基于LS-SVM的捷联大失准角初始对准技术

An initial alignment technique for strapdown inertial navigation systems based on large misalignment angles

针对捷联惯性导航系统（SINS）大失准角初始对准情况下非线性模型线性化导致模型不准确和影响对准精度的问题，设计了一种基于最小二乘支持向量机（LS-SVM）的大失准角对准算法。该方法采用基于加性四元数误差（AQE）的大失准角误差方程，采用简化的无迹卡尔曼滤波器（UKF）来模拟LS—SVM训练样本。捷联惯性导航系统和全球定位系统（GPS）的速度和位置误差作为LS-SVM的输入样本，简化UKF得到的失准角经小波去噪后作为输出样本。LS—SVM算法采用交叉验证法选择最佳的核函数参数。仿真结果表明，在大失准角下LS-SVM算法在对准时间和对准精度上与简化UKF和EKF相比均表现出较好的性能。

To solve the problem that the linearization of the nonlinear model causes the model inaccurateness and the influences on alignment accuracy during the initial alignment of a strapdown inertial navigation system （SINS） in the circumstances of the large misalignment angle, the paper proposes an algorithm for large misalignment angle alignment based on least squares support vector machines（LS-SVM）. The method introduces large misalignment angle error equations based on additive quaternion error（ AQE）, and uses the simplified unscented Kalman filter（SUKF） to simulate training samples. The velocity and position errors between SINS and GPS are set as input samples of the LS-SVM and misalignment angles which are the outputs of the SUKF are set as output samples after de-noised by wavelet. Cross validation is used to choose the best kernel function parameters. The simulation results demonstrate that the LS-SVM algorithm has the better performance on alignment time and accuracy than the SUKF and EKF.