Modeling and control of COVID-19 disease using deep reinforcement learning method

Area of Science:

• Epidemiology and Public Health
• Computational Modeling and Control Theory

Background:

• The COVID-19 pandemic has had devastating global health and economic impacts, necessitating effective disease control strategies.
• Existing epidemiological models often lack the granularity to fully capture the complex dynamics of disease spread and intervention effectiveness.

Purpose of the Study:

• To introduce and evaluate a novel SQEIAR epidemiological model for COVID-19.
• To apply optimal control methods to minimize disease prevalence and intervention costs.
• To assess the robustness of the proposed control strategy under various uncertainties.

Main Methods:

• Developed the SQEIAR (Susceptible, Quarantined, Exposed, Infectious-Symptomatic, Infectious-Asymptomatic, Recovered) model incorporating six population groups.
• Integrated three control inputs: quarantine of susceptible individuals, vaccination, and treatment.
• Employed the Deep Deterministic Policy Gradient (DDPG) algorithm for optimal control to minimize symptomatic individuals and costs.
• Simulated the model under different control scenarios and evaluated outcomes, including robustness testing with noise and parameter uncertainty.

Main Results:

• The optimal control strategy reduced deaths by 60% and symptomatic infections by 74% compared to the uncontrolled model.
• The DDPG algorithm successfully identified optimal control inputs for various intervention scenarios.
• The control system demonstrated significant robustness when subjected to noise in observer variables, control inputs, and model parameters.

Conclusions:

• The SQEIAR model with DDPG-based optimal control offers a powerful framework for managing infectious disease outbreaks like COVID-19.
• The proposed intervention strategies are effective in mitigating disease spread and mortality.
• The control system's robustness ensures reliable performance even in the presence of real-world uncertainties.