Computationally Efficient Dynamic Traffic Optimization Of Railway Systems

TL;DR

This paper introduces a real-time, model predictive control approach for dynamic railway traffic optimization that guarantees safety constraints while reducing computational complexity through minimal prediction horizons.

Contribution

It develops procedures to determine safe, minimal optimization horizons ensuring safety and feasibility, enabling efficient, real-time railway traffic management with solution reuse capabilities.

Findings

Guarantees safety despite limited prediction horizons

Provides upper bounds on computational requirements

Demonstrates effectiveness on a real-world freight railway system

Abstract

In this paper we investigate real-time, dynamic traffic optimization in railway systems. In order to enable practical solution times, we operate the optimizer in a receding horizon fashion and with optimization horizons that are shorter than the full path to destinations, using a model predictive control (MPC) approach. We present new procedures to establish safe prediction horizons, providing formal guarantees that the system is operated in a way that satisfies hard safety constraints despite the fact that not all future train interactions are taken into account, by characterizing the minimal required optimization horizons. We also show that any feasible solution to our proposed models is sufficient to maintain a safe, automated operation of the railway system, providing an upper bound on the computations strictly required. Additionally, we show that these minimal optimization horizons also characterize an upper bound on computations required to construct a feasible solution for any arbitrary optimization horizon, paving the way for anytime algorithms. Finally, our results enable systematic solution reuse, when previous schedules are available. We test our approach on a detailed simulation environment of a real-world railway system used for freight transport.