Integrated Forecast and Optimization for Retailer Allocation in a Two-Echelon Inventory System

The paper reports on a real-world application case of integrated data forecasting and process optimization for a tactical-level study in the context of the final stage of two-echelon logistics support for a large retail chain. The allocation of final stores to distribution centers had to AQ1 be redefined in view of the expected increase in sales volume during the Christmas season. For this purpose, time series of past sales were projected up to the period of interest as a basis for the reoptimization of the store allocation. The latter problem was identified as an extended generalized assignment problem, which was solved using a Lagrangian matheuristic. Computational results are presented showing a comparison of different forecasting algorithms on the actual data and the advantages of using a matheuristic for optimization in this industrial setting, even when compared to results obtained by proprietary third-party solutions.