Analysis And Prediction Of Harbin Air Quality Based On Optimized BPNN

With economic development and population growth,air pollution is becoming increasingly serious.The use of scientific and accurate methods to predict and an-alyze the air quality in Harbin is of great significance for helping the government formulate corresponding emergency plans and ensuring the normal production ac-tivities of the people.This article uses descriptive statistics to analyze the temporal changes in air quality from 2017 to 2021.The results showed that the concentration changes of PM_2.5、PM₁₀、SO₂and CO showed a“U-shaped”trend consistent with AQI every year,while NO₂remained within a certain concentration range throughout the year and showed a fluctuating trend,while O₃showed an inverted“U-shaped”trend.Secondly,the correlation coefficient method is used to analyze the correla-tion between AQI and air pollutants,and select representative pollutants for the prediction of air quality.The results show that PM_2.5、PM₁₀and AQI have the strongest correlation.Among them,PM_2.5、PM₁₀、CO、NO₂and SO₂are positively correlated with each other,while O₃is negatively correlated with the other five in-dicators.At the same time,based on the lasso regression,the correlation between AQI and twelve social factors was studied,and the regression model was built to obtain the most influential social factors on air quality.The results showed that the energy consumption of industrial added value and public green space area had a significant impact on air quality index.Then,in response to the problem that the traditional grey correlation analysis method did not objectively assign weights to the comparison sequence when calculating the correlation degree between the refer-ence sequence and the comparison sequence,this article adopts the improved entropy weight method to correct its shortcomings in weight determination,and then forms the entropy weight grey correlation degree to search for seasonal main air pollutants in Harbin,Provide representative indicators for the comparative empirical analysis of air quality prediction models in this article.The results show that PM_2.5,PM₁₀and O₃are the three main pollutants affecting the air quality of Harbin in the four seasons,PM₁₀is the main air pollutant in autumn,and PM_2.5is the main pollutant in winter;The air pollutant affecting spring and summer is O₃.Finally,this arti-cle constructs a new multi-scale prediction model based on BP neural network for the first time to predict PM_2.5、PM₁₀and AQI.Firstly,the BP neural network is used as a single time series prediction model.Then,PSO is used to optimize the weight matrix and threshold of the BP network model.CEEMD is introduced to decompose the time series,and the final prediction value of the CEEMD-PSO-BPNN multi-scale model is obtained by integrating the prediction results of intrinsic mode components.The two models are compared and evaluated in terms of model pre-diction performance,training time,and generalization ability.The results indicate that the CEEMD-PSO-BPNN model established for the first time in this article has better performance in predicting air quality.In terms of the average absolute er-ror of CEEMD-PSO-BPNN and the single BP model for predicting AQI,PM_2.5and PM₁₀,they decreased by 60.49%,53.81%and 55.246%respectively.At the same time,the multi-scale prediction model basically fits the trend and fluctuation of the concentration of air pollutants,and can be used as an effective prediction model in the air field.