Abstract: To address the issue of limited accuracy of state of health (SOH) of electrochemical energy storage batteries evaluation, firstly, through the analysis of data correlation, the voltage variation values in 10 minutes after the voltage reaches 3.25 V, 20 minutes before reaching 3.35 V and 30 minutes before reaching 3.4 V during the charging process are selected as health factors. These factors are utilized as input parameters of the model in a long short-term memory neural network optimized through genetic algorithms to facilitate SOH assessment for energy storage batteries. To validate the effectiveness of this evaluation method, five soft pack LiFePO4 batteries are subjected to cyclic aging experiments, obtaining health factors alongside actual SOH values for model development. The evaluation results are then compared with the actual SOH values obtained from testing. The evaluation results are quantified using mean square error (MSE) and mean absolute percentage error (MAPE), which shows that the SOH evaluation of the model is accurate,and MSE and MAPE are reduced by 48.3% and 74.1% respectively, compared with the unoptimized long short-term memory neural network.

Key words: energy storage batteries, state of health, fragment data, long short-term memory