Robust simulation optimization for supply chain problem under uncertainty via neural network metamodeling

Abstract

Real-world supply chain management problems are highly complicated such that their optimization procedure is computationally expensive due to the extensive dimensions and uncertainty of their critical variables. Simulation optimization is a commonly applied technique to determine the optimal variables since the problem is too complex. Due to the uncertain nature of the real-world systems, it is also worthy to consider the robustness of the optimal solutions. To address this issue, this study investigates the problem of determining near-optimal safety stock levels in a multi-product supply chain with regard to deviations of its overall cost. A new framework is proposed to define the decision as well as environmental variables. This novel framework results in a significant reduction of the solution space while maintains the essential supply chain control parameters. The prediction performance of artificial neural networks (ANNs) with different structural settings are compared, and the best-fitted ANNs are selected to obtain the robust solutions. Consequently, we employ a robust metamodel-based simulation optimization approach based on Taguchi’s view and optimize the multi-objective supply chain problem with respect to supply chain operational costs and customer satisfaction criteria.