Inventory Optimization in Retail Supply Chains Using Deep Reinforcement Learning

Abstract

Inventory management is a critical component of retail supply chains, directly affecting operational efficiency, customer satisfaction, and profitability. Traditional approaches to inventory optimization often rely on heuristic rules or static mathematical models, which struggle to cope with the high-dimensional, stochastic, and dynamic nature of modern retail environments. This paper proposes a novel framework utilizing deep reinforcement learning (DRL) to optimize inventory control decisions in end-to-end retail supply chains. The supply chain system is modeled as a Markov Decision Process (MDP), where the agent observes states such as stock levels, sales trends, supplier lead times, and demand forecasts. A DRL agent, trained with the Deep Deterministic Policy Gradient (DDPG) algorithm, learns to generate real-time replenishment and ordering strategies that maximize long-term performance by minimizing costs and avoiding stockouts. Experimental evaluations using both simulated and real-world retail data demonstrate that the proposed method outperforms classical baselines such as economic order quantity (EOQ) and safety stock models in terms of inventory turnover, service level, and total cost. The results suggest that DRL can serve as a robust and adaptive solution to inventory optimization under uncertainty.