水库优化调度ANN模型隐层节点数经验公式比较  

Empirical Formulae Comparison for Number Selection of Hidden Layer's Nodes in ANN Model Applied for Reservoir Optimal Operation

作　　者：刘宇[1] 钟平安[1,2] 张梦然[1] 顾锦 孔艳[1] 

机构地区：[1]河海大学水文水资源学院,江苏南京210098 [2]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 [3]浙江省衢州市水文勘测站,浙江衢州324000

出　　处：《水力发电》2013年第5期65-68,共4页Water Power

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

摘　　要：人工神经网络(ANN)应用于水库中长期调度时,隐层节点数尚无公认的选择标准。选取常用的4种确定隐层节点数的经验公式,采用合格率、确定性系数、平均绝对误差、指标综合占优数等指标,以全年、汛期、非汛期为统计时段,选取三峡、池潭、隔河岩等3个水库为对象,评价各经验公式对不同水库的普适性及可替代性。结果表明,各经验公式对不同水库、不同时段的应用效果存在一定差异,但对发电量最大为目标的模拟调度效果相差不大,可优先考虑使用隐层节点数最少的Lippmann公式建模。There are no general rules to calibrate node number of hidden layer when Artificial Neural Network （ANN） is applied to study reservoir optimal operation. For assessing the substitutability and universality of four typical empirical formulae which often used to calibrate the node number of hidden layer, a ANN optimal operation model is used to study the optimal operation of Three Gorges, Chitan and Geheyan reservoirs respectively for evaluating models＂ calibration and validation performance, in which, the qualified ratio, deterministic coefficient, mean absolute error and comprehensive index are used as indicators and the flood season, dry season and whole year are selected as statistical periods respectively. The results show that the four typical empirical formulae have some differences for different reservoirs and different periods, but no significant difference when taking maximum generation as optimal target. The Lippmann＇s formula with least node numbers is recommended.

关 键 词：水库中长期调度 调度规则提取 人工神经网络(ANN) 隐层节点数 

分 类 号：TV697.11[水利工程—水利水电工程]