A new data-driven framework to select the optimal replenishment strategy in complex supply chains

Abstract

Motivated by the high variability of markets occurred in the last years, which in turns determined significant uncertainty in lead times and supply chain dynamics, this paper introduces a data-driven framework based on machine learning and metaheuristic optimization to dynamically select the most suitable replenishment strategy for a complex two-echelon (supplier-inventory-factory) supply chain (SC) problem with perishable product and stochastic lead times. Since the supplier dispatches the product (i.e., the raw material) with a fixed expiration date, the product shelf-life strictly depends on the related delivery lead time, which is subject to uncertainty. In addition, a minimum order quantity has to be fulfilled and the time between two consecutive orders cannot be less than one month. The aim of the work is to select the most suitable replenishment strategy able to minimize the average stock level, which is a surrogate cost metric, while respecting a target fill rate. Considering a smoothing order-up-to policy, the data-driven prediction-optimization framework makes use of Artificial Neural Network (ANN) and Particle Swarm Optimization (PSO) to select the best replenishment parameters (i.e., forecasting factor, proportional controller and safety stock factor) able to dynamically enhance the SC economic performance under the fill rate constraint. The ability of the framework under the predictive and the optimization perspective is assessed and a sensitivity analysis on the influence of replenishment parameters is presented as well.