Three-phase Three-level (NPC) Shunt Active Power Filter Performances based on PWM and ANN s Controllers for Harmonic Current Compensation

Transcription

1 International Journal on Electrical Engineering and Informatics - Volume 6, Number 2, September 214 based on PWM and ANN s Controllers for Harmonic Current Compensation Cennai Salim 1 and Bencouia Moamed Toufi 2 1 Electrical Engineering Department, Birine Nuclear Researc Center, Algeria, 2 Laboratory L.G.E.B., Bisra University, Algeria cenaisalimov@yaoo.fr Abstract: Tis paper presents a sunt active power filter performances based on treelevel (NPC) inverter using conventional pulse widt modulation (PWM) and Ann s controllers to mitigate current armonics generated by non-linear loads. Sunt APF is te best solution to eliminate armonics drawn from nonlinear load especially for low power system, te conventional sceme based on te two level inverter controlled by ysteresis controller presents several drawbacs suc us ig inverse voltage applied to power components, degraded output voltage inverter waveforms, etc. Today tree-level (NPC) inverter and intelligent controllers are widely used in different industrial applications. To tae advantages of tese inverter topology and artificial intelligent controller performance a novel control sceme based on Ann s current controller is proposed in tis wor. Te control strategy adopted is te syncronous current detection metod, it is easy to implement and permits a good extraction of reference currents compensation. Te numerical simulation is performed using MATLAB-Simulin and SimPowerSystem Toolbox. Te simulation results obtained in steady and transit states sow te effectiveness and te superiority of Ann systems compared to PWM controller. Keywords: Sunt active power filter, intelligent controller, Ann s, Tree-level (NPC) inverter, Harmonic currents compensation, power quality improvement. 1. Introduction Wit te proliferation of nonlinear power electronics loads, te problem of armonic is severity, wic influences te quality of power grid. Harmonic currents produced by nonlinear loads, suc as rectifiers, power electronics converters, controllers for adustable-speed motor drives, electronic power supplies, DC motor drives, battery cargers, electronic ballasts are responsible for te rise in PQ related problems [1]. Passive power filter is a traditional armonic restraint metod, tis solution ave been used to eliminate current armonics and to increase te power factor, wic are simple and low cost. However, te use of passive filter as many disadvantages, suc as large size, series or parallel resonance wit te system impedance. [2]. One of te most popular active filters is te Sunt Active Power Filter; Sunt APF ave been researced and developed, tat tey ave gradually been recognized as a feasible solution to te problems created by non-linear loads. Active filter is used, tese last years, to improve power quality on te load side from te grid by inecting compensating currents. Te performance of any active filter mainly depends on te reference current generation strategy, ysteresis or pulse widt modulation signals, topology of te power converter, etc Instantaneous power teory [3] and syncronous reference frame detection metod [4] are widely used in different researc wor, tese tecniques provide good results under different voltage source conditions but present some drawbacs suc us a muc calculations number, necessitate complex matematical transformation and difficult implementation in practice [5]. Syncronous current detection metod is anoter interesting metod; it is concise and requires less computational calculation compared te two oter metods. In tis wor, tis tecnique is adopted to use wit tree-level (NPC) inverter after several modifications. Received: November 9 t, 213. Accepted: Marc 16 t,

2 Cennai salim, et al. In oter side, tree-level inverter is one of te most multilevel converters used successfully in medium and ig power applications [6], [7]. Teir advantages are well nown as very low armonic distortion, smaller common-mode (C-M) portions, and lower switcing frequency and power loss. However, te control of a tree-level inverter is more complex tan tat of a two-level inverter because of large numbers of inverter switcing states. Te controller is te main part of any active power filter operation and as been a subect of many researces in recent years. Among te various current control tecniques, ysteresis current control is te most extensively used tecnique. It is easy to realize wit ig accuracy and fast response. In te ysteresis control tecnique te error function is centred in a preset ysteresis band. Wen te error exceeds te upper or lower ysteresis limit te ysteretic controller maes an appropriate switcing decision to control te error witin te preset band. However, variable switcing frequency and ig ripple content are te main disadvantages of ysteresis current control [8]. To improve te control performances tere s a great tendency to use intelligent control tecniques, particularly fuzzy logic or artificial neural networ controllers [9], [1]. Teir principal advantages are robustness, ability to control nonlinear systems, no need accurate matematical model, etc. Tis paper presents te performance of te tree-level (NPC) sunt active filter based on PWM and Ann s controllers. Te model and numerical simulation in transient and steady states conditions are developed and performed using Matlab-Simulin and SimPowerSystem Toolbox. 2. Sunt active power filter Te basic bloc diagram including non linear load compensation principle of a sunt active power filters based on tree-level (NPC) inverter is sown in Figure 1. It is controlled to draw/supply a compensated current from/to te utility, suc tat it cancels armonic currents of te non-linear load and maes te source current in pase wit te different waveforms. Te current drawn from te power system at te coupling point of te sunt APF will result sinusoidal [11], [12]. Figure 1. Tree-level (NPC) Sunt active filter 533

