On Wide-Band Oscillation Localization in Power Transmission Grids: Explainability and Improvement

The broadband oscillation caused by the large-scale integration of new energy generation units into the power grid poses hidden dangers to the stable operation of the power grid. Fast and accurate positioning of the oscillation source is the basis for cutting off the oscillation source. In order to improve the interpretability and accuracy of the broadband oscillation positioning model, this paper proposes an interpretability framework for the transmission network broadband oscillation positioning model, mainly including the improved broadband oscillation model and its interpretation framework. This model integrates graph convolutional neural network and long short-term memory network, takes transmission network measurement sampling data as input, and establishes a broadband oscillation localization model in a data-driven manner; An explanatory framework was constructed for the proposed wideband oscillation localization model, which combines global and local interpretations based on the additive interpretation of Shapley values to improve the interpretability of the wideband oscillation localization model. Based on the explanatory results, an attention feature mechanism is introduced into the localization model to enhance the wideband oscillation localization model. This article uses MATLAB/Simulink (version 2024b) to build a power grid model, produces a sample dataset, and verifies the feasibility and effectiveness of the proposed explanatory framework through numerical simulation.