State of art on state estimation: Kalman filter driven by machine learning

Abstract

The Kalman filter (KF) is a popular state estimation technique that is utilized in a variety of applications, including positioning and navigation, sensor networks, battery management, etc. This study presents a comprehensive review of the Kalman filter and its various enhanced models, with combining the Kalman filter with neural network methodologies. First, we provide a brief overview of the classical Kalman filter and its variants, including the extended Kalman filter (EKF), unscented Kalman filter (UKF), cubature Kalman filter (CKF), etc. It is pointed out that the traditional Kalman filter faces two main problems: system model and noise model parameter identification. To overcome these obstacles, researchers have developed novel solutions by integrating machine learning techniques with the Kalman filter. Secondly, this paper classifies the related models into two categories: both the internal cross-combination of the Kalman filter and neural network and their external combinations. Two different hybrid models and typical structures show that the hybrid model performs more accurately and robustly overall. Finally, the characteristic of the two hybrid models is summarized so that readers can understand them more intuitively.