黄龙滩水库中长期径流预报方法研究  

Research on the Medium and Long-term Runoff Forecast Method of Huanglongtan Reservoir

作　　者：常新雨 周建中[1] 方威 王彧蓉 黄靖玮 CHANG Xinyu;ZHOU Jianzhong;FANG Wei;WANG Yurong;HUANG Jingwei(School of Civil and Hydraulic Engineering,Huazhong University of Science and Technology,Wuhan 430074,Hubei,China)

机构地区：[1]华中科技大学土木与水利工程学院,湖北武汉430074

出　　处：《水力发电》2021年第8期10-14,93,共6页Water Power

基　　金：国家自然科学基金重点支持项目(U1865202);国家自然科学基金重点项目(52039004);中央高校基本科研业务费专项资金资助项目(2021yjsCXCY018,2021yjsCXCY020,2021yjsCXCY-042)。

摘　　要：基于黄龙滩水库和潘口水库历史旬月径流数据,选取其2012年~2018年的径流、降雨数据进行灰色关联分析,筛选出与黄龙滩水库入库径流关联度最高的7个预报因子,建立深度神经网络(DNN)、Elman神经网络和支持向量机(SVM)径流预测模型,对模型参数进行训练,统计模型训练期和检验期的确定性系数、洪峰合格率、均方差和平均相对误差。预报效果表明,3种模型在黄龙滩水库中长期径流预测上效果较好,精度较高,误差较小,预报结果对于黄龙滩水库水文预报上具有重要意义。相比于深度神经网络和Elman神经网络,支持向量机在洪峰预报上误差更小,且具有更高的预测精度。Based on the historical runoff data of Huanglongtan Reservoir and Pankou Reservoir,the runoff and rainfall data from 2012 to 2018 are selected for grey correlation analysis,and seven predictors with the highest correlation with the inflow runoff of Huanglongtan Reservoir are screened out.Then the deep neural network(DNN),Elman neural network and support vector machine(SVM)runoff prediction models are established,the model parameters are trained,and the certainty coefficient,peak qualification rate,mean square error and average relative error of the model training period and test period are calculated respectively.The forecast results show that above three models have good results in the medium and long-term runoff forecast of Huanglongtan Reservoir with higher accuracy and smaller errors.The forecast results are of great significance to the hydrological forecast of Huanglongtan Reservoir.Compared with DNN and Elman neural network,the SVM has smaller error in flood peak forecasting and has higher prediction accuracy.

关 键 词：径流预报 灰色关联分析 深度神经网络(DNN) ELMAN神经网络 支持向量机(SVM) 黄龙滩水库 

分 类 号：P338.2[天文地球—水文科学]