Capacitated Vehicle Routing Using Hybrid Genetic Search and Deep Reinforcement Learning

A distribution centre must deliver known demands to 100-300 customers with trucks of fixed capacity and wants the shortest total distance: the capacitated vehicle routing problem (CVRP). This project compares the classical Clarke- Wright savings heuristic with local search, Google OR-Tools guided local search, the hybrid genetic search of PyVRP (the state of the art) and a POMO attention model trained by deep reinforcement learning on a laptop CPU, on the standard CVRPLIB X benchmark instances with their best-known solutions, and checks small instances against an exact MILP. A web dispatch planner with route maps, solver comparisons and dispatch plans, and a complete MATLAB version are included.

Category: Industrial Engineering

Industrial Engineering Project IE05 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.