Original Article

Development of a dynamic control and dispatching platform for natural gas pipeline emergency situations based on reinforcement learning

¹ School of Safety Science and Emergency Management, Wuhan University of Technology, Wuhan 430000, Hubei, China
² Zhejiang Business College, Hangzhou 310000, Zhejiang, China
³ The Chinese University of HongKong, Shenzhen 518100, Guangdong, PR China

^* e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

Received: 28 October 2025
Accepted: 16 March 2026

Abstract

Natural gas pipeline network emergencies are frequent. Existing monitoring and dispatching systems are mostly static and rule-driven, making them difficult to adapt to changing environments and coupled risks and lacking self-learning and adaptive optimization capabilities. To address this, this paper constructs a dynamic control system for natural gas pipeline emergencies based on reinforcement learning. First, SCADA (Supervisory Control and Data Acquisition) and sensor data fusion are used to achieve multi-source state perception and establish a temporal state space suitable for training. A double deep Q-network intelligent decision engine is then introduced to learn stable policies through experience replay and a target network. A state-action adaptive mapping mechanism is designed to achieve intelligent adjustment of valves, pressure, and flow under different emergency levels. A multi-objective reward function is constructed by combining safety, timeliness, and energy consumption to achieve dynamic system balance. Finally, a visual control platform based on Python and TensorFlow is developed to complete the closed-loop optimization from data perception to policy execution. Experiments show that the average response time of the proposed method is only 0.96 s, a significant improvement over traditional method. After training, the valve control stability index reaches 0.98, and the adjustment time is shortened to 2.1 s. Under level 6 emergency conditions, the safety retention rate still reaches 91.7%, energy consumption is reduced by 13.2%, and the average reward under complex disturbances is 75.8, verifying its high efficiency and robustness in dynamic regulation.