EUVIMEDEUVIMED entdecken
Crossref

Delayed Load-Pattern Mining for Operational Irregularities in Smart Grids

Kevin Chun-Ho Wong, Emily Hoi-Yan Chan, Marcus Ka-Wai Lee

The Journal of Applied Engineering and Technologies · 01.06.2026 · 1 (2)

DOI: 10.64744/tjaet.2026.269 ↗

Kurz gefasst

Fachartikel in der Zeitschrift The Journal of Applied Engineering and Technologies, online erschienen am 01.06.2026. Verfasst von Kevin Chun-Ho Wong und 2 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Smart grid operation systems produce continuous data streams from substations, smart meters, distributed energy resources, transformer sensors, voltage regulators, and control commands. These streams are affected by load variation, weather conditions, renewable generation, demand response, and grid reconfiguration. Conventional online clustering models may adapt too quickly to repeated abnormal patterns, especially when abnormal voltage fluctuations or malicious control behaviors occur over extended periods. This study develops a stream clustering method with delayed adaptation for detecting abnormal smart grid operation patterns. The model maintains dynamic clusters for normal operating states and applies an adaptation-delay mechanism to preserve sensitivity to repeated but risky deviations. A grid-state deviation score is calculated using voltage variation, current imbalance, frequency drift, control-command repetition, and feeder-level load changes. Experiments are conducted on a smart grid streaming dataset containing 1,920 feeder lines, 48 substations, 136,000 smart meters, and 64 operation indicators collected at 10-second intervals over 90 days. The dataset contains 746 million streaming records and 3,180 annotated abnormal segments, including transformer overload, abnormal voltage sag, feeder switching error, suspicious control-command repetition, and distributed generation instability. The proposed method reduces median detection delay from 18.9 minutes to 5.6 minutes compared with DenStream. False alerts decrease to 2.8 cases per feeder-month. The online clustering engine processes 64,000 records per second, with 1.14 GB memory consumption during full-grid monitoring. The delayed-adaptation module prevents 2,260 abnormal operating segments from being merged into normal clusters during repeated load fluctuation periods. These findings indicate that anti-habituation clustering can strengthen online anomaly detection in dynamic smart grid data streams.

Zitieren

APA

Wong, K. C., Chan, E. H., & Lee, M. K. (2026). Delayed Load-Pattern Mining for Operational Irregularities in Smart Grids. The Journal of Applied Engineering and Technologies, 1(2). https://doi.org/10.64744/tjaet.2026.269

Vancouver

Wong KC, Chan EH, Lee MK. Delayed Load-Pattern Mining for Operational Irregularities in Smart Grids. The Journal of Applied Engineering and Technologies. 2026;1(2). doi:10.64744/tjaet.2026.269

BibTeX
@article{Wong2026,
  author = {Wong, Kevin Chun-Ho and Chan, Emily Hoi-Yan and Lee, Marcus Ka-Wai},
  title = {Delayed Load-Pattern Mining for Operational Irregularities in Smart Grids},
  journal = {The Journal of Applied Engineering and Technologies},
  year = {2026},
  volume = {1},
  number = {2},
  doi = {10.64744/tjaet.2026.269}
}
Kostenlos starten →