Accurately predicting the State of Charge (SOC) of lithium batteries is one of the key technologies to ensure the safe and reliable operation of electric vehicles. Based on the standard Sparrow Search Algorithm (SSA), an improved SSA based on Circle mapping and genetic algorithm mutation strategy (CSSA), which is combined with the Back Propagation (BP) neural network, has been proposed. The combined algorithm is named as CSSA-BP, which utilizes the CSSA to optimize the initial weights and thresholds of the BP neural network Using the Panasonic 18650PF test dataset, a comparative analysis of the CSSA-BP, BP neural network and SSA optimized BP neural network (SSA-BP) were conducted. After experiments, the maximum absolute error of CSSA-BP is −0.70, and the root mean square error (RMSE) is 1.54% and the mean absolute error (MAE) is 1.13%, both of which are lower than those of BP and SSA-BP. This indicates that CSSA-BP has good accuracy and robustness.