The SOH estimation of lithium battery based on health index and improved LSTM

A method based on health index and IPSO-LSTM was proposed to estimate the health status (SOH) of lithium battery. Firstly, based on the NASA battery dataset, Spearman correlation analysis was used to select health indexes for estimating lithium battery SOH. Secondly, the IPSO algorithm was used to optimize the hyperparameters such as the number of hide-layer neural cell, dropout ratio and batch size in LSTM. Then, pre-selected health indexes and battery SOH values were used as inputs and outputs to construct IPSO-LSTM neural network for predicting lithium battery SOH. Finally, the datasets of batteries from NASA were chosen to confirm the effectiveness of the constructed SOH prediction model of battery. What’s more, the predictive results based on SVR, LSTM and PSO-LSTM were compared and analyzed. The results indicated that compared to SVR, LSTM and PSO-LSTM models, the IPSO-LSTM model had the smallest mean absolute error and root mean square error, within 0.05, and the model had higher prediction accuracy.