SOH estimation of lithium-ion batteries based on raw charging data and TCN

Accurate estimation of the state of health (SOH) of lithium-ion batteries is crucial for improving battery lifespan and device reliability. This study proposes a novel SOH estimation method for lithium-ion batteries, which is based on raw charging data and temporal convolutional networks (TCN), enhanced by the integration of a Bayesian optimization algorithm to improve estimation accuracy. Addressing the limitations of traditional data-driven methods in feature engineering and model generalization, this study first establishes a three-dimensional time series data structure to reduce dependence on complex feature engineering and enhance model practicality. Furthermore, a temporal convolutional network that combines stacked causal convolutions and dilated convolutions is designed, which can accurately capture the multi-scale temporal dependency features of batteries during the charging process. Additionally, the Bayesian optimization algorithm is employed to tune the network’s hyperparameters, automating parameter selection and improving the model’s prediction accuracy and generalization capability. Experiments conducted on the open battery dataset from the University of Maryland demonstrate that the proposed method outperforms existing traditional methods in terms of both accuracy and robustness in SOH estimation.