Research On Timing Complexity Risk Assessment Technology And Its Applicatio

As social and organizational structures become more and more complex,more and more people are aware of the complexity of risks.The complexity of risk and risk assessment has been recognized by specialists in risk management.However,due to the different requirements of different industries,organizations,and risks,quantitative assessment methods need to be enriched and supplemented.The existing fractals and chaos theory in complex systems and nonlinear science have not been applied to com-plexity risk assessment.The cross-study of the complex system,nonlinear science and complexity risk assessment may be helpful.The complexity risk is too wide to cover all risks.A feasible direction is to study part of the complexity first,then enrich risks gradually.In this dissertation,we first distinguish the temporal complexity risk from general complexity of risks.By utilizing the nonlinear scientific knowledge,based on data-driven and machine learning methods,the technology and application of temporal complexity risk assessment are been studied.After giving the definition and scope of temporal complexity risk,the existing con-cepts,theories,and methods in nonlinear science can be introduced into the complexity risk assessment technology in this dissertation.The theories of multifractal,multifractal cross-correlation,and chaos can be extended to multifractality complexity risk assess-ment,multifractal cross-correlation complexity risk assessment,and chaos complexity assessment.Furthermore,new studies and innovations are been developed.The recent research results of multifractal and multifractal cross-correlation show that the causes of multifractality should be expanded to three parts,and the nonlinear correlations are the genuine source of the multifractality.How to qualify intrinsic multifractality has become a problem that needs to be solved.The MCASS-MFDFA method and MCASS-MFDCCA method are proposed to quantitatively evaluate the intrinsic multifractality.It is further applied to multifractality and multifractal cross-correlation complexity risk assessment.Aiming at the problem of multi-scale selection and prediction of chaotic time series,by integrating multi-scale chaotic analysis and machine learning algorithm,the TS-CSVR method is proposed.Then,the three methods are applied to five application cases,including insurance stock market multifractal complexity risk assessment,insurance stock market multifrac-tal cross-correlation complexity risk assessment,air pollution index multifractal com-plexity risk assessment,multifractal cross-correlation between PM_2.5and O₃complexity risk assessment and multi-scale chaos complexity risk assessment and prediction of avi-ation accidents and aircraft hull insurance losses.The five application cases belong to three types of complex systems.The first two application cases belong to the insurance stock market in the financial system,the second and the third application cases belong to the atmospheric system in the natural environment,and the last one belongs to the aviation and insurance system in the real economy.The effectiveness of the three meth-ods proposed in this dissertation is verified through five application cases.The study of application cases improves the deep understanding of three complex systems and gives some meaningful conclusions for specific application fields.The contributions of the dissertation are summarized as follows:（1）A data-driven multifractal causes analysis method based on stochastic simula-tion and MFDFA（MCASS-MFDFA method）is proposed.By integrating the MFDFA method,traditional multifractal causes analysis,and random stochastic simulation method,the MCASS-MFDFA can give more objective and more robust quantification results of three multifractal causes and intrinsic multifractality of a single complex time series.Two application cases show that the apparent multifractality is overestimated,which shows the effectiveness of the MCASS-MFDFA method.In addition,this method can also deeply analyze the components of multifractality and compare the difference between apparent and intrinsic multifractality.（2）A data-driven multifractal causes analysis method based on stochastic sim-ulation and MFDCCA（MCASS-MFDCCA method）is introduced.By integrating the MFDCCA method,traditional multifractal causes analysis,and random stochas-tic simulation method,the MCASS-MFDCCA can give more objective and more ro-bust quantification results of three multifractal causes and intrinsic multifractal cross-correlation between two complex time series.Two application cases show that the ap-parent multifractal cross-correlation is overestimated,which shows the effectiveness of the MCASS-MFDCCA method.In addition,this method can also deeply analyze the components of multifractal cross-correlation and compare the difference between apparent and intrinsic multifractal cross-correlation.（3）A two-stage SVM prediction method based on multi-scale chaotic analysis（TS-CSVR method）is developed.By integrating multi-scale chaotic analysis and traditional SVR prediction methods,this method can improve the forecasting accuracy of a chaotic time series.An application case verifies the effectiveness of the method.（4）The MCASS-MFDFA method is utilized to study the multifractal complexity evaluation of the insurance stock market and air pollution.Because the two application cases are in different modules of different complex systems,the empirical results are quite different.Firstly,it illustrates the”scissors difference”between the insurance stock market index and the composite index before and after the COVID-19 outbreak.Through the dynamic risk assessment of the multifractal complexity,it shows that there is a larger intrinsic multifractality of the three insurance indexes after COVID-19,but intrinsic multifractality is still not the main reason,and the increase in the intrinsic multifractality is weaker than the apparent multifractality.The dynamic risk analysis of the multifractal complexity of the AQI index in Shanghai before and after the COVID-19 lockdown shows that there is little difference between the apparent and intrinsic multifractality.（5）The MCASS-MFDCCA method is employed for the multifractal cross-correlation complexity assessment of the insurance stock market and air pollution.The multifractal cross-correlation complexity dynamic risk assessment between the insur-ance stock market index and the composite index before and after the COVID-19 out-break shows that the intrinsic multifractal cross-correlation has become the main reason after the COVID-19 outbreak.However,the increased percentage of intrinsic multi-fractal cross-correlation is weaker than that of apparent multifractal cross-correlation.Moreover,the”seesaw effect”between PM_2.5and O₃is observed in the second case.The dynamic risk analysis of the multifractal cross-correlation complexity between the two pollutants shows that although the COVID-19 lockdown contributes to the improve-ment of multifractal cross-correlations between PM_2.5and O₃,their effects are limited from the perspective of intrinsic multifractality.（6）The TS-CSVR method is used to study the multi-scale chaotic complexity as-sessment and prediction of aviation accidents and aircraft hull insurance losses.The empirical analysis shows that the four subsequences all have chaotic characteristics,and there is no obvious linear relationship between the chaotic characteristic parame-ters and the size of the time scale.The risk of chaotic complexity of the EFD-scale series is the lowest.Further prediction results show that the TS-SVR method has better forecasting accuracy.Each CSVR prediction model presents a U-shaped structure,the volatility of the prediction error descends firstly and then ascends.Dividing aircraft accidents into a fine-grained subsequence help increase the prediction accuracy.How-ever,the smallest granularity is not the best.The performance of different granularities needs to be evaluated to find an optimal value.