An Integrated Knowledge Distillation Framework for Battery SOH Estimation

State of health (SOH) estimation plays an important role in Battery Management Systems (BMS), especially in electric vehicles (EVs). However, deploying deep learning-based SOH estimation models in EVs is challenging due to their large computational demands and model sizes. Moreover, battery datasets often contain environmental noise, which affects the accuracy of predictions. We propose an integrated knowledge distillation (KD) framework for model compression. Under this framework, the size of the model is compressed by knowledge distillation. This framework uses the Wasserstein distance to solve the extreme value problems during the knowledge distillation process. Additionally, we refine Wasserstein distance with a Gradient Penalty (G-P) technique. This method controls the gradients of the model outputs within a certain range, which can solve the gradient explosion or disappearance problems. The experimental results indicate the validity of our framework.