期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2012
卷号：45
期号：2
页码：535-541
出版社：Journal of Theoretical and Applied
摘要：A novel nonlinear filter called model-enhanced Gaussian process square root cubature Kalman filter (MEGP-SRCKF) is proposed to estimate the state of nonlinear dynamic systems where their state-space models are unknown or insufficiently accurate. The algorithm integrates Gaussian process regression (GPR) into square root cubature Kalman filter (SRCKF). Given the training data, GPR model is used to learn and represent the residual of system after factoring the contributions of the parametric model. The combination of GPR and parametric models enhances the performance of either model alone. It improves the accuracy of the transition and measurement models. The resulting MEGP-SRCKF algorithm has several advantages over standard extended Kalman filter (EKF), nonaugmented unscented Kalman filter (UKF), augmented UKF, and SRCKF. Two cases are used to test these filters and the superiority of the proposed filter is demonstrated, where the MEGP-SRCKF can obtain the better results whether an accurate parametric state-space model is obtained.
关键词：Bayesian Filter; Cubature Rule; Gaussian Process Regression; Model-Enhanced; Parametric Model; State-Space model; Nonlinear Dynamic System
