Random-Key Optimizer for the Problem of Routing Vehicles Empowered with Reverse Deliveries and Pickups

This study addresses the Capacitated Vehicle Routing Problem with constraints, within the business context of a logistics company. The problem involves auctioning routes for a heterogeneous fleet that performs deliveries and reverse pickups, aiming to optimize costs and distances, based on a predefined distance matrix. The approach includes a pre-planning stage with route clustering into micro-regions and distribution of high-cost product prices. The objective is to develop an optimization method for product delivery and reverse logistics from the warehouse to the final customer. The Random-Key Optimizer (RKO) method is proposed and compared against other metaheuristics used in the process. Additionally, parameter tuning is performed using a reinforcement learning algorithm: Q-Learning. Preliminary results show that metaheuristics are effective in solving the problem, achieving good solutions within reasonable computational time. Among the methods tested, VNS and ILS performed well, but RKO stood out as the most effective approach tailored to the problem.