Research on Enterprise Production Decision-Making: Integrating Monte Carlo with Double Deep Q-Network Methods

This paper proposes an optimization method for production decision-making by integrating Monte Carlo and Double DQN (Deep Q-Network) models. It addresses defect rate sampling issues in two-process production, aiming to improve efficiency. By constructing state transition equations and reward functions, a production cycle is defined as the process of selecting components and finished products for sale or discard. In scenarios where step decisions are fixed and information is not shared, Monte Carlo simulation generates various strategies to estimate expected returns under six scenarios. When step information is shared, the Markov Decision Process (MDP) and Bellman equation are used to derive the Q-values, which are then estimated using memory replay, Online DQN, Target DQN, and Double DQN. The optimal strategy is chosen based on the maximum Q-value. Sensitivity analysis indicates the model's stability and predictability under market price fluctuations. The combination of the two methods enables this study to optimize production profits and make optimal decisions under different scenarios where there is difficulty in information communication or the ability to dynamically adjust the defect rate in real-time during the second process, thereby generating the optimal decision-making path.