## 2.3 Case study

The use of the calendar display is illustrated on smart meter energy usage from four households in Melbourne, Australia. Individuals can download their own data from the energy supplier, and the data contained in the paper is made available from four colleagues of the authors. The calendar display is useful to help people understand their energy use. The data contains half-hourly electricity consumption in 2017 and 2018. The analysis begins by looking at the distribution over days of week, then time of day split by work days and non-work days, followed by the calendar display to inspect the daily schedules.

Figure 2.13 shows the energy use across days of week in the form of letter value plots (Hofmann, Wickham, and Kafadar 2017). Letter value plots are a variant of boxplots for large data, with other quantiles represented by boxes. Letters indicate the fraction of the data divisions, for example, F indicates fourths or quartiles, and the two outer ends of the box are the 25th and 75th percentile, the same traditional ends of the box as a boxplot. The letter E indicates eighths, with box ends being 12.5th and 87.5th percentiles of the data. These additional boxes replace the whiskers in a traditional boxplot. The letter value plots for the households, show a line indicating the median (M) and the innermost boxes corresponding to the fourth (F) and the eighth (E) quantile divisions. Inspecting the medians across households tells us that household 3, a family size of one couple and two kids, uses more energy over the week days, than other households. The relatively larger boxes for household 2 indicates greater variability in daily energy consumption with noticeable variations on Thursdays, and much higher usage over the weekends. The other two households (1 and 4) tend to consume more energy with more variation on the weekends relative to the week days, reflecting of work and leisure patterns.

Figure 2.14 shows energy consumption against time of day, separately by week day and weekend. Household 1 is an early bird, starting their day before 6 and going back home around 18 on week days. They switch air conditioning or heating on when they get home from work and keep it operating until mid-night, learned from the small horizontal cluster of points around 0.8 kWh. On the other hand, the stripes above 1 kWh for household 2 indicates that perhaps air conditioning or heating runs continuously for some periods, consuming the twice the energy as household 1. A third peak occurs around 15 for household 3 only, likely when the kids are home from school. They also have a consistent energy pattern between week days and weekends. As for household 4, their home routine starts after 18 on week days. Figures 2.13 and 2.14, part of a traditional graphical toolkit, are useful for summarizing overall deviations across days and households.

Figure 2.15-2.18 display the data in calendar layout individually for each household, unfolding their day-to-day life. Glancing over household 1, we can see that their overall energy use is low. Their week day energy use is distinguishable from their weekends, indicating a working household. The air conditioner appears to be used in the summer months (Decemberâ€“February) in the evening and weekends. In household 2, heating keeps functioning for consecutive hours, which is evident in the mid July. In contrast, household 1 uses heating cautiously. These observations help to explain the stripes and clusters in Figure 2.14. The calendar plots speak the stories about vacation time that are untold by previous plots. Household 1 is on vacation over three weeks of June, and household 2 was also away for vacation in late December and in the second week of June. Figure 2.17 shows household 3 takes two one-month-long family trips in September until early October and in June/July. Household 4 is away over two or three weeks in early October, December, early April, and late June. The use of air conditioning and heating leaves no trace in these two households.