Leakage detection of low-pressure gas distribution pipeline system based on linear fitting and extreme learning machine

Abstract

Pipelines are the lifeblood of the natural gas market. Compared with the high-pressure long-distance pipeline system and medium-pressure municipal pipeline system, the leak detection of low-pressure gas distribution pipeline system inside buildings that only relies on the combustible gas alarm is relatively backward and passive. Based on the principle of the negative pressure wave method, we proposed a leak detection method combining linear fitting and extreme learning machine leak detection. We designed and built a low-pressure gas pipeline system, which truly reproduces the gas leakage scenes inside buildings. By establishing multi-dimensional sliding windows, combined with a linear fitting method, the original pressure signal can be initially screened. Small leakage and pipe network fluctuations can be effectively identified by the windows. For medium leakage, large leakage, and gas usage that are easy to be confused, the windows will accurately identify the inflection point of the pressure drop for subsequent processing. Aiming at the characteristics of negative pressure waves, we extracted 15-dimensional feature vectors by dividing different stages. After introducing extreme learning machine for training, the detection accuracy of the final model can reach more than 90 %. Compared with support vector machines and back propagation neural networks, the trained extreme learning machine model has higher accuracy and higher speed. The joint leak detection algorithm based on linear fitting and extreme learning machine can greatly reduce the number of gas leakage accidents of low-pressure gas pipeline system through single point pressure monitoring. We applied the artificial intelligence algorithm and negative pressure wave to the low-pressure pipeline system for the first time, and established a new pressure wave processing method, characteristic model and algorithm model, which has a good engineering application prospect and theoretical significance.