A systematic method for the optimization of gas supply reliability in natural gas pipeline network based on Bayesian networks and deep reinforcement learning. (September 2022)

• Record Type: Journal Article
• Authors: Fan, Lin; Su, Huai; Wang, Wei; Zio, Enrico; Zhang, Li; Yang, Zhaoming; Peng, Shiliang; Yu, Weichao; Zuo, Lili; Zhang, Jinjun
• Is Part Of: Reliability engineering & system safety. Volume 225(2022)
• Publication Date: 2022-09
• Subjects: Natural gas pipeline network -- Gas supply reliability -- Preventive maintenance -- Bayesian network -- Reinforcement learning; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2022.108613
• Languages: English
• ISSNs: 0951-8320

Abstract

Highlights: Bayesian network approach to optimize the reliability of gas supply is proposed. Probabilistic safety analysis and prognostic health management are integrated. Deep reinforcement learning is proposed for decision-making. Shortage risk eliminations and operational efficiency improvements are achieved. Abstract: This study proposes a method based on Bayesian networks (BNs) to optimize the reliability of gas supply in natural gas pipeline networks. The method integrates probabilistic safety analysis with preventive maintenance to achieve the targets of minimizing gas shortage risk and reducing maintenance costs. For this, the tasks of unit failure probability calculation, system maximum supply capacity analysis, gas supply reliability assessment and system maintenance planning are performed. A stochastic capacity network model is coupled with a Markov model and graph theory to generate the state space of the pipeline network system. BN, is then, proposed as the modeling framework to describe the stochastic behavior of unit failures and customer gas shortage. The system maintenance problem is converted into a Markov decision process (MDP), and solved by using deep reinforcement learning (DRL). The effectiveness of the proposed method is validated on a case study of a European gas pipeline network. The results show that the proposed method outperforms others in identifying optimal maintenance strategies. The DRL-optimized maintenance strategy is capable of responding to a dynamic environment through continuous online learning, considering the randomness of the unit failures and the uncertainty in gas demand profiles.