Você está na página 1de 6

Optimized Resonant Control for Shunt Active Power Filters

Antonio Dell'Aquila, Maria Marinelli, Vito Giuseppe Monopoli, Agostino Lecci DEE - Politecnico di Bari BARI - ITALY

Abstract- This paper presents an optimization technique of the resonant controller based on the minimization of an objective function through the Nelder-Mead method. This function allows to evaluate the behavior of the system in steady state and transient conditions. The optimized controller has been compared with the Naslin polynomial based one. The results show that by means of the optimization technique it is possible to achieve better performances also in presence of harmonics the control system has not been designed for.

I. INTRODUCTION

The compensation for harmonics, reactive power, unbalance and neutral current which cause poor power factor and low system efficiency has become of remarkable importance with the proliferation of electronic equipment. The negative effects on the power system determined by the presence of such disturbances have encouraged the development of equipment able to improve the power quality and enhance the reliability of power utility. Among these equipments, the active power filters (APF) are the most effective because of their flexibility to load and network condition change. The most utilized configuration is the shunt active power filter shown in Fig. 1. It acts as a current generator by injecting into the network a current equal but opposite to the harmonic content of the load current. As a consequence, the supply feeds only the fundamental component. The effectiveness of shunt active filters greatly depends on the method to determine the current references, the current control strategy and the modulation technique. The methods for extracting harmonic references from the load currents are separated in compensation techniques in the frequency domain and compensation techniques in the time domain. The compensation techniques in the frequency domain are mainly based on the use of the Fourier harmonic analysis [1], requiring a large computational effort and consequently introducing a long time delay in the response. A large number of compensation techniques in the time domain are present in literature, but the most common are based on the PQ theory [2,3] and the Synchronous Rotating Frame (SRF)

or dq method [4,5]. These methods provide a faster response, ensuring a more effective compensation. Once the current references have been determined, the APF must have the capability to track accurately such references even in presence of sudden slope variations. The objective is the reduction of tracking error, such as in any feedback control, and this aim is supposed to be accomplished processing the error by means of standard proportional-integral (PI) controllers. In three phase systems such as the proposed application, several solutions have been adopted in order to achieve better results with reduced number of single phase controllers. Among the current control techniques that have demonstrated the best compensating performance there are the hysteresis control [6], the predictive control [7,8] and the use of single and multiple synchronous reference frames [9]. Synchronous controllers are based on PI compensators that operate on current components in rotating synchronous coordinates. In this rotating frame, fundamental error components become dc components that can be easily reduced to zero. Anyway this method has some key disadvantages: it requires two coordinates transformation and the knowledge of the synchronous frequency. Good results with a simpler control system can be obtained by means of a resonant controller, which can achieve zero error in the stationary coordinates [1012]. Some procedures to calculate the resonant controller parameters have been proposed in literature. Some of them allow to calculate the parameters directly in the stationary frame [11]. While some authors propose to carry on calculation in the rotating reference and then to transform the resulting parameters in the stationary frame[10, 12]. The aim of this paper is to present an optimization technique of the resonant controller by proposing a function to be minimized through the Nelder-Mead method. The controller parameters have been also calculated using an approach based on Naslin polynomial as proposed in [ 1]. The controller performance obtained using such a design procedure has been compared with the results achieved through the proposed optimization technique. In this way transient and steady-state current tracking has been improved. 877
EPE-PEMC 2006, Portoro2, Slovenia

1-4244-0121-6/06/$20.00 2006 IEEE

where Ki are the gains that should be chosen in order to tune the controller and wl, (05, 0w7 are the angular frequencies corresponding respectively to the fundamental, the fifth and the seventh order harmonic component. Several procedures have been proposed in literature for resonant controller gains design. In this work the method based on Naslin Polynomial is applied

[1 1].

Vd-

A. Naslin polynomial design procedure The design procedure based on Naslin Polynomial leads to the values of the controller gains by means of an intermediate stage where coefficients ci are computed. In fact it is possible to describe the transfer function of a resonant controller constituted by only a resonant term either by means of K gains or by means of ci coefficients:
G (s) = K
+
52sKb + C)2

Fig. 1- Scheme of an APF


II. RESONANT CONTROL

+C1S+C2S s2 + O2

(2)

