Abstract: Domestic train control has the characteristics of multi-objective, large delay, uncertainty and so on. With the wisdom development of urban rail transit, these problems are increasingly prominent. There is an urgent need for self-learning algorithm to adapt to the changing characteristics of the circuit and reduce the cost of manual debugging. The self-learning algorithm can identify the environmental changes and improve the algorithm automatically, which is suitable for the current development stage of rail transit. This paper summarizes the new challenges of the existing algorithms of automatic train operation system, introduces the key technologies and indicators, current situation, problems and shortcomings of the existing algorithms , and then analyzes the application direction of self-learning algorithm in the automatic train operation system,  finally prospects the self-learning algorithm to help the construction of intelligent transportation.

Key words: self-learning, automatic train operation, train control

摘要： 国内列车控制存在多目标、大延时、不确定性等特点，城市轨道交通面向智慧化发展，这些问题日益凸显，迫切需要具备自学习能力的算法来适应多变的线路特点，减少人工调试成本。自学习算法能够主动识别环境变化并自动完善，适用于当前的轨道交通发展阶段。对自动驾驶系统既有算法面临的新挑战进行总结，介绍自动驾驶系统控制算法的关键技术与指标、既有算法的现状和不足，然后分析自学习算法在列车自动驾驶系统中的应用方向，最后对自学习算法助力智慧交通建设进行展望。

关键词: 自学习, 车辆自动驾驶, 列车控制