地震事件的多重时空聚类算法研究

Multi-step spatiotemporal clustering algorithms for seismic events

地震序列是一组在一定的时间和空间范围内连续发生的地震事件,其发震机制具有某种内在联系,包含着地震孕育、发生、发展过程的丰富信息,是研究地震孕育、发震构造、区域应力特征及地震活动趋势的重要依据.实际的地震序列经常在时间和空间上重叠并淹没在海量的背景地震事件中,使得准确高效地识别地震序列仍然存在难度.本文基于ST-DBSCAN和DBSCAN两种聚类算法,采用先松后紧的参数设置策略,通过多次聚类逐渐寻找最佳的地震序列的方法构建了多重时空聚类算法,并同时获得每一个地震序列的展布方向、持续时间、序列类型等地震序列的特征参数.通过对理论合成地震目录和六盘山及其邻区内的实际地震目录的测试表明,该算法不需要预先设置地震序列的个数,可以有效的识别小震群;避免使用时空耦合距离参数,使参数的设置更具有实际意义;采用多次聚类策略,提高了聚类的有效性和准确性;计算效率高节约内存,能够对数量庞大的地震目录进行聚类;能够获得地震序列的相关统计参数,为进一步的定量分析提供依据;能够在含有噪点的数据集中有效识别具有复杂时空分布的地震序列,解决了因发震构造形状复杂所造导致的震群分布复杂无法识别的问题.

Earthquake sequence is a set of earthquake that occur in continuous time and space. Its seismic mechanism has strong correlation. Earthquake sequence is important basis to study earthquake breeds, seismic mechanism, Stress feature and Seismicity, because earthquake sequence contains the information of earthquake preparation, ccurrence and development. In fact, earthquake sequences are submerged in massive background earthquake, and overlap in time and space, which makes it difficult to identify earthquake sequences accurately and efficiently. Base on ST-DBSCAN and DBSCAN, this paper constructed multi-step spatiotemporal clustering algorithms to Search for the best earthquake sequence. The algorithm’s main process: the first step is using ST-DBSCAN algorithm for spatiotemporal clustering by loose clustering parameters;the second step is to use DBSCAN clustering algorithm to spatially cluster by tight clustering parameters;the third step is using DBSCAN clustering algorithm to time clustering by time clustering;finally, counting key parameters of earthquake sequence, such as the distribution direction、duration、sequence type, etc. Using theoretically synthesized earthquake catalogue and actual earthquake catalogue in Liupanshan to test the method. The test result shows:(1)This method can effectively identify earthquake sequences which submerged in massive background earthquake and overlaped in time and space, can identify earthquake sequence with complex distribution due to complex shape of seismogenic structure;(2)There is no need to preset the number of seismic sequences, so the hidden earthquake sequences can be effectively distinguish;(3)Avoid using space-time coupling distance to make parameter setting more practical;(4)Multi-step spatiotemporal clustering can not only avoid missing earthquake events, but also improve the accuracy of clustering;(5)The analysis of earthquake arameters can provide quantitative data basis for further analysis;(6)Under the current monitoring capability of the seismic network, Changing the clustering parameters has little effect on the clustering results.(7)This method improves the computational efficiency, saves memory, and can cluster huge amount of earthquake catalogs.