Optimizing Inventory Policies Under Supply Chain Disruption Risks: A Deep Learning-Based Surrogate Modeling Approach

This study proposes a Deep Neural Network (DNN)-based surrogate model methodology to overcome the computational limitations of traditional simulation-based optimization in managing supply chain disruption risks and lead-time variability. By learning from stochastic disruption simulation data within a 3-echelon supply chain, the Multi-Layer Perceptron (MLP) surrogate model accurately approximated complex cost surfaces. The results quantitatively demonstrate a buffering mechanism where upstream suppliers preemptively increase safety stock to absorb system-wide shocks under escalating risks. The proposed model serves as an effective real-time Decision Support System (DSS), instantly determining the optimal Reorder Point (ROP) without computational bottlenecks. Ultimately, this study lays a vital academic foundation for developing Deep Reinforcement Learning (DRL)-based adaptive inventory policies for highly dynamic supply chains.