基于改进McDalNet的电力系统暂态电压稳定自适应评估方法OA

To address the degradation in model generalization caused by frequent switching of power system operat-ing scenarios in transient voltage stability assessment(TVSA)models,an adaptive assessment method based on a modified multi-class domain adversarial learning networks(McDalNet)is proposed.First,the modified McDalNet uses the Wasserstein distance to construct the loss function to more effectively capture domain distribution discrep-ancies before and after scenario switching,while a center loss is introduced to enhance intra-class feature cluster-ing,thereby improving the separability of samples from different classes.Subsequently,the feature extractor and la-bel classifier are trained using features from three sampling moments:steady-state,fault occurrence,and fault clearance,to build a high-precision assessment model for the original scenario.Finally,domain alignment is achieved through an auxiliary classifier and a small number of target-domain samples,enabling adaptive model up-dating so that it can be applied to TVSA in new scenarios.Case studies demonstrate that the proposed method can align the data distributions of the source and target domains,effectively enhancing the generalization performance and continual learning capability of TVSA models under multiple operating scenario transitions in power systems.