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(四川大学 建筑与环境学院, 四川 成都, 610065)；邓红霞(四川大学 水利水电学院, 四川 成都, 610065)；朱兵(四川大学 水利水电学院, 四川 成都, 610065)
Author：(School of Architecture and Environment, Sichuan Univ., Chengdu 610065, China)；(School of Water Resource and Hydropower, Sichuan Univ., Chengdu 610065, China)；(School of Water Resource and Hydropower, Sichuan Univ., Chengdu 610065, China)
收稿日期：2006-06-12 年卷（期）页码：2007,39(2):25-29
期刊名称：工程科学与技术
Journal Name：Advanced Engineering Sciences
关键字：水文预测；非线性预处理；数据平滑度；神经网络
Key words：hydrology prediction;non-linear pretreatment;data smoothness;ANN
基金项目：国家自然科学基金资助项目(50679047)
中文摘要
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
英文摘要
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%.
【关闭】
