Research On The Prediction Of PM_2.5 Daily Average Concentration In Nanjing Based On The ARIMA-BP Neural Network Model And Its Optimized Model

With the economic development and the acceleration of urbanization,the increase in the discharge of various pollutants has caused a huge impact on the atmospheric environment.Air pollution has become one of the issues of widespread concern in today's society.PM_2.5,as the main pollutant of the atmospheric environment,has negative effects on people's health,transportation,and environmental quality.Because of its long transmission distance and long stay time,PM_2.5 is often used to evaluate air quality standards.Studying the daily average concentration change of PM_2.5 is conducive to strengthening the understanding of the PM_2.5 concentration change rule,influencing factors,and future development,and laying a foundation for pollution forecasting,prevention of respiratory diseases,and treatment of the atmospheric environment.Based on this,we mainly study the following three contents:1.We analyzed the influencing factors of the daily average PM_2.5 concentration in Nanjing.First,we selected the daily average PM_2.5 concentration data in Nanjing and observed its concentration change.We established a regression model of the daily average concentration of PM_2.5 in Nanjing and other pollutant gases such as SO₂,NO₂,CO,and O₃.Besides,we established a regression model of the daily average concentration of PM_2.5 and meteorological factors such as maximum temperature,precipitation,minimum temperature,wind,and weather conditions and applied the principal component analysis to obtain the contribution of the influencing factors.The results show that the daily average concentration of PM_2.5 in Nanjing is the lowest in summer and the highest in winter and the trend is the same in the past several years.The daily average concentration is positively correlated with SO₂,NO₂,CO,and the classification of weather conditions?including four weather states:sunny,cloudy,overcast,and rain?,and negatively correlated with O₃ and meteorological factors such as maximum temperature,precipitation,minimum temperature,and wind.2.We established an ARIMA-BP neural network combination model to predict the daily average PM_2.5 concentration in Nanjing.First,we applied the ARIMA model to build a time series model and make predictions(the linear part of PM_2.5 daily average concentration prediction).The prediction results met the corresponding range of the air quality level.Then,the BP neural network was used to predict residual part?the nonlinear part?.Results show that the PM_2.5 daily average concentration predicted by the combined model has significantly improved the prediction accuracy of the original time series model.3.We established ARIMA-two-layer BP neural network prediction model to optimize the combined model.We analyzed the reason for the gap between the combined model and the actual data and optimized it by increasing the correlation between the input and output layers of the neural network.First,we established a time series model of the input variables and used BP neural network to construct its fitting curve,and it is the first layer of the neural network.Then,we used the first layer of network fitting values as the input layer and the real PM_2.5 daily average concentration as the output layer to establish the second layer of the neural network.The results show that the optimized model has improved prediction accuracy compared to the original combined model.