Bairrão, Diego (Polytechnic of Porto), Ramos, Daniel (Polytechnic of Porto), Faria, Pedro (Polytechnic Institute of Porto), Vale, Zita (Polytechnic Institute of Porto)

Improving Load Forecasting with Data Partitioning: A K-Means Approach to an Office Building

Scheduled for presentation during the Invited Session "Optimal operation and control in smart grids" (ThuS2T2), Thursday, July 11, 2024, 11:50−12:10, Grand Amphitheater

12th IFAC Symposium on Control of Power & Energy Systems, July 10-12, 2024, Rabat, Morocco

This information is tentative and subject to change. Compiled on January 2, 2025

Abstract

In recent years, the energy landscape has undergone significant transformations, characterized by the integration of renewable energy sources, smart grids, and the proliferation of IoT-enabled devices. As a result, the efficient management of energy resources has become paramount, requiring advanced methodologies in load forecasting and clustering. This article presents an enhanced methodology for short-term load forecasting that focuses on load consumption profile recognition within a smart building environment. The methodology is designed to analyze and identify recurring load consumption profiles and measures of sensors, thereby enhancing load consumption profile recognition capabilities within the smart building context. The interaction between single and grouped datasets is explored to enhance the accuracy and interpretability of predictions, contributing to optimized energy consumption and providing valuable information for demand response programs. The default forecasting methods used in the methodology are artificial neural networks and k-nearest neighbors. For comparing results and evaluating the proposed approach, XGBoost is also employed. The dataset is selected from a specific database, and the clustering method, partitioning type, is applied with k-means. The results, validated with error evaluation models and statistics, reveal the advantages of the proposed approach, especially with three clusters, where the results achieved by the Artificial Neural Network are the best. The clustering process, particularly the partitioning type, demonstrates a strong capability in improving load forecasting in smart buildings and helps understand load consumption patterns and achieve energy savings.