Energy Management Meets the AI Revolution
- Oct 18, 2018
Artificial intelligence is a buzzword in many industries, and when it comes to energy management, it can be the catalyst for innovative, money-saving operations. Determining whether one of today’s AI programs is the right fit for a property is a major decision.
Yardi Energy Vice President Matt Eggers speaks frequently at national industry events about these issues, as well as about Yardi Pulse, an integrated online platform designed to enhance energy efficiency and occupant comfort. In a wide-ranging conversation, he discusses the evolving role of artificial intelligence in energy management.
When it comes to building management systems, what does “artificial intelligence” refer to?
There are different types of artificial intelligence, and the type that’s getting the most buzz these days is machine learning, which is very recent. That said, there are very few machine learning applications for real estate in the marketplace.
For example, Yardi’s product isn’t a machine learning product, it’s an expert system. When it’s applied to AI, an expert system basically means that you take all the rules in a human expert’s head, and you put it into a system. You do have to codify the system and tell the machine what to do through a series of “if-then” rules. You can’t build an expert system to teach you how to drive a car around San Francisco, for example—that’s machine learning. (But you can build an expert system that will run a property much more efficiently.)
How does an AI program fit into an existing building management system?
It works with data like an app on your phone. Think of the AI platform as the app, and the BMS like the operating system. The AI app sits on top of the BMS. There’s a lot of BMS management stuff that our app does not replicate. We add a bunch of basic intelligence to that control system. We give it a series of resets and a new set of instructions. For instance, the app might tell the BMS to change the temperature on the air handler by a quarter of a degree.
You could also compare the relationship of the AI and BMS to a driver and a race car. I could get in and drive it around the track. I’m not a professional race car driver, so when I put my foot on the gas pedal, it probably doesn’t go all that fast. But you can take those same controls, and put an expert behind the wheel, and the car will go much faster around the track.
Would you explain how AI changes the property’s equipment operation?
A BMS, even without a product like Pulse, has a lot of sensors. Most large buildings have a central chilling plant. It produces cold water that’s set at a certain temperature, regardless of what’s happening in the building. Regardless of what’s happening in the building, the chiller makes water at 42 degrees in the summer and 48 degrees in the winter, when you don’t need as much cooling. Every 30 seconds, Pulse is re-evaluating the water temperature and changing it. It’s taking into consideration a bunch of data (or variables):
A lot of building engineers program re-sets into the BMS, but they might change them only once a season. The AI app is pulling in a variety of data from different sources, evaluating the data every 30 seconds, and sending continuous updates to those controls. Going back to the example of our race car driver, he’s not going to change the controls—he’s just going to give them a different strategy.
Another important distinction is that AI tools are cloud-based, whereas the BMS is not. The BMS is in the basement of the building, and it’s very hard to see what’s happening to the system. With the Yardi product, if we find a bug, we can push it out immediately. We add features and we fix bugs all the time. We have a cloud-based AI system, and it’s constantly getting updated.
What are the potential benefits of implementing the technology?
First, I’d start with comfort. The AI platform is responding to changes every 30 seconds, so it can make the building much more comfortable.
Another one is energy efficiency. Typically, the BMS produces the resources in case they need them, and then they constantly choke them off. It’s like driving with your foot on the accelerator and keeping your feet on the brake all the time.
It doesn’t have to be that way, if a very intelligent component is controlling the building and making all these resources every 30 seconds. It’s paying attention to the rate of change, and the rate of change of the rate of change. Say the system figures out that it’s getting warmer on the fourth floor. It can ramp up resources to that area and keep it comfortable. As a result, the system can use less resources in a steady state.
To go back to the car analogy, you don’t need to floor the accelerator and control the speed of the car with the brake. Instead, you can control the car with the accelerator.
The third benefit is operational. If the system is keeping the space more comfortable and using less energy, it will free up resources to do other stuff. When you talk to building engineers, they’ll often say, “There’s a million things that I could fix, but I don’t have time.” A system like this takes some of the load off the building engineers by reducing the numbers of emergencies that they have to deal with.
For owners and managers who want to consider adding an AI program to their BMS, what kinds of properties tend to be the most suitable? Are there any minimum requirements?
First, the buildings need variable-speed motors. Over the last 10 to 20 years, building owners have been replacing older equipment with motors that have more than one speed. We can’t do much with motors that are just on and off.
Generally, the AI piece works best in buildings that are roughly 200,000 square feet or larger. That’s both because we achieve a critical mass and because it works best with HVAC systems that have a lot of ways to control them.
What’s new for AI as it applies to energy management? What can we expect?
A: People are collecting ever-bigger data sets on how BMS performs. The data sets are still relatively small, but they’re getting bigger. Another trend is training these AI algorithms. Going back to the car analogy, you could compare it to self-driving cars. In San Francisco you’ll see Chevy and Google testing self-driving cars all the time. All those sensors and cameras are collecting the data that will be necessary to operate self-driving cars. In the same way, data about buildings is being collected that will enable them to use AI.