Development of an Energy Management Agent for Particulate Matter Control in Railway Stations Considering Regenerative Braking Energy

This paper presents an enhanced energy management approach for underground train stations by integrating regenerative braking power into an artificial intelligence-based control framework. Air quality in stations is often degraded due to particulate matter (PM2.5 and PM10) generated from train operation, passenger flow, and limited ventilation. To improve both air quality and energy efficiency, we propose a learning-based energy management agent that combines an artificial neural network (ANN) with a Deep Q-Network (DQN). The ANN serves as a transition model to predict future dust concentration levels based on fan and HVAC operation as well as available renewable energy. The DQN then determines optimal control actions for fans, air conditioning units, and an energy storage system (ESS) to minimize energy cost while maintaining air quality. In this study, regenerative braking power was quantified using real train timetables and incorporated as a state variable, enabling the agent to utilize surplus energy effectively. A case study using data from Namgwangju Station demonstrates that the proposed method achieves faster dust concentration reduction, more stable ESS operation, and significant reductions in grid power consumption and energy cost compared with conventional strategies.