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

Area of Science:

• Epidemiology
• Computational modeling
• Public health

Background:

• Traditional Susceptible-Exposed-Infectious-Recovered (SEIR) models lack individual heterogeneity, limiting their predictive accuracy for infectious diseases.
• The COVID-19 pandemic (as of 2021) highlighted the need for more sophisticated modeling to inform public health decisions.
• Multi-agent simulation (MAS) allows for the flexible representation of complex systems through autonomous micro-agent interactions.

Purpose of the Study:

• To propose and demonstrate the utility of multi-agent simulation (MAS) for developing optimal COVID-19 prevention strategies.
• To compare the effectiveness of MAS against traditional analytical models in pandemic modeling.
• To explore the integration of reinforcement learning with MAS for enhanced strategy optimization.

Main Methods:

• Development of a MAS model simulating a metropolitan city in Japan.
• Inclusion of individual heterogeneity (age, background information) within the agent-based model.
• Integration of reinforcement learning techniques with the MAS framework to identify optimal intervention strategies.

Main Results:

• The MAS model realistically captured the effects of interventions like vaccinations on infection spread by accounting for individual differences.
• The study demonstrated that MAS provides more nuanced insights compared to simpler analytical models.
• Combining MAS with reinforcement learning yielded effective recommendations for pandemic suppression.

Conclusions:

• Multi-agent simulation (MAS) presents a powerful and flexible tool for pandemic preparedness and response.
• MAS, by incorporating individual heterogeneity, offers a more realistic approach to modeling infectious disease dynamics.
• The integration of MAS with AI techniques like reinforcement learning can significantly enhance the development of optimal public health strategies.