摘要
采用小波分解(WD)将济南市科干所监测站PM_(2.5)浓度的一维时间序列(2013年1月1日—2017年8月15日)分解为高维信息,获得了该监测站附近PM_(2.5)浓度的时频变化特征,重点分析了PM_(2.5)的随机性和趋势性问题.然后构建了基于小波分解的多级残差修正的最小二乘支持向量回归预测模型(AMLRC-WLSSVR),结果发现,该模型能够很好地对济南市PM_(2.5)浓度做出预测,特别是针对重污染天气的预测有很好的精度.为了避免预测结果的不确定性问题,提出了一种基于方差估计给出预测值置信区间上界的方法,同时,有效弥补了单点预测的不稳定性及预测精度不足的缺点,该方法能够为实际空气污染预警提供技术支持.
By wavelet decomposition（ WD）,the PM2.5 concentrations in Kegansuo stations of Jinan City which are one-dimensional time series（ from January 1,2013 to August 15,2017） were decomposed into high-dimensional information. Through this,the time-frequency characteristics of the PM2.5 concentrations near the monitoring station were obtained. In this study,the randomness and tendency of PM2.5 were emphatically analyzed. Then the prediction model of the least squares support vector regression under wavelet decomposition and adaptive multiple layer residual correction（ AMLRCWLSSVR） was developed to forecast the PM2.5 concentration in Jinan City. The experimental results showed that the proposed method would be effective to improve the prediction accuracy,especially for the heavy pollution weather. Besides,in order to avoid the uncertainty of the prediction,the novel prediction method in this paper was also based on variance estimate,which can help to obtain the upper bound of the prediction under certain confidential interval. At the same time,it can make up for the lack of the instability of single point prediction and increase the precision of the prediction effectively. In general,the proposed method can provide technical support for the earlier warning on the air pollution.
作者
尹建光
彭飞
谢连科
徐毅
刘辉
巩泉泉
王坤
YIN Jianguang;PENG Fei;XIE Lianke;XU Yi;LIU Hui;GONG Quanquan;WANG Kun(State Grid Shandong Electric Power Research Institute, Jinan 250002;Shandong Universtity of Science and Technology, Qingdao 266590;Chinese Academy for Environmental Planning, Beijing 100012)
出处
《环境科学学报》
CAS
CSCD
北大核心
2018年第5期2090-2098,共9页
Acta Scientiae Circumstantiae
基金
国家电网公司科技项目~~
关键词
PM2.5
小波分解
自适应多级残差修证
最小二乘支持向量回归
PM2.5
wavelet decomposition
adaptive multiple layer residual correction
least squares support vector regression
