Busca em vizinhança variável para um problema de estoque e roteamento

Abstract

This paper discuss a case of the Inventory Routing Problem, which considers the distribution and storage costs of a certain product mix, considering not only the manufacturer’s storage but also the buyers’, and also the delivering schedule, aiming the best decisions not only for the manufacturer but for the whole supply chain. A mathematical model was used to describe the problem, for this particular case, will be considered a single factory and one product to be distributed to a group of buyers M between 5 and 50 and also varying the timeframe considering 3 and 6 units of time. First of all, an exact method was tested, the model was implemented in the software AMPL© in order to verify if the it would be able to solve the whole set of instances, but the results showed that it was only able to process problems with 20 buyers when considering 3 units of time or 15 buyers when considering 6 units, therefore another method was necessary in order to solve the bigger problems, to do so, the metaheuristic Variable Neighborhood Search (VNS) was selected for those cases. At the end of the research the code showed good results in some of the tests, but not always, because the search found local optimum results which impact not only the final value, but also the computational time. An alternative that improved the total time was a search structure that generated multiple starting points, although the instability regarding the final values remained, therefore, although the metaheuristic approach proved to be an alternative to the exact methods, the VNS was not able to provide the performance expected, in order to solve this problem another structure able to escape from the local optimums would be necessary or evaluated other metaheuristics that accepts worse results while searching, that would show a similar effect.