Transient Stability Assessment of Power System Based on Two-stage Ensemble Deep Belief Network

With the development of artificial intelligence, deep learning technology has been applied to power system transient stability assessment, while a series of difficulties in selection of the model structure and optimization of the assessment performance have to be faced with. In order to give full play to the advantages of deep learning, this paper proposes an assessment method based on a two-stage ensemble deep belief network (DBN). In the first stage, the transient stability of power system is predicted after fault. Since the importance of input features and model structure, this paper chooses the relevant features such as the original electrical features, the manual experiencefeaturesand the features extracted from the stacked denoising auto-encoder as the inputs to train DBN with different structures respectively. According to the output probability, the DBN models with the different input features are synthesized to generate the prediction results of the ensemble DBN model with the credibility of the result evaluated. In the second stage, the stability margin or instability degree of the credible samples are further evaluated with DBN-based transient stability degree regression model. The simulation results in the New England system with 10 machines and 39 buses demonstrate that the proposed method has high prediction accuracy and good fault screening performance. At the same time, it can accurately measure the stability margin or instability degree of the credible samples, providing reliable referencefor the subsequent control. © 2020, Power System Technology Press. All right reserved.