This is the authors' abstract. We don't add key points for this paper.
Accurate estimation of battery state of charge (SOC) under complex and varying operating conditions remains challenging. This paper proposes a variational Bayes-based multiple neural network (VBMNN) method for battery SOC estimation. Within the proposed algorithm, a hierarchical probabilistic model based on Dirichlet and categorical distributions is constructed to characterize the model indicator vector and model probability vector. Variational Bayesian inference is employed to obtain approximate posterior distributions of the state, model indicator, and model probability vector. The posterior expectation of the model indicator serves as the fusion weight to combine the information provided by different models, thereby forming a single state posterior distribution. Through a pseudo-measurement formulation, neural networks trained under different operating conditions are incorporated into recursive state estimation as measurement models. In addition, a Student’s t noise model with a latent Gamma variable is introduced to mitigate the effects of outliers and non-Gaussian measurement noise. Experimental results obtained from the tested battery show that VBMNN achieves root mean square errors (RMSEs) of 0.40%, 0.49%, and 0.49% under three driving cycles not used for training. After artificial non-Gaussian noise is introduced, the corresponding RMSEs remain below 0.60%, demonstrating the robustness of the proposed algorithm. Additional tests at different temperatures and cross-cell validation further demonstrate its adaptability to operating conditions not encountered during training and to cell-to-cell variations.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Zhao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: