State-of-Health Estimation With Random Constant-Current Charging Segments of a Lithium Battery Using an Improved Transfer Learning Algorithm

An accurate state of health (SoH) is important for the optimal performance and safety of Li-ion battery (LIB) storage systems. In this study, an estimation method for SoH based on random constant-current charging segments is proposed. First, a new SoH indicator based on the time interval of unit charging voltage differences is proposed to simulate random constant-current charging segments. Second, on the basis of transfer learning (TL), an improved domain adaptive TL algorithm is proposed to avoid negative transfer and reduce the training load of the network. Again, on the basis of the new SoH indicator and neural network, an estimation approach based on random constant-current charging segments is established. Finally, the progressiveness of the proposed method is validated on test data from a self-built experimental platform. The verification results show that the proposed method has a different degree of reduction in estimation error than the same type of methods do.