Adaptive fuzzy modelling and identification with its applications

Abstract

A novel fuzzy identification approach of complex systems is proposed based on an updated version of pi-sigma neural network. By taking advantage of the learning and nonlinear mapping abilities of the neural networks, the proposed method has the following characteristics. The consequence function of each fuzzy rule can be a nonlinear function, which makes it capable to deal with the nonlinear systems more efficiently. Each parameter of the consequence functions is convenient to be adjusted on-line so that automatic rule modification can be realized. The membership function of each fuzzy subset can be modified easily on-line. In this way, the cumbersome task to form fuzzy rules from a set of given data and to prescribe an appropriate membership function for each premise variable can be avoided. Examples of fuzzy modelling in weather forecast, burden optimization and fuzzy control system design show that our algorithm is effective in quite wide fields. We also apply our scheme to the decoupling control of a rigid robot manipulator. With the help of the self-organizing fuzzy mechanism, the controller acquires a satisfying performance under large parameter uncertainties