Influencing Factors And Prediction Of Urban Haze In Beijing,Tianjin And Hebei

The haze weather has occurred frequently in recent years,and its damage is extremely great.Therefore,the study of the causes of haze and the effective prediction of haze have become the research hotspots in the current scientific field.This paper studies the influencing factors affecting the smog in the 13 prefecture-level cities of Beijing-Tianjin-Hebei from air pollution factors,meteorological factors and spatial distribution,and finally makes effective predictions for the future smog weather,which can be smoggy,and finally achieve a reasonable plan to prevent and control smog.Firstly,this paper analyzes the temporal and spatial evolution of the smog in Beijing-Tianjin-Hebei by spatial analysis and Kriging interpolation,and finds the distribution of time and space of smog.From the perspective of space,there are obvious regional differences in smog pollution as "high pollution in the southern region and low pollution in the northern region",from the time point of view,there are obvious seasonal,monthly,and daily changes,as well as "non-working day effect" and "Spring Festival effect".Secondly,the global spatial correlation and local spatial correlation of PM2.5 were quantitatively studied by Moran’I index and Local Moran’s I index.PM2.5 was found to have significant spatial aggregation and spatial dependence,and the phenomenon in winter is particularly obvious.Thirdly,the paper analyzes the direct and indirect effects of various impact factors on PM2.5 through the PLS1 model of PM2.5.It is found that each impact factor not only acts directly on PM2.5,but also acts indirectly through other influencing factors.PM2.5,indirect effects can not be ignored.Finally,the BP neural network model and the deep learning model were built respectively to predict whether there will be haze weather in the next 24 hours.The results show that the accuracy of the latter model is 88.66%,the precision is 90.26 %and the recall rate is 81.36%,better than that of the former model.This not only shows that deep learning has better characterization characteristics,but also that smog is a multi-factor interaction,and there is a very strong nonlinear influence relationship.