基于声音识别与温度诊断的天然气泄漏监测

Research on Intelligent Leak Detection of Western Pipeline valve chamber based on Neural Network Multi-Algorithms

阀室天然气泄漏检测存在误报率、漏报率较高、实用性和可推广性较差等问题,通过对阀室声音样本进行采集,采用时域转频域短时傅里叶技术,提取声音的梅尔倒谱系数特征,基于声音识别和大数据分析技术,构建阀室天然气泄漏声音识别的神经网络模型,同时结合大量模拟泄漏温度实验数据进行辅助诊断。对西部管道公司输气管道2处阀室展开工业性试验,试验结果表明,基于音频相似性与温度诊断的智能诊断方法,诊断准确率在97%以上,取得了较好的预期效果。

There are many problems in the detection of natural gas leakage in the valve chamber,such as high false alarm rate,high false alarm rate,poor practicability and scalability.This paper collects the sound of the valve chamber,extracts the sound characteristics,and builds a neural network model for the sound identification of natural gas leakage in the valve chamber based on the sound recognition and big data analysis technology.At the same time,it combines a large number of simulated leakage temperature experimental data to carry out auxiliary diagnosis.Through the industrial test of two valve chambers in the gas transmission pipeline of Western Pipeline Company,the test results show that the intelligent prediction method based on sound recognition and temperature diagnosis has a diagnostic accuracy of more than 97%,and has achieved good expected results.