Accurate state of charge (SOC) estimation of a lithium-ion battery is of great significance for prolonging battery life, improving battery utilization, and ensuring battery safety. An SOC estimation algorithm based on a temperature-dependent dual-polarization equivalent circuit model was established after the basic performance test and dynamic condition test of the lithium-ion battery were performed at different ambient temperatures. The traditional extended Kalman filtering algorithm was replaced by the H-infinity filtering algorithm, and accurate SOC estimation was realized without assuming that the process noise and measurement noise obeyed Gaussian distribution. The proposed model was verified considering the temperature change and the battery model error. The results show that the maximum error of SOC estimation under different temperature conditions can be kept within ±0.03, which proves that the proposed SOC estimation algorithm has higher temperature adaptability and robustness.