Bayesian filtering is a state estimation method based on the Bayesian inference framework, which can be applied to state estimation of linear, nonlinear, and non-Gaussian systems, and is an essential tool in signal processing. However, modern dynamic systems often exhibit complexity and uncertainty, making it challenging for traditional Bayesian filtering to perform robust estimation under such conditions. With the rapid rise of artificial intelligence technologies, renowned for their powerful ability to extract features from statistical data, data-driven deep learning approaches have introduced new perspectives to traditional model-based estimation methods. This has led to the emergence of hybrid methods that combine data-driven and model-driven approaches with distinct functionalities. This paper summarizes and organizes these methods, elaborates on their advantages, and discusses future development directions.
Liu et al. (Thu,) studied this question.