Robust hybrid intelligent control using probabilistic feature for active power filter

Abstract

Harmonic problem in the power grid is becoming increasingly serious, which greatly threatens the stable operation and security of power system. As a reliable solution to suppress harmonics, active power filter (APF) faces with severe challenges in dealing with uncertainties for control design. To tackle this issue, an improved terminal sliding mode control (TSMC) scheme based on modified wavelet fuzzy neural network (MWFNN) is designed for APF. This fills the gap in the application of MWFNN with probabilistic feature in APF system and remedies the flaw that WFNN can't effectively handle the stochastic uncertainties before. Firstly, the dynamic model of APF considering external disturbances and component parameter perturbations is obtained according to its internal structure. Then TSMC is designed for APF, and the stability and finite time convergence are proved by Lyapunov theory. In addition, the MWFNN using probabilistic feature to enhance the advantages of wavelet neural networks is constructed to approximate the uncertain function, and adaptive laws of network parameters are also designed. Finally, compared with recurrent neural network (RNN) and TSMC by simulation and experiment, it can be found that the MWFNN has superior performance in harmonic suppression.Index Terms—Terminal sliding mode control, fuzzy neural network, wavelet fuzzy neural network, active power filter