A novel intelligent health prediction method for lithium‐ion batteries within a variable voltage range

Summary As the Internet of Vehicles and cloud computing have rapidly developed, they have become increasingly relevant to the online prediction of the state of health (SOH) of lithium‐ion batteries (LIBs). To accurately and robustly predict the SOH of LIBs within a variable voltage range, a novel intelligent SOH prediction model for LIBs is proposed by combining a one‐dimensional convolutional (Conv1D) layer, gated recurrent unit (GRU) network and an attention‐based encoder‐decoder framework. First, based on a battery aging data set, multiple health features are extracted, and then the Conv1D layer is utilized to learn the local trends of these features. Then, the attention‐based framework is used to extract the most relevant cycle information for prediction from the encoder output of the GRU network. The experimental results show that the proposed model achieves prediction results with a root mean square error within 0.9%, which consistently outperforms existing models. An acceptable prediction can be performed within an appropriate range of available voltage data, which is verified and analyzed based on the incremental capacity curves. Furthermore, the robustness of the model is experimentally demonstrated. Even with 150 mV of voltage noise input in an incomplete process, the proposed model yields superior and robust results. Intelligent cloud battery management system is constructed to realize battery state online prediction and this paper focuses on the real‐time SOH prediction part based on the self‐built dataset. Using available data of variable voltage range to predict is closer to practical application based on a one‐dimensional convolution. Employing the attention‐based encoder‐decoder model, SOH can be accurately predicted by extracting the most relevant cycle information.