Latency Aware Metaheuristics for Earthquake Early Warning Network Design

Abstract

Earthquake Early Warning(EEW) network design couples a physical deadline with uncertain sensing, processing, and communication delays. An improved optimization score is useful only if it increases the probability that an actionable alert arrives before the relevant shaking. This paper develops a latency based interpretation of metaheuristic network design through a critical literature synthesis, a precise timing model, and a reproducible placement experiment. The model treats station reports as concurrent arrivals, represents confirmation by an order statistic, and includes central processing, delivery, action time, and station outages. We establish monotonicity under a fixed exogenous latency model, exhibit the failure of submodularity under multiple station confirmation, and specialize a finite family concentration bound to distinguish optimization error from sampling error. Five search methods are evaluated in six constructed planning cases using 900 runs and 1.62 million charged objective evaluations. Complete enumeration of 405,273 feasible placements supplies training optima for every case, and independent scenario sets assess generalization. Simulated Annealing(SA) and restarted swap search attain every small case optimum in all 90 runs per method, whereas performance becomes more variable in the larger cases. For prespecified placements, adding four seconds to the event path reduces weighted timely coverage by 9.33–10.71 percentage points, exceeding the largest median training gap among the tested solvers. The findings support deadline based evaluation and offline optimization, while remaining conditional on the stated geometry and delay distributions. They establish neither a universal algorithm ranking nor operational performance for a deployed warning network.