Enhancing electrical load profile segmentation for smart campus energy management

Efficient electrical load profile segmentation is essential for optimizing energy management, facilitating data-driven decision-making and operational efficiency. This paper addresses a significant gap in existing clustering methodologies by proposing an adaptive framework designed to identify and characterize minority groups in unbalanced datasets, a challenge often overlooked by traditional clustering techniques. By leveraging similarity analysis through metrics such as Cosine Similarity, Pearson Correlation Coefficient, Minkowski Distance, and Dynamic Time Warping, the methodology incorporates sensitivity analyses to establish dynamic thresholds for robust classification. The research benchmarks nine clustering techniques under complex and imbalanced conditions, evaluating their performance using established metrics, including the Silhouette Coefficient, Calinski-Harabasz Index, and Davies-Bouldin Index. Key contributions include the development of adaptive thresholds and iterative refinement processes to improve minority group detection, ensuring more accurate and representative clustering outcomes. When applied to extensive datasets from a smart campus, the proposed approach demonstrates enhanced clustering accuracy and the ability to uncover nuanced consumption patterns. The results confirm the methodology's effectiveness across diverse clustering techniques, highlighting its scalability and adaptability. These findings contribute to advancing load profile segmentation, underscoring the importance of addressing dataset imbalances and presenting a versatile solution to improve clustering interpretability in smart grid applications.