基于育种算法的BP网络学习研究  

A BP Neural Networks Learning Algorithm Research based on Breeding Algorithm

作　　者：王羽[1] 马光文[2] 杨道辉[1] 

机构地区：[1]二滩水电开发有限责任公司,四川成都610021 [2]四川大学水电学院,四川成都610065

出　　处：《水力发电》2008年第8期12-13,42,共3页Water Power

基　　金：国家自然科学基金重点项目(50539140);国家自然科学基金项目(50679098)

摘　　要：基于育种算法的BP网络学习算法,用育种算法替代传统BP算法中的梯度下降法,使得改进后的算法具有不易陷入局部极小、泛化性能好等特点。该算法应用于径流预测的实验证明:这种算法能够明显减少迭代次数、提高收敛精度,其泛化性能也优于传统BP算法。The BP neural networks learning algorithm based on Breeding Algorithm is proposed in this paper. Among this algorithm, breeding algorithm has substituted the gradient descent method in traditional BP algorithm; this new algorithm does not easily trapped local minima and has better generalization. The algorithm is applied to neural network＇s training in runoff forecasting. The results show： this algorithm can reduce number of training and error obviously; it has better generalization than traditional BP algorithm.

关 键 词：BP网络 育种算法 径流预测 

分 类 号：O157.5[理学—数学]