Blending deep-learning and observers for improved state of charge estimation in vanadium flow batteries

This paper presents the design and implementation of a deep-learning-based observer for accurately estimating the State of Charge (SoC) of a vanadium flow battery. The novelty of the proposal lies in its direct use of terminal voltage and the application of a machine learning algorithm to model the battery's overpotentials, leading to greater accuracy and reduced complexity compared to classical models. The overpotentials model consists of a neural network trained using data generated by a classical observer that estimates species concentration using a physical electrochemical model and the open-circuit voltage measurement. The trained model is then integrated with the observer to improve SoC estimation accuracy. The proposed method is validated through comprehensive numerical simulations and experimental studies using a real vanadium flow battery setup, demonstrating its effectiveness in providing reliable SoC estimation with a relative error of less than 5%.