基于滑窗自适应的故障估计与补偿及其在组合导航系统中的应用

Fault Estimation and Compensation Based on Sliding Window Adaptive and Its Application in Integrated Navigation System

针对以往方法对缓变故障存在较大的故障检测延时,从而降低导航系统可靠性的问题,实时的故障估计及补偿十分必要。将故障动力学建模为随机过程,提出了一种基于不准确状态和测量噪声协方差矩阵自适应学习的滑动窗口变分自适应卡尔曼滤波器。它由前向卡尔曼滤波、后向卡尔曼平滑和噪声协方差矩阵的在线估计组成。通过对滑动窗口状态向量的平滑后验分布进行逼近,利用变分贝叶斯方法将噪声协方差矩阵的后验分布解析更新为逆Wishart分布,避免了不动点迭代,获得了良好的计算效率。最后,基于捷联惯导/卫星(SINS/GPS)组合导航系统,进行了导航系统的故障估计及补偿的仿真验证。仿真结果表明,在不需要了解量测噪声和故障系数矩阵的情况下,该方法能够较好地估计出目标故障信号,然后利用估计出来的故障值对导航系统进行补偿,极大地提高了SINS/GPS组合导航系统的实时估计精度与可靠性。

In view of the problem that the traditional methods have a large fault detection delay for slow fault and thus reduce the reliability of navigation system,real-time fault estimation and compensation are very necessary.In this paper,fault dynamics is modeled as a random process and a sliding window variational adaptive Kalman filter is proposed based on adaptive learning of inaccurate state and measured noise covariance matrix.It consists of forward Kalman filtering,backward Kalman smoothing and online estimation of noise covariance matrix.By approximating the smooth posterior distribution of the sliding window state vector,the posterior distribution of the noise covariance matrix is analytically updated to the inverse Wishart distribution by using the variable decibel Bayesian method,which avoids fixed point iteration and gets good computational efficiency.Finally,based on SINS/GPS integrated navigation system,the fault estimation and compensation of navigation system are verified by simulation.Simulation results show that this method can estimate target fault signals well without knowing the measurement noise and fault coefficient matrix,and then use the estimated fault values to compensate the navigation system,which greatly improves the accuracy and reliability of SINS/GPS integrated navigation system.