Fei Xiao
Senior Engineer
Southwest Oil & Gas Field Company, PetroChina
Dr. Fei Xiao was awarded a doctorate in oil-gas storage and transportation at Southwest Petroleum University in 2020. Subsequently, he worked as a post PhD in PetroChina. He is now working as a senior engineer of pipeline management in the Gas Transmission & Management department. His research interests include risk assessment and control. Presently, Fei Xiao is focusing on the application of AI in the risk management of pipeline to achieve objectives of adaptive optimization and automation.
Participates in
TECHNICAL PROGRAMME | Energy Infrastructure
Pipelines, Storage and SPRs
Forum 08 | Digital Poster Plaza 2
28
April
12:30
14:30
UTC+3
Pipeline safety serves as a foundational element in ensuring the operational integrity and security of energy infrastructure. Due to the factors, such as geological instability, material fatigue, and unauthorized third-party encroachment, persistent challenges continue to impede real-time risk detection and coordinated emergency response in pipeline systems. Operators now leverage advanced monitoring technologies such as distributed fiber optic sensing (DFOS), unmanned aerial systems (UAS), and computer vision-enabled cameras (CVCs) to achieve holistic perimeter threat assessment. These solutions have driven a 94.4% decrease in pipeline failure rates, with incident frequency dropping from 1.78 (2015) to 0.1 (2024) per 1,000 km-year.
Despite these technological improvements, significant issues persist. Interoperability constraints between monitoring systems have resulted in disparate data ecosystems that hinder unified risk management. Specifically, alarms generated through DFOS lack automated correlation with UAS and CVC data streams, requiring human operators to manually reconcile multi-system verification. Furthermore, post-detection processes dependent on manual data aggregation and decision-making introduces substantial response delays during emergency operations, potentially unsatisfying the requirement of 5-minute emergency action prescribed in API RP 1174:2020 for pipeline leak containment.
The past decade has seen unprecedented advances in artificial intelligence, marked by developments in large language models (LLMs) such as ChatGPT and DeepSeek. These innovations have fundamentally redefined operational paradigms in big data analysis. To address issues mentioned above, a DeepSeek-based intelligent agent was developed for pipeline risk management, enabling intelligent multi-system coordination and decision support across pipeline safety operations. This pioneering architecture positions the AI agent as a cognitive command hub, achieving seamless external system interoperability through industry-standard API protocols while enabling bidirectional data/control flows.
The system architecture further incorporates a AI-powered alarm risk verification engine, which integrates multi-source alert feeds with rule sets and historical incident repositories. This risk verification module autonomously executes various system tasks, such as drone takeoff or camera rotation, and generates priority rankings.
The framework also culminates in an intelligent emergency response knowledge base that synthesizes real-time risk analysis with operational metadata, including pipeline specifications and High-Consequence Area (HCA) geospatial parameters. This cognitive agent autonomously executes: risk stratification through machine learning classifiers, resource allocation optimization via constraint-based algorithms, activation of emergency plans, and automated notification workflows—delivering decision support throughout the emergency management lifecycle.
This technological innovation introduces pipeline risk management into a framework that comprehensively addresses the full operational lifecycle from risk control to emergency response. By implementing intelligent cross-system data integration and leveraging DeepSeek's advanced analytical capabilities, the solution achieves significant operational efficiencies - reducing manual intervention while accelerating risk verification and response times. Operational data demonstrates the system's proven effectiveness, currently delivering annualized cost savings exceeding RMB 3.6 million through optimized resource allocation and incident prevention measures.
Co-author/s:
Zixiao Chen, Senior Engineer, PetroChina.
Liwen Tan, Senior Engineer, PetroChina.
Jingdong Chen, Senior Engineer, PetroChina.
Despite these technological improvements, significant issues persist. Interoperability constraints between monitoring systems have resulted in disparate data ecosystems that hinder unified risk management. Specifically, alarms generated through DFOS lack automated correlation with UAS and CVC data streams, requiring human operators to manually reconcile multi-system verification. Furthermore, post-detection processes dependent on manual data aggregation and decision-making introduces substantial response delays during emergency operations, potentially unsatisfying the requirement of 5-minute emergency action prescribed in API RP 1174:2020 for pipeline leak containment.
The past decade has seen unprecedented advances in artificial intelligence, marked by developments in large language models (LLMs) such as ChatGPT and DeepSeek. These innovations have fundamentally redefined operational paradigms in big data analysis. To address issues mentioned above, a DeepSeek-based intelligent agent was developed for pipeline risk management, enabling intelligent multi-system coordination and decision support across pipeline safety operations. This pioneering architecture positions the AI agent as a cognitive command hub, achieving seamless external system interoperability through industry-standard API protocols while enabling bidirectional data/control flows.
The system architecture further incorporates a AI-powered alarm risk verification engine, which integrates multi-source alert feeds with rule sets and historical incident repositories. This risk verification module autonomously executes various system tasks, such as drone takeoff or camera rotation, and generates priority rankings.
The framework also culminates in an intelligent emergency response knowledge base that synthesizes real-time risk analysis with operational metadata, including pipeline specifications and High-Consequence Area (HCA) geospatial parameters. This cognitive agent autonomously executes: risk stratification through machine learning classifiers, resource allocation optimization via constraint-based algorithms, activation of emergency plans, and automated notification workflows—delivering decision support throughout the emergency management lifecycle.
This technological innovation introduces pipeline risk management into a framework that comprehensively addresses the full operational lifecycle from risk control to emergency response. By implementing intelligent cross-system data integration and leveraging DeepSeek's advanced analytical capabilities, the solution achieves significant operational efficiencies - reducing manual intervention while accelerating risk verification and response times. Operational data demonstrates the system's proven effectiveness, currently delivering annualized cost savings exceeding RMB 3.6 million through optimized resource allocation and incident prevention measures.
Co-author/s:
Zixiao Chen, Senior Engineer, PetroChina.
Liwen Tan, Senior Engineer, PetroChina.
Jingdong Chen, Senior Engineer, PetroChina.
