Smart Train Operation Algorithms Based on Expert Knowledge and Reinforcement Learning

During decades, the automatic train operation (ATO) system has been gradually adopted in many subway systems for its low-cost and intelligence. This article proposes two smart (STO) algorithms by integrating expert knowledge with reinforcement learning algorithms. Compared previous works, proposed can realize control of continuous action optimize multiple critical objectives without using an offline speed profile. First, through historical data experienced drivers, we extract rules build inference methods to guarantee riding comfort, punctuality, safety system. Then develop optimizing energy efficiency operation. One is STO algorithm based on deep deterministic policy gradient named (STOD) other normalized advantage function (STON). Finally, verify performance via some numerical simulations real field from Yizhuang Line Beijing Subway illustrate that developed are better than manual driving existing ATO terms efficiency. Moreover, STOD STON adapt different trip times resistance conditions.