A hybrid optimization framework for dynamic rolling-stock assignment at congested railway terminals

Abstract

Delayed arrival at busy stations results in cascading effects. Late arrivals mean that an incoming rake would not be available for reuse and the subsequent departure would be affected. The research investigates this issue for New Delhi's train station (NDLS), India's busiest rail terminus and a case study based on 58,698 train movements throughout 2023. It reveals that, as 79% of arriving express trains reached NDLS with delays (mean 53 minutes), these delays significantly transferred to departing trains (r=0.491). Of 6.86 million available buffer minutes, only 13% was actually utilized, rest were "frozen" by an inflexible static assignment method. This research formulates the rolling stock assignment problem as a constrained bipartite matching problem and develops a hybrid assignment algorithm (HAA) that integrates a cost matrix based optimization with auction-based assignment and a dynamic re-assignment process. HAA provides 98.9%, 99.9%, and 88.6% delay absorption capacities for three different levels of traffic congestion while reducing originating train delay by up to 39.8 percentage points compared to the present system. Thoughtful assignments not more capacity are the key to improving punctuality in a heavily utilized rail.