Natural gas furnaces burn gas, but the components that make them run like igniters, blowers, control boards are all powered by electricity. At a high enough resolution of electricity monitoring those components are distinguishable, and using Pecan Street’s one-second resolution residential electricity data we built an algorithm that identifies gas furnace cycles directly from the electricity data with no gas meter input. At this resolution the data reveals not just that a gas furnace is running, but what type of furnace is installed in the home. Individual furnace components have unique power draws that can be isolated for identification. This post walks through how that is possible and what it implies for the kinds of residential energy research Pecan Street’s data is built to support.

What a natural gas furnace startup sequence looks like

A typical gas-furnace ignition sequence takes a little over a minute and proceeds in four phases:

  1. The draft-inducer motor energizes first. It clears the heat exchanger of residual combustion products, and a pressure switch confirms airflow before the sequence continues.
  2. The hot-surface igniter receives line voltage and heats to roughly 1,800°F. The gas valve opens, the flame is proved, and the igniter de-energizes.
  3. A short stabilization period follows while the heat exchanger warms toward fan-on temperature.
  4. The main blower engages.

At one-second resolution, each of these phases shows up as a distinct power level. The trace in Figure 1 walks through the full sequence on a furnace startup cycle from a Pecan Street home. When the algorithm detects this power draw sequence we know the furnace is starting to burn natural gas.

Figure 1. A gas furnace startup at one-second resolution. The inducer prepurge holds a flat low band for about fifteen seconds. The igniter warm-up steps up to a higher band for another fifteen. A brief stabilization period follows, and the main blower engages with a brief inrush spike before settling into steady operation. The full sequence takes roughly 90 seconds.

Different components, different signatures

Since the components activate sequentially in the furnace startup we can compare them across furnaces. The hot-surface igniter is the most diagnostic component of the sequence. Two generations of igniters are common in modern residential furnaces:

  • Silicon-carbide (SiC) igniters were standard through the mid-2010s.
  • Silicon-nitride (SiN) igniters dominate new installations today.

Both glow red-hot to ignite the gas-air mixture at the burners, but they draw very different amounts of power doing it. Figure 2 shows two gas furnace startups from two different Pecan Street homes, plotted on the same y-axis. A silicon-carbide igniter pulls 200-500 watts during warm-up. A modern silicon-nitride element pulls 40-150 watts. This power draw difference allows us to classify the igniter generation directly from the startup trace.

Figure 2. Two gas furnace startups, sampled at one-second resolution. On the left, a silicon-carbide igniter holds its warm-up band at about 320W. On the right, a silicon-nitride igniter holds its warm-up band at about 80W. They share the same function but the older platform draws four times the power.

The igniter is not the only component you can fingerprint this way, the blower motor also leaves a distinct signature. The furnace on the left of Figure 2 shows a sharp inrush spike the moment the blower engages. That’s characteristic of a permanent-split-capacitor (PSC) induction motor, which briefly draws locked-rotor current before settling to its steady running point. The furnace on the right ramps gently from its warm-up plateau to a higher steady level with no inrush at all. That’s the signature of an electronically commutated motor (ECM) or variable drive motor, whose onboard electronics smooth the start.

Figure 3. Blower startup traces from two gas furnaces, both equipped with silicon-nitride igniters. The PSC induction motor on the left spikes to about 1,265W before settling near 616W with the locked-rotor current of an induction motor coming up to speed. The ECM motor on the right ramps smoothly to roughly 515W with no inrush because its startup is managed by onboard drive electronics.

Identifying furnace activity at scale

The startup traces above show what’s possible with our data, and the same analysis applied across hundreds of homes unlocks almost a decade of natural gas furnace usage in our participant homes. For each detected furnace cycle, the furnace runtime multiplied by burner capacity gives an estimate of natural gas consumption. Figure 4 shows one heating season of detected cycles for a representative home in Central Texas, with an estimate for consumption based on the igniter age and blower motor.

Figure 4. One full heating season of detected gas-furnace activity for a representative home in Central Texas. Each bar is the number of furnace starts on that day. The pink line is daily mean temperature. The header reports the season totals: detected events, total furnace runtime, and the gas-usage estimate derived from runtime and burner capacity.

Why the data resolution matters

Most utility advanced metering infrastructure records consumption at 15-minute or hourly intervals. At that frequency the entire ignition sequence collapses into a single reading which is indistinguishable from most residential loads. The same principle applies to anything else cycling in a residential electrical system. A refrigerator compressor starts with its own inrush signature. A dishwasher’s heating element switches in and out on a duty cycle that traces the wash program. An EV charger negotiates current with the vehicle in steps that reveal the onboard charger’s design. None of these are visible at meter level resolution. This furnace example is one of many in our dataset.

Waveform data unlocks even more capabilities

One-second power data is enough to fingerprint a gas furnace because the startup sequence is an orderly cascade of events, but not every load is that neatly defined. Modern electronics switch their current on and off many times per second, and at one-second resolution those loads look like a smooth average. Our new waveform dataset makes these timescales visible. Instead of one power sample per second, we record voltage and current at kilohertz rates, capturing the actual shape of the current waveform on every cycle of the line. Variable-frequency drives, inverter-based heat pumps, EV chargers, LED lighting, and electric water heaters all become individually identifiable from the harmonic signatures they impose on the line. The fingerprinting technique developed here for gas furnaces generalizes to almost every major load in a home once the data is fast enough to resolve the underlying switching.

High-resolution data also opens the door to spotting equipment problems before they become service calls. For our furnace example, igniter elements oxidize and crack with age, bearing wear in blower motors pushes steady-state running power up over time, and a partially blocked flue can alter the inducer’s prepurge wattage. Our data can pick up a struggling component before the homeowner notices anything wrong at startup. The same data supports a wider set of applications such as non-intrusive load monitoring research, degradation studies across appliance generations, demand-response modeling at the device level, and product development for manufacturers who want to know how their equipment actually behaves in the field rather than on the test bench.

Pecan Street has been collecting residential electricity data from instrumented homes since 2011. The Dataport platform makes this data available to researchers, utilities, and energy-modeling teams.

Request Dataport access at Dataport

See pricing at Commercial Pricing Guide