Desigh Of Neural Network PI Controller For Active Power Filter

In our country, electric power industry develops vigorously, according with power load increasing sharply at the same time. Then nonlinear load in distribution system is increasing, which causes serious pollution to power quality of the power supply system. On the other hand, modern automation and intelligent industrial electrical equipment also put forward higher requirement of power supply system. How to suppress harmonic wave in the power grid and improve the power quality has become a research hotspot recently. Therefore, the harmonic management has become urgent problems to be solved in electrical engineering field.Active Power Filter (APF) can eliminate harmonic current by generating compensating current which has the same amplitude and opposite phase with source harmonious wave. To track changes of reference current quickly and accurately, controller design method plays an important role. In order to obtain ideal compensation characteristics, considering the nonlinearity, time-varying characteristics and parameters uncertainty, difficulty to establish precise mathematical model, Neural Network (NN) PI controller is designed firstly. PI controller is designed to control the APF while neural network is trained to set the parameters of PI controller, which can gain higher control precision. Because PI controller is difficult to achieve no deadbeat control, a recursive integral PI controller together with neural network is proposed to achieve non-steady-state error control of APF secondly. However, while neural network is used to tune parameters of PI controller, selection of initial parameters has a great influence on the final result. Finally, an iterative NN recursive integral PI controller is presented. It can not only reduce initial tracking offset, but also speed up initial tracking speed. The simulation results prove effectiveness of the algorithms above.