Research On Leak Diagnosis Method Of Urban Water Supply Pipeline Based On Pressure Signal Identification

With the continuous improvement of urbanization in China,the length of the water supply pipeline in our country shows an accelerating trend.The water delivered in the water supply network is all treated by local water plants which quality is in accordance with the nation's standard.Leakage of the pipe network will not only causes the waste of water resources,resulting in a drop in the water supply capacity of the municipal system or even a partial collapse,but also a huge loss to the local economy and serious environmental damage.Negative pressure is a signal that is very effective for feature extraction of liquid delivery piping conditions,but some conventional operations will generate negative pressure signal which causes false alarms.For the leakage diagnosis of urban water supply pipelines,how to identify the leakage from the complex interference is the key difficulty of the entire technology.Water supply pipeline leak diagnosis is essentially a kind of fault diagnosis.The entire process of diagnosis is divided into: noise reduction of the original acquisition signal,selection the characteristics of original signal after noise reduction and system condition identification based on selected features.Because negative pressure wave shows a high degree of relevance to the liquid transport pipeline,it is used as original signal in this thesis.To make the solution according to the above steps respectively,the main contents of this article are as follows:1)The main decomposition principle of Variational Mode Decomposition(VMD)is introduced in detail.A method of sample entropy based on the Sample Entropy criterion to optimize the decomposed IMF to reconstruct signal is proposed.Through the decomposition and reconstruction of the negative pressure wave signal of the water supply network with this method,filter by sample entropy of decomposed IMF base on the Shannon entropy criterion for signal reconstruction can be effectively applied to the diagnosis of pipeline leakage.2)The principle of Random Forest(RF)is introduced in detail.Twelve common characteristics of the negative pressure wave signal of the water supply network that can reflect the change of signal waveform are input as the total feature set.After the RF iterations,the features are ranked for the accuracy of the classification,the first five of them as the preferred features are selected for the subset for classification.The signal obtained by the previous VMD decomposition for noise reduction is regarded as the original signal in this step.After extracted feature sets is input,an optimal feature subset that can reflect the negative pressure wave signal of the water supply pipeline after multiple iterations of the RF algorithm is obtained which include kurtosis,kurtosis factor,mean amplitude,root mean square and margin factor.3)The principle of Support Vector Machine(SVM)is introduced in detail.Particle swarm optimization algorithm is proposed to optimize the kernel parameters g and the penalty factor c in radial basis function kernel function in SVM.At least,the pluralistic improved SVM based on one-to-many rule is established which can identify water supply pipeline conditions.Experimental results shows,PSO algorithm can quickly find the optimal value of key parameters in SVM.Compared with the traditional back propagation algorithm and the support vector machine algorithm based on genetic algorithm and ant colony algorithm,the improved algorithm has improved the training time and accuracy.