Abstract: Based on the predicted power results, constructing the wind power output scenarios set is an important factors for the stochastic optimal dispatch of the power system. The existing generation methods of multi-regional wind power output scenarios are mainly sampled randomly by the spatial-temporal correlation coefficients constraints. Due to the nonlinearity of the spatial-temporal correlationship characteristics of the wind cluster, the generated scenarios are quite different from the actual outputs of the clusters. This paper proposes a scenarios set constructing method of multi-regional wind power based on conditional generative adversarial network (CGAN). The method combines the spatial and temporal correlation characteristics of multi-regional outputs of wind power, and uses a three-dimensional convolutional network to design a network structure suitable for the generation of multi-regional output scenarios of wind power. The mapping relationship between the input data and the expected output scenarios is setup. The proposed method is verified by five wind power regions in northwest of China. The proposed method is also compared with the traditional Latin hypercube sampling (LHS) method constrained by the correlationship coefficients. The results show that the characteristcs of the multi-regional output scenatios constructed by the proposed method are more in line with that of the real wind power.