对传神经网络算法的改善及其应用.pdfVIP

第 39 卷第 5 期 中南大学学报自然科学版( ) Vol.39 No.5 2008 年 10 月 J. Cent. South Univ. (Science and Technology) Oct. 2008 对传神经网络算法的改进及其应用 宋晓华，李彦斌，韩金山，牛 达 (华北电力大学 工商管理学院， 北京 102206) 摘 要：针对传统对传神经网络(Counter propagation networks，即 CPN)要求输入向量必须均匀分布以及隐含层神 经元个数难以确定，其应用受到很大局限等问题，对 CPN 算法进行改进并运用于电力负荷预测。研究结果表明： 通过改进 CPN 算法的初始权重设置规则，克服了对输入向量限制过于严格的不足；通过优化算法运行步骤，提 高了算法的运行效果；改进后的 CPN 算法比 BP 算法所得预测结果误差小，比目前电力负荷预测研究中 RBF 和 Elman 神经网络所得预测结果误差也小；与 BP 算法相比，CPN 改进算法的预测精度提高 4%左右，运算时间减少 45% ，适用于电力负荷的预测。 关键词：负荷预测；人工神经网络；BP 神经网络；对传神经网络 中图分类号：TP39 文献标识码：A 文章编号：1672−7207(2008)05−1059−05 An improved counter propagation networks and its application SONG Xiao-hua, LI Yan-bin, HAN Jin-shan, NIU Da (School of Business Administration, North China Electric Power University, Beijing 102206, China) Abstract: Based on the fact that the input vectors of counter propagation networks(CPN) are supposed to be uniform distribution,，and it is difficult to choose the number of their hidden lay neurons, its application is restricted in a few fields, CPN algorithm was improved and was applied to power load forecasting. The results show that through changing the setting rules of the initiation weight, the problem of too strict limitation to input vectors can be solved. Based on the optimization of the operation process, the efficient of the CPN can be enhanced. The simulation error using the ameliorated CPN is lower than those with BP, RBF and Elman networks. Compared to traditional BP networks, the forecasting accuracy using the ameliorate ameliorated PN improves about 4%, and the com