Apparatus and Method for Neural Network-Based State of Charge Estimation for Lithium-Ion Batteries According to Temperature

Vehicle Battery SOC Estimation Using a Temperature-Specific Deep Neural Network Bank

This technology estimates the State of Charge (SOC) of a battery by storing deep neural network models (MNN, LSTM, GRU) trained under different temperature conditions in a temperature-based model bank and selectively running the model that matches the real-time battery temperature.

The internal resistance of a lithium-ion battery changes non-linearly with temperature. Conventional static table-based estimation methods have limitations, as they suffer from low accuracy during temperature fluctuations and are highly dependent on data reliability.

This technology uses voltage, current, temperature, and time data as inputs to build specialized models for different temperature ranges using deep neural networks such as MNN, LSTM, and GRU. It reduces estimation error (MAE) by selecting the most suitable model for the real-time measured temperature to calculate the SOC. It can be applied to BMS for electric and hybrid vehicles operating in extreme cold or heat, as well as outdoor energy storage systems with significant ambient temperature variations, ensuring stable driving range predictions regardless of season or location.

‍

‍

Key Features:

• A battery data acquisition unit that receives battery data consisting of voltage, current, and temperature data from the battery installed in the vehicle.
• A temperature data acquisition unit that receives temperature data from the battery data and outputs selection information for the temperature-based model.
• A temperature-based model selection unit that selectively transmits battery data to the temperature-based model corresponding to the selection information.
• A temperature-based model bank unit equipped with deep neural network models, each trained using only battery data corresponding to specific temperatures.

‍