支持向量机在非点源污染负荷预测中的应用

针对非点源污染形成过程复杂、基础资料不完备的特点,尝试将支持向量机技术用于小样本渭河流域华县站非点源总氮年污染负荷量预测.支持向量机(SVM)能在有限样本情况下,采用结构风险最小化准则,把学习问题转化为一个二次规划问题来获得全局最优解,从而克服了神经网络易陷于局部极小值的缺点.采用华县站1976-1993年总氮非点源负荷及与其产生关系密切的径流、泥沙、降雨资料,前15 a资料用作训练,后3 a资料用作检验.经过与最小二乘支持向量机、BP神经网络和最小二乘回归方法预测结果比较,表明SVM方法预测精度要优于后两者,可用于有限资料条件下非点源负荷预测.

Application of support vector machine method in prediction of non-point source pollution load

The prediction of non-point source pollution(NSP) is faced with several difficulties,including complicated origins and poor data availability.New techniques are needed to improve its prediction.This study applied a Support Vector Machine(SVM) method in estimating annual NSP load at the Huaxian section in the Weihe River basin.The novel SVM method can transform the learning process into a secondary planning problem,and the global optimal solution could be obtained.This method avoids the potential shortcoming of artificial neural networks(ANN),which may be trapped in local optimums.The data series used in the study included NSP total-N load from 1976 to 1993,and the environmental factor series such as flow,sediment,precipitation,etc.The first 15 years' data were used for training,and the last 3 years' data for testing.The SVM method performed better than ANN and the least square regression methods.This study indicated that the prediction of NSP could be improved by the SVM approach under limited data availability.