Conference Paper

Proceedings of SimBuild Conference 2024

Autonomous Load Forecast Framework with Dynamic Model Selection

Christoph Gehbauer ¹, Nicholas Deforest ¹, Peter Grant ¹, Manfred Tragner ³, José Baptista ², Douglas Black ^1
1 Lawrence Berkeley National Laboratory, United States of America
² University of Trás-os-Montes and Alto Douro, Portugal
³ University of Applied Sciences Joanneum, Austria

Abstract: The accurate forecasting of weather conditions and electricity demand is of great importance in smart building and distributed energy resource operations. It ensures efficient resource allocation, operational cost reduction, and mitigation of environmental impacts. Traditional forecast methods often fall short due to their reliance on simplistic statistical techniques and limited adaptability to complex, dynamic patterns. Machine learning has emerged as a powerful (but complex) approach for improving the accuracy of such predictions by leveraging advanced algorithms. In this context, we introduce a framework which hosts a publicly available library of traditional and state-of-the-art machine learning models tailored for weather and electricity demand forecasting in buildings. Models are dynamically selected and combined based on an internal optimization algorithm to autonomously operate without user interaction and to ensure optimal forecast performance over the lifetime of the installation.
Keywords: forecast, occupancy forecast, machine learning, dynamic selection, hybrid model
Pages: 229 - 240
Paper: