On the Design and Implementation of Hybrid Intelligent Controllers

1 During the last years, in the eld of \intelligent control" [1], several methodologies have shown up, like fuzzy, neural and genetic control, providing for the rst time practical solutions for nonlinear control problems, or supplying better or at least alternative solutions for some classic problems. Each of these new approaches has its own advantages and drawbacks, and that explains why they have been applied only to speci c elds. Thinking that most applications are better solved with an integrated approach, we developed a hybrid approach that merges some of the following control paradigms: fuzzy control, neurocontrol, linear control, optimization algorithms like simulated annealing and genetic optimization, and nite state automata [2, 3]. An immediate synergy can be found between fuzzy and neural control. The former exploits an important feature of Fuzzy Systems (FSs) [2], that is the capability of building the rule base by mapping the knowledge of human experts onto it. On the other hand, Neural Networks (NNs) can be trained with a suitable set of data samples, without being able to map available human knowledge to network weights. Recently, NNs and FSs have been merged into Weighted Radial Basis Function (WRBF) networks [4], in which fuzzy rules can be immediately mapped into the network structure, and then trained or optimized in the same way as traditional NNs. This can lead to a noticeable