Study On PM2.5 Prediction Of Beijing Subway Based On Bp Neural Network

With the development of the urbanization process,the congestion of ground transportation is increasing.Due to its advantages of punctuality,speed and convenience,the subway has developed rapidly in recent years.At the same time,the air quality in subway stations has gradually attracted people's attention.Compared with outdoor,subways are densely crowded and poorly ventilated,and related studies have shown that PM2.5 in subways is rich in metal elements and more toxic.PM2.5 in the subway should be valued.At present,the research on PM2.5 in subway is mainly based on the actual measurement,but there are some limitations in the separate actual measurement research,the research results can not be directly applied to the air quality regulation in the subway station.Therefore,this paper combines the actual measurement research with the prediction research,establishes the prediction model of PM2.5 concentration in the subway station through the actual measurement data of the subway station.It provides the reference for the next step of adjusting the PM2.5 concentration in the station and improving the air quality in the station.In this paper,the measurement of PM2.5 concentration in multiple stations in Beijing subway is carried out.The distribution rule of PM2.5 concentration in subway stations is studied.The influencing factors of PM2.5 concentration in subway stations are analyzed.Using SPSS software to quantitatively analyze the possible influencing factors of the PM2.5 concentration of the platform and hall,the following conclusions were drawn:(1)There is a correlation between the PM2.5 concentration at the subway station platform and the outdoor PM2.5 concentration.The PM2.5 concentration at the subway station station hall is also correlated with the outdoor PM2.5 concentration.Compare the correlation,The PM2.5 mass concentration at the station hall is greater than the PM2.5 mass concentration at the platform.(2)The PM2.5 concentration in subway stations is affected by the PM2.5 concentration in the tunnel.Different types of platform doors have different barrier effects on PM2.5 in the tunnel.Therefore,the type of platform door will affect the PM2.5 concentration at the platform.(3)When the outdoor PM2.5 concentration is at the same pollution level for two consecutive days,the PM2.5 concentration of platform and station hall is measured,and it is found that the PM2.5 concentration of platform and station hall is correlated with the PM2.5 concentration of the previous day,indicating that there is a cumulative PM2.5 concentration of platform and station hall.Based on the above-mentioned measured data and research results,imitating the weather prediction model,this paper establishes a PM2.5 concentration prediction model in a short-term subway station at intervals of one day and a short-hours prediction.model at intervals of several hours.The prediction errors of various prediction models of PM2.5 concentration in subway station based on BP neural network are calculated,and the causes of the errors are analyzed.The main conclusions obtained are as follows.(1)The prediction error of the short-term subway platform PM2.5 concentration prediction model is 10.58%,and the prediction error of the short-term subway hall PM2.5 concentration model is 12.62%.Based on the actual measured data of different subway stations,the models of platform and station hall prediction for each station were established.Yong'anli Station have the half height safety door sysytem.Pingleyuan Station and Shilipu Station have the full height safety door sysytem.The errors of the platform of Yong'anli Station,Pingyuan Station,and Shilibao Station station prediction model are 10.49%,3.15%,and 15.72%,respectively;the station hall prediction models of Yong'anli Station,Pingleyuan Station,and Shilipu Station are 6.68%,15.46%,16.55%.(2)During the operation of the subway,according to the passenger flow,the subway operation period is divided into the peak period of passenger flow and the low peak period of passenger flow.During the low peak period of passenger flow,the short-hours prediction model of the subway station platform with a delay of 5h is established,and the model prediction error is 10.32%.The short-hours prediction model of the subway station platform with a delay of 8h is established,and the model prediction error is 15.34%.During the peak period of passenger flow,a prediction model for PM2.5 concentration during the late peak hours of the platform was established,and a prediction model for PM2.5 concentration during the early peak hours of the platform was established.The errors of these two models were 17.95% and 16.95%.(3)By comparing the prediction models of station platform and hall,it found that the accuracy of the prediction model of station platform with half height safety door is lower than that with full height safety door,and the prediction error of the prediction model of station hall is just the opposite.According to previous studies,for the prediction of PM2.5 concentration in subway stations,the prediction error based on theoretical derivation is 20%,and the prediction error based on various data-driven methods is greater than 20%.The prediction errors of various PM2.5 concentration prediction models of subway stations based on BP neural network are less than 20% in this paper,which shows that it is feasible to establish PM2.5 concentration prediction models in subway stations based on BP neural network.The research of this subject is of great significance to the health of subway staff,and also to the health of passengers taking the subway.A preliminary prediction model has been established for the prediction and warning system of air quality adjustment in subway stations,which can be used as the concentration of subway PM2.5.This paper provides a reference for the follow-up related research of subway PM2.5 concentration.The paper has 55 figures,17 tables and 66 references.