Owing to the shortcomings of current state of charge (SOC) estimation algorithms, such as poor stability and large error, a new algorithm based on the integration of adaptive extended Kalman filter (AEKF) and short term memory (LSTM) based on real vehicle cloud discharge data was proposed to predict SOC of small-power electric vehicles. Adaptive forgetting factor least square method (AFFRLS) was used to identify the second order RC equivalent circuit model parameters of the battery. Secondly, the real-time discharge data collected by the cloud is taken as the research target, and the AEKF-LSTM fusion algorithm is used to predict the battery SOC of small-power electric vehicles. The AEKF-LSTM fusion algorithm takes the terminal voltage, current, temperature at the current moment and the SOC of the battery at the previous moment as inputs, and uses the updated SOC as the output to train the estimation model. Finally, compare the battery SOC prediction results of the AEKF-LSTM fusion algorithm and the single AEKF algorithm with the actual SOC values. The experimental results show that the root mean square error (RMSE) of the AEKF-LSTM fusion algorithm is 0.005 8 V, and the mean absolute error (MAE) is 0.004 1 V. Compared with the AEKF algorithm, its RMSE is reduced by 0.008 7 V and its MAE is reduced by 0.116 4 V, and both RMSE and MAE of AEKF-LSTM fusion algorithm are less than 0.6%. It is proved that the fusion algorithm has high estimation accuracy and strong robustness.