Real-time Planning of Target Speed for Automatic Train Operation Based on Robust Optimization

To address the issue of passenger load uncertainty affecting the normal operation of urban rail trains, this paper proposes a real-time target speed planning algorithm for automatic train operation based on robust optimization. First, a reinforcement learning model of train operation is constructed, with reward design taking into account the robust optimization of performance evaluation indices and manipulation strategies. Second, reward shaping techniques are used to improve the convergence performance of model training. The value function is estimated through a Deep Q-Network (DQN), enabling real-time response to changes in train load. Finally, simulation cases verify the effectiveness of the algorithm. Results show that the proposed algorithm can achieve real-time calculation of the target speed, and the generated speed profiles have better robustness and lower energy consumption compared to the traditional DQN algorithm.