Advanced start method for solving vehicle routing problems

This study addresses a routing problem at a last-mile company in Vietnam. It proposes using a machine learning technique-Nearest Neighbor Search-to identify the most similar known solution, which is then used as the initial solution for a branch-and-bound algorithm. Firstly, the machine learning investigated a dataset of 500 pre-solved problems to find solutions for new problems. Then, the solution found is fed to the solver as a hint. The experimental results indicate that the algorithm can retrieve solutions for similar cases in just a few milliseconds. For entirely new problems, the method reduces computation time compared to the traditional branch-and-bound algorithm. This significant improvement in efficiency demonstrates the method’s practical value in real-world logistics operations. The combination of machine learning and classical optimization in this approach shows promise for enhancing decision-making processes in the logistics industry, particularly in scenarios requiring rapid and reliable route planning.