基于高斯过程回归模型的径流短期预测研究  

Short-term Runoff Prediction Based on Gaussian Process Regression Model

作　　者：黄亚 易灵[1] 肖伟华[2] 侯贵兵 李媛媛[1] HUANG Ya;YI Ling;XIAO Weihua;HOU Guibing;LI Yuanyuan(China Water Resources Pearl River Planning,Surveying&Designing Co.,Ltd.,Guangzhou 510610,Guangdong,China;State Key Laboratory of Simulation and Regulation of Water Cycle in River Catchment,China Institute of Water Resources and Hydropower Research,Beijing 100038,China)

机构地区：[1]中水珠江规划勘测设计有限公司,广东广州510610 [2]中国水利水电科学研究院流域水循环模拟与调控国家重点实验室,北京100038

出　　处：《水力发电》2020年第12期9-12,共4页Water Power

基　　金：国家重点研发计划项目(2018YFC1508200,2017YFC-0404701);国家自然科学基金资助项目(51669003);广西研究生教育创新计划资助项目(YCBZ2018023)。

摘　　要：为寻求更为精确的径流量预测方法,研究利用传统BP神经网络、支持向量机(SVM)以及高斯过程回归(GP)三种模型对径流量进行预测,并以广西天峨水文站日入库径流量为例进行预测实践和分析。结果表明,高斯过程回归模型对径流短期预测具有较高精度,预测平均相对误差绝对值为1.29%,最大相对误差绝对值为2.71%,预测精度和泛化能力均优于传统BP神经网络模型和支持向量机模型,是进一步提高径流预测精度的有效方法。To seek a more reasonable runoff prediction method,three kinds of models including traditional BP neural network,support vector machine(SVM)and Gaussian process regression(GPR)are used to predict runoff,and the average daily runoff of Guangxi Tiane Hydrological Station is used as study case for runoff prediction and analysis.The results show that the shortterm prediction of runoff based on GPR is more accurate,and its absolute value of average relative prediction error is 1.29%and the maximum absolute value relative error is 2.71%.The prediction accuracy and generalization ability of GPR is superior to the traditional BP neural network and SVM.It is one of the effective methods to improve the prediction precision of runoff.

关 键 词：径流预测 高斯过程回归 BP神经网络 支持向量机 

分 类 号：TV12[水利工程—水文学及水资源]