3 Multilevel inverters are being investigated and recently used for active filter topologies. Tree-level inverters are becoming very popular today for most inverter applications, suc as macine drives and power factor compensators. Tese advantages are reduction of te armonic content generated by te active filter and decrease te voltage or current ratings of te semiconductors. Figure 2, sown te circuit topology of a diode-clamped tree-level inverter based on te six main switces (T11, T21, T31, T14, T24, T34) of te traditional two-level inverter, adding two auxiliary switces (T12, T13, T22, T23, T32, T33) and two neutral clamped diodes on eac bridge arm respectively, te diodes are used to mae te connection wit te point of reference to obtain Midpoint voltages. Suc structure allows te switces to endure larger dc voltage input on te premise of not raising te level of teir witstand voltage. Moreover, tae pase-a as example, tree inds of voltage level Udc/2, and Udc/2 can be output corresponding to tree inds of switcing states A,, B (Table. 1) [13], [14]. Figure 2. Circuit topology of tree-level (NPC) inverter 534

4 Cennai salim, et al. Table 1. Switcing states of tree-level (NPC) inverter Switcing states Voltage output T11 T12 T13 T14 A Udc/2 ON ON OFF OFF OFF ON ON OFF B Udc/2 OFF OFF ON ON 3. Control strategies Te strategy control used is te syncronous reference current detection metod is sown in Figure 2. It s concise and requires less computational efforts tan many oters metod control [12], [13]. Te compensating currents of active filter are calculated by sensing te load currents, te current delivered by DC voltage regulator I*smd, pea voltage of AC source (Vsm) and zero crossing point of source voltage. vsa( t) = Vsm.sin( ωt) vsb ( t) = Vsm.sin( ωt 2 π ) (1) 3 vsc ( t) = Vsm.sin( ωt 4 π ) 3 In order to compensate te current armonics, te average active power of AC source must be equal wit PLav, considering te unity power factor of AC source side currents te average active power of AC source can be: 3 * P s = Vsm. Ismp = PLav 2 (2) From tis equation, te first component of AC side current can be calculated: I * smp 2 = 3 P V Lav sm (3) Te second component of AC source current I*smd is obtained from DC capacitor voltage regulator. Te desired pea current of AC source can be calculated as bellow: * sm * smp * smd I = I + I (4) Te AC source currents must be sinusoidal and in pase wit source voltages, teses currents can be calculated wit multiplying pea source current to a unity sinusoidal signal, tese unity signals can be obtained from equation (5): vsa i ua( t) = Vsm vsb i ub( t) = Vsm (5) vsc i uc( t) = V sm Te desired source side currents can be obtained from equation (6): * * i sa ( t) = I sm. iua * * i ( t) = I. i (6) sb sm ub * * i sc ( t) = I sm. iuc 535

5 Finally, te reference currents can be obtained from (7): * * * i ( t) = i ( t) i ( t) ca sa La * * * icb ( t) isb ( t) ilb ( t) * * * icc ( t) = isc ( t) ilc ( t) = (7) To compensate te inverter losses and regulate te DC lin voltage Udc, a proportional integral voltage controller is used. Te control loop consists of te comparison of te measured voltage (Udc=Udc1+Udc2) wit te reference voltage Udc-ref and ΔUdc=Udc-ref-Udc. Te loop generates corresponding current I*smd as given by: I = K p ΔU dc + K i ΔU dc dt * smd. (8) Figure 3. Syncronous current detection metod control 536

