基于主成分分析的三种中长期预报模型在柘溪水库的应用  

Application of Three Kinds of Medium and Long Term Hydrological Forecast Models in Zhexi Reservoir Based on Principal Component Analysis

作　　者：李薇[1,2] 周建中[1,2] 叶磊[1,2] 卢韦伟[1,2] 姚翔宇[3] 

机构地区：[1]华中科技大学水电与数字化工程学院,湖北武汉430074 [2]华中科技大学数字流域科学与技术湖北省重点实验室,湖北武汉430074 [3]中国建筑西北设计研究院有限公司,陕西西安710018

出　　处：《水力发电》2016年第9期17-21,共5页Water Power

基　　金：国家自然科学基金重点项目(51239004);国家自然科学基金资助项目(51309105)

摘　　要：基于柘溪断面历史旬径流资料，选择1980年－2012年共33a的降雨和流量数据经主成分分析处理后，分别作为多元线性回归模型、BP神经网络模型、Elman神经网络模型的训练样本，对模型参数进行训练；然后对样本进行模拟预报，统计模拟绝对误差和相对误差，同时预报柘溪断面2013年、2014年和2015年的年、汛期、季节和月尺度的流量，预报结果可精确到旬尺度，对比分析三种模型各时间尺度的预报结果，最终确定各模型在柘溪流域中长期水文预报过程中的作用。Based on the history runoff data of Zhexi Reservoir, days from 1980 to 2012 are selected and treated by principal a total of 33 years of rainfall and flow data in a period of ten component analysis as the inputs of muhiple linear regression model, BP neural network model and Elman neural network model respectively to calibrate model parameters. The samples are then simulated, and the absolute error and relative error of simulations are counted. And at the same time, the annual, flood, seasonal and monthly scale runoffs in 2013, 2014 and 2015 at Zhexi Reservoir are forecasted respectively, and the forecast results can be accurate to the scale of ten days. The forecast results of three models are comparatively analyzed and the application of each model in the long term hydrological forecasting of Zhexi Basin is finally determined.

关 键 词：多元线性回归 BP神经网络 ELMAN神经网络 中长期径流预报 主成分分析 柘溪水库 

分 类 号：P338.2[天文地球—水文科学]