THREE-PHASE POSITIVE AND NEGATIVE SEQUENCES ESTIMATOR TO GENERATE CURRENT REFERENCE FOR SELECTIVE ACTIVE FILTERS F. Ronchi , A. Tilli Dept. of Electronics, Computer Science and Systems (DEIS) University of Bologna Viale Risorgimento n.2, 40136 Bologna, ITALY fax: +39 051 20 93073 e-mail: fronchi, atilli @deis.unibo.it Keywords: active power filters, harmonic distor- ing software parameters. Besides, shunt active filters have the goal to generate current equal but opposite tion, positive and negative sequences. to the harmonic currents in the load waveform. This leads to a cheaper design of the filter components Abstract with respect to series active filters [6]. This article deals with the on-line estimation of The performance of active power filters (APF) is three-phase three-wire signal harmonics. It is perbased on the inverter parameters, control algorithm formed in the d-q synchronous reference frame and and on the method of obtaining current reference. In exploits the Luenberger observer concept to isolate conventional load current detection methods [2],[4], positive and negative sequences of each harmonic. the generation of the active filter current reference is This estimator is suitable to be implemented as refbased on the harmonic detection of load currents, userence generator for selective active power filters. ing the well known instantaneous power theory [2], Simulation and experimental results are presented. time-domain correlation techniques [7], FFT, etc. 1 Introduction The objective of this paper is to provide an on-line method to estimate load current harmonics, separating positive and negative sequences. An approach based on Luenberger observer is proposed. The estimated harmonics can be used both to simply monitor the load and to generate current references for the APF controller, to perform selective compensation. The use of nonlinear devices, e.g. power electronics, generates harmonics, subharmonics, and interharmonics in voltage and current mains spectra. It is necessary to measure and decrease this harmonic distortion, thus reducing the power losses and the risk of equipment damage or malfunctioning. Moreover, in order to cope with the delay of the voltage-source inverter current loop, each estimated Current harmonics have been traditionally compensated with passive filters, which have several draw- harmonic can be processed inverting the transfer backs: their operation depends on the network function of the closed-loop control system APF. The impedance, they have to be tuned on fixed frequen- current control is usually [8] performed in a d-q syncies, etc. Active power filters [1],[5] based on digital chronous reference frame. Hence the estimation and controllers can be more expensive respect to the pas- isolation of positive and negative sequences of each sive ones, but they are less network-dependant and harmonic is carried out in this reference frame. This can be tuned on different frequencies simply chang- paper is organized as follows. In section (2) some β concepts about positive, negative sequences and harb monics description are presented; in section (3) the estimation-isolation algorithm is reported. The last two paragraphs show simulation and experimental results. Conclusions summarize the contents ofPSfrag the replacements paper. a α c 2 Preliminaries According to Fortescue’s theorem [3], an unbalanced set of N phasors can be resolved into N systems of phasors called the symmetrical components Figure 1: Fixed reference frames of the original phasors. For a three-phase systems (i.e. N 3), the three sets are: with kc arbitrarily chosen constant, typically kc 2 3 2 3 . p p p 1. Positive sequence. Three phasors Ia Ib Ic , Let consider the positive sequence equal in magnitude, 120o apart, with the same iap t Re Iap e jω t I p cos ω t sequence (a-b-c) as the original phasors. i t Re I e i t Re I e 2. Negative sequence. Three phasors Ian Ibn Icn , equal in magnitude, 120o apart, with the opposite sequence (a-c-b) of the original phasors. p b p c 2π 3 I cos ω t 2π 3 I p cos ω t p It can be calculated that i t 3. Zero sequence. Three identical phasors Ia0 Ib0 Ic0 : equal in magnitude, with no relative phase displacement. p jω t b p jω t c p iα t p β 2π 3 kc I p cos ω t 2 3 kc I p sin ω t 2 Defining r e j 3 the operator that rotates a phasor of 120o , the relationships among the sequence com- that correspond to a phasor rotating with frequency ω in the complex plain. ponents for a-b-c are: Consider the negative sequence p 2 Ia Ia 1 r r 1 ina t Re Ian e jω t In cos ω t 2 n 1 r r Ia Ib n n j ω t 3 ib t Re Ib e In cos ω t 2π 3 1 1 1 Ia0 Ic n n jω t ic t Re Ic e In cos ω t 2π 3 The system under study is three-phase three-wire: the sum of the three a-b-c currents is identically zero It can be calculated that due to Kirchoff’s current law and therefore there is 3 iαn t kc In cos ω t no zero sequence current. 2 3 Positive and negative sequences correspond to two kc In sin ω t inβ t 2 vectors rotating in opposite directions in the complex plane. In order to better understand the direction that correspond to a phasor rotating with frequency of rotation of the vector, the α β fixed reference ω in the complex plain. frame is considered. The matrix that changes coorThe effect of changing from a fixed reference frame dinates from a-b-c to α β is, according to Fig. 1 to the rotating d-q frame is the following. Consider the matrix that describes the change of refer1 1 2 1 2 αβ Tabc kc ence frame from the fixed α β one to the d-q one, 0 3 2 3 2 which rotates synchronously with the voltage mains. 3 dq Tαβ ω t cos sin ω t m m Estimation Let consider the three-phase signal u, its coordinates in the d-q reference frame are: sin ωm t cos ωm t ! ! " Id0 ∑M Harmonics rotating with frequency n ωm (indicated n 1 Idn cos nωmt ϕdn u M I q0 ∑n 1 Iqn cos nωmt ϕqn as PR, Positive Rotating) are shifted into frequency n 1 ωm in the d-q reference frame. At first, let assume that there is only one d-q harmonic in the signal u whose PR and NR compo3 idp t kc I p cos n 1 ωmt nents have to be estimated. Let hωm be its frequency. 2 The proposed estimator based on the Luenberger ob3 kc I p sin n 1 ωmt iqp t server scheme is the following: 2 In the same way, harmonics rotating with frequency n ωm (indicated as NR, Negative Rotating) are where shifted into frequency n 1 ωm in the d-q reference frame. A linear time-invariant (LTI) state-space representation of the PR and NR sequences at frequency ω is given by the following oscillators Ax t ẋ t with A A Ar Ac 0 ω ω 0 0 ω ω 0 0 (PR) if ω 0 (NR) In a digital control system, signals are sampled with period Ts . Hence the following discrete-time model of previous oscillators is considered ii kk 11 A ii kk α α β β with A A Ar Ac cs where ch sh ch sh h h sh ch sh ch if ω if ω cos hωm Ts sin hωm Ts # K $ u t # x̂˙ t &% Ah x̂ t Ah Arh Arh 0 0 Ach 0 hω m Ch x̂ t hω m 0 (1) h0ω hω0 C 10 01 10 01 and x̂ '$ x̂ x̂ x̂ x̂ % is the state of the estimator, where: m Ach m h if ω dr qr dc qc T • x̂dr is the d-component of the estimated positive sequence; • x̂qr is the q-component of the estimated positive sequence; • x̂dc is the d-component of the estimated negative sequence; • x̂qc is the q-component of the estimated negative sequence. 0 (PR) 0 (NR) The system Ah Ch is completely observable, therefore all the eigenvalues of the estimator dynamic matrix can be arbitrarily chosen by means of K. Let u be decomposed in the sum of of u1 , the part to be estimated, and u2 , the part to be rejected, with u1 x ẋ Ch x xdr xqr xdc xqc Ah x $ % T Defining x̃ Mh x̂ x estimation error Ah KCh matrixes Ah Ch Mh of (1) and (2) have to be replaced by the following A C M : (2) A the estimation error dynamic can be described as follows x̃˙ Mh x̃ Ku2 11 1 Ah 0 0 Aj 0 0 and then, Laplace-transforming ('$ sI M *% ) K U s X̃ s h 1 C M 2 6 676 4 0 0 6676 666 . Ch C j A KC .. 0 0 0 322 2 (6) 0 Az 666 C 5 (7) (8) z The matrix Mh must be shaped in order to ensure that where h j z are the numbers of the harmonics considered for estimation. The system A C is still 1. Mh is Hurwitz. This guarantees the conver- completely observable and therefore all the estimator dynamic matrix eigenvalues can be arbitrarily gence of the harmonic h estimation. chosen by means of K. The conditions to take in jω I Mh 1 K 1 2. account are: ω nω m n h All the harmonics different from the h one have 1. M must be Hurwitz. This guarantees the conto be greatly decreased. vergence of harmonics h j z estimation. / +$ % ) 0 ,+.