An online prediction method for airplane landing risk in extremely windy conditions based on stream learning

Accurate prediction of landing risk during extremely windy conditions is vital to ensuring flight safety . However, most existing methods rely on pre-trained single-risk models, which often have poor prediction accuracy under extremely windy conditions such as strong winds. This study proposes an online multi-risk prediction method for airplane landing during extremely windy conditions, leveraging a streaming Bi-LSTM-Transformer (BLT) architecture. First, data preprocessing is performed to ensure the high-quality input for analysis. A comprehensive multi-risk assessment framework is then developed to evaluate the potential landing risks, utilizing the processed data. Subsequently, time-series modeling is conducted using a Bi-LSTM network. To improve the models’ global time-series modeling ability, a Transformer component is integrated into the Bi-LSTM output, forming the BLT model. Finally, stream learning is introduced to maintain predictive performance under extremely windy conditions. Using real-time flight data streams, the Stream BLT model applies eligibility traces and a dynamic learning rate to enable continuous, adaptive learning . Experimental results demonstrate that the proposed method accurately predicts airplane landing risks under extremely windy conditions, particularly when wind speeds exceed 33knots. The model achieves an average relative error of approximately 0.01–0.07 and a prediction latency of under 30 ms, offering efficient, reliable real-time airplane landing risk assessment.