Edge Computing Enabled Battery State of Charge(SoC) Estimation Using TinyML Techniques

Redox flow batteries (RFBs) are electrolyte-based batteries that have become a research hotspot due to their modular nature and decoupled power-energy characteristics. Household and microgrid environments extensively utilize RFBs as storage systems. The nonlinear chemical nature of RFBs hinders the modeling process, estimation of critical parameters like State of Charge (SoC), and consequently, overall energy management. The current methods in the literature use computationally complex multi-physical models that are unsuitable for deployment at the edge. Therefore, this study focuses on developing a resource-optimized machine learning-based SoC estimation model fit for deployment on an ESP32-based edge device. The study also outlines and enumerates the experimental setup for two types of RFBs: vanadium-based (AVRFB) and zinc bromide-based (ZBRFB), its data preprocessing steps, and the edge deployment process. Additionally, model quantization has also been targeted to reduce computational requirements. Using a dynamically quantized model, the devised framework achieved a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> score of 0.975 for AVRFB, performing inference in 1.72 seconds. In the case of ZBRFB, the SOC estimation model achieved a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> score of 0.818 with the float32 quantized model, outputting inference in 3.18 seconds.