Deep Reinforcement Learning with Convolutional Neural Networks for Optimizing Supply Chain Inventory Management

Abstract

With the increasing complexity of global supply chains and frequent fluctuations in market demand, inventory management faces severe challenges, and traditional inventory control methods are difficult to meet the needs. This paper constructs an inventory control model that combines deep reinforcement learning (DRL) with convolutional neural networks (CNN). By defining the state space, action space and reward function, the Q-learning algorithm is used to optimize inventory decisions. At the same time, CNN is used to extract historical demand data features to improve the accuracy of demand forecasting. This study uses historical sales data from a medium-sized clothing retailer as a dataset, which contains sales, inventory, and replenishment records for the past 4 years. The model was trained for 500 episodes. This model was compared with the economic order quantity (EOQ) model, the periodic ordering model, and the simple moving average forecasting model as the benchmark model. The model's demand forecast error of 3.2% was measured on independent actual test data. The experimental results show that the model has a demand forecast error of only 3.2%, the total inventory cost is 14,500 yuan, the cost reduction rate is - 22%, the average inventory turnover rate is 10.5 times, and the average out-of-stock rate is only 2.1%. All indicators are significantly better than the economic order quantity (EOQ) model and the periodic ordering model. The study proves that the model can effectively cope with demand fluctuations and uncertainties, optimize inventory management, and provide a new and effective method for supply chain inventory control.