*4.3. Device Categorization*

Let us next introduce the device categorization used during our evaluation studies. A complete overview of the appliances under consideration and their corresponding classes according to the three categorization approaches are given in Table 4. Categorizing appliances by their classes allows us to not only run analyses on the entire dataset, but also on subsets of the data in which all devices share a commonality (e.g., an inductive load curve). For example, the device of type mini oven is a member of the Single-State devices class, belongs to the large consumers (classified by its power consumption), and is an ohmic appliance (due to the load type of its heating rods).

#### 4.3.1. Categorization by Number of Operational States

Classifications by an appliance's number of operational states can be found in works considering load monitoring [2,34,35]. The typical device classes listed in the context of Non-Intrusive Load Monitoring (NILM), based on the complexity of their consumption patterns, are given as follows.


We included this categorization for two reasons: First, different works in load monitoring have reported that Single-State appliances are often easier to disaggregate [2]. Second, we expect greater differences in the importance of ambient features when appliances exhibit multiple operational states.

#### 4.3.2. Categorization by Power and Energy Consumption

We wish to note at this point that NILM is not the only use case for energy data analysis. For example, applications like demand side managemen<sup>t</sup> favor the availability of loads with a large energy consumption. As large energy consumers typically emit more excess heat and generate stronger magnetic fields, differences in the feature importance are likely to occur. Finding the optimal feature subset for such application cases is thus a prerequisite for the realization of such services. Hence, we classify the appliances under consideration by their power consumption, which we define as follows:


#### 4.3.3. Categorization by Load Type

Finally, we consider device classification according to their load type, according to the classification scheme proposed in [34]. The load type is determined by the phase shift between mains voltage and an appliance's current consumption as well as the presence of non-linear loads within devices. To allow to determine the load type for the devices in our evaluation, we followed the definition and examples given in [34], and evaluated the devices' data sheets.


• Switched-Mode Power Supplies (SMPS): Devices with a large amount of harmonic content, whose characteristics may moreover vary depending on the currently exhibited power demand.

It needs to be noted that we excluded the "composite load" class proposed in [34], as for our test only the washing machine and dryer could have fit the corresponding definition. Furthermore, the proposed "Capacitive Consumers" presented in [34] do not have a representation in our set of devices, as they are not typically found in residential spaces.
