Abstract
Due to the flourishing development in the field of energy storage power station, there has been considerable attention directed towards the prediction of battery system states and faults. Voltage, as a primary indicative parameter for various battery faults, holds paramount importance in accurately forecasting voltage abnormity to ensure the safe operation of battery systems. In this study, a prediction method based on the Informer is employed. The Bayesian optimization algorithm is utilized to fine-tune the hyperparameters of the neural network model, thereby enhancing the accuracy of voltage abnormity prediction in energy storage batteries. With a sampling time interval of 1 minute and a one-step prediction, where the training set constitutes 70% of the total data, this approach reduces the root mean square error, mean square error, and mean absolute error of the prediction results to 9.18 mV, 0.0831mV, and 6.708 mV, respectively. The impact of actual grid operation data on the prediction results at different sampling intervals and data training set ratios is also analysed, resulting in a dataset that balances efficiency and accuracy. The proposed Bayesian optimisation-based method can achieve more accurate voltage anomaly prediction.