TY  - JOUR
AU  - Aleksandrov, Saša
PY  - 2026
DA  - 2026/09/12
TI  - Comparative Analysis of Smart Meter Testing According to International Standards with AI-Based Methods for Anomaly Detection
JO  - ICCK Transactions on Electric Power Networks and Systems
T2  - ICCK Transactions on Electric Power Networks and Systems
JF  - ICCK Transactions on Electric Power Networks and Systems
VL  - 2
IS  - 3
SP  - 149
EP  - 160
DO  - 10.62762/TEPNS.2026.407661
UR  - https://www.icck.org/article/abs/TEPNS.2026.407661
KW  - anomaly detection
KW  - machine learning
KW  - metrological validation
KW  - smart meter
AB  - Standard type verification of three-phase smart meters is performed under idealized balanced test conditions that do not fully reproduce the current unbalance, single-phase loading and voltage deviations found on real distribution networks, so that a channel-specific calibration fault can pass undetected. This paper presents a methodology combining standard IEC~62052-11 and EN~50470-3 verification with an extended set of near-real-network test conditions (unbalanced current distribution, single-phase loading, and voltage deviation from nominal), together with a statistical/machine-learning screening step applied to aggregated laboratory verification results and used to flag units for targeted extended validation. Three three-phase static meters were analyzed: two units flagged by the screening procedure and one reference unit. Under standard balanced testing all three units remained within their declared accuracy class, but under extended single-phase loading on one specific current channel the two flagged units exhibited systematic registration errors of up to approximately 4%, traced to an incorrectly recorded calibration coefficient on that channel; the reference unit remained within tolerance throughout. The results show that screening of aggregated verification data can identify meters with hidden channel-specific faults that standard balanced verification alone does not reveal, and support extending routine Type C verification with a small set of unbalanced/single-phase test points. The primary contribution of this work is the extended verification methodology and its validation on a real production case; the AI-assisted screening is employed as a triggering mechanism.
SN  - 3070-2607
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Aleksandrov2026Comparativ,
  author = {Saša Aleksandrov},
  title = {Comparative Analysis of Smart Meter Testing According to International Standards with AI-Based Methods for Anomaly Detection},
  journal = {ICCK Transactions on Electric Power Networks and Systems},
  year = {2026},
  volume = {2},
  number = {3},
  pages = {149-160},
  doi = {10.62762/TEPNS.2026.407661},
  url = {https://www.icck.org/article/abs/TEPNS.2026.407661},
  abstract = {Standard type verification of three-phase smart meters is performed under idealized balanced test conditions that do not fully reproduce the current unbalance, single-phase loading and voltage deviations found on real distribution networks, so that a channel-specific calibration fault can pass undetected. This paper presents a methodology combining standard IEC~62052-11 and EN~50470-3 verification with an extended set of near-real-network test conditions (unbalanced current distribution, single-phase loading, and voltage deviation from nominal), together with a statistical/machine-learning screening step applied to aggregated laboratory verification results and used to flag units for targeted extended validation. Three three-phase static meters were analyzed: two units flagged by the screening procedure and one reference unit. Under standard balanced testing all three units remained within their declared accuracy class, but under extended single-phase loading on one specific current channel the two flagged units exhibited systematic registration errors of up to approximately 4\%, traced to an incorrectly recorded calibration coefficient on that channel; the reference unit remained within tolerance throughout. The results show that screening of aggregated verification data can identify meters with hidden channel-specific faults that standard balanced verification alone does not reveal, and support extending routine Type C verification with a small set of unbalanced/single-phase test points. The primary contribution of this work is the extended verification methodology and its validation on a real production case; the AI-assisted screening is employed as a triggering mechanism.},
  keywords = {anomaly detection, machine learning, metrological validation, smart meter},
  issn = {3070-2607},
  publisher = {Institute of Central Computation and Knowledge}
}