Hybrid Optimization Framework for Closed-Loop Supply Chains: Embedding Neural Networks in Two-Stage Stochastic Programming
TL;DR
The model combines a neural network with Two-Stage Stochastic Programming (2SP) to handle demand uncertainty and significantly improves strategic and operational decision-making by integrating critical elements such as location planning and material flows.
Abstract
This paper develops an integrated surrogate model for optimizing a sustainable Closed-Loop Supply Chain (CLSC) in an Additive Manufacturing (AM) context. The model combines a neural network with Two-Stage Stochastic Programming (2SP) to handle demand uncertainty. This model significantly improves strategic and operational decision-making by integrating critical elements such as location planning and material flows, leveraging machine learning to manage uncertainties and enhance adaptability to dynamic demand behaviours. The efficacy of our approach is supported by computational experiments showing that the integration of 2SP and machine learning can provide solutions that are more scalable and computationally efficient than Sample Average Approximation (SAA) for the problem instances considered. The results suggest that this model may offer a useful tool for practitioners working on sustainable CLSC design under demand uncertainty, though further validation on larger real-world instances is needed. From a managerial perspective, the proposed framework can support sustainable investment planning and risk management in circular economy strategies.