Research On Identification Of Leakage Aperture In Water-supply Pipeline Based On Compressed Sensing And Deep Learning

With the expansion of the scale of water-supply network in cities and towns,the water supply-pipeline often leaks.The amount of water leaked by different leakage apertures will have different effects on the destruction of urban foundations.Thus,the size of the leakage aperture of the pipeline has a direct relationship with the degree of danger caused by the leakage.Data collection under the traditional Nyquist will generate massive amounts of data,and the data information redundancy is overmuch,which is not conducive to the data storage,network communication,leak detection and identification of the water-supply pipeline monitoring system.So,it is necessary to introduce leading-edge compressed sensing(CS)theory and deep learning into the field of water-supply pipeline leakage diagnosis.CS can compress and collect data,thereby reducing the network load of the pipe network.Combined with deep learning technology,it can realize the intelligent detection and diagnosis of water-supply pipe leakage,which can provide a new direction for the research of water-supply pipeline leakage aperture identification.The main contents are as follows:(1)Aiming at the problem of nonlinear and non-stationary water-supply pipeline leakage vibration signal when compressed sensing is performed under the analytical dictionary,the reconstruction mean square error is large and important leakage information in the signal is lost,a sparse method based on variational mode decomposition and ksingular decomposition algorithm(VMD-K-SVD)is proposed.The experimental results show that the compressed sensing method based on VMD-K-SVD sparse representation to construct over complete dictionary has better reconstruction performance and sparsity,and the reconstructed signal accuracy is higher than that based on FFT,DCT and K-SVD.(2)Aiming at the problem that traditional pipeline leakage diagnosis methods cannot realize automatic and efficient leakage aperture identification in the face of massive watersupply pipeline leakage signals,this paper proposes a water-supply pipeline leakage aperture identification method based on convolutional neural network(CNN).Experiments show that this method has a high accuracy rate.meanwhile,the accuracy of the leakage aperture identification under different CNN structures is analyzed and compared.Among them,the average accuracy of the leakage aperture identification of the water-supply pipeline based on the residual neural network(ResNet)is higher,reaching 99.13%.(3)Aimed at the problems of excessive data volume and excessive information redundancy in the intelligent water-supply pipeline leakage aperture identification method under the Nyquist data acquisition mode,it is not conducive to data storage and network communication.This paper proposes a water-supply pipeline leakage aperture recognition method based on compressed sensing and ResNet.The compressed collection is carried out under the condition that the observation matrix guarantees the integrity of the signal information,and the redundant information and volume of the data are reduced.Experiments show that using a small of compressed observations can get a relatively high accuracy rate and shorter model training time;when the compression rate is 70% and the observation matrix is a gaussian random observation matrix,the average accuracy rate reaches 96.67%,and the training time is only 50% of the uncompressed acquisition;when the compression rate is 80% and the observation matrix is a partial Fourier matrix,the average accuracy of the leakage aperture identification of the water-supply pipeline is98.77%,which is close to the accuracy of the uncompressed acquisition.