Efficient active-passive vehicle coordination in multimodal transportation networks

Description

This dataset accompanies a study submitted to "Transportation Research Part B: Methodological", focusing on a modular, demand-responsive transport system. The system uses passive, modular containers, or “pods,” that are transported across multiple vehicle types like trucks, trains, and boats to fulfill door-to-door customer requests. The study's main objective is to improve customer satisfaction and reduce operator costs, thus providing a more flexible alternative to traditional public transportation systems. The unique challenge lies in synchronizing the movements of both the pods and the vehicles they rely on, which creates interdependencies between routes. To address this, a mixed-integer linear program (MILP) and an adaptive large neighborhood search (ALNS) heuristic were developed, alongside a clustering approach to accelerate solution times for real-world applications. This dataset validates the performance of these optimization methods, providing a practical framework for testing advanced multimodal routing solutions. Since no standard benchmark data existed for this novel system, the dataset consists of three custom instance sets - A, B, and C - with increasing levels of complexity and sizes, containing 30 instances each. Set A includes scenarios with 4 customer requests, 2 pods, 2 bases, and 6 hubs, representing smaller, simpler problems. Set B scales up to 10 requests, 6 pods, 4 bases, and 8 hubs, while Set C represents the most complex instances with 20 requests, 10 pods, 5 bases, and 10 hubs. Each instance includes randomly generated hubs in a two-dimensional space, with minimum pairwise distances maintained for spatial realism. The transportation network is constructed by assigning edges randomly within each modality (e.g., road, rail) while ensuring full connectivity. Additionally, each base and pod is assigned a capacity, and time windows for customer requests vary, with each subsequent set having a longer time window range. Travel times are computed as the Euclidean distance between hubs divided by a constant speed factor, with adjustments for loading and boarding times to reflect real-world operational delays.

Steps to reproduce

A link to the paper will be provided once it is published, including a detailed description of the instance creation process.

Institutions

Categories

Funding

Licence