基于支持向量机的城市道路降雨径流污染预测模型

为实现城市降雨径流污染有效预测,以文献中的实测数据作为样本,选取雨型、平均雨强、峰值雨强、降雨历时、雨前干期、大气降尘量、PM 10、车流量、路面材料及城市功能区等10项影响因子作为模型输入量,选取径流污染指标COD的场次降雨平均浓度EMC及初期冲刷指数FF30作为模型输出量,基于支持向量机(SVM)构建了城市道路径流污染预测模型。结果表明:EMC-SVM及FF30-SVM模型均具有较高的预测精度,EMC-SVM模型校验参数RMSE、MBE远小于数据集EMC均值,CE、CC达到0.815及0.933;FF30-SVM模型校验参数RMSE、MBE远小于数据集FF30均值,CE、CC分别为0.866及0.932;选用径向基函数(RBF)作为核函数,使用k折交叉验证法对模型参数进行寻优,对于EMC-SVM及FF30-SVM模型寻得的最优参数(c,g)分别为(64.0,0.001953125)、(2.0,0.0625)。

Pollution prediction model for urban road rainfall runoff based on support vector machine

In order to effectively predict urban rainfall runoff pollution,a road runoff pollution prediction model was built based on support vector machine(SVM)taking the data in previous literatures as sample.This model selected 10 influence factors(rain type,average rain intensity,peak rain intensity,rainfall duration,dry period before rain,atmospheric dust amount,PM 10,vehicle flow,surface material and urban functional area)as the model inputs,and selected event mean concentration(EMC)and fist flush index FF30 as the model outputs.The results show that:①Both EMC-SVM and FF30-SVM models have high prediction accuracy.The calibration parameters(RMSE and MBE)of EMC-SVM model are much lower than the average EMC.The CE and CC of EMC-SVM model are 0.815 and 0.933,respectively.The calibration parameters(RMSE and MBE)of FF30-SVM model are much lower than the average FF30,The CE and CC of FF30-SVM model are 0.866 and 0.932,respectively.②Using radial basis function(RBF)as the kernel function and the k-folding cross validations as the validation method,the optimal parameters of EMC-SVM and FF30-SVM models are found to be(64.0,0.001953125)and(2.0,0.0625),respectively.