Evolutionary Learning in Neural Fuzzy Control Systems

Neural networks and fuzzy logic have emerged recently as powerful techniques for implementation of complex nonlinear control systems. A persistent bottleneck, however, remains the manner in which suitable control strategies can be derived in both approaches. Each method requires optimization of several parameters, which is usually not a trivial problem for many optimization techniques. This chapter will discuss important issues related to rule-based fuzzy and neural systems. These issues are knowledge representation and knowledge acquisition. Knowledge representation refers to the manner in which knowledge can be expressed in connectionist neural systems in order to preserve the advantages of fuzzy-based systems. Knowledge acquisition, on the other hand, is the manner in which the rules that represent fuzzy knowledge are obtained and stored taking advantage of neural learning techniques. The primary objective of using fuzzy-neural approaches is to develop systems that are capable of automatic acquisition of knowledge for given network representations. In that regard the chapter will also discuss the implementation of learning using genetic algorithms in such neural-network-based fuzzy systems. Typical applications for such systems include the control of unknown nonlinear plants and this will be demonstrated in this chapter via a case study involving aero-engine control.