Making buildings energy-smart using artificial intelligence

Today's buildings are becoming increasingly more complex, particularly those that produce and store their own energy. Such buildings are equipped with systems that can communicate with each other, producing vast amounts of data.

"In such systems, it is challenging to optimise performance," says Sven Opalic.

He recently defended his doctoral thesis at the University of Agder (UiA), ‘Advanced Warehouse Energy Storage System Control Using Deep Supervised and Reinforcement Learning’.

Complex technical rules

Opalic works as an energy technology and innovation specialist at Relog. One of his main tasks has been to optimise energy consumption in a large warehouse building located in Vagle Business Park in Sandnes, Norway.

Early on, he realised the extent of the challenge. With a size of nearly 30,000 square metres, the building serves as a refrigerated warehouse for Rema Distribution. It generates electricity through rooftop solar panels and stores energy in a battery bank equivalent to five Tesla cars.

In addition, the building features a 300,000-litre water pool that serves as part of the sprinkler system but can also be used to store heating and cooling energy.

"When preparing the building for operation in 2018, we had eight full pages of technical operating rules. It wasn't manageable. We could have simplified it, but that would have compromised efficiency," he says.

Opalic contacted Professor Morten Goodwin at the Centre for Artificial Intelligence Research (CAIR) at UiA. This led to the doctoral thesis based on the energy system in the warehouse building in Vagle and how artificial intelligence can be used to optimise it.

Creating a simulation 

"Each building is unique, so introducing energy management systems requires significant expertise and time," says Opalic.

To be able to implement the technology in all kinds of buildings, Opalic's system employs reinforcement learning, in which artificial intelligence experiments and improves by learning from its mistakes.

Having a software program experiment and fail in a real building would be unfortunate. Therefore, a model is first created that simulates the energy systems of the building.

Once the system demonstrates satisfactory performance in the simulation, it can be applied to the actual building’s systems.

Here you can see how the system works in practice

"Inspiring Collaboration" 

Synnøve Gautesen Berg is the director of responsibility and sustainability at Rema Distribution. The company delivers goods to all supermarkets in the Rema1000 chain in Norway and is an energy-intensive business that operates many refrigerated and frozen warehouses.

"Having forward-looking property owners like our sister company Relog is extremely valuable. Having access to the kind of expertise that Opalic has acquired is crucial to achieving innovative and effective solutions," says Berg.

She adds that smart energy management is essential in realising the sustainability goals set by Rema1000.

"This is a very inspiring collaboration that is taking us in the right direction," she says.

May deliver significant savings

Opalic believes that automation of this kind can lead to significant savings, although the extent depends on the specific building.

"If a skilled programmer has designed the system in a new building, there may be fewer gains. However, in inefficient buildings that haven't undergone much optimisation, savings of 30 to 50 percent are possible," he says.

The research has yielded positive results, and parts of the results have already been implemented at the Rema1000 warehouse outside Sandnes. Opalic now aims to facilitate the development of software that can be utilised across multiple buildings.

"Our goal has always been to eliminate bottlenecks. Now, our goal is to test the software on our own buildings to see if our excellent results still hold up in the face of reality," he says.