Deep-Learning-Based Docking Error Prediction and Safety Warning of Aerial Refueling

Autonomous aerial refueling (AAR) is a key technology in both civilian and military aviation fields, and its docking process involves complex dynamic environments and high-precision requirements. To enhance safety during the aerial refueling docking process and prevent safety incidents, this paper proposes a method for docking error prediction and safety warning of aerial refueling based on deep learning. First, a high-fidelity aerial refueling simulation platform is constructed to generate a time-series dataset. Second, deep feature extraction of the time-series data is implemented through a convolutional neural network, and the inherent spatiotemporal feature representations are mined. A bidirectional long short-term memory network is used to model the temporal dynamic relationships of historical data from aerial refueling docking, and the attention mechanism is integrated to construct a docking error prediction model with the ability to capture temporal dependencies. Then, residual analysis of the model’s predicted values for docking error is conducted using the sliding window method to determine the appropriate safety warning threshold and rules. Finally, the experimental results indicate that the proposed combined model has high prediction accuracy and can achieve safety warnings of abnormal docking states during aerial refueling. This verifies the feasibility of the proposed method and provides technical support for the implementation of high safety in AAR.