Design of a real-time prediction device for lithium battery SOC based on BP neural network and STM32 chip

Abstract

To overcome the limitations of traditional State of Charge (SOC) prediction methods such as complex models and poor environmental adaptability, this paper proposes a real-time lithium battery SOC prediction device that integrates a Back- Propagation (BP) neural network with a high-performance STM32 micro-controller. In terms of hardware design, a system based on the STM32F407 chip, combined with AD620 instrumentation amplifiers and an AD7606 synchronous data acquisition module, is developed to achieve high-precision synchronous acquisition of the lithium battery’s voltage and current. For the algorithm design, a three-layer fully connected feed-forward BP neural network architecture is constructed. The training performance of the gradient descent method, the Levenberg-Marquardt algorithm, and the Bayesian regularization back-propagation algorithm is compared. The Bayesian regularization algorithm is then optimized to effectively suppress over-fitting and improve the model’s generalization level. Finally, the trained network weights and threshold parameters are embedded into the STM32 chip to complete the integration of the algorithm with the hardware, enabling real-time SOC prediction of the battery. Experimental results show that under a 400 mA constantcurrent discharge condition, the real-time SOC prediction error of the proposed device is less than 5%, demonstrating good prediction accuracy and real-time response capability. The device offers advantages such as a compact structure, controllable cost, and flexible deployment. It may provide a reliable technical solution of battery management systems for electric vehicles, portable electronic devices, energy storage systems, and other related fields.