Abstract: To address the issue of significant estimation errors in the SOC state estimation of lithium-ion power batteries, a SOC estimation method considering temperature variations is proposed. A second-order RC equivalent circuit model is established as the foundational framework, with the dynamic characteristics of the battery described using Kirchhoff's voltage law. Recursive least squares (RLS) with a forgetting factor is employed to identify the model parameters. Subsequently, a binary mapping relationship between parameters and SOC/temperature is constructed through polynomial fitting, enabling temperature-adaptive adjustment of the model parameters. Additionally, a transfer learning mechanism is introduced to dynamically correct the model parameters using online temperature signals. This approach aims to mitigate uncertainties caused by temperature variations and enhance the robustness of SOC estimation. Experimental results demonstrate that within the temperature range of 0 ℃ to 40 ℃, the root mean square error (RMSE) between the SOC estimates and actual values does not exceed 0.001.