论文部分内容阅读
首先,采用对角神经网络(DRNN)对陀螺的随机误差进行建模。给出DRNN的结构与数学模型,说明DRNN作为内递归网络在建模时不必知道模型阶次;然后,利用LM算法优化BP算法,导出DRNN适用的LMBP算法对网络进行训练;最后,利用DRNN对光纤陀螺随机误差建模,并与陀螺随机误差建模常用的ARMA模型和外递归BP网络进行对比,通过仿真定量分析DRNN的建模效果。仿真结果表明,DRNN比其他方法更加有效便捷。
First, the random error of the gyro is modeled using diagonal neural network (DRNN). The structure and mathematic model of DRNN are given. It shows that DRNN does not need to know the model order when modeling as inner recursive network. Then, LM algorithm is used to optimize BP algorithm and the LMBP algorithm which is suitable for DRNN is derived to train the network. Finally, The error of optical fiber gyroscope is modeled. Compared with the ARMA model and the external recursive BP network, which are commonly used in gyro random error modeling, the modeling effect of DRNN is quantitatively analyzed through simulation. Simulation results show that DRNN is more efficient and convenient than other methods.