확률적 리드타임과 반품이 존재하는 공급망에서DRL 기반 재고정책의 설계와 (s, S) 정책 비교

Modern supply chains are increasingly complex due to globalization, interdependence, and real-time information processing, with stochastic lead times and product returns emerging as major sources of uncertainty. These uncertainties directly impact inventory and customer service levels, leading to issues like excess inventory, increased costs, and forecast errors that traditional inventory policies struggle to resolve. While conventional (s, S) policies perform well in stable environments, they show limitations in flexibility and efficiency in complex real-world scenarios with stochastic lead times, high demand variability, and irregular return flows. Furthermore, existing research on inventory management incorporating returns often relies on simplified models assuming single items, fixed demand, and constant lead times, failing to reflect actual operational complexities. This study aims to bridge this academic gap by analyzing inventory management in uncertain supply and demand environments where both stochastic lead times and returns are present. We propose a new reinforcement learning (RL)-based algorithm to overcome the limitations of traditional (s, S) policies in such complex settings. The performance of our developed algorithm will be empirically validated by comparing its superiority and practical applicability against the conventional (s, S) policy. The findings will demonstrate the potential of data-driven, learning-based approaches to effectively manage the intricate challenges of modern supply chains in complex situations involving both stochastic lead times and product returns.