Accurate and reliable state of charge (SOC) estimation can provide a guarantee for the safe and efficient use of battery management systems. Given that there is insufficient accuracy of SOC estimation of lithium batteries, this paper proposes the artificial bee colony algorithm (ABC) and the random forest optimized EKF algorithm (RFEKF) respectively to realize parameter identification and SOC estimation of the battery model. Based on the establishment of dual polarization model, to solve the problem of the accumulation of initial errors in online identification, the ABC algorithm is used to search for the global optimal impedance parameter value under the minimum model voltage error, and realize accurate identification of model parameters. Based on obtaining accurate model parameters, this paper uses random forest (RF) to online compensate for the SOC posterior estimation error, and achieves the purpose of making up for the error of the high-order term of the traditional EKF algorithm. Then it achieves high-precision SOC estimation. Combining a hardware-in-the-loop simulation system and battery test platform, it realizes rapid control prototype verification for the SOC estimation algorithm under EPA urban power conditions. The results show that the error indicators of the lithium battery SOC estimation algorithm based on ABC-RFEKF are lower than the traditional SOC estimation algorithm. The average error is around 1%, thereby meeting actual engineering needs. This work is supported by the National Natural Science Foundation of China (No. 51775426).