Wet Gas Pipeline Maintenance Process Using Reinforcement Learning

Abstract

Oil and gas extraction is one of the essential businesses globally, since petroleum and natural gases, including wet gas, a liquid-based natural gas are necessary for people's lives. Wet gas pipeline systems often face internal corrosion problems leading to gas leakage, environmental pollution, and human fatalities. In addition, the wet gas pipeline system is usually installed underground or under the sea, making it difficult to maintain and resulting in high costs. In this project, the pipeline maintenance scheduling system has been developed by applying Reinforcement Learning, a type of Machine Learning widely and efficiently used in condition-based maintenance problems with a stochastic environment. A combination of Q-learning technique and epsilon-greedy policy had been utilized as the algorithm for the learning process. According to the results, the pipeline maintenance scheduling process from our developed system could prevent leakage and rupture during the experimental period, which was 40 years. It had significantly reduced the cost of periodic maintenance process, from 19,455.28 USD to 8,463.60 USD per month. Furthermore, our pipeline maintenance schedule system can be developed to a greater extent, to be more ecologically friendly with environmental impact in mind.