*3.2. Data Statistics*

Overall, we collected 53,328 h of raw current and voltage waveforms using the proposed framework. Figure 7 highlights the richness of the captured data for the SmartMeter and the PowerMeter units. Figure 8 shows the apparent power extracted for each individual appliance. It further emphasizes the contribution of each appliance to the total power consumption on this day.

According to [33], the average consumption of a comparable three person household in Germany is 7.12 kW h per day. Evaluating the SmartMeter data of the FIRED dataset results in an average electricity consumption of 6.06 kW h per day, which is slightly lower compared to the average, but this is expected as the data does not contain the electricity consumed by the washing machine, dryer, and freezer.

Figure 12a shows the consumption of six appliances at the hour of day averaged over the whole recording duration. This delivers a good indication for usage patterns. For example, the 'espresso machine' shows two distinct peaks, one in the morning at around 9:00 a.m. (morning coffee) and one in the afternoon at 3:00 p.m. (coffee break). In comparison, the 'router' does not show any significant peak. It can also be seen that the 'office PC' has a high standby consumption of around 35 W and is used mainly between 9:00 a.m. and 6:00 p.m. Figure 12b shows the distribution of power demand for the same appliances. Some state information can already be derived from these plots. The 'hairdryer' shows two distinct states representing two different temperature settings. The 'office PC' shows three peaks. The peak around 35 W represents the already mentioned standby consumption, the peak around 50 W represents the PC in its *On* state, and the 140 W peak includes the *On* state of the 27-inch monitor connected to the same meter. The 'espresso machine' consumes a huge amount of power (1200 W) during its heating cycles but is mostly idle (5 W) in between.

**Figure 12.** Appliance usage over the complete recording duration. (**a**) shows the daily usage patterns of the devices as the average power for each hour-of-day; (**b**) shows the histogram of the power demands. A 2 W threshold was set to omit data in which no power is drawn.

## *3.3. Technical Validation*

Measurements of the FIRED dataset are provided without applying any pre-processing or filtering rather than the conversion to physical quantities. The recording framework is equipped with different mechanisms to cope with real world effects such as network dropouts or clock drifts, and the data integrity was analyzed in detail.
