Time-dependent heterogeneous vehicle routing problem with path flexibility based on traffic flow prediction and TDACO-NSGA-II

The Green Time-Dependent Vehicle Routing Problem (G-TDVRP) has garnered growing attention. However, its heterogeneous and path-flexible variant—the Green Time-Dependent Heterogeneous Vehicle Routing Problem with Path Flexibility (G-TDHVRP-PF)—remains underexplored. To address the spatiotemporal variations in vehicle speeds caused by traffic uncertainty, this study proposes an integrated framework combining traffic flow prediction and vehicle routing optimization. An improved Spatio-Temporal Graph Convolutional Network (STGCN) with a composite adjacency matrix effectively captures the spatial attenuation of traffic flow, thereby enhancing the accuracy of speed prediction. A hybrid Time-Dependent Ant Colony Optimization-NSGA-II algorithm balances delivery cost and carbon emissions by integrating a probability suppression mechanism and a time-sensitive 2-opt operator, thereby enhancing routing flexibility and efficiency. Experiments using real data from a Macao logistics company (96 time intervals, 30 customers, 52 intersections) show an 63.7% larger solution set and 5.65–7.69% higher diversity, demonstrating strong adaptability to dynamic urban traffic conditions.