Research On Leakage Location Of Water Supply Network Based On Deep Learning Model Of Spatiotemporal Data Analysis

The leakage of water supply pipe network is one of the important problems in the water supply system.The leakage of the pipe network will not only cause the waste of water resources,affect the normal water use of the surrounding residents,but also cause the growth of bacteria in the surrounding environment,affect the water quality,and in severe cases can endanger the lives of residents.The realization of the leakage location of the water supply pipe network can help the staff of the water supply department to quickly determine the location of the pipe network leakage,gain time for subsequent pipeline repairs,and greatly improve the quality of residents’ living water.Therefore,timely and accurate determination of the leakage location of the pipeline network is of great significance for controlling the leakage of the pipeline network and preventing the waste of water resources in our country.Water supply network data includes time series data,such as water pressure,flow,etc.It also contains spatial sequence data such as pipe diameter and pipe material.Deep learning can learn the characteristics of various types of data well,and find the trend of data in different characteristics.Therefore,we choose a deep learning model to analyze the spatiotemporal data leaked by the pipeline network,and abstract the leak location problem into a multi-category problem.Different categories indicate the leakage of the pipe network at different nodes,and then the location of the pipe network leakage can be found.After the water supply network leaks,the water pressure and flow rate in the pipeline will change with time.Therefore,the pipeline network leakage data is time series data.In this paper,the Bi-LSTM neural network,which is good at analyzing time series data,is used to analyze the time series data when the pipe network is leaking,and get the location of the leaking point.First,the pipeline network online simulation platform built on the EPANET software based on the pipeline network simulation platform built by the laboratory is used to simulate leakage and obtain data on the platform,and then use Bi-LSTM for analysis.Finally,through the multi-classification operation,the location of the leakage node and the probability of leakage of each leakage node are obtained.Experimental results show that the model can quickly and accurately locate the leakage of water supply pipes,and its accuracy is improved compared with the BP neural network and the traditional LSTM model.Pipe network leakage is not only related to time series data,but also to spatial series data such as pipe diameter,pipe age and pipe roughness coefficient.Based on the temporal and spatial characteristics of the pipeline network leakage data,this paper uses the attention mechanism-based CNN-LSTM model to analyze the pipeline network data,which can realize the comprehensive capture of the pipeline network leakage information and achieve accurate leakage location.The model still uses a Bi-LSTM neural network to analyze time series data,and then uses Convolutional Neural Networks(CNN)to learn spatial series data.Since the influence of spatial attributes on the leakage of the water supply network is relatively small compared with the temporal attributes,the attention mechanism is introduced into the model to assign weights to the dimensions of data input,which can achieve more accurate leakage Positioning.Similar to the leakage location model based on Bi-LSTM,the model will also return to the specific location of the water supply pipeline network leakage and the probability of leakage at each leakage node through a multi-classification operation.The model is compared with the BP neural network and the Bi-LSTM neural network.The experimental results show that the network effectively improves the accuracy of classification and realizes the accurate location of the leakage points of the water supply network.