What is an Intelligent building? Theory & practice

Interview with Heikki Ihasalo by Lada Hensen Centnerová

Professor of Practice Heikki Ihasalo (D.Sc.) works at the Department of Electrical Engineering and Automation at Aalto University His professorship focus is on smart building technologies and services. His research activities are within building data semantics, advanced control solutions, smart workplace applications and smart building rating systems. He has over 20 years’ experience in research and development within smart buildings both in academia and industry. In addition to Aalto University, he works as an innovation director at Granlund.

LHC: You are a professor of practice focusing on smart building technologies. What is your definition of intelligent building?

HI: That is a good question. I would start with the basics. Intelligent building must provide a good environment and at the same time must be using as little energy as possible. And nowadays it is not only about energy consumption but also about the flexibility of energy demand. That is the traditional point of view to intelligent buildings. Already four years ago we widened this traditional point of view on user centricity a bit.

LHC: Yes, I think you mean the paper ‘Smart workplace solutions – can they deliver the offices that employees have been waiting for?’ based on the PhD research of Laura Remes [1].

HI: Indeed. We noticed already a few years ago that there were coming a new kind of startups to our area, providing mobile applications for building users to find and book meeting rooms but also to see where printers are located, type and opening hours of cafes in the building. Especially new students at our university really appreciate this kind of applications.

LHC: This kind of features are part of BREAM and LEED certification schemas, so if building owners want to score high in such a certification, they need to have these applications.

HI: That’s true. Recently introduced SRI (Smart Readiness Indicator) under the EPBD rates buildings in 3 key functionalities [1]:

·         Energy efficiency

·         Grid flexibility

·         Adaptation to users’ needs

The adaptation to users’ needs is defined in terms of four attributes: comfort, convenience, promoting health and wellbeing and the building’s ability to provide information to occupants. The last attribute are mobile applications we are here speaking about.

LHC: I was shocked by one figure in the paper we are referring to. Here in Figure 1. Could you explain this figure, please?

HI: In our research we were looking at 30 different mobile applications where we identified 36 smart features and then based on the frequency of their occurrence, we made a graph of the most relevant features (for the users) and that is here the Figure 1.

Figure 1. The core and two layers.

LHC: Are you saying that the building users are not very much interested in the quality of their indoor environment, but they are more concerned about the safety of having an appropriate working place?

HI: Yes, exactly. These were results based on the technology available to building users (features that were mentioned in the applications). In our later research we interviewed many facility managers who use this kind of applications in their buildings and asked their point of view [2]. Also, here IEQ and even building performance were not in the core of the graph.

LHC: Let’s come back to the definition of intelligent building. T. Derk and J. Clements-Croome wrote in 1997 in their paper, “What do we mean by intelligent building?” [3] …intelligent buildings can cope with social and technological change and also are adaptable to human needs… Is that the good definition?

HI: That sounds like AI (Artificial Intelligence). After collecting data from using a building, you see, for example, how people are using the meeting rooms. If there is more demand for smaller meeting rooms, you can adapt the building and make from one big meeting room a few small meeting rooms. The space is to meet the requirements of the users. That is how buildings adapt to human needs, so yes, I like that definition.

LHC: This is very different from what was originally seen as an intelligent building in the ‘80s [4]: The first intelligent buildings were mainly used to control heating, ventilation and airflow (HVAC).

HI: I would say that to some extent that is still very important. Also, SRI (in EPBD) concentrates more on energy efficiency and flexibility but there are also features related to comfort and convenience of use.

LHC: OK. Let’s talk about practice. How can you make a house smart/intelligent in terms of energy efficiency?

HI: I can explain it with an example on room level presented in Figure 2. The traditional way would be to have a PI controller with a set point (blue line). In this case is the set point most of the time on 21,5°C. Then you have a valve (red line) which goes from 0 to 100% open. The PI controller is balancing the room temperature (orange line) by opening and closing the valve.

Figure 2. Room heating demand response.

If we want to control the room temperature and at the same time optimize the cost of heating, we need to add a smarter solution, machine learning. The yellow line in our example represents the price of electricity, which has mostly its peak between 7.00 and 9.00 in the morning. This system preheats the room to avoid heating indoors in the price peak. This example still needs a bit of tuning to the model because right now, there is a short period of heating between 8.00 and 9.00.

LHC: You used the term ‘machine learning’. What does it mean exactly? And what is the difference with ‘reinforcement learning’?

HI: Reinforcement learning is one of the methods used in machine learning. We usually speak about three groups of machine learning [5].

Perhaps the most traditional way is ‘supervised learning’. You train the model with a data set of input and output data and then the model learns the relationship between the input and output data. Using this kind of model, you can predict the future.

Then there’s the unsupervised learning that is more like using different kind of statistical methods to find typical clusters in the data set. For example- it is possible to find abnormal energy consumption patterns from a group of buildings.

And then you have the reinforcement learning where you don’t need any kind of human to tech system. It learns by itself. Just by trial and error.

LHC: Yes, I read about this. Sometimes it’s called ‘rewards and punishment’. Can you give an example of a building (system) where you can better use machine learning and not just a PID controller?

HI: PID is really good in simple control like single input and single output but when you need a multi objective optimization, like we spoke earlier about that, or you have a more complex system or when you need predictions for the future, then you use machine learning.

LHC: I read somewhere that the problem with using machine learning is having the operational data to teach the algorithm.

HI: That is true. You need data to teach the algorithm but that can be data from the existing building (historical data), or you can create them with simulation. The supervised machine learning model learns itself from historical data. It can only do things that have been already happening previously. But if something new happens, it’s like, what do I do? But reinforcement learning, which learns from reward and punishment, can also adapt to the new situation.

LHC: When I supervise students’ projects at our university, I ask them what they think is more sustainable: high-tech building or low-tech building. What is your thought about this?

HI: Well, if I take example from my own life, having a second home from the 40s in the countryside with natural ventilation, it works quite nice during autumn and spring but during summer it’s too hot and during winter a bit cold. And if I compare this to a new building with good insulation and balanced mechanical ventilation, where I live in the city. I prefer when everything is automated in the background.

Sources

[1]     L. Remes, K. Dooley, J. Ketomäki, and H. Ihasalo, “Smart workplace solutions – can they deliver the offices that employees have been waiting for?,” Facilities, vol. 40, no. 15/16, pp. 40–53, Jan. 2022, doi: https://doi.org/10.1108/F-04-2021-0032.

[2]     K. A. Einola, L. Remes, and K. Dooley, “How can facilities management benefit from offices becoming more user-centred?,” Facilities, vol. 42, no. 15/16, pp. 17–29, Dec. 2024, doi: https://doi.org/10.1108/F-01-2023-0003.

[3]     T. Derek and J. Clements-Croome, “What do we mean by intelligent buildings?,” Automation in Construction, vol. 6, no. 5–6, pp. 395–400, 1997, doi: https://doi.org/10.1016/S0926-5805(97)00018-6.

[4]     “A History of Smart Buildings.” [Online]. Available: https://hereworks.io/tech-talk/a-timeline-in-the-history-of-smart-buildings/.

[5]     S. Sierla, H. Ihasalo, and V. Vyatkin, “A Review of Reinforcement Learning Applications to Control of Heating, Ventilation and Air Conditioning Systems,” Energies, vol. 15, no. 10, pp. 1–26, 2022, doi: https://doi.org/10.3390/en15103526.