Co-evolutionary Algorithm For Hierarchical Fuzzy Control Of The Inverted Pendulum

In this paper we design a hierarchical fuzzy controller for the inverted pendulum problem. A co-evolutionary algorithm is used to learn a complete set of fuzzy rules in the hierarchical structure and also to learn key parameters of the input and output membership sets in each layer in order to achieve a more ‘ damped’ convergence of the state variables and smoother controller function. We compare our results with previous research which examined the problems of hierarchical decomposition associated with determining the number of layers and which input variables are associated with each layer. This research had assumed a uniform distribution of fuzzy membership sets for input and output variables.