【摘要】 针对捷联惯性导航系统动基座初始对准的误差模型具有非线性、时变性的特点 ,提出将前向神经网络应用于该误差模型 ,并采用线性H∞ 鲁棒滤波算法在线调整网络权值 ,得到一种适合于该系统模型的改进的自适应前向神经网络 .仿真结果表明 ,在存在模型误差或噪声不确定情况时 ,该网络不仅具有较好的鲁棒性能 ,而且能使误差在很短的时间内收敛 ,提高了系统的实时性 ,而对准精度与采用推广卡尔曼滤波器的精度相当更多还原

【Abstract】 According to the characteristic of nonlinear time varying in strapdown inertial navigation system (SINS) initial alignment’s error model, the application of feedforward neural network to the error model is proposed by using linear H ∞ robust filter to adjust the network’s weights in real time. As a result an adaptive neural network is brought about for SINS. The simulation results show that when there exists model error or uncertain noises, the method has high convergence speed and better robustness comparing to the extended Kalman filter (EKF). In addition the precision is similar to EKF’s.更多还原