【摘要】 为实现对污水处理厂出水总氮（TN）浓度的准确预测，收集并分析了某污水处理厂的317条实际运行监测数据，将数据进行预处理后，结合关联度较高的进水TN等6项常规水质指标，利用随机森林、KNN及SVR 3种机器学习模型对出水TN进行预测，并将预测值进行对比分析。结果表明：KNN和SVR 2种模型的决定系数分别为0.759和0.723，低于随机森林模型的0.814;KNN与SVR对应的平均绝对比分比误差分别为12.02%和12.49%，相较于随机森林的10.54%更高；KNN与SVR的均方误差分别为1.09和1.26，高于随机森林的0.85。因此，在有一定进水数据的基础上，通过随机森林构建的出水TN预测模型的精度更高，该研究方法能为污水预测提供一种新的尝试。更多还原

【Abstract】 To accurately predict the total nitrogen(TN) concentration of effluent from a sewage treatment plant, collected and analyzed 317 actual operational monitoring data of a sewage treatment plant, upon data pretreatment, in combination with six conventional water quality indexes with high correlation such as inlet TN, three machine learning models of random forest,KNN and SVR were used to predict effluent TN, and the predicted values were compared and analyzed. The results indicate that: The determination coefficients of KNN and SVR were 0.759 and 0.723 respectively, lower than 0.814 of the random forest model. The mean absolute percentage error of KNN and SVR was 12.02% and 12.49% respectively, which was higher than that of random forest(10.54%). The mean square error of KNN and SVR is 1.09 and 1.26 respectively, which is higher than that of random forest(0.85). Therefore, the prediction model of effluent TN built by random forest has a higher accuracy on the basis of certain influent data, and this research method can provide a new attempt for sewage prediction.更多还原