Abstract: It is difficult for traditional modeling methods of wind power to get higher accuracy due to the large range fluctuation of wind power under the condition of same wind speed. A modeling strategy of wind power curve based on optimal smoothing order is proposed. Firstly, smoothing preprocessing method based on time series is utilized to achieve new input wind speed, optimal smoothing order with maximum correlation coefficient is chosen as objective function, and BP neural network is applied to fit wind power curve. Based on the measured data of the wind turbine in a wind farm located in southwest China, experimental results show that the accuracy of wind power curve model built by taking wind speed as input obtained by the optimal smoothing preprocessing is obviously better than that of the existing methods which utilize original wind speed data as input.