It can be seen from Figure 1 that the RMSEs obtained by the ASCKF are significantly smaller than those of the other two filters, indicating that the proposed ASCKF can achieve higher estimation accuracy compared with CKF and SCKF. The reason is that the mean and time-varying covariance of the noise can be estimated online effectively to revise the filter by the noise statistic estimator in ASCKF, as shown in Figure 2.
It can be seen from the table that the three states of ASCKF are improved by 38.55%, 32.61%, and 16.19%, respectively, compared with the SCKF, thus verifying the validity of the proposed filter.
The simulation results show that the proposed ASCKF has achieved better performance compared with the CKF and the SCKF in the case of inaccurate time-varying noise statistic, and the validity of the proposed filter has been verified.
Filters State 1 State 2 State 3 CKF 0.4693 0.3410 0.1423 SCKF 0.4677 0.3404 0.1421 ASCKF 0.2874 0.2294 0.1191
The IMM approach carries out a "soft switching" between the model-based SCKF and the model-based SCRHKF by the model probability.
The SCKF and SCRHKF are processed to estimate the vehicle sideslip angle, respectively.
The initial model probabilities for the model-based SCKF and the model-based SCRHKF are identical with [[mu].sup.1.sub.0] = [[mu].sup.2.sub.0] = 0.5.
The results based on the hybrid Kalman filter follow the trend of measured values quite well and they are always within an acceptable range, while the estimated lateral velocity and sideslip angle using SCKF generate some large errors due to the modeling mismatch in nonlinear region and sensor noise.
The SCKF is influenced by the modeling mismatch and sensor noise and shows some large errors for the estimated lateral velocity and sideslip angle.
The proposed method combines the merit of SCKF that has high convergence speed and numerical stability with the merit of SCRHKF that has great robustness against model uncertainty and temporary noise.