基于XGBoost算法的WRF-Chem模式优化模拟

Optimization of WRF-Chem model results by XGBoost algorithm

采用人工智能算法XGBoost结合大气化学模式WRF-Chem,利用北京地区大气污染物的模拟结果及站点监测数据,构建XGBoost统计预报算法模型,并对两种大气污染物(PM_(2.5)和O_(3))进行优化模拟,同时分析其特征贡献要素.结果表明,该统计预报模型能够很好地优化大气化学模式模拟的大气污染物浓度,降低模拟误差,对于北京地区站点模拟浓度优化呈现出城区>近郊>远郊的优化特点,且算法模型对O_(3)浓度优化程度更高,优化后相关系数提高达128%.此外,通过特征要素的贡献量分析表明,CO是影响O_(3)优化的重要特征变量,城郊区特征贡献得分均高达1000以上,Q2(近地面2m比湿)是影响PM_(2.5)优化的重要气象特征变量,城郊区特征贡献得分分别为950和824.

The artificial intelligence algorithm XGBoost was used to build the statistical prediction algorithm model, combined with the atmospheric chemical model WRF-Chem, the simulation results of air pollutants in Beijing and the site monitoring data. The PM_(2.5) and O_(3) were optimized and simulated, and their characteristic contribution factors were analyzed. The results showed that the statistical forecast model of XGBoost can optimize the atmospheric pollutant concentration simulated by the atmospheric chemical model and reduce the simulation error. Moreover, the simulation concentration optimization presented the optimization characteristics of urban> suburbs> outer suburbs in Beijing area site, and the algorithm model optimized the O_(3) concentration to a higher degree, and the correlation coefficient was increased by 128% after optimization. In addition, the contribution analysis of feature elements showed that CO was an important feature variable which affects the optimization of O_(3) and the feature contribution scores of urban and suburban areas were as high as 1000 or more. Q2 (2m specific humidity near the ground) was an important meteorological characteristic variable that influenced the optimization of PM_(2.5) and the characteristic contribution scores of urban and suburban areas were 950 and 824, respectively.