基于STL和BI-MLLSTM的航向误差故障事件预测研究

Research on Heading Error Fault Prediction Based on STL and BI-MLLSTM

摘要:

航班密集化及航线复杂化对飞机航向精度提出更高要求,因此准确预测航向误差故障事件(Heading error fault event,HEFE)对飞行安全及优化维修策略意义重大。然而,HEFE机理复杂,呈现出季节性、非线性及不规则波动性等特点。鉴于此,本文提出一种局部加权回归季节趋势分解(Seasonal and trend decomposition using Loess,STL)与双向堆叠式长短期神经网络(Bi-directional multi-layer long short-term memory,BI-MLLSTM)相结合的方法对HEFE进行预测。首先,采用STL方法将HEFE数据分解为解释性较强的周期项、趋势项及剩余项。其次,结合BI-MLLSTM的双向学习及时序处理优势搭建预测模型。最后,本文模型与传统时序预测模型对比,MAE和RMSE误差平均降低33.6%与33.2%,能够有效实现HEFE预测。

Abstract:

The increasing density of flight schedules and the complexity of flight routes have raised higher demands for the accuracy of aircraft heading precision. Therefore, accurately predicting Heading Error Fault Events (HEFE) is of significant importance for flight safety and the optimization of maintenance strategies. However, the mechanism of HEFE is complex, showing seasonal, nonlinear and irregular fluctuations. In light of this, this paper proposes a method that combines Seasonal and Trend Decomposition using Loess (STL) with a Bi-directional Multi-layer Long Short-Term Memory Network (BI-MLLSTM) for predicting HEFE. Firstly, the STL method is employed to decompose the HEFE data into highly interpretable seasonal components, trend components, and residuals. Secondly, a predictive model is constructed by leveraging the bidirectional learning and temporal processing advantages of BI-MLLSTM. Finally, compared with the traditional time series prediction model, the MAE and RMSE errors of the proposed model are reduced by 33.6 % and 33.2 % on average, which can effectively realize the HEFE prediction.