As shown in [10], an equivalent single-phase stationary ac current controller which achieves the same dc control response as synchronous controllers can be designed. In order to achieve zero error in steady state, an infinite gain is needed at reference signal frequency. The resonant controller exhibits a function similar to an integrator for a specific frequency component. Thus, it can eliminate the steady-state control error completely [13]. In an active filter application, the compensating system should be capable to generate a waveform characterized by a high harmonic content. The harmonic content of the compensating current should be made up of all the undesired harmonics components or by the main polluting ones, that is those harmonic components with higher amplitude. In three-phase systems main polluting harmonics are the fifth and the seventh order ones. As a consequence, proper reference to track should be characterized by the fifth and the seventh order harmonic to deal with harmonic pollution and by an additional fundamental, which is output by the PI controller belonging to the external dc voltage control loop. In order to properly track such a highly complex reference waveform, the control system of the compensating device should ensure zero error at every harmonic component. As a consequence high gain in the transfer function of the error controller should be ensured at the corresponding frequencies. This goal can be achieved by means of a multiresonant controller, since its frequency response is characterized by infinite gain at resonant frequencies; as a consequence, in the transfer function of the current controller the presence is only needed of a resonant term for each specific harmonic order to be

where Ka and Kb are the gains to be designed and w, is the resonant frequency of the term, while ci are coefficients to be designed. The design procedure requires the comparison between the characteristic equation of the controlled system and the Naslin Polynomial related to the single resonant term. The design procedure will then be repeated for each single resonant term. The characteristic equation can be derived by the block diagram described in Fig. 2 where the PWM inverter transfer function has been considered equal to 1:

D(s) =co +cis+c2s2 +(R+sL)(s2 +OS2).

(3)

R and L are the sum of supply (Rs) and filter resistance

(RF), and supply (Ls) and filter inductance (LF) respectively. The original equation introduced by Naslin [14] is:

8(s) =85s

s . -s

-...... +

S+

50
c52
n-2 n

(4)

where 6, is a real positive coefficient. The characteristic ratios can be defined as:

a,1 = '52 o808C2

82

a2

(52

82

...a

(5)

compensated. The transfer function of a proper multiresonant controller is the following:


G, (s)= K,
+
s

sK2
2

+C)2

++S2sK3

+C)2 wts+

++S2sK4

(1)

Fig. 2 - Block diagram of the APF system

878

and the generalized time constant as:


1r

61 6'0
an-l = 91-1
(7)

and the characteristic pulsatances as:

the third-order Naslin Polynomial to be utilized in this design procedure, which is the following one:

. ....n5 By considering ul12=..... = OCn1 . it is possible to obtain

00

go =-

91

o =-I 1

PN (S)

11
( G(

accO

(8)

From equations D(s) and PN(S) the following relation can be derived:

wo

=-s

(9)

and by using the optimal value of the characteristic ratio: a=2 (10) it is possible to compute the values of the gains by means of the system parameters knowledge. The procedure is reiterated in order to compute the K gains for the fifth and the seventh order harmonics too. B. Nelder-Mead optimization design procedure The Nelder-Mead simplex method, also known as the Amoeba method [15], is a direct method for multidimensional unconstrained function minimization. Since it does not require derivatives but only polynomial calculations, it is very effective for coefficient optimisation when computational burden is relevant. The description of such a method starts from the definition of the search space, based on a geometrical analogy. A search space is the simplex, defined as an N-dimensional non-zero closed space (N is the number of parameters under optimization) having N+1 vertices connected through straight lines. For instance, a triangle is the twodimensional simplex, a tetrahedron is a three-dimensional one and so forth. The simplex is redefined at each step in order to discard those parts of the initial search space where the function is supposed to be larger. The general idea for minimisation is to decrease the simplex space while maintaining the minimum within the simplex. The initial N+1 vertices are chosen in order to define the initial simplex. The algorithm iterations define new vertices and evaluate the corresponding function value discarding those parts of the search space where the function is expected to have larger values. Accordingly, the search space is resized and the new vertices become closer to the optimum value. The most important part of the optimisation steps consists of a mirroring of the worst point for the simplex since a better search place is supposed to be opposite; in this way the simplex can expand itself toward a

