Optimizing lockdown schedules for emerging pandemics using multi-objective evolutionary algorithms

Lockdown policies are critical non-pharmaceutical interventions for controlling emerging infectious disease outbreaks, yet their design involves fundamental trade-offs between epidemic suppression and socioeconomic costs. This study develops a framework combining agent-based epidemic modeling on small-world networks with multi-objective optimization using the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to identify optimal lockdown strategies. We optimize three competing objectives – minimizing peak infection ratio, final epidemic size, and social production loss – across a population of 10,000 individuals with moderate infectivity (R0=2.5). Under unconstrained conditions, Pareto-optimal solutions cluster into three qualitatively distinct strategies: an aggressive early-intervention strategy that achieves near-complete epidemic elimination, a minimal-intervention strategy that allows largely natural epidemic progression, and a sustained mild-lockdown strategy that suppresses the peak while permitting herd immunity acquisition. However, the elimination strategy requires maintaining lockdowns until near-zero active cases remain—a criterion that demands extremely high detection accuracy and sustained policy compliance. While theoretically achievable, such conditions may be difficult to guarantee under practical surveillance uncertainty and behavioral fatigue. Introducing explicit lower bounds on lockdown release thresholds (0.5%–5% of population) removes the elimination-dominated cluster and leaves a predominantly continuous trade-off surface. Under these constraints, stricter release criteria enable more effective peak suppression but require higher lockdown intensity, and final epidemic sizes consistently fall within 0.6–0.8 regardless of release threshold stringency, indicating that herd immunity acquisition remains the fundamental resolution mechanism under all practically implementable policies. These results show that ignoring implementation constraints can qualitatively alter the Pareto frontier and shift policy recommendations toward regimes requiring highly stringent monitoring and compliance conditions, underscoring the necessity of constraint-aware optimization for pandemic response planning.