6 Cennai salim, et al. 4. Sunt APF Control A. PWM current controller Te main component of any active filter is te controller; te conventional sceme is based on ysteresis controller, it is very complicated and requires many calculations for its implementation [15]. To replace te conventional ysteresis control sceme a novel logic controller to use wit treelevel (NPC) inverter given by Figure (4) is proposed. Te difference between te inected current and te reference current determines te reference signal (e). Tis signal is compared wit two triangular-carrying identical waves sifted from one to te oter by a alf-period of copping and generating te switcing pulses. Te control of inverter is summarized in te two following stages: Figure 4. Tree-level (NPC) inverter PWM logic control 537

7 Determination of te intermediate signals Vi1 and Vi2: If error Ec carrying 1 Ten Vi1=1, If error Ec < carrying 1 Ten Vi1=, If error Ec carrying 2 Ten Vi2=, If error Ec < carrying 2 Ten Vi2=-1. Determination of control signals of te switces Ti (i=1, 2, 3; =1, 2, 3, 4): If (Vi1+Vi2)= 1 Ten Ti1=1, Ti2=1, Ti3=, Ti4=, If (Vi1+Vi2)= Ten Ti1=, Ti2=1, Ti3=1, Ti4=, If (Vi1+Vi2)=-1 Ten Ti1=, Ti2=, Ti3=1, Ti4=1. Te bloc diagram of te tree-level (NPC) sunt active power filter based on conventional current controller is sown in Figure (5). Figure 5. Tree-level (NPC) sunt APF bloc diagram based on conventional controller 538

8 Cennai salim, et al. B. Ann s current controller Artificial Neural Networs ave provided an alternative modeling approac for power system applications. It is essentially a cluster of suitably interconnected non-linear elements of very simple form tat possess te ability of learning and adaptation. Tese networs are caracterized by teir topology, te way in wic tey communicate wit teir environment, te manner in wic tey are trained and teir ability to process information. Te MLPNNs is one of te most popular topologies in use today [16], te structure model of tis topology is sown in Figure 6. It consists of input layer, output layer and idden layer, eac layer as many neurons; coosing appropriate number of neurons of te idden layer and activation function and tresold value of neurons, and determining te weigt of connections between neurons troug learning and training, it can realize approximation of any non-linear obect wit small error [17]. Figure 6. Tree-layer neural networ topology structure Data flows into te networ troug te input layer, passes troug te idden layers and finally flows out of te networ troug te output layer. Te networ tus as a simple interpretation as a form of input-output model, wit networ weigts as free parameters. Te training cycle as two distinct pats, te first one is Forward propagation (it is te passing of inputs troug te neural networ structure to its output), te second one is te error bacpropagation (it is te passing of te output error to te input in order to estimate te individual contribution of eac weigt in te networ to te final output error) [18]. Te weigts are ten modified so as to reduce te output error. To train te neural networ current controller, te Quasi-Newton Levenberg-Marquardt Training algoritm is used, it is efficient, easy to implement and is not time consuming. Computations of te algoritm proceed as follows: 1. Initialize te interconnection weigts and te biases of te nodes randomly, 2. Calculate te idden layer outputs as: x ni = f + ( x i w i, ) b wni+ 1, (9) i= 1 Were x is te output of te idden node, connecting input node i wit idden node, w ni 1, xi is te i't input, w i, is te weigt b is te input bias to idden node (normally =1), + is te weigt connecting te bias to te idden node, ni is te number of input nodes, and f is sigmoid function defined as: 1 f ( x) = (1) 1 x + e b 539

9 Calculate te output layer outputs as: n x = f ( x w, ) + b wn + 1, (11) = 1 Were x is te output of te output node, w, is te weigt connecting te idden w + 1, node wit output node, b is te input bias to te output node (normally b =1), n is te weigt connecting te bias to te output node and n is te number of idden nodes. Calculate δ of eac of te output nodes as: T δ = x 1 x )( x x ) (12) ( Wereδ, is te error (target-output) at te output of te neuron multiplied by te derivative of f (x), Calculate T x is te target output (desired output) of te output node. δ eac of te idden nodes as follows: no ( 1 x ) w, = 1 δ = x δ (13) Were δ te derivative of is f (x) multiplied by te summation of te weigts multiplied by te output delta, Adapt te weigts of te output layer as: w ( t + 1) = w ( t) + ηδ x + αδw ( ) (14),,, t Were <η < 1 is te learning constant, < α < 1 is te momentum constant and, Δw ( t) = w ( t) w ( t 1) (15),,, Adapt te weigts of te idden layer as: w Were, i, ( i, i i, t t + 1) = w ( t) + ηδ x + αδw ( ) (16) Δw, ( t) = w, ( t) w, ( t 1) i i i (17) Repeat steps 1 to 7 until te error e is less tan a prescribed small valueε. e = no = 1 T ( x x ) 2 (18) 54

10 Cennai salim, et al. Te flow cart for training ANN controller is given by Figure (7). Figure 7. Flow cart for training Ann s controller For te tree-level (NPC) inverter, te proposed Ann s current controller is sown in Figure 8. Te input pattern is te error values (Eca, Ecb and Ecc) between te measured filter currents (ifa, ifb and ifc) and te compensating reference currents (ifa*, ifb* and ifc*) [9], [1] and [14]. Wereas te outputs values are te switcing states T11, T12, T21, T22, T31 and T32. Te idden layer contains 12 neurons wit a sigmoid activation function and te output layer contains six neurons wit a linear activation function. Te networ was trained wit 1 training examples using Levenberg-Marquardt bac propagation algoritm. 541

11 Figure 8. Ann s current controller 542

12 Cennai salim, et al. Te bloc diagram of te tree-level (NPC) sunt APF based on Ann s current controller is sown in Figure (9). Figure 9. Tree-level (NPC) sunt APF bloc diagram based on Ann s current controller 543

13 Figure (1) sows te Matlab-Simulin simulation bloc diagram of te proposed Sunt APF based on artificial neural networ current controller. Figure 1. Tree-level (NPC) sunt APF sim ulation model based on Ann s current controller 544

14 Cennai salim, et al. 5. Simulation results and discussion Te simulation results are provided to verify te performance and effectiveness te treelevel (NPC) sunt active power using PWM and Ann s current controllers. Te active filter is composed mainly of te tree-pase source, tree-level (NPC) inverter, nonlinear load (Rectifier & R,L). Te parameters of te simulation are: Lf = 3 mh, C1 = C2=3 μf, Vs = 22 V/5 Hz and Udc-ref = 8 V. A. Simulation results using conventional controller Figure (11) sows te simulated waveforms of te source current before compensation. Te corresponding armonic spectrum is sown in Figure 12. Te source current and inected current before and after APF application are sown in Figures 13 and 14, respectively. Te DC voltage is presented in Figure 15. Te waveforms of source voltage and source current after compensation are simultaneously sown in Figure 16. Te armonic spectrum of te source current after compensation is sown in Figure 17. Figure 11. Source current witout Sunt APF Figure 12. Source current spectrum witout Sunt APF: THDi (%) = 28.16% 545

15 3 isa(a) 2 Source current isa (A) Figure 13. Source current isa (A) before and after compensation using conventional controller 2 15 ia(a) Inected current Ia (A) Figure 14. Inected current ia (A) 82 8 Udc(V) Udc (V) Figure 15. DC side capacitor voltages Udc (V) 546

16 Cennai salim, et al. 4 3 vsa(v) isa(a) 2 vsa(v), isa(a) Figure 16. Current source isa (A) and voltage source vsa (V) before and after compensation 1.9 Magnitude compared to fundamental Harmonic order Figure 17. Source current spectrum wit Sunt APF: THDi (%) = 4.32% B. Simulation results using conventional controller wit sudden step cange in load To evaluate dynamic responses and test robustness of te proposed sunt active filter a step cange in load is introduced between t1=.2 sec and t2=.4 sec. Figures 18 and 19 sow te respective waveforms of te source current and inected current before and after compensation. Te dc side capacitor voltage is sown in Figure 2. Te current and te voltage source waveforms before and after compensation are simultaneously presented in Figure isa(a) Source current isa (A) Figure 18. Source current isa (A) wit step cange in load (between t1=.2 s and t2=.4 s) 547

