基于投影寻踪学习网络模型的光纤陀螺漂移估计器设计(英文)

Estimator for Fiber Optical Gyro Drift Using Projection Pursuit Learning Network

环境温度变化造成的复杂漂移(温度漂移)始终是制约光纤陀螺(FOG)性能提高的重要因素。FOG温度漂移本质上是一组与温度有关的多变量非线性时间序列。在这一领域首次采用投影寻踪学习网络(PPLN)方法设计FOG温度漂移在线估计器。相对于传统的神经网络技术,PPLN采用批量学习和参数交替优化的训练算法,可以自适应确定神经网络的规模、参数和神经元函数,不仅具有简捷的网络结构和较强的鲁棒性和模型辨识能力,还可以有效克服学习过程局部极限问题。基于该方法设计PPLN漂移估计器对某型FOG温度漂移进行估计。采用试验实测数据对所提方法进行验证,并采用传统反向传播神经网络(BPNN)的方法进行比较,计算分析结果表明,PPLN漂移估计器具有更好的估计精度和鲁棒性,尤其在陀螺温度不正常变化时对当前漂移的估计精度可以提高至少2倍。

The large drift of fiber optical gyroscope （FOG） caused by variation of environmental temperature is the main factor affecting its performance. Considering the fact that the drift is a group of multi-variable nonlinear time series related with temperature, for the first time, the projection pursuit learning network （PPLN） was employed to model the FOG drift with respect to the environmental temperature. Different from conventional back propagating neural networks （BP NN）, the PPLN uses the batch learning technique to acquire adapting networks size, weights and also hidden unit functions, hence can obtain simpler architecture and better approaching capability. Numerical results from the real drift data of FOG verify the effectiveness of the proposed PPLN-based method. In the situation when unexpected external temperature change happens, the proposed estimator can provide better robustness and more than twice prediction accuracy in comparison with the BP-based method.