In office and apartment buildings across the country, operators are adjusting equipment systems and responding to issues so the rest of us remain comfortable.
Like referees in sports, when operators do their job well, no one notices. And when they don’t, everyone notices. Also like referees, the job of a building operator requires making judgement calls based on incomplete data.
In a good scenario, this data set is derived from an extensive (and expensive) building management system (BMS).
In most cases, where there isn’t a BMS, the only data operators have to form judgments is pieced together from spreadsheets, preventative maintenance software, utility bills, and their own intuition.
Either way, building operators are lacking the tools they truly need to excel at their jobs. That’s exactly where machine learning comes in.
Unfortunately, terms like big data, machine learning, smart buildings and artificial intelligence have become prominent so quickly, both operators and their managers are confused with how close these technologies are to market-ready, and what exactly they do.
As a result, building operations have continued to lag behind on technology adoption, even while the industry has aggressively adopted leasing, acquisitions, and tenant amenity solutions.
To exemplify this, VTS CEO Nick Romito was recently quoted as saying "Machine learning, AI, cryptocurrency is not the basics and is not important in our space today, most of this stuff will not impact business at all in the next five to 10 years."
And he’s right. For the most part.
What he’s missing is that machine learning is here today, and is being deployed across some of the most well-run portfolios in the world.
First, a quick foray into the distinction between two terms: Artificial intelligence (AI) and machine learning.
In its true sense, artificial intelligence implies a technology that can improve itself based on an understanding of its own performance. The world is not there yet.
The AI products we have today mimic human decision-making when performing a specific task (stock trading, fraud detection, customer care -> chatbots, etc.). Artificial intelligence can solve many problems, but it won’t replicate the human brain anytime soon.
On the other hand, machine learning is the process of teaching computers to make and improve predictions based on the data available, without being explicitly programmed.
Now that a sizable, diverse and granular dataset can be extracted from equipment-level submetering and IoT sensors, the tools available through machine learning represent a panacea for building operators.
When applied to a specific vertical (a pump on a factory floor runs differently than a pump in a basement of a commercial building), the application of machine learning can solve the existing problems faced by operators.
Here are some specific examples:
The schedules for critical equipment systems, such as lighting and HVAC, are often set by the BMS control engineer who originally installed the system or are operated manually by the on-site staff.
Either way, the setup in the vast majority of buildings is both imprecise and relatively static in that the schedule is rarely analyzed or updated.
This is because the data points involved in determining the degrees of occupancy is too complicated for any human, especially an operator who has many other responsibilities. Different groups of occupants arrive at different times, in different parts of the building, and put different strains on building systems.
Machine learning can take thousands of data points from equipment usage and various sensors to “learn” the true schedule of a building. By smoothing out these data points (a topic that requires an entire article to explain), this technology can provide operators with insights on exactly how to change equipment schedules to maximize efficiency, thus reducing operating costs.
If occupant behavior changes, the same algorithms can notify the building operator to make the necessary adjustments to maintain tenant comfort.
One of the issues with aspiring to maximize operational efficiency without the support of technology is the number of variables that are in play. Occupancy, temperature, precipitation, humidity, even holidays have complex and interrelated effects on building operations.
When it comes to predicting the needs of a building in the future, the effect of these variables becomes even more unclear.
Again, experienced building operators can account for each variable in an imprecise way, but it is simply impossible to expect these decisions to be optimal without the insights of machine learning. Even with granular data about each equipment system in a building, operators would still struggle to form a manual analysis.
As a thought exercise, try to forecast what adjustments should be made to a multifamily building in two days if the temperature is supposed to be unseasonably warm, there is a 50% chance of rain, and it's a national holiday.
Algorithms run this situation against the historical data of the same building as well as other similar buildings to make a prediction about operational needs. The machine learning aspect means that these predictions are constantly improved as forecasts are compared to outcomes. As more building deploy cloud-connected data collection, predictions become even more accurate.
No matter how well an operator makes adjustments to equipment schedules and forecasts operating needs, unexpected issues will still arise.
Mechanical systems break for a number of reasons and the effects may or may not be immediately noticeable.
Fault detection and diagnostics (FDD) is not a brand new concept in building operations, but it has historically been specific and static. This means that energy tracking can be used for a set number of faults and building operators can be notified whenever an issue is detected.
What is new, and only now possible due to machine learning, is fault detection based on a data set that has never been seen before. The big advantage of machine learning is that it is not limited to explicit programming, it can monitor all equipment systems in a building and detect anomalies that could not possibly have been programmed ahead of time by humans.
As building systems become even more complex and occupants become more dependent on the indoor environment, a host of unforeseen problems will occur. It’s crucial to be able to catch these issues, even if they have never happened before.
As we covered earlier, weather has a significant impact on the operational needs of a building. Good operators know this and act accordingly.
But no operator can hope to understand the complex relationships among each of the different variables. Machine learning can use a set of statistical models to do just this.
Operational decisions are not black or white, they are a matter of degrees. Being able to normalize operations against the effects of weather allows operators to compare apples to apples and have real transparency into what is happening over time.
We’ve been talking a lot about equipment systems. But this is a simplification of the complexity that makes up building operations.
Equipment systems are groups of individual pieces of equipment that function in a similar way or towards the same goal.
Unfortunately, many buildings have not codified the different levels of operation of equipment. In cases like this, machine learning can be implemented to understand the cohesion and separation of different systems.
By automatically identifying equipment correlations based on usage parameters, operators can make better decisions around runtime hours and also be notified of inefficiencies, such as short cycling.
This is important because while buildings have a lot of the same fundamentals, each has its own distinctive operating characteristics. Humans tend to extrapolate lessons learned in one situation to many others. For building operations, this is not ideal and machine learning can help avoid that scenario.
The companies that embrace machine learning early will have a competitive edge in the long run. In the same ways that large companies can flex their financial muscle, soon some real estate companies will have an insurmountable data advantage over their competition.
As the market tightens, a major differentiator in the operating income (and thus asset value) of portfolios will be in the intelligence that can be inferred from a myriad of data sources.
With all the benefits promised by machine learning, commercial real estate companies may wonder whether they should build the technology in-house or contract a vendor.
While the focus of machine learning is to make life more simple for building operators, the actual development of these technologies is incredibly complicated.
Most companies are opting to partner with a vendor, but there are a couple pitfalls to be sure to avoid.
It is important to check a vendor’s data expiration schedule. Because of the cost of data storage can add up, some vendors only maintain data for one day.
This is simply is not conducive for long-term improvements, and constraints on data retention is anathema to machine learning tools. Without long-term data storage, there is no way to continually find new value from historical data.
If you're interested in seeing firsthand how machine learning algorithms can enhance building operations, schedule a demo today!