Learning-based Model Predictive Control for Passenger-Oriented Train Rescheduling with Flexible Train Composition

This paper focuses on passenger-oriented real-time train rescheduling, considering flexible train composition and rolling stock circulation, by integrating learning-based and optimization-based approaches. A learning-based model predictive control (MPC) approach is developed for real-time train rescheduling with flexible train composition and rolling stock circulation to address time-varying passenger demands. In the proposed approach, first, the values of the integer variables are obtained by pre-trained long short-term memory (LSTM) networks; next, they are fixed and the values of continuous variables are determined via nonlinear constrained optimization. The learning-based MPC approach enables us to jointly consider efficiency and constraint satisfaction by combining learning-based and optimization-based approaches. In order to reduce the number of integer variables, four presolve techniques are developed to prune a subset of integer decision variables. Numerical simulations based on real-life data from the Beijing urban rail transit system are conducted to illustrate the effectiveness of the developed learning-based MPC approach.
Comment: 14 pages, 14 figures, submitted to IEEE Transactions on Intelligent Transportation Systems