基于机器学习的人为活动减小条件下中国大陆地区空气污染分析和模拟  

Analysis and simulation of Chinese Mainland air pollution under the condition of reduced human activities based on machine learning

作　　者：马敏劲[1,2] 谈昌蓉 赵侦竹 康国强 MA Min-jin;TAN Chang-rong;ZHAO Zhen-zhu;KANG Guo-qiang(Key Laboratory of Arid Climate Change and Disaster Reduction of Gansu Province,Lanzhou University,Lanzhou 730000,China;College of Atmospheric Sciences,Lanzhou University,Lanzhou 730000,China)

机构地区：[1]兰州大学甘肃省干旱气候变化与减灾重点实验室,兰州730000 [2]兰州大学大气科学学院,兰州730000

出　　处：《兰州大学学报（自然科学版）》2021年第6期743-752,共10页Journal of Lanzhou University(Natural Sciences)

基　　金：中国气象局兰州干旱气象研究所干旱气象科学研究基金(IAM202002);国家重点研发计划重点专项项目(2017YFC1501800)。

摘　　要：分析2019年末新型冠状病毒肺炎疫情爆发以来,在人为活动减小条件下武汉、京津冀、长三角、珠三角等地区的大气污染状况,使用随机森林算法(RFA)对武汉、北京等城市的污染物分布进行模拟.结果表明,疫情前期,京津冀、长三角、关中地区及四川盆地颗粒物污染较为严重,空气质量指数(AQI)增长趋势明显,以PM_(2.5)为首要污染物,自2月开始颗粒物质量浓度呈下降趋势, AQI减小,表明疫情期间各大城市停工停产使空气质量明显变好.武汉疫情期间ρ(PM_(2.5))呈降低趋势, RFA对疫情前后的ρ(PM_(2.5))的模拟效果较好,但对ρ(PM_(2.5))峰值的模拟结果偏低,对于ρ(PM_(2.5))<75μg/m;的时段模拟效果较好. RFA能较好地模拟出ρ(O_(3))的一般性变化趋势,对突发高质量浓度时段的模拟能力较弱.检验结果表明RFA的模拟结果平均拟合度指数高,均方根误差和平均绝对误差较小.Since the outbreak of COVID-19 pandemic at the end of 2019, human activities had been greatly reduced in order to prevent and control the spread of the epidemic. This article mainly analyzed the air pollution in Wuhan, Beijing-Tianjin-Hebei, Changjiang Delta, Zhujiang Delta under the conditions of the reduced human activities, and used a random forest algorithm to simulate pollutants in several cities including Wuhan and Beijing. The results showed that, in the early stage of the epidemic, the pollution of particulates in the Beijing-Tianjin-Hebei, Changjiang Delta, Guanzhong(central Shaanxi) and Sichuan Basin was more serious than the other regions in Chinese Mainland. PM_(2.5)was the primary pollutant,whose decrease since February 2020 made for an obvious positive air quality index growth trend and its negative growth indicated that the suspension of work and production in the major cities had signifi-cantly improved the air quality during the epidemic. After Wuhan started its lock-down, the ρ(PM_(2.5))showed a decreasing trend. Simulation of the algorithm on the ρ(PM_(2.5)) before and after the lock-down was good, but the simulation of the ρ(PM_(2.5)) peak value was lower than real time value. For the time period when the ρ(PM_(2.5)) was lower than 75 μg/m;, the simulation effect was better. Together with ρ(PM_(2.5)),the general trend of ρ(O_(3)) could be simulated, and the simulation ability for the sudden high concentration was weak. The verification results showed that the average index of agreement in the test method was high, while the root mean square error and mean absolute error were both small.

关 键 词：新冠肺炎 空气质量 随机森林算法 

分 类 号：P421[天文地球—大气科学及气象学]