Line planning under crowding: A cut-and-column generation approach

TL;DR

This paper presents a novel optimization approach for public transit line planning that incorporates crowding effects, using a mixed-integer second-order cone programming formulation and advanced solution algorithms, demonstrating scalability and effectiveness on large networks.

Contribution

It introduces a new MI-SOCP formulation and tailored cut-and-column generation algorithms for line planning under crowding, addressing non-linearity and scalability issues.

Findings

Algorithm scales to large networks with hundreds of stations and OD pairs.

Considering crowding reduces crowding significantly with minimal impact on travel time.

User-equilibrium routing closely aligns with system-optimal solutions.

Abstract

Problem definition: To mitigate excessive crowding in public transit networks, network expansion is often not feasible due to financial and time constraints. Instead, operators are required to make use of existing infrastructure more efficiently. In this regard, this paper considers the problem of determining lines and frequencies in a public transit system, factoring in the impact of crowding. Methodology: We introduce a novel formulation to address the line planning problem under crowding and propose a mixed-integer second-order cone programming (MI-SOCP) reformulation. Three variants of the cut-and-column generation algorithm with tailored acceleration techniques find near-system-optimal solutions by dynamically generating passenger routes and adding linear cutting planes to deal with the non-linearity introduced by the crowding terms. We find integral solutions using a diving heuristic. In practice, passengers may deviate from system-optimal routes. We, thus, evaluate line plans by computing a user-equilibrium routing based on Wardrop's first principle. Results and implications: We experimentally evaluate the performance of the proposed approaches on both an artificial network and the Beijing metro network. The results demonstrate that our algorithm effectively scales to large-scale instances involving hundreds of stations and candidate lines, and nearly 57,000 origin-destination pairs. We find that considering crowding while developing line plans can significantly reduce crowding, at only a minor expense to the travel time passengers experience. This holds both for system-optimal passenger routing and user-optimal passenger routing, which only differ slightly.