Lithium-Ion SOC Diagnosis Using Lebesgue Sampling and ECM Updates

The patent segments the SOC estimation process into two distinct phases: an offline training phase where the electrochemical model is used to generate accurate training data, and an online estimation phase where a simplified neural network model performs real-time SOC estimation. This segmentation allows the computationally intensive electrochemical model to be used only when necessary for training, while the lightweight neural network handles continuous online estimation with minimal computation cost.

The patent applies preliminary action by performing extensive OCV measurements and data collection during the offline training phase before the battery is actually deployed for use. During this preliminary stage, the battery is discharged at various rates and allowed to rest to establish accurate OCV-SOC relationships. This pre-acquired knowledge is then stored in the neural network model, eliminating the need for time-consuming OCV measurements during online operation.

The patent transforms the static fixed-parameter ECM into a dynamic adaptive model by integrating a neural network that continuously updates the ECM parameters based on real-time terminal voltage measurements. The neural network learns the nonlinear relationships between SOC, SOH, and ECM parameters during offline training, enabling the parameters to adapt dynamically during online operation. This dynamic approach maintains high simulation accuracy while keeping the computational burden manageable through efficient neural network inference.