A Novel Semantic Representation of Airport Surface Trajectory for Taxiing Pattern Recognition

Many operational modeling and analysis tasks in airport surface operations rely on accurate representation of aircraft taxiing paths, which can be expensive to generate. Clustering techniques offer an efficient and convenient approach to pattern discovery in trajectory analysi. However, challenges arise in the feature extraction and pattern recognition processes of high-dimensional trajectory data, particularly due to the presence of noise. This paper proposes a semantic surface trajectory clustering framework that overcomes these challenges, enabling feature extraction from high-dimensional trajectory data and automated learning of typical taxiing patterns. Firstly, geometric matching algorithms and topological relationship algorithms are employed for map matching, enabling the transformation of trajectory data into a new semantic representation. Then, by combining Bi-LSTM and K-means, deep clustering of the new semantic representation of trajectories is performed to explore taxiing patterns in airport surfaces. Bi-LSTM models the temporal behavior of each trajectory, learning the feature representation of the trajectories. The clustering module iteratively assigns trajectories based on sample means, achieving iterative updates. The combined training process integrates feature learning with traditional clustering through the optimization of a comprehensive loss function, which minimizes the overall error of the clustering results. Finally, the proposed framework is validated on real datasets, demonstrating superior performance compared to traditional methods across various evaluation metrics.