Aiming at the optimal scheduling problem of virtual power plant ( VPP ) with multiple uncertainties on the source-load side, this paper proposes a two-stage stochastic robust optimal scheduling method considering the uncertainty of the source-load side. This method combines the characteristics of robust optimization and stochastic optimization to model the source-load uncertainty differentiation. The Wasserstein generative adversarial network with gradient penalty ( WGAN-GP ) is used to generate electric and thermal load scenarios, and then K-medoids clustering is used to obtain several typical scenarios. The min-max-min two-stage stochastic robust optimization model is constructed, and the column constraint generation ( C & CG ) algorithm and dual theory are used to solve the problem, and the scheduling scheme with the lowest operating cost in the worst scenario is obtained.