Intelligent Building Agents

Summary

As of 2020, commercial buildings use approximately 18 % of primary energy (35 % of electricity) in the US at a cost of around $190 billion. Approximately 35 – 40 % of that energy is used for the operation of heating, ventilating, and air conditioning (HVAC) equipment. In addition, the HVAC industry is facing a workforce shortage that impacts its ability to maintain and repair equipment in a timely manner.  The application of artificial intelligence (AI) to advanced control and monitoring systems (including fault detection and diagnostics) is increasing in the industry, but serious challenges exist. These technologies are not currently widespread for reasons including the cost of implementation, a lack of trust, and the lack of building automation systems (BAS) in most commercial buildings in the US. While 60 % of commercial buildings over 50,000 ft² have a BAS, 87 % of smaller commercial buildings do not, so the use of AI to reduce operating costs and mitigate the impact of workforce shortages is limited. The focus of this project is to demonstrate, develop, and evaluate AI technologies for applications in commercial buildings in the US.

Description

Objective
Demonstrate the feasibility of AI control techniques for reducing the energy costs of HVAC systems in commercial buildings and create a research infrastructure, the Intelligent Building Agents Laboratory (IBAL), suitable for ongoing demonstration, development, and evaluation of advanced AI technologies. One key objective is to address concerns about the practical application of AI to HVAC operations across a wide array of building types. 

Technical Idea
AI can be used to operate HVAC equipment and systems more efficiently by making more intelligent control decisions, automatically finding and diagnosing faults, and allowing existing facilities managers and technicians to do their jobs more effectively. More efficient operations lead to lower operating costs and a building that is more responsive to grid needs, including transactive energy signals.

Intelligent agents are one AI approach that is being investigated. They have been successfully implemented in applications including search engines and robotic systems, and a considerable amount of information already exists in the AI community on different agent architectures (e.g., deliberating, reactive, and hybrid), agent design and implementation, and agent programming. Intelligent agents know or can learn the performance and status of the systems and equipment they monitor and can communicate and collaborate with other agents to achieve a common goal, such as minimizing the cost of operations, maximizing occupant comfort, identifying and diagnosing problems, etc.  Intelligent agents can be deployed in a distributed manner, which will decrease the computational requirements of the system. Agents also learn from data, so the controller can adapt to different systems automatically. This work requires simulation and prototyping tools for rapid testing of AI approaches as well as a laboratory facility where promising approaches are tested.

Research Plan  
The Intelligent Building Agents project has two primary components: 1) the development and evaluation of AI-based control algorithms for HVAC operations and 2) support for other research efforts that require access to real HVAC systems.

The Intelligent Building Agent Simulation (IBASIM) program is a simulation platform that includes a validated and calibrated simulation of the IBAL, a virtual building model, and baseline control algorithms. IBASIM is used to quickly test a new control algorithm, revise the algorithm, and examine its performance under different scenarios, all while comparing it to the baseline controllers. The most promising algorithms are then tested in the IBAL to determine how well they perform in a real system. As with all simulations, IBASIM does not capture all of the dynamic, transient, and stochastic behavior of real equipment, so it is critical to assess a promising algorithm using real equipment. This can lead to further refinement of the algorithm, provide a more realistic assessment of how much that approach reduces operating costs in a real system, and demonstrate how complex it would be to implement.

It is critical to consider how these algorithms will be implemented in the real-world building stock; the cost of implementing the approach cannot be greater than the savings realized from operating the system more efficiently. Approximately 90 % of the commercial building stock in the US is under 50,000 ft², and only 13 % of these buildings have a building automation system, so affordably deploying AI approaches to the control and monitoring of these buildings will require creative solutions. There will not be a single approach that works in all buildings – some will require a BAS, some will only work with a new type of BAS that is developed to support these advanced algorithms, and some will have to work in buildings that do not have a BAS. In addition, the algorithms will need to be implemented by facilities managers and technicians and enable them to do their jobs efficiently.

The IBAL also supports other research needs. It provides a flexible platform on which ideas can be tested under repeatable conditions before they are deployed in larger, more complex commercial buildings. For example, the IBAL is being used to test automated fault detection and diagnostics algorithms and in support of the development of a semantic framework for the interoperability of building systems. Additional opportunities for collaborations that support American commerce continue to be explored.

Major Accomplishments

• The Intelligent Building Agents Laboratory (IBAL) is a unique, flexible laboratory facility consisting of a mixed system of chillers, thermal energy storage, and air distribution components capable of supporting building system optimization research under reproducible operating conditions.
• Data and software are available:
  • https://data.nist.gov/od/id/mds2-2751
  • https://data.nist.gov/od/id/mds2-3058
  • https://github.com/usnistgov/IBA-Project