Achieving national goals of net zero energy buildings requires substantial reduction in the energy consumption of commercial building systems. Although significant progress has been made in the integration of building control systems through the development of standard communication protocols, such as BACnet and BACnet/IP, little progress has been made in making them "intelligent" about optimizing building system-level performance. The focus of this project is to demonstrate the potential for distributed, intelligent software agents to perform this optimization and to develop a research infrastructure suitable for development and testing of advanced agent-based optimization techniques that can improve the energy and comfort performance of building systems.
Objective - To demonstrate the feasibility of intelligent agent control techniques for reducing energy consumption in commercial buildings by optimizing the performance of interacting building systems and to create a research infrastructure suitable for ongoing development and testing of advanced agent-based optimization techniques.
What is the new technical idea?
The new technical idea is to adapt advances in intelligent agent technology to solve the control optimization problem. Intelligent agents have been successfully implemented in a variety of applications, including search engines and robotic systems, and a considerable amount of information already exists in the artificial intelligence 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 cost of operation and/or energy consumption, maximizing comfort, identifying and diagnosing problems, etc. Intelligent agents make it possible to solve the problem of building system optimization in a distributed manner, which greatly simplifies the computational methods required. Developing intelligent agent technology for this application requires a combination of simulation and prototyping tools along with laboratory facilities that use real mechanical systems that operate in a reproducible but realistic manner to test agent performance.
What is the research plan?
An intelligent building agent simulation program (IBAS) has been developed and used to evaluate some prototype intelligent agents for their suitability to optimize the performance of building systems. The results demonstrate that intelligent agents have the potential to solve the optimization problem for building systems and to be used for automated fault detection and diagnostics. A new laboratory has been constructed that will enable research on intelligent agent control of a mixed system of chillers, thermal storage, and air distribution components under realistic and reproducible operating conditions.
All equipment in the hydronic and air systems in the laboratory have been commissioned. All systems can be operated manually, though most of the laboratory can also be operated in an automated manner. Improvements to and analysis of various modes of automated control, including the use of agents, will be part of the research program. The laboratory was designed so that different systems can be isolated and tested independently.
A software model of the hydronic system in the laboratory has been developed and will be combined with an air system model that is currently under development. The final model will enable rapid evaluation of potential control algorithms. The most promising algorithms can then be deployed in the actual laboratory for evaluation in real systems. Data acquired from laboratory operations will be used to validate the model at both the component and system levels. The validated model can also be used to generate data for use in the development of machine learning models.
Work is currently underway to develop machine learning models of components and reinforcement learning based methods for distributed optimization. These techniques will be tested in the laboratory, evaluated, and improved. As promising techniques emerge, the research will be expanded to include field testing that is expected to involve CRADAs with control system manufacturers and building owners in order to verify laboratory results and speed commercialization of the technology.
Lessons learned from this research will also enable development of draft specifications for several intelligent building agents (e.g., an Air Handling Unit agent, a Chiller agent), including a list of functions to be included in each agent, technical design features, software architecture, and methods for system identification, communication, and negotiation among agents. NIST will work with standards writing organizations, such as ASHRAE, to develop guidelines and/or standards based on these specifications to facilitate the implementation of intelligent agents in building control applications.