Professor Gregor Henze's Research Group

High-Fidelity Building Performance Simulation
We developed an open-source building performance simulation test bed, the Advanced Controls Test Bed (ACTB), that interfaces high-fidelity Spawn of EnergyPlus building models, with advanced controllers implemented in Python. The ACTB leverages the Building Optimization Testing and Alfalfa platforms for managing simulations, providing an external clock, a representational state transfer (REST) application programming interface (API), and key performance indicators for evaluating the effectiveness of control strategies. The REST API allows the development of external controllers programmed in languages such as Python, which provides flexibility and a rich choice of scientific libraries for designing control sequences. We also developed three test cases based on the U.S. Department of Energy’s Reference Small Office Building to demonstrate the ACTB’s capabilities: (a) rule-based controls compliant with ASHRAE Guideline 36 control sequences; (b) an economic model predictive control (MPC) implemented using do-mpc; and (c) a deep Q-network reinforcement learning agent implemented using OpenAI Gym. The ACTB is an open-source development effort and available for download at ; development is active and ongoing.

The ACTB capitalizes on scalable, open-source software and interfaces that aspire to provide both the user-friendliness and robustness expected in commercial software, as well as the freedom to expand the ACTB beyond its intended scope. The Spawn Modelica-based models provided with the ACTB and presented in this paper are examples of high-fidelity building models, but the ACTB may be enriched with a variety of libraries, allowing for example the modeling of large-scale power grids, advanced materials, or water treatment systems. The simulation environment, based on BOPTEST-service, provides general purpose input/output interfaces that do not limit the software to a specific field or application. Finally, the RESTful API and the Python examples provided in the ACTB allow for interfacing of a multitude of external libraries out of the box and encourages users to interface or develop their own custom modules.

��

Figure 1: Overview of the ACTB framework

Figure 1 details the ACTB architecture with its two core interfaces: the OpenAI Gym interface and the do-mpc interface. The ACTB integrates high-fidelity Spawn models, for which the project team is building a U.S. Department of Energy (DOE) commercial reference buildings ensemble. The models are compiled into FMUs that are simulated using the tools included in the BOPTEST-service framework, at the center of the figure. The building emulators are provided with low-level and supervisory-level control interfaces, programmed in Python and interfaced using Modelica blocks taken from the Modelica Building Library package. Finally, the ACTB implements two interfaces to popular environments for developing advanced controllers, do-mpc and OpenAI Gym. These interfaces are complemented by a metamodeling module based on the SIPPY package, that enables the derivation of reduced order models from Spawn models using linear system identification methods, in order to propose a computationally-efficient solution for MPC prediction models and pre-training of reinforcement learning agents.

��

Figure 2: Reinforcement learning control research utilizing the Advanced Controls Test Bed

Reinforcement Learning Control
Reinforcement learning (RL) has gained significant traction recently due to its success in autonomous vehicles, robotics, and gaming industries learning through interaction without developing complex models. This has attracted the attention of the building automation and controls field due to its potential application to residential and commercial buildings to optimize complex objective functions, such as balancing multiple goals of occupant comfort, energy efficiency, and grid flexibility in place of sub-optimal rule-based or heuristic controls. However, RL control (RLC) has several drawbacks in building control applications. They may require significant training times and exhibiting unstable early training behavior, making it unrealistic to be applied directly to building control tasks. The research aims to address these issues with the assistance of simplified surrogate models and transfer learning techniques to improve the scalability of RLC applications, providing recommendations at the end harnessing the capabilities of the Advanced Controls Test Bed. The work’s contribution is to broaden the knowledge of RL's practical application in building controls where there is a lack of advanced controls with an unrealized potential for saving energy and carbon emissions, providing thermal comfort and good indoor air quality, increasing workspace productivity, and providing electric grid response.

��

Figure 3: Visual representation the WHISPER system and its features
Building Occupancy Detection
HVAC systems often operate in a continuous fashion without regard to actual human presence, leading to wasted energy consumption. We developed the WHISPER (Wireless Home Identification and Sensing Platform for Energy Reduction) system to address the energy and power demand triggered by human presence in homes. The presented system includes a maintenance-free and privacy-preserving human occupancy detection system wherein a local wireless network of battery-free environmental, acoustic energy, and image sensors are deployed to monitor homes, record empirical data for a range of monitored modalities, and transmit it to a base station. Several machine learning algorithms are implemented at the base station to infer human presence based on the received data, harnessing a hierarchical sensor fusion algorithm. Results from the prototype system demonstrate an accuracy in human presence detection higher than 95%. Using machine learning, WHISPER enables various applications based on its binary occupancy prediction, allowing situation-specific controls targeted at both personalized smart home and electric grid modernization opportunities. ��

