Artificial intelligence to save energy

AIBUILD+ is another research project research has recently received funding from the regional government as part of its call for strategic research and development projects. It studies how to apply artificial intelligence and digital twins to optimize energy efficiency and comfort in buildings. Mabel Morales, a researcher at the Institute of data Science data Artificial Intelligence DATAI, presents the project explains the work being carried out. 

Q. What is the main goal this project?

AIBUILD+ goal innovative solutions to reduce energy consumption in buildings. These are achieved through the management of HVAC systems in buildings, ensuring thermal comfort for occupants and adequate indoor air quality.

These solutions must be capable of replacing traditional control systems, based on set schedules, with management and dynamic management capable of responding to changes in weather conditions and actual building usage.

The project also project the use of Artificial Intelligence (AI), particularly in the implementation of Reinforcement Learning algorithms applied to energy control in buildings.

Q. How is Artificial Intelligence used in this project?

A data integration platform has been developed data as the core of the system. This platform will enable the collection, storage, and structuring data the thermal sensation of buildings, management systems, and HVAC equipment, integrating heterogeneous information from different devices and formats. From this platform, the data will be pre-processed and tagged, as well as made available to digital and control models.

Digital twins of buildings and their HVAC systems database created from this database . These twins will combine two types of models: on the one hand, physical subject models, implemented in energy simulation tools widely used in architecture, such as DesignBuilder and EnergyPlus; and, on the other hand, models based on subject " data , such as artificial intelligence models, for example, recurrent neural networks.

These digital twins will enable the thermal and energy behavior of buildings to be reproduced and their performance to be simulated in different usage scenarios, under different outdoor conditions. Reinforcement learning algorithms will be developed and implemented on these models. This will facilitate integration with simulation tools, data processing data subsequent deployment in real buildings.

Reinforcement learning will be used to train intelligent controllers capable of optimally deciding temperature settings, ventilation, and operating schedules for air conditioning systems. These controllers will learn by interacting with digital twins and optimizing a reward function that balances energy consumption reduction with maintaining thermal comfort and indoor air quality. Once trained, the artificial intelligence models will be integrated into a hybrid cloud and edge computing architecture, from which they will run in real time in the pilot buildings.

Q. How is the DATAI Institute collaborating with the School of Architecture the development project?

Without the School of Architecture , we would School of Architecture understand the context: the physical properties of the building, how it behaves, the geometry, all the variables of the machines... At DATAI, we contribute the knowledge of statistical models and artificial intelligence.

In the case of the digital twins created by the School of Architecture, we will change the building settings to see how they behave. For example: what happens if I open all the windows in the building and it cools down to the maximum? How much do I save if I lower all the settings to the minimum? These things cannot be tested in a real context, so we test them in virtual reality.

We are a consortium made up of research agents research companies, such as the Navarre association , Saltoki, Satelec, and Vors Control. It is precisely these companies that make it possible for us to have all the data systems necessary to manage it. They provide the integration platforms that allow us to connect the multiple devices that are currently collecting information.

Q. What types of buildings do you work on?

We now have two prototypes on which we are going to develop and test the models, validate them, calibrate them, etc. These are the Amigos and School of Architecture buildings. However, broadly speaking, the idea behind project for these models to then be generalizable to any other non-residential subject .

We collect data temperature, humidity, CO₂, etc. from the virtual models of these buildings. In addition, the School has a weather station that provides us with all the data on outside temperature, such as rainfall, wind direction, and wind strength. This project all these factors in these two buildings so that it can be extended.

Q. This project at the end of 2025. What do you hope to have achieved by its completion in 2027?

At the end of project would like to have a program, a software, in which any Username interact: "What happens if I now lower the temperature in my office by one Degree? How much can I save with that?" 

We would also like it to be an automatic controller that the university maintenance team could use.