Optimal Automatic Train Operation Via Deep Reinforcement Learning

The energy consumption occupies a considerable part in the total cost of the high-speed train operation. This paper focuses on minimizing the energy consumption of high-speed train by providing an optimal trajectory planning method. In this case, several other conditions including punctuality standard, comfort standard and varying speed limitation are taken into consideration in order to systematically evaluate the performance of a designed trajectory. The air resistance related to the current speed of the train and the regenerative braking system related to the braking force enhance the difficulty o ft he trajectory planning problem. In previous studies of trajectory planning, either the effort produced by the high-speed train or the distance is regarded as a discrete variable, while the modern train usually provides continuous effort. This paper will propose an algorithm to deal with the trajectory planning problem mentioned based on the Deep Deterministic Policy Gradient.