Evaluation And Prediction Of Air Quality In Changsha Zhuzhou Xiangtan Area

Air,like water,is a necessity for our survival.In this paper,python crawler is used to obtain the air data and meteorological data of Changsha Zhuzhou Xiangtan area from December 1,2013 to December 31,2020.After preprocessing the data,Pearson correlation analysis and principal component analysis are used to evaluate the air quality in Changsha Zhuzhou Xiangtan area,and a variety of machine learning models are used to predict the air quality in Changsha Zhuzhou Xiangtan area.The summary is as follows:1.The number of days of various grades in Changsha Zhuzhou Xiangtan city is mainly left skewed distribution,and the number of days higher than light pollution is less.In the observation period,the excellent and good rate of Zhuzhou City was the highest,which was 76.6%.In each year,the number of days of AQI reaching the standard will reach the highest in 2020.2.The monthly average AQI of Changsha Zhuzhou Xiangtan city shows the characteristics of high pollution degree in spring and winter,and relatively low pollution degree in summer and autumn.The annual average of AQI shows a decreasing trend year by year,but there is a weak rebound in 2017 and 2019.3.PM_2.5and PM₁₀in Changsha Zhuzhou Xiangtan area seriously exceeded the concentration limit in autumn and winter;NO₂and O₃in various cities exceeded the concentration limit in some time in autumn and winter;while CO and SO₂were relatively stable and basically did not exceed the concentration limit.4.In terms of prediction accuracy,Random Forest model and XGBoost model are improved by 3.63%and 4.79%respectively compared with SVM.XGBoost model has better prediction accuracy.In the time of predict,Random Forest model is 2.4S shorter than SVM model,and XGBoost model is 1.9s shorter than SVM model.In terms of macro indicators,due to the limitations of a single model,SVM performs generally in three macro indicators,while random forest and XGBoost model perform well in all indicators.In terms of cost sensitive errors,XGBoost model performs best,followed by random forest model and SVM model.