Machine Learning Optimization of Target Velocity Curves in Hybrid Electric Trains

This study tackles the challenge of refining the target velocity curves for hybrid electric trains, governed primarily by onboard Automatic Train Operation (ATO) systems. These systems take into account various factors, such as the interstation line conditions and the specific traction and braking characteristics of hybrid trains. Traditional approaches, which rely on fixed speed–position sequences to navigate trains and ensure safety through the Automatic Train Protection (ATP) system, struggle to adapt to dynamic environmental changes, leading to compromised operational efficiency. In response, our research adopts a machine learning framework, with a particular emphasis on reinforcement learning, to devise a real-time, flexible optimization model for determining the train’s target velocity curve. This model harnesses the potential of the double-depth Q network to enhance the optimization process. The primary objective is to improve the punctuality and energy efficiency of train operations while simultaneously increasing passenger comfort through better adaptation to environmental variations. Simulation results demonstrate that the newly optimized target velocity curve notably diminishes the on-time errors for hybrid trains and achieves approximately 0.98% in energy savings compared to traditional heuristic algorithms. These outcomes highlight the significant advantages of integrating sophisticated machine learning techniques like double-depth Q network to boost the efficiency and sustainability of hybrid electric train operations.