This paper studies the optimization of inventory control strategy in supply chain management by using data analysis and machine learning. Traditional inventory control strategy relies on static assumptions and historical data, so it is difficult to adapt to the realistic environment of intensified demand fluctuation, complicated supply chain and frequent emergencies. In this paper, by integrating multi-source data and extracting key variables by feature engineering, a high-dimensional demand forecasting model is constructed, and dynamic forecasting and catastrophe risk identification are realized by combining time series analysis and ensemble learning. At the same time, the reinforcement learning (RL) framework is constructed by using the Deep Deterministic Policy Gradient (DDPG) algorithm, and the dynamic replenishment strategy is designed, and the multi-objective collaborative optimization is realized through the constrained multi-objective optimization model and NSGA-III algorithm. The empirical research is based on the operation data from January 2022 to March 2024, covering three product lines. The results show that this research method is significantly superior to the traditional method and benchmark machine learning model in the core indicators such as inventory turnover rate, order satisfaction rate, average inventory days, emergency replenishment frequency and forecast error, which effectively reduces the inventory holding cost, improves the operational efficiency and resilience of the supply chain, and has good economic benefits and feasibility.
Wang et al. (Sun,) studied this question.