两种新型流域非点源污染负荷估算模型的比较

Comparison of two machine learning models for non-point source pollution load of watershed

为探索简单快捷、实用性强的流域非点源估算模型,基于降雨量差值法的基本思想,应用径向基函数(RBF)神经网络和支持向量机(SVM)2种技术,以渭河流域历年降雨量资料为基础建立了相应的流域非点源污染负荷估算模型.并将2种模型估算结果与相关分析法估算结果进行了比较,综合分析渭河流域历年径流量与三者所推得的历年点源负荷的标准差.结果表明,支持向量机的估算结果最为合理,且计算简单,操作便捷,具有实用价值.

To find good and practical methods for predicting watershed non-point pollution load, two watershed machine learning models for non-point source pollution were established by using radial basis function （RBF） network and support vector machines （SVM） based on rainfall deduction method. Correlation of the two models＇ results for the Weihe Basin were analyzed. By analyzing runoffs and the standard deviations of corresponding point source pollution loads of the Weihe Basin in 9 years, SVM model was proved more accurate and reasonable than RBF.