Design And Research On Controller Based On Wavelet Network

Wavelet neural network is a novel network combined with wavelet analysis and artificial neural network. Because the wavelet neural network inherits the self-learning ability of neural network and the time-frequency localization of wavelet analysis, it can tolerant more fault and approach function more closely. The convergence speed, fault tolerance and forecasting results are superior to the traditional feed-forward neural network in dealing with multivariate, nonlinearity and inaccuracy, so it has extensive application prospects. The study algorithms of wavelet network, parameters initialization and methods of structure optimization are researched in the paper, and a controller is designed with excellent performance.A classification scheme based on wavelet neural network is proposed, the method avoids"the dimension disaster"in view of mapping study of wavelet network. In order to overcome shortcoming based on the back propagation algorithm of wavelet neural network to choose the parameters, using genetic algorithm to optimize parameters of wavelet neural network is researched. Combination the wavelet neural network of self-learning ability, time-frequency localization and genetic algorithm of global search capability, an effective method is proposed. The initial parameters of the network are determined by the genetic algorithm, because of genetic algorithm in local search space does not have the ability to fine-tune, when the error is no longer significant change, wavelet neural network is used to train. the search direction and step have calculation error in using the conjugate gradient algorithm, so the direction of the search can not guarantee each conjugate, the algorithm is improved to couple with search position and narrow range as soon as possible to find a small point in linearity. In so many wavelet function, wavelet selection criteria is not an optimum, in the paper, using DOG (Difference of Gaussian) wavelet, Morlet wavelet, Mexihat wavelet, Shannon wavelet as the wavelet function, the training effects of them are obtained and compared. Then the effectiveness of the wavelet neural network controller is tested by the double inverted pendulum, and the controller has strong anti-interference ability.