Accurate and smooth parking is an essential goal for automatic driving braking control of maglev trains.The strong coupling of the electro-hydraulic hybrid braking state affects the medium and low-speed maglev trains during the stopping braking process,and the traditional braking control method based on the theoretical model of braking features makes it difficult to guarantee the parking accuracy and comfort of the maglev train.This paper proposed a reinforcement learning braking control method for maglev trains based on self-learning of hybrid braking features.First,a long short-term memory(LSTM)network was used to establish a hybrid braking feature model for maglev trains,and the self-learning of dynamic braking features was performed based on the operating environment and status data of maglev trains.Then,the reward function and learning strategy of reinforcement learning were updated according to the learning results of dynamic features,and a train braking optimization control method based on deep reinforcement learning was proposed.Finally,simulation experiments were carried out by using on-site operation data of medium and low-speed maglev trains.The experimental results show that the braking control method proposed in this paper improves comfort and parking accuracy by 41.18％and 22％,respectively,compared with the traditional method.It proves the effectiveness of the modeling and braking optimization control method in this paper.
关键词
磁浮列车/制动控制/多目标优化控制/强化学习/制动特征自学习
Key words
maglev train/braking control/multi-objective optimal control/reinforcement learning/self-learning of braking features
维普
万方数据