Machine Learning Could Help Buildings Manage Peak Energy Demand

Machine learning could help commercial buildings anticipate heating and cooling demand, allowing operators to adjust equipment before electricity use peaks, according to PhD researcher Guaraang Malik.

In an interview with Energi Media publisher Markham Hislop, Malik described research that starts with simulated building data and aims eventually to inform energy management in actual buildings. His project is still at an early stage, with no measured energy savings presented in the interview.

The approach uses simulations to represent building operations, schedules and weather conditions. Machine learning algorithms then use those data to predict energy consumption, including how demand changes from hour to hour.

Once a model predicts consumption accurately enough, Malik hopes to apply it to buildings equipped with automation systems. He is evaluating algorithms using simulated data and seeking access to buildings with BACnet systems.

Anticipating the Afternoon Peak

Malik illustrated this with a building expecting a cooling peak at 4 p.m. If the system knows some rooms are empty beforehand, it could begin cooling those spaces earlier, reducing the cooling demand that coincides with the peak.

“So in that way, you could reduce the peaks,” he said.

The example points to a possible application of predictive energy management: changing when equipment operates while considering the needs of occupants. It does not establish how much electricity a particular building would save or whether total consumption would fall.

BACnet, which Malik identified as Building Automation and Control Network, provides a protocol for sharing building information. In his proposed approach, access to information about rooms could help a model forecast demand and guide building controls.

Hislop discussed the potential relevance for warehouses, workshops and other commercial facilities, where managing energy use must also account for worker comfort and building operations.

AI Inside the Building

Malik also outlined a longer term idea: placing computing capacity inside a building to run AI locally. A server could support building management and potentially other AI applications, reducing the need to send requests to cloud services.

Keeping processing on site could give operators more control over where information goes. However, local processing should not be confused with a guarantee of complete privacy or security; those outcomes depend on how the system is configured and managed.

Malik suggested distributed computing might also reduce the energy impact associated with cloud servers and their cooling requirements. He distinguished that possibility from a result.

“It’s my hypothesis,” he said.

Research Still Needs Real Building Data

At the time of the interview, Malik said he was six months into his PhD and focused on choosing a machine learning algorithm for predicting building energy use.

The next step is access to operational building data. That would allow the research to move beyond simulations and examine how the approach performs under real conditions.

For building operators, the interview raises questions about forecasting demand, shifting cooling loads and managing data locally. The potential is to make energy management more responsive. Whether that translates into lower costs or reduced energy consumption remains a question for testing.