Knowledge-Informed Deep Learning Method for Multiple Oscillation Sources Localization

This letter presents a novel knowledge-informed deep learning method for the fine-grained localization of forced oscillation sources (FOSs). This method can effectively identify multiple FOSs under anomalous measurements. First, a knowledge-guided block based on dissipated energy flow (DEF) is proposed. In this block, phasor measurement unit (PMU) signals are disassembled and reconstructed in the time-frequency domain to extract DEF knowledge. Subsequently, a spatial-temporal graph attention (ST-GAT) network is employed. Topology information is embedded into this network to capture the spreading patterns of FOSs. Simulation results demonstrate that the proposed method exhibits superior accuracy and robustness compared to the conventional methods.