论文部分内容阅读
针对无陀螺捷联惯导系统(GFSINS)中传统角速度算法解算精度不高的问题,提出一种可避免复杂代数运算的反向传播(BP)神经网络算法来求解角速度.基于一种十加速度计构型方案,选择10个加速度计输出、采样周期和臂杆距离等12个已知量作为网络输入,以对数法得到的角速度值作为期望输出,针对5 000个样本在不同的隐含层层数、单层神经元个数以及学习步数等情况下进行网络训练,构建了一个含有30个隐含层神经元的3层BP网络模型.采用此模型对角速度进行实时预测,结果表明:网络具有很好的适应能力和实时性,角速度实时预测时间与对数法相当,且其预测精度比对数法提高大约3倍.
Aiming at the problem of low accuracy of traditional angular velocity algorithm in GFSINS, a BP neural network algorithm which can avoid complicated algebraic computation is proposed to solve the angular velocity.Based on a ten-acceleration In this scheme, 12 known quantities such as the output of 10 accelerometers, sampling period and arm distance were selected as the network inputs, and the angular velocity values obtained by logarithm method were output as expected. For the 5 000 samples under different implicit Layer number, the number of single neurons and the number of learning steps and so on, a 3-layer BP neural network model with 30 hidden layer neurons was constructed.With this model, the angular velocity was predicted in real time and the result showed : The network has good adaptability and real-time performance. The real-time prediction time of angular velocity is comparable with logarithmic method, and its prediction accuracy is about 3 times higher than that of logarithmic method.