Online State of Charge Estimation for Battery Based on the Improved Double Fraction-Order Multi-Innovation Unscented Kalman Filter With Full Parameters Identification and Self-Rectification of Covariance Matrices

Abstract

The state of charge (SOC) is an extremely important indicator of the lithium cell. Currently, traditional identification algorithms and integer-order models are poorly adaptive and unable to respond to the dynamic characteristics of batteries in real time, whereas unscented Kalman filtering is often accompanied by the problem that the covariance matrix is not positively determined, leading to algorithmic collapse. This study combines the multi-innovation theory with online identification and provides an improved fractional-order multi-innovation double unscented Kalman filter (DFOMIUKF) joint estimation algorithm. The algorithm contains a multi-timescale framework that allows full parametric identification of fractional-order models, which guarantees the algorithm's ability to cope with complex working conditions, and the problem of failure of the Kalman filtering algorithm is solved by modifying the covariance matrix in real time by singular value decomposition (SVD). On the basis of the fractional RC model, the FOMIUKF₁ algorithm allows real-time updating of model parameters on a macro scale, and the obtained parameters at the microscale are passed to the FOMIUKF₂ algorithm for real-time updating of the charge level of the lithium battery. Then, the DFOMIUKF algorithm is validated under different working environments, respectively. The findings indicate that the proposed algorithm predicts the SOC with the maximum values of root mean square error (RMSE) and mean absolute error (MAE) not exceeding 0.77% and 0.66%, respectively, and maintains high accuracy even at low temperatures. It is illustrated that the proposed algorithm solves the problems of the above offline algorithms with high precision and robustness.