基于改进CGAN和ARIMA组合模型的风光典型场景生成OA

With the high penetration of renewable energy,the complexity of uncertainties faced by power systems has intensified,making operational planning increasingly reliant on refined scenario generation.To address the issues of insufficient interpretability and difficulty in ensuring physical feasibility in a single deep generative model,a typical wind and PV output scenario generation method based on an improved conditional generative adversarial network(CGAN)and autoregressive integrated moving average(ARIMA)hybrid model is proposed,which integrates the implicit learning capability of deep generative models with the explicit modeling capability of parametric statistical models.First,the ARIMA model is used to capture the low-frequency trend components of the output data,while extracting the high-frequency residual sequences.Then,an improved CGAN model integrated with long short-term memory network is constructed.A statistical regularization term is introduced into the generator loss function to refine the distribution characteristics of residuals,thereby achieving high-frequency residual sequence modeling.Moreover,kernel principal component analysis and a Gaussian mixture model are applied to perform nonlinear dimensionality reduction,adaptive clustering,and probability weighting on the generated high-dimensional scenario matrices,thereby extracting typical scenarios with probability weights.Finally,case studies indicate that the proposed approach reduces the Wasserstein distance index by more than 50%and the conditional value-at-risk error by over 40%,confirming its effectiveness.