Design And Research On Wavelet Network Controller Based On QPOS

The Wavelet Neural networks is a new type of neural network based on theartificial neural network and wavelet analysis combining the benefits of artificialneural network and wavelet analysis. At first, the frequency localization of theimportant characteristics of wavelet transform is fully utilized, Secondly, theartificial neural network itself own learning ability, the ability of fault toleranceand strong approximation are fully played. However, a certain problem may beproduced in the wavelet neural network structure when the artificial neuralnetwork with the combination of wavelet analysis and the problem of localminimum or not converge appear easily in the wavelet neural network model.The research of learning algorithm and structural properties of the waveletneural network is the first part of main content in this paper. The goodperformance wavelet controller can be designed according to the above researchresults. The characteristics will be researched in the specific simulationexperiments. In the process of research the algorithm many inadequacies in thetraditional training algorithms of the wavelet neural network algorithm. In orderto improve traditional algorithm, quantum-behaved particle swarm algorithm(QPSO) was introduced in this paper. The basic principle and flow of QPSO isanalyzed briefly. The advantages of QPSO are researched and the taking themodel of quantum δ as the basic form of the QPSO algorithm. The probabilitydensity function and the wave function of the particle position can be calculatedand the convergence of the QPSO algorithm is researched. The basic evolutionequation of the QPSO algorithm will be derived according to the Monte Carlomethod and the convergence for the particle position in the basic evolutionequation can be judged. Therefore the two good research strategies can beobtained and the complete algorithm process can also be obtained. The advantages of the QPSO can be obtained finally by comparing the algorithms ofQPSO and PSO.Secondly, according to analysing the algorithm above, the research designsa wavelet controller with better performance. The parameters including learningfactors, speed of evolution and particle aggregation degree are added to theQPSO algorithm and the parameter iteration update of the QPSO algorithm isimproved to increase the control parameters and to obtain the complete trainingresults improved. The wavelet neural network controller is optimized by applyingquantum behaved particle swarm optimization.Finally, the simulation on its performance is studied. Selection of the mostcommonly used Morlet wavelet as the wavelet basis function, and to vote for thedouble inverted pendulum control object, using software to build the simulationmodule. The simulation module is built by the software. The strong stability andanti-interference ability of the controller can be verified by the simulation andphysical control experiments.