Behavior Planning of Trains with Trajectory Predictioin

Due to the development of high-speed railway, Virtual Coupling has been paid more attention to shorten the tracking distance so as to enhance the transportation capacity. Trajecory prediction is critical to Virtual Coupling when considering more dynamic train ahead behavior. This paper presents a trajectory basis method for trajectory prediction of trains based on greedy algorithm and neural network and combines it with MPC controller to plan the safe motion of trains. The main idea is to use a trajectory basis to represent all possible trajectories (known as set covering problem), which can deal with high dimension of the output in traditional trajectory prediction methods. An trajectory prediction of trains with basis algorithm has been proposed, in which greedy algorithm is used to find the trajectory basis, and the affordance vectors are defined to describe the scenario, whereas neural network is applied for classifying and identifying with affordance as input to obtain the predicted trajectory of train ahead. MPC controller is designed to solve an optimal problem by leveraging the predicted trajectory to plan the motion of controlled vehicle. We extracted more than 20000 trajectory segments from historical train state dataset and obtained an trajectory basis consisting of 4 base trajectories. The result of neural network for trajectory identification shows that the prediction model has low loss and faster running speed with small-cardinality output. The output of MPC controller proves that the proposed method is available to generate the safe and efficient control of trains.