We choose household #2 electricity data for further study because of the diversity of appliances it has. The details of household #2 electricity data are as follows (Note: For details of data description, please consult 02_doc.txt within the data set):

• Household 2 smart meter data (Eco_DataSet/house#02/02_sm_csv)

Measurement period: 01.06.12 ~ 31.01.13. Each rows consist of the following comma separated values every second:

1. powerallphases: Sum of real power over all phases
2. powerl1: Real power phase 1
3. powerl2: Real power phase 2
4. powerl3: Real power phase 3
5. currentneutral: Neutral current
6. currentl1: Current phase 1
7. currentl2: Current phase 2
8. currentl3: Current phase 3
9. voltagel1: Voltage phase 1
10. voltagel2: Voltage phase 2
11. voltagel3: Voltage phase 3
12. phaseanglevoltagel2l1: Phase shift between voltage on phase 2 and 1
13. phaseanglevoltagel3l1: Phase shift between voltage on phase 3 and 1
14. phaseanglecurrentvoltagel1: Phase shift between current/voltage on phase 1
15. phaseanglecurrentvoltagel2: Phase shift between current/voltage on phase 2
16. phaseanglecurrentvoltagel3: Phase shift between current/voltage on phase 3

• Household 2 plug data (Eco_DataSet/house#02/02_plugs_csv)

Measurement period: 01.06.12 ~ 31.01.13. Each file within the above folder denotes one specific appliance:

1. Tablet
2. Dishwasher
3. Air exhaust
4. Fridge
5. Entertainment
6. Freezer
7. Kettle
8. Lamp
9. Laptops
10. Stove (no. days: 28, which is insufficient for prediction)
11. TV
12. Stereo

• Household 2 occupancy data (Eco_DataSet/house#02/02_occupancy_csv)

The occupancy data is available for summer (02_summer.csv) and winter (02_winter.csv). Each row in the csv file represents a day (specified by the first column). Each day consists of 86400 fields indicating presence (1) and absence (0) for each second of the day. Thus the csv file contains 86401 columns.

Visualization of appliance level power consumption on 03-Jun-2012 is shown below.

1. The first graph shows the total power consumption of tablet, dishwasher, air exhaust, fridge, entertainment, freezer as well as ground truth occupancy state. Note that the tablet is used for monitoring when people enter or leave the room, and is not considered appliance, so we will ignore it.
2. The second graph shows the power consumption of kettle, lamp, laptops, stove, TV, stereo as well as ground truth occupancy state. Note that TV and stereo combined is considered as entertainment in the first graph, so we will ignore these two. (The first two graphs are generated by Eco_DataSet/codes/graph_house2_plugs.py by changing index number at around line 50.)
3. The third graph shows the total power, power phase 1, power phase 2, power phase 3 as well as ground truth occupancy state. (The third graph is generated by Eco_DataSet/codes/graph_house2_sm.py by changing index number at around line 47.)

As shown in the graphs above, residential power loads can be classified as background loads and foreground loads.

• The background loads runs in the background without active user intervention. e.g. Fridge is controlled by an internal thermostat that periodically turns on the compressor. Freezer has similar characteristics as fridge.
• The foreground loads are ones that are actively controlled by a user. The usage depends on the type of the load, time-of-day, day-of-week and seasonal patterns associated with them. e.g. Lamps are more likely to be turned on at night.