小波神经网络用于光纤陀螺漂移误差辨识

Application of wavelet neural network for identification of drifts errors in fiber optical gyroscope

提出了采用小波消噪和小波神经网络两个模型对光纤陀螺漂移误差进行辨识。应用小波分析方法消除高频噪声,改善信噪比,把消噪信号作为神经网络期望输出,然后采用带遗忘因子的递推最小二乘(DRLS)算法训练网络并调整权值。该算法不进行任何矩阵运算,在保持收敛速度快和精度高的前提下,极大地减少了计算量,提高了小波神经网络的实时性能,仿真结果表明辨识误差在1.5%以内。

A wavelet de-noise model and wavelet neural network model are presented to predict drift errors of FOG and to enhance identification precision. The method utilizes wavelet analysis to remove high frequency noise and to improve its signal-to-noise ratio. De-noise signal is looked on as the desired output. Then, the network is trained and the weight matrices of neural network are modified by the recursive least square method with forgetting factor. This algorithm without any matrix operation can reduce the computation time and greatly improve real-time performance at fast convergence speed and high precision. The simulation results show that the identification error is within 1. 5%. The method establishes the foundations for identification and compensation of the drift error effectively and accurately in inertial navigation systems.