Accurately predicting the state of charge(SOC) of power batteries is the key technologies of battery management system(BMS), which is of great significance to the performance and safety of power batteries. However, the battery will produce noise during the process of data collection, which will affect the SOC estimation accuracy of power battery. Therefore, a SOC estimation method based on the particle swarm optimization and deep belief network and adaptive extended Kalman filter/adaptive H-infinite filtering fusion (PSO-DBN-AEKF/AHIFF) was proposed. Firstly, the particle swarm optimization (PSO) was used to improve the deep belief network (DBN) to obtain the particle swarm optimization and deep belief network (PSO-DBN) for the SOC estimation, so as to avoid the local optimal problem and increase the speed of computation. Next, with a focus on the issue of battery data noise, the adaptive extended Kalman filter (AEKF) was used to reduce the impact of white noise and the adaptive H-infinite filtering fusion (AHIFF) was used to reduce the impact of colored noise, and the SOC fusion estimator was used to achieve the optimal estimation strategy under different noise conditions. In comparison to the conventional algorithm, the experimental results show that the PSO-DBN-AEKF/AHIFF algorithm has higher prediction accuracy and robustness of SOC estimation results under the interference of Gaussian white noise and colored noise.