Anomaly and seasonal trend identification of electricity demand in university facilities using clustering algorithms

Abstract

The continuous increase in energy demand, driven by population growth, industrial expansion, and digitalization, places increasing pressure on electrical distribution systems. In this context, machine learning algorithms have been increasingly adopted in recent years to process electrical data and improve the operational performance of power networks. Evaluating their performance using real-world datasets is essential to understanding their potential in practical applications. This article presents a study of electrical demand time series from two public buildings using unsupervised clustering algorithms (K-Means, Fuzzy C-Means, and Gaussian Mixture Models). Although these algorithms have been widely applied in various engineering domains, the analyzed electrical demand datasets exhibit a combination of seasonal patterns and anomalous events that can hinder accurate cluster identification and, consequently, affect clustering quality. The quality of the results is assessed using the Silhouette and Davies–Bouldin indices. This study aims to explore the ability of clustering techniques to extract relevant information from multi-year consumption time series, laying the groundwork for future research focused on optimizing energy management in public buildings and in other customers with similar demand profiles.