promising direction. When the simplex shrinks itself to the minimum value, it contracts itself in the transverse direction starting a new minimization. Therefore, it can move towards all directions and pull itself into the best point, the lowest value of the function, just like an amoeba. The Nelder-Mead minimization method was utilised according to the following scenario. First of all a function, the range of which spans from 0 to infinity, is needed, and if a combination of terms is supposed to be utilized, it is necessary that each term has a range starting from zero. Coefficients can be utilized in order to weight the influence of the different factors. The function to be minimised is a combination of two parameters able to evaluate the quality of current control. Two important figures are utilized: the mean of the RMS value of the current tracking error and a figure related to the current overshoot. Since in proposed tests a step in the amplitude of the current reference has been issued at time instant 0.15 s, the time interval, during which the RMS error has been averaged, spans from 0.20 to 0.26 s, when steady state is achieved. After a step reference variation it is important to monitor the eventual presence of a current overshoot that could damage the system components. In order to monitor this event, the maximum value of the inverter current, in a time interval centred around the step change, is measured and is normalized with respect to the maximum value of the reference current in the same interval (Norm_OVR). This normalized figure spans from unity to infinity, as a consequence a unity has been subtracted from it in order to obtain a zero-infinity range. Weighting coefficients have been chosen according to the following criteria. The limit acceptable for normalised overshoot and RMS error have been established based to the system under investigation: * 2 for normalised overshoot and * 5 A for mean RMS current error. When these values are reached, the function to be minimised has the value 100, while values beyond 100 constitute an even worse situation. So the minimising function is defined as:
f =(Norm OVR -1) 50+mean RAIIS error -10 (11)

where Norm OVR is the normalised current overshoot and mean RMS error is the mean RMS current tracking error, measured as previously described. For this function 0 is the optimum value while higher than 100 values should be avoided. The parameters to be chosen in order to minimise the function are the four gains of the resonant controller. In this case the search space, as well as any moving simplex has no more a 2D or 3D geometrical analogy since it spans in a four-dimensional space. Each simplex vertex is a 4-ple made up of a candidate group of controller gains. The vertices are redrawn according to the algorithm policy as described in [15].

879

described in the following section.

functionf are obtained from simulations of the system as


111. DESCRIPTION OF THE SYSTEM

The parameters used to compute the value of the

IV. SIMULATION RESULTS

As shown in Fig. 3, the inverter, used as an active filter, is connected in parallel with the supply through filter impedance RF-LF and operates as a closed loop controlled current source. Since the object of this work is the control of the converter used as active power filter, the derivation of the current reference has not been considered. The converter works as a VSI controlled with a Space Vector Modulation (SVM) technique. The current control and the SVM are implemented in the alpha-beta reference frame. The modulator derives the modulating signals from the current error filtered by the resonant regulators and the measured supply voltages. The system of Fig. 3 has been simulated using Matlab/Simulink environment, which offers numerous toolboxes together with a user-friendly graphical interface. The model reproduces a shunt APF prototype built in our laboratory and it is based on measurements taken all over the system. The modeled APF comprises a commercial 7.5 kW IGBT inverter with a 340 [tF capacitor on the dc-link. The inverter switching frequency is fs=10200 Hz. The rated filter inductance is 10 mH, with a rated resistance of 0.3 Q. An RC type antialiasing filter is used downstream the voltage and current transducers, with a rated cutoff frequency of 2 kHz. A Space Vector Modulation technique has been used because it shows the best behavior in such applications as grid connected converters.

Using Naslin polynomial the resonant controller parameters have been calculated aiming to control the first, the fifth and the seventh current harmonics. Initially, simulation tests have been carried out issuing a step change of the fifth and seventh harmonic current reference of 5A occurring at t=0.15 s. The same test has been repeated with parameters obtained from the optimization procedure. Results are shown in Tab. I, where the values of the controller parameters and objective function f are reported, and in Figs. 4(a)-(b), where the current reference tracking can be observed. The improvement of functionf is 388% in the case of optimized parameters. Such improvement is mainly due to a reduced overshoot during transient conditions. Other tests have been carried out considering an additional 1 1 harmonic current reference, without modifying the controller structure. The goal of this test is to assess the performance of the system in presence of harmonics the control system has not been designed for. In this case, as can be derived from Tab. II, the improvement of functionf is 4500 and the improvement is due to the better current tracking both in transient and in steady-state conditions

(Figs. 5(a)-(b)).

TABLE I CONTROLLER PARAMETERS AND FUNCTIONfVALUES

Naslin Polynomial Nelder-Mead

114.5 150.5

Ki

2956.8 6734.5

K2

K3 73919 125288.3

K4 145130 29517.6

10.1 6.3

Naslin Polynomial Nelder-Mead

TABLE II CONTROLLER PARAMETERS AND FUNCTIONfVALUES WITH 1 1th CURRENT HARMONIC K1 K2 K3 K4 f

114.5 365.5

2956.8 6864.9

73919 91030.9

145130 279.6

114 62.7

Fig. 3 - Scheme of the APF simulation model

880

10

-5

-z

0.14

0.145

0.15

0.155

0.16

0.165

0.17

0.175

0.18

0.185

0.19

time (s)

