This work focuses on the probability distribution function of wind power variation. After analyzing the characters of the power fluctuation data, normal distribution function, t location-scale distribution function and mixed second-order one-dimensional Gaussian distribution function are chosen to describe the wind power variation. Then K-S test(Kolmogorov-Smirnov) test and Pearson product-moment correlation coefficient are used to evaluate the fitting effect of the three distribution functions respectively, which indicates that the mixed second-order one-dimensional Gaussian distribution is the most appropriate one. At last, the factors affecting the parameters of Gaussian mixture distribution and to what degree they can achieve are investigated.