基于神经网络的无陀螺捷联惯导系统姿态预测  

Attitude forecast of gyroscope-free SINS based on neural network

作　　者：曹咏弘[1] 张慧[1] 马铁华[1] 范锦彪[1] 祖静[1] 

机构地区：[1]中北大学仪器科学与动态测试教育部重点实验室,太原030051

出　　处：《中国惯性技术学报》2008年第2期159-161,170,共4页Journal of Chinese Inertial Technology

基　　金：武器装备预研基金项目(9140A17080307BQ0409)

摘　　要：在无陀螺捷联惯导系统中,以工程问题为研究背景,针对以往解算载体角速度精度不高,导航误差随时间积累较快的问题,提出了基于信息融合理论的BP神经网络模型预测飞行体姿态的系统,并采用LM算法,提高学习速度。在研究了加速度传感器输出信号对飞行体姿态影响的基础上,将加速度计的输出信息作为输入变量,飞行体的实时三轴角速度作为目标信号建立网络模型。选取测试样本进行训练,得到较高精度的角速度输出,再运用四元数法解算姿态角,从一定程度上抑制了误差的积累。仿真结果表明该优化算法收敛速度快,对角速度的预测精度较高,并且合理选择及增加样本信息可以提高网络的泛化能力,为该系统走向工程实践提供了理论依据。In the gyroscope-free strapdown inertial navigation system, a BP network model system with the information fusion theory was set up for forecasting flying body posture based on improving angular velocity calculation precision and reducing navigation error accumulated over time. And LM algorithm was used to enhance the learning speed. By regarding the information of accelerometer output as input variable, and regarding the real-time three-axis angular velocity of aerocraft as the target signal, the model was built by studying the effect of the information of accelerometer output factors on the testing of flyer attitude. Test samples were selected for training to get higher precision of the angular velocity. Then the quatemion was used to solve the attitude angle and inhibit the accumulation of errors in a certain extent. The results show that the algorithm has quick convergence speed and high accuracy in forecasting the angular velocity. And Reasonable choosing and increasing the sample information can improve the generalization capacity of network.

关 键 词：无陀螺捷联惯导系统 BP神经网络 角速度 姿态解算 

分 类 号：U666.1[交通运输工程—船舶及航道工程]