A study from the University of Maryland suggests that deep learning, a sophisticated form of artificial intelligence (AI), could significantly reduce energy consumption in future heating, ventilation, and air conditioning (HVAC) systems. The research, conducted at UMD’s Center for Environmental Energy Engineering (CEEE), focused on the impact of AI on predicting power use in a variable refrigerant flow (VRF) system. This particular HVAC technology, featuring an outdoor and several indoor units, was tested in Glenn L. Martin Hall. Their findings will be published online in the January 2025 issue of the International Journal of Refrigeration. HVAC systems account for about half of a building's electricity use, and optimizing a VRF system's control hinges upon accurate power consumption predictions. The UMD researchers evaluated two AI models: the traditional Artificial Neural Network (ANN) and the newer, data-intensive Long-Short-Term Memory (LSTM) model.

Both analyze data to recognize patterns and make predictions, but LSTM requires more data. Unexpectedly, the LSTM model not only predicted power consumption more accurately but also demanded less computing power and memory than ANN. According to Po-Ching Hsu, a mechanical engineering graduate student and lead author, the ANN model attempts to boost accuracy by constructing a more complex model during the optimization process but still falls short compared to the LSTM. Co-authors include Lei Gao Ph. D. ’22, now at Oak Ridge National Laboratory, and Yunho Hwang, a mechanical engineering research professor and CEEE Co-director. Although the LSTM model's application in HVAC technology is still being explored, this study indicates it might significantly enhance energy efficiency. The LSTM model was developed using a year's worth of data. Hsu added, “The challenge now is to determine if we can achieve the same accuracy with just a few days or weeks of data. ”