Research On Inventory Control And Dynamic Pricing Of Fresh Produce Based On Deep Reinforcement Learning

The demand for fresh produce is increasing year after year and the market size is expanding accordingly.However,the high cost of inventory due to the perishable nature of fresh produce and the complex environment make fixed inventory and pricing strategy no longer applicable.For this type of problem,it is especially important for companies to develop an efficient inventory and pricing strategy.Moreover,fresh produce e-commerce platforms have already accumulated a large amount of product and consumer-related data,and making full use of the big data resources stored in the enterprise for inventory and pricing strategies will bring a key competitive advantage to the enterprise.In this paper,we study the joint inventory control and dynamic pricing problem of fresh products based on deep reinforcement learning methods to explore the optimal ordering and pricing strategies in complex environments.First of all,this paper analyzes the main business models of large fresh produce ecommerce websites,models the joint optimal inventory control and dynamic pricing problem for single and multiple products of fresh produce respectively with the characteristics of fresh produce products for real application scenarios,and defines the Markov decision process according to the research problem.Secondly,this paper designs a simulation environment for the application scenario and solves the problem based on online deep reinforcement learning methods for the case that there is difficult and incomplete data collection for certain commodities;and designs a solution framework based on offline reinforcement learning methods for the case that the platform has a large amount of commodity data,but it may be an imperfect data set.Finally,the feasibility and effectiveness of the algorithm are compared and analyzed through the performance experiments of the solution algorithm.The experimental results show that the joint inventory control and dynamic pricing strategies based on deep reinforcement learning have the best performance;compared with the online reinforcement learning methods,the joint inventory control and dynamic pricing strategies based on offline learning can combine with historical data sets to obtain higher returns.The innovation of this paper mainly lies in enriching and supplementing the theoretical system in this field by studying the problem of joint optimized inventory control and dynamic pricing for high-dimensional fresh products under changing state transfer;designing two reasonable simulation environments based on singleproduct and multi-product research problems;conducting research for scenarios with lack of data and sufficient data respectively,and promoting the implementation of reinforcement learning methods in the field of revenue management.