A robust machine learning-based SOC estimation approach for vanadium redox flow battery

The vanadium redox flow battery (VRB) is recognized as an effective large-scale energy storage solution for mitigating the renewable intermittency and ensuring grid reliability. Accurate estimation of the state of charge (SOC) is crucial for the optimal operation of VRB. This paper presents a novel machine learning-based estimation algorithm to overcome the long-lasting problem of model dependency in the existing SOC estimation approaches for VRB. Compared to the conventional model based methods, such as Kalman filter and sliding mode observer, the proposed algorithm does not need any knowledge of the VRB model. In addition, the proposed algorithm employs recurrent equilibrium network (REN), which has “built in” behavioral guarantees of stability and robustness compared to the traditional machine learning algorithms. Furthermore, the proposed algorithm employs the nonlinear direct parameterization technique to substantially simplify the neural network training. Its efficacy is verified by experimental results. • Present a learning-based method for battery charge level estimation. • Achieves high accuracy in both charging and discharging conditions. • Delivers stable results under complex and noisy measurement conditions. • Validated using data from a real industrial battery system.