BP神经网络预测日径流序列的数据适应性分析  

Data Adaptability of BP ANN in the Prediction of Daily Flow

作　　者：李存军[1] 邓红霞[2] 朱兵[2] 王文圣[2] 

机构地区：[1]四川大学建筑与环境学院,四川成都610065 [2]四川大学水利水电学院,四川成都610065

出　　处：《四川大学学报（工程科学版）》2007年第2期25-29,共5页Journal of Sichuan University (Engineering Science Edition)

基　　金：国家自然科学基金资助项目(50679047)

摘　　要：对于不同数据特点的序列神经网络的逼近能力有较大的差异。为了使BP神经网络预测河流日径流的效果有较大改善,分析了S型神经元的训练和数据调整过程,提出了数据对神经网络的主动适应性的表征和判断标准,在提高其平滑度的基础上结合水文数据的结构特点,给出了日径流的非线性变换的几种基本形式。以广西平乐站29年的日径流量为例,通过适当的非线性平滑预处理后用神经网络进行预测,相对误差<10%的天数平均提高47.8%,相对误差<20%的天数平均提高35.6%。In order to improve the prediction precisionof BP ANN for runoff in day, the training and adjusting process of S nerve-cell was analyzed, the standard of data initiative adaptability to ANN was proposed, and several kinds of non-linear transfer for daily runoff with a view to characteristic of hydrology data was givenon the basis of increasing smoothness. The daily runoff in 29 years of Pingle Station in Guangxi province was predicted by ANN after non-linear smoothing transferring. The results showed that days in 10% relatively error averagely increase 47.8%, and days in 20%, relatively error averagely increase 35.6%

关 键 词：水文预测 非线性预处理 数据平滑度 神经网络 

分 类 号：P333[天文地球—水文科学]