基于Stacking的地面PM2.5浓度估算

为了解决地面PM_2.5监测网络在空间和时间覆盖受限的问题，提出了基于宽时空覆盖的卫星气溶胶光学厚度AOD，利用Stacking方法建立地面PM_2.5浓度估算模型，将AOD、PM_2.5和各气象参数以及与PM_2.5排放有关的数据进行训练，使用改进网格搜索对模型超参数进行优化，通过对多重共线性分析，建立基于Stacking的最优PM_2.5浓度估算模型。选取2016-01-01-2016-12-31的数据作为实验对象，结果表明:相比于随机森林、GBRT和XGBoost 3种模型，使用岭回归作为元学习器的Stacking模型性能更优，可见Stacking适用于大范围地理区域的大气污染监测。

APPLICATION OF STACKING IN GROUND-LEVEL PM2.5 CONCENTRATION ESTIMATING

In order to solve the problem that the ground PM_2.5 measurement was limited in spatial and temporal coverage, satellite aerosol optical depth AOD with wide spatial-temporal coverage and stacking method were proposed to establish a PM_2.5 concentration estimation model. The AOD and meteorological parameter and PM_2.5 emissions related data were trained for building the model, and the improved grid search algorithm was used to optimize the hyperparameters of each model. Based on multi-collinearity analysis, the optimal PM_2.5 concentration estimation model based on Stacking was established. The data between January 1st, 2016 to December 31st, 2016, was selected as the experimental object. The experimental results showed that the performance of the stacking model using ridge regression as the meta-learner was better than that of the random forest, GBRT and XGBoost model. It was concluded that the stacking model was applicable for air pollution monitoring in a large geographical area.