Statistical Inference of Dynamic Conditional Generalized Pareto Distribution with Weather and Air Quality Factors

Air pollution is a major global problem, closely related to economic and social development ecological environment construction. data for most regions of China have close correlation with time seasons are affected by multidimensional factors such as meteorology air quality. In contrast classical peaks-over-threshold modeling approaches, we use deep learning technique three new dynamic conditional generalized Pareto distribution (DCP) models weather quality fitting the time-dependence pollutant concentration make statistical inferences about their application in analysis. Specifically, proposed DCP models, autoregressive exponential function mechanism applied time-varying scale parameter tail index distribution, sufficiently high threshold chosen using two selection procedures. The probabilistic properties model maximum likelihood estimation (MLE) investigated, simulating showing stability sensitivity MLE estimations. fit PM2.5 series Beijing from 2015 2021. Real used illustrate advantages DCP, especially compared volatility GARCH AIC or BIC criteria. involving both mixed performs better than other alone. Finally, prediction based on long short-term memory (LSTM) predict concentration, achieving ideal results.