Clustering prosumers' data to identify their demand response potential in smart grids

Abstract

In modern energy systems, the rise of prosumers—individuals or households that both consume and produce electricity—presents opportunities and challenges for efficient energy management. This study explores using advanced machine learn- ing techniques to cluster prosumers and analyze their consumption and production patterns to enhance demand response (DR) programs. We utilized two datasets: the Ausgrid dataset, containing data from 300 solar-equipped households in Aus- tralia, and the Danish dataset, which provides detailed electricity consumption data. For the Ausgrid dataset, we preprocessed the data, extracted relevant features, and applied clustering algorithms like K-Means, Fuzzy C-Means, K-Medoids, Hier- archical Clustering, and Spectral Clustering. The results, evaluated using metrics such as the Silhouette Score and Davies-Bouldin Index, revealed distinct clus- ters of prosumers with similar energy patterns, offering insights for targeted DR programs. In the Danish dataset, we used subsequence clustering to identify recurring patterns within the time series data. This involved dividing the time series into overlapping subsequences and clustering them. The analysis showed minimal sea- sonal influence on the clusters, suggesting other factors significantly shape energy behaviors. These insights have practical implications for energy utilities and policymak- ers. Understanding prosumer patterns can help develop effective DR programs, enhancing grid stability and energy savings. The minimal seasonal impact on energy patterns suggests that DR strategies could be universally applicable year- round. This study demonstrates the applicability of machine learning techniques to clustering analysis in energy management, providing a blueprint for optimizing energy systems in a prosumer-dominated landscape.