Application Research of Multisource Data Fusion and Multimodel Ensemble Methods in Aircraft Approach State Prediction

With advances in sensing technology, situational awareness and prediction based on multisource data have become crucial for ensuring flight safety. Compared to single-model approaches, multimodel ensemble prediction algorithms offer higher accuracy and robustness, yet current ensemble strategies are insufficient for further development. Therefore, this article introduces a new multimodel ensemble prediction algorithm - the minimum consensus cost ensemble (MCCE) prediction algorithm - specifically for predicting the final approach state of civil aircraft. The MCCE algorithm incorporates the "minimum consensus cost"concept from group decision-making, quantifying the bias of individual models toward different prediction outcomes and dynamically adjusting the ensemble strategy to select the result with the highest consensus among models as the final output. This approach addresses decision bias issues in traditional ensemble models, enhancing stability and accuracy. In experiments, a fusion model was designed to maximize the silhouette coefficient, combining convolutional neural networks (CNNs) and multihead attention (MHA) mechanisms to efficiently integrate multisource data and isolate final approach state features. Through independent modeling and performance comparisons with multiple classification algorithms such as SVM and CNN, results show that the MCCE algorithm achieves approximately 99.97% accuracy, maintaining about 94.71% accuracy even under 20-dB signal-to-noise ratio (SNR) noise interference, significantly outperforming other models. Based on MCCE's consensus cost, the key parameters leading to abnormal approach states in civil aircraft can be identified, helping pilots make timely, effective flight decisions to reduce approach-phase accident risks.