Prediction of vanadium redox flow battery storage system power loss under different operating conditions: Machine learning based approach
Nawin Ra, Ankur Bhattacharjee
Abstract
Prediction of battery storage system loss is necessary to further improve the performance reliability and efficiency of the battery storage system. The prediction of the overall system power loss of Vanadium Redox Flow Battery (VRFB) using different machine learning (ML) algorithms has been demonstrated for the first time. Under different operating current levels and electrolyte flow rates, the internal resistance variation and pump power consumption of the practical 1 kW 6 kWh VRFB system dataset have been considered for prediction. The prediction accuracy of ML algorithms has been analysed in detail based on the regression metrics such as correlation coefficient (R2), mean absolute error (MAE) and root mean square error (RMSE). It is observed that Ensemble learning (EL) based Adaptive Boost (AdaBoost) algorithm is superior in predicting VRFB system loss compared to that of linear regression (LR), support vector regression (SVR) algorithms. The AdaBoost algorithm statistically presents the best prediction results with prediction error parameters obtained as R2, MAE and RMSE of around 0.99, 3.242 and 4.073, respectively, for 50 A stack current profile with a flow rate of 18 L/min. The prediction results obtained in this paper claim to be beneficial for designing optimized interfacing of VRFB storage with renewable energy sources and other power system applications.