基于RBF网络的低成本惯组中陀螺仪的温度补偿

RBF Networks for Temperature Compensation of Gyros in Low-Cost SINS

针对在低成本捷联惯性导航系统中动力调谐陀螺仪的输出信号随温度漂移严重的问题,使用径向基函数神经网络进行补偿。采用RBF神经网络建立温度补偿模型,以温度信号为网络的输入训练样本,以陀螺仪的温度漂移为网络的目标输出。神经网络的结构为输入层和输出层各有1个节点,中间隐层含有4个节点,隐层节点的聚类中心均匀分布在温度的变化范围之内。结果表明,该方法能有效将低成本惯组中动力调谐陀螺仪输出信号中的温漂误差减小一个数量级以上,从而提高惯组的导航精度。

In the view of serious temperature excursion of dynamically tuned gyros in a low-cost strap-down inertial navigation system, a method of using RBF neural network to compensate the temperature excursion is presented. Use RBF neural network to build the compensation model, the network is trained by the temperature signal, and its output is the value of the gyro＇s bias. The input layout and the output layout have one node separately, the middle layout has four nodes, and the aggregation center points of the nodes in the middle layout are scattered evenly among the changing area of the temperature. It＇s proved by the experiment that this method could reduce the gyro＇s temperature excursion at least 1 magnitude in comparison with traditional method, and could improve the accuracy of the low-cast strap down inertial navigation system.