Investigation of the Effects of Noise and Plant Parameter Variations for an Adaptive Neuro Fuzzy Inference Controller with Evolutionary Tuning

摘要

Adaptive neuro fuzzy inference systems (ANFIS) are a class of adaptive networks which provide modeling through the neural network and compensation for uncertainty using fuzzy inference. Gradient based learning, such as back propagation, is typically used as the training algorithm. Inherent problems associated with the gradient based approaches include a high probability of getting trapped in local maxima, sensitivity to noise and plant parameter variations, and the inability to provide an optimum solution for non-differentiable or multimodal problems. Instead of using the gradient techniques and/or estimation, evolutionary learning may be used for offline or online tuning of the ANFIS control parameters. In this paper, simulation studies on the effects of noise and plant parameter variations for nonlinear systems employing an ANFIS controller with evolutionary tuning are performed. Results show that the tuner in combination with the adaptive nature of the ANFIS structure compensate for noise or parameter variations when compared to classical ANFIS control.