With the increasing proportion of wind power access in southern China, the impact of the randomness and volatility of wind power generation on power system operation cannot be ignored. Therefore, it is necessary to study the characteristics and random distribution of wind power generation in southern China. Firstly, the influence of time and geographical factors on wind power generation characteristics is analyzed. Secondly, an improved non-parametric kernel density estimation method considering the optimal bandwidth selection of kernel function is proposed to establish the random distribution models of wind power generation in five southern provinces under multiple time scales. The evaluation index system of model fitting effect is also established. Thirdly, the effectiveness of the improved non-parametric kernel density estimation method proposed is verified by comparing with two parameter estimation methods based on typical empirical distribution function. Finally, the differences of probability characteristics of wind power generation in different provinces of southern China are analyzed by daily, quarterly and annual time scales, which can play an auxiliary role in decision-making of power system dispatching, power planning, site selection and capacity determination of wind farms.