Because lithium-ion battery is a rigid system, it presents rich nonlinear dynamic characteristics and high complexity. The parameters of the equivalent circuit model and SOC state show time-varying slowly with the use of the battery, and the SOC estimation method of using the gauge usually has some shortcomings such as low accuracy and poor timeliness. The charge control memristor was connected to a first-order RC model as a load to establish a fourth-order chaotic system. The unknown parameters of fourth-order chaotic system were identified online by using a state observer, and the R0, R1 and C1 values of first order RC model parameters were obtained in real time. The mathematical expression of the accurate first-order RC model was established by using the online parameter identification values. Then AEKF and SVR models were used to estimate SOC time series in real time, and SOC estimated values of the two models were obtained. Using the LSTM model nonlinear combination of AEKF and SVR model estimated values, the final lithium-ion battery SOC estimated values were finally obtained. The experimental results show that the nonlinear combined estimation model can accurately estimate SOC in real time, which indicates that the proposed nonlinear combined estimation model has better nonlinear dynamic estimation ability, higher accuracy and generalization ability.