Study On PM_2.5 Pollution Characteristics And The Establishment Of Its Statistical Prediction Model Of Typical Cities In Anhui Province During Winter

Air pollution affects public health seriously and limits socialeconomic development.Grasping the basic characteristics of air pollution and provide timely and accurate forecast results will be helpful to the implementation of pollution control measures to reduce air pollution.In this paper,Hefei,Bengbu and Chizhou are selected as the research objects in Anhui province,because their different climate,topography and economic development characteristics.First,lucubrate the PM_2.5 pollution characteristics in the three cities with meteorological data and pollutant concentration monitoring data during the winter of 2015-2018,including spatial and temporal distribution characteristics of PM_2.5,characteristics of potential source areas and typical case analysis.Then,on this basis,through the threshold analysis and combined with the ensemble empirical mode decomposition algorithm（EEMD）capturing peak and valley values,the prediction effect of the PM_2.5 statistical prediction model is improved step by step.The main conclusions are as follows:（1）The spatial distribution of PM_2.5 concentration during winter in Anhui province is decreasing from north to south,among which Bengbu city has the highest PM_2.5 concentration,Hefei is the second,and Chizhou is the lowest.The interannual change showed an increase before 2016 and a decline afterwards.The PM_2.5 concentration variation characteristics of the three cities are different during winter.Hefei city has a significant decline,Bengbu city is relatively stable,Chizhou city’s fluctuation is big,and have aggravating trend.The daily variation of PM_2.5 in the three cities was bimodal.The morning and evening peaks were at 10:00and 22:00 respectively both in Hefei and Bengbu,and at 12:00 and 20:00 respectively in Chizhou.Chizhou’s lowest daily value appears at 7:00 in the morning.（2）The pollution characteristics of the three cities are quite different.Blocked by urban buildings,Hefei city’s wind speed is small,and the wind direction is relatively divergent during winter.The potential sources of PM_2.5 pollution are large and scattered,most of which come from the north and southwest,and a small part from the southeast coast in Hefei.In addition to wind speed and wind direction,Hefei city’s pollution characteristics is greatly influenced by human activities and stability,Hefei have the urbanization characteristics obviously.The dominant wind in Bengbu is the northeast wind,followed by the east wind.Under the condition of gentle breeze,the concentration of PM_2.5 in Bengbu was high in all directions,and local emissions accumulated significantly.When the wind was above gentle breeze,Bengbu city has obvious long-distance transportation in NE.The change of PM_2.5 concentration in bengbu city is greatly affected by the change of weather situation,boundary layer height and ventilation rate.Affected by the topography,Chizhou city’s dominant wind frequency（NE）is more than40%,and the wind speed is relatively high.The concentration of PM_2.5 is less affected by human activities,and is mainly influenced by the north and southwest airflow.Massive,continuous rainfall could completely remove pollutants.（3）After establishing the statistical forecast model for the above three typical cities,it is found that the method of threshold analysis proposed in this paper to filtrating the prediction factors can effectively improve the prediction effect of the model.The MAE,MAPE and RMSE of BP model were reduced by 30～34%,34～42%and 30～34%respectively,and IA and R were increased by 11～12%and 20～22%respectively compared with filtrated by Pearson correlation coefficient.In addition,the EEMD-BP hybrid model is more accurate in capturing the peak and valley values of PM_2.5 concentration compared with the BP model.Compared with the multiple regression（MLR）model commonly used in the business,the MAE、MAPE and RMSE of the EEMD-BP hybrid model were found to be reduced by 46～55%、16～49%and45～49%respectively,and R and IA were improved by 17～23%and 28～32%respectively.The optimization effect of the EEMD-BP hybrid model is significant,especially for high PM_2.5concentration prediction.To sum up,the EEMD-BP model shows good stability and strong generalization ability in the application of PM_2.5 prediction in the three cities.