Research on Train Dynamic Coupling Strategy Based on Distributed Model Predictive Control

Abstract

This paper mainly studies the problem of dynamic coupling under multiple optimization objectives and multiple constraints. Virtual coupling (VC), as a new train operation mode, effectively improves the operating efficiency of the line, but reduces the safety margin between trains, and requires higher control accuracy. As an important scenario of virtual coupling, dynamic formation has not formed a set of applicable train control strategies. Aiming at the dynamic train formation scenario of virtual coupling, propose a control architecture based on distributed model predictive control (DMPC). Numerical simulation using Matlab verifies the effectiveness of the algorithm in this paper. The simulation results show that the train is under the condition of satisfying the operating constraints. Compared with PID control, the method in this paper is more effective and more accurate.

Railway Virtual Coupling: A Survey of Emerging Control Techniques

“…Y MPC [34] 3rd MPC [55] 2nd Y Y Y MPC [56] 2nd MPC [57] 2nd Y Y MPC [39] 2nd Y ML [37] 2nd Y Y ML [32,33] 2nd Y Y Y CBC [58][59][60] 2nd CBC [64] 2nd Y CBC [61] 3rd CBC [36] 3rd Y Y CBC [62,63] 2nd Y Y Y CFC…”

“…In a VC system, the design of the minimum gap reference needs to consider various factors such as: train location errors, speed errors, actuator lag, information latency (delay), train speeds, and train accelerations. Researchers in [37,57] introduced a method to determine a minimum reference gap by adapting the approach specified in the IEEE 1474.1 as shown in Fig. 3.…”

“…Felez et al [43] applied a constraint for maximum power (the product of speed and force) of the system. Yang and Yan [57] used a maximum jerk constraint for the consideration of ride comfort. Researchers in [32,50,56] considered a stability constraint which guaranteed the actual distance between any two trains asymptotically approaches the reference; and the relative speed error between any two trains asymptotically approaches zero:…”

“…In [34], an estimator was used to update control law and to minimize the differences between predicted train states and actual train states. Researchers in [44,49,57] used the gap error element and the actuator force/power variation element without the force/power element. The variation term not only served a similar role as the force/power element but also helped to minimize the jerk, i.e., the variations of forces/accelerations.…”

“…8) with the increases of prediction steps. Most of the MPC applications for VC studies used no more than 10 prediction steps: Su et al [51] used 10 prediction steps; researchers in [52,54,55,57,65] used 5 prediction steps; researchers in [34,50,56] used 4 and 3 prediction steps respectively.…”