103 / 2025-04-17 17:14:07

Voltage Prediction of Vanadium Flow Battery Based on CNN-BiLSTM-Attention

RFBs,Voltage prediction,CNN,LSTM

This paper proposes a vanadium flow battery voltage prediction method based on a CNN-BiLSTM-Attention hybrid model, optimized using MBGD, Adam optimizer, and Log-cosh loss function. The model is trained using a publicly available vanadium flow battery dataset. By capturing the temporal features of the voltage data, the model utilizes historical voltage data to predict voltage trends. Experimental results show that the model achieves an average RMSE of 0.0737 under different charge/discharge conditions in this dataset, demonstrating high prediction accuracy. However, the model’s generalization ability under random charge/discharge conditions still requires further improvement.