Study On Leakage Identification Of Water Supply Network Based On Data-driven Method

Urban water supply network is one of the important infrastructures of the city.The leakage of the network often occurs,resulting in the loss of a large number of water resources.At present,more and more cities have established SCADA monitoring system and intelligent water affairs system for water supply pipe network.However,the utilization rate of monitoring data is not high,and there are few attempts to diagnose pipe network accidents based on monitoring data.At the same time,the in-depth mining of monitoring data is still in the initial stage of exploration.Therefore,combining big data,statistics,artificial intelligence and other methods as well as traditional flow analysis,model analysis and other methods to evaluate,predict and identify leakage is conducive to further reduce the leakage rate and improve the management level of water companies.Based on the monitoring data of water supply network,this paper studies the water consumption analysis and prediction,hydraulic model construction and leakage identification and diagnosis of water supply network by using the methods of data statistical analysis,random forest,probability density function and neural network,and puts forward a leakage identification and diagnosis method of water supply network based on the data-driven of hydraulic model.Firstly,this paper cleans the monitoring data of water supply network,analyzes the law and characteristics of users’ water use based on quantile method,and classifies users’ meters based on dynamic time warping method.Then the concept of water use level is put forward from the perspective of statistics to reflect the trend of water consumption.Finally,the water consumption is predicted based on random forest and probability density function.The results show that the difference between the predicted and the actual water consumption is small and the effect is good.In this paper,the water quantity of the water supply network is calculated,and the hydraulic model is established.On the basis of the monitoring data,the sparrow optimization algorithm is used to check the hydraulic model of the water supply network.The results show that the checked hydraulic model is more consistent with the hydraulic situation of the actual pipe network and can reflect the operation condition of the real pipe network.Based on the hydraulic model,this paper simulates the leakage and generates the training data set.BP neural network and LSTM neural network are used to train and identify the leakage identification based on the flow data series of 24 h and 36 h,including leakage event identification,leakage time identification and leakage quantity identification.The results show that it has a good identification accuracy.Then the time window is used to improve the leakage identification,which improves the efficiency and flexibility of leakage identification and diagnosis,and a leakage quantity identification method based on dynamic time warping is proposed.Finally,the actual flow data is used to test the leakage identification and diagnosis system,and two abnormal events are successfully identified.The results show that it has a certain identification accuracy and has a certain reference value for leakage identification and diagnosis based on flow detection.At the same time,the leakage identification and diagnosis system has the characteristics of convenient calculation,fast-update and high identification and diagnosis efficiency.It is suitable for the current actual water supply network.