AI Face-off: Researchers Pit Old AI Tech vs. New in HVAC Efficiency Test

A sophisticated type of artificial intelligence (AI) known as deep learning could play an important role in reducing energy usage in the next generation of heating, ventilation and air conditioning (HVAC) systems, a new University of Maryland study found.

Researchers in UMD’s Center for Environmental Energy Engineering (CEEE) explored AI’s impact on predicting power consumption in a variable refrigerant flow (VRF) system—a type of complex HVAC technology that has an outdoor unit and multiple indoor units—located in Glenn L. Martin Hall. They presented their findings online for the January 2025 issue of the International Journal of Refrigeration.

HVAC accounts for around half of a building’s electricity consumption, and optimized control of a VRF system requires accurate prediction of power consumption. The researchers compared the prediction ability of two types of AI models: a traditional machine learning model known as Artificial Neural Network (ANN) and a newer deep learning model called Long-Short-Term Memory (LSTM). Both models use data to recognize patterns and produce insights and predictions, but LSTM requires more data.

As expected, the team found that the more up-to-date, more data-intensive LSTM model more accurately predicted power consumption. The big surprise, though, was that the LSTM model appears to require less computing power and memory than the ANN model.

The ANN tries to improve its accuracy during the optimization process by building an increasingly complex model to predict the power consumption, said lead author mechanical engineering graduate student Po-Ching Hsu. “But even with that, it still cannot achieve the same performance as LSTM.”

The paper’s other co-authors are former CEEE researcher Lei Gao Ph.D. ’22, now on the R&D staff at Oak Ridge National Laboratory, and CEEE Co-director Yunho Hwang, a mechanical engineering research professor.

While LSTM is still in the exploratory stages for use in HVAC technology, the UMD study indicates that it could become a powerful tool for improving energy efficiency. In this study, the LSTM model relied on a year’s data for development. The challenge is: “Is there a way we can do this with data from a few days or a few weeks and still make very good predictions?” Hsu said.