Research On Combination Prediction Model Of Residual Chlorine In Water Distribution Network Based On Support Vector Machine And Neural Network

The residual chlorine in the urban water supply network is a non-stabilizing material which will decrease continuously in the process of net working.According to the Hygienic Standard for Drinking Water?GB5749?,while the residual chlorine reduce to 0.02mg/L?Cl O₂?,its bactericidal capacity will be weanken,causing bacterial growth,meawhile,resulting in the deterioration of water quality and seriously threatening the health of users.The accuracy prediction about the residual chlorine in the urban network is important for the water supply companies,in this way they could detect the trend of water deterioration in time and take necessary measures,adjust chlorine dosage reasonably,reduce disinfection by-products,and select reasonable secondary chlorination points.Thus,it is indispensable to study the appropriate model which can predict the residual chlorine in the network with quickly,simply and accurately.This thesis established PSO-SVR,LM-BP,ABC-Elman and RBF models by MATLAB,based on the theory of Support Vector Regression and Artificial Neural Network.Then intergrated the PSO-SVR model with the LM-BP,ABC-Elman and RBF models to establish a combined residual chlorine prediction model in the network.The residual monitoring data about the four pipe networks monitoring bases on a city water plant in Yunnan Province were used.The above single model and combined model are verified by examples and comparative analysis.The main research content and conclusions of the paper are as follows:?1?Analyzing the prediction model of residual chlorine in water supply network of Support Vector Regression and ANN.It was concluded that these two types of models do not involve deeper mechanism analysis of residual chlorine attenuation,only analyzes the changes of residual chlorine.The prediction of the sample data does not require an experimental study and analyze of the reaction mechanism of residual chlorine in the water supply pipeline,it is easy for practical applications,hwoever,the prediction accuracy and stability are inadequate.?2?SVR?Support Vector Regression?water supply pipe network residual chlorine prediction model is established based on Support Vector Machine,but the selection of penalty parameter C and kernel function parameter?in SVR model has a great influence on the prediction performance of the model,in order to make the model achieves the best predictive performance.The PSO particle swarm algorithm is used to optimize the two parameters of the SVR,and established the PSO-SVR residual chlorine prediction model.The results of living example validation shows that in the four monitoring points,the predictive mean MAPE of the model is 7.23%,R² is 0.9105,and RMSE is 0.0043.Compared with the SVR prediction model,the mean predicted MAPE?Mean Absolute Percentage Error?is only decreased.With3.29%,R²?Decision Coefficient?increased by 0.0256,and RMSE?Root Mean Square Error?decreased by 0.0011,prediction accuracy was still not ideal.?3?The LM algorithm is used to optimize the problem of BP neural network falling into local minimum,and established the LM-BP residual chlorine prediction model;the initial weights and thresholds of the Elman neural network are set at liberty so we us the ABC?Artificial Bee Colony?algorithm to solve the problen,and established the ABC-Elman residual chlorsine prediction model.Established the RBF residual chlorine prediction model.Example verification results shows that the average prediction accuracy of the three residual chlorine prediction models LM-BP,ABC-Elman,and RBF are not significantly different.The predicted average MAPE,R²,and RMSE of the four monitoring points were between 12.17%to 13.65%,0.8485 to0.8757,and 0.0058 to 0.0070,and the prediction accuracy was relatively low and the prediction results were unstable.?4?Combining the PSO-SVR model with the LM-BP,ABC-Elman and RBF models and estabilished the water supply residual predictive combination model of SVR+BP,SVR+Elman,and SVR+RBF water.In the case verification of the four monitoring points,the predicted average MAPE of residual chlorine in SVR+BP model is 3.69%,R² is 0.9566,and RMSE is 0.0020;SVR+Elman's MAPE is 2.16%,R² is 0.9664,and RMSE is 0.0016;SVR+RBF has a MAPE of 3.90%,R² of 0.9557,and RMSE of 0.0020.Compared with the four single models,namely,PSO-SVR,ABC-Elman,LM-BP,and RBF,the three predictive combinatorial models have significant improvements in mean MAPE,R²,and RMSE,indicating that the combined model has higher prediction accuracy.It can more accurately predict the residual chlorine in the water supply network.The residual chlorine prediction combined model based on Support Vector Machine and Artificial Neural Network established in this paper can provide more accurate information on the trend of water quality changes in the pipeline network,enabling water companies to identify the trend of water quality deterioration as soon as possible and take relevant measures in a timely manner;Under the premise of reducing the dosage of chlorine as much as possible and reducing the disinfection by-products,it also provides a reference for the reasonable choice of the locat the secondary chlorination position.