Geographic diversification of wind farms can smooth out the variability of wind power. The paper applies copula function and mean-variance model to study the wind speed spatial correlation and optimal wind power allocation. The maximum likelihood method is utilized to choose appropriate copula function to describe the correlation of wind speeds, and the pairwise rank correlation coefficients of wind speeds are calculated by copula, while the relationship between rank correlation and wind farms distance is fitted by least squares method. A new mean-variance model is constructed to optimize wind power allocation, where return is defined as the capacity factor, and risk is defined as the standard deviation of hourly wind power variation, and linear correlation is replaced by rank correlation. Taking Holland 40 wind farms as an example, the results show that Gumbel copula and t copula present a better fit for wind speeds correlation, and the rank correlation tends to decrease by 0.1 with a increasing distance of 100 km. By solving the mean-variance model, the optimal combination of wind power is obtained, and wind power variability drops by a maximum of 70% comparing with the single wind farm, where off-shore wind farms play a more important role. Under the guidance of this model, an optimal wind power allocation strategy can be used to reduce the system risk and cost due to wind power variability.