GNSS失锁下基于混合预测模型的POS误差估计方法

Aiming at POS observation loss caused by GNSS signal occlusion and interference, a hybrid prediction method based on multiple linear regression (MLR) and radial basis function neural network (RBFNN) is proposed for the prediction of POS errors during GNSS outages. Hodrick-Prescott (HP) filter is utilized in the proposed method to decompose the POS errors sample data into the trend and fluctuation series, which are separately intended for use with MLR and RBFNN modeling in order to fully characterize both the linearity and nonlinearity of POS errors. The results of vehicle test show that the percentage improvement of the proposed MLR/RBFNN hybrid prediction method in position and velocity accuracy is found to improve by 72.9%~89.1% and 50.1%~60.8% compared with the standard Kalman filter, which is superior remarkably to the single MLR and RBFNN.