Genetic optimization for the design of an n-step fuzzy controller

The paper introduces a new methodology for automatic design of fuzzy controllers through the use of genetic algorithms for optimization of the controlled plant trajectories. The approach consists of two phases: a first one, where genetic optimization is used, that allows the extraction of a numerical control law in accordance with the desired specifications, and a second one where the rules and the parameters of the fuzzy controller are identified from the optimized control law through the use of supervised and unsupervised neural networks. The main feature of the proposed technique is the possibility of designing a fuzzy controller that leads the plant to the setpoint through desired paths. Genetic algorithms are used to compute, on the discretized input variable domain, sequences of control values that minimizes a suitably defined fitness, representing the desired specifications. The main feature of the methodology is its independency from the problem and its unsupervised approach. In the absence of a model of the plant, the neural tool above quoted, allows extraction of a fuzzy model from I/O measures. The proposed technique is presented for the fuzzy control of a DC motor