Deep learning model-based real-time state-of-health estimation of lithium-ion batteries under dynamic operating conditions

Accurate real-time estimation of the State-of-Health (SOH) of lithium-ion batteries is crucial for the safe and reliable operation of fully electric vessels. However, estimating SOH in real time under complex working conditions is challenging. This work proposes a deep-learning model for precise real-time SOH estimation during ship navigation. Firstly, we analyze public battery datasets from NASA, MIT, and CALCE under various working conditions, extracting universal health factors from partial charging and discharging data. The distance correlation coefficient verifies the correlation between factors and SOH sequences. Next, an improved CNN-BiGRU-AM model is developed, using the Kepler Optimization Algorithm (KOA) to determine hyperparameters. The model is trained with battery datasets under diverse conditions, using factors as inputs and real SOH sequences as outputs, to establish prior knowledge of the mapping between factors and SOH under varied operating conditions. Finally, the model is tested using batteries under both simple and complex conditions. The results show that the model achieves a maximum absolute error of only 1.4 % in SOH estimation under simulated ship sailing conditions, with MAE and RMSE evaluation indexes below 0.7 %. The proposed model outperforms existing methods under complex conditions.