基于机器学习算法的2023年土耳其地震显著持时预测模型

Prediction model of ground-motion significant duration of 2023 Turkey earthquake based on machine learning algorithm

发展可靠的地震动持时预测模型对结构抗震设计与评估等至关重要,随着强震动观测数据数量和质量的提升,基于数据驱动的机器学习方法建立地震动持时预测模型能够取得可靠的预测结果。2023年2月6日土耳其发生两次7级以上地震及多次余震,震源深度均在20 km以内,地震持续时间较长且造成了严重人员伤亡。本文基于机器学习算法对土耳其地震中获得的660组地震动记录的显著持时进行预测,建立了预测模型并开展了残差分析,进一步将预测模型与传统持时预测公式预测结果进行了对比。结果表明:采用机器学习算法建立的预测模型具有较好准确性,可以取得良好预测效果。研究结果和结论可作为地震动参数区划、结构抗震设计和概率地震危险性分析等工作参考。

Developing reliable prediction model for ground-motion duration is crucial for structural seismic design and evaluation.With the improvement of the quantity and quality of ground-motion data,the reliable results can be achieved by using ground-motion prediction model based on machine learning algorithm.On February 6,2023,two earthquakes with magnitude above 7 and several aftershocks occurred in Turkey,with focal depth within 20 km,long duration and serious casualties.In this paper,the significant durations of 660 ground-motion records in the Turkey Earthquake were predicted based on machine learning algorithm,the prediction models were developed and the residuals were analyzed,and further compared the prediction models with the traditional prediction equations.The results show that the prediction model established by machine learning algorithm is more accurate and can achieve satisfactory results.The results and conclusions can be used as reference for the seismic ground motion parameter zonation,structural seismic design and probabilistic seismic hazard analysis.