Renewable energy generation is characterized by high uncertainty, and its large-scale integration into power systems poses new challenges to operational reliability. To address this, a capacity allocation strategy for wind–solar–storage microgrids is proposed, which accounts for system reliability and leverages the efficiency of a Swift Wasserstein Generative Adversarial Network (Swift-WGAN). The Swift-WGAN is employed to generate typical wind and solar power output scenarios, followed by scenario reduction using the K-medoids clustering algorithm. A reliability cost model is then developed using the expected energy not supplied (EENS) as a performance metric. Based on this model, an energy storage capacity configuration scheme is formulated with the goal of minimizing total system cost. Finally, a simulation is carried out using the data of a region in Inner Mongolia as an example. The results show that the proposed method has obvious economic benefits and offers practical guidance for capacity configuration in real-world microgrids.