关键词: QAR数据, 飞行操作绩效, 梯度提升, 特征提取, 评估方法

Abstract: To address the problems that traditional flight operation performance evaluation methods were generally subjective, analyzed parameters one-sidedly, and could not be comprehensively and objectively quantified, a Multi-dimensional Feature Analysis based on Gradient Boosting Regression Tree (MFA-GBRT) was proposed. Time-domain and trend features of Quick Access Recorder (QAR) data were extracted. An improved gradient boosting regression tree, combined with a threshold cumulative importance screening mechanism, was adopted to realize adaptive quantification of parameter weights, construct a performance level evaluation model, and establish an evaluation index system covering "attitude control-power management-environmental response". Experimental results based on simulator and flight base data showed that the average evaluation accuracy of the proposed method reached 87.83%, which was 10.78%, 6.06% and 3.55% higher than that of existing methods, i.e., Long Short-Term Memory-Deep Neural Network (LSTM-DNN), curve similarity method and wavelet analysis method, respectively. Cross-scenario validation showed that the model had an adaptability of over 95% (high adaptability) in three different flight scenarios. The method realized the full-process objective quantitative evaluation of the flight process and provided a scientific scheme with engineering practicability for flight operation performance evaluation.

Key words: Quick Access Recorder (QAR)data, flight operation performance, gradient boosting, feature extraction, evaluation method