Optimizing Drainage Network Monitoring And Pollution Tracing Using Machine Learning And Markov Chain Monte Carlo

The rapid industrialisation and urbanisation in China have led to an increase in the generation of domestic and industrial wastewater.The surreptitious discharge,leakage,and over-discharge of untreated industrial wastewater into the drainage network exacerbate the workload of wastewater treatment plants,decrease their efficiency,and negatively impact the urban water cycle and entire urban water ecosystem,as well as the environment.Therefore,improving the daily monitoring of drainage systems is necessary to enhance rapid supervision and monitoring response to pollution events in the drainage network and to precisely control the sources of pollution.This study aims to provide strong scientific and technological support to water management authorities.Specifically,the research object is the drainage network in Pinghu Street,Longgang District,Shenzhen,and the work involves drainage network monitoring selection and pollution tracing using machine learning(ML)and Markov Chain Monte Carlo(MCMC)to improve the emergency response to drainage network pollution events and post-pollution assessment work.The main research elements and findings are as follows:(1)Optimal sensor selection based on aggregated hierarchical clustering and analysis of variance.A sensor optimization selection method is proposed based on aggregated hierarchical clustering and analysis of variance(ANOVA)to analyze water quality data obtained from a Python programming-driven stormwater management model(SWMM)using a clustering algorithm.Appropriate distance judgment and distance metrics are selected to cluster the node water quality data,which is then divided into clusters to serve as the basis for selecting the number of sensors.ANOVA and multiple comparison analysis are used to select outlier points within each cluster that can acutely reflect changes in water quality for a faster response to pollution events.In the simulation example,a total of eight monitoring nodes were arranged with a contour coefficient of 0.58 and a reliability of 88.36%,which can meet the needs of monitoring and responding to drainage pollution events.(2)Convolutional neural network(CNN)-based pollution source localization model.A CNN-based pollution source localization model is proposed to extract the data features of node pollution and pollution sources using the CNN model,transforming the pollution localization problem into a classification problem.When new data from monitoring nodes are input,the trained CNN network can output the probability that each node is a potential pollution source,thus enabling fast localization of pollution sources.The training results of the dataset show that the proposed model is able to achieve fast and accurate pollution source localization even in the case of partial sampling time and lack of certain sensor data.(3)Combined convolutional neural network and Markov Monte Carlo algorithm for pollution information inversion.A combined CNN and Markov Monte Carlo(MCMC)algorithm for pollution information inversion is proposed.The source probabilities generated by the CNN are used as the prior data for the Bayesian method to improve the efficiency and accuracy of traceability.The MCMC method performs statistical sampling inversion based on the prior data and the downstream observations of the pipe network to derive the posterior probability distribution describing the estimated values of the parameters to be traced.The standard M-H sampling method is improved by incorporating an adaptive mechanism in the initial sampling and making a judgment on whether subsequent sampled values satisfy the conditions of the posterior distribution,improving the efficiency of MCMC sampling.The case application results show that the proposed CNN-MCMC method exhibits better adaptability and effectiveness under both nodal instantaneous and nodal continuous pollution discharge methods.The method proposed in this study provides a scientific basis for water management authorities to improve the daily monitoring of urban drainage systems and enhance the rapid supervision and monitoring response to pollution events in drainage networks.