Research on Multi-Stage Decision Optimization Based on Monte Carlo Simulation and Deep Q-Network Algorithm

This paper presents a comprehensive study on multi-stage decision optimization leveraging Monte Carlo simulation and Deep Q-Network (DQN) algorithms. We integrate Monte Carlo simulation with sequential sampling, progressively drawing samples and dynamically calculating the log-likelihood ratio to determine the inspection plan. Furthermore, the paper delves into the quality control decision-making problem in multi-stage production processes. Utilizing a DQN model coupled with Bayesian optimization for hyperparameter tuning, our approach employs reinforcement learning where an agent interacts with the environment to progressively learn the optimal decision-making strategy. Experimental results demonstrate that the DQN model effectively reduces inspection and disassembly costs. Notably, when the defective rate is low or inspection costs are high, the agent tends to minimize inspection frequencies.