Research on earthquake localization method for multiple stations based on hybrid genetic algorithm

Earthquake localization plays a critical role in Earthquake Early Warning (EEW) systems, as its real-time capability and accuracy directly determine the timeliness and reliability of warnings. Conventional localization methods typically rely on complex waveform analyses and multiple seismic parameters, which often entail substantial computational costs and exhibit sensitivity to data quality and environmental noise. To reconcile efficiency with precision under the real-time requirements of dense seismic networks, this study introduces a multi-station localization approach that integrates Voronoi diagrams with a Genetic Algorithm (GA). The procedure begins with the STA/LTA algorithm to pick P-wave arrivals at each station and determine their sequence. Based on this chronological order, the Voronoi diagram is applied to confine the source region to a limited area, serving as the initial search space for the subsequent GA. A double-difference travel-time fitness function is constructed to iteratively refine candidate source locations, ultimately yielding precise event coordinates. Tests using 20 events recorded by Japan’s K-NET and KiK-net networks demonstrate that the method converges to an optimal fit within eight iterations, with an average localization time of ≤3.3 s and epicentral errors consistently within 0–5 km. Compared to the Gaussian Mixture Model Association (GaMMA) method, the proposed approach reduces computational time by 1.349 s and lowers the mean square error by 51.18%. These results indicate that the method effectively balances real-time performance and robustness in dense network environments, offering reliable technical support for EEW systems.