Deep learning and polarization equilibrium based state of health estimation for lithium-ion battery using partial charging data

With the rapid advancement of battery technology, the application of lithium-ion batteries in automotive and energy storage sectors has significantly increased. In order to ensure the safety and stability of batteries during operation, as well as to assist in the recycling and reuse, accurate estimation of battery state of health (SOH) has become crucial. Firstly, this paper proposes a new estimation method in the voltage–differential voltage (v−Δv) coordinate space, which is more concise and practical than existing methods, by integrating the degradation mechanisms of batteries and the data characteristics during constant current (CC) charging processes. Secondly, based on the polarization equilibrium characteristics of batteries during charging and the data characteristics of partial charging data with polarization processes, a comprehensive estimation strategy for partial charging data with polarization processes is proposed, which is closer to real-life scenarios compared to traditional partial charging data estimation methods. Subsequently, an adaptive sampling deep neural network (ASDNN) is designed and established according to the proposed estimation strategy. After multiple validations, with no overfitting observed, the average absolute error (MAE) is 0.0145, the root mean square error (RMSE) is 0.0185, and the average estimation time is 0.0117 ms.