17 15 ia(a) 1 Inected current I (A) Figure 19. Inected current ia (A) wit step cange in load (between t1=.2 s and t2=.4 s) 82 8 Udc(V) Udc (V) Figure 2. DC side capacitor voltage wit step cange in load (between t1=.2 s and t2=.4 s) 5 vsa(v) isa(a) vsa(v), isa(a) Figure 21. Current and voltage source before and after compensation wit step cange in load (between t1=.2 s and t2=.4 s) 548

18 Cennai salim, et al. C. Simulation results using Ann s currant controller Te source current and inected current before and after APF application using Ann s controller are respectively sown in Figures 22 and 23. Te output DC capacitor voltage is presented in Figure 24. Te waveforms of source voltage wit source current after compensation are simultaneously sown in Figure 25. Te armonic spectrum of te source current after compensation is sown in Figure isa(a) Source current isa (A) Figure 22. Source current isa (A) before and after compensation using Ann s controller 15 ia(a) 1 Inected current Ia (A) Figure 23. Inected current ia (A) Udc(V) 76 Udc (V) Figure 24. DC side capacitor voltages Udc (V) 549

19 5 vsa(v) isa(a) vsa(a),isa(a) Figure 25. Current source isa (A) and voltage source vsa (V) before and after compensation 1.9 Magnitude compared to fundamental Harmonic order Figure 26. Source current spectrum wit APF using Ann s current controller: THDi (%) = 3.96% Troug visualization (Figs. 13, 18, and 22), we are able to conclude tat te operation of te proposed tree-level sunt APF based on conventional and Ann s current controllers is successful. Before te application of sunt APF, te source current is equal to non-linear load current; igly distorted and ric in armonic. After compensation, te THDi is considerably reduced from 28.16% to 3.96% for Ann s controller and to 4.32% for conventional controller. Te dc voltage is maintained at a constant value wic is equal to te reference value Udc-ref = 8 V by using PI voltage controller. Figures 2 and 21 illustrate te dynamic response of te proposed sunt APF. It is observed tat te dc voltage pass troug a transitional period of.6 s before stabilization and reaces its reference Udc-ref=8 V wit moderate pea voltage approximately equal to 7 V wen a step cange in load current is introduced between t1=.2 s and t2=.4 s. Figures 16, 21, and 25 demonstrate tat after active filter application, te current source is sinusoidal and in pase wit te voltage source. Te performance of te tree-level sunt active filter based on Ann s controller in terms of armonics elimination and reactive power compensation is very satisfactory and better tan conventional current controller. Te THDi values obtained wit te two controllers respect te IEEE standard Norms of THDi 5%. 55

20 Cennai salim, et al. 6. Conclusion In te present paper, a tree-pase tree-level (NPC) sunt active power filter wit neutralpoint diode clamped inverter based on PWM and Ann s current controllers are presented. Use of te filter is aimed at acieving te elimination of armonics introduced by nonlinear loads. Several simulations wit nonlinear loads composed by AC/DC converter wit R,L loads under different conditions are performed. Te simulation results sow te robustness and effectiveness of te proposed Ann s controller in terms of eliminating armonics compared to conventional currant controller. Te THDi is significantly reduced from 28.16% to 4.32% using conventional controller and to 3.96% using Ann s controller in conformity wit te IEEE standard norms. Te control strategy based on syncronous current detection metod permits a good extraction of reference currents compensation. Te PI Controller ensures tat te dc voltage across capacitor is maintained constant and equal to reference vale (Udc-ref=8V). Te steady state of te source current is obtained after.3 sec in te two cases. Te current source after compensation for te two proposed control sceme is sinusoidal and in pase wit te line voltage source; te power factor is nearly equal to unity. Hence, te proposed Ann s current controller is excellent candidates to control sunt active filters based on multilevel inverter topology toward eliminating te armonic currents and improving te power factor compared to conventional current controller. 7. References [1] N. Gupta, S. P. Sing, and S. P. Dubey, Neural networ based sunt active filter for armonic and reactive power compensation under non-ideal mains voltage, IEEE, pp , 21. [2] J. Wang, Simulation of tree-pase tree-wire sunt active power filter, International Journal of Sciences and tecniques of automatic Control & Computer engineering, Vol. 3, N 1, pp , 29. [3] R-H. Herrera, P. Salemeron, H. Kim, Instantaneous Reactive Power Teory Applied to Active Power Filter Compensation: Different Approaces, Assessment, and Experimental Results, IEEE, Trans. on Industrial Electronics, pp , 28. [4] S. Battacrya, D. Divan, Syncronous frame based controller implementation for a ybrid series active filter systems, IEEE, pp , [5] A. Candra, B. Sing, K. Al-Haddad, An improved control algoritm of sunt active filter for voltage regulation, armonic elimination, power factor correction, and balancing nonlinear loads, IEEE Tran. on Power Electr, pp , 2. [6] O. Vodyao, T. Kim, S. wa, Comparison of te space vector current controls for sunt active power filters, IEEE, pp , 28. [7] O. Vodyao, D. Hacstein, A. Steimel, T. Kim, Novel direct current-space vector control for sunt active power filters based on tree-level inverters, IEEE, pp , 28. [8] B. Lin, C-H Huang, T-Y. Yang and Y-C. Lee, Analysis and Implementation of Sunt Active Power Filter wit Tree-Level PWM Sceme, IEEE, pp , 23. [9] A. Zouidi, F. Fnaiec, K. Al-Haddad, Neural Networ tree-pase tree-wire sunt active filter, IEEE, pp. 5-1, 26. [1] A. Zouidi, F. Fnaiec, K. Al-Haddad, Neural networ controlled tree-pase tree-wire sunt active power filter, IEEE, ISIE, 26. [11] E.E. El-Koly, A. El-Hefnawy, H.M Marous, Tree-pase active power based on current controlled voltage source inverter, Elsevier, Electric power and Energy Systems 28, pp , 26. [12] S. Cennai, M-T. Bencouia, Intelligent controllers for sunt active filter to compensate current armonics based on SRF and SCR control strategies, International ournal on Electrical Engineering and Informatics, IJEEI, Vol.3, No.3, pp , 211. [13] Y. Wan, J. Jiang, Te study of FPGA-based tree-level SVM NPC inverter, IEEE, pp ,

21 [14] S. Cennai and M-T. Bencouia, Sunt active power filter performances based on treelevel (NPC) inverter to compensate current armonics using intelligent controller, 8 t Electrical Engineering Conference CGE 8, 16 and 17 April 213, Algeria (213). [15] S. Cennai and M-T. Bencouia, Simplified Control Sceme of Unified Power Quality Conditioner based on Tree-pase Tree-level (NPC) inverter to Mitigate Current Source Harmonics and Compensate All Voltage Disturbances, Journal of Electrical Engineering & Tecnology, JEET (KIEE), Vol.8, No.3, pp , May 213. [16] G. Liu, S. Su, P. Peng, Intelligent Control and Application of All-function Active Power Filter, IEEE, International Conference on Intelligent Computation Tecnology and Automation, pp , 28. [17] V-S. Kumar, D. Kavita, K. Kalaiselvi and P-S. Kannan, Harmonic Mitigation and Power Factor Improvement using Fuzzy Logic and Neural Networ Controlled Active Power Filter, Journal of Electrical Engineering & Tecnology, JEET (KIEE), Vol.3, No.3, pp , 28. [18] Y-H. Jiang, Y-W. Cen, Neural networ control tecniques of ybrid active power filter, International Conference on artificial Intelligence and computational intelligence, pp. 26-3, 29. Cennai Salim Was born in Bisra, Algeria, on February 3, He obtained is engineering degree in Electrotecnics from Bisra University in He was recruited in 1993 as senior engineer in power electronics in Birine Nuclear Researc Center, Algeria. Since 2, e as been woring as researcer in te Electrical Engineering Department. He obtained is M.Sc degree in electrical engineering in 29 from Media University, Algeria and is P.D. degree in Electrical Engineering from Bisra University in 213, Algeria. His researc interests are electrical drives, power electronics, energy quality, power systems, and intelligent control. Bencouia Moamed Toufi Was born in Bisra, Algeria. He received is Engineer degree in Electrotecnics and Magister degree in electrical engineering from Bisra University in 1991 and 1998 respectively. He received is P.D. in electrical engineering from Bisra University in 26. Since 21 e as eld a teacing and researc position in te Department of Electrical Engineering of Bisra University, Algeria. His researc interests are in electrical drives, power electronics and power systems. 552