Design And Research On Wavelet Network Controller Based On Improved Particle Swarm Optimization

As the perfect combination of artificial neural network theory and wavelettheory, wavelet neural network has both of self-learning function, adaptabilityand robustness of neural network and time-frequency local characteristics ofwavelet. Its fault tolerance and approximation ability is strong. In dealing withuncertainty, unknown and complex nonlinear problem, wavelet neural networkhas more advantages than the conventional feedforward neural networks, so ithas the more broad application prospects. This paper mainly studies learningalgorithm and structural characteristics of wavelet neural network, and thendesigns a controller which has excellent performance, and verifies its validity inthe simulation and real-time control experiments.Similar to the genetic algorithm, the particle swarm optimization algorithmis also a kind of optimization tool based on iteration, but the difference is, it doesnot have the variation operation and crossover operation of genetic algorithm, butthe particles determines flight direction to search optimal value throughmeasuring the flight experience of population and itself. According to thecharacteristic that basic particle swarm algorithm is easy to fall into the localoptimal value, in order to strengthen the ability that algorithm escaped from thelocal optimal value, this paper brings in cross factor of genetic algorithm andindividual average extremum, and puts forward an improved particle swarmalgorithm to expand the cognitive scope of particle while maintaining thediversity of the population, which enables particle to get more information toadjust their own states. In addition, in order to weigh the convergence precisionand convergence rate of algorithm, weighting factor is added to the algorithm.The improved particle swarm optimization algorithm is used for optimizingwavelet neural network controller.Then Morlet wavelet is selected as wavelet function to design wavelet neural network controller. Lastly double invertedpendulum is selected as a controlled object to build the simulation module, andstability and anti-jamming ability of the controller is proved effectively insimulation experiment and real-time control experiment.