Study On Prediction Of Roadside PM_2.5 And Ozone By Using Back Propagation Neural Network

With the development of economy and the continuous progress of urbanization,the pressure on the urban environment in China is gradually increasing.Among the air pollutants,fine particulate matter(PM_2.5)and ozone（O₃）have become the main objects of attention due to their greatly harm to human health and their high concentration levels.Motor vehicle emission is one of the major contributors of PM_2.5 and ozone.Motor vehicles mainly drive in densely populated areas in cities,and the number of motor vehicles keeps increasing.The concentrations of pollutants in roadside areas is generally higher than the average concentrations in cities.When people are near roads,they are likely to be directly exposed to highly polluted areas,which has adverse effects on health.Therefore,it is of great significance to predict the concentration of pollutants near roads accurately.This could help people understand the pollutant concentration in advance and reduce the exposure of pollutants.In this paper,PM_2.5.5 and O₃ are taken as the research objects to establish a 1-3 hour BP neural network model.The research time is from 0:00 on May 13,2014 to 23:00 on September30,2015.The input parameters mainly include meteorological parameters,traffic parameters,roadside pollutant concentration parameters and pollutant concentration parameters of national monitoring stations.Also,this paper proposes a systematic BP neural network model optimization scheme,which optimizes the maximum training times,activation function,number of hidden layer nodes and other important parameters in BP neural network model.Based on the preliminary optimization of the model,this study further proposes a stepwise screening method,which optimizes the input parameter group of the model,and eliminates redundant or irrelevant input parameters.The results show that the performance of the model is significantly improved by the above optimization scheme.Compared with the previous one,the prediction R² of the model after the input parameter optimization increases by 5.63%-87.4%,and the R² range of the 1-hour prediction model is 0.806-0.880,the R² range of the 2-hour prediction model is 0.611-752,and the R² range of the 3-hour prediction model is 0.440-0.657.Based on the BP neural network model and the proposed optimization scheme,this paper further studies the 1-3 hour prediction of PM_2.5 and O₃ concentration in national monitoring stations.The input parameters are meteorological parameters and pollutant concentration parameters of national monitoring stations,and the time is from 0:00 on May 13,2014 to 23:00on September 30,2015 and 2017-2018 respectively.In the optimized prediction results,for PM_2.5 and O₃,R² range of 1-hour prediction model is 0.961-0.965 and 0.936-0937,R² range of2-hour prediction model is 0.854-0.885 and 0.773-0.799,R² range of 3-hour prediction model is 0.754-0.813 and 0.624-0.440,respectively.Finally,different traffic control policies are simulated,in which traffic flow is reduced by 25%,50%,75%and 100%,respectively.The impact of traffic control policies on road pollutant concentration is discussed.The results show that the traffic control policy has certain effect on the reduction of pollutants.For example,when the traffic flow is reduced by 50%,the average reduction range of the pollutant concentration of the two roads is 3.32%-5.94%,and the increase range of the proportion of standard hours is 3%-18%;under the 100%reduction scenario,the average reduction range of the pollutant concentration is 13.6%-20.38%,and the ratio of standard hours is 13.6%-20.38%The increase was 21%-55%.The simulation results also show that the role of a single traffic control measure is still limited,and it is necessary to coordinate the control of other pollution sources when the road pollutants exceed the standard.