Deriving knowledge from local optima networks for evolutionary optimization in inventory routing problem.

This paper proposes an evolutionary approach to solve the Inventory Routing Problem (IRP) using knowledge extracted from Local Optima Networks (LONs). Solving the IRP involves a simultaneous optimization of the transportation routes and the inventory levels. One of important steps in solving IRP is determining the optimal route of each supplying vehicle for each date of the planning horizon. As the transportation cost is based only on the distance matrix, constant in time and independent of the supplying vehicle, this step consists of solving a TSP problem on a certain subset of facilities. This paper aims at improving solving IRP by deriving some knowledge on the full TSP problem and reusing it in solving TSP sub-problems. Experiments carried out on popular benchmark IRP instances prove that using the knowledge derived from LONs increases the efficiency of the evolutionary algorithm and the proposed approach outperforms simple evolutionary algorithms in solving IRP.