A Data-Driven Heuristic for the Dynamic Vehicle Routing Problem with Multiple Soft Time Windows

This project proposes a data-driven heuristic for the dynamic vehicle routing problem with multiple soft time windows (Dynamic VRP-MSTW). It applies machine learning forecasting techniques to estimate delivery earliness/tardiness probabilities at vertices and arcs of the VRP graph. These probabilities determine the softness of each time window at every customer. The refined data is to be fed to a dynamic hybrid adaptive large neighborhood search (Dynamic-HALNS) algorithm, which provides a primary near-optimal routing solution, and performs real-time re-optimization of remaining portions of routes during their execution in case of emergency. The overall objective of the project is to improve the real-time performance of the vehicle routing operations. The machine learning and heuristic' parameters are to be calibrated following different scenarios. Under different real-life scenarios, the competitiveness of the proposed algorithm is to be tested on real-life instances to show the efficiency of the proposed heuristic approach and its adaptability.