摘要
天顶对流层延迟(Zenith Tropospheric Delay, ZTD)与雾霾之间的相关性,为雾霾的监测、预报提供了新手段.近年来,利用ZTD结合其他因素主要采用多元线性回归模型、多元时间序列ARMAX模型预测PM2.5质量浓度的变化,并取得了一定的成果.本文在此基础上,提出了基于小波分析之后的多元线性回归模型预测PM2.5质量浓度的变化,并利用北京市2016年第355~357天和2018年第50~52天的每小时对流层数据、气象数据以及环保提供的污染物数据,分别建立了每小时PM2.5浓度变化多元回归模型、小波分析之后的多元回归模型、多元时间序列模型(ARMAX).从检验回归模型的4个统计量可知,基于小波分析之后的多元回归模型优于传统的多元回归模型,而ARMAX模型的拟合优度值在多元回归模型和小波分析之后的多元回归模型之间.对以上三种模型进行跟踪验证,并分别应用于预测2016年第358天24 h的PM2.5浓度的变化,从预测图、模型均方根误差RMSE和拟合优度可知:基于小波分析的多元回归模型优于ARMAX模型,而ARMAX模型优于传统的多元回归模型,而三种模型的拟合优度R2都大于0.9这表明能够在小时尺度上预测PM2.5质量浓度的变化.
The change rule between Zenith Tropospheric Delay(ZTD) and haze provides a new means for monitoring and forecasting haze. In recent years, some results have been achieved to predict the change of PM2.5 mass concentration using ZTD combined with other meteorological data with multiple regression model and ARMAX model. On this basis, a multivariate linear regression model based on wavelet analysis is proposed to predict the change of PM2.5 mass concentration. At same time, based on the hourly tropospheric data, meteorological data and pollutant data provided in Beijing for DOYs 355~357 in 2016 and DOYs 50~52 days in 2018, the multivariate regression model, the multivariate regression model based on wavelet analysis and the multivariate time series model(ARMAX) were established respectively to predict PM2.5 mass concentration. From the statistics of regression model,F,P, based on the results of experiments, the results show that the multivariate regression model based on wavelet analysis is superior to the traditional multivariate regression model, while the multivariate ARMAX model is between the multivariate regression model and the multivariate regression model based on wavelet analysis. Based on the above three models to predict the change of PM2.5concentration in 24 h for DOY 358, 2016, the result from the prediction graph, RMSE and goodness of fit, proves that the multivariate regression model based on wavelet analysis is superior to ARMAX model, while the ARMAX model is superior to the traditional multivariate regression model from the forecasting chart, RMSE of root mean square error and goodness of fit. And the goodness of fit of the three models is greater than 0.9, which shows that the change of PM2.5 mass concentration can be predicted on an hour scale.
作者
郭敏
张捍卫
张红利
GUO Min;ZHANG Han-wei;ZHANG Hong-li(School of Surveying and Land Information Engineering,Henan Polytechnic University,Jiaozuo 454003,China)
出处
《地球物理学进展》
CSCD
北大核心
2020年第6期2068-2074,共7页
Progress in Geophysics
基金
国家自然科学基金(41905027,41474021)
河南省科技攻关项目(182102210315)
河南省高校基本科研业务费专项资金资助(NSFRF180406)联合资助。
