SOH probability prediction for lithium-ion batteries based on improved DeepAR

As a key and core component of energy storage system, the safety and reliability of lithium-ion battery directly affect the comprehensive performance of energy storage system to a large extent. Aiming at the safety problem of energy storage batteries, a method for predicting the health state (SOH) of lithium batteries based on improved autoregressive recurrent neural network (DeepAR) was proposed. Firstly, the characteristic information such as capacity, internal resistance and time interval are extracted from the operating data of lithium batteries, and the features with strong correlation with battery health status are found through correlation analysis method as the input of health factors. Then, residual connections are introduced into the DeepAR model to reduce gradient disappearance during training, and the adaptive Dropout regularization method is used to improve the generalization ability of the model. Finally, the data set of lithium battery from University of Maryland is used to verify the validity of the model. The experimental results show that the maximum root mean square error (RMSE) is less than 0.01, and the mean absolute error (MAE) is less than 0.011. This method has high accuracy for SOH prediction of lithium-ion battery, and has certain application value for safety research of energy storage battery.