Research Of Active Power Filter On Neural Network Control

With the rapid development of the industrial manufacture and electronic technology,the electronic equipment and trading devices become more prevalent in the various fields.The interference of some non-linearity loads as frequency conversion, commutate equipment, transfuses a large quantity of harmonic wave into the power system. The growing tendency of the abnormity and accident which caused by the harmonic wave has already become the social harmfulness in the power system.Passive filters and active power filter(APF) are the two primary approaches for eliminating harmonics. Comparing with passive filters, APF have many advantages, which are not easily affected by system impedance and can be used to dynamically compensate harmonic. APF will play an important role in eliminating harmonics. Researches on APF are focused on main circuit structure forms(series-type, shunt-type, series-and shunt-type),harmonic current detecting approaches and control strategies to compensate current etc.This paper mainly deals with compensate current control approaches based on neural network.With the premise of harmonic measurement accuracy and the real-time guaranteed on,current compensation methods of quality control direct impact on the compensation effect of load current. The main control method of current compensation includes hysteretic loop comparison control and triangular wave comparison control. Hysteretic loop control is a real-time control, high control accuracy and fast current response speed, and simple hardware. However, because of their fixed hysteretic width, the current scope of the following error is fixed and the switching frequency of power semiconductor device is changed; With the control of triangular wave comparison control, the switching frequency of power semiconductor device fixed, but the hardware is more complicated with a large error, low control accuracy and response speed. As APF device current-voltage inverter control is a non-linear process control, it is very suitable for using neural network to control the APF. Therefore, this paper presents a method based on artificial neural network to control the APF. The simulation results show that when the ANN can be trained for suitable to output the switching control signal of APF compensate current, there is a good control results, and when non-linear load has a change, there is also a good generalization ability.APF neural network are used to control through off-line training, at the same time, online learning can also track changes in the non-linear load current, even necessary to compensate for the harmonic parts of the non-linear load have a large change, the great changes compensation can still achieve good results without having to off-linear training.This neural network control can be used to prove the feasibility and correctness of APF.