AQI Time Series Feature Analysis Based On Additive Regression Model

Air quality index（AQI）is a comprehensive index to measure regional air environment.It can be obtained by calculating the Individual Air Quality Index（IAQI）of SO₂,NO₂,CO,O₃,PM₁₀,and PM_2.5.In this paper,the time series analysis method is used to study the AQI daily mean historical data of 12 automatic air stations in Beijing from May 2014 to December 2017.And using the Prophet program creativity to predict the change of AQI,which was based on additive regression model and usually used to economic and business prediction.Through data sorting and processing,defining models,training models,building prediction sets and forecasting,we analyze the changing rules of research objects.Research on the prediction of AQI can help environmental management departments to formulate environmental protection decisions and pollution prevention measures.It can also provide reference for people’s life and work,thereby reducing the harm caused by air pollution to human body.From a geographical point of view,the growth trend of AQI in the northern areas of Beijing was smaller than that in the southern areas.From the analysis of time,Beijing’s AQI shows a declining trend gradually,which showed the first effect of air pollution prevention and control;Holidays have a great impact on AQI.Among them,AQI was increased by human factors（fireworks and firecrackers）before and after the Spring Festival,and decreased during the Qingming Festival.In the year AQI growth showed a winter>spring>autumn>summer rules.The weekly trend showed a continuous increase from Monday to Friday,and began to decline on Saturday and Sunday,which is conformed to the law of people’s work and rest.This shows that AQI change is not only influenced by climate and meteorological conditions,but also closely related to human factors.After forecasting the trend of AQI from January to March 2018,it was verified by the measured values that the average prediction accuracy of MAE is 46.3,and the model has a better prediction effect on the overall trend of the research objects.However,the handling of unexpected events which do not meet the historical rules will lead to a larger deviation of the forecast results.It is necessary to optimize the whole model with multiple factors in order to improve the accuracy of prediction.