Repair Crew Routing for Power Distribution Network Restoration

TL;DR

This paper introduces the Power Restoration Traveling Repairman Problem (PRTRP), a complex routing challenge for power grid repair crews that considers interactions between road and power networks to minimize service disruption time.

Contribution

It formulates PRTRP, develops an exact bi-directional dynamic programming solution with bounds and heuristics, and compares its performance against benchmark heuristics.

Findings

Exact method effectively solves PRTRP instances.

Heuristic variants provide near-optimal solutions efficiently.

Proposed approach outperforms benchmark heuristics in computational tests.

Abstract

As extreme weather events become more frequent and disruptive, service restoration is increasingly important for many infrastructures, e.g., power grids and communication networks. In many studies on service restoration, the logistics issue of traveling over the road network, however, is often overlooked due to the complexity of considering multiple networks simultaneously, resulting in prolonged disruption time. In this work, we address such a problem arising in power systems, where technical crew and utility trucks travel to a number of sites to repair damaged equipment, with the goal of minimizing the total service disruption time within the service region. We call this problem the Power Restoration Traveling Repairman Problem (PRTRP). What makes it significantly more challenging than a typical routing problem is that the service disruption time in a location depends on the interaction of the routing sequence with both networks, i.e., the road network and the power grid. To solve the problem, we develop an exact method based on bi-directional dynamic programming. We then improve the method by reducing the search space with solution upper and lower bounds, and threshold rules derived from the precedence relations in the power grid. We also propose efficient heuristic variants of the method. We present computational results and compare our method with benchmark heuristics.