Dynamic adaptive square-root unscented Kalman filter and rectangular window recursive least square method for the accurate state of charge estimation of lithium-ion batteries

Building an accurate battery model and realizing a high-precision state of charge (SOC) estimation is the key to ensuring the safe and efficient operation of electric vehicles. In this paper, a novel dynamic adaptive square-root unscented Kalman filtering (DSAR-UKF) method is proposed to accurately estimate the SOC of lithium-ion batteries. It adaptively identifies the battery parameters online based on a rectangular window recursive least squares (RW-RLS) method. To solve the problem of data saturation of the traditional RLS method, the novel framework of the RW-RLS method is constructed based on a second-order Thevenin equivalent circuit model to simulate and monitor the battery's characteristics. Aiming to eliminate the uncertainties and inaccuracies caused by the statistical characteristics of the measurement noise of the time-varying system in the traditional square-root unscented Kalman filtering (SR-UKF) method, a noise covariance matching method based on an adaptive filter is introduced. The method uses a dynamic threshold adjustment factor to modify the window size and realize the real-time correction of the system noise matrix. The SOC estimations are carried out at noise interference and ambient temperature under complex hybrid pulse power characterization (HPPC), Beijing bus dynamic stress test (BBDST), and dynamic stress test (DST) working conditions to verify the effectiveness of the proposed method. The verification results show that the average error of RW-RLS method is 0.00503 V, the relative error is 0.119 %, which is less than that of the traditional RLS method. The rationality of the proposed identification method is verified. At room temperature, the estimated value of SOC of each method is closer to the true value of SOC. By comparing different methods, the DSAR-UKF method has fewer error values and fast computation, and its maximum error is 0.878 %, 0.822 %, and 0.415 % respectively, which satisfies the need for critical SOC estimation. The introduction of the adaptive noise and dynamic threshold adjustment factors improves the accuracy and stability of the proposed method, which provides a theoretical basis for the efficient operation of the battery management system.