We have presented a hybrid approach to solving the Vehicle Routing Problem with Stochastic Demands (VRPSD). The approach combines Monte Carlo simulation with reliability indices and a well-tested metaheuristic for the Capacitated Vehicle Routing Problem (CVRP). One of the basic ideas of our methodology is to consider a vehicle capacity lower than the actual maximum vehicle capacity when designing VRPSD solutions. This way, this capacity surplus or safety stocks can be used when necessary to cover route failures without having to assume the usually high costs involved in vehicle restock trips. Another important idea is to transform the VRPSD instance to a limited set of CVRP instances – each of them defined by a given safety-stocks level – to which efficient solving methods can be applied. Our approach provides the decision-maker with a set of alternative solutions, each of them characterized by their total estimated costs and their reliability values – the former reflecting the probability of that solution being a feasible one – leaving to him/her the responsibility of selecting the specific solution to be implemented according to his/her utility function. Although other previous works have proposed to benefit from the relationship between the VRPSD and the CVRP, they usually require hard assumptions that are not always satisfied in realistic scenarios. On the contrary, our approach relaxes most of these assumptions and, therefore, it allows for considering more realistic customer demand scenarios. Thus, for example, our approach can be used to solve CVRPSD instances with hundreds of nodes in a reasonable time and, even more important, it is valid for virtually any statistical distribution – the one that best fits historical data on customer demands. Also, the methodology can be naturally extended to consider: (a) different distributions for different customer demands, (b) possible dependences among these demands, (c) multiple failures per route, and (d) multiple recourse strategies. A complete set of tests have been performed to illustrate the methodology and analyze its efficiency as well as its potential benefits over previous works.
As future work, we plan to compare the efficiency and robustness of the proposed approach against alternative optimization methods lying on mathematical or constraint programming models. More specifically, the main idea is to replace the metaheuristic used in the step 4 of the proposed methodology by a rigorous two-stage stochastic optimization approach. In this way, the solution generated will simultaneously consider multiple scenarios for the customer demands instead of the one based only on the expected value of each random demand. By considering the uncertain information of demands in a proactive way, we expect to be able to generate cost-effective solutions with higher reliability, although at the expense of a potential significant increase of the computational times. The trade-off between solution quality and computational times will be carefully evaluated. In addition, since routes failures may be reduced but never eliminated, we also plan to develop efficient dynamic optimization methods for quickly updating the original solution after the occurrence of route failures.