| In recent decades,with the rapid development of China’s modernization,the air quality is deteriorating,which hinders the sustainable development of society.The air quality of a city affects the comprehensive competitiveness of the whole city.Due to the improvement of people’s quality of life,higher requirements are put forward for the living environment.The continuous deterioration of air quality will cause serious harm to people’s health.Improving air quality has become the main work of our country at present.The important function of air quality index is to judge the air condition.The accurate forecast of air quality not only allows the public to understand the future air conditions in advance,and provides convenience for travel,but also help environmental protection departments to formulate air pollution prevention and control measures.In this context,this paper uses correlation analysis to study the relationship between meteorological,Socio-economic factors and air quality.BP neural network is used to set up the classification prediction model to realize the classification prediction of air quality grade in Nanchang City;whale algorithm and Grey relational analysis are introduced to improve the model to improve the measurement accuracy of the model.The central contents of this dissertation are summarized as follows:1.The air quality and meteorological data of Nanchang City from January 2016 to May 2019 are selected,Pearson correlation method and Grey relational analysis are used to study the correlation between air quality pollutants and between pollutants and meteorological factors.Based on the social economy data of Nanchang from 2003 to2018,the random forest features were extracted,and the important factors in social economy indicators were used for stepwise regression analysis to determine the economic factors that have the greatest impact on the air quality level of Nanchang.2.The preprocessed data are divided into four new data sets according to the seasons,and the data are input as BP neural network model respectively,and the training result of this model is determined.The Grey relational analysis and whale algorithm are introduced to improve BP neural network,and compared with other classification models to determine the advantages and disadvantages of the model used in this dissertation.3.Using the PM10 concentration from April 1,2019 to May 21,2019,the auto regressive XGBoost time series prediction model is used to predict the PM10 concentration of the primary pollutant to analyze the accuracy of PM10 prediction.Through the analysis of the experimental results,it is found that among the influencing factors of air quality,the correlation coefficient between PM2.5 and PM10 is the highest,reaching 0.896,and there are different degrees of correlation between other pollutants;the correlation degree between meteorological influencing factors and air quality index is different in different seasons,and the correlation degree of relative humidity is higher in different seasons;PM10 has the highest grey correlation with air quality index,and relative humidity and maximum wind speed are the most important meteorological factors,and the correlation degree is 0.93 and 0.90 respectively.Per capita GDP is the main Socio-economic factor affecting air quality.The improved BP neural network based on GRA-WOA proposed in this dissertation increases the accuracy of classification and prediction by about 20% compared with the simple BP neural network,reaching 95.19%.Compared with other classical classification algorithms,the classification and prediction performance of the model presented in this dissertation is better.The predicted trend of PM10 Concentration Based on the Auto regressive XGBoost time series prediction model is basically consistent with the real change,and the relative error for the next 10 days is less than4.5%.This work not only enriches the optimization theory of BP neural network algorithm,but also provides a theoretical basis for air quality classification and prediction. |
