Study Of Nonlinear Systems Modeling Based On Wavelet And Fuzzy Technology

Along with wavelet network development and mature, wavelet network applies in each aspect research more and more. It widely has applied in technical domain, such as signal processing, imagery processing, pronunciation recognition and synthesis, pattern recognition, machine vision, mechanical breakdown diagnosis and monitoring, and has yielded the very big achievement. And people start to study its application in the domain of the system identification, especially in the nonlinear system identification. At the same time, the fuzzy neural network have also many researches in various aspects. In order to use the superiority of fuzzy, wavelet and neural network well, we combine them to constitute new model, research the study algorithm, and its application in the nonlinear system. On the base of multi-resolution analysis theory and the fuzzy model, combining fuzzy with wavelet network, we propose the fuzzy wavelet network model. This network draw fuzzy model into wavelet network. It enhances the approximating precision of network, and simplifies system structure with no increasing the wavelet base function.In this paper, the first and second chapter mainly introduce the research goal, the significance, the present situation and the elementary knowledge.The third chapter introduces some concrete contents of fuzzy neural network and wavelet network. The fourth chapter introduces fuzzy wavelet network.In the fuzzy wavelet network model, each fuzzy rule corresponds a sub-wavelet network, in which wavelet base has the samescale factor. Namely under the identical scale the wavelet base group is obtained by translating wavelet function. Therefore under these different scale factors the sub-wavelet network can catch the characteristic of the approximated function under different time-frequency range. And fuzzy rules decide contribution degree of various sub-wavelet network to the entire network output. By this, it reduces the difficulty of choosing the wavelet by fuzzy inference. Moreover in the fuzzy rule the wavelet base of different scale value also can fully appear the system essential characteristic in various aspects. Comparing with traditional wavelet network, the part approach ability, the convergence rate, the accuracy and extensive ability have been enhanced by learning displacement parameter the wavelet function and adjusting the fuzzy rules in fuzzy wavelet network model. And what's more, through the MATLAB simulation, it has been proved its superiority and application in the system identification. The last chapter introduces the dynamic recurrent fuzzy wavelet network model. The main thought is increasing some feedback links, namely, the output of each membership function feedback to itself, which can obtain the recurrent characteristic. Dynamic recurrent fuzzy wavelet network has following merits: it may effectively approximate function and has the on-line learning capability, has the smaller network size and the quicker study speed which are proved by simulation.The development of fuzzy wavelet network will be certainly exciting. Although the research work of fuzzy wavelet network is an arduous and long-term duty, it has a point which causes the human gratified: Now it only shows talent for the first time, has huge potentiality and opportunity. It has glorious prospect and very big research value.