Accurate estimation of lithium ion battery state of charge(SOC)is the key to ensure safe and stable operation of battery management system. The commonly used ampere-hour integral method has cumulative errors. And the Kalman filter algorithm needs to establish an accurate battery model. The neural network method does not rely on a specific lithium battery model and can reflect the nonlinear relationship of lithium batteries, and thus has received extensive attention. However, traditional neural network has long training time and low precision when estimating SOC. For the low accuracy of SOC estimation in the past, the particle swarm optimization(PSO) of extreme learning machine(ELM) neural network method is proposed. In the PSO-ELM model, voltage, current and temperature are used as input vector and the value of SOC is used as output vector. The experimental data is imported into the model for training and testing, and the input weight matrix and hidden layer threshold of ELM are optimized by PSO. In addition, the simulation results show that compared with the prediction results of BP neural network, the method of estimating SOC in this paper has higher precision.