To enhance the operational security of integrated energy systems （IES） under extreme weather conditions，it is essential to fully consider the impact of extreme scenarios in production scheduling.However，because historical samples of extreme scenarios are scarce，IES face challenges in operational scenario modeling and scheduling，making effective methods for generating extreme scenarios highly significant.To address this issue，a method based on a Wasserstein generative adversarial network with gradient penalty （WGAN-GP ） was proposed for generating source-load extreme scenarios in IES.By introducing the Wasserstein distance and gradient penalty，the proposed method improves model training stability，increases the diversity of generated samples，and expands the original scenarios.Subsequently，a two-stage source-load screening strategy was designed.Thresholds were set based on multiples of the standard deviation to identify samples located at the extremes of the distributions of renewable energy output and multi-energy load data，respectively，thus constructing source-load extreme scenarios.Case study results show that the scenario set generated by the proposed method is highly consistent with historical data in terms of probability distribution，temporal characteristics，and load correlation，while also exhibiting good diversity.Finally，two typical types of extreme scenarios，were identified，validating the effectiveness of the proposed method.