The optimization of state of charge and state of health estimation for lithium‐ions battery using combined deep learning and Kalman filter methods

Abstract

Summary An accurate estimate of the battery state of charge and state of health is critical to ensure the lithium‐ion battery's efficiency and safety. The equivalent circuit model‐based methods and data‐driven models show the potential for robust estimation. However, the state of charge and state of health estimation system's performance with a parallel comparison has been rarely investigated. In this study, the performances of state of charge and state of health with equivalent circuit model‐based methods and data‐driven estimations are analyzed by different aged and capacity batteries through methods including extended Kalman filters, fully connected deep network with drop methods, and the combination (extended Kalman filters—fully connected deep network with drop methods). Besides the battery state of the voltage and current, the relationship between inner resistance, temperature, and capacity are also considered. Finally, a suggested method is promising for online state estimation of battery working at different temperatures and initial working state. The results indicate that the maximum state of charge estimation errors of the fully connected deep network with drop methods is 0.56% for the fully charged battery. Simultaneously, with an uncertain initial state of charge, the extended Kalman filter shows the lowest maximum state of charge estimation errors (1.4%). For the state of health estimation, the optimized method uses extended Kalman filters to do the monitor first; after 5 testing points, if the state of health drops to lower than 0.95, extended Kalman filters—fully connected deep network with drop methods is suggested. And finally, estimation errors for this method decreased from 30% to 2%.

A novel collaborative multiscale weighting factor‐adaptive Kalman filtering method for the time‐varying whole‐life‐cycle state of charge estimation of lithium‐ion batteries

“…(4) The voltage is measured after being shelved for 30 min. (5) The voltage is measured when the SOC value equals 0.05, 0.1, 0.15, …, 0.95 sequentially. (6) Steps (1) to ( 5) are repeated to obtain the CCV value for the current rates 0.3, 0.4, and 0.5 C. The voltage changing curve is obtained for different current rates with the CCV difference between the internally connected battery cells for the same SOC level F I G U R E 7 Experimental results under the complex BBDST working condition adaptive to the time-varying charge-discharge current rates, as shown in Figure 9.…”

“…An electrical equivalent circuit model (ECM) is necessary to obtain the battery characteristics. The mathematical expression is a prerequisite for reliable SOC estimation, which is conducted for accurate SOC estimation with the state‐space equation of the ECM model 5 . As for the asymptotic reduction and homogenization, a thermo‐electrochemical model is constructed 6 .…”

“…The mathematical expression is a prerequisite for reliable SOC estimation, which is conducted for accurate SOC estimation with the state-space equation of the ECM model. 5 As for the asymptotic reduction and homogenization, a thermo-electrochemical model is constructed. 6 By building a physics-based model, the SOC estimation is realized by considering the battery aging characteristics.…”

A novel streamlined particle‐unscented Kalman filtering method for the available energy prediction of lithium‐ion batteries considering the time‐varying temperature‐current influence

“…The propagated and updated Cholesky factor can form abrupt sigma points in subsequent iterations. The error covariance correction is used to describe the state variables, according to which the experimental measurement of y k is performed at the time point k, as shown in Equation (22).…”

“…The Recurrent Neural Network illumination searching method is introduced to realize the available energy prediction, which is combined with Particle Filtering (PF) and other methods [19][20][21][22] to improve the residual energy prediction effect of lithium-ion batteries. The temperature compensation modeling strategies can be also considered to predict the energy state of the retractable charger.…”