支持向量回归机在水质预测中的应用与验证  

Water Quality Prediction and Validation by Using Support Vector Regression Machines

作　　者：武国正[1] 徐宗学[1] 李畅游[2] 

机构地区：[1]北京师范大学水科学研究院,水沙科学教育部重点实验室,北京100875 [2]内蒙古农业大学水利与土木建筑工程学院,呼和浩特010018

出　　处：《中国农村水利水电》2012年第1期25-29,33,共6页China Rural Water and Hydropower

基　　金：国家水体污染控制与治理科技重大专项(2008ZX07526-001)

摘　　要：以干旱区浅水湖泊乌梁素海的多年实测pH值为例,在分析支持向量回归机算法(ε-SVR)核函数选取的基础上进行了回归分析及预测,并与线性回归、BP神经网络、RBF网络等算法进行了比较。研究结果显示:①基于径向基核的支持向量回归机模拟效果优于其他核函数;②ε-SVR模拟结果与线性回归(LR)、BP神经网络和RBF网络等算法模拟结果相比,其拟合精度与预测精度均比其他三种方法要高。计算结果充分证明了支持向量回归机有较强的学习能力和泛化能力且该方法可以应用于水质预测研究。Based on the observed pH values of the Wuliangsuhai Lake, the support vector regression （SVR） method of ε-SVR and kernel functions are analyzed. Then the model using ε-SVR with radial basis kernel is established to predict the pH values. In addition, the algorithms of linear regression, back propagation neural network and radial basis function network are introduced to verify the accuracy of the SVR model. The results indicate that： ①the accuracy of simulated results using RBF kernel in ε-SVR are superior to using other kernels; ②the ε-SVR has more excellent fitting accuracy and prediction accuracy than the other methods such as linear regression, back propagation neural network and radial basis function network. The research results suffice to support the conclusion that the SVR has outstanding learning capacity and generalization capacity and the method can be used in water quality prediction.

关 键 词：支持向量回归机 核函数 参数寻优 水质预测 

分 类 号：TU991.21[建筑科学—市政工程]