Pipeline Safety Early Warning by Multifeature-Fusion CNN and LightGBM Analysis of Signals From Distributed Optical Fiber Sensors

• Published in: IEEE transactions on instrumentation and measurement Vol. 70; pp. 1 - 13
• Main Authors: Yang, Yiyuan, Zhang, Haifeng, Li, Yi
• Format: Journal Article
• Language: English
• Published: New York IEEE 2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Damage detection; Distributed optical fiber sensor; Energy; Gas pipelines; industrial signal processing and monitoring; Inspection; lightGBM; Machine learning; Monitoring; multifeature fusion convolutional neural network (MFCNN); Natural gas; Optical communication; Optical fibers; Optical network units; Optical sensors; Optical signal processing; Petroleum pipelines; Pipeline safety; pipeline safety early warning (PSEW); Pipelines; Real time; Real-time systems; Reflectometers; Safety; Signal processing algorithms
• DOI: 10.1109/TIM.2021.3092518
• ISSN: 0018-9456

Energy pipelines are the backbones of global energy systems. Monitoring their safety and automatically identifying and locating third-party damage events are crucial to energy supply. However, most traditional methods lack in-depth consideration of distributed fiber signals and have not been tested on real-world long-distance pipelines, making it difficult to deploy them in operating long-distance pipelines. In this study, we utilize a novel real-time machine-learning method based on phase-sensitive optical time domain reflectometer technology to monitor the safety of oil and gas pipelines. Specifically, we build a multifeature-fusion convolutional neural network and LightGBM fusion model based on two novel complementary spatiotemporal features. The method was applied to a large amount of data collected from real-world oil-gas transportation pipelines of the China National Petroleum Corporation. The proposed method could accurately locate and identify third-party damage events in real-time under conditions of strong noise and various types of system hardware, and could effectively handle signal drift in the time and space dimensions. Our methodology has been deployed at real long-distance energy pipeline sites and our work will contribute to energy pipeline safety and energy supply security. Furthermore, the proposed solution could be generalized to other fields, such as industrial inspection, measurement, and monitoring.