Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optimizing train dispatching for the Union Pacific Railroad

Maurizio Boccia, Veronica Dal Sasso, Leonardo Lamorgese, Carlo Mannino, Paolo Ventura

TL;DR

The paper tackles automated real-time train dispatching on Union Pacific's large network, framing it as a resource-constrained planning problem on block sections. It introduces AMP as the planning engine behind an in-house A-TMS, modeling dispatching with a dispatching graph $G=(N,A)$ consisting of routing, timetable, and conflict arcs $A^R$, $A^Q$, and $A^C$, and solving for a consistent subset $y$ that yields schedule $t(y)$ when $G(y)$ contains no positive cycles. Key contributions include the continuous rolling horizon Branch-and-Bound (CRH-B&B) planner, Safeplacer and Overlength modules for deadlock avoidance, two deadlock detectors (2trains and multi-train), and two-level scalability via geographical and line-station decompositions. The system achieved live regional deployment in May 2021 and full rollout by November 2021, delivering deadlock-free, optimized plans every 10 seconds and demonstrating the practical feasibility and impact of automated dispatch on a national-scale rail network.

Abstract

Union Pacific (UP) is one of the largest transportation companies in the world, with over 50.000 kms of rail network covering 23 states in the United States. In 2017 Union Pacific embarked on a project that within 5 years would lead it to become the only rail operator in the world equipped with a technology capable of fully automating the real-time management and optimization of train traffic. In 2021 the main milestone of such project has been reached with the first deployment of the automated dispatching system we present here. To attack such large and complex problem, we decomposed it into distinct but interrelated functional components, and developed optimization models and methods to handle such components. The models communicate with each other through variables and constraints, and by a careful timing of invocations. In this paper we give an overview of the overall approach.

Optimizing train dispatching for the Union Pacific Railroad

TL;DR

The paper tackles automated real-time train dispatching on Union Pacific's large network, framing it as a resource-constrained planning problem on block sections. It introduces AMP as the planning engine behind an in-house A-TMS, modeling dispatching with a dispatching graph consisting of routing, timetable, and conflict arcs , , and , and solving for a consistent subset that yields schedule when contains no positive cycles. Key contributions include the continuous rolling horizon Branch-and-Bound (CRH-B&B) planner, Safeplacer and Overlength modules for deadlock avoidance, two deadlock detectors (2trains and multi-train), and two-level scalability via geographical and line-station decompositions. The system achieved live regional deployment in May 2021 and full rollout by November 2021, delivering deadlock-free, optimized plans every 10 seconds and demonstrating the practical feasibility and impact of automated dispatch on a national-scale rail network.

Abstract

Union Pacific (UP) is one of the largest transportation companies in the world, with over 50.000 kms of rail network covering 23 states in the United States. In 2017 Union Pacific embarked on a project that within 5 years would lead it to become the only rail operator in the world equipped with a technology capable of fully automating the real-time management and optimization of train traffic. In 2021 the main milestone of such project has been reached with the first deployment of the automated dispatching system we present here. To attack such large and complex problem, we decomposed it into distinct but interrelated functional components, and developed optimization models and methods to handle such components. The models communicate with each other through variables and constraints, and by a careful timing of invocations. In this paper we give an overview of the overall approach.

Paper Structure

This paper contains 19 sections, 2 theorems, 3 equations, 2 figures.

Table of Contents

  1. Introduction
  2. An overview of the real-time train traffic management problem
  3. A-TMS state-of-the-practice
  4. Pilot studies.
  5. Real-life implementations.
  6. Introducing the next sections
  7. A model for the train dispatching problem
  8. Features of the system
  9. Train dispatching by Branch&Bound
  10. Deadlock avoidance
  11. Safeplacer module
  12. Overlength train module
  13. Deadlock detection
  14. Two-train deadlock detector 2trains
  15. Multiple-train deadlock detector
  16. ...and 4 more sections

Key Result

Lemma 1

A (feasible) schedule $t^y$ for $G(y)$ exists if and only if $G(y)$ does not contain positive cycles.

Figures (2)

  • Figure 1: A dispatcher operating center
  • Figure 2: An overview of the system and its continuous planning process.

Theorems & Definitions (2)

  • Lemma 1
  • Theorem 4