基于多变量随机森林模型的空气污染过程分析及应用

Analysis and Application of Air Pollution Process Based on Multivariate Random Forest Model

在深入大气污染防治、改善人居环境的大背景下,针对传统空气质量监测站受地理位置和数量限制,无法对整体区域的污染程度及过程进行全面评估的问题,利用随机森林算法,模拟多种污染物浓度的时空分布,选择南京市北部区域进行空气污染过程研究和实践。结果表明,建立的多变量随机森林模型性能良好,模拟近地面PM_(2.5),PM_(10),NO_(2),O_(3)时均浓度的R^(2)值分别达到0.82,0.85,0.77和0.89,利用模型结果可更直观地展现区域污染变化过程,辨别观测区污染来源,为合理开展污染防治工作提供依据。

Under the background of in-depth air pollution control and environmental improvement of human settlements,the traditional air quality monitoring stations are limited by geographical location and number,and cannot comprehensively evaluate the pollution degree and process of the whole area.The random forest algorithm was used to simulate spatiotemporal distribution of the concentration of various pollutants,and applied to the research and practice of air pollution process in the northern area of Nanjing.The results showed that the established multivariate random forest model had good performance,the hourly estimations of PM_(2.5),PM_(10),NO_(3),and O_(3)near the ground gave average model-fitting R^(2)values of 0.82,0.85,0.77,and 0.89,respectively.The model results could be used to show the process of regional pollution changes intuitively,identify the sources of pollution in the observation area,and provide a basis for reasonable pollution prevention and control.