In engineering applications, a large number of sensors are needed to observe the system status. However, we have to face the observation challenges brought by sensor noise or damage in some exceptional cases. Because of the strong nonlinearity and uncertainty, it is difficult to build the model of dynamics. Aiming at this problem, this paper proposes a data-driven Kalman filter with deep neural networks for state estimation with unknown nonlinear dynamics. The main idea is to first represent the nonlinear systems in the feature space via deep neural networks and use the linear Kalman filter design to construct full state information. We test the proposed algorithm for state estimation of ground vehicle dynamics. The simulation results show that our algorithm is valid in a large portion of state space. We also compare our approach with the Extended Kalman Filter and report it has better tracking and stability in modeling accuracy.
Data-Driven Kalman Filter for Nonlinear Systems with Deep Neural Networks
Lect. Notes Electrical Eng.
International Conference on Autonomous Unmanned Systems ; 2021 ; Changsha, China September 24, 2021 - September 26, 2021
Proceedings of 2021 International Conference on Autonomous Unmanned Systems (ICAUS 2021) ; Chapter : 272 ; 2777-2784
2022-03-18
8 pages
Article/Chapter (Book)
Electronic Resource
English
Diffusion nonlinear Kalman filter with intermittent observations
SAGE Publications | 2018
|Nonlinear filters: beyond the Kalman filter
IEEE | 2005
|Nonlinear Filters: Beyond the Kalman Filter
Online Contents | 2005
|