- A way to satisfy these requirements is to impose complex conjugate eigenvalues for Mh , with the peak frequency in hωm and a very low damp. Choosing K 11 k 3 22 k k1 k2 k1 k2 2 666 2. +$ jω I M % ) K ,+.- 1 / ω nω n 0 h j 6766 z This guarantees that all the harmonics different from the h j 666 z ones are greatly decreased. 1 m 1 (3) These conditions can be satisfied forcing complex conjugate eigenvalues for M, with the peak frequencies at the harmonics that have to be estimated, and it can be found that the characteristic polynomial of a very low damp. In particular, if all the harmonics from 1 to hM have to be estimated, then the rejection Mh is must be ensured only for the harmonics higher than 2 2 2 hM ; hence j ω I M 1 K can be shaped s 2k1 s hωm 2 hω m k 2 as a low pass filter that greatly attenuates harmonics Therefore, k1 and k2 can be chosen to im- higher than hM . pose polynomial damp δ and natural frequency In order to implement the proposed estimator on ωn 1 2 δ 2 1 h ωm a DSP board, its time-discrete version must be investigated. All the considerations made about the k1 δ ωn (4) continuous-time estimators can be repeated for the ωn2 ω 2 discrete-time ones. In particular, in the case of one (5) k2 d-q harmonic PR and NR components to be isolated, 2ω the discrete-time solution is Starting from the previous result, let consider the x̂ k 1 general case of many harmonics to be estimated. The Ah x̂ k K u k Ch x̂ k (9) 4 k2 k1 5 ) $ %) 8 # $ 9 :% ( with uβ t Ah Arh Arh 0 0 Ach ch sh sh ch cs cs 10 01 10 01 Ach Ch h h h h 10 sin 10 sin 20 sin 10 sin 10 sin 10 sin 10 sin 20 sin 10 sin 4ωmt 5ω m t 7ω m t π 5 8ω m t π 5 10ωmt 11ωmt 12ωmt 13ωmt π 5 14ωmt π 5 with ωm 2π 50 rad s. Assume that the goal is to estimate harmonics 5 7 11 13. The 5th harmonic has NR sequency only, while the 7th one has the PR one only. Hence both are mapped as a 6th harmonic in the d-q reference frame. The 11th harmonic is completely NR, the 13th one is PR. Hence both are 2 mapped as a 12th harmonic into the d-q reference z2 2 cos ωn 1 δ 2 Ts e δ ωn Ts z e 2δ ωn Ts frame. A continuous-time estimator having dynamic the same gain matrix K reported in (3) can be con- matrix (8) is considered. sidered, with Ar6 0 0 0 The considerations about the placement of the eigenvalues are the same seen for the continuous-time estimators. In particular, if there is only one d-q harmonic to divide into its PR and NR components, and the following characteristic polynomial is desired ; k1 k2 ) c cos ω 1 e) 2s2k c h n δ ωn Ts 1 h < ) ) δ 2 Ts e 1 δ ωn Ts h 11 A C 0 0 0 22 Ac6 0 0 0 Ar12 0 0 0 Ac12 1 0 1 0 1 0 1 0 0 1 0 1 0 1 0 1 If the harmonics to estimate are several, then the matrixes to consider have the same structure of (6), (7) In order to ensure high rejection to the other harmonand (8). ics, the matrix K is chosen to obtain for M the following characteristic polynomial: s 4 Simulation results 2 In order to verify the performances of the estimawith tor, the following simulation has been executed. Let u α uβ T , consider the three-phase signal u uα t m m m m m '$ % 10 cos 4ω t 10 cos 5ω t 20 cos 7ω t π 5 10 cos 8ω t π 5 10 cos 10ω t 10 cos 11ω t 10 cos 12ω t 20 cos 13ω t π 5 10 cos 14ω t π 5 m m m m 2δ ωn6 s 2 ωn6 δ ωn6 ωn12 s 2 2 2δ ωn12 s 0 6 60ω01 1 2δ 112 ω2δ 2 ωn12 2 m 2 m 2 All the estimator state variables have initial values equal to zero. Each of the resulting estimated harmonics is converted from d-q to α β coordinates and compared with the original ones. The α component of the resulting estimation errors is presented in Fig. 2. The β component is similar and therefore is omitted. PR, 6th harmonic 20 magnitude 1.5 0 1 -20 0 0.5 10 1 1.5 2 2.5 NR, 6th harmonic 3 3.5 0.5 0 200 0 220 240 260 280 300 320 340 360 380 400 220 240 260 280 300 320 340 360 380 400 -10 0 0.5 20 frag replacements 1 1.5 2 2.5 3 PR, 12th harmonic 150 3.5 phase 100 PSfrag replacements 50 0 0 -50 -20 0 0.5 1 1.5 2 2.5 3 -100 3.5 NR, 