BP神经网络辅助的SINS/GPS组合导航姿态误差补偿方法研究

Research on Attitude Error Compensation of SINS/GPS Integrated Navigation Based on BP Neural Network

针对SINS/GPS组合导航姿态估计精度不足问题,提出一种BP神经网络辅助的导航姿态估计的补偿方法。首先,以QR分解Kalman滤波的增益矩阵并构建其降维特征向量;然后,以此为输入,以姿态估计误差为期望输出对BP神经网络进行训练;最后,利用BP神经网络的输出辅助修正SINS/GPS组合导航的姿态估计结果。数值仿真表明,相对于仅依赖传统Kalman滤波的方法,使用降维特征向量训练的BP神经网络获得补偿误差的SINS/GPS组合导航系统姿态估计可大大降低计算耗时,同时精度可提高2个数量级,对提高SINS/GPS组合导航精度具有较高参考价值。

To enhance accuracy of attitude estimation in the SINS/GPS integrated navigation system,a compensation algorithm for attitude estimation based on BP neural networks is proposed.Firstly,the dimensionality-reduced characteristic vectors input to a BP neural network are formed by QR decomposition of the gain matrix obtained during Kalman filtering.Secondly,taking the characteristic vectors as the input,the BP neural network is trained with attitude estimation error as the expected output.Thirdly,the attitude estimation results of the SINS/GPS integrated navigation system are calibrated by the BP neural network output.Finally,numerical simulation results show that,compared to those with common Kalman filters,computation time is greatly reduced under the proposed method,while attitude estimation accuracy is improved by two orders of magnitude higher.These technical features are significant and useful for improving the attitude estimation accuracy of SINS/GPS integrated navigation systems.