Operations Research 2025
Abstract Submission

2358. Improving Scalability in Flat Yard Shunting Optimization: Refining Graph-Based Exploration in the HEROS Algorithm

Invited abstract in session FA-1: Railway Optimization, stream Mobility, Transportation, and Traffic.

Friday, 8:45-10:15
Room: Audimax

Authors (first author is the speaker)

1. Felix Rauschert
Big Data Analytics in Transportation, TU Dresden
2. Pascal Kerschke
Big Data Analytics in Transportation, TU Dresden & ScaDS.AI

Abstract

Efficient operation of industrial railways and freight transport networks relies heavily on optimizing shunting processes within flat yards. Our shunting algorithm, HEROS, employs a multi-stage approach tailored specifically for this purpose.
The core of HEROS is an effective combination of exploration and exploitation of a graph-based representation of the solution space. Initially, a deterministic opening heuristic sets a preliminary path within the graph, establishing a baseline solution. Subsequently, the exploration phase incrementally extends this graph by introducing new edges and nodes, representing additional shunting operations and yard states. During the exploitation phase, HEROS focuses on discovering new paths within this expanded graph to enhance the objective value. The cycle of exploration and exploitation continues until termination criteria are met, consistently refining the best-known solution from the last exploitation phase.

Previous studies have demonstrated HEROS' strong performance, particularly in smaller yards, highlighting robustness and reliability in optimization. However, as problem size increases, the rate of improvement decreases. To address this, our current investigation focuses on refining the graph creation process and enhancing the network structure. Drawing from these insights, we aim to enhance the algorithm's exploration phase, thereby improving scalability for larger instances that involve more tracks and longer trains.

Keywords

Status: accepted

Back to the list of papers