Closed-loop state of charge estimation of Li-ion batteries based on deep learning and robust adaptive Kalman filter

The state of charge (SOC) is among the most crucial monitoring states in battery management systems. To accurately and robustly estimate the battery SOC, a closed-loop SOC estimation method based on deep neural networks is proposed in this study. Initially, a hybrid deep learning model, TGMA, is established by combining a temporal convolutional network (TCN) with a gated recurrent unit (GRU) and incorporating a multi-head self-attention mechanism (MHA) to estimate SOC. Subsequently, to further enhance the accuracy of estimating the battery SOC and the ability to resist outliers, the pre-estimated SOC from the deep learning model TGMA is used as the noisy observation in the Kalman filter, and robust factor and noise adaptive filtering are introduced into the Kalman filter, proposing a closed-loop estimation method based on deep learning and robust adaptive Kalman filter (RAKF). Finally, through comparison with other similar methods, the proposed TGMA-RAKF method is validated to have higher accuracy, generalization, and robustness against outliers under different operating conditions and temperatures, with root mean square error, mean absolute error, and maximum absolute error all less than 1 %, 1 %, and 2.5 % respectively.