In view of the inaccurate estimation in the SOC(state of charge) for SOAH LiFeP04 cell of electricvehicle，an improved PSO(particle swarm optimization ) algorithm based on the original neural network was intro-duced to optimize the weights and thresholds of BP neural network. Furthermore，the optimized network was appliedto SOC prediction，so as to reduce the error of SOC estimation. The object of this research was SOAH lithium bat-tery. Firstly，the mutation operator was introduced into the PSO algorithm to improve the shortcomings of PSO，such as low search precision and poor later iteration efficiency. Then，through the experiment the relationship between temperature，voltage，electric current of three main parameters and SOC was analyzed，respectively，and thedischarge experiment method was employed to draw the relation curve between the SOC and the open circuit volt-age. In the end，based on a large number experimental data，the BP neural network and improved PSO一BP neuralnetwork was used respectively for training and SOC prediction. Both theoretical analysis and experiment resultsshow that this method can effectively reduce the error of SOC within 2%，improve the precision of prediction andmeet the requirements of SOC estimation.