Battery state of health estimation under dynamic operations with physics-driven deep learning

Accurate assessment of battery aging is crucial for the effectiveness of electrochemical energy storage systems. This study focuses estimation of the state of health for lithium-ion batteries under multi-dynamic operations, leveraging data from different states of charge interval. The approach begins with online identification of the battery model to swiftly gather aging-related physical information. This physical information, when combined with feature engineering techniques, is transformed into fused features capable of generalizing multi-dynamic operations. This is adopted as an input to a recurrent neural network based on an encoder-decoder framework, which establishes a mapping relationship between the fused features and the state of health (SOH). Notably, the framework showcases universality and flexibility, with an encoder that integrates with most recurrent neural networks, bypassing the need for intricate structures to deliver estimation with excellent accuracy. Specifically, with five common RNNs, the battery with a rated capacity of 5.0 Ah has a root mean square error of less than 0.68%. Moreover, by simply fine-tuning the weights, the decoder facilitates SOH estimation across different battery types. This methodology underscores the efficacy of merging physical information with a universal deep learning framework, enabling precise SOH estimations under multi-dynamic operations.