Multi-Level Scenario-based Model Predictive Control for Optimal Maintenance Planning of Railway Networks

Zhou Su, TU Delft

Abstract:  We develop a multi-level decision making approach for the optimal condition-based maintenance planning of a railway network divided into multiple sections with independent deterioration dynamics. The high-level Model Predictive Control (MPC) controller determines the section-wise intervention plan at each time step, while the low-level Capacitated Arc Routing Problem (CARP) produces the optimal schedule for the suggested interventions. We consider parameter uncertainty in the deterioration model, and a local chance constraint is imposed to each section to ensure that the condition of each section stays below the maintenance threshold with a probabilistic guarantee. A tractable scenario-based approach is adopted to approximate each chance constraint with a set of deterministic constraints. The resulting large Mixed Integer Linear Programming (MILP) problem is solved using Dantzig-Wolfe decomposition. The proposed approach is illustrated by a numerical case study on the optimal planning of tamping and ballast replacement of a regional railway network.