A machine-learning based hybrid algorithm for strategic location of urban bundling hubs to support shared public transport

The location problem of Bundling Hubs (BHs) remains a contentious issue for efficient shared transportation systems. In this respect, the strategic configuration of BHs plays a crucial role in saving supply costs, covering demand, and minimizing the external effects of Shared Passenger and Freight Public Transportation (SPFPT). As urban areas become crowded, they show a significant increase in congestion and transport demand. Thus, sites where logistic operations, sales, or services are likely to occur, imply the final customers whose transport demand is a key factor that could affect cargo distribution using SPFPT systems. Since each BH should help efficiently to satisfy the transport demand of allocated customers, they would not play their key role if such factor of demand is not involved upstream in the long-term scheduling horizon. This paper focuses on locating BHs, using a Hybrid Robust Machine Learning-Heuristic Algorithm (HR-MLHA), among established ones and existing demand nodes while assessing a dynamic process so that the configured BH network is robust. The feature of robustness is supported by a robust command to keep BHs attractive and demand-responsive for the long-term in dynamic environments, i.e., the cities. To reduce complexity, the conceptual and computational approaches are structured in two main axes. The first axis includes a machine-learning-based zoning approach that helps with targeting the implementation area and assessing demand behavior. The second axis presents a mathematical model of the capacitated two-echelon BH location problem. When looking across the two-echelon location process, we aim at conducting dynamic location analysis using both current and predicted demand. In order to validate our approach, a set of benchmarks has been performed comparing with existing heuristics and using a whole package of experimental and real-life instances. The experimental results provided through the proposed approach have allowed valuable insights into successfully implementing our methodology.