Prediction Of Xinjiang Earthquake Based On Time Series Data

Xinjiang is located in the front of the collision of Indian Ocean plate and Eurasian plate.There are five major seismic belts,namely,Altay seismic belt,North Tianshan seismic belt,South Tianshan seismic belt,West Kunlun seismic belt and Altun mountain seismic belt.High frequency,large magnitude and wide range are the characteristics of earthquakes in Xinjiang.However,most of the earthquakes in Xinjiang occur in mountainous areas,far away from cities.Compared with several seismic belts in the east of China,the losses of people and property are relatively small.In order to explore the recessive relationship between the occurrence frequency and time of earthquakes in Xinjiang,the BP Neural Network Algorithm in machine learning is used to predict the earthquake frequency according to the earthquake catalogue data of Xinjiang in recent ten years.In order to explore the movement rule of Jinghe crustal deformation in theory,ARIMA prediction model is constructed based on the deformation data of Jinghe seismic station in the past ten years.According to the 2009-2018 earthquake catalogue data in Xinjiang,the earthquake frequency is predicted and the recessive relationship between frequency and time is analyzed.In this paper,the logarithm of earthquake frequency is selected as the output of BP Neural Network to predict the frequency of earthquakes with MS3,4 and 5 in 2017 and 2018 in Xinjiang.In the preprocessing of seismic catalogue data,firstly,we screen the time and magnitude of the earthquake,count the number of earthquakes with magnitude 3,4 and 5 in each year of Xinjiang,and take the logarithm of the frequency after statistics;secondly,we reduce the training time of BP neural network,avoid the problem of slow convergence,and normalize the input value of the network.According to the processed data,the software is used to train the network model and adjust the network weight at the same time.When the network weight basically converges,the regression prediction shows that the earthquake frequency in 2017 and 2018 in Xinjiang region has high accuracy and small error.Verify that there is a linear or non-linear relationship between the frequency and time of earthquakes in Xinjiang,and verify the feasibility and relevant value of BP Neural Network model to express this relationship.Finally,using the method of commensurability to make the commensurability formula of earthquakes with m≥6,to predict the Jinghe earthquake on August 9,2017,and to verify that there are certain laws for the occurrence of strong earthquakes in Xinjiang.Based on the whole point time data of crustal deformation,this paper intercepts the data from 00:00 on January 1,2009 to 23:00 on January 27,2019,carries out noise processing and establishes ARIMA model.Using the software to analyze the deformation data,using the time sequence diagram to judge the sequence stability,ADF test to further determine the sequence as non-stationary sequence,using the difference method to make the non-stationary sequence stable,using the model identification and model test to get the prediction model.ARIMA(1,1,1)model is used in the NS and EW directions of water pipes,ARIMA(5,1,8)model and ARIMA(4,1,2)model are used in the NS and EW directions of horizontal pendulum,ARIMA(1,1,8)model and ARIMA(1,1,5)model are used in the NS and EW directions of extensometer,The modelR~2(29).095,the significance of all models?(27).00 1,BIC value is also very small,each AR,Ma,?(27).001,t value is also very significant,through the residual test and heteroscedasticity test,it shows that the model fitting is good,can describe the law of crustal deformation.Through the establishment of ARIMA model to predict the change trend of crustal deformation,comparing the observed value with the predicted value,according to the data of water pipe,horizontal pendulum and extensometer,the error before and after the earthquake(August 9,2017)is greater than the normal.It can be seen that ARIMA model provides a reliable background to distinguish crustal deformation anomalies.The analysis of Xinjiang earthquake is helpful to develop the research of earthquake prediction and numerical prediction,which shows that the monitoring environment of Xinjiang seismic station is better and the monitoring data of monitoring instrument is more accurate.