Method and System for Online State of Charge and State of Health Estimation of Lithium Batteries Based on a Neural Network Model Bank

Degradation-Adaptive SOC Estimation Using Neural Network Model Switching Based on SOH Estimation Results

This technology uses real-time battery operating data—such as voltage, current, temperature, and operating time—to first estimate the current State of Health (SOH), then selects the neural network model optimized for that SOH from a model bank to precisely estimate the State of Charge (SOC).

Conventional SOC/SOH estimation methods, such as Coulomb counting or simple Extended Kalman Filters (EKF), have struggled to flexibly account for characteristic changes caused by battery degradation. Furthermore, these methods often suffer from a sharp decline in accuracy under specific operating conditions.

This technology constructs a neural network model bank consisting of multiple pre-trained models (MNN or LSTM) categorized by health status—such as normal, caution, and fault—and dynamically selects and switches to the most suitable model based on the SOH estimation result to calculate the SOC. It can be applied to EV BMS, ESS using repurposed aging batteries, and online diagnostics for drone and robot battery packs, helping to reduce the widening gap in remaining capacity displays as batteries age and enabling more accurate determination of replacement timing.

‍

‍

Key Features:

• A data management unit that receives the initial SOC, and subsequently the previously estimated SOC, to output them along with operating time, voltage, current, and temperature.
• An SOH estimation unit that receives output data from the data management unit to estimate the SOH of lithium batteries using a neural network model.
• An SOC estimation unit that selects a model corresponding to the estimated SOH from a bank of neural network models trained for different health states to estimate the SOC.
• An averaging unit that calculates the mean value of the SOCs estimated by multiple neural network models and provides it as the initial SOC to the data management unit.

‍