Research On Key Techniques Of Abnormal Recognition Of Seismic Precursor Data

China is an earthquake prone country. Under the combined action of the Eurasian plate, the Pacific plate and the Indian plate, the seismic activity of China is high frequency, high intensity and range wide. The occurrence of earthquake disaster will bring great harm to people's life and social economy. The research of earthquake prediction in China has been carried out for many years. At present, there are more than 50 kinds of observations in the National Earthquake Precursory Network of China (NEPNC). After years of observation, abundant data has been accumulated on continuous observation of crustal movement, Geophysics and geochemistry which is the key product of NEPNC and the basement for investigation of earthquake prediction. However, due to the complex and variability of earthquake event, the formation mechanism of the earthquake and the pregnancy process are still in the initial stage of exploration. The seismic precursor observation data collected at the present stage contains the details of signal variation and the mechanism of dynamic evolution process. The analysis of these data can accurately identify the abnormal pattern of earthquake precursor and provide a theoretical and technical reference for the earliest and fast analysis on earthquake event. Currently, the most outstanding problem of the limitation of precursor observation data in earthquake prediction application effectively is that the massive high sampling data does not match with the traditional data processing model by hand for each station day by day. The traditional calculation method and processing of the model has been difficult to quickly capture anomalies in the massive data automatically. Therefore, how to use big data technology and abnormal model automatic identification technology on the earthquake precursory observation data, to exclude anomalies of environmental and the instrument factors, to study the relationship between the anomaly and earthquake and to carry out the practice of earthquake prediction is a very significant work. In the application of it on the research works, the seismic precursor observation data anomaly recognition technology has become the most suitable technology which could fully reflect characteristics such as the limitation of seismic precursor observation system resources, and the instability, multi-source, heterogeneous of massive observation data.At present, the research on anomaly recognition technology of seismic precursor faces the following problems:(1) Because observation system has several data types, massive data, different requirements of data transmission, and high failure rate of instruments and network, we cannot obtain timely, accurate and complete observation data and guarantee the reliability of data acquisition and transmission. (2) There are many unfavorable factors, such as high failure rate of observation equipment, complex and changeable external environment disturbance, and some interference factors are not aware of by the current technical means or people have not been aware of and ruled out. Such instable data quality yields great difficulties to the seismic precursor observation data anomaly pattern recognition. (3) With the scale increasing of earthquake precursory observation data analysis system, the data type enriching and time span increasing, the amount of data to be comprehensive analysis accordingly in the network has increased dramatically hand by hand. In addition, the corresponding analysis of massive data needs to process as quickly as possible, in order to find the abnormal pattern of earthquake precursor and deal with it immediately.To solve the above technical problems, the paper summarizes the existing work. And according to its own characteristics of earthquake precursor observation data, this paper researches the reliability, the accuracy and the high efficiency of data acquisition for multi-source heterogeneous massive data, the method of accurate anomaly pattern recognition algorithms and the rapid identification method with big data platform. The results can provide new solution for the existing seismic precursor observation data processing system and perfect it. The innovative work of this paper mainly includes the following aspects.(1) In the aspect of reliability data acquisition, low latency, high data precision and overall reliability of data collection in underground observation environment, a prototype system of hierarchical structure is designed. Based on regional data collection algorithm and automatic dormancy and wakeup mechanism, the core idea of the system increase data processing speed and reduce unnecessary energy consumption monitoring in order to detect earthquake precursor phenomenon such as supplementary observation (meteorology and hydrology) immediately. The experimental results show that compared with the existing monitoring system, the layered system designed in this paper with faster data processing speed, and can reduce unnecessary energy consumption for monitoring, more timely and accurate predict the environment change. A novel data forwarding protocol for energy saving optimal ring is proposed. The approximate optimization is used to solve the problem of the edge of the sensor network to optimize the data compression ratio in the data fusion, so as to reduce the adverse effect of the funneling effect, which can greatly release the inefficient data transmission and serious energy waste. Experimental results show that the new designed protocol could balance and reduce energy consumption which can extend the network lifetime and support large-scale network.(2) The traditional pattern recognition method only compares the amplitude of point value in the original signal time series. And the recognition accuracy influenced by the noise of the signal greatly. To solve this problem, a porous wavelet algorithm is used. Firstly, the original data time series are decomposed; Then, we use the multi-resolution anomaly pattern recognition algorithm to match the amplitude and local frequency of each sampling point at different levels; Finally, this method can search the abnormal pattern with repeated signal and arbitrary length and structure automatically. Porous wavelet algorithm ensures that the time shift is invariance in the transformation process and the time accuracy of the identification results, which also has the stronger anti-interference ability. We carried out the experiment on simulated data and the observed data with above method. The results show that the algorithm not only can accurately identify noise anomaly pattern which artificially added in the simulated data, also can detect some tiny signal we usually difficultly to feel from different scales on data analysis in the practical observation data. Compared with traditional algorithm, this algorithm's biggest advantage is that the algorithm can reset the amplitude and frequency index weightings respectively for the practice application.(3) The traditional model on data processing by hand for each station day by day already far falls behind massive high sampling rate observation data processing system. And the traditional method has been difficult to capture the abnormal automatically and quickly, and low recognition rate from the massive data. Referencing graphics processing theory and data processing technology and improving the SURF algorithm in order to better adapt to the seismic waveform image matching, we define a new similarity measure function for the description of earthquake precursor abnormal waveform change trend form, and complete the SURF feature extraction and feature matching based on HADOOP program development. We also carry out the single serial and parallel cluster algorithm of SURF anomaly pattern recognition experiment on real seismic data, and analyze the differences of efficiency and accuracy for the single machine and cluster algorithm. At the same time, we verify the effectiveness and practicability of the SURF algorithm to identify anomalies of seismic data with different sampling scale and waveform.