基于门控注意网络模型的天然气管道泄漏检测新方法  

A new gas pipeline leak detection method based on the SGAN model

作　　者：董宏丽[1,2,3] 孙桐 王闯 杨帆[1,2,3] 商柔 DONG Hongli;SUN Tong;WANG Chuang;YANG Fan;SHANG Rou(Sanya Offshore Oil&Gas Research Institute,Northeast Petroleum University,Sanya,Hainan 572025,China;Artificial Intelligence Energy Research Institute,Northeast Petroleum University,Daqing,Heilongjiang 163318,China;Heilongjiang Provincial Key Laboratory of Networking and Intelligent Control,Daqing,Heilongjiang 163318,China)

机构地区：[1]东北石油大学三亚海洋油气研究院 [2]东北石油大学人工智能能源研究院 [3]黑龙江省网络化与智能控制重点实验室

出　　处：《天然气工业》2025年第1期25-36,共12页Natural Gas Industry

基　　金：国家自然科学基金区域创新发展联合基金项目“基于分布式算法及大数据驱动的微地震信号去噪与反演研究”(编号:U21A2019);国家自然科学基金青年科学基金项目“面向不完备数据场景的油气管网故障诊断和异常预警方法研究”(编号:62403119);国家资助博士后研究人员计划B档资助项目“面向深海油气管道典型小样本场景的知识迁移方法研究”(编号:GZB20240136);中国博士后科学基金第75批面上资助“地区专项支持计划”项目“非完备数据约束下的油气管网智能运维关键技术研究”(编号:2024MD753911)。

摘　　要：准确的泄漏检测对维护天然气管道运行安全至关重要。近年来,深度学习已成为天然气管道泄漏检测的常用方法,但由于天然气管道数据具有复杂的时间动态特性,进而导致大多数深度学习方法在识别泄漏类型方面难以取得优异的性能。此外,检测模型的初始超参数选择通常是随机的,这也可能会导致识别性能不稳定。为了提升天然气管道泄漏检测的准确性,提出一种基于麻雀搜索算法的门控注意网络模型(Sparrow Search Algorithm-based Gate Attention Network, SGAN)。首先,为了提取有效且具有鲁棒性的数据特征,采用带交叉熵函数的麻雀搜索算法对门控循环单元的初始超参数进行全局搜索;然后,设计了一种异常注意力机制,通过对数据特征进行加权来放大正常和泄漏数据之间的区分差异;最后,将所提算法应用于天然气管道的泄漏检测。研究结果表明:(1) SGAN模型能够实现模型超参数的自适应优化,并加快了模型的收敛速度,使模型性能更加稳定;(2) SGAN模型通过对正常与泄漏特征进行加权处理,显著提升了数据特征的区分效果;(3) SGAN模型的学习表示能力和泛化能力得到了明显加强,以此提高了对数据的分类性能;(4) SGAN模型能够显著提高天然气管道泄漏检测的准确率和召回率,可减少误报率和漏报率,并且其性能明显优于常规分类算法。结论认为,SGAN模型通过自适应优化和异常注意力机制结合,能精准识别泄漏特征,并快速响应天然气管道中的泄漏情况,有效提升了检测的准确性和可靠性,显著降低了安全事故风险,为天然气管道泄漏检测提供了一种高效、智能的解决新方案。Accurate leak detection is crucial for maintaining the safe operation of gas pipelines.In recent years,deep learning has become a commonly used method for the leak detection of gas pipelines.Due to the complex temporal dynamics of gas pipeline data,however,most deep learning methods can hardly achieve excellent performance in identifying leak types.In addition,the initial hyperparameter selection of the detection model is usually random,which may lead to unstable performance recognition.In order to improve the accuracy of gas pipeline leak detection,this paper puts forward a sparrow search algorithm-based gate attention network(SGAN)model.Firstly,in order to extract effective and robust data features,a sparrow search algorithm(SSA)with cross entropy function is adopted to globally search for the initial hyperparameters of the gated recurrent unit(GRU).Then,an abnormal attention mechanism(AM)is designed,which amplifies the discriminative differences between normal and leak data by weighting data features.Finally,the proposed algorithm is used for the leak detection of gas pipelines.The following results are obtained.First,the SGAN model can achieve the adaptive optimization of model hyperparameters and increase the convergence speed of the model,making its performance more stable.Second,the SGAN model significantly improves the discriminative effect of data features by weighting normal and leak features.Third,the learning representation and generalization abilities of the SGAN model are significantly enhanced,and thus its performance in data classification is improved.Fourth,the SGAN model can significantly improve the accuracy and recall rate of gas pipeline leak detection,and reduce false positive rate and false negative rate,and has significantly a better performance than conventional classification algorithms.In conclusion,combining the adaptive optimization with the abnormal attention mechanism,the SGAN model can accurately identify leak characteristics and quickly respond to leak situations in gas pipelines,

关 键 词：天然气管道 泄漏检测 麻雀搜索算法 门控循环单元 异常注意力机制 自适应优化 智能 

分 类 号：TE88[石油与天然气工程—油气储运工程]