(a)

0,14

0,145

0.15

0.155

0.16

0.165

0.17

0.175

0.18

0.185

0.19

time (s)

(b) Fig. 4

-Filter phase

current (red) and its reference (blue) in the case of calculated (a) and optimized (b) controller parameters

20rT
15__
10

rTrr
_
-

-15

-20 0.14

0.145

0.15

0.155

0.16

time (s)

0.165

0.17

0.175

0.18

0.185

0.19

(a)
20
15 10

r- 5-

a,

0 -5 -5

-10 -15 -20 0.14


0.145 0.15 0.155 0.16

time (s)

0.165

0.17

0.175

0.18

0.185

0.19

(b)
Fig. 5 Filter phase current (red) and its reference (blue) with an additional 1 1th harmonic in the case of calculated (a) and optimized (b) controller parameters
-

881

V. CONCLUSIONS

This paper presents an optimization technique of the resonant controller performance based on the minimization of an objective function through the NelderMead method. The objective function is based on the average of the current error rms value in steady-state conditions and the current overshoot in transient conditions. The optimized controller has been compared with the Naslin polynomial based one. The results show that with the optimization technique it is possible to achieve better performances, even if in the reference signal harmonics with a frequency different from the resonant ones are present.
REFERENCES
[1] [2]
W. M. Grady, M. J. Samatyj, A. H. Noyola, "Survey of Active Power Line Conditioning Methodologies", IEEE Trans. on Power Delivery, Vol. 5, July 1990. H. Akagi, "Trends in Active Power Line Conditioner", IEEE Trans. on Power Electronics, Vol. 9, May 1994. J. Nastran, R. Cajhen et al., "Active Power Filter for Nonlinear AC Loads", IEEE Trans. on Power Electronics, Vol. 9, January 1994. P. Verdelho, G. D. Marques, "An Active Power Filter and Unbalanced Current Compensator", IEEE Trans. on Industrial Electronics, Vol. 44, n. 3, June 1997. V. Soares, P. Verdelho, G. D. Marques, "An Instantaneous Active and Reactive Current Component Method for Active Filters", IEEE Trans. on Power Electronics, Vol. 15, n.4, July 2000. L. Malesani, P. Mattavelli and P. Tomasin, "High-performance hysteresis modulation technique for active filters", IEEE Trans.

[7]

[8]

[9]

[10]

[11]

[12] [13]

[3]
[4]

[14] [15]

[5] [6]

on Power Electronics, Vol. 12, n. 5, pp. 876-884, September 1997. D. Butt, M. Sumner and J. C. Clare, "Harmonic compensation in active shunt filters using controllers employing rotating frames of reference", Proc. of 8th European Conference on Power Electronics and Applications EPE'99, Lausanne, September 1999, CD-ROM. N. Mendalek, F. Faiech, K. Al-Haddad, and L. Dessaint, "Non-linear optimal predictive control of a shunt active power filter", Proc. of 37th IAS Annual Meeting IEEE Industry Applications Conference, 2002. vol. 1, pp. 70 -77. Z. Lu and T.C. Green, "Neural network based predictive control strategy of a shunt active power filter for electric drives", Proc. of IEE Power Electronics and Variable Speed Drives Conference, pp287-291, September 1998. D. N. Zmood, D. G. Holmes, and D. H. Bode, "Frequencydomain analysis of three-phase linear current regulator". IEEE Trans. on Industry Applications. Vol 37, No. 2, March/April 2001, pp. 601-610. M. P. Kazmierkowski, R. Krishnan, F. Blaabjerg, "Control in Power Electronics", Academic Press, London, 2002. P. Mattavelli, "A Closed-Loop Selective Harmonic Compensation for Active Filters", IEEE Trans. on Industry Applications, Vol. 37, No. 1, January/February 2001, pp. 81-89. Y. Sato, T. Ishizuka, K. Nezu, and T. Kataoka, "A New Control Strategy for Voltage-Type PWM Rectifiers to Realize Zero Steady-State Control Error in Input Current" IEEE Trans. on Industry Applications, Vol. 34, No. 3, May/June 1998, pp. 480486. P. Naslin, "Essentials of Optimal Control", Boston Technical Publishers, Inc., 1969. J. C. Lagarias, j. A. Reeds, M. H. Wright, P. E. Wright, "Convergence Properties of the Nelder-Mead Simplex Method in Low Dimensions", SIAM Journal of Optimization. 1998, Vol. 9, No. 1,pp. 112-147

882

Você também pode gostar