The lithium-ion battery is an indispensable energy storage component in a power system. To predict the State of Charge (SOC) of a lithium-ion battery under high pulse rate conditions, an improved Long Short-Term Memory (LSTM) neural network is used to build the SOC prediction model of a ternary lithium-ion battery. Two gating units are added to the original LSTM to improve the dynamic approximation ability of the model by enhancing the interaction between input and output. Compared with Back Propagation (BP) and LSTM neural networks, the prediction performance of the algorithm under high pulse rate conditions is proved superior. The results show that the improved method can accurately characterize the operating characteristics of ternary lithium batteries, and meet the actual needs of SOC estimation. This work is supported by the Natural Science Foundation of Shandong Province (No. ZR201910230295) and the Key Project of Natural Science Foundation of Shandong Province (No. ZR202010290147).