With the continue increasing of the penetration rate of renewable energy

how to effectively describe the uncertainty of output is a major challenge of day-ahead scheduling of nowadays power system. This paper proposed a conditional generative adversarial networks based scenario generation algorithm

in which Wasserstein distance was used as the loss function of discriminator

and a specific network structure was designed which suits for generation of day-ahead scenarios. Through the game training of GAN

the mapping between the noise distribution and day-ahead scenario set under the prediction condition can be learned. Actual wind farm data（including forecasting and actual generation data） was used to validate the proposed model

simulation results shows the proposed model has better performance for describing uncertain nature of wind power compared with traditional Markov chain scenario generation method.