Pandemic response optimisation using multi-objective reinforcement learning approach

Pandemic management demands the careful balancing of multiple, often conflicting objectives, including public health protection, economic resilience, and social well being. Conventional policy optimisation techniques frequently prove inadequate in navigating these complex trade offs, particularly in dynamic and uncertain environments. Recent developments in Multi-Objective Reinforcement Learning (MORL) have demonstrated considerable promise in addressing such challenges. Yet, many existing approaches struggle with generalisability, adaptivity to non-stationary conditions, and the explicit incorporation of ethical and behavioural constraints crucial to public health decision-making. In this work, we present a novel MORL based policy optimisation framework designed explicitly for pandemic response planning. Our approach unifies epidemiological modelling, economic impact assessment, and social utility evaluation within a single intelligent agent. The proposed method employs Pareto front learning augmented with dynamic constraint regularisation and adaptive objective reweighting, facilitating the exploration of a diverse set of optimal strategies while ensuring robust performance in rapidly evolving scenarios. We validate our framework using a high-fidelity agent-based simulation informed by real-world COVID-19 data. Empirical results demonstrate that our approach substantially outperforms existing MORL baselines, achieving a 17% reduction in cumulative infections, a 22% increase in economic activity retention, and a 28% enhancement in aggregate social utility relative to state-of-the-art alternatives. These results highlight the potential of MORL as a foundational tool for future pandemic preparedness and real-time policy adaptation, offering transparent, adaptive, and ethically aligned decision support for diverse stakeholder objectives.