Research And Software Development On Seismogram Records Vetorizing And Earthquake Catalogs Data Mining

There have been many destructive earthquakes in human history. These earthquakes have caused great loss to human life, production and life, and provide valuable historical data for understanding, researching and predicting earthquakes.A large amount of analog seismograms in the past 70 years are still of great value. These original analog seismograms are closely associated with the earthquake-generating process and can be applied for analysis and study of earthquakes. At the same time, some existing earthquake catalogues in the world, including a large number of information about seismic energy radiation and mechanism solution, contains a wealth of seismic time and space correlation. These correlations with potential value are mostly implicit, prior unknown. It expects people to adopt new methods and techniques to find the important law implied in them and can help to predict the future earthquakes. We can vectorize analog seismograms, and mine information from earthquake catalogues by developing digital image processing and data mining algorithm, which has important practical scientific value for research and forecast earthquakes.Using only one digital image processing algorithm to vectorize analog seismograms has certain limitations and one-sidedness. Integrating a variety of digital image processing and data mining algorithms, and based on big data technology, a set of algorithms with excellent anti-noise proper and good robustness proper are proposed. Paper analog seismograms contain many noises, a variety of digital image processing algorithms are adopted to remove background noise and extract useful waveforms. Some data mining algorithms are used for dynamic analysis, intelligent inversion of the waveform records after the vectorization of seismograms, and rapidly positioning of seismic waveforms. A new program on digitizing analog seismograms based on big data platform can handle more drawings per unit of time, and its processing speed is faster. Which can achieve accurately, large-scale vectorization of different types, different batches of analog seismograms. In this paper, we focused on about 500 seismograms acquired in 1991 stored in the seismic station of Chengde. Based on the full analysis of a variety of seismograms abroad and at home, the advantages and disadvantages of all kinds of digital image processing and data mining algorithms are deeply studied. We present an algorithm called color scene filed method(CSF) to automatic digitize seismograms, and an intelligent inversion algorithm based on an improved K-means algorithm to rapidly position seismic waveforms. Then an interactive program based on big data platform is developed using C# to digitize seismogram traces from raster files quickly and accurately.At the present, the most researches of earthquake catalogs data mining still pay too much attention to the physical relationship between seismic parameters, but ignore the statistics correlation between seismic parameters, which does not meet the technical characteristics of the era of big data. Now, the types of earthquake catalogs used in the aftershocks forecast research is less, and the representativeness of the samples is not high. The most works remain in the numerical simulation based on partial data, without efficient supported algorithm and software system. In this paper, supported by various data mining algorithms, a statistical correlation model between main-shock magnitude, main-shock apparent stress, main-shock seismic moment, main-shock focal depth and so on, and the biggest after-shock is established, which may be an important basis for prediction of after-shock. Based on big data platform, using C# language, a program is developed based on earthquake catalogs for after-shock prediction.The main research results and innovations of this research are as following:1. This paper systematically summarizes vectorization methods for analog seismogram records and data mining methods for earthquake catalogs. We analyzes and compares various analog seismograms vectorization methods abroad and at home, summarizes the advantages and disadvantages of each method, and discusses the development trend in the research field. Using high precision industrial scanners, we scan many analog seismograms to get their raster image files, which are sorted, classified and marked using artificial way. The two earthquake catalogs are collected, which are NEIC broad-band radiated energy catalog and Harvard CMT catalogue. Between May 31, 2013 and January 1, 1977, 46587 seismic events between M5 and M9 are cleaned and sorted systematically.2. A raster image file can be pretreated by graying, binarizing and erasing successively to decrease noise and obtain the proper image format(PNG) by an optional pretreatment model designed in the study. A smoothing waveform tracing algorithm based on the linear max weight algorithm is designed, which is used to eliminate smoothing waveforms and realize further de-noising in a seismogram. A tracing algorithm is proposed for the first time, which is named Color Scene Filed Method(CSFM) and is used to trace seismic waveforms in a seismogram. It greatly improves the accuracy and robustness of seismogram vectorization. A corresponding error analysis based on vision is given out.3. The corresponding time mosaic algorithm and curves mosaic algorithm are provided base on two linear mathematical models. A waveform inversion algorithm has also been discussed based on linear mathematical and K-Means models, using which we realize the automatic quantitative extraction and display of seismic waveform.4. A big data platform is built, which is based on Mongo and used to store and distribute analog seismograms and their vectorization results. We discuss some new methods and new techniques used for sharing of analog seismograms by seismic scientists. A software system has been designed and developed, which is used to vectorize analog seismograms. The altogether 10 G analog seismograms, which come from Chengde Seismic station in 1991, have been digitalized by using the software. The treatment results were analyzed.5. Based on the analysis and research on the relationship between the various seismic parameters, a main-after shocks sequence classification model and an earthquake type's classification model are established. Based on an improved SVM regression model, a statistical correlation between main-shocks and afters-hocks is designed, and a prediction work of aftershocks is carried out. A software system for prediction global after-shocks based on the two earthquake catalogs is developed. An accuracy estimation model based on cross validation is presented, and the robustness analysis is carried out by the measured data.