| 摘要: |
| 为进一步提高PM2.5浓度预测精度，以2016～2021年逐小时地面气象数据和空气污染物数据，构建K近邻算法（KNN）、支持向量机（SVR）、BP神经网络（BPNN）、随机森林（RF）、极限梯度提升（XGBoost）等5个PM2.5浓度预测模型，进行PM2.5预测精度和稳定性评估。结果表明：在5种PM2.5浓度机器学习模型预测中，XGBoost模型的预测精度优于其他模型，其R2、MAE、RMSE、MAPE分别为0.89、0.24 μg/m3、0.11 μg/m3、1.6%。XGBoost模型在不同季节预测PM2.5浓度结果有差异，春季预测精度最高，秋季次之，夏季最低。其中春季R2、MAE、RMSE、MAPE分别为0.88、0.26 μg/m3、0.12 μg/m3、1.32%。气象因子重要性评价结果表明，能见度、降水量对PM2.5浓度预测模型精度贡献显著；污染物浓度重要性评价结果表明，CO、NO2浓度对模型精度贡献显著。本研究可为PM2.5浓度精细化预测提供参考，对于大气污染防控和环境监测管理具有现实意义。 |
| 关键词: PM2.5 预测模型 机器学习 XGBoost 气象因子 |
| DOI： |
| 分类号:X513 |
| 基金项目:云南省气象局青年科技创新团队；云南省气象局科研项目；昆明市气象局科研项目 |
|
| Hourly PM2.5 Concentration Prediction Based on Various Machine Learning Models: The Case of Kunming |
|
Yang Ke1, Ma Bitao2, Niu Xi2
|
|
1.Guandu Meteorological Bureau;2.Kunming Meteorological Bureau
|
| Abstract: |
| To enhance the prediction accuracy of PM2.5 concentrations, five predictive models were developed: K-Nearest Neighbors (KNN), Support Vector Regression (SVR), Back Propagation Neural Network (BPNN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). These models utilized hourly ground meteorological and air pollution data from 2016 to 2021 to assess their predictive accuracy and stability. Results indicated that the XGBoost model outperformed the other models, achieving R2, MAE, RMSE, and MAPE values of 0.89, 0.24 μg/m3, 0.11 μg/m3, and 1.6%, respectively. Seasonal variations were observed in the predictions, with the highest accuracy in spring, followed by autumn, and the lowest in summer;Specifically, the values for R2, MAE, RMSE, and MAPE in spring were 0.88, 0.26 μg/m3, 0.12 μg/m3, and 1.32%.The evaluation of meteorological factors highlighted that visibility and precipitation significantly contributed to the accuracy of the predictive models, while the assessment of pollutant concentrations showed that CO and NO2 levels substantially impacted model performance. This research offers valuable insights for the precise forecasting of PM2.5 concentrations and is significant for air pollution control and environmental monitoring. |
| Key words: PM2.5 prediction models machine learning XGBoost meteorological factors |
