Distributionally robust optimization for the multi-period multi-item lot-sizing problems under yield uncertainty

Yield uncertainty is an important issue in various industries such as agriculture, food, and textile where the production output relies on uncontrollable factors and fluctuating raw material quality. To systematically leverage data to deal with uncertainty cost-effectively, distributionally robust optimization combines the strengths of stochastic programming and robust optimization by optimizing the expected costs against an ambiguity set that defines possible distributions. In this work, we leverage a data-driven robust optimization framework and formulate a mixed-integer distributionally robust multi-item lot-sizing model with uncertain production yield to determine a robust production plan. To this end, we use a scenario-wise formulation that partitions the available data into scenarios that define different patterns influencing the quality of the product and production process. In addition, we apply the proposed approach to real-world data of a case study to demonstrate the effectiveness of the proposed framework in dealing with yield uncertainty. Our experimental results show that distributionally robust plans lead to more effective cost-saving strategies and decreased risk of stock-outs. Additionally, our findings suggest that the proposed model exhibits lower sensitivity to variations in production yield realizations and it is more proficient in incorporating historical data into the decision-making process. This results in a more effective response to challenges encountered within the production system under yield uncertainty.