{"paper":{"title":"Latency-Aware Deep Learning Benchmark for Real-Time Cyber-Physical Attack and Fault Classification in Inverter-Dominated Power Grids","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Deep learning models classify power grid anomalies in under 15 ms but require 50 to 90 ms for complete inference.","cross_cats":["cs.AI","cs.LG","cs.SY"],"primary_cat":"eess.SY","authors_text":"A.P. Sakis Meliopoulos, Emad Abukhousa, Saman Zonouz","submitted_at":"2026-05-17T04:57:30Z","abstract_excerpt":"This work introduces a latency-aware benchmarking framework for evaluating deep learning models in power system anomaly detection using high-fidelity, time-domain signals generated from an industry-grade electromagnetic transient simulator. Eight neural network architectures, ranging from MLPs to Transformers, were systematically evaluated on streaming datasets representing both physical faults and cyber-attacks in inverter-dominated networks. All models successfully classified two representative multi-event sequences in real time with sub-cycle response times below 15 ms. However, although cl"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"All models successfully classified two representative multi-event sequences in real time with sub-cycle response times below 15 ms. However, although classification decisions occurred within one cycle, the end-to-end inference latency consistently exceeded three cycles, ranging from 50 to 90 ms. These results highlight a critical gap between algorithmic capability and protection-grade deployment.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the high-fidelity, time-domain signals generated from the industry-grade electromagnetic transient simulator accurately represent the behavior of real inverter-dominated power grids under both physical faults and cyber-attacks, allowing the benchmark results to inform real-world deployment decisions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Benchmark of eight neural networks on simulated power grid data finds sub-cycle classification but 50-90 ms end-to-end latency, indicating a deployment gap.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Deep learning models classify power grid anomalies in under 15 ms but require 50 to 90 ms for complete inference.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9fc529879e09fd6195db7cf63c3434d19c999487b473c290c5504d160ca45d71"},"source":{"id":"2605.17256","kind":"arxiv","version":1},"verdict":{"id":"cd5c5ba6-cca1-47a7-8380-9e09fc36988b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T23:24:24.271740Z","strongest_claim":"All models successfully classified two representative multi-event sequences in real time with sub-cycle response times below 15 ms. However, although classification decisions occurred within one cycle, the end-to-end inference latency consistently exceeded three cycles, ranging from 50 to 90 ms. These results highlight a critical gap between algorithmic capability and protection-grade deployment.","one_line_summary":"Benchmark of eight neural networks on simulated power grid data finds sub-cycle classification but 50-90 ms end-to-end latency, indicating a deployment gap.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the high-fidelity, time-domain signals generated from the industry-grade electromagnetic transient simulator accurately represent the behavior of real inverter-dominated power grids under both physical faults and cyber-attacks, allowing the benchmark results to inform real-world deployment decisions.","pith_extraction_headline":"Deep learning models classify power grid anomalies in under 15 ms but require 50 to 90 ms for complete inference."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.17256/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T23:31:20.282066Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T23:31:06.196027Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T22:01:57.855253Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T21:33:23.787057Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"6d7be1b7f882e454847ff1422fa43ba3b82e7fe066843378b89a8b2102904f18"},"references":{"count":11,"sample":[{"doi":"","year":2021,"title":"Influence of inverter-based resources on microgrid protection: Part 1: Microgrids in radial distribution systems,","work_id":"127b75c9-627b-48fb-917e-aca58fea176b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2012,"title":"Scpse: Security-oriented cyber-physical state estimation for power grid critical infrastructures,","work_id":"50e6d56e-9b2e-4bab-ab26-8a96c7b2cd5b","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Dynamic estimation-based protection and hidden failure detection and identification: Inverter-dominated power systems,","work_id":"e2e8fd1a-1f38-44d2-9c51-e595dc80a106","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Cnn-based transformer model for fault detection in power system networks,","work_id":"93299780-cc9e-4940-ac78-d45cf1d96a0e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Deep machine learning model-based cyber-attacks detection in smart power systems,","work_id":"4b08efac-4a8e-45fa-9210-e2db650bee2e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":11,"snapshot_sha256":"ae515ebc33e4919fb238bae02dc55328f53caf96a97fd89148ba62d54f728db2","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ae3c362bd56f9eb544a93b5ff055160cd7ba360c46ea61a3a13f2712fae66bb4"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}