State of health estimation for all-vanadium redox flow batteries based on interaction measurement analysis and genetic algorithm

Abstract

Accurate state of health (SOH) estimation of all-vanadium redox flow batteries (VRFBs) is essential for ensuring safe and reliable operation. In this paper, a series of VRFB charge/discharge cycle experiments is conducted to extract the best health feature subset by the interaction measurement analysis, based on which an SOH estimation method combining the extreme learning machine and a genetic algorithm (GA) is proposed. Compared with the sand cat swarm optimization algorithm, GAs are more suitable for SOH prediction, as they can not only estimate accurately but also predict recession characteristics. The best activation function is linear, in which the maximum error is 0.53%. The results demonstrate an average mean absolute error of 0.003, an average mean squared error of 0.000018, an average mean absolute percentage error of 0.1%, and a coefficient of determination of 0.99998. The influences of noisy conditions on predicting performance is also taken into consideration, and the most suitable models for varying levels of data acquisition precision are determined. The results confirm that the proposed method possesses accurate predicting ability and robustness, which provides an effective technical solution for SOH estimation. This work can contribute to operation in a healthy state in industrial applications of VRFBs.