基于无监督机器学习的噪声信号聚类分析——以郯庐断裂带潍坊段短周期密集台阵观测为例

Using unsupervised machine learning for clustering seismic noise:a case study of a dense seismic array at the Weifang segment of the Tanlu Fault

非构造活动震源所引起的地面震动通常被看成是地震记录中的噪声信号,此类噪声与微震或非火山震颤等弱构造活动信号往往在时间域或频率域都难以区分,从而会影响利用常规方法对弱构造活动信号的识别与检测.即使利用最新的机器学习方法对微地震信号检测,若缺乏对噪声信号特性的了解,也会对监督模型的训练产生不利的影响.因此,有必要剖析地震噪声信号,理解其特征属性,以及背后可能的物理震源.本研究中,我们利用一个布设于华北东部地区的短周期密集观测台阵,使用K-means算法聚类分析不同类型的地震噪声信号.分析表明密集台阵可以观测到6类噪声信号,噪声来源包括轨道交通、风和附近的电力输送线.

Ground motion induced by non-tectonic sources is commonly regarded as seismic noise in seismic recording.Seismic noise has similar characteristics of weak tectonic signals such as microearthquakes or nonvolcanic tremors,making the recognition and detection of weak tectonic signals very challenging with traditional signal processing techniques.Lack of knowledge on seismic noise features can also negatively affect model training when applying the latest machine learning techniques for weak signal detection.Thus,it is necessary to conduct investigations on seismic noise,understanding its features and potential sources.In this study,we apply the Kmeans analysis to continuous recordings of a dense seismic array deployed at the Weifang segment of the Tanlu fault in north China to characterize various types of seismic noise.The analysis indicates that noise field recorded by the array comprises 6types of signals related to various nontectonic sources,including road and rail traffic,wind,and nearby power lines.