Interactive reinforcement learning innovation to reduce carbon emissions in railway infrastructure maintenance

• Published in: Developments in the built environment Vol. 15; p. 100193
• Main Authors: Sresakoolchai, Jessada, Kaewunruen, Sakdirat
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.10.2023; Elsevier
• Subjects: Carbon emission; Environmental impact; Maintenance; Railway defects; Railway system; Reinforcement learning; Railway defects; Carbon emission; Railway system; Maintenance; Environmental impact; Reinforcement learning
• ISSN: 2666-1659; 2666-1659
• DOI: 10.1016/j.dibe.2023.100193

Carbon emission is one of the primary contributors to global warming. The global community is paying great attention to this negative impact. The goal of this study is to reduce the negative impact of railway maintenance by applying reinforcement learning (RL) by optimizing maintenance activities. Railway maintenance is a complex process that may not be efficient in terms of environmental aspect. This study is the world's first to use the potential of RL to reduce carbon emission from railway maintenance. The data used to create the RL model are gathered from the field data between 2016–019. The study section is 30 km long. Proximal Policy Optimization (PPO) is applied in the study to develop the RL model. The results demonstrate that using RL reduces carbon emission from railway maintenance by 48%, which generates a considerable amount of carbon emission reduction and reduces railway defects by 68%, which also improves maintenance efficiency significantly. •The first reinforcement learning model for railway carbon emissions reduction.•Field data robustly enables the creation of customized environments for the model.•Environment's states are obtained from defective track geometry and track component.•A complex combination of maintenance activities is adopted as an action space.•Reduce the defect and carbon emissions using an interactive and dynamic approach.