Optimization and Implementation of Self-Organization Fuzzy Neutral Network Control Algorithm

ABSTRACT

Traditional fuzzy neural network tends to be used for parameters identification and the network structure is separated by grids, so there are obvious defects in this mode of control design. This paper introduces a flexibly model of self-organizing fuzzy neural network according to specific network structure. We analyze its structure and containing parameters and propose an improved nearest neighbor clustering algorithm first for the predicting model of online identification. For parameter optimization, the parameter value acquired at self-organizing learning phase is adopted as the initial value of supervised learning. Then it adopts BP algorithm to adjust the parameter to optimal value based on the same training set, so as to acquire the final model of FNN. The experiments demonstrate that our algorithm can solve the predicting problems of nonlinear system with constraints, and the range and changing rate of control signal. It shows rapid computing speed, better stability and strong anti-disturbance capacity. It is also verified to be suitable for actual engineering control environments.

Keywords: