Joint Estimation Method of SOC and SOH Based on Fusion of Equivalent Circuit Model and Data-Driven Model

The accuracy of state of charge (SOC)can be significantly affected by battery aging, leading to misguidance in the calibration of state of health(SOH). Existing studies often estimate SOC and SOH separately, neglecting their close relationship and resulting in reduced estimation accuracy. This paper proposes a joint estimation method for SOC and SOH based on the fusion of an equivalent circuit model and a data-driven model. The influence mechanism between battery SOC and SOH is revealed, mitigating their mutual influence and enhancing the accuracy of SOC and SOH estimation. Firstly, by constructing a second-order RC equivalent circuit model of the battery considering aging and SOC, the recursive least square method with a forgetting factor isused to identify battery parameters online under different SOC and SOH conditions. Secondly, the required time from 20%SOC to the end of the constant-current charging stage is extracted. Pearson and Spearman relationships between constant current charge time and SOH of lithium-ion batteries arecalculated. Thirdly, the actual time required from 20%SOC to the end of the constant-current charging phase of lithium-ion batteries is taken as input and battery SOH as output to train the GPR model offline. The trained GPR model is optimized by hyper parameters and used for SOH prediction. Finally, the estimated SOH output ismultiplied by the rated capacity of the cell to obtain the actual cell capacity, which is used to update the second-order RC state space equation. Based on the second-order RC equivalent circuit model, the battery SOC was estimated by the EKF. The Oxford University battery degradation data set and NASA random battery data set are used to verify the joint estimation method. The results show that the proposed method achieves low average MAE and RMSE for SOC estimation (typicallyless than 0.04). In aging experiments of Cell 1～Cell 8 and RW 3～RW 6 under different working conditions, the average MAE and average RMSE are stable. The actual initial SOC value is 1, and the initial value is set to 0.7 in this paper. With the decline in battery capacity, the joint estimate of battery SOC can follow the actual SOC more accurately. The joint estimation algorithm is robust and accurate. Meanwhile, the reservation-one method is used to verify the Gaussian process regression model. The MAE and RMSE predicted by SOH for Cell 1～Cell 8 are less than 0.5%, and the MAE and RMSE predicted by SOH for RW 3～RW 6 are about 0.05. All the predicted SOH values are in a narrow confidence interval. The following conclusions can be drawn from the simulation analysis: (1) Compared with the existing battery model, the dynamic second-order RC equivalent circuit model considering battery aging and SOC is constructed. In the case of battery aging, the voltage obtained by fitting the identified circuit parameters can track the actual voltage well. (2) The joint estimation method applies the real-time online modified battery parameters and battery SOH to ensure that the battery SOC is adjusted with battery aging. The SOC estimation is accurate. (3) The combined method applies the estimated SOC to ensure effective health feature extraction and improve the accuracy of SOH prediction. © 2024 China Machine Press. All rights reserved.