基于多元时间序列的PM2.5预测方法

PM2.5 Prediction Method Based on Multiple Time Series

针对利用多元线性回归和时间序列模型预测PM2. 5时,存在信息利用不全面和预测精度不高的问题,提出了基于多元时间序列(ARMAX)的PM2. 5预测方法;方法在回归项中引入了PM2. 5影响因子在时间序列上的滞后性阶数,并对残差序列进行信息提取,建立了PM2. 5浓度预测模型;首先通过"天气后报网"采集了合肥市2017年和2018年污染物数据;完成了数据的预处理及相关性分析;分别建立了PM2. 5浓度预测的多元线性回归模型、时间序列模型和ARMAX模型;最后通过RMSE、MAE和Theil不相等系数3个评价指标,将3个模型预测精度进行比较;结果表明:ARMAX模型的预测精度显著高于单一的时间序列模型或多元线性回归模型。

In view of the incompleteness of information utilization and the low prediction accuracy of PM2. 5 prediction by multiple linear regression and time series models,a PM2. 5 prediction method based on multiple time series( ARMAX) is proposed. This method introduces the delay order of PM2. 5 influence factor on the time series in the regression term,extracts the residual error sequence information,and establishes the PM2. 5 concentration prediction model. Firstly,pollutant data of Hefei in 2017 and 2018 were collected through the ' Post-Weather Network';Next the data preprocessing and correlation analysis were completed;Then,the multiple linear regression model,time series model and ARMAX model of PM2. 5 concentration prediction were respectively established;Finally,the prediction accuracy of the three models was compared by the three evaluation indicators of RMSE,MAE and Theil inequality coefficient. The result shows that the prediction accuracy of ARMAX model is better than that of single multiple linear regression model and time series model.