Signal noise interference, adaptability of battery model to temperature and aging, and inconsistency of the battery pack are vital factors which have the influence on the accuracy of State of Charge（SOC） estimation. To estimate the SOC of battery pack accurately, this paper proposes a novel method that combines the Interacting Multiple Model (IMM) and the Adaptive Battery State Estimator（ABSE）. Firstly, the battery interaction models are established based on the comprehensive characteristics of the battery pack. The SOC is estimated by ABSE and embedded in the IMM model. Then, the information distribution factors of each model are calculated, and the SOC of each model is probabilistically fused according to the information distribution factors to obtain a battery pack SOC with higher precision. Finally, the robustness and universality of the algorithm are evaluated under combined conditions of different temperature. The experimental results show that this method is effective for various conditions including the input signals with noise，complicated condition under the whole climate, and inconsistency between batteries.The estimation error can be controlled within the range of 2% during effective charging and discharging cycles.