%0 Journal Article %T 基于气象因素的PM2.5质量浓度预测模型<br>Prediction models of PM2.5 mass concentration based on meteorological factors %A 刘杰 %A 杨鹏 %A 吕文生 %A 刘阿古达木 %A 刘俊秀< %A br> %A LIU Jie %A YANG Peng %A LYU Wensheng %A LIU Agudamu %A LIU Junxiu %J 山东大学学报（工学版） %D 2015 %R 10.6040/j.issn.1672-3961.0.2014.214 %X 摘要： 为得出拟合效果最佳的预测模型,建立了多元回归和机器学习预测模型对PM2.5质量浓度进行预测。在输入气象因素的基础上,引入污染物质量浓度基础值和周期因素两类变量作为预测输入,并对4种预测模型进行对比研究。研究结果表明:对预测输入进行改进后,多元线性回归预测模型拟合优度由0.52提高至0.64,所选取的气象参数、污染物质量浓度基础值和周期因素能较好地描述PM2.5质量浓度的日变化情况;与多元线性回归预测模型相比,BP神经网络和支持向量机两种预测模型能较好地捕捉PM2.5质量浓度与预测输入之间的非线性影响规律,整体拟合优度分别达0.69和0.74,预测准确度较高;支持向量机预测模型可作为PM2.5质量浓度预测的首选方法。<br>Abstract: In order to get the optimal prediction model, the prediction models of PM2.5 mass concentration based on multiple linear regression and machine learning were developed. Basic values of pollutants mass concentrations and periodical factors were introduced as predictive inputs based on meteorological factors. Then four prediction models were developed for comparison. Results showed that goodness of fit of multiple linear regression model based on improved predictive inputs was increased from 0.52 to 0.64. The selected meteorological factors, basic values of pollutants mass concentrations and periodical factors could accurately describe daily variation of PM2.5. BP neural network and support vector machine models could be trained to model the highly non-linear relationships between PM2.5 mass concentration and predictive inputs. They provided satisfactory results with goodness of fit of 0.69 and 0.74, respectively. Support vector machine model was proved to be optimal prediction model of PM2.5 mass concentration %K 机器学习 %K BP神经网络 %K 支持向量机 %K 多元线性回归 %K PM< %K sub> %K 2.5< %K /sub> %K < %K br> %K PM< %K sub> %K 2.5< %K /sub> %K multiple linear regression %K support vector machine %K machine learning %K BP neural network %U http://gxbwk.njournal.sdu.edu.cn/CN/10.6040/j.issn.1672-3961.0.2014.214