Optimizing Facility Operations by Applying Machine Learning to the Army Reserve Enterprise Building Control System
Emily Wendel | Pacific Northwest National Laboratory
This effort will apply machine learning methods to building operation data to automate the identification of operational issues and energy-savings opportunities. Established machine learning algorithms will be trained to recognize patterns in combinations of whole-building utility and building system-level data in near-real-time, allowing for continuous monitoring and insight across an entire portfolio of buildings. Optimization opportunities will be delivered to building management staff through the Army Reserve’s Enterprise Building Control System (EBCS) Building Health Monitor report, enabling Army Reserve to establish accountability for responding to identified measures.
Advanced meters and building control systems (BCS) have the potential to enable significant reductions in energy use through demand response and controls optimization, contribute significantly to meeting U.S. Department of Defense (DOD) energy-reduction goals, and reduce the mission impact associated with downtime and outages. Applying advanced analytic tools, like machine learning, to these data has the potential to identify complex relationships between dependent variables that basic rules-based tools cannot.
The key technology demonstrated by this project is a class of data-driven, self-improving algorithms categorized as machine learning algorithms. A machine learning algorithm is capable of adjusting parameters automatically from the data fed into the algorithm. This feature makes machine learning effective in solving nonlinear, dynamic, and ever-changing problems such as image recognition, voice detection, and computer visual analysis. Machine learning algorithms are used to make predictions about a target output, which become more accurate as the algorithm is exposed to diverse samples in the training data. Typical machine learning tasks include regression (the prediction of a continuous value), classification (the prediction of categorical value), and optimization strategy identification. In a buildings context, example predictions could be baseline energy use (regression) and fault detection (classification). This could occur at the whole-building level or for subsystems (e.g., boilers, chillers). This project will implement proven machine learning techniques to automate the identification of operational issues and energy-savings opportunities.
Machine learning methods have not yet been deployed on large sets of buildings for the military. The Army Reserve has been progressive in developing data-driven processes to optimize operation of facilities. By integrating a BCS to the Enterprise system and optimizing controls, the actual energy savings have ranged from 5% to 30% (average of 14%). Implementation of machine learning is expected to add an additional 5% to 10% savings from advanced optimization. Because Army Reserve buildings are geographically dispersed, the benefits of fault detection and optimization for buildings on the EBCS are also expected to generate savings from reduced travel and contracted trouble calls, which will be validated through the work order system data.
The Army Reserve is a willing partner to share ideas with other commands within the Army and would likewise willingly share with others within DOD. The machine learning tools are readily applicable to commands that are connecting their BCS to a central server and facilities with BCS at larger DOD installations. These tools are not limited to a single platform which makes this Environmental Security Technology Certification Program (ESTCP) applicable and relevant to all of DOD.