Machine learning-driven optimal control method for subway station VAC systems

Ventilation and air conditioning (VAC) systems in subway stations play a crucial role in ensuring passenger comfort and operational efficiency. Conventional control strategies, such as fixed-schedule operation and PID feedback control, often fail to achieve optimal energy efficiency due to lack of predictive capabilities. While model predictive control (MPC) offers improved foresight, it requires accurate system modeling and high computational resources, making it challenging for real-time applications. Reinforcement learning (RL), particularly Deep Q-Networks (DQN), has demonstrated promising results in multi-step decision-making but traditionally demands extensive training data and learning time.

This study proposes an intelligent VAC control strategy based on deep reinforcement learning, aiming to optimize energy consumption while maintaining indoor comfort. A comprehensive environmental model of a subway station was developed using historical operational data and deep learning techniques, providing a high-fidelity virtual environment for training the RL agent. The DQN-based control strategy was trained to dynamically adjust VAC operations, learning optimal control policies that minimize energy use and operational costs.

Simulation results indicate that the proposed DQN-based control strategy achieves a 30% reduction in energy consumption compared to conventional PID control, demonstrating the effectiveness of reinforcement learning in complex, dynamic environments. These findings highlight the potential of integrating deep learning-based environmental modeling with reinforcement learning to develop data-driven, adaptive control systems for subway station VAC management.

This study contributes to the advancement of AI-driven energy optimization strategies, paving the way for more intelligent, sustainable, and energy-efficient building operations in underground transit infrastructure.