Research on Pilot Workload Identification Based on Stacking Ensemble Learning

Research on identifying pilot workload under complex task conditions and reducing human error is crucial for the improvement of aircraft safety. However, traditional workload identification methods often suffer from low accuracy and poor generalization. To address these issues, physiological data related to pilots, such as EEG, ECG, and NASA-TLX scores, were collected through high-fidelity flight simulation experiments. Techniques including independent component analysis, wavelet packet transform, and time-frequency analysis were applied to extract feature indicators, which were then analyzed and screened based on their performance across different tasks. A stackingbased ensemble learning model integrating RF, SVM, LSTM, and Bi-LSTM algorithms was developed for workload identification. Compared to traditional algorithms, the proposed approach demonstrated significant improvements in both accuracy and generalization. Specifically, it achieved an accuracy of 89.05%, representing respective improvements of 12.46%, 10.48%, 6.51%, and 3.93% over the base models. Moreover, this research offers significant advancements in pilot workload management and intelligent assistance in civil aircraft cockpits.