IPIGN: An Interpretable Physics-Informed Graph Network Multitask Cascading Failure Diagnosis Method for Distributed Energy Systems

Cascading failure such as false tripping of nonfaulty branches and distributed generators forced off-grid in distributed energy systems, which significantly complicates originating fault diagnosis base on feeder terminal unit (FTU). To address this issue, an interpretable physics-informed graph network (IPIGN) multitask cascading failure diagnosis method is proposed for distributed energy systems. First, a multichannel self-masking mechanism is proposed to solve the problem of originating fault being hidden due to the superposition of multiple features, reconstructing the fault feature through spatial attenuation characteristic in the spectral domain. Second, a leapfrog neighborhood awareness strategy is proposed to solve the problem of incorrect diagnosis caused by similarity of local node features, performing topological local to global differentiated feature aggregation. Further, an interpretable multitask fault diagnosis framework is proposed to ensure the transparency of the cascading failure diagnosis process, which supervises the feature decoupling through the physical mechanism. IPIGN overcomes the limitations of cascading failure features that are complex to parse, while enabling the detection, location, phase selection and identification of originating fault. Finally, the validity of the proposed method is verified by actual field failures and a semiphysical hardware platform.