基于GMDH神经网络的智能多源自主导航方法  

Intelligent Multi-source Autonomous Navigation Method Based on GMDH Neural Network

作　　者：冯会硕 邢朝洋[1] 薄凡 周睿阳 南子寒 孟凡琛 FENG Huishuo;XING Chaoyang;BO Fan;ZHOU Ruiyang;NAN Zihan;MENG Fanchen(Beijing Institute of Aerospace Control Devices,Beijing 100039)

机构地区：[1]北京航天控制仪器研究所,北京100039

出　　处：《导航与控制》2024年第4期45-53,共9页Navigation and Control

基　　金：国家自然科学基金(编号:62388101);航天科技集团九院科技委青年基金(编号:KJWQN202402)。

摘　　要：针对复杂受限、博弈对抗等场景下飞行器自主导航系统的“可重构性”需求,提出了一种基于GMDH神经网络的多源自主导航方法。利用GMDH神经网络的动态建模能力增强传统卡尔曼滤波器的状态估计矩阵,通过神经网络构建时变状态转移模型,预测并代替GNSS系统时间序列中的位置与速度信息,实现卫星拒止条件下的多源自主导航滤波器的平滑快速收敛,有效应对卫星信号中断情况下导航误差发散等难题,提高其应对复杂任务场景的自主判别与重构能力。仿真结果表明,相比于传统松组合导航和长短时记忆神经网络滤波优化算法,所提方法在卫星短时拒止条件下,速度精度提升了27.3%,位置精度提升了20.1%,为新一代国家综合PNT体系端侧自主导航系统设计提供参考。To enhance the reconfigurability of autonomous navigation systems for aircraft operating in complex and adversarial scenarios,a multi-source autonomous navigation method based on Group Method of Data Handling(GMDH)neural network is proposed.The dynamic modeling capability of GMDH neural network is utilized to improve the state estimation matrix of the traditional Kalman filter,and a time-varying state transition model is constructed through the neural network,which predicts and substitutes the position and velocity information in the GNSS system time series.The proposed method allows the multi-source autonomous navigation filters to achieve smooth and rapid convergence under satellite rejections,effectively mitigating error divergence during GNSS signal interruptions.Furthermore,it enhances the system’s autonomous judgment and reconfiguration capabilities in complex mission scenarios.Simulation results show that compared to traditional loosely coupled navigation and Long Short-Term Memory neural network filter optimization algorithms,the proposed method improves the velocity accuracy by 27.3%and the position accuracy by 20.1%under short-term GNSS-denied conditions,providing a reference framework for designing end-side autonomous navigation systems within the new generation of national integrated Positioning,Navigation and Timing(PNT).

关 键 词：卫星拒止 GMDH神经网络 智能多源自主导航 可重构性 

分 类 号：V249[航空宇航科学与技术—飞行器设计]