
the harmonics complies with the EN61000-3-2 Class D [12] M. Sunwoo, K. C. Cheok and N. J. Huang, "Model
regulation as shown in Fig. 7. Consequently, one can reference adaptive control for vehicle active suspension
clearly seen from these experimental results, no matter systems," IEEE Trans. Ind. Electron., vol. 38, no. 3, pp.
where in steady or transient-state, the supply current iS and 217–222, June 1991.
line voltage vS are in phase achieving unity power factor [13] P. Crnosija, Z. Ban and R. Krishnan, "Application of
model reference adaptive control with signal adaptation to
operation.
PM brushless DC motor drives," in Proc. IEEE
International Symp. Ind. Electron., 2002, vol. 3, pp.
VI. CONCLUSION 689–694.
[14] E. Leksono and Pratikto "Adaptive speed control of
In this paper, a single-phase shunt APF with MRAC has induction motor with DSP implementation," in Proc. IEEE
been proposed. We study the problem of developing a Ind. Electron. Conf., 2004, vol. 2, pp. 1423–1428.
[15] H. K. Khalil, Nonlinear systems. 2nd Edition, Upper
control law which can force the output of a bilinear plant
Saddle River, NJ: Prentice-Hall, 1996.
to track the output of a stable linear time-invariant (LTI) [16] dsPIC30F Family Reference Manual. Microchip
reference model. Furthermore, the MRAC approach Technology Inc., 2003.
improves many failures of the fixed-gains controller, such [17] V. B. Sriram, S. SenGupta and A. Patra, "Indirect current
as: some large overshoots and undershoots lead to burnout control of a single-phase voltage-sourced boost-type
of the devices at the transient, need to retune gains for bridge converter operated in the rectifier mode," IEEE
different operation regions and pressed for robustness, etc. Trans. Power Electron., vol. 18, no. 5, pp. 1130–1137,
The experimental results have demonstrated that the Sep. 2003.
proposed approach effectively eliminates the reactive and

78