Accurate wind power forecasting is very crucial for the security and stability of power system operation. From a statistical standpoint, a dynamic harmonic regression method is proposed for very-short-term wind power forecasting. Cubic polynomials between wind power and wind speed at different heights are used to construct the regression model. Then, autoregressive integrated moving average model is proposed to model the regression residual in order to fully use historical information of wind power time series. Finally, according to the daily seasonal characteristics of wind power, fourier series is introduced and the final model is established. Results from real-world wind farms show that this method can effectively improve the traditional autoregressive integrated moving average model and regression methods, which can reduceroot mean squared error and improve the prediction accuracy. Compared with two commonly-used existing approaches, persistence and autoregressive integrated moving average model, the proposed model is verified to have higher prediction accuracy, indicating that this method has certain practical application value.