Machine Learning: Making Sense of Big Data
OTHER PARTS OF THIS ARTICLEPt. 1: Technology Shaping the Future of Buildings Pt. 2: Virtual Reality, Augmented Reality: Better Visualization Aids OperationsPt. 3: This PagePt. 4: Electric Vehicles: Should Facility Managers Invest In High-Speed Charging? Pt. 5: Wearables: The Body Is The New Credential Pt. 6: Drones: An Opportunity and a ChallengePt. 7: Autonomous Vehicles: Lowering Demand for Facility Parking?
The term “machine learning” may sound like just another new marketing buzzword for the growing category of smart building technology. But computer professionals have been studying and developing machine learning for nearly 60 years. In 1959, Arthur Samuel came up with the term to describe the ability of computer systems to improve performance on specific tasks even though the answers are not pre-programmed into them. Machine learning relies on algorithms that analyze data using pattern recognition and computational learning theory.
Two branches of machine learning — supervised and unsupervised — are impacting the marketplace. Supervised machine learning uses labeled responses. For example, an algorithm could be developed to predict when a replacement belt should be budgeted based on rising operating costs.
Unsupervised machine learning looks at raw data and spots patterns. For instance, a building’s traffic patterns could be analyzed based on sensor or security camera data. The unsupervised machine learning could correlate these patterns with energy consumption and cleaning requirements in specific areas of the building.
Machine learning can help make sense of the vast amount of data that buildings can produce. “Buildings must be managed and operated with accurate data,” says Jim Sinopoli, managing principal of Smart Buildings. “Artificial intelligences via machine learning can manage and analyze masses of data. Humans can’t data mine.”
Machine learning algorithms have begun to enter the facilities management arena. A handful of companies is already applying machine learning in their facilities, says Sinopoli.
In late July, Google released some information on its use of machine learning algorithms to cut data center energy bills, according to Sinopoli. Early results include a 40 percent reduction in cooling energy needed. Google is unique: It paid a reported $500 million for an artificial intelligence start up called DeepMind. The work on cutting data center energy bills is only one DeepMind project. Another, more publicized example: artificial intelligence software that defeated a world champion at the game of Go.
Amazon, Netflix, and Google are just a few companies that have been applying advanced levels of machine learning for more than a decade, according to Shaun Klann, executive vice president of Intelligent Buildings. Now, even the risk-adverse real estate industry needs to look at machine learning algorithms, Klann suggests.
“The time horizon and tolerance levels for the end-user’s experiences in the building are growing short,” Klann says. “After all, our buildings are already run off of computers and ‘static’ programs. The next evolution toward machine learning is simply turning these static programs into dynamic ones.”
To show how facility managers could benefit from machine learning, Klann uses a commercial office building HVAC and lighting example. Historically, buildings have binary schedules: weekday and weekend or holiday, Klann explains. “But we all know that how we use and load our buildings is more complex than that,” says Klann. “Let’s look at Fridays in the summertime, the day after Thanksgiving, or the day after the Super Bowl. We know our buildings are loaded differently on these days and leveraging historical data we can automatically react to real-life conditions rather than acting as if the day after the Super Bowl was just a regular day in the office.”
Numerous tech giants are — or soon will be — offering machine learning engines for the cloud. IBM offers the Watson Internet of Things platform, and Microsoft and Amazon both have machine learning services, while. Google’s has an offering to assist others in easily building machine learning models intended to work on any type of data, of any size.
In addition, there are other avenues for enterprising companies interested in entering the facility management field. For example, Torchnet is an open-source toolkit that includes a collection of boilerplate code, key abstractions, and reference implementations that can be combined or taken apart for later reuse that are speeding development of deep machine learning.
Major building controls, HVAC, and lighting companies also are entering the field. “Companies already are developing and selling machine learning for buildings,” says Sinopoli. He believes that new building management systems eventually will be based on machine learning.
Rita Tatum, a contributing editor for Building Operating Management, has more than 30 years of experience covering facility design and technology.
Email comments and questions to email@example.com.
Machine Learning: Making Sense of Big Data