Nonlinear Systems Design by a Novel Fuzzy Neural System via Hybridization of EM and PSO Algorithms

摘要

In this paper, we propose a hybridization of electromagnetism-like mechanism (EM) and particle swarm optimization algorithm (PSO) algorithms to design the proposed functional-link based Petri recurrent fuzzy neural system (FLPRFNS) for application of nonlinear system control. The FLPRFNS has a TSK-type fuzzy consequent part which uses functional-link based orthogonal basis functions and a Petri layer is added to eliminate the redundant fuzzy rule for each input. In addition, the FLPRFNS is trained by a hybrid algorithm-modified EMPSO. The main modification is that the randomly neghiborhoodly local search is replaced by particle swarm optimization algorithm with an instant update particles velocity strategy. Each particle updates its velocity instantaneously one by one and every particle can get best information from system. The modified EMPSO combines the advantages of multipoint search, global optimization, and faster convergence. Simulation results show that the modified EMPSO has the ability of global optimization, advantages of faster convergence and FLPRFNS has effect of higher accuracy.