Research On Data Driven Algorithm For Time Series B-value Calculation

The b value of time series has an important application in the risk analysis of natural earthquakes and industrial mining induced earthquakes.However,for a long time,the traditional calculation methods have been affected by the strong artificial subjectivity in the setting of calculation rules,the reliability of calculation results and the low accuracy on changepoints identification,which restrict the comparability of different results and the refinement of consensus scientific knowledge.In order to solve the above problems,a time series b-value calculation method based on data-driven(TbDD)is constructed by using OK1993 model in the form of continuous function,time axis random partitioning,model selection of Bayesian information criterion(BIC).Using OK 1993 model in the form of continuous function,random segment division of time axis,model selection of BIC and other technical links.In this paper,the theoretical test of TbDD method is carried out by using the synthetic earthquake catalogues,and compared TbDD method with the traditional fixed window method.The results show that the TbDD method can restore the input parameter bo value of the synthetic earthquake catalog,and has obvious advantages in the objectivity of the calculation rule setting and the accurate identification of the changepoints of b values.The TbDD method based on BIC was applied to the 2021 Yangbi Ms6.4 earthquake sequence in Yunnan Province,and the influence of the TbDD method on the calculation results was further tested from the perspective of the number of random models and the setting of random sections of the time axis.The test results show that the b-value is less affected by the number of random models and has strong stability.The setting of random sections of the time axis will affect the disclosure of the micro fluctuation of the b-value time series.In order to verify the effectiveness of different model selection criteria,we use BIC,Akaike information criterion(AIC),AIC and BIC mixed criterion for model selection,and conduct theoretical tests on the recovery of ensemble results and ensemble median to input parameters considering weight.The test results show that the initial parameters can be recovered well by using AIC to select the best model with a certain proportion,and using the ensemble median as the final calculation result.We also apply TbDD method based on AIC to the case study of the 2021 Yangbi Ms6.4 earthquake sequence in Yunnan and the 2022 Menyuan Ms6.9 earthquake sequence in Qinghai,as well as the hydraulic fracturing induced earthquake sequences of Geysers and Newberry EGS projects.The paper also summarizes the physical mechanism of spatiotemporal heterogeneity of seismic b value induced by hydraulic fracturing at home and abroad.The TbDD method developed in this paper has certain stability and objectivity compared with the traditional method.The optimalcriterion of model selection obtained through theoretical test can further improve the accuracy of TbDD.The TbDD method,which uses the AIC model selection criterion and the ensemble median as the final result,has good application potential in the fields of aftershock trend tracking,industrial mining induced earthquake risk management and control,etc.,which have high requirements for the accuracy and precision of time series b value calculation.