余辉, 袁晶, 于旭耀, 张力新, 陈文亮
AuthorsHTML:余辉, 袁晶, 于旭耀, 张力新, 陈文亮
AuthorsListE:Yu Hui, Yuan Jing, Yu Xuyao, Zhang Lixin, Chen Wenliang
AuthorsHTMLE:Yu Hui, Yuan Jing, Yu Xuyao, Zhang Lixin, Chen Wenliang
Unit:天津大学精密仪器与光电子工程学院，天津 300072
Unit_EngLish:School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:针对目前缺乏小时尺度上PM_2.5浓度统计预测模型以及已有模型预测精度对训练数据的依赖问题, 利用天津市环保物联网监测到的污染物及气象数据, 建立了PM_2.5小时浓度预测的多元时间序列模型(ARMAX), 并提出一种模型在线自适应改进方法：设定模型评价指标并实时监测, 当模型预测精度超标时对模型进行在线更新．将改进后的模型应用于天津市的9个监测站点, 用2013—2014年的监测数据对模型进行验证．结果表明：模型均方根误差RMSE＜20 μg, 平均绝对误差MAE＜20 μg, 拟合优度R²＞0.9, 能够在小时尺度下有效地预测PM_2.5浓度, 可以为突发性PM_2.5污染事件的应急处理提供决策支持．
Abstract_English:To solve the lack of statistical prediction model in hourly scale for PM_2.5 concentration and the training data dependence of the existing model，a multivariate time series model ARMAX was established. The pollutants and meteorological data monitored by Internet of Things for Environmental Protection in Tianjin were used to establish the model. An online adaptive improved method was proposed，in which three model evaluations were set and monitored online，the model was updated when prediction accuracy was reduced to an extent. The improved prediction model was applied to nine monitoring sites in Tianjin，and the data of 2013—2014 was used to validate the model. Results showed that with the model root mean square error RMSE＜20 μg，the mean absolute error MAE＜20 μg，and the goodness of fit R²＞0.9，the model can predict the next hourly scale PM_2.5 concentration effectively，which can provide a decision support for the emergency treatment of PM_2.5 pollution incidents.
Keyword_Chinese:PM_2.5; 小时浓度预测; 多元时间序列模型; 跟踪预测
Keywords_English:PM_2.5; hourly concentration prediction; multivariate time series model; tracking prediction

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=5773