Research On Long-wave Radiation Anomaly Before Earthquakes Based On Time Series Prediction Models

China is one of the countries that have suffered the most severe earthquake disasters in the world,and actively conducting earthquake research and earthquake prevention and disaster reduction work is an important task for earthquake workers in China.In recent years,machine learning and deep learning have developed rapidly.The advantage of these models lies in processing complex multi feature data and discovering hidden information in the data.Many disciplines of seismic research have accumulated massive data with different forms and characteristics.This data feature meets the requirements of machine learning for sample data,and further promotes the research of earthquake prediction based on multidisciplinary and multi-source data of machine learning.Since former Soviet scientist V.I.Gorny et al.(1988)discovered anomalies in preearthquake thermal infrared remote sensing images while studying earthquakes in Central Asia,seismic researchers from various countries have begun to use thermal infrared data to study seismic activity.Infrared longwave radiation data has strong temporal persistence.Since the 1970 s,NOAA series satellites have continuously obtained longwave radiation data and are open for use.The traditional longwave radiation anomaly extraction algorithm is a mathematical model that lacks learning ability and has poor universality for different earthquake cases.A major characteristic of time series prediction models is their ability to process data at long time scales,which has certain advantages in processing infrared longwave radiation data recorded in long time series.In addition to traditional time series prediction models,deep learning time prediction models all have good data learning and feature extraction capabilities.This paper first introduces the research progress of machine learning in Earthquake prediction based on multi-source data of different seismic disciplines and summarizes different data sources.Different seismology disciplines include Seismology,Geodesy,geochemistry,seismic Electromagnetism and seismic geology.Secondly,based on infrared long wave radiation time series data,different time series prediction models are used to predict 5 ° with the epicenter as the center × Infrared longwave radiation values within a range of 5 ° and at different predicted time ranges before earthquakes.By selecting algorithms to evaluate earthquake cases,five prediction algorithms,namely,ARMA,ARIMA,SVM,XGBoost and BILSTM,were used to predict the long wave radiation values in different time ranges before the earthquake.After comparing the Root-mean-square deviation of all grids,five days were determined as the best prediction days,and BILSTM was the best model.The anomaly extraction stage uses a95% confidence interval as the standard to extract the amplitude and corresponding date that exceed the confidence interval range.Eight earthquakes with magnitudes ranging from 6.0 to 7.0 within the scope of the China Earthquake Science Experimental Site were selected as research examples for anomaly extraction research.Finally,analyze the anomalous features of eight strong earthquakes based on time and space dimensions,and summarize the distribution patterns of anomalous features.Research conclusion:(1)Among the eight earthquakes,except for the thrust type Lushan earthquake,which mainly showed positive anomalies,the other seven strike slip types of earthquakes all showed negative anomalies.Based on time series prediction models to detect long wave radiation anomalies before earthquakes,strike slip earthquakes often exhibit negative anomalies,and this conclusion needs to be verified by a large number of earthquake examples in the future.(2)In the eight earthquake cases,the distribution of Lushan earthquake,Ludian earthquake,Jiuzhaigou Valley Scenic and Historic Interest Area earthquake and Changning earthquake anomalies is significant,while the distribution of other earthquake anomalies is irregular.The abnormal spatial distribution of earthquake cases has a certain degree of randomness;(3)The anomalous time concentration of the eight earthquakes is strong,with multiple grids experiencing longwave radiation anomalies simultaneously within one day.A strong and concentrated long wave radiation anomaly occurred during the Yao’an earthquake,Jinggu earthquake,and Yangbi earthquake;The other five earthquakes have multiple strong and concentrated long wave radiation anomalies.Based on this feature,an evaluation index for detecting long wave radiation anomalies before earthquakes using time series prediction models can be preliminarily constructed based on long wave radiation data.This article combines deep learning time series prediction models with infrared longwave radiation data to apply to seismic anomaly research,enriching the methods for extracting seismic anomaly information and having certain positive significance for thermal infrared seismic anomaly research.