<
doi:  10.12013/qxyjzyj2020-016
基于BP人工神经网络的鹰潭市PM2.5和PM10浓度预测模型

Prediction model of PM2.5 and PM10 concentration in Yingtan city based on BP neural network
摘要点击 154  全文点击 80  投稿时间:2020-03-24  修订日期:2020-05-20
查看HTML全文  查看全文  查看/发表评论  下载PDF阅读器
基金:  2019年江西省青年人才项目“BP人工神经网络在鹰潭市空气污染指数预报中的应用”
中文关键词:  大气颗粒物,预测模型,BP人工神经网络,气象要素,气体污染物
英文关键词:  atmospheric particulate matter  prediction model  BP neural network  meteorological elements  air pollutants
        
作者中文名作者英文名单位
刘懿枢Liu Yishu鹰潭市气象局
戴熙敏Dai Ximin鹰潭市气象局
齐永胜Qi Yongsheng鹰潭市气象局
引用:刘懿枢,戴熙敏,齐永胜.2020,基于BP人工神经网络的鹰潭市PM2.5和PM10浓度预测模型[J].气象与减灾研究,43(2):123-129
中文摘要:
      利用2015—2019年鹰潭市5个大气成分监测站数据和气象站地面观测数据,运用主成分分析法,提取气象要素、气体污染物对PM2.5和PM10浓度影响的主要成分,调整BP人工神经网络的隐藏层个数和隐藏层节点数,构建基于BP人工神经网络的鹰潭市PM2.5和PM10浓度预测模型。结果表明:1) 气象要素中,共提取3个影响PM2.5、PM10浓度的主成分,分别为相对湿度、降水,气温、气压和风速,其中湿度、气温、风速与PM2.5、PM10浓度显著相关。2) 气体污染物中,共提取2个主成分,分别为SO2、NO2和O3,其中NO2、SO2与PM2.5、PM10浓度显著相关。3) 所建立的PM2.5、PM10浓度逐小时预测模型在20 h内预测性能良好,预测准确率分别为88%、86%,逐日预测模型在5 d内的预测性能良好,预测准确率分别为94%、92%,准确率较高,具有良好的预报性能。
Abstract:
      Based on the observation data of five atmospheric composition monitoring stations and ground observation data of meteorological stations in Yingtan City from 2015 to 2019,the main influence factors of meteorological elements and air pollutants impacting on the concentration of PM2.5 and PM10 were extracted by principal component analysis, the number of hidden layers and hidden layer nodes of BP artificial neural network were adjusted, and then a PM2.5 and PM10 concentration prediction model based on the BP artificial neural network in Yingtan City was established. The results showed that: 1) For meteorological factors, relative humidity, precipitation, temperature, air pressure and wind speed were the main components, humidity,temperature and wind speed presented significantly correlations with the concentration of PM2.5 and PM10. 2) For air pollutants, SO2、NO2 and O3 were the main components, NO2 and SO2 presented significantly related to the concentration of PM2.5 and PM10. 3) The PM2.5 and PM10 concentration prediction models exhibited good performance in hourly prediction within 20 h and daily prediction within 5 d, the hourly accuracies of PM2.5 and PM10 concentration prediction were 88% and 86%,respectively, and the daily prediction accuracies of PM2.5 and PM10 concentration prediction reached 94% and 92%. The accuracy of daily prediction was relatively higher.
主办单位:江西省气象学会 单位地址:南昌市高新开发区艾溪湖二路323号
联系电话: (0791)82713175传真:0791-82713175 邮编:330096 Email:qxyjzyj@163.com
本系统由北京勤云科技发展有限公司设计