Research On Prediction Method Of PM2.5 Concentration Grade Based On Deep Forest

The people’s daily life and work are terribly affected by haze,what’s worse,our country’s sustainable development strategy is also negatively influenced.Particulate Matter PM2.5 is the primary factor that causes the haze,and it has been widely focused by human beings,which becomes the hot topic in recent years.Accurately forecasting Particulate Matter PM2.5 is beneficial to strengthing the people’s prevention consciousness,in addition,it can also offer the data material for handling the Particulate Matter PM2.5.This text aims at the hot topic about Particulate Matter PM2.5 to acquire a concentration class forecasting method of PM2.5 of gcforest.(1)The PM2.5,environment and meteorologic factors in Huangdao are selected as the researching objective,and the experimental data of PM2.5 collected from 2014 to 2019 are acquired by Python,including pollutants(PM2.5,PM 10,NO2,SO2,CO,O3),meteorological conditions(wind,humidity,precipitation,temperature)and PM2.5 concentration data from adjacent sites,which achieves efficient data acquisition and storage.The integrity and consistency of the data set are enhanced through data preprocessing.(2)The PM2.5’s space-time correlation of concentration is explored by analyzing the relation between the changes of PM2.5 concentration at multiple scales and other characteristic factors.Secondly,the lag of PM2.5 and various influencing factors in time series is analyzed by the Time Lag Cross Correlation.Finally,filtering the characteristic factors by the univariate feature selection method and a total of 9 characteristic values were selected,including PM2.5,NO2,SO2,CO,temperature and PM2.5 concentrations from Licang subdistrict site,Chengyang subdistrict site,Sifang subdistrict site,Yangkou subdistrict site,and constitute a time series data set.(3)Concentration class forecasting method of PM2.5 based on random forest and gcforest depend on the time series data.Comparing with the neural network,the model is evaluated by four aspects:precision,recall,F1-Score and Kappa coefficient.The result shows that the concentration class forecasting method of PM2.5 based on gcforest has the comparatively accurate forecast degree.