Enhanced Robustness of DFIG Rotor Speed Estimation Using a Correntropy-Based Weighted Extended Kalman Filter

In this paper, we propose a correntropy weighted extended Kalman filter (CWEKF) method to address the challenges of low estimation accuracy and poor robustness in sensorless rotor speed estimation for doubly-fed induction generators (DFIGs). Firstly, based on Faraday’s law of electromagnetic induction and the mechanical motion equation, we derive a DFIG nonlinear state-space model. This model quantifies the sources of nonlinearity arising from cross-coupling terms and product terms, providing a precise model foundation for rotor speed estimation. Secondly, we introduce correntropy theory to design a residual dynamic weighting scheme. By quantifying the local similarity between current and historical residuals, the scheme adaptively adjusts the noise covariance estimation weights, suppressing the interference of outdated data. Combined with the Chi-squared test, we derive an adaptive kernel bandwidth mechanism, balancing the response speed to noise variations and the estimation accuracy in steady-state. Additionally, we further integrate Huber robust weighting and regularization techniques for constructing a hybrid weighting mechanism and optimizing the covariance positive-definiteness correction to address the numerical stability deficiencies of the original algorithm. Using the Lipschitz condition and Lyapunov theory, we prove the mean-square exponential boundedness of the CWEKF estimation error. Finally, we build a DFIG vector control model using MATLAB R2021a and conduct comprehensive experiments, including simulation comparative experiments, open-loop speed identification experiments, and closed-loop sensorless control experiments. Comparative simulation experiments are conducted with EKF, AEKF, and RWEKF under three operating conditions; open-loop experiments verify that the constructed platform meets variable-speed constant-frequency (VSCF) power generation requirements, and closed-loop experiments compare CWEKF with MRAS under different speeds and parameter variations. The results show that the CWEKF has a maximum rotor speed estimation error