Improved Battery Degradation Modelling with Coupled Physical and Machine Learning Modelling

The ability to accurately estimate the state-of-health (SOH) of Lithium-ion (Li-ion) batteries is important for assuring their optimal operational management, reliability, and safety compliance. This is a challenging problem because various internal and external ageing mechanisms occur and lead to battery capacity loss. In this paper, a fusion approach that combines an integrated empirical model for battery degradation with a back propagation neural network model (BPNN) is developed. The integrated empirical model for SOH estimation is built upon a simplified capacity fade model and an internal resistance growth model. The error in the integrated empirical degradation model were corrected and updated using a BPNN model. The difference between the empirical model predictions and the actual measured battery SOH values was used as output target in the BPNN model. The accuracy of this fusion model was evaluated using different performance criteria and also compared against with other non-fusion models. The result shows that the fusion model improved the accuracy of SOH estimation in all tested battery set with minimal error.