Research On Seismic Precursor Anomaly Detection Method And Realization Of Early Warning Platform

Seismic precursory data play an important role in earthquake early warning,postearthquake trend analysis,petroleum exploration and other geophysical fields.However,it is still challenging to obtain high-quality seismic precursory data.The traditional method of manual cleaning and anomaly detection of precursor data has the problems of low efficiency and subjectivity.In this regard,scholars at home and abroad continue to make in-depth research and propose different anomaly detection algorithms for precursor data.However,the current anomaly detection algorithm of precursor data is mainly based on off-line batch processing,which can not meet the real-time requirements of station staff and can not be well applied to the real scene.Moreover most of the methods only for a certain type of precursor anomaly detection data,each type of precursor data need to design the corresponding detection model,thus inevitably require multiple model deployment at the same time,increase the difficulty of deployment and the demand of the computing resources,on the other hand for single type of precursor data in anomaly detection to ignore the multidimensional time series data potential nonlinear relationship among different dimensions,and studies have shown that make full use of the data the correlation between each dimension on anomaly detection work has important guiding significance.Therefore,this paper will try to explore an online multi-dimensional precursor data synchronization anomaly detection algorithm.The main work and contributions of this paper are as follows:(1)A data cleaning method based on EWT-IFOREST-LI is proposed.Multidimensional precursory data as there is a dimensional data resulted from abnormal noise interference or the total moment data is not available,use the experience of the wavelet transform to decompose of precursor data of each dimension,the decomposed high frequency part of the modal component isolated forests method was used to clean the data to eliminate one of the mutations and noise data points,linear interpolation method to eliminate the data points used for filling to ensure data integrity.(2)A precursor anomaly detection algorithm based on CNN-BiLSTM network prediction is proposed.In the prediction stage,the online multi-dimensional synchronous prediction model CNN-BiLSTM is proposed.BiLSTM is used to capture the long-term characteristics of the time-series data,and CNN is used to capture the potential non-linear relationship inside the time-series data,so as to improve the loss function,so that the multi-dimensional data can achieve a good prediction effect.In the outlier evaluation stage,the original outlier score was obtained by comparing the prediction error with the mean error.In the abnormal judgment stage,sliding window and normal distribution right-tail function were used to adjust the threshold value for judgment,and the overall abnormal score was evaluated at the moment.(3)According to the actual needs of Chuzhou Seismological Bureau,a seismic precursor anomaly detection and early warning platform based on Spring Boot and VUE front and rear end separation was built.The algorithm proposed in this paper was applied to the anomaly detection module in the platform to realize the effective transformation of scientific research achievements.