Applications Of Wavelet Transform And Wavelet Neural Network In Chemical Process Control

This thesis fully discusses the theory of wavelet transform, and two applications of it—online real time wavelet filter and wavelet neural network. In order to apply wavelet neural network in predictive control, this essay used genetic algorithm to search the best inputs of the given outputs of wavelet neural network and got good result.In order to make the theory of wavelet transform easy to understand, this thesis try to state on application basis. Tedious mathematical deductions are avoided as much as possible. As in the discussion of the essential of wavelet transform, some simple concepts such as vector and component, instead of linear transform, are use to explain the wavelet transform. Figures and flowcharts are also added to help to make the explanations easy to follow.Wavelet filter maps the given signals to wavelet domain, and then remove the components which stands for fast changing parts of signal, which interpreted as the noises. Because this process is independent of the models, wavelet filter is robust. The validations by artificial and industrial signals prove that online real-time wavelet filter can restore contaminated signals better than low pass filter. The wavelet and transform level used are two parameters in online real-time wavelet filter. Their effects are also studied. The complexity of the online real-time wavelet filter is presented. The asymptotic running time of wavelet filter is related to the wavelet filter length F and transform level L, and can be stated as 6 (F~2·2~L).The wavelet neural network combines the wavelet transform and neural-network, and possess the advantages of them. Thus it can be use as a promising modeling tool in lots of areas. In order to prove the function approximation ability of the wavelet neural network, several group of SISO and MISO modeling tests were carried out to validate the learning algorithm and the generalization ability of the wavelet neural network. Modeling compare of the wavelet neural network, the simplified wavelet neural network, which imple- mented by former researchers, and RBF neural network with industrial data of circulating fluidized bed boiler (CFBB) were presented. The results suggest that the wavelet neural network presented in this thesis gave better modeling result than RBF neural network and the simplified wavelet neural network. Finally, the suggestions about applications of wavelet neural network are given.The genetic algorithm is the extended part of wavelet neural network. It is the preceding work for the predictive control with wavelet neural network. This thesis gives the fully discuss of the principles and programming flow of genetic algorithm. The experiment results of searching input of wavelet neural network by genetic algorithm were quite good.