基于机器学习的郑州市大气PM_(2.5)与O_(3)浓度预测方法及气象因子的影响分析  

Prediction Method and Meteorological Factors Impact Analysis of Atmospheric PM_(2.5) and O_(3) Concentration in Zhengzhou City Based on Machine Learning

作　　者：张容硕 谢沛远 陈宏飞[2] 杨清荣 关民普 马南 尉鹏[1] 朱仁成[1,2] ZHANG Rongshuo;XIE Peiyuan;CHEN Hongfei;YANG Qingrong;GUAN Minpu;MA Nan;WEI Peng;ZHU Rencheng(Atmospheric Environment Institute,Chinese Research Academy of Environmental Sciences,Beijing 100012,China;School of Ecology and Environment,Zhengzhou University,Zhengzhou 450001,China;Technology Center for Ecology and Environment of Henan Province,Zhengzhou 450046,China)

机构地区：[1]中国环境科学研究院大气环境研究所,北京100012 [2]郑州大学生态与环境学院,河南郑州450001 [3]河南省生态环境技术中心,河南郑州450046

出　　处：《环境科学研究》2024年第3期469-478,共10页Research of Environmental Sciences

基　　金：河南省自然科学基金面上项目(No.232300421243);郑州大学2023年度基础研究(自然科学)培育项目(No.32213970/23)。

摘　　要：近年来,我国面临着细颗粒物(PM_(2.5))污染形势依然严峻以及臭氧(O_(3))污染日益凸显的双重压力.为进一步准确预测郑州市大气PM_(2.5)与O_(3)浓度并探明气象因子的影响,本研究使用2018−2022年郑州市大气污染物和气象因子逐时数据,结合统计学单因素分析和机器学习LightGBM模型多因素分析,建立了一种基于长时间序列数据的PM_(2.5)与O_(3)浓度预测及气象因子影响分析的综合分析方法.结果表明:①训练后的LightGBM模型能够较好地预测PM_(2.5)污染,准确率达80.8%;对O_(3)污染预测的准确率为52.5%.②郑州市大气PM_(2.5)浓度与气压呈正相关,与比湿和环境温度均呈负相关;大气O_(3)8 h滑动平均浓度(O_(3)-8 h浓度)与比湿和太阳辐射均呈正相关,与气压呈负相关.③有利的气象条件可能是2021年PM_(2.5)年均浓度得到显著改善的重要因素;同时,不利的气象条件也促使2021年和2022年6月O_(3)月评价值(O_(3)日最大8 h滑动平均90百分位浓度)有所上升.研究显示,这种基于长时间序列的综合分析方法适用于大气PM_(2.5)与O_(3)浓度的气象因子影响分析,也能有效预测PM_(2.5)与O_(3)的浓度.In recent years,China has faced the dual challenges of the continued serious fine particulate matter(PM_(2.5))pollution and increasingly prominent ozone(O_(3))pollution.To further accurately predict and explore the meteorological factors influence the atmospheric PM_(2.5)and O_(3)concentrations in Zhengzhou City,this study applied hourly data on air pollutants and meteorological variables in Zhengzhou City from 2018 to 2022.A comprehensive method for predicting PM_(2.5)and ozone concentrations and analysing the impact of meteorological factors based on long-term time series data was established.The findings highlight the key meteorological factors that have the greatest impact on both PM_(2.5)and ozone 8 h concentration average concentrations(O_(3)-8 h).The results showed that:(1)The welltrained LightGBM model showed good performance in predicting PM_(2.5)pollution with 80.8%accuracy,while the accuracy of predicting O_(3)pollution was 52.5%.(2)The PM_(2.5)concentration in Zhengzhou City shows a positive correlation with air pressure and a negative correlation with specific humidity and temperature;The concentration of O_(3)-8 h is positively correlated with humidity and radiation,and negatively correlated with air pressure.(3)Favorable meteorological conditions may be a significant factor contributing to the notable improvement in the annual average concentration of PM_(2.5)in 2021.Simultaneously,adverse meteorological conditions have led to an increase in the monthly evaluation values of O_(3)(90th percentile concentration of the maximum 8-hour sliding average of O_(3))during June 2021 and 2022.The research shows that this comprehensive analysis method based on long-term time series can be applied to the analysis the impact of meteorological factors in atmospheric PM_(2.5)and O_(3)concentrations.In addition,it can also effectively predict the concentrations of PM_(2.5)and O_(3).

关 键 词：大气污染防控 细颗粒物(PM_(2.5)) 臭氧(O_(3)) 气象因素 机器学习 郑州市 

分 类 号：X511[环境科学与工程—环境工程]