Approach and Landing Safety Early Warning Based on Spectral Temporal Graph Neural Network

Civil aviation accident safety investigations show that approach and landing are most risky phases of the whole flight. Effectively extract features from QAR data by some deep learning methods can help reduce flight safety risks at these phases. Based on the flight dynamics principle, StemGNN spatial-temporal prediction model, and multivariate time series extracted from QAR data, this paper proposes a flight safety early warning analysis method for the aircraft approach and landing phases. The method can effectively extract the inter-series correlation and improve the accuracy of accident warning. We believe that the method has made a significant contribution to the civil aviation approach and landing safety warning.