Non linear dynamic system identification using Legendre neural network and firefly algorithm

Abstract

An accurate identification is very essential for the stability of the nonlinear dynamic plant. It is highly challenging job to findout an accurate model with minimum prior knowledge in the system that are highly nonlinear and dynamic such as robotics and autonomous system. Fuzzy neural networks are highly used to deal with the real world identification and classification issue. But if the system carries nonlinear behavior then the FNN fails to provide an accurate solution. The rate of learning speed and computational efficiency are the most essential parameter for identification point of view. In this paper a novel method for nonlinear dynamic plant identification is proposed i.e. Pseudo inverse Legendre neural network (with tanh functions in the hidden layer matrix) optimized by using firefly algorithm (LNNT-FF). The proposed model LNN deals with the chaotic variation of the plant by expanding the input pattern. For better performance the firefly optimization technique is used. The firefly algorithm is inspired by the motion and metaheuristic behavior of the fireflies. Light intensity and the attractiveness variation are the two major issues related in FF algorithm.