当前位置：首页 > 过刊浏览->2026年第52卷第1期
基于t-SNE和PGNN的输气管道泄漏工况识别技术
Identification technology for leakage working conditions of gas transmission pipeline based on t-SNE and PGNN
浏览(442) 下载(1)
- DOI:
- 10.3969/j.issn.1001-2206.2026.01.014
- 作者:
- 钱学峰
QIAN Xuefeng
- 作者单位:
- 中国石油天然气股份有限公司河南储气库分公司, 河南郑州 450000
Henan Gas Storage Branch of PetroChina Co., Ltd., Zhengzhou 450000, China
- 关键词:
- t-SNE算法; PGNN模型; 泄漏; 压降速率; 压缩机抽吸
t-SNE algorithm;PGNN model;leakage;pressure drop rate;compressor suction
- 摘要:
- 压降速率是识别输气管道泄漏工况的重要判别指标，为降低不同工况下管道沿线截断阀的误动作或不动作现象，利用模拟软件获取了几组对比工况的压降速率信号，分析了各工况信号的差异性和重复性，基于t分布随机邻域嵌入（t-SNE）算法对信号进行非线性降维和可视化表征，将降维数据代入物理引导神经网络（PGNN）模型，通过构建物理约束损失函数对分类结果进行修正，在现场利用放空阀模拟泄漏过程，验证模型的准确性。结果显示，t-SNE算法将压降速率信号降至三维后，数据结构变得更为紧凑，数据量由600×50个降低至600×3个；PGNN模型和BP神经网络模型的整体正确率分别为98.16%、87.33%；泄漏孔径越大，压降速率峰值越大；泄漏位置距离截断阀越远，压降速率峰值越小，PGNN模型可以捕捉到上述变化规律。研究成果可模拟管道运行现场数据并及时识别泄漏工况，避免发生更大的泄漏事故。
Pressure drop rate is a vital determination value for identifying leakage working conditions of gas transmission pipeline. To reduce the misoperation or non-operation of the shut-off valves along the gas transmission pipeline under different working conditions, the pressure drop rate signals under different groups of comparison working conditions were obtained by simulation software. The differences and repeatability of the signals were analyzed under different working conditions. The nonlinear dimensionality reduction and visual characterization of the signals were carried out based on the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm. The dimensionality reduction data was substituted into the Physical Guided Neural Network (PGNN) model. The classification results were corrected by constructing the physical constraint loss function. The leakage process was simulated on site using the vent valve to verify the accuracy of the results. The results show that after the t-SNE algorithm reduces the voltage drop rate signal to three dimensions, the data structure becomes more compact, with the data volume dropping from 600×50 to 600×3. The overall correct rate of the PGNN model and the BP neural network model is 98.16% and 87.33%, respectively. The larger leakage aperture increases the peak value of the pressure drop rate. The farther leakage position from the shut-off valve decreases the peak value of the pressure drop rate. The PGNN model can capture the variation laws above. The proposed model can identify leakage working conditions in a timely manner on site to avoid larger leakage accidents.