引用本文:伍亚,邹凤娟,高丽洁,刘晓波,马文军,梁晓峰,朱穗.基于机器学习的大气PM2.5中金属浓度预测模型的研究[J].环境监控与预警,2023,15(5):8-16
WU Ya,ZOU Fengjuan,GAO Lijie,LIU Xiaobo,MA Wenjun,LIANG Xiaofeng,ZHU Sui.Prediction Models of Metal Components in Ambient PM2.5 Based on Machine Learning[J].Environmental Monitoring and Forewarning,2023,15(5):8-16
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 4584次   下载 685 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于机器学习的大气PM2.5中金属浓度预测模型的研究
伍亚1,邹凤娟1,高丽洁1,刘晓波2,马文军1,梁晓峰3,朱穗1*
1. 暨南大学基础医学与公共卫生学院,广东 广州 510632;2. 哈尔滨市疾病预防控制中心,黑龙江 哈尔滨 150030;3. 暨南大学疾病预防控制研究院,广东 广州 510632
摘要:
基于2013—2018年哈尔滨市气象数据、大气污染物数据和细颗粒物(PM2.5)中金属成分数据,采用机器学习方法探索大气PM2.5中金属浓度预测模型,并选择最优模型进行污染物浓度预测。结果表明,多元线性回归(MLR)、人工神经网络(BP-ANN)、支持向量机(SVM)和随机森林(RF)4种模型中,RF对大气PM2.5中5种金属[锑(Sb)、砷(As)、铅(Pb)、镉(Cd)、铊(Tl)]的浓度预测效果最佳,在训练集和测试集中表现均较稳定,其中相关系数(r)均>0.7, 平均绝对误差(MAE)和均方根误差(RMSE)数值较小。RF在大气PM2.5中金属浓度预测上具有较好的表现,可在缺乏监测和实验数据的情况下,实现对大气颗粒物中金属浓度的快速预测,为全面了解颗粒物中金属污染特征提供数据基础。
关键词:  细颗粒物  金属  机器学习  预测模型
DOI:10.3969/j.issn.1674-6732.2023.05.002
分类号:X513
基金项目:广东省基础与应用基础研究基金项目(2021A1515012578)
Prediction Models of Metal Components in Ambient PM2.5 Based on Machine Learning
WU Ya1, ZOU Fengjuan1, GAO Lijie1, LIU Xiaobo2, MA Wenjun1, LIANG Xiaofeng3, ZHU Sui1*
1.School of Basic Medical Sciences and Public health, Jinan University, Guangzhou,Guangdong 510632, China;2.Harbin Center for Disease Control and Prevention, Harbin, Heilongjiang 150030, China;3.Disease Control and Prevention Institute, Jinan University, Guangzhou, Guangdong 510632, China
Abstract:
Based on the meteorological data,air pollutant data and metal components in PM2.5 in Harbin from 2013 to 2018,the machine learning method was used to explore the metal concentration prediction model in ambient PM2.5, and the optimal model was selected for prediction. After comparing four models of multiple linear regression(MLR), back propagation artificial neural network(BP-ANN), support vector machine(SVM) and random forest(RF), the results showed that the prediction effect of RF on the five metals concentration [antimony(Sb), arsenic(As), lead(Pb), cadmium(Cd), thallium(Tl)]in PM2.5 was the best, and the predicted performance in the training set and test set was relatively stable, and the correlation coefficient(R) values were all greater than 0.7, and the mean absolute error(MAE) and root mean square error(RMSE) value were smaller. RF had a good performance in predicting the concentration of metal components in PM2.5. Our results provide an effective approach for the prediction of airborne metal concentrations in the absence of monitoring and experimental data, and provide data basis for a more comprehensive understanding on metallic pollutants in particulate matters.
Key words:  PM2.5  Metal  Machine learning  Prediction model