EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

1515. A Genetic Algorithm with Several Fitness Functions for Line Planning

Invited abstract in session TC-51: Network Design and Line Planning for Public Transportation 1, stream Public Transport Optimization.

Tuesday, 12:30-14:00
Room: M5 (building: 101)

Authors (first author is the speaker)

1. Maarten Wens
KU Leuven
2. Pieter Vansteenwegen
Institute for Mobility - CIB, KU Leuven

Abstract

Most algorithms solving the transit network design problem (TNDP) or Line Planning Problem (LPP) focus only on one objective function and the provided input parameters, e.g. the distance matrix. By changing these parameters or the objective, algorithms can escape from local optima. In this work, a genetic algorithm is created that includes several alternative fitness functions across a population, focusing on different aspects of a solution. This can range from putting a focus on low-demand areas to putting a focus on minimising the transfers. Our implementation introduces the concept of sub-populations, which work in parallel and aim for optimal solutions for slightly different fitness functions. These sub-populations share their found improvements, allowing a synergy between them. When this idea is implemented in an algorithmic framework for network design, the results improve the current literature for all Mumford networks, a commonly used benchmark instance. For these benchmark instances, the objective function value, considering the average travel time of passengers, is reduced by (1, 0.5, 3, 2.5)%. Not only does this algorithm improve current benchmark results, but the output contains a set of high-quality, yet substantially different networks, which is interesting for transport operators and for the follow-up research we have in mind. Finally, this algorithm is successfully used in a case study in cooperation with the main public bus operator of Flanders, Belgium.

Keywords

Status: accepted

Back to the list of papers