融合高斯过程回归的UKF估计方法

高精度滤波估计是SINS/GNSS组合导航系统的关键技术之一，其估计精度直接影响了导航精度。传统滤波估计方法一般只基于惯导误差模型，未考虑惯导误差模型不确定性的影响。针对此问题，提出了一种采用高斯过程回归（GPR）增强无迹卡尔曼滤波（UKF）预测和估计能力的高精度滤波估计方法。一方面，能在有限的训练数据条件下通过UKF估计误差状态量；另一方面，高斯过程既考虑了噪声，也考虑了UKF的不确定性。将所提方法应用于SINS/GNSS组合导航系统中，车载实验结果表明，所提方法能有效提高滤波估计精度。 

UKF estimation method incorporating Gaussian process regression

The high-precision filter estimation is a key technology in the SINS/GNSS integrated navigation system, and its estimation accuracy has direct influence on the accuracy of navigation. The traditional filter estimation method is based on inertial navigation error model, and does not take its uncertainty into account. Aimed at the problem, a high-precision filter estimation method is presented, which uses Gaussian process regression (GPR) to enhance the capabilities of prediction and estimation for parametric unscented Kalman filter (UKF). On one hand, it can estimate the state vector of the nonlinear parametric UKF on condition that trained data is limited; on the other hand, GPR can also take both the noise and the uncertainty in the nonlinear parametric UKF into consideration. The real vehicle-mounted experiment results show that the proposed method can effectively enhance filter estimation precision through applying the enhanced GPR-UKF into the SINS/GNSS integrated navigation system.