基于时序分解的飞机平均故障间隔飞行时间组合预测

Combination forecasting of mean flying hours between failures for aircraft based on time series decomposition

针对飞机平均故障间隔飞行时间(MFHBF)指标值的预测问题,提出一种基于时间序列分解的组合预测方法。首先利用STL(Seasonal and Trend Decomposition using Loess)方法分解出MFHBF指标的长期趋势项和季节变动项,然后以灰色模型预测指标的长期趋势,以BP神经网络和支持向量机回归组合模型分别预测指标的季节变动,根据误差权重计算季节变动的加权值,最后以加法模型合并趋势和季节的预测值获得最终结果。利用服务点积累的指标数据对方法进行检验,与单独使用支持向量机回归预测得到的结果相比,平均绝对误差由45%减小至21%,证明该方法能够有效提高预测精度,为保障人员提供可信的指标预测结果。

Concerning the prediction accuracy of aircraft＇ s Mean Flying Hours between Failures （MFHBF） index, a combination forecasting method based on time series decomposition was proposed. Firstly, the MFHBF data was decomposed into the long term trend and seasonal variation by using the Seasonal and Trend Decomposition using Loess （ STL）. Secondly, the long-term trend of the index was predicted by the grey model, and the seasonal variation was separately predicted according to Back Propagation （BP） neural network and support vector machine regression model. The weighted value of seasonal variation was calculated according to error weight. Finally, the final result was calculated by combining the seasonal and trend prediction values under the addition model. The practical prediction effect of the method was verified by the measured data from the service point. The mean absolute error was reduced to 21% by using this method in contrast to 45% by using support vector machine regression model alone. This method can effectively improve the prediction accuracy, and provide credible index prediction results for security personnel.