Joint estimation of SOH and SOC of lithium battery based on feature fusion model

Abstract To address the issue of low accuracy in independently estimating the state of health (SOH) and state of charge (SOC), and considering the mutual influence between SOH and SOC, this paper proposes a joint estimation method for SOH and SOC based on a cascaded feature fusion model. This approach mitigates the impact of the reduced maximum usable capacity due to battery aging on SOC estimation accuracy. First, the improved Aquila optimization algorithm is employed to optimize the initial weights and thresholds of back propagating neural network, effectively enhancing SOH estimation accuracy under small-scale data; Second, the estimated SOH values are fed as features into the SOC estimation model, which consists of a temporal convolutional network (TCN), bidirectional gated recurrent units (BiGRUs), and an attention mechanism. The TCN extracts multi-scale temporal features through dilated convolutions, the BiGRU further learns temporal dependencies between features, and the attention mechanism focuses on key features. This approach reduces the impact of battery aging on SOC estimation. Finally, the model was trained and tested on the NASA lithium-ion battery dataset. Extensive experimental results demonstrate that the proposed joint estimation method outperforms the independent SOC estimation model under various temperatures and discharging conditions. For the B5 battery, the root mean square error of SOC estimation was reduced by 2.65% and 3.66% during charging and discharging, respectively. Furthermore, generalization and comparative experiments validate the robustness and superiority of the proposed framework.