Rev. Roum. Sci. Techn.– Électrotechn. et Énerg. Vol. 61, 4, pp. 408–413, Bucarest, 2016 ARTIFICIAL NEURAL NETWORK BASED SYNCHRONOUS REFERENCE FRAME THEORY IN THE dq0 AXES FOR REFERENCE HARMONIC CURRENTS GENERATION OF A FOUR LEG SHUNT ACTIVE POWER FILTER ALI CHEBABHI 1 , MOHAMMED-KARIM FELLAH 2 , MOHAMED-FOUAD BENKHORIS 3 , ABEDELHALIM KESSAL 1 Key words: Four-leg shunt active power filter, Synchronous reference frame (SRF) theory, Artificial neural network (ANN), Backstepping, 3D space vector modulation (SVM), Harmonics, Zero sequence current. This paper presents the modelling and simulation of a three-phase four leg shunt active power filter (SAPF) using artificial neural network (ANN) based synchronous reference frame theory (SRF) in the dq0-axes for power quality improvement. It is used for harmonics and zero-sequence current elimination, and reactive power compensation under unbalanced nonlinear loads. Backstepping regulators are used to control the reference currents and the dc bus voltage of four leg inverter based SAPF to robustness, stabilizing the system and to improve the response and to reduce the overshoot and undershoot of traditional proportional-integral (PI). The success of the proposed control is demonstrated through simulation using MATLAB SIMULINK. 1. INTRODUCTION The degradation of power quality distribution system is the major problem in present day [1, 2]. Single phase non- linear loads such as single phase rectifiers, large number of low-power electronic-based appliances, speed drives, electric traction, are the major sources of harmonics and zero-sequence current, these generate significant distortions in the three-phase four-wire electrical networks and causing power quality problems such as production loss, equipment damage or even be destructive to human health [2, 3]. Various solution power devices and filters are used to eliminate these power quality problems for power factor correction, harmonics and zero-sequence current elimination, and reactive power compensation to the level of international standards [4, 5]. One of the solution celebrated as four leg shunt active power filter (SAPF) is widely used for harmonics and zero-sequence current elimination correlated power quality problems in the four-wire electrical networks [2, 3, 6]. The performance of SAPF depends on the strategies selected for reference currents generation, reference currents and dc bus voltage regulation and gate switching pulses generation. Patel et al. [7] have explicated the instantaneous active and reactive current components (id– iq) called synchronous reference fame theory (SRF) for reference currents generation with fuzzy logic (FLC) for dc bus voltage control and hysteresis PWM for switching signals generation based SAPF for decrease of power quality problems. For improvement the SAPF performances, authors [6, 8] have described the pq theory on the performance of SAPF to reduce these problems. In all of them publications, have confirmed that these methods are valid only if the source and line voltages are non-distorted and them publications has been shown that the SAPF with SRF theory has simple structure, easy to realized and good harmonics extracting than pq. Some other works have described the performance of several control algorithms based SAPF under unbalanced conditions [9–13]. In these efforts, authors [10, 11] have used the direct power control of SAPF using high selectivity filter (HSF) [10] and FLC [11]. In others various publications many modifications on SRF theory based SAPF are suggested for improvement the performances. Benchouia et al. [14] have enhanced the SRF theory by a sSelf-tuning filter (STF) called multi variable filter (MVF) and [15] have described the phase locked loop (PLL) based SRF theory for extracted the fundamental current components of an SAPF for improved performance. These modifications are also not fully effective in the application of SAPF due to the limitation of PLL [15]. In [16] we are improved the PLL based pq0 theory by an MVF for extracted the fundamental components of the source voltages, directly in the αβ0-axes of a four leg SAPF for improved performances. This PLL with MVF or PLL MVF gives very good performances. Several researchers described the effect of switching frequency and gate switching pulses on the performance of SAPF [3, 17]. Authors [17] have compared three techniques of gate switching pulses generation and it has been shown that the SAPF controlled by 3D-SVM has operating with fixed switching frequency, simple simulation in Matlab, lower commutation losses and gives good performances than the hysteresis and PWM. Also same others researchers have demonstrated the effect of DC bus voltage and reference currents regulation on the performance of SAPF. Yi et al. [13] demonstrated the good performances gives in the application of PI regulators. However, under loads unbalanced and parameter variations, the PI regulator presents in the DC capacitor voltage the overshoot which affect the SAPF performances [5]. Therefore, the dynamic performance of SAPF with PI is not robust, not suitable due to loads unbalanced and parameter variations [5, 9]. The backstepping regulators (BSRs) possibly will be used for improving the dynamics of PI to have increase the robustness, the stabilization, and gives the good regulation on the DC bus voltage to eliminating the overshoot [10, 18]. In this paper, an enhanced SRF theory by an artificial neural network (ANN) based of PLL MVF is proposed for on four leg SAPF under unbalanced loads with an backstepping regulators to dc bus voltage and reference currents regulation and 3D SVM for gate switching pulses generation. 1 University of Bachir el Ibrahimi, Department of Electromechanical, Bordj Bou Arreridj, Algeria, E-mail: chebabhiali@gmail.com 2 University of Sidi-Bel-Abbes, ICEPS Laboratory, Department of Electrical Engineering, Sidi Bel Abbès, Algeria 3 University of Nantes at Saint Nazaire, IREENA Laboratory, France