AI-assisted digital twin modeling technology for lithium-ion batteries

Accurate assessment of state of health (SOH) for lithium-ion batteries throughout their entire lifecycle and real-time voltage tracking prediction have become urgent requirement to ensure battery system safety and reliability. This paper develops a cloud-edge collaborative digital twin architecture to estimate SOH for LiCoO₂ cells and predict the real-time voltage. First, a long short-term memory (LSTM) network model is adopted in the cloud side, where health features are extracted from the current segments of the constant voltage charging phase to estimate the cell's SOH. Subsequently, based on the estimated SOH, the cloud side matches electrochemical parameters and transmits them to the edge-side battery management system (BMS), triggering the rapid simulation by an improved single-particle (SP⁺) model to obtain preliminary voltage predictions. Considering the impact of cell-to-cell variations and temperature fluctuations on voltage simulation accuracy, an online correction model based on a fully connected neural network (F-CNN) is further constructed to refine the simulation output, ultimately yielding high-accuracy voltage predictions. Experimental results demonstrate that the developed digital twin model enables precise SOH estimation throughout the entire cell lifecycle and achieves high-precision real-time voltage prediction, providing robust support for battery health management and safety applications.