Whose preferences matter more? Handling unbalanced panel data for choice modelling

Abstract

Public transport systems are typically designed based on estimated passenger demand and supply patterns, yet may often be called to operate under vastly different operational settings. To systematically design resilient transit systems, it is necessary to “weave” resilience-oriented thinking into the established public transport network design process, moving from an abstract concept to an implementable methodology. This study aims to effectively and efficiently design resilient public transport networks through the integration of Reinforcement Learning (RL), Local Search operators and Particle Swarm Optimization. We present a redundancy indicator and integrate it within a hybrid RL-enhanced metaheuristic solution framework to design more resilient route structures. We apply the proposed Memetic algorithm to an established benchmark from the literature and validate the proposed approach under a series of random and targeted attacks, simulating link disruptions. Results demonstrate that resilience can be enhanced through redundancy without adversely impacting average travel times.