Transient Stability Assessment Of Power System Based On GAN And Multi-channel CNN

With the widespread application of wide area measurement systems(WAMS),data driven machine learning methods are gradually being applied to power system transient stability assessment.However,this method has certain difficulties in identifying system states that lie between stable and unstable boundaries,and it is also difficult to take into account the accuracy and rapidity of online assessment.To address this issue,this paper proposes a transient stability assessment method based on generating adversarial networks(GAN)and multi-channel convolutional neural network(multi-channel CNN),which improves the accuracy of transient stability assessment for systems under transient stability boundary conditions.The effectiveness of the proposed method has been verified on IEEE39 nodes and a provincial power grid.Firstly,a temporary stable sample set construction method and a boundary sample enhancement method based on GAN are proposed,and the t-SNE algorithm is used to visualize the distribution of boundary samples.The fault trajectory of generator electrical quantities is extracted and the input features of the model are constructed;Building a generated adversary network GAN enhances boundary samples,solves the problem of fewer samples at stable and unstable boundaries,and improves the number and diversity of boundary samples;The distribution of generated boundary samples is verified using the t-SNE algorithm.Secondly,a transient stability evaluation model based on multi-channel CNN is established to evaluate the transient stability state of the system and predict the transient stability margin of the system.Based on traditional CNN with strong feature extraction capabilities,a multi-channel CNN is constructed,and a weight sharing mechanism between channels is designed to improve the feature extraction capabilities of multi-channel CNN;The transient stability state prediction model and margin prediction model based on multi-channel CNN are obtained through training using transient stability sample sets and boundary sample sets.The prediction process and results of the multi-channel CNN model are explained based on the LIME algorithm and the t-SNE algorithm.Then,a transient stability assessment method based on multi-channel CNN and GAN is proposed.The method.is divided into two stages for power system transient stability state prediction and transient stability margin prediction:The first stage is a transient stability evaluation model with cascaded multi-channel CNN,and the front and rear models predict the transient stability state of the system with non boundary samples and boundary samples respectively;The second stage predicts the system transient stability margin.Experimental results on IEEE 39 bus and a provincial 2128 bus power grid show that the proposed model performs well in transient stability assessment.Finally,in order to solve the problem of degradation of model evaluation when power grid conditions change,an adaptive transient stability evaluation method based on deep migration learning is proposed.The structure of a multichannel CNN model is adaptively optimized based on a deep network,improving the evaluation effect of the model when operating conditions change,and shortening the simulation time for model updating.The simulation results on an IEEE 39 bus power grid show that the proposed model has good transient stability evaluation performance in the face of significant changes in power grid conditions.