Fuzzy controller synthesis with neural network process models

The general problem considered is the optimization of a fuzzy controller using a neural network model for the process in the optimization procedure. A fuzzy controller is synthesized for a simple nonlinear process. An algebraic feedforward neural network is used for modeling the process, and an optimization criterion based on set point error is selected. Initially, a subset of the fuzzy controller parameters is optimized using a genetic algorithm. Subsequently, the search is extended to other controller parameters, including the scaling factors used to map measured variables to the appropriate universe of discourse, and the slopes of piecewise linear membership functions. Significant improvement in control performance is observed relative to an unoptimized fuzzy controller.<<ETX>>