Revealing Cyclic Dynamics in Electricity Consumption: A Hybrid Feature-Based Clustering Framework

Date

2026-01-01

Author

Karakaya, Şule Şevval
Purutçuoğlu Gazi, Vilda
Bursalı, Ahmet
Erkuş, Ekin Can

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

50
views

0
downloads

Cite This

The rapid growth of smart metering and sensing systems generates vast amounts of electricity consumption data that providers must analyse carefully to manage resources and costs effectively. This study presents a novel clustering framework that significantly enhances consumption profile segmentation by incorporating cyclic signal characteristics which capture inherent periodic behaviour. We derive phase-based descriptors using the Hilbert Transform, including circular mean, circular variance, and chord distance, that accurately represent temporal cycles in the data. We compare two experimental scenarios: Case 1 combines these cyclic descriptors with Principal Component Analysis components for feature generation, while Case 2 relies exclusively on Principal Component Analysis. We apply both feature sets to three diverse electricity consumption datasets and execute two clustering algorithms, DBSCAN and Spectral Clustering, which handle nonconvex shapes and complex affinities effectively. We evaluate performance using the Davies–Bouldin Index for cluster compactness and separation, and supervised accuracy for alignment with known labels. The results demonstrate clearly that adding cyclic descriptors yields notably better clustering quality, especially for datasets with pronounced temporal patterns, and that embedding functional data analysis methods into classic clustering pipelines improves both interpretability and robustness.

Subject Keywords

Cyclic Transformation, DBSCAN, Electricity Consumption, Feature-Based Clustering, Principal Component Analysis, Spectral Clustering

URI

https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=105030337701&origin=inward
https://hdl.handle.net/11511/118843

Collections

Department of Statistics, Conference / Seminar