An optimized spatial target trajectory prediction model for multi-sensor data fusion in air traffic management

With the evolution of air traffic safety management, the traditional single-sensor approach no longer meets the demands for spatial target surveillance. Consequently, there is increasing research interest in multi-sensor data fusion. This paper proposes an innovative network model based on the improved snow ablation optimizer algorithm. It employs convolutional neural network, structured within a bidirectional gated recurrent unit framework, combined with a multi-head attention mechanism, for spatial target trajectory prediction. We segment data from various sensors within the automatic dependent surveillance-broadcast system using a designed sliding window of equal time steps, inputting them into the feature extraction network to capture spatiotemporal features. The improved snow ablation optimizer algorithm optimizes hyperparameters of this network automatically, while the multi-head attention mechanism redistributes weighted features, integrating them into comprehensive features. Finally, predictions of spatial target trajectories are derived from outputs of fully connected layer. Through experiments on the constructed real dataset, it is evident that the improved snow ablation optimizer algorithm exhibits superior performance in optimization tasks. The sensor missing experiment underscore the advantages of multi-sensor data fusion. Furthermore, the ablation studies elucidate the functional disparities among various network architectures. In comparative analyses, the proposed network significantly outperforms prevailing trajectory prediction models across multiple dimensions. In this paper, we propose a new deep learning network, and apply it to the real-world engineering challenge of spatial target trajectory prediction in the air traffic management domain.