摘要： 随着对惯性导航系统中对准时间要求的不断提高，初始对准需要在大方位失准角条件下进行，此时需采用非线性滤波方法来实现初始对准。基于此，提出高斯过程回归平方根中心差分卡尔曼滤波算法(GP-SRCDKF)。将高斯过程回归融入到SRCDKF算法中，利用高斯过程得到系统回归模型及噪声协方差，用回归模型代替状态方程和观测方程，对相应的噪声协方差进行实时自适应调整。该算法不仅克服了扩展卡尔曼滤波滤波精度低、需要计算雅可比矩阵的不足，而且可解决传统滤波容易受系统动态模型不确定和噪声协方差不准确的限制。仿真实验结果验证了该算法的有效性和优越性。

关键词: 大方位失准角, 捷联惯导, 初始对准, 高斯回归, 高斯过程回归平方根中心差分卡尔曼滤波, 自适应调整

Abstract: In order to improve the alignment time, initial alignment is carried on with large azimuth misalignment, and the nonlinear filtering methods are utilized. Therefore Gaussian Process regression Square Root Central Difference Kalman Filtering(GP-SRCDKF) is proposed, and which is taken Gaussian process regression into SRCDKF algorithm to get system regression model and noise covariance, regression model is taken instead of state equation and observation equation, and the corresponding noise covariance makes real-time adaptive adjustment, which not only overcomes the deficiencies that Extended Kalman Filtering(EKF) has low precision and needs to calculate the Jacobian matrix, but also solves the problems that traditional filter is limited by the uncertain system dynamic model and inaccurate noise covariance. Simulation results verify the effectiveness and superiority of the proposed algorithm.

Key words: large azimuth misalignment, strapdown inertial navigation, initial alignment, Gaussian regression, Gaussian Process regression Square Root Central Difference Kalman iltering(GP-SRCDKF), adaptive adjustment

中图分类号: 

• TP18