12th harmonic 10 -150 200 0 -10 0 0.5 1 1.5 2 2.5 3 3.5 Figure 4: FFT of current signal to be processed, phase a Figure 2: Estimation errors, α axis vma 400 200 vma 400 0 -200 -400 3.966 3.968 200 3.97 3.972 3.974 3.976 3.978 0 1 400 3.97 3.975 3.98 3.985 3.99 3.995 3.98 3.982 3.984 3.986 3.98 3.982 3.984 3.986 0 10 5 0 -5 -10 PSfrag replacements -0.5 -1 3.966 3.968 3.97 3.972 3.974 3.976 3.978 NR 2 1 3.975 3.98 frag replacements 3.985 3.99 3.995 4 0 -1 -2 ilb 3.966 3.968 10 5 0 -5 -10 3.97 3.982 3.984 3.986 0.5 4 ila 3.97 3.98 PR 200 3.975 3.98 3.985 3.99 3.995 4 3.97 3.972 3.974 3.976 3.978 Figure 5: harmonics estimated in simulation, d axis Figure 3: Phase voltage vma and current signal to be K is chosen to obtain for M6 a characteristical polynomial with damp δ 0 001 and natural frequency processed ωn 6ωm 1 2δ 2 . The d-components of the estimator outputs are shown in Fig. 5. They match the In order to make comparisons among simulation amplitudes in Fig. 4. In order to verify if also the and experimental results, the three-phase signal of phases are well estimated, they have been converted Fig. 3 is considered. In a fixed reference frame, into a-b-c coordinates and subtracted from the origithis signal is characterized by the following harmon- nal signals. The results have 5th and 7th near to zero, with the fundamental at therefore estimations are good. ics: 1 5 7 11 13 17 19 fm 50 Hz. The goal is to estimate the 5th and the 7th harmonics, that are shown in Fig. 4. They 5 Experimental results both are shifted into 6th harmonic on the d-q reference frame. The three-phase signal is sampled The current signal to be processed is the one dewith frequency fs 7 kHz and then converted from scribed in the previous section and is shown in Fig. 3. the fixed a-b-c reference frame to the rotating d-q The three-phase signals are acquired and sampled one. A time-discrete estimator having the structure with frequency fs 7kHz. The DSP on-line exeof (9) is considered, with h 6. The gain matrix cutes all the calculus to change reference frame from 6 666 vma 400 200 0 -200 -400 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 PR 1 0.5 0 frag replacements -0.5 -1 0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02 0.012 0.014 0.016 0.018 0.02 NR 2 1 [4] L. Malesani, L. Rossetto, P. Tenti. “Active filter power filter with hybrid energy storage’, IEEE Trans.Power Electron., volume 6, pp. 392–397, (July 1997). [5] P. Mattavelli. “A closed-loop selective harmonic compensation for active filters”, IEEE Trans.Ind.Applicat., volume 37, pp. 81–89, (January/February 2001). 0 -1 -2 0 0.002 0.004 0.006 0.008 0.01 Figure 6: harmonics estimated in laboratory a-b-c to d-q and implements a time-discrete estimator having the structure of (9), with h 6. The dcomponents of estimated harmonics are shown in Fig. 6. They match the ones resulting from simulation. 6 Conclusions A positive and negative sequence on-line estimator has been presented. It allows to separate the positive from the negative sequence, that hence can be separately processed by the active power filter controller. Simulation and experimental results have been presented. References [1] H. Akagi. “New trends in active filters for power conditioning”, IEEE Trans.Ind.Applicat., volume 32, pp. 1312– 1332, (Nov./Dec.1996). [2] H. Akagi, A. Nabae. “Control strategy of active power filters using multiple voltage source PWM converters”, IEEE Trans.Ind.Applicat., volume IA-22, pp. 460–465, (May/June 1986). [3] C.L.Fortescue newblock “Method of symmetrical coordinates applied to the solution of polyphase networks”, Trans.AIEE volume 37, pp. 1027–1140, (1918). [6] F. Ronchi, A.Tilli “Design methodology for shunt active filters”, Proc. 10th EPE-PEMC 2002, (Cavtat & Dubrovnik, Croatia, September 2002), to be published, . [7] G. L. Van Harmelen, J.H.R. Enslin. “Realtime dynamic control of dynamic power filters in supplies with high contamination”, IEEE Trans. Power Electron., volume 8, pp. 301– 308, (July 1993). [8] P. Verdelho, G. D. Marques. “An active power filter and unbalanced current compensator”, IEEE Trans.Ind.Applicat., volume 44, pp. 321–328, (June 1997).