Application of adaptive extended Kalman algorithm based on strong tracking fading factor in Stat-of-Charge estimation of lithium-ion battery

• Published in: Energy (Oxford) Vol. 284; p. 129095
• Main Authors: Zhan, Mingjing, Wu, Baigong, Xu, Guoqi, Li, Wenjuan, Liang, Darong, Zhang, Xiao
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2023
• Subjects: Adaptive strong tracking extended Kalman filter; algorithms; energy; lithium batteries; Lithium-ion battery; management systems; Measurement noise covariance; SOC; Strong tracking fading factor; variance covariance matrix; Adaptive strong tracking extended Kalman filter; Lithium-ion battery; Strong tracking fading factor; Measurement noise covariance; SOC
• ISSN: 0360-5442
• DOI: 10.1016/j.energy.2023.129095

Accurately estimating the state of charge (SOC) of a lithium-ion battery is the key to a battery management system (BMS). This paper proposes an adaptive extended Kalman algorithm (ASTEKF) based on strong tracking fade factor to address the issues that the extended Kalman filter algorithm cannot track the system state in real time and cannot accurately estimate the measurement noise covariance matrix when estimating the state of charge of lithium-ion batteries. The extended Kalman algorithm incorporates the fade factor of the strong tracking filter to improve tracking performance. The adaptive algorithm uses the relationship between prior residual and posterior residual to re-determine the value of the fading factor. On this basis, the window-opening method is introduced to ensure the stability of the fading factor while making it self-adaptive. The measurement noise covariance matrix is updated according to the predicted and estimated values, and the measurement noise covariance matrix is estimated and corrected in real time. Finally, a pulse discharge experiment is performed, and the estimated and experimental results are compared. According to the findings, the maximum error of the ASTEKF algorithm in SOC estimation is reduced by 24.6%, the average error is reduced by 25.5%, and the root mean square error is reduced by 64.7% when compared to the conventional EKF algorithm. •The observation noise and the theoretical residual are used as the new conditions to judge the value of the fading factor.•The fenestration factor is updated by the windowing method combined with the prior and posterior residuals.•A method of generating real time positive definite measurement noise covariance matrix from first and second residual.