Lightweight CNN-based state of charge estimation for lithium-ion batteries with embedded deployment on STM32

Abstract

State-of-Charge (SOC) Estimation for lithium-ion batteries is critical for reliable Battery Management Systems. This paper proposes a lightweight one-dimensional convolutional neural network (1D-CNN) for real-time SOC estimation, which establishes the nonlinear correlation between measurable battery parameters and SOC. The proposed model is trained, validated, and tested on the NASA lithium-ion battery aging dataset. The trained model is exported to ONNX format using MATLAB Deep Learning Toolbox and then converted into optimized C code for STM32H743 by STM32Cube.AI. The proposed model demonstrates strong performance on the test set, with RMSE, MAE, and R² values of 0.0610, 0.0453, and 0.9628, respectively. Embedded deployments show that the model contains 3809 parameters, occupies about 31.07 KiB Flash and 3.35 KiB RAM. The results verify the feasibility of the proposed model for embedded SOC prediction.