Research on Order Acceptance Strategy for Military manufactures Based on Semi-Markov Average Reward Reinforcement Learning

Order acceptance is an important decision-making issue that military manufacturers need to consider, which is crucial for military manufacturers to complete their production tasks and improve their benefits. In this paper, we first analyze the customer orders pattern and production pattern of military manufacturers, and study the order acceptance decision algorithm for the military products orders with the characteristics such as small orders, multiple batches, strong dynamics, emphasis on customer priority, and complex inventory requirements. The order acceptance problem model of traditional manufacturing enterprises mainly considers factors such as delay penalty cost, rejection cost and production cost under static state conditions, On the basis of these factors, we further take the inventory cost of orders completed before the lead time and multiple customer priorities as influencing factors of order acceptance problem into consideration under dynamic demand. Combine with the idea of reward management, we solve the order acceptance problem with a semi-Markov decision process (SMDP) model using a model-free reinforcement learning method: Semi-Markov Average Reward Technique (SMART). Finally, We present a detailed study of this algorithm on a military order acceptance problem in the simulation environment. We compare the average reward, order acceptance rate and product acceptance rate of different algorithms, and verify the superiority of the order acceptance strategy model proposed in this paper over the traditional decision model.