Abstract
Globally, the outbreaks of infectious diseases have exerted an extremelyprofound and severe influence on health security and the economy. During thecritical phases of epidemics, devising effective intervention measures poses asignificant challenge to both the academic and practical arenas. There isnumerous research based on reinforcement learning to optimize interventionmeasures of infectious diseases. Nevertheless, most of these efforts have beenconfined within the differential equation based on infectious disease models.Although a limited number of studies have incorporated reinforcement learningmethodologies into individual-based infectious disease models, the modelsemployed therein have entailed simplifications and limitations, rendering itincapable of modeling the complexity and dynamics inherent in infectiousdisease transmission. We establish a decision-making framework based on anindividual agent-based transmission model, utilizing reinforcement learning tocontinuously explore and develop a strategy function. The framework's validityis verified through both experimental and theoretical approaches. Covasim, adetailed and widely used agent-based disease transmission model, was modifiedto support reinforcement learning research. We conduct an exhaustiveexploration of the application efficacy of multiple algorithms across diverseaction spaces. Furthermore, we conduct an innovative preliminary theoreticalanalysis concerning the issue of "time coverage". The results of the experimentrobustly validate the effectiveness and feasibility of the methodologicalframework of this study. The coping strategies gleaned therefrom prove highlyefficacious in suppressing the expansion of the epidemic scale and safeguardingthe stability of the economic system, thereby providing crucial referenceperspectives for the formulation of global public health security strategies.