Design And Research Of Wavelet Neural Network Controller Based On The Improved Genetic

Wavelet neural network is an algorithm that combines the neural networkwith wavelet analysis theory. It not only has the characteristics of time-frequencylocalization of wavelet analysis but also inherits the artificial neural network’spowerful self-learning function. Because of its powerful approximation abilityand fault-tolerant function, wavelet neural network is better than the traditionalfeed-forward neural network in the aspect of fault-tolerant function, the forecastresults and the convergence speed etc, when handling a series problems ofcomplex nonlinear, unknown and uncertain systems. Wavelet neural network haswidely practical significance.This paper has mainly studied the learningalgorithm and structure of the wavelet neural network, then designed a controllerwith excellent properties.In this paper, the genetic algorithm is selected to optimize wavelet neuralnetwork. Although the genetic algorithm has many unique characteristics inglobal optimization, there are still several problems in it. The problems areshown as follows: the initial parameter selection is subjective; the convergencerate is slow; the premature convergence is easy to occur during optimization;each iterative error is not stable and so on. In order to slove those problems, thispaper proposes an improved genetic algorithm based on the niche technology.According to the number of generations and individual fitness, the improvedgenetic algorithm changes the adaptive crossover probability. And the improvedgenetic algorithm combines with the niche technology, which makes theindividual in population survive and evolve in specific environment. Not only awide variety of population is ensured, but also the premature convergence is improved. This algorithm gets significantly improvement in the speedconvergence and global optimization. Experiments prove the effectiveness of theimproved algorithm used in the wavelet neural network controller.Based on the former research and the characteristics of the secondaryinverted pendulum, this paper constructs a module of the wavelet neural networkcontroller which utilized the improved genetic algorithm. Then the actual controlexperiments prove the effectiveness of the controller.