Data-driven Procurement Optimization in Fresh Food Distribution Market under Demand Uncertainty: A Two-stage Stochastic Programming Approach

Lu Xu, Xinglu Liu, Li Xiao, Qiruyi Zuo, Jianfeng Liu, Wai Kin Victor Chan

Abstract: Multiple purchases are necessary for fresh food distribution supply chain management due to the high uncertainties of demand. To optimize the procurement policies of decision-makers (i.e., minimizing the total purchasing cost), we develop a two-stage stochastic programming (TS-SP) formulation in which the purpose of the first stage is to determine the optimal quantity of advance purchase, while the second stage aims at making appropriate replenishment purchase decisions. However, a challenging task in the problem is that the involved scenarios might fail to be generated via the conventional Sampling Average Approximation (SAA) method because the demand may not follow any known probability distribution (or the demand prediction results may be highly inaccurate). To overcome this issue, we propose a novel prediction-model-based scenario generation method that can handle any type of demand uncertainty scheme and incorporate multiple trained demand prediction models. Extensive numerical experiments on real datasets reveal that the TS-SP approach is superior to the 12 considered predict-then-optimize methods in terms of overall purchase cost and stability. Furthermore, the advantage of using the TS-SP approach is gradually enlarged as the number of involved SKUs or considered scenarios increase.