Dual-optimization for a MEMS-INS/GPS system during GPS outages based on the cubature Kalman filter and neural networks

Abstract

Abstract To improve the performance of a microelectromechanical-system (MEMS)-based inertial navigation system (INS)/Global Positioning System (GPS) integrated navigation system, a dual optimization scheme comprising a cubature Kalman filter (CKF)-multiple layer perceptron (MLP) and radial basis function (RBF)-CKF is proposed for the compensation of position and velocity errors during GPS outages. The proposed method has advantages: (i) The generalization ability of the CKF-MLP is much better than that of other methods, such as the extended Kalman filter (EKF)-MLP and unscented Kalman filter (UKF)-MLP, and the proposed CKF-MLP provides highly accurate position information even during long GPS outages; (ii) The accuracy of the error estimation of the RBF-CKF is higher than that of other neural-network-assisted CKF methods, such as the Adaboost-CKF and random forest (RF)-CKF, and the proposed CKF-MLP can estimate the weights of the MLP adaptively and achieve an appropriate internal structure when the GPS signal is available; (iii) The RBF is added to the CKF to establish the relationship between filter parameters and estimation error in the proposed RBF-CKF; (iv) During a GPS outage, position errors and velocity errors are predicted and compensated for in the proposed dual optimization scheme. Field test data are collected to evaluate the proposed solution. Experimental results show that (i) the root-mean-square error of the velocity for an eastern position is reduced by 77% to 0.34 m/s using the proposed CKF-MLP strategy; (ii) the root-mean-square error of a northern position determined by the proposed RBF-CKF remains at 23.11 m even during long outages of 500 s, and the RBF-CKF effectively suppresses the divergence seen for other methods; and (iii) the dual optimization process using different estimators provides better error compensation results than a single optimization method, which demonstrates that the proposed solution leads to the better performance of a MEMS-based INS/GPS navigation system.