Figure 2 shows that WHISPER is composed of 1) a set of sensor nodes, with each sensor element (2) using wireless communication based on digital backscattering and (3) powered by solar panel without the need for batteries. The sensors (4) include image and acoustic energy modalities but can be expanded to any sensor of similarly low power requirements. The (5) ultra-low power backscatter-based sensor network, composed of these sensor nodes, and (6) a sensor fusion algorithm leverage the spatiotemporal interactions of measurements from a residential building to ensure that (7) privacy preservation and low power are possible by collecting camera information in an extremely restricted fashion. (8) Camera gating, which leaves the image sensor nodes off until lower-power, and less obtrusive sensors indicate possible human activity to reduce system power consumption. Furthermore, (9) image sensors can be commanded to capture obfuscated images composed of horizontal and vertical bars that indicate the level and location of activity and require 100× less power than full frame capture from the 10k pixel array, while (10) portions of the image will be collected at higher resolution if human activity has not been ruled out. (11) Human vs. pet discrimination is performed on higher quality sub-images; (12) individual, low-level sensor modality algorithms based on spatiotemporal pattern networks, random forests, and convolutional neural networks will be fed by the various sensor data streams (image, acoustic energy, and environmental features), and a (13) high-level sensor fusion system that ingests data processes instantaneous and past low-level occupancy inferences to achieve high-accuracy occupancy detection.

��

Figure 4: Evolution of d

istrict energy systems.

Novel District Energy Systems
A scientific consensus has emerged that widespread electrification of space heating will be necessary to meet targets for deep decarbonization in the U.S. while minimizing the need for costly technologies to capture and sequester carbon emissions. The U.S. National Academies of Science, Engineering, and Medicine recently called for electrification of heating in new construction in much of the United States by 2030, for the country to reach net-zero carbon emissions by 2050 and avert the worst consequences of climate change. The European Commission concluded that averting global temperature rise of more than 1.5°C by 2100 would likely require electrification of more than 60% of energy end uses in buildings by 2050, and that a similar degree of electrification would be required in buildings in China to avert global temperature rise of more than 2°C. Given worldwide trends towards urbanization, it is expected that 68% of people will be living in cities by 2050. District thermal energy systems (DES) operating at near-ambient temperatures facilitate beneficial electrification of heating in urban districts, as well as significant reductions in source-energy use intensity (EUI) in appropriate applications. However, such systems face barriers to adoption, especially in the United States, due to their high infrastructure costs, and the very large search space of potential network configurations, which complicates their design.

Advanced district energy systems are the evolution of traditional district energy systems as they become more energy- and exergy-efficient. Historically, district energy systems started as steam-based and have been progressing towards near-ambient temperatures; further qualified as 1GDH systems to currently adopted 4GDHC systems. The next generation is termed an ambient loop system or a 5GDHC system. Figure 4 is a graphical representation on the district energy systems generations clearing showing the inverse relationship of operative temperatures and efficiency.

Commercial building data has become increasingly accessible over the last decade. Contributing factors to this increased accessibility include new building technologies, ubiquitous internet access within buildings, smart grid initiatives, updates to building infrastructure with smart devices, building facility manager requests for easier access to data, public building disclosure data through benchmarking and transparency ordinances, and many others. Regardless of the main drivers for more universal data access, commercial building datasets are now readily available and can be leveraged for advanced analyses as long as the data resolution meets the analysis requirements. The hypothesis for this research thrust is that city planners, designers, and city managers are able to screen the potential energy and decarbonization impacts of district energy systems by leveraging public building disclosure data and open-source analysis platforms such as the SEED, URBANopt District Energy Systems, topology analysis and optimization, the metamodeling framework, and core building modeling engines. Beyond direct decarbonization benefits, district energy systems combined with thermal energy storage and controls will help create grid-interactive efficient districts providing energy flexibility across a district of real buildings and fostering the power system transition away from carbon-based fuels and to renewables. A detailed bottom-up analysis is underway that will provide actionable insights into how a grid-interactive efficient district should be designed and operated for the case study of Washington, DC. This work is undertaken in the context of the International Energy Agency IEA-EBC Annex 82 Energy Flexible Buildings Towards Resilient Low Carbon Energy Systems ().

��

��Ƶ18