On the use of Hamming distance tuning for the generalized adaptive neural network fuzzy inference controller with evolutionary simulated annealing

The design of robust controllers to handle noise and system variation is a challenging problem and research in this area continues; one approach which has gained success is to employ a neural network to learn about the unknown plant and fuzzy inference to compensate for the uncertainty (GANFIS control). As with most controllers, the design of GANFIS controllers requires proper parameter selection as well as tuning due to variations in the plant or environment. Evolutionary learning has been shown to be feasible to tune these parameters, but there may be some difficulty in selecting the learning factors. Previously, we addressed this problem by designing a fitness function containing both prediction and heuristics for measuring system performance and extending the approach by integrating simulated annealing features for faster convergence. In this paper, we employ a tuning strategy based upon the Hamming distance to select the key convergence parameters of the evolutionary learning algorithm: the cross-over and mutation rates. To study the approach, we compare the case when simulated annealing is applied for constant values; then an adaptive tuned version of the simulated annealing algorithms is applied. Results show that by tuning, the GANFIS controllers produce an overall tracking performance with a lower error trajectory, thus illustrating the feasibility of this additional feature of the GANFIS controller.