融合STFT和多流动态网络的空战轨迹预测方法  

Air-combat Trajectory Prediction Method Integrating STFT and Multi-stream Dynamic Network

作　　者：岳圣智 邓向阳 付宇鹏 徐俊 宋婧菡 林远山 YUE Shengzhi;DENG Xiangyang;FU Yupeng;XU Jun;SONG Jinghan;LIN Yuanshan(School of Information Engineering,Dalian Ocean University,Dalian 116023,China;Aviation University Combat Services College,Yantai 264001,China)

机构地区：[1]大连海洋大学信息工程学院,辽宁大连116023 [2]海军航空大学,山东烟台264001

出　　处：《弹道学报》2025年第1期60-67,共8页Journal of Ballistics

基　　金：辽宁省教育厅基本科研项目(LJKZ0730,QL202016);辽宁省自然基金资助计划(2020-KF-12-09);辽宁省重点研发计划(2020JH2/10100043);山东省自然基金青年基金(ZR2024QF094)。

摘　　要：现代空战对抗过程的复杂多变使得空战决策模糊多变,有效的轨迹预测能够极大提升决策的准确性。针对空战轨迹预测中复杂时间序列的特性,提出了一种融合短时傅里叶变换(STFT)与多流Transformer网络的轨迹预测方法,旨在提高空战机动轨迹预测的精度。空战机动过程中,飞行器的轨迹变化频繁且复杂,因此,首先通过高阶差分对轨迹数据进行预处理,以消除噪声并保留轨迹的时空特征。随后,利用短时傅里叶变换对预处理后的轨迹进行频域特征提取,分析轨迹的动态变化。为了更好地捕捉位置轨迹和姿态轨迹的差异性,设计了轨迹解耦策略,将这两类轨迹分开处理。接着,基于多流动态注意力机制的Transformer网络处理这些时空特征,从而实现对飞行轨迹中深层依赖关系的捕捉。该网络通过多头注意力机制对多个数据流进行加权处理,增强了模型对不同数据流的时空依赖关系的捕捉能力。实验结果表明,相比传统的预测方法,该文提出的方法预测精度提高了3.88%;STFT与多流Transformer结合的方法有效提高了对复杂空战机动轨迹的预测精度,验证了其在高精度空战场景预测中的适用性。The complexity and changeability of modern air-combat confrontation makes air-combat decisions fuzzy and changeable.Effective trajectory prediction can greatly improve the accuracy of decision-making.Aiming at the characteristics of complex time series in air-combat trajectory prediction,a trajectory prediction method integrating short-time Fourier transform(STFT)and multi-stream transformer network was proposed to improve the accuracy of predicting air-combat maneuver trajectory.During air-combat maneuvers,the trajectory of aircraft changes frequently and complexly.Therefore,the trajectory data were first preprocessed by high-order difference to eliminate noise and retain the spatiotemporal characteristics of trajectory.Subsequently,the short-time Fourier transform was used to extract frequency-domain features of the preprocessed trajectory and analyze the dynamic changes of the trajectory.In order to better capture the differences between position trajectory and attitude trajectory,a trajectory decoupling strategy was designed to process these two types of trajectories separately.Then,the transformer network based on the multi-stream dynamic attention mechanism was used to process these spatiotemporal features,thereby capturing the deep dependencies in the flight trajectory.The network weights multiple data streams through multi-head attention mechanism,enhancing the model's ability to capture the spatiotemporal dependencies of different data streams.Experimental results show that compared to traditional prediction methods,the proposed method has a 3.88%improvement in prediction accuracy.The combination of STFT and multi-stream transformer effectively improves the prediction accuracy of complex air-combat maneuver trajectory,verifying its applicability in high-precision air-combat scene prediction.

关 键 词：轨迹预测 多维特征处理 短时傅里叶变换 

分 类 号：V249[航空宇航科学与技术—飞行器设计]