基于时间序列模型的航班运行风险短期预测

针对国内航班运行风险预测技术匮乏的现状,采用移动平均自回归(ARMA)方法,构建航班日运行风险的单变量预测模型;采用向量自回归(VAR)方法,构建航班日运行风险的多变量预测模型;经稳定性检验后,对比2种方法的短期预测效果,发现使用ARMA的单变量预测模型,未来第3天预测精度达到80.76%,可用预测周期为1~3天;而VAR多变量预测模型计算出未来第1天预测精度可高达92%,第7天预测精度仍达到80.64%,适用预测周期为1~7天。结果表明:基于ARMA和VAR的时间序列模型可用于航班运行风险的短期预测,而VAR模型精度更好,更加符合实际需求。

Short-time prediction of flight operation risk based on time series models

In order to address the lack of flight operation risk prediction technology in China,ARMA method was used to build a univariate prediction model of flights’daily operation risk.Then,a multivariate prediction model was constructed by using VAR method.Finally,short-term prediction efficiency of two models was compared through stability test.The results show that the 3 rdday prediction accuracy of ARMAbased single variable prediction model can be 80.76%,and its available forecast period is 1-3 days while that of VAR-based model can be as high as 92%for the 1 stday and still keep at 80.64%for 7 thday with an applicable prediction period of 1-7 days.It is proved that ARMA and VAR-based time series models can predict flight operation risk in a short term,but the VAR-based multivariate prediction model has higher accuracy,which meets airlines’actual needs better.