IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 54, NO. 1, FEBRUARY 2007 61 A Unified Artificial Neural Network Architecture for Active Power Filters Djaffar Ould Abdeslam, Patrice Wira, Jean Mercklé, Damien Flieller, and Yves-André Chapuis Abstract—In this paper, an efficient and reliable neural active power filter (APF) to estimate and compensate for harmonic distortions from an AC line is proposed. The proposed filter is completely based on Adaline neural networks which are organized in different independent blocks. We introduce a neural method based on Adalines for the online extraction of the voltage com- ponents to recover a balanced and equilibrated voltage system, and three different methods for harmonic filtering. These three methods efficiently separate the fundamental harmonic from the distortion harmonics of the measured currents. According to either the Instantaneous Power Theory or to the Fourier series analysis of the currents, each of these methods are based on a specific decomposition. The original decomposition of the currents or of the powers then allows defining the architecture and the inputs of Adaline neural networks. Different learning schemes are then used to control the inverter to inject elaborated reference currents in the power system. Results obtained by simulation and their real-time validation in experiments are presented to compare the compensation methods. By their learning capabilities, artifi- cial neural networks are able to take into account time-varying parameters, and thus appreciably improve the performance of traditional compensating methods. The effectiveness of the algorithms is demonstrated in their application to harmonics compensation in power systems. Index Terms—Active power filter (APF), adaptive control, arti- ficial neural networks (ANNs), harmonics, selective compensation, three-phase electric system. I. INTRODUCTION U NEXPLAINED computer network failures, premature motor burnouts, humming in telecommunication lines, and transformer overheating are only a few of the damages that quality problems may bring into home and industrial installa- tions. What may seem like minor quality problems may lead to crucial situations in the future with the increasing proliferation of nonlinear loads. Indeed, nonlinear load currents and voltages are nonsinu- soidal, and it is necessary to compensate for the generated harmonics and to correct the load power factor. For several Manuscript received June 28, 2005; revised February 9, 2006. Abstract pub- lished on the Internet November 30, 2006. D. O. Abdeslam, P. Wira, and J. Mercklé are with the Faculté des Sci- ences et Techniques Laboratoire MIPS-TROP 4, Université de Haute Alsace, 68093 Mulhouse Cedex, France (e-mail: djaffar.ould-abdeslam@uha.fr; patrice.wira@uha.fr; jean.merckle@uha.fr). D. Flieller is with the Department of Electrical Engineering, Institut Na- tional des Sciences Appliquées (INSA), 67084 Strasbourg, France (e-mail: flieller@mail.insa-strasbourg.fr). Y.-A. Chapuis is with the Institute of Industrial Science, University of Tokyo, Tokyo 106-8558, Japan (e-mail:chapuis@fujita3.iis.u-tokyo.ac.jp). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIE.2006.888758 years, active power filters (APFs) have been recognized as advanced techniques for harmonic compensation [1]. Their objective is to recover balanced and sinusoidal source currents by the injection of compensation currents. APFs are very able to suppress the current harmonics and to correct power factor, especially in fast-fluctuating loads, in comparison to other compensation devices. In addition to their performances, APFs can favorably be inserted in existing power systems and are thus widely used in practical applications. A lot of recent research work investigates and tries to improve APFs by developing new topologies with power electronics technologies or new control laws. Mathematical modeling is also involved to formulate and characterize active and nonactive current and power for nonsinusoidal and nonperiodic waveforms in electrical sys- tems. It also concerns numerical analysis, harmonic detection algorithm, control theory, and artificial intelligence techniques like artificial neural networks (ANNs). For a few years, ANN techniques have been applied with suc- cess in control of APF [2] and are very promising in the field. In- deed, the learning capacities of the ANNs allow an online adap- tation to every changing parameter of the electrical network, e.g., nonlinear and time-varying loads. Most of these control constraints are quite still very challenging with classic control methods. In this paper, the authors analyze and synthesize most of the ANN strategies already known in order to propose a novel, unified and complete ANNs approach to control APFs. Indeed, the main motivation of this paper is to propose a uni- fied neural approach of the entire adaptive harmonic compensa- tion system. The proposed approach is unified in the sense that it is only based on a single type of ANN: the Adaline neural network [3]. This approach is motivated by a need of simplicity and flexibility in ANNs-based control strategies used in elec- trical systems, but also to optimize the hardware resources re- quired for the neural algorithmic implementation. Based on a functional decomposition, each computing task of our neural APF is achieved with ANNs. These blocks are an Adaline-based disturbance voltage regulation [4], an Adaline- based harmonic estimation [5]–[7], and neural control schemes of the inverter [8]. New voltage decompositions are developed for the two first blocks and are transferred to the Adalines by fixing their structure and their inputs. The control of the inverter, the third block, uses multilayer perceptrons (MLPs) which can be considered as a combination of several simple unit neurons, i.e., Adalines. As will be shown through numerous simulation and experi- mental results, the proposed neural APF is better than the con- ventional APF for the determination and compensation of the harmonic distortions. Moreover, results show excellent behav- iors and performances, as well as robustness and usefulness. The 0278-0046/$25.00 © 2007 IEEE