To improve the State of Charge （SOC） estimation accuracy， an improved Cubature Kalman Filter （CKF） algorithm is proposed， in which an error compensation module based on Kernel Extreme Learning Machine （KELM） algorithm is integrated into CKF properly. The Recursive Least-Square method （RLS） is used to online estimate the parameters of the equivalent circuit model of the battery， and the preliminary SOC value is estimated using the CKF algorithm accordingly. Then， taking the operation voltage and current of the battery， and residuals mean and variance of the preliminary SOC estimation values as the input， and the corresponding SOC estimation error as the output， the SOC estimation error prediction model based on the KELM algorithm is obtained， and trained by using the Federal Urban Driving Schedule （FUDS） condition data. The regression prediction ability of the KELM model is utilized to perform error compensation on preliminary estimation， so as to reduce the error of SOC estimation. To verify the effectiveness and advancement of the proposed method， the experimental data of the Arbin battery test platform are used for simulation analysis. The results show that the proposed method has higher estimation accuracy， stronger generalization， and robustness under various operating conditions.