Reinforcement learning has made great success in recent years. Generally, the learning process requires a huge amount of interaction with the environment before an agent can achieve acceptable performance. This motivates many techniques, such as incorporating prior knowledge which is usually presented as experts’ demonstration, and using a probability distribution to represent state-and-action values, to accelerate the learning process. The methods perform well when the prior knowledge is genuinely correct and no much change occurs to the learning environment. However, the requirement is not perfectly realistic in many complex applications. The demonstration knowledge may not reflect the true environment and even be full of noise. In this paper, we introduce a dynamic distribution merging method to improve knowledge utilization in a general reinforcement learning algorithm, namely Q-learning. The new method adapts a normal distribution to represent state-action values and merges the prior and learned knowledge in a discriminative way. We theoretically analyze the new learning method and demonstrate its empirical performance over multiple problem domains.

After the coronavirus disease 2019 (COVID-19) outbreak erupted, it swiftly spread globally and triggered a severe public health crisis in 2019. To contain the virus’s spread, several countries implemented various lockdown measures. As the governments faced this unprecedented challenge, understanding the impact of lockdown policies became paramount. The goal of addressing the pandemic crisis is to devise prudent policies that strike a balance between safeguarding lives and maintaining economic stability. Traditional mathematical and statistical models for studying virus transmission only offer macro-level predictions of epidemic development and often overlook individual variations’ impact, therefore failing to reflect the role of government decisions. To address this challenge, we propose an integrated framework that combines agent-based modeling (ABM) and deep Q-network (DQN) techniques. This framework enables a more comprehensive analysis and optimization of epidemic control strategies while considering real human behavior. We construct a pandemic simulator based on the ABM method, accurately simulating agents’ daily activities, interactions, and the dynamic spread of the virus. Additionally, we employ a data-driven approach and adjust the model through real statistical data to enhance its effectiveness. Subsequently, we integrated ABM into a decision-making framework using reinforcement learning techniques to explore the most effective strategies. In experiments, we validated the model’s effectiveness by simulating virus transmission across different countries globally. In this model, we obtained decision outcomes when governments focused on various factors. Our research findings indicate that our model serves as a valuable tool for decision-makers, enabling them to formulate prudent and rational policies.