基于CWGAN-GP的新型配网源荷场景生成方法OA

A method for generating source and load scenarios of novel distribution networks based on conditional Wasserstein generative adversarial network with gradient penalty(CWGAN-GP)was proposed to address the issues of training instability,mode collapse,and insufficient controllability in traditional generative adversarial networks(GANs)in source and load scenario generation for novel power system.The theoretical foundation of CWGAN-GP was first discussed,including the advantages of Wasserstein distance and the gradient penalty mechanism,as well as the conditional control capabilities of conditional generative adversarial networks.By optimizing the network structure and training strategy,a CWGAN-GP model for source and load scenario generation was constructed.To comprehensively evaluate the quality of the generated scenarios,a multi-dimensional evaluation system was developed,which includes the cumulative distribution function(CDF),Fréchet inception distance(FID),and mean absolute percentage error(MAPE).The results of the case study demonstrate that the proposed method significantly outperforms traditional methods in terms of FID and MAPE.The generated load and photovoltaic output time-series curves are highly consistent with real data in terms of periodicity and fluctuation characteristics,validating the effectiveness and robustness of the model.