Episode Description:
Join us for Episode 479 of the Smart Buildings Academy Podcast, where we dive into a topic that's rapidly transforming the world of building automation - AI and Machine Learning in HVAC Energy Optimization.
Whether you're an HVAC technician, BAS professional, engineer, sales specialist, or building manager, this episode is for you. Discover how AI-driven solutions are redefining the way we approach energy efficiency, predictive maintenance, and system performance - all without the need for complicated programming.
By tuning in, you'll gain insights into how these cutting-edge technologies are shaping the future of our industry and how you can stay ahead of the curve. Expect real-world examples, expert analysis, and a glimpse into what 2025 holds for AI integration in building automation systems.
In This Episode, We Cover:
- How AI analyzes building data to forecast and optimize energy usage.
- Machine learning’s role in predictive maintenance, helping you avoid costly failures.
- Real-time dynamic adjustments that improve occupant comfort while reducing energy costs.
- Anomaly detection techniques to identify inefficiencies before they become expensive problems.
- Real-world case studies showing how companies are saving thousands with AI-driven strategies.
Click here to download or listen to this episode now.
Resources mentioned in this episode
- Free Technical Skill Assessments
- Smart Buildings Academy Workforce Development Program
- Smart Buildings Academy Courses
- Control Sequence Fundamentals HVAC Course
Podcast Video
Transcript
Speaker 1 0:00
Hello everyone. Michael Roper here and welcome to episode number 479, of the smart buildings Academy podcast. In today's episode, I'm going to be discussing the role of artificial intelligence and machine learning and how we use them in HVAC energy optimization for our commercial buildings. So as I go through the podcast, I'm going to break down how AI analyzes massive amounts of data from our building systems to predict future energy usage. We'll also get into the machine learning algorithms or programs to help optimize HVAC equipment performance in real time. And then I got a couple of case studies I want to share with you on how real world buildings are already saving 1000s of dollars through using these advanced technologies. So whether you're an HVAC technician, building automation technician, engineer, you're in sales, maybe you're even a manager, or someone that's just interested in the future of bas technology, you're going to want to stick around for this one, because in 2025 I'm predicting you're going to see a lot more integration with AI models into your building automation systems. So let's go and get started. And what I want to do to get started here is define what is artificial intelligence and what is machine learning. Okay, now as I'm going through this. This is going to be a very high level overview. I'm not going to get into how to program artificial intelligence models and things like that. This is going to be like an overview on how these two technologies interface with building automation systems and HVAC. So AI Artificial Intelligence. AI is the abbreviation refers to computer systems, specifically like servers designed to perform tasks that typically require human intelligence, only these can do it much quicker, faster, and we they can do it 24/7, without taking a break, and usually can do it more accurately. Now, in the context of HVAC, energy optimization. AI analyzes vast amounts of real time and historical data, and it gets that data from sensors in the building automation system, maybe a weather forecast on the internet, and it interfaces to it, and gets that information and also building occupancy sensors, and it uses all this data to make intelligent decisions, to improve system efficiency and also performance. So that's AI. Now machine learning also known as ml, as the abbreviation, if you see ml, that's what that means. Now, machine learning is actually a subset of AI. So machine learning is in the AI model, and it uses algorithms or programs to learn patterns from your building data, and it uses that to improve system performance over time, without this complicated programming and stuff that we have To do so in HVAC systems, machine learning models predict equipment failures. Okay, so it could predict a failure before it happens. Maybe it's getting some alarms, you know, maybe on the fan keeps failing, maybe an actuator keeps failing. This helps us predict equipment failures in the future, to get a technician out there to fix it, repair or replace it before it completely fails. It also helps us optimize energy use and adjust operations in real time based on environmental and operational data. So maybe it's outside air humidity or outside air temperature, and we're have to bring in a certain amount of outside air, it's predicting how much we have to bring in and how it's going to affect our building spaces. Okay, so that's the definition of machine learning and artificial intelligence. What I want to do now is talk about how artificial intelligence analyzes and predicts future energy usage, and we do that by data collection. One of our main points here is data collection and integration. So like I talked about in the opener here, we're going to be gathering data from various sources. Maybe it's an IoT sensor. If you've ever heard that buzz word, IoT stands for Internet of Things, and all that means is these IoT sensors are connected to the internet, where we can view those off the internet. So every temperature sensor, humidity sensor, CO two, maybe it's just an occupancy motion sensor that data is being transmitted to the AI system. Also, we talked about weather forecast for predicting external conditions. Maybe the weather forecast is forecasting a very hot week next week. So that means we're going to probably have to pre cool our building before the tenants get in there, so they don't start complaining to the maintenance department. Okay. Also, it helps us gather information about energy usage patterns. How much kilowatts is that building using at certain times of the day? Certain times of the month or certain times of