CNN and transfer learning based online SOH estimation for lithium-ion battery
Yang Li, Jili Tao
Abstract
Accurate estimation of state of health (SOH) is extremely important for lithium-ion (Li-ion) rechargeable batteries. An improved strategy based on convolutional neural network (CNN) architecture is proposed for online SOH estimation, in which the features can be automatically extracted, instead of manual extraction. Accelerated aging data from different dynamic conditions, including overcharging and over-discharging, is utilized to pretrain a base model. And then by transfer learning method, the base model can be fine-tuned with just 15% normal-speed aging data and migrated as a new model for testing on the remaining 85% normal-speed aging data. The transfer learning method can reduce the laboratory costs for large amount of cycling data. Only the constant current charging data is selected as the input of model. And results show that the proposed deep learning method owns great generalization ability between different aging scenarios.