Water supply network leakage identification method based on deep learning

Abstract

The invention belongs to the technical field of water supply network leakage detection, and provides a water supply network leakage identification method based on deep learning in order to solve the problems that leakage characteristics are difficult to extract, parameters are sensitive and the like in a conventional noisemeter processing method, and the method comprises the steps that firstly, an audio signal is converted into a frequency spectrum thermodynamic diagram, and time-frequency domain information of an original signal is fully reserved; and secondly, a large number of existing actual samples are utilized, features are automatically extracted through the self-learning ability of the CNN model for recognition, the limitation of human experience is broken through, and compared with a conventional signal processing method, the method has more accurate leak detection ability and stronger generalization characteristics.

Description

technical field [0001] The invention relates to the technical field of water supply pipe network leakage detection, in particular to a water supply pipe network leakage identification method based on deep learning. Background technique [0002] The acoustic leak detection method for water supply pipelines is based on the acoustic emission signal or pipeline vibration signal generated by pipeline leakage to determine the occurrence of leakage and the location of the leakage point, which has been used in the industry for a long time. For example, leak detection sticks and electronic listening equipment used by water leak inspectors. In recent years, a large number of noise meters have been deployed in the water supply pipe network, and the Internet of Things technology has been used to collect and upload on-site signals to monitor the leakage of the pipe network. [0003] Whether it is artificial hearing loss or noise meter monitoring, since the acoustic signal is easily dist...

AI Technical Summary

Problems solved by technology

[0005] Although the literature [5] uses a one-dimensional convolutional neural network, the processing of the original sound vibration signal still uses the traditional method of "preprocessing the original signal through a filter and converting the time domain signal into a frequency domain signal". , therefore, the loss of time-frequency joint characteristics in the original signal will directly affect the accuracy of leakage accident diagnosis

Images