To deal with the changes of internal characteristics in lithium-ion battery at different temperatures and different charge and discharge rates, a design method of adaptive unscented Kalman filter in this paper is proposed to improve the adaptive ability of the state of charge (SOC) in the complex environment. To solve the problem on the estimations of unknown parameters and order in the lithium-ion battery model effectively, the fractional-order and the SOC are mapped to the effective value ranges via the Sigmoid function to prevent the failure of SOC estimation. Using the augmented vector method, the augmented state equations in terms of system states including SOC, order and parameters are established. Based on the method of unscented transformation, the nonlinear function of the lithium-ion battery described by a fractional-order model is processed, and the SOC estimation with a high estimation accuracy is obtained. In addition, the adaptive estimation algorithms for noise covariance matrices are discussed to further increase the adaptive ability of the proposed algorithms in complex environment. Finally, the effectiveness of the proposed algorithms is verified by several groups of experimental results under different working conditions.