State of Charge Estimation of Power Batteries Based on a Multi-Model Fusion Method

Power batteries have significant time-varying nonlinearity, resulting in significant differences in the estimated state of charge (SOC) obtained from different equivalent circuit models. A fusion method for the SOC estimation of power batteries based on multiple model probabilities is proposed in this paper. Firstly, select three battery models, namely Thevenin model, DP model, and fractional order model, and the forgetting factor recursive least squares method(FFRLS) is used to identify the model parameters. Secondly, considering that traditional FFRLS parameter identification methods rely on fixed forgetting factors in the algorithm, the algorithm is optimized into a variable forgetting factor recursive least squares (VFFRLS) and is used for online identification of model parameters. Thirdly, the adaptive unscented Kalman filtering(AUKF) technique is applied to estimate the SOC of each of the three models shows that SOC estimation outcomes based on a sole model cannot always guarantee reliability. Additionally, to address the issues of low utilization of historical data when using the UKF for SOC estimation, as well as the potential failure of the algorithm in simulations due to non positive definite covariance matrices, a novel SOC framework is proposed. This framework combines the Multiple Innovations (MI) theory, Singular Value Decomposition (SVD), and Adaptive Unscented Kalman Filter (AUKF) algorithm, namely SVD-AMIUKF, and employs a Bayesian-based probabilistic method to integrate the SOC estimates obtained from these three models with optimal weighting, achieving accurate SOC estimation. Finally, the proposed model is validated under different experimental conditions. The experimental results show that under DST working conditions, the average absolute percentage error (MAPE) of the proposed model is 0.00753. Compared with the estimation results of the Thevenin model, DP model, and fractional order model, the estimation accuracy has improved by 69.8%, 12.9%, and 55%, respectively. Under FUDS working conditions, the proposed model estimates a MAPE of 0.0028 for SOC. Compared with the above three models, the estimation accuracy has been improved by 70.5%, 69.8%, and 74.5% respectively, which verifies the effectiveness of the proposed method.