Energy Disaggregation from Single Point Measurements

This direction is merged into another direction: non-intrusive load monitoring. Its last version stays readable and citable here.

Version 1 of 144 papers52 conceptsfrozen 2026-10-04

To manage a modern power grid, operators need to know exactly what is happening inside a building. If a utility company wants to "nudge" a consumer to save energy or coordinate a "Virtual Power Plant" to stabilize the grid, they need to know if the air conditioner is running or if the dishwasher just started.

The obvious solution—installing a smart sensor on every single appliance—is a logistical nightmare. As Oliver Parson and Jason Koh point out, the cost of hardware, professional installation, and the invasive nature of per-device tracking make sub-metering impractical for large-scale deployment.

The research direction of Non-Intrusive Load Monitoring (NILM), or energy disaggregation, solves this by treating the building’s total power signal as a composite "riddle" to be solved. By analyzing the aggregate data from a single main meter, researchers can "unmix" the signal into its constituent appliance-level parts.

The Intuition of Electrical Signatures

Every appliance has a unique way of "breathing" electricity. When a toaster turns on, it creates a sharp, resistive spike. When a refrigerator compressor starts, it creates a complex inductive "transient" as the motor spins up.

Lillian J. Ratliff and Henrik Ohlsson frame this as a dynamical systems problem. Instead of looking for static on/off signatures, they model appliances using state-space equations. By treating the total power signal as the output of multiple hidden dynamical systems, they can use adaptive filtering to track switching events in real-time. The goal is to find the simplest set of appliance inputs that mathematically explain the total observed power.

Probabilistic Unmixing

If we view a house as a collection of states (e.g., "Fridge Defrosting" or "TV Standby"), the problem becomes one of sequence estimation. Alex Rogers and Sarvapali D. Ramchurn have shown that Hidden Markov Models (HMMs) are particularly effective here.

In this framing, each appliance is a hidden state that emits a certain amount of power. By observing the total power, a Factorial HMM can estimate which combination of appliances is most likely active. Arnaud Cadas extended this to real-time demand dispatch, using particle filters to track appliance states so grid operators can coordinate loads without needing individual sensors.

Scaling with Low Resolution Data

While high-frequency data (sampling thousands of times per second) provides the clearest signatures, it is computationally expensive. A major breakthrough in the field is the ability to disaggregate energy using much coarser data.

Hongning Wang and Nipun Batra 0001 demonstrated that matrix factorization can break down monthly utility bills into appliance-level estimates by comparing a home's characteristics against a wider population. Similarly, Guoming Tang developed a "Plug and Play" model that uses only basic rated power data from user manuals—rather than complex training—to identify appliances, making the technology far easier to scale to millions of homes.

From Waveforms to Computer Vision

The most recent shift in NILM research involves changing the representation of the data entirely. Researchers like Yiwen Xu and Luyao Ma have begun transforming one-dimensional electrical current signals into two-dimensional color images or "date-time matrices."

By treating energy disaggregation as an image segmentation problem, they can use powerful deep learning architectures like U-Nets. This allows the model to "see" patterns in the power flow that are difficult to capture in time-series models, such as when a high-power appliance "masks" the signature of a smaller one.

Why Disaggregation Matters

The ultimate goal of this research is not just to know what is running, but to act on it.

  • Behavioral Nudges: Roy Dong and S. Shankar Sastry use disaggregated data to design personalized incentives, helping utilities "learn" consumer preferences and encourage efficiency.
  • Grid Stability: Fengji Luo envisions buildings as "Virtual Power Plants" where disaggregated insights allow decentralized systems to trade energy and stabilize the grid.
  • Fault Detection: In industrial settings, Oliver Niggemann uses these techniques to localize faults in complex machinery using only a single global sensor.

By turning a single point of measurement into a window into the entire system, energy disaggregation provides the granularity of a thousand sensors with the infrastructure of just one.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_energy_disaggregation_from_single_point__88ecaaa9,
  title = {Energy Disaggregation from Single Point Measurements},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/energy-disaggregation-from-single-point-measurements/rdr_88ecaaa9ea48475698ad2976ea58ee72},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_energy_disaggregation_from_single_point__88ecaaa9_v1,
  title = {Energy Disaggregation from Single Point Measurements},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/energy-disaggregation-from-single-point-measurements/rdr_88ecaaa9ea48475698ad2976ea58ee72/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}