函数型数据下广义分位数回归模型的PM_(2.5)分析  

Analysis of PM_(2.5) Based on Generalized Quantile Regression with Functional Data

作　　者：郑禹 罗双华[1] 周锦涛 ZHENG Yu;LUO Shuanghua;ZHOU Jintao(School of Science,Xi’an Polytechnic University,Xi’an 710600,China)

机构地区：[1]西安工程大学理学院,西安710600

出　　处：《湖北民族大学学报（自然科学版）》2023年第3期375-380,共6页Journal of Hubei Minzu University:Natural Science Edition

基　　金：陕西省自然科学基金项目(2020JM571,2021JM002)。

摘　　要：为了得到PM_(2.5)浓度更稳健、更精准的预测结果,将广州市PM_(2.5)浓度由离散数据化为函数型数据,引入温度、湿度、风级3个变量并结合广义分位数回归模型来研究2022-2023年广州市PM_(2.5)问题。利用季节项分解、函数型主成分分析等方法,得出广义分位数函数,以此对未来时刻PM_(2.5)浓度值进行预测,并利用函数型主成分分析描述PM_(2.5)浓度函数的动态特征。分析结果显示,温度、湿度对PM_(2.5)浓度有显著影响。将离散数据化为函数型数据后进行分析将极大提高数据的稳定性,并能更好地揭示PM_(2.5)的规律和特征;在分位点τ=0.8时的广义分位数函数能使最终预测值与真实值拟合最好,4个主成分函数即可概述复杂多变且高维的PM_(2.5)实时浓度。该研究达到了分析、解释、预测的目的,能为出行及PM_(2.5)防控等提供参考。To get a more robust and accurate prediction of PM_(2.5) concentration,we transform the discrete data of PM_(2.5) concentration in Guangzhou into functional data,and introduce the temperature,humidity,and wind intensity as well as the generalized quantile regression model to study the PM_(2.5) problem in Guangzhou in 2022-2023.The generalized quantile function was obtained by using seasonal decomposition and functional principal component analysis.Based on this,the future PM_(2.5) concentration is predicted,and functional principal component analysis is used to describe the functional dynamic characteristics of PM_(2.5) concentration.The results showed that temperature and humidity had significant effects on PM_(2.5) concentration.Transforming discrete data into functional data for analysis will greatly improve the stability of data and better reveal the rule and characteristics of PM_(2.5).The predicted values of the generalized quantile function withτ=0.8 fit the real values best.The four principal component functions can summarize the complex,variable,and high dimensional real-time concentration of PM_(2.5).It achieves the purpose of analysis,interpretation,and prediction,and provides a reference for travel and PM_(2.5) prevention and control.

关 键 词：函数型数据 广义分位数回归 PM_(2.5) 温度 湿度 风级 

分 类 号：O212.1[理学—概率论与数理统计]