Research on Neural Network Training Algorithm Based on PSO-Genetic Algorithm

摘要：

It is not only access to global optimal solution, it also improve the neural network generalization performance when GA is used to optimize the design of neural network. But in the application of the GA, there are also the problems of slow convergence at the latter part and easily trapped into local optimal solution because of the gene deletion. To this end, This article from PSOs ideology, Using a modified mutation operator and population segmentation strategy so that PSO and genetic algorithm combining to design the neural network structure, the initial connection weight and threshold, the learning rate and the momentum factor of evolutionary. And then find the optimal weights and thresholds again by the training samples from the evolution of neural networks. Finally, based on improvement of the PSO-GA neural network training algorithm formed.

作者: Zhijie Pei Li Zhu Guoping Wu

作者单位: School of Computer.China University of Geosciences.Hubei 430074 China Faculty of Mechanical & Electronic Information.China University of Geosciences.Hubei 430074 China

会议类型: 国际会议

会议名称: Third International Symposium on Intelligence Computation and Applications(ISICA 2008)(第三届智能自动化、计算与制造国际研讨会)

会议地点: 武汉

会议语种:英文

页码: 83-86

在线出版日期: 2008-12-19（万方平台首次上网日期，不代表论文的发表时间）