Research On Leakage Location Of Water Supply Pipe Network Based On Deep Learning

In an era of rapid population growth,per capita water consumption has increased sharply,and the shortage of water resources cannot be underestimated.Pipe network leakage is one of the main reasons among the various factors which cause water waste;at the same time,leaky pipe network leakage seriously affects people's quality of life.How to effectively control the leakage and find and locate the leak location in time,are of great significance to reduce the waste of water resources and ensure the normal use of water for people's lives.With the development of big data technology,machine learning and deep learning technology,new technologies and methods are applied to the field of water supply pipe network,back propagation neural network and support vector machine are the first.According to in-depth analysis of all aspects of water supply pipe network influencing factors,this paper built water supply pipe network leakage damage location models under the different circumstances.This paper focused on the following work:(1)There are two kinds of hydraulic monitoring points in the water supply pipe network: pressure monitoring points and flow monitoring points.Among them,the installation and maintenance cost of the flow monitoring points is high,in the large-scale laying period of the previous pipe network,sensors and other technologies were not widely used.There are areas where the pipe network traffic data is not easy to obtain.In view of this situation,the pipe network leakage location model based on back propagation neural network was designed and implemented.Firstly,the application environment of the water supply pipe network and the pipe network structure were described in this paper.Further,this paper exported the pipe network data from the water supply pipe network simulation test platform.The pipe network data includes five monitoring point pressure values,pipe diameter,pipe material,pipe length,pipe resistance coefficient and other data.The data was used as input,and the leakage point was used as the output to establish the back propagation neural network model.However,we finded that the back propagation neural network was easy to fall into the local optimal characteristics by the analysis of the weight and threshold in the construction.This paper used the genetic algorithm according to the principle of "survival of the fittest" to optimize the weights and thresholds of the back propagation neural network.The comparative analysis experiments were carried out under differentinitial population numbers and crossover probabilities,positioning accuracy was improved compared to traditional back propagation networks.(2)With the development of Internet of Things technology,the collection and collection methods and channels of pipe network traffic data have gradually developed.According to the improved back propagation neural network model by genetic algorithm,which was not considered the important influence factors which was pipe network traffic,and the characteristics of time series data of pipe network traffic cannot be utilized.At the same time,there is no concept of timing in back propagation network.Based on this,this paper used a two-channel deep learning based on pipe network leakage location model.Firstly,the ADF root test was used to test data.According to the t statistic and p value in the test result,the nonlinear and non-stationary characteristics of the pipe network flow data were obtained.And the nonstationarity of the pipe network flow data was proved.The ensemble empirical modal decomposition algorithm was used to decompose and smooth the flow data,and the intrinsic mode function and the residual term obtained by the decomposition were entered into the deep learning models of the long short term memory network for feature extraction.At the same time,back propagation neural network with non-time series data such as monitoring point pressure value,pipe network diameter and tube age as input were used as the dual channel,and the characteristic analysis is respectively carried out.The extracted features were connected to jointly locate the network leakage.This paper conducted a comparative analysis experiment.The positioning accuracy of the two-channel deep learning based on pipe network leakage location model was improved compared with back propagation networks which was optimized by genetic algorithm.In this paper,the pipe network leakage model based on back propagation neural network improved by genetic algorithm and the leakage location model based on deep learning in water supply network are suitable for pipe network leakage in different situations.The data of the model is the data derived from the pipe network simulation test platform.The real and effective experiments carried out under the derived data are of great significance for the safe operation of the water supply network,laying a good foundation for the next step of timely taking leakage measures.