A Novel Data-Knowledge Co-Driven Method for Forced Oscillation Sources Localization

Accurately and quickly localizing forced oscillation sources (FOSs) is a critical task in power systems. For that, this paper proposes a data-knowledge co-driving method, which integrates dissipating energy flow (DEF) knowledge and system topology knowledge into a neural network framework. First, a frequency domain knowledge enhancement module is proposed, and this module reconstructs the signals in the frequency domain with the significance of the DEF mechanism. Second, the reconstructed signals are characterized as multimodal spatial-temporal graph data, i.e., each type of variable corresponds to a graph structure. Finally, a spatial-temporal graph sparse network (ST -GSAT) is used to extract the distribution of spatial and temporal properties of forced oscillations (FO). Meanwhile this network can effectively learn the coupling relationship between multimodal graph data. The case study shows that the proposed method has excellent localization performance in multiple FOSs scenarios. And compared with other methods, the proposed method has better robustness against PMU missing and sampling noise.