Abstract: Low efficiency in QAR(Quick Access Recorder) data analysis highlights necessity of feature extraction from QAR data. In response to issue of insufficient focus on temporal trend features in QAR data feature extraction, Piecewise Cubic Hermite Interpolating Polynomial (PCHIP) and ordering relation algorithm (G1) were integrated to form the interpolative weighting PG, and combined with a convolutional autoencoder (CAE) to construct PG-CAE model. a flight pitch operation feature extraction method based on PG-CAE model was developed to support analyses such as flight level anomaly detection. First, PCHIP interpolation is used to standardize data length; second, G1 weighting module determines weight between flight operations and flight attitudes, quantifying temporal importance of pitch operation data; then, a CAE module is employed to extract features from weighted data; finally, model is validated by pitch operation data from 406 flight segments of a certain airline's A319 aircraft. Results indicate that PG-CAE model significantly outperforms CAE model by introducing PCHIP and G1 modules, Reconstruction error was employed as baseline criterion to measure compliance of individual data samples with original data, determining model acceptability. Standard deviation was used to assess model’s capability to extract overall trend features from dataset. CAE5 model with a 5-layer convolutional-pooling architecture was ultimately identified as the optimal configuration, demonstrating a reconstruction error of 0.03284 and standard deviations of (0.1621, 0.2805). Furthermore, by K-means algorithm, a comparison of point clustering effect after PG-CAE feature extraction with curve clustering effect without feature extraction demonstrates that PG-CAE model can transform temporal trend data into two-dimensional feature point clusters, serving research on flight level anomaly detection in QAR data.

Key words: flight operation, feature extraction, Quick Access Recorder &#40, QAR&#41

摘要： 快速存取记录器(Quick Access Recorder，QAR )数据分析效率低，对QAR数据进行特征提取至关重要。针对QAR数据特征提取对于时序趋势特征关注不足的问题，融合分段三次Hermite插值（PCHIP）和序关系分析法(G1)形成模型插值赋权部分PG，结合卷积自编码器(CAE)构建PG-CAE模型，提出一种基于PG-CAE模型的飞行俯仰操作特征提取方法，为飞行级异常检测等分析提供支持。首先，利用PCHIP插值统一数据长度；其次，利用G1赋权模块，根据飞行操作与飞行姿态的因果时序相关性，确定权重量化飞行俯仰操作数据的时序重要性；然后，使用CAE模块对赋权后的数据进行特征提取；最后，基于某航司A319机型406个航段俯仰操作数据进行模型验证。结果表明：通过引入PCHIP与G1模块，PG-CAE模型结果明显优于CAE模型，以重构误差度量单一数据样本与原始数据符合度，作为模型是否可接受的底线标准，以标准差度量模型对数据集整体趋势特征的提取能力，最终确定具有5重卷积池化层的CAE5模型，重构误差为0.03284、标准差为(0.1621,0.2805)，为最优模型结构。此外，结合K-means算法，对比PG-CAE特征提取后的点聚类效果与未经特征提取的曲线聚类效果，进一步证明PG-CAE模型可将时序趋势数据线簇数据提取为二维特征点簇数据，服务于QAR数据飞行级异常检测等研究。

关键词: 飞行操作, 特征提取, 快速存取记录器, 卷积自编码器, 序关系分析法