An ALNS-based matheuristic algorithm for a multi-product many-to-many maritime inventory routing problem

Abstract

AbstractIn this paper, we propose an adaptive large neighborhood search-based matheuristic algorithm to solve a multi-product many-to-many maritime inventory routing problem. The problem addresses a short sea inventory routing problem that aims to find the best route and distribution plan for multiple products with a heterogeneous fleet of vessels through a network including several producers and customers. Each port can be visited a given number of times during the planning horizon, and the stock level for each product should lie within the predefined bound limits. The problem was introduced by Hemmati et al. (Eur J Oper Res 252:775–788, 2016). They developed a mixed integer programming formulation and proposed a matheuristic algorithm to solve the problem. Although their proposed algorithm worked well in terms of running time, it suffers from disregarding a part of the solution space. In this study, we propose a new matheuristic algorithm to find better solutions by exploring the entire solution space for the same problem. In our solution methodology, we split the variables into routing and non-routing variables. Then in an iterative process, we determine the values of the routing variables with an adaptive large neighborhood search algorithm, and we pass them as input to a penalized model which is a relaxed and modified version of the mathematical model introduced in Hemmati et al. (2016). The information from solving the penalized model, including the values of the non-routing variables, is then passed to the adaptive large neighborhood search algorithm for the next iteration. Several problem-dependent operators are defined. The operators use the information they get from the penalized model and focus on decreasing the penalty values. Computational results show up to 26% improvement in the quality of the solutions for the group of instances with a large feasible solution space. We get the optimal value for the remaining instances matched with the reported results.