The Research Of Wavelet-Fuzzy Networks And Its Application In Robots Compensating Control

There are two factors that make a full impact on the advancement of control theory. First, industry and other technologies have higher requirement. Second, there is a close relationship between control theory and the development level of technology, especially mathematic tools. Mathematics makes a great difference on control theory. Especially fuzzy logic and wavelet analysis have gotten far-reaching in applications of intelligent control.This paper expatiates the structure of wavelet neural network (WNN) and fuzzy neural network (FNN). Because BP algorithm often settles in undesirable local minima of the error surface and converge too slowly. This paper proposes a fast algorithm based on non-orthogonal wavelet-based functions for WNN and an adaptive BP algorithm that can get global optimal solution for FNN.Based on the research of wavelets and fuzzy inference, the paper has some tentative study of combining wavelets with fuzzy theory. First, a fuzzificated wavelet neural network and its training algorithm are proposed. Second, this paper presents a new model of wavelet-fuzzy network (WFN), and expatiates its training algorithm. Finally, the author proposes a new control strategy, on-line fuzzy adaptive controller with wavelet neural network. This strategy avoids the fluctuation and saturation problems, which exist in the process of training the direct neural controller, and realizes the fast non-overshoot control. Moreover, this strategy can effectively eliminate the effect of noise existing in industrial processes, and then improve the control quality.According to this paper, we can believe that the combination of intelligent control with wavelet analysis and fuzzy theory will advance the development of control field.