Scenario generation technology

by simulating the randomness and diversity of various typical scenarios

provides critical data support for addressing the limitations of historical data samples and the difficulty of covering extreme scenarios. This paper proposes a feature-constrained and target-driven WGAN-GP (Wasserstein GAN with gradient penalty) based photovoltaic multi-scenario generation method. By introducing the Wasserstein distance and gradient penalty

the method enhances the stability of the generator and the diversity of generated samples. Based on this

key features of three typical scenarios-sunny

rainy

and cloudy-are extracted to define generation targets. A dynamic gate function is embedded in the generator's output stage to dynamically determine the boundary between daytime and nighttime

ensuring zero output during nighttime and realistic irradiance variations during daytime. Additionally

a weighted sampling optimization strategy is employed to dynamically adjust sample weights and selection probabilities

further strengthening the learning of critical feature samples. This enables the generator to more accurately capture target features and significantly improve the generation of scarce scenarios. Case study results demonstrate that the proposed method effectively captures the key features of various weather scenarios

producing samples with excellent target feature alignment and physical consistency

offering a reliable solution for photovoltaic scenario generation.