顾及PWV的广西地区多尺度PM2.5浓度预测
Multi-scale PM2.5 concentration prediction considering PWV in Guangxi
谢劭峰 1张亚博 1黄良珂 1魏朋志 1张继洪 1唐友兵1
作者信息
- 1. 桂林理工大学 测绘地理信息学院,广西 桂林 541006;桂林理工大学 广西空间信息与测绘重点实验室,广西 桂林 541006
- 折叠
摘要
针对现有的雾霾预测方法较少考虑可降水量的影响,且大部分预测方法都没有对模型回归残差进行有效处理因而预测精度不是很高的问题,利用广西南宁、桂林、梧州和百色四市 2017 年的PM2.5 日均值数据,结合大气污染物、气象因子和大气可降水量PWV等因素,分别建立全年和分季度的ARIMA模型对该地区PM2.5 日均浓度进行短期预测,并将ARIMA模型预测残差分别用前馈神经网络径向基函数RBF和多层感知器MLP进行拟合,以达到优化ARIMA模型的目的.结果表明,除桂林外,分季度ARIMA模型预测效果优于全年ARIMA模型,季度ARIMA-MLP神经网络预测精度优于分季度ARIMA模型,表明该类模型可以用于区域PM2.5 浓度预测.
Abstract
The existing smog-haze forecast methods less consider about the influence of precipitable water va-por,and most of the prediction methods do not effectively handle the model regression residuals,so the predic-tion accuracy is not very high.For these issues,the PM2.5 daily mean value data of four cities in Guangxi(Nan-ning,Guilin,Wuzhou and Baise)in 2017 combining with the factors of air pollutants,meteorological factors and precipitable water vapor(PWV),ARIMA models of the whole year and each quarter are respectively estab-lished to make short-term prediction of daily average PM2.5 concentration in this region.The forecast residuals of ARIMA model are respectively fitted with the feedforward neural network radical basis function(RBF)and multi-layer perceptron(MLP)in order to optimize ARIMA model.The results show,except Guilin,the predic-tion effect of quarterly ARIMA model is better than the annual ARIMA model,and the quarterly ARIMA-MLP neural network prediction accuracy is better than the quarterly ARIMA model,indicating that this kind of model can be used for regional PM2.5 concentration prediction.
关键词
PM2.5/PWV/ARIMA/前馈神经网络Key words
PM2.5/PWV/ARIMA/feedforward neural network引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据