Gas pipeline leakage recognition method based on deep learning in strong background noise environment

Abstract

The invention provides a gas pipeline leakage recognition method based on deep learning in a strong background noise environment. Firstly, valve leakage and gasket leakage are simulated on a laboratory pipeline leakage simulation system and a leakage sound signal is acquired, meanwhile, a strong background (fan) noise signal is acquired in an actual pipe rack, and the leakage signal and the actualbackground signal are mixed to simulate the leakage condition in the actual pipe rack. Each each type of audio data is segmented to obtain a large number of short-term audio signals, short-time Fourier transform is performed on the audio signals, then feature enhancement processing is performed, and an enhanced feature matrix is mapped into a time-frequency map to form a dataset for underground comprehensive pipeline rack leakage detection. Then, a convolutional neural network model suitable for gas leakage detection is constructed. The feature enhancement and neural network are combined to constitute a complete and effective gas pipeline leakage detection scheme. The gas pipeline leakage recognition method can effectively improve the leakage recognition accuracy under the interference ofstrong background noise.

Description

technical field [0001] The invention belongs to the field of pipeline leakage detection, and in particular relates to a method for identifying gas pipeline leakage in a strong background noise environment based on deep learning. Background technique [0002] Pipeline transportation, with its characteristics of low cost, fast transportation, high efficiency and high safety, has become the most important mode of gas transportation at present, especially in the fields of oil and gas and chemical industry. In the underground space of densely populated modern cities, there are a large number of high-pressure gas pipelines such as natural gas and heating. These pipelines will have loose joints, corrosion of pipe walls, or man-made damage during long-term use. If the leakage point is not detected in time and corresponding maintenance is carried out, the gas in the pipeline will continue to leak, and it is very easy to reach the explosion limit range in a closed space, and serious s...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem of signal recognition under the condition of strong background noise, the present invention proposes a new method based on the combination of time-spectrogram feature enhancement and convolutional neural network for the leakage recognition of gas pipelines

Images