Determination of optimal prevention strategy for COVID-19 based on multi-agent simulation

This study proposes a direction for the utilization of multi-agent simulation (MAS) to consider an optimal prevention strategy for the spread of the coronavirus disease of 2019 (COVID-19) through a pandemic modeling example in Japan. MAS can flexibly express macroscopic phenomena formed through the interaction of micro-agents modeled to act autonomously. The use of MAS can provide a variety of recommendations for bringing a pandemic under control, even in the case of the COVID-19 pandemic, which has become more intense as of 2021. However, models that do not consider individual heterogeneity, such as analytical Susceptible-Exposed-Infectious-Recovered (SEIR) models, are often used as predictive models for infectious diseases and the main reference for decision-making. In this study, we show that by constructing a MAS that simulates a metropolitan city in Japan in a simple manner while considering the heterogeneity of age and other background information, we can capture the effects of various measures such as vaccinations on the spread of infections in a more realistic setting. Moreover, it is possible to offer various recommendations for optimal strategies to suppress a pandemic by combining reinforcement learning with MAS. This study explicates the potential of MAS in the development of strategies to prevent the spread of infection.