摘要
利用粒子群(PSO)算法替代BP算法对小波神经网络(WNN)进行训练,针对局部极小值问题提出了改进的PSO算法,即判断当粒子陷入局部极小时将其重新初始化,并对小波的平移和伸缩参数的初始化进行了研究,避免了网络的盲目搜索,减少了迭代次数.通过非线性函数逼近的仿真结果表明,上述措施有效提高了网络搜索成功率,在一定程度上解决了局部极小值的问题.
PSO algorithm to instead of the BP algorithm for the wavelet neural network training was used. The advanced PSO algorithm is proposed by aiming at the local minimum problem. The initialization method of the scale parameters and the translation parameters of the wavelet are proposed in order to avoid the network blind searching. According to these means, the network convergence rate is greatly enhanced and the iteration is decreased. Proven by the simulations of the nonlinear function approximation, the network convergence rate is improved efficiently and the local minimum problem is solved in a certain degree.
出处
《华中科技大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2006年第8期43-45,共3页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
航天技术创新基金资助项目
关键词
小波神经网络
粒子群优化算法
平移参数
伸缩参数
wavelet neural network (WNN)
particle swarm optimizer (PSO)
translation parameter
scaling parameters