Abnormity mining based on error and key-point in seismic precursory observation data
一种基于误差和关键点的地震前兆观测数据异常挖掘算法*

Seismic precursory observation data is the very important basis for seismic analysis and forecast. However, the artificial methods are the main mode to deal with the huge data. In order to solve this problem, it need a practical abnormity mining algorithm. This paper brought forward a segment algorithm named EKTW and an abnormity analysis method based on local outlier factor of time domain neighbor(TLOF). The conventional segment algorithm had a poor approximate ability under the high resolution, which brought some bad effect in the process of discovering short-time high-frequency abnormity. Compared with the defect of the conventional segment algorithm, EKTW segment algorithm identifies and holds the key points in time series, which enhances the approximate ability under high resolution. Taking the time attribute into account, the index TLOF evaluates the abnormal degree of an object with its time domain neighbors. Experiments show that the algorithms described above have a good effect in finding the two kind of representative abnormity in seismic precursory observation data.