Research On Leakage Location Technology Of Urban Water Supply Network

The pipe leakage and burst of urban water supply network is a common problem in the water supply industry both at home and abroad.The leakage of water supply pipeline not only causes the waste of water resources,but also affects the quality of life and living cost of residents.How to control and reduce the leakage of water supply pipeline,reduce the waste of water resources and ensure the normal operation of pipelines is of great significance.In recent years,with the rapid development of artificial intelligence,large data,and other new technologies,more and more new methods are applied to fault detection.Among these methods,the Support Vector Machines(SVM)shows its superior performance.SVM can maximize the hidden classification information on the basis of obtaining limited feature information.It is more suitable for the practical engineering problems of fault diagnosis.Least Squares Support Vector Machines(LSSVM)is an extension of standard SVM.It has faster solution speed and less solution resources).If the Particle Swarm Optimization(PSO)is used to optimize LSSVM parameters,it can produce better diagnosis results.This paper studies the pressure changes before and after the leakage of pipe network.The PSO-LSSVM diagnosis model of the leakage,leakage position,and pressure change at different conditions is established.The selection of learning samples in the model is further studied,and the influence of sample’s type and number on the accuracy of model is analyzed.It is concluded that increasing the learning sample density to a certain scale can improve the diagnostic accuracy of the model correspondingly.For large and complex water supply network,it is very difficult to collect learning sample data when leaking,especially for collection of multi point leaks.In order to solve this problem,this paper proposes an improved algorithm that does not require a large increase in the number of samples and still has a good positioning effect.The algorithm only needs to select one learning sample,increase the sample density and form the enhanced sample.The mean square error(MSE)of enhanced sample and the learning sample are used as the fitness function of the PSO-LSSVM algorithm.The PSO algorithm should consider both the MSE of the learning sample and enhanced sample in the optimization of the parameters,so as to improve the generalization ability of the model.This paper studies the selection method of enhanced sample,and analyzes the type and quantity of enhanced sample.The research shows that the enhanced sample can be selected arbitrarily in learning samples and it only needs to select one enhanced sample to improve the generalization ability of the model.Finally,a large complex water supply network in a city is simulated by EPANETH software,and the pressure data under different leakage conditions are collected.Simulation experiments on low density learning samples,high-density learning samples and improved algorithm are carried out.The simulation results show that the PSO-LSSVM diagnostic model constructed by low-density learning samples(algorithm 1)has an average absolute error of 11.73;the model constructed by increasing learning sample density(algorithm 2)has an average absolute error of 0.33.The average absolute error of the improved algorithm is 2.12,which is 9.61 lower than that of algorithm 1,which is only 1.79 higher than that of algorithm 2.Combined with the output number rule,the accuracy rate of algorithm 1 is only 37.70%,which cannot meet the prediction accuracy requirement.The algorithm 2 is 98.36% and the improved algorithm is correct in 100%.The improved algorithm reduces the number of 150 groups from algorithm 2 to 55 groups.It simplifies the algorithm operation,and illustrates the superiority of the improved algorithm.