Dissertation > Industrial Technology > Automation technology,computer technology > Automated basic theory > Artificial intelligence theory > Artificial Neural Networks and Computing

Adaptive mutation algorithm and structure of wavelet neural network learning based on particle swarm optimization

Author TangLiangYu
Tutor LinZuo
School Shanghai Normal University
Course Applied Computer Technology
Keywords Wavelet transform Wavelet neural network Particleswarm optimization self-adapting mutation Wavelet frame
Type Master's thesis
Year 2012
Downloads 59
Quotes 0
Download Dissertation

Wavelet transform, a mathematic theory and method, is widelyused in research and application fields currently. Wavelet neuralnetwork (WNN) is a kind of feed-forward neural network in which thecells activations function is wavelet functions. Wavelet neural network isalso used in research and application fields. Based on the research ofstudy algorithms of WNN and wavelet frame theory, this paper proposea improved study algorithms for WNN and a new neural networkstructure.This paper contrapose the defects of BP study algorithm inwavelet neural network, propose a new study algorithm: adaptivemutation cooperative particle swarm optimization. This study algorithmcombines the high efficiency of cooperative particle swarmoptimization algorithm and global optimization ability of adaptivemutation. This new algorithm has strong search capability in local region,and with the adaptive mutation method, this algorithm can overcomethe restriction of the local extremum. This paper uses this new studyalgorithm for training of continuous type wavelet neural network tosearch the neural network parameters. After the research about theframe wavelet decomposing and composing process based on theoryof normal orthogonal basis, this paper propose a neural networkstructure that based on the theory of frame wavelet. Also use thealgorithm of adaptive mutation cooperative particle swarmoptimization to train this type of neural network.In the end of this paper, propose using wavelet neural networkbased on the new algorithm for the forecast process on the time seriesof Chinese volume of trade. We also compare with other typicalforecast models for this forecast sample. We also use four types ofsamples that downloaded from UCI repository to test the capability of the classifying based on the study algorithm that proposed in this paper.

Related Dissertations
More Dissertations