基于卷积神经网络的气井油管液位下方气体泄漏声波检测  

Acoustic detection of gas leakage under tubing liquid level based on convolutional neural network in gas wells

作　　者：马凡凡 樊建春[1,2] 杨云朋 丁建敏 房奕霖 MA Fanfan;FAN Jianchun;YANG Yunpeng;DING Jianmin;FANG Yilin(College of Safety and Ocean Engineering,China University of Petroleum(Beijing),Beijing 102249,China;Key Laboratory of Oil and Gas Safety and Emergency Technology,Ministry of Emergency Management,Beijing 102249,China;Research Institute of Safety and Environmental Protection Technology Co.,Ltd.,China National Petroleum Corporatio,Beijing 102206,China;College of Artificial Intelligence,China University of Petroleum(Beijing),Beijing 102249,China)

机构地区：[1]中国石油大学(北京)安全与海洋工程学院,北京102249 [2]应急管理部油气生产安全与应急技术重点实验室,北京102249 [3]中国石油集团安全环保技术研究院有限公司,北京102206 [4]中国石油大学(北京)人工智能学院,北京102249

出　　处：《应用声学》2024年第4期765-774,共10页Journal of Applied Acoustics

摘　　要：气井油管泄漏问题一直普遍存在于油气生产中。针对油管液位下方泄漏声波特性不明、识别困难的问题,建立了基于声波和卷积神经网络的气井油管液位下方泄漏检测方法。首先搭建了可视化的油管泄漏模拟试验装置,进行了典型工况下的泄漏声波检测试验,然后在时频域内分析了液位下方泄漏的声波信号,最后利用自主搭建的卷积神经网络模型用于油管泄漏检测,将声波信号经过短时傅里叶变换获取的时频图作为模型的输入进行模型参数训练。结果表明,液位下方泄漏声波均方根值和绝对均值随着泄漏流量、液位深度的增加而增大,液位下方泄漏声波时频谱明显区别于其他工况,所提模型泄漏识别准确率可达99.33%,与基于极限学习机,支持向量机的识别模型相比较,所提模型识别准确率更高,验证了所提方法的有效性。Tubing leakage in gas wells has always been a common problem in oil and gas production.Aiming at the problem that the acoustic characteristics of tubing leakage under liquid level are unknown and difficult to identify,a gas well tubing leakage detection method under liquid level based on acoustic wave and convolutional neural network(CNN)is established.Firstly,a visual tubing leakage simulation experiment device was built,and the leakage acoustic wave detection experiment under typical working conditions was carried out.Then,the leakage acoustic wave signal under the liquid level was analyzed in the time domain and frequency domain.Finally,the self-built CNN model was used for tubing leakage detection.The time-frequency diagram obtained by the short-time Fourier transform of the acoustic wave signal was used as the input of the model for model parameter training.The results show that the root mean square value and absolute mean value of leakage acoustic wave under liquid level increase with the increase of leakage flow rate and liquid level depth.The time-frequency spectrum of leakage acoustic wave under liquid level is obviously different from other working conditions.The accuracy of leakage identification of the proposed model can reach 99.33%.Compared with the recognition model based on extreme learning machine and support vector machine,the proposed model has higher recognition accuracy,which verifies the effectiveness of the proposed method.

关 键 词：气井 油管液位下方泄漏 卷积神经网络 泄漏识别 声波 

分 类 号：X924.2[环境科学与工程—安全科学]