Abstract: Firstly LM neural network model was used to predict the wind power. In order to determine the optimal weights and thresholds of the neural network and avoid local optimum, ant colony algorithm was used to optimize it. Then, the time points with low prediction accuracy were determined by the predicted values of wind speed, and these time points were further predicted by the curve of wind power characteristics. The results showed that the prediction accuracy based on ant colony optimization neural network model was improved by 16.272 percentage points, the qualified rate was increased by 18.735 percentage points, and the mean square error was reduced by 3.117.