A data-driven heuristic for the dynamic vehicle routing problem with multiple soft time windows

In the Dynamic Vehicle Routing Problem with Multiple Soft Time Windows (Dynamic VRPMSTW), customer requests arrive in real time and must be scheduled within flexible service intervals. This problem is complicated by operational constraints, such as vehicle capacities, travel durations, and heterogeneous fleets, which make it difficult for classical optimization methods to adapt quickly to changing conditions. Following recent trends in contextual optimization, we propose a Data-Driven Dynamic Heuristic that integrates Artificial Neural Networks for predicting travel times and demands into a Dynamic Hybrid Adaptive Large Neighborhood Search (DD-Dynamic HALNS). Using cluster assignment and genetic crossover operators, the method generates high-quality initial solutions and continuously re-optimizes them as new requests emerge, ensuring adaptability and service reliability. The effectiveness of the proposed method is evaluated on real-world logistics data and benchmark instances. Results from real-world delivery operations demonstrate an average distance reduction of 11.6% compared with the current solution, with further improvements up to 15.5% when a 10-minute time window flexibility is introduced. These findings highlight the practical benefits of integrating predictive analytics with heuristic optimization, leading to improved cost efficiency, reduced operational constraints, and enhanced service reliability.