ASHRAE Journal - May 2021 - 36

TECHNICAL FEATURE

FIGURE 5 Longest expected workday for the 37 occupants, calculated by subtracting a zone's earliest expected arrival time from its latest expected departure
time.

Fraction of Occupants

0.25

Fraction of Absent Days

µ=11.5 h

0.2

FIGURE 6 The fraction of absent workdays for the 37 occupants.

14 h Occupancy for the
Whole-Building from
6 a.m. to 8 p.m.

σ =2.4 h

0.15
Longest Workday
Distribution for
Each Occupant

0.1

0.25

5
10
15
Longest Workday for Each Occupant (h)

A less conventional but likely more effective zone-level
OCC metric to exploit is latest expected arrival times.
This metric highlights absent workdays, which are
becoming increasingly common, particularly for whitecollar workers, due to working from home, work-related
travel, sick days, meetings outside the office and so on.
In our small sample of 37, occupants on average spent
one in every four days away from their offices. Because
the data were collected well before the COVID-19 lockdown, we can assume that the fraction of absent workdays will be higher after the restrictions are lifted.
Figure 6 presents the distribution, ranging from 10% to
90%, of absent workdays. For this population, a simple
control logic that reinstates temperature setback when
an arrival has yet to occur by the latest expected arrival
time can increase the average weekday temperature
setback period by 3.3 hours. Assuming 250 workdays
in a year, use of zone-level latest expected arrival times
can increase zone-level setback periods by another
825 hours (9% of the full year). Previous research demonstrated that zone-level OCC metrics, when used to
implement a ±3°C (±5°F) setback from a 22°C (72°C)
default setpoint, can deliver a 27% reduction in HVAC
energy use in Ottawa, Canada.16 Temperature setbacks
tend to be more effective in moderate climates; they
account for a larger fraction of the environmental loads.
Because we collected the occupancy data from singleoccupant offices, we cannot demonstