【摘要】 应用差分自回归移动平均模型(ARIMA)和最小二乘支持向量机(LS-SVM)的组合模型,对某航空公司的月度事故征候万时率进行了预测分析。对2008—2016年某航空公司的事故征候、飞行小时、航空器数量等历史数据建立ARIMA模型,应用SPSS软件进行模型拟合,获得事故征候万时率的线性部分;随后利用LS-SVM分析ARIMA模型的残差,获取非线性部分,最终通过二者之和获得ARIMA+LSSVM组合模型。对2017年1—3月的月度事故征候万时率进行了预测,并用实际数据验证。结果表明:ARIMA(1,1,1)(1,1,1)12模型较好地拟合了事故征候万时率的历史序列,LS-SVM模型对残差的拟合获得了较好的精度;组合模型的短期(3个月)预测值与航空公司事故征候万时率的趋势完全一致,且预测精确度可接受。更多还原

【Abstract】 This paper is aimed at providing a data-base for decision making and flying trend analysis of the safety operation in the airline control. For the said purpose,we have first of all established an auto-regressive integrated moving average( ARIMA)model and the least square supporting vector machine( LS-SVM)model on the basis of analyzing and forecasting the accidents per10 000 flight hours each month with the Chinese airline in accordance with the data of the accidents,the flight hours,the numberof airplanes and that of pilots during the aircraft flying years from Jan. 2008 to Dec. 2016 in the said airline. What is more,the model can also be made to construct the seasonal ARIMA model to be used to predict the linear components of the accidents with the SPSS software. Besides,in the modeling process,the parameters of ARIMA should also be taken as stationary R2= 0. 608,with the normalized BIC =-1. 93. The kernel function was selected by radial basis function( RBF),and the parameters are set to: C = 1000,σ = 0. 001. Furthermore,we have also chosen ARIMA + LS-SVM model to predict the monthly incidents per10 000 flight hours from Jan. 2017 to Mar. 2017. Thus,the results of our calculation and investigation indicate that they are exactly in accord with the number of accidents per 10 000 flight hours of the historical recording series whereas the LS-SVM model helps to obtain a nice precision of the fitted value of residuals.Thus,the real values and the fitted values have also been proved to be in a nice accord,with the maximum absolute error of the fitted value being 0. 76 of Sep. 2012. And,in comparison,the minimum absolute errors of fitted value turn to be 0. 01 in Oct.2009,Apr. 2011,March. 2012,Jan. 2014,and Apr. 2014.And,so,it can be concluded that the combined model can be used to make a short-term prediction( 3 months) with the predicted data well in accord with the trend of the incidents per 10 000 flight hours of the airline. Hence,the accuracy of the predicted results can be said acceptable and,therefore,can be taken as a scientific basis and a guideline proposal to establish a prevention and control plan of the accidents in the airline control.更多还原