基于贝叶斯框架的LS-SVM中长期径流预报模型研究  

Study of mid-long term runoff forecast based on LS-SVM in Bayesian evidence framework

作　　者：邵骏[1] 袁鹏[1,2] 张文江[3] 钱晓燕[1] 

机构地区：[1]四川大学水利水电学院,成都610065 [2]四川大学水力学与山区河流开发保护国家重点实验室,成都610065 [3]中国科学院地理科学与资源研究所,北京100101

出　　处：《水力发电学报》2010年第5期178-182,189,共6页Journal of Hydroelectric Engineering

基　　金：国家自然科学基金项目(40801175)

摘　　要：为解决最小二乘支持向量机模型采用交叉验证方法确定模型参数耗时较长的问题,将贝叶斯证据框架理论用于最小二乘支持向量机模型参数的优选。选用径向基核函数,建立了中长期径流预报模型。采用岷江紫坪铺水文站1937～2007年的年径流资料进行模型的预测和检验,并与交叉验证方法优选参数确定的最小二乘支持向量机模型及BP神经网络模型进行比较。研究结果表明,基于贝叶斯框架下的最小二乘支持向量机径流预报模型具有较好的预报精度。As the least squares support vector machine（LS-SVM） model with its parameters determined by cross-validation is quite time consuming,a combination method of mid-long term runoff forecast based on LS-SVM within the Bayesian evidence framework is proposed in this paper.By the Bayesian evidence framework theory,the radial basis function（RBF） kernel is used to select and tune the LS-SVM regression parameters.This method is applied to the forecasting of annual runoff at the Zipingpu station of Min River,and a comparison is made with the traditional LS-SVM of cross-validation and BP neural network.The results show that the proposed method is effective in improving forecast accuracy.

关 键 词：水文学 径流预报 贝叶斯证据框架 最小二乘支持向量机 

分 类 号：P333[天文地球—水文科学] TV121[水利工程—水文学及水资源]