基于机器学习的长沙市空气污染物浓度预报研究  

Research on air pollutant concentration forecast in Changsha based on machine learning

作　　者：陈金车 迪里努尔·牙生 王田宇 王金艳[2] 孙彩霞 谢祥珊 冯薇 CHEN Jinche;DILINUER Yasheng;WANG Tianyu;WANG Jinyan;SUN Caixia;XIE Xiangshan;FENG Wei(Lanzhou Meteorological Bureau,Lanzhou 730101,China;College of Atmospheric Sciences,Lanzhou University,Lanzhou 730000,China)

机构地区：[1]兰州市气象局,甘肃兰州730101 [2]兰州大学大气科学学院,甘肃兰州730000

出　　处：《环境保护科学》2022年第4期103-112,共10页Environmental Protection Science

基　　金：国家重点研发计划项目(2020YFA0608402);国家自然科学基金面上项目(41575138)。

摘　　要：基于2014~2019年长沙市6种空气污染物日均浓度的监测数据、同期的气象数据,利用随机森林重要性评估的方法对影响污染物浓度的预报因子进行筛选,构建了基于随机森林算法和支持向量机算法的2种机器学习预报模型对6种空气污染物浓度分别进行预报。结果表明:各污染物浓度预报结果的均方根误差随着AQI指数的增加而变大;经随机森林变量筛选优化之后2种模型对各种污染物浓度的预报准确率都有所提升,且预报准确率都随着预报时效的增大而降低。整体而言,支持向量机回归模型对长沙市空气污染预报具有更强的泛化能力,误差更小。Based on the monitoring data of daily average concentrations of six air pollutants in Changsha from 2014 to 2019 and the meteorological data during the same period,using the random forest importance assessment method to screen the forecast factors affecting the pollutant concentration,two machine learning forecasting models based on random forest algorithm and support vector machine algorithm were constructed to forecast the concentrations of the six air pollutants.The results showed that the root mean square error of the forecasting results for each pollutant became larger with the increase of AQI index.The forecast accuracy of the two models for various pollutant concentrations improved with screening and optimizing the random forest variables.And the forecast accuracy decreased with the increase of the forecast timeliness.Overall,the support vector machine regression model had a stronger generalization ability and less error for the air pollution forecasting in Changsha.

关 键 词：空气污染 预报 随机森林 支持向量机 机器学习 

分 类 号：X51[环境科学与工程—环境工程] P404[天文地球—大气科学及气象学]