多机器学习模型下空气质量等级分类预测研究  

Research on Air Quality Classification and Prediction under MultipleMachine Learning Models

作　　者：李玄 刘佳 刘馨语 LI Xuan;LIU Jia;LIU Xinyu(Panzhihua Meteorological Bureau,Panzhihua 617000,China;Fucheng Natural Resources Bureau,Mianyang 621000,China;Renhe Meteorological Bureau,Panzhihua 617000,China)

机构地区：[1]攀枝花市气象局,四川攀枝花617000 [2]涪城区自然资源局,四川绵阳621000 [3]仁和区气象局,四川攀枝花617000

出　　处：《沙漠与绿洲气象》2024年第6期146-153,共8页Desert and Oasis Meteorology

基　　金：高原与盆地暴雨旱涝灾害四川省重点实验室青年专项(SCQXKJQN202211)。

摘　　要：采用2015—2017年攀枝花市污染要素(PM_(2.5)、PM_(10)、SO_(2)、NO_(2)、O_(3)及CO)逐小时实测数据,分季节构建了K近邻模型(KNN)、BP神经网络模型(BPNN)、XGBoost模型和长短期记忆模型(LSTM)对逐小时空气质量指数(Air Quality Index,简称AQI)进行预测分析。结果表明:AQI与各类气象要素以及时间序列都具有良好的相关性,作为特征变量在空气质量等级分类预测中效果良好。模型预测评估方面,LSTM模型预测效果最好,综合准确率达88.1%,XGBoost次之。LSTM模型对污染天气预测效果良好,总体呈现冬季>秋季>春季>夏季。气象要素与空气质量等级相关性呈冬高夏低,能见度与空气质量等级相关性最好,而露点温度、水汽压相关性在夏季较差。本研究通过对比不同季节多种机器学习模型对逐小时AQI等级的预测效果,找出最适宜的模型,为环保部门准确预报、预测空气质量指数,提高环境治理效果提供参考。In this paper,hourly data of pollution factors in Panzhihua from 2015 to 2017are used to predict hourly Air Quality Index(AQI)by constructing K-nearest neighbor(KNN),BP neural network(BPNN),XGBoost,and Long Short-Term Memory(LSTM)models by season.The results show that the AQI has a strong correlation with meteorological elements and time series.In terms of model performance,the LSTM model yields the best prediction results,with an overall accuracy of 88.1%,followed by XGBoost.The LSTM model performs well in predicting pollution events,with the order of winter>autumn>spring>summer.The correlation between meteorological factors and air quality levels is highest in winter and lowest in summer.Visibility shows the strongest correlation with air quality levels,while dew point temperature and vapor pressure exhibit poorer correlation in summer.By comparing the prediction performance of multiple machine learning models in different seasons on hourly AQI level,this study finds out the appropriate model,which provides a reference for the environmental protection department to accurately predict the air quality index and improve the environmental management efforts.

关 键 词：空气质量指数 气象要素 机器学习 长短期记忆模型(LSTM) 

分 类 号：X831[环境科学与工程—环境工程]