摘要
针对含硫天然气中H2S等腐蚀性物质易导致井站设备、管线等发生泄漏,易引发火灾,但常用的火焰和烟雾检测仪器、算法易受井站复杂环境影响,且含硫井站人工巡检存在一定风险,提出一种基于深度学习目标检测模型的含硫井站火焰和烟雾检测方法。首先,将能在移动端实时检测的YOLO V4目标检测模型先对公开火焰、烟雾数据集进行训练;接着,将训练好的模型采用迁移学习方法对井站火焰、烟雾数据集进行训练,提取井站火焰、烟雾特征;最后,经迁移学习训练后的YOLO V4模型对火焰、烟雾检测的平均精度均值高达99.62%,配合巡检机器人将对含硫井站有更好的火灾预警和救援侦察能力。
Aiming at the H2S and other corrosive substances sulfur-containing natural gas can easily lead to leakage of well sta-tion equipment and pipelines,which can easily lead to fire.However,the commonly used flame and smoke detection instruments and algorithms are easily affected by the complex environment of well station,and there are certain risks in manual inspection of sour well station.It proposes a flame and smoke detection method for sour well station based on deep learning target detection model.First of all,we train the open flame and smoke data set with the YOLO V4 target detection model which can detect real-time in the mobile terminal;then,we train the well station flame and smoke data set with the migration learning method to ex-tract the well station flame and smoke features;finally,after the transfer learning the average accuracy of.YOLO V4 target de-tection model is 99.62%.YOLO V4 target detection model cooperating with the inspection robot will have better fire warning and rescue reconnaissance ability for the sulfur-containing well station.
作者
向伟
龚云洋
李华昌
XIANG Wei;GONG Yun-yang;LI Hua-chang(No.3 Gas Production Plant of Sinopec Southwest Oil&Gas Company,Sichuan Deyang 618000,China)
出处
《机械设计与制造》
北大核心
2024年第1期261-264,共4页
Machinery Design & Manufacture