隐层节点数经验公式在水库调度规则提取中的应用效果评价  

Evaluation on Empirical Formula of Number of Hidden Nodes Applied in Reservoir Operation Rules Extraction

作　　者：刘宇[1] 钟平安[1,2] 张梦然[1] 孔艳[1] 徐斌[1] 

机构地区：[1]河海大学水文水资源学院,江苏南京210098 [2]河海大学水资源高效利用与工程安全国家工程研究中心,江苏南京210098

出　　处：《水电能源科学》2012年第11期42-44,213,共4页Water Resources and Power

基　　金：全球变化研究国家重大科学研究计划(973计划)基金资助项目(2010CB951102);国家科技支撑计划课题基金资助项目(2009BAC56B03);国家自然科学基金资助项目(51179044)

摘　　要：隐层节点数是确定人工神经网络模型结构的重要参数,但目前尚无通用的确定方法。以水库中长期优化调度规则提取为例,选择了四种典型的隐层节点数经验确定公式,将合格率、确定性系数、平均绝对误差、指标综合占优数等作为评价指标,以全年、汛期、非汛期为统计时段,评价了各经验公式对四项评价指标的拟合和检验效果,并进行了摄动分析。结果表明,Lippmann R P公式应用效果、适应性及稳定性均较好,更适合建立水库优化调度规则提取模型。The number of nodes in hidden layer is a significant parameter of artificial neural network structure and currently there is no general method to calibrate the parameter.In order to study the conventional model structure in extracting reservoir operation rules of neural networks,this paper compares four typical empirical formula of calibrating nodes number in hidden layer,and the qualified ratio,deterministic coefficient,mean of absolute error,comprehensive index are used as indicators to evaluate model calibration and validation performance based on the output of different periods as flood season,dry season and the whole year.The results show that Lippmann R P＇s formula has best overall performance in applicability and robustness,which is fit for extraction of reservoir optimal operation rules.

关 键 词：水库调度规则 人工神经网络模型 隐层节点数 经验公式 

分 类 号：TV697.1[水利工程—水利水电工程] TP183[自动化与计算机技术—控制理论与控制工程]