Wind power prediction is significantly important to the safe and economic operation of power system with large-scale wind power. The effect of the characteristic of wind speed and wind power, the algorithm and input variables of prediction model on power prediction error was analyzed. Taking the measured data in a wind farm as an example, the error evaluation indices were used to evaluate the predictive results, and the method of model modification to gradually reduce the prediction error of wind power has been proposed. Then, the flow chart of prediction error modification was presented. The method can improve the prediction accuracy of wind power, and make the wind power prediction system better for the practical production.