In view of the nonlinear and non-stationary characteristics of runoff prediction data series in the hydrological prediction, a new intelligent optimization algorithm, artificial electric field algorithm (AEFA), is combined with LSTM neural network to optimize the network parameters, then the AEFA-LSTM prediction model is established. The measured annual runoff of Xuanwu Hydrological Station of Wohe River in Zhaokou Large Irrigation Area was then used as the sample data for the network optimization training and prediction analysis, and the results were compared with those of the GA-LSTM model and PSO-LSTM model which were established using conventional optimization algorithms (GA and PSO). The comparison analysis shows that compared with GA-LSTM model and PSO-LSTM model, the average relative error of the predicted value calculated by the AEFA-LSTM model is reduced by 7.59% and 5.22% respectively and its average absolute error, mean square error and root mean square error are the smallest of the three models, indicating that the AEFA-LSTM model can predict the runoff more accurately, and provide a new high-precision runoff prediction method for the hydrological prediction.