Data-Driven State Estimation and Robust Nonlinear Filtering
组织者
演讲者
陶飏天择
时间
2024年03月27日 15:00 至 15:30
地点
理科楼A-304
摘要
State estimation is a crucial topic in automatic control and signal processing, aiming to accurately estimate hidden states or parameters in the presence of noise. This paper focuses on two main aspects: Real-time and efficient algorithms for high-dimensional nonlinear state estimation: Using a data-driven approach, filtering and smoothing problems are transformed into sequential optimization problems. Neural networks are employed to approximate the underlying space, enabling offline training and online inference. Theoretical proofs of convergence and bounds on estimation errors are provided.
Robust nonlinear filtering algorithms: Addressing uncertainty in state space models, particularly inaccuracies in noise modeling. A framework for robust iterative extended Kalman filtering is proposed, considering cases with noise outliers. Theoretical contributions include proving outlier robustness and providing estimates for estimation errors.
Numerical tests validate the proposed algorithms, demonstrating their effectiveness compared to commonly used state estimation methods.