Paper Conference

2018 Building Performance Analysis Conference and SimBuild co-organized by ASHRAE and IBPSA-USA


Adaptive Multi-Agent Control of HVAC Systems for Residential Demand Response Using Batch Reinforcement Learning

José Vázquez-Canteli, Stepan Ulyanin, Jérôme Kämpf, Zoltán Nagy
The University of Texas at Austin, Austin, TX
Georgia Institute of Technology, Atlanta, GA
Haute Ecole d'Ingénierie et d'Architecture Fribourg, Fribourg, Switzerland

Abstract: Demand response allows consumers to reduce their electrical consumption during periods of peak energy use. This reduces the peaks of electrical demand, and, consequently, the wholesale electricity prices. However, buildings must coordinate with each other to avoid delaying their electricity consumption simultaneously, which would create new, delayed peaks of electrical demand. In this work, we examine this coordination using batch reinforcement learning (BRL). BRL does not require a model, and allows the buildings to adapt over time to the optimal behavior. We implemented our controller in CitySim, a building simulator, using TensorFlow, a machine learning library.
Pages: 683 - 690