Study On Abnormal Detection Method Of Earthquake Precursor Observation Data

Earthquake precursor observation is the basis of the analysis and prediction of the earthquake. After years of effort, precursor observation system has been digitized and networked. The improved accuracy of output data and sampling rate led to the surge in the amount of data. Manual and semi-manual data processing methods can’t meet the actual work needs. In addition, high precision and high frequency sampling data carry more interference information, resulting in more different kinds of the shapes of the data changes and the types of abnormalities, increasing the difficulty of the data processing work. Precursory observation data anomalies recognition is an important basis for earthquake prediction and related geophysical research, and the anomaly caused by various interference factors will affect the real seismic anomaly. Therefore, preprocessing abnormal data is needed before making earthquake prediction and other analysis by precursor observation, and that has become a routine work of the precursor production process. But because of the huge amount of data, manual data processing works hard to meet the demand of actual work. Therefore, it’s needed to study the abnormal data detecting method according to the characteristics of precursory data.After analyzing the characteristics of precursory observation data, this paper studied data pre-processing technology of data mining, and designed the anomaly detection method which combined with precursory data characteristics, and proposed a multi-item detection results integrated application method, which based on big data analysis idea, to determine group anomalies from abnormal detection results. This paper research content, from data processing, data mining to results application, formed a complete set of precursory data analysis application methods.The main work of this paper are as follows:(1) In data preprocessing step, focusing on precursory observation data missing and breaking, data interpolation method suited to the characteristics of precursory data was studied. Because of the precursor data variety and the complexity of shape changes, conventional interpolation methods can’t meet all items’ interpolation precision requirement. So a interpolation method using ARMA forecasting model was provided. It’s effect is better than other methods, and it can be use in the interpolation of consecutively data missing.(2) Based on abnormality detection algorithms of the data mining, combined with the precursor data features, a precursor abnormality detecting pattern method was proposed. By using this method, the obvious abnormal cases, such jumping sharply, steps, and others, could be detected rapidly. And the detection of abnormalities can be controlled by parameters setting. The results of the actual observation data application shows that this method works very effectively on the anomaly detection of massive data. It can effectively solve the precursory network efficiency slowly problem of manual handling mass data, and makes significant senses for the preprocessing work of earthquake precursor observation data.(3) Based on big data analysis idea, a group anomalies dectection method of earthquake precursory observation was proposed. It detected group anomalies by synthesizing abnormality detection results of a wide range of different observation items and long time data, and then analyzed the correlativity between group anomalies and events such as the earthquake. And, the analysis of several moderately strong earthquakes cases in recent years showed that correlativity between group anomaly and systematic events probably exists. The method provides a new way for the innovative application of precursory observation.The features of this paper are that the research content is aimed at solving the existing problems in the practical application of precursory data and the research results has a good practicability. The automatic anomaly detection method can greatly improve the efficiency of data processing, and solve the problem of precursory data automatic detection method is less, and can’t meet the demand of practical application. Big data analytics was introduced to precursor data analysis application, to consider the issue as a whole, to avoid tangling local details. It was the attempt and exploration of the new data application mode in the field of precursory data analysis application.