DVL数据失效时辅助SINS的神经网络改进算法

Improved Neural Network Algorithm for Assisting SINS When DVL Data Fails

摘要:

SINS/DVL 水下组合导航时，受外界因素影响，DVL 信号不稳定和丢失情况时有发生，容易造成定位结果不连续或精度减弱。将 DVL 正常时段采集数据作为训练样本，采用径向基函数神经网络算法（RBF） 对 DVL 丢失时段信号进行填补。为降低系统噪声影响，选择采用扩展卡尔曼滤波（EKF）和自适应渐消 Sage-Husa 扩展卡尔曼滤波（SHEKF）2 种模式进行组合导航计算，得到不同计算结果。分析表明，RBF 算法能够用于处理 DVL 信号丢失情况，相同条件下，SHEKF 滤波模式能够得到更优计算结果，E 方向上位置误差相比 EKF 滤波减少约 50%。

Abstract:

In underwater SINS/DVL integrated navigation，the instability and loss of DVL signal often occur due to external factors，which may easily lead to discontinuous positioning or weakened accuracy. In this paper，the data collected during the normal period of DVL are used as training samples，and the radial basis function neural network algorithm（RBF）is used to fill the signal during the period of DVL loss. To reduce the influence of system noise，two modes of extended Kalman filter（EKF）and adaptive fading Sage-Husa extended Kalman filter（SHEKF） are selected for integrated navigation calculation，and different calculation results are obtained. The analysis shows that RBF algorithm can be used to deal with the loss of DVL signal. Under the same conditions，SHEKF filter mode can get better calculation results，and the position error in the direction of E is reduced by about 50% compared with EKF filter.