A novel evolutionary algorithm (EA) approach is developed for the implementation of rule based controllers, which increases the diversity of individuals to reduce the opportunities of falling into local optima. The rule based controller is simply composed of a look up table whose dimensions are determined by the numbers of the subranges of error variables required for operation. Many experiments are successfully made on the evolution of the rule based controllers by the novel EA for three illustrative examples: DC servomotor control, ball and beam control, and cart pendulum control. The computer simulations show that the proposed EA is an effective approach to generating control rules which manipulate nonlinearly dynamical systems exceedingly well. It reveals that the EA's learning behavior is very like that of a human expert.
展开▼