Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA

     

Advanced Control of Dynamic Facades and HVAC with Reinforcement Learning based on standardized co-Simulation

Christoph Gehbauer 1, Andreas Rippl 1,2, Eleanor Lee 1
1 Lawrence Berkeley National Laboratory
2 University of Applied Sciences Technikum Wien


DOI: https://doi.org/10.26868/25222708.2021.30432
Abstract: With increased complexity due to time-variable renewable electricity supply and associated variable cost, it has become evident that management of building energy demand as a resource is essential for grid stabilization. However, techno-economic and human constraints make such solutions non-trivial. Reinforcement learning (RL), a discipline of machine learning, was explored to take on this challenge. The open-source Functional Mock-up Interface - Machine Learning Center (FMI-MLC) was developed to provide a standardized interface of RL and simulation environment, through the FMI industry standard for co-simulation. RL demonstrated its ability to operate an electrochromic window and heating, ventilation, and air conditioning system (HVAC), and reduce demand during critical periods of the electric power grid.
Keywords: reinforcement learning, machine learning, co-simulation, functional mock-up interface
Pages: 231 - 238
Paper:
bs2021_30432