Physical Prior knowledge as constraints to predict State of Health for lithium-ion batteries

In the field of battery State of Health (SOH) prediction, machine learning methods have emerged as a research focal point. However, traditional machine learning models often decline in performance with battery aging due to limitations in adaptability across different battery types and conditions. The challenge of developing models with robust generalization capabilities is significant, and to address this, we propose a Constraint Knowledge-based Neural Network (CKNN) method. The CKNN model, developed using an artificial neural network and integrating physical prior domain knowledge-monotonicity and smoothness-thereby enhances the robustness of the SOH estimation. The efficacy of CKNN is empirically validated using open-source and self-tested datasets, affirming the method's better robustness and precision than contrastive models.