用GRU循环神经网络优化CMAQ预测结果  

Optimizing the Prediction Results of the CMAQ Model Using GRU Recurrent Neural Network

作　　者：张彦虎 范敬勇 ZHANG Yanhu;FAN Jingyong(Xingtai Ecological Environment Monitoring Center of Hebei Province,Xingtai 054000,China)

机构地区：[1]河北省邢台生态环境监测中心,河北邢台054000

出　　处：《中国环境监测》2024年第6期70-79,共10页Environmental Monitoring in China

摘　　要：提高空气质量预报的准确度对于区域大气污染精准防控具有重要意义。针对邢台市空气质量预报情况,使用WRF气象模型输出数据和CMAQ空气质量模型输出数据结合GRU循环神经网络,建立了WRF-CMAQ-GRU模型,对2022年7月邢台市PM_(2.5)、PM_(10)、SO_(2)、NO_(2)、O_(3)、CO等6种污染物的预测结果进行优化。实验发现:该模型对PM_(2.5)及O_(3)的优化效果最明显,PM_(2.5)数据优化后的相关系数由0.28提高到0.85,O_(3)数据优化后的相关系数由0.29提高到0.70。初步验证了GRU循环神经网络对WRF-CMAQ模型预报结果的显著优化作用,使空气质量预报准确度得到较大提升。Improving the accuracy of air quality forecast is of great significance for precise prevention and control of regional air pollution.In this study,aiming at the air quality forecast in Xingtai area,the WRF-CMAQ-GRU model was established using the WRF meteorological model output and the CMAQ air quality model output combined with GRU recurrent neural network.Correction experiment were conducted to optimize the prediction of six pollutant in Xingtai City in July 2022,including PM_(2.5),PM_(10),SO_(2),NO_(2),O_(3),and CO.It was found that the optimization effect of the model on PM_(2.5) and O_(3) was the most obvious,and the overall correlation coefficient of PM_(2.5) increased from 0.28 to 0.85,the correlation coefficient of O_(3) increased from 0.29 to 0.70.It's preliminary verified that the GRU recurrent neural network can significantly optimize the prediction results of the WRF-CMAQ model,which greatly improves the accuracy of air quality prediction.

关 键 词：环境空气 预测预报 多尺度空气质量模型(CMAQ) WRF-CMAQ-GRU模型 循环神经网络 

分 类 号：X823[环境科学与工程—环境工程]