基于随机森林算法的石家庄市首要污染物分类预报技术研究

Primary Pollutants Classification and Forecast Based on Random Forecast Algorithm in Shijiazhuang

基于2018年1月-2022年6月石家庄市逐日首要污染类型数据和ERA5逐6 h再分析气象要素资料,构建了机器学习所需的多维特征量数据集,并利用随机森林算法学习训练,得到石家庄市首要污染物分类预报最佳模型,实现了不同气象条件下首要污染物分类识别及预报。结果表明,随机森林模型预报首要污染物分类准确率达到76%,对PM10、PM2.5首要污染物分类结果最好,召回率达到93%、89%,O3首要污染物次之,召回率为74%。与中国气象局下发的空气质量指导产品(CMA-ZD)和国家级雾霾数值预报业务系统产品(CUACE)相比,预报准确率分别提升11%、36%,明显优于指导产品。

Based on the daily primary pollution type data and ERA5 reanalysis of meteorological elements data for 6h from January 2018 to June 2022,this paper constructed the multidimensional eigenquantity data set required by machine learning and realized the classification and prediction of primary pollutants under different meteorological conditions by random forest algorithm to learn training.In the results,the daily temperature at 14:00,sequence,the average relative humidity were important characteristic quantities.The classification accuracy of primary pollutants predicted by random forest model reached 76%and the recall rates of PM 10,PM 2.5 and O 3 reached 93%89%and 74%.The accuracy of the model is 11%and 36%higher than that of the Air Quality Guidance Product(CMA-ZD)and the National Haze Numerical Forecast Business System Product(CUACE)issued by the China Meteorological Administration.