the year. So this data is fed into what we call an AI model. So we're getting all this data together in a model, and we're going to use that model to identify patterns and anomalies or issues. Okay, okay, another part of AI is we have energy modeling. We talked about modeling just just now, and it creates predictive models based on historical and real time data to forecast energy demands. So we look at a time like the weather forecast example, if we're looking a week ahead and we're predicting it's probably going to increase our energy demands next week, we're going to make some modifications for that. An example would be a building's AI might predict increased HVAC usage on maybe Mondays, whenever occupancy is very high in a certain space, okay, or a certain building, everybody comes in on Monday, but they're off on Tuesday, or maybe you have half the staff on Tuesday through Friday. So that means we're gonna it's gonna predict that we're going to have to leave our equipment on longer to cool those spaces. Also, we have what we call dynamic adjustments, so AI can doesn't only monitor inputs and weather information and things like that. We can also do use this to make adjustments to system settings based on real time forecast and live data. So we can adjust HVAC set points on predicted occupancy. So maybe it's looking ahead to see if a space or a building is going to be unoccupied in the future, and in the summertime, if it's going to be unoccupied, you know, all day or certain times of the day, we can raise that set point up in the summertime, we can lower that back down in the wintertime, okay? Since we have no tenants in the building, we don't have to worry about, you know, our tenants being comfortable or comfort. Another example would be like modulating air flows and temperatures in different zones to balance comfort and energy use. Okay? The next point we have an AI is called anomaly detection. And what this means is the AI model can identify like an unusual energy spike or maybe equipment inefficiencies. For example, we might be having some alarms on a damper. Every once in a while, you might be getting those in, or the AI model might be seeing that it's not the damper might not be getting enough air flow or too much air flow, so that would cause excessive air flow and energy waste into the space, and we could get a technician out there to look at that actuator and maybe calibrate it before it becomes a bigger problem introducing too much outside air, temperature or humidity into our building. Next is energy cost optimization, and this is where AI is going to integrate with our utility pricing models from your electric company, or maybe the natural gas company, wherever you get your energy source from, it can integrate to that, and by doing that, it can tell what the peak hours are. So peak hours are usually in the summertime, whenever the electricity is the most expensive on the rate. So you want to start load shedding some of your equipment during the summertime. And it integrates to that to know exactly what those times are, to be aware of that and when we can raise set points in the summertime and then lower those back down in the wintertime. So okay, so that covers artificial intelligence. We're going to go down to machine learning now, algorithms and how that helps us with HVAC energy optimization. The first one is supervised learning. Okay, so these, this is referred to when models are trained on historical data, and then they predict specific outcomes. For example, the daily energy demand for a building we our model is trained by the information over time, so it's going to learn that over time, at certain times of the day, we might be consuming more energy than others. Okay, so it's going to predict those specific outcomes and put that into the AI model. It can also help us with the likelihood of equipment failure, like we talked about before. If we're seeing a lot of alarms coming in, like on air flow or something like that, that's a trigger that to send maintenance out there, or a technician out there to take a look at the problem and see if we can make some adjustments before it becomes a bigger problem. So that's supervised learning.
Speaker 1 9:29
The next one we have is reinforcement learning. So this is where we have an algorithm or a program to learn trial by error, to optimize control strategies. Now back in the day, we used to have to do this out on the job site manually, right? So we, you know, we tweak a schedule, a time of day schedule, and maybe, you know, we turn the equipment off too early or too late in the building. The load conditions got out of control. So and we'd have to go out to the job site, and we'd have to sit there and spend cost. Time tweaking the system to get it to work right. Reinforcement Learning uses the AI model to perform this software trial and error and hopefully come up with a way to make sure that works well. So as more data is collected into the system, the system is going to improve the decision making on that. Next we have our clustering algorithms, and this is used for grouping similar data points, such as building zones, with similar usage patterns. I mean, so the AI model is going to predict like you might have accounting and you might have marketing, and they have similar usage patterns because they don't use their building their zones, they don't have a high heat load in them, but maybe your server room over here, or maybe engineering department, they have a very high heat load. So we're going to put those in their other cluster over here, and this enables us to have some strategies, like we're going to put these in building zones, and we're going to schedule these, these zones to turn on and off at the same time, and also be able to raise the set points in the summertime and lower them in the wintertime. Okay, next we have what we call deep learning, and this is neural networks analyze complex non linear relationships in HVAC systems, and this has the interaction between weather occupancy and energy use. So an example of deep learning model would be like it can predict the optimal ventilation rate for a mixed use building. Okay, so we need to determine, you know, how much fresh air we need inside that building based on all the conditions in there. So that's probably going to be with the use of CO two sensors as well for demand control, ventilation. Okay. Next the last one in machine learning, we have what we call digital twins. So this is power digital twins simulate building systems allowing operators to test energy optimization strategies in a virtual environment before implementation. So what you'll do is you'll have two different models. You'll have one building system that you have set up right now. Then you'll have a digital twin over here too, as well. These are not physical systems by any means. These are digital. So we can make a simulation on these. And what we can do is we can tweak the system. We can say what's going to happen if we raise the set point to 78 degrees at two o'clock in the afternoon, and then we'll run that simulation see what happens. And we can also do at the same time the same conditions in the digital twin. Well, let's do it at 79 and see how far past set point we're going to go to as well. So that just is a representation of simulating two different building systems and seeing which one works better. Okay, so that wraps up my overview of AI and machine learning, just a high level overview of some different techniques that's involved in that. What I want to do now is I want to go over some real world building applications that's used in the industry. And the first one we have here is comes from this company called Siva logistics. All right, so what they did is they, they use this, this case study reducing energy consumption with AI driven HVACR technology. So this is a warehouse, temperature controlled warehouse, and what they did is they teamed up with this company called be bright. Now be bright is a consulting company, and they work with companies like Siva to reduce energy usage with the use of AI. So on this project, it looks like we have a warehouse two of its multi level refrigerated Singapore warehouses where temperature control is critical. And what they did is they tied their IoT sensors, their temperature sensors, their humidity sensors, and their power meters and real time monitoring dashboards, and they enable, they enabled automated and predictive control of their auto air conditioning systems. And what they did is they were able to achieve 30% reduction in energy savings. As we can see down here, they saved 30% on their building or two warehouses and 850,000 kilowatts of energy saved per year by doing that. Okay, so they integrated AI, and they were looking for better ways to control their building by doing this, I got another case study over here, and this was done at La Jolla University. This is from the brain box AI. This is a company that actually does AI software algorithms for companies, and they worked with La Jolla University on this. So La Jolla University, they're really big into LEED. It looks like they're a LEED Gold certified building. So they're into energy efficiency and environmental type. Stuff. The building type is a university that on this one, it was built in 2015 looks like it's a retail first floor. It's got some classrooms in it, lecture halls and offices. It's 150,000 square feet and 10 floors. So the H back description on this building, it looks like it had electric water cooled chiller. It had some gas boilers, Vav air handler with vav boxes, fan cool units, radiant heat dedicated outdoor air unit, probably in a kitchen, maybe heat recovery unit and some CO two control. So what happened? What they did here is they integrated Brain Box AI with their building automation system through BACnet, the BACnet protocol. Because remember, we can integrate our BACnet objects into the AI modeling system. Okay, we just pass those points right to the AI modeling system. So that's what they did here. And they used another company called what time that's and they're kind of like a sub partner, I think of brain box. And what they did is they were looking for some load shedding strategies here. So during low emission events, they decreased space temperature. Okay. Now, low emission events. Means that during the the the electric company, whenever they're consuming less electricity, that's what that means. So low emissions, low electricity consumption, they're going to decrease the space temperature set points to cool their building before it gets too hot later on in the day, so it's pre cooling the building, then they return the set point to normal, okay, allowing the building to kind of drift. Okay, keeping that cool in there a little bit lower than usual, and it's going to drift when, when the grid relied more on fossil fuels or electricity from the power company. So what they did here, I don't know if it shows the energy savings on this, but yeah, it does right here. So it took, you know, a reduction in HVAC carbon equivalent of 15% and it reduced their HVAC electrical consumption by 10% so they saved 10% on their first year by doing that. And it also got rid of some CO two emissions, and that's kind of important too, because some cities and some states have regulations on and requirements to reduce greenhouse emissions. So this can also be powerful in that case. Otherwise, they give you fines. So we got to be careful with that too, with our customers to make sure they're not consuming too much energy. Okay, so those are two examples I wanted to talk about and kind of share with you that's being used out in the real world at this time here, and that's really going to wrap up our presentation on this podcast. So thank you very much for joining me and listening to this podcast on the smart buildings Academy, and we hope you learned a lot from this episode. If you found this episode valuable, please leave us a five star review on Apple products or Spotify. If you're watching on YouTube, be sure to like, comment and subscribe, and if you're tuning in on LinkedIn, please share this with your network to help us reach more industry professionals. As always, this episode will be on our website, at podcast, dot smart buildings academy.com, forward slash 479, we look forward to seeing you on the next episode. Take care and goodbye.
