In This Story
In homes and buildings across the country, electricity flows through a single-entry point before branching out to power everything inside. A refrigerator hums to life; a laptop charges, lights switch on, and a dishwasher runs its cycle. Yet the electric meter at the side of the building records only one number: the total amount of power consumed.
For Anomadarshi Barua, an assistant professor in the Department of Electrical and Computer Engineering at George Mason University, that single number represents an opportunity to learnmuch more. He recently received a grant of $840,000 from the Office of Naval Research to develop his work in this field.
Barua is developing techniques that can break down aggregated electrical data and reveal which individual devices are using power. The approach, known as non-intrusive load disaggregation, analyzes the electrical signal entering a building and separates it into the distinct signatures created by different appliances and electronic systems.
Barua’s research focuses on improving the accuracy of this process using generative AI models that capture subtle electrical details hidden in power data. When devices operate, they produce distinctive electrical patterns, including harmonic signals that appear at higher frequencies. These patterns can act as identifying features, allowing researchers to determine which appliances are active at any given moment.
But the details are often difficult to observe directly. Standard power meters record electrical signals at relatively low sampling rates, meaning some high-frequency information is lost. Barua and colleagues are developing algorithms that reconstruct the missing details from the available data. “We are trying to recover the high-frequency information that is not captured by the meter,” he said. “That information helps create the fingerprint of each device.”
The approach draws on concepts from chaos theory and signal processing. By studying patterns in the lower-frequency electrical data, the model can estimate the higher-frequency harmonics that would normally accompany them. Once those features are reconstructed, it becomes easier to distinguish between different types of electrical loads.
Even identical appliances can produce slightly different signatures depending on the environment in which they operate. Variations in wiring length, electrical resistance, and circuit configuration can alter the behavior of the signal. These subtle differences make the disaggregation problem particularly challenging. “Two houses might have the same model of refrigerator, but the electrical environment is different,” Barua said. “Those differences affect how the signal appears at the main power meter.”
The research has potential applications in both residential and commercial settings. For consumers, more detailed energy information could help identify which devices account for the largest share of electricity use. That knowledge could encourage more efficient behavior, such as reducing the use of energy-intensive appliances or replacing older equipment.
The technology could also be applied in more complex electrical environments, such as data centers, where numerous devices operate simultaneously and generate intricate electrical patterns.
Barua and PhD students Tarikul Tamiti and Sajid Fardin Dipto, along with several undergraduate students, published several papers exploring related techniques since he joined the university in 2023, laying the foundation for the current project. “We have been working on these ideas for the last year and a half,” he said. “The publications helped demonstrate the progress of our algorithms and supported the proposal.”