Intelligent controllers based on fuzzy systems and neural networks

This dissertation deals with the tuning of intelligent control systems composed of fuzzy systems and neural networks. In general, there are three options for tuning fuzzy controllers: scaling factor tuning, membership function adjustment and rules modification. The objective of tuning is to select the proper adjustment of fuzzy logic controller (FLC) parameters so that the resulting performance would satisfy the desired criteria. The dissertation topics are summarized as follows:; A new analytical scaling factor tuning using an approximation relation of the input variables and the output control action with traditional linear strategy (off-line tuning) and a maximum reduction of number of rules is introduced. Furthermore, we propose a new approach for scaling factor self-tuning to implement an on-line tuning based on a multi-layer perceptron integration with the fuzzy controller.; Frequency specifications are employed to achieve a systematic utilization of fuzzy neural systems. Instead of determining the controller parameters by trial and error graphically using bode plots or using numerical methods, a new tuning method using an adaptive neural fuzzy inference system (ANFIS) is proposed to determine the controller parameters more efficiently.; Finally, FLC tuning based on the inconsistent IF-THEN rules is introduced. The overlap index between two fuzzy sets is utilized to measure the inconsistency and its influence on the performance of a control system is studied. A new algorithm that employs a similarity measure to identify similar fuzzy sets is proposed. A proper inferences scheme for weighted rule base based upon neural network is developed.; The proposed tuning schemes and algorithms are comparatively evaluated by computer simulations studies with exists tuning schemes.