Adaptive State Estimation Using Kalmannet for Systems with Time-Varying Noise

This paper addresses the challenging state estimation problem in real-world systems characterized by nonlinearities and time-varying measurement noise. Data-driven approaches like KalmanNet, while enhancing state estimation for nonlinear systems, struggle to adapt to changing noise environments. To overcome these limitations, this paper proposes a noise-parameter adaptive KalmanNet state estimation method that combines the strengths of both model-driven and data-driven paradigms, enabling adaptive state estimation in dynamic environments with time-varying measurement noise. Specifically, the proposed method leverages data-trained networks to extract the state transition characteristics of nonlinear systems and designs a dual model and data driven state estimation framework. This framework decouples the prior state error covariance from Kalman gain computation. Furthermore, an online adaptive estimation strategy incorporating Bayesian inference is introduced to estimate time-varying measurement noise, thereby improving the robustness and adaptability of the model's state estimation performance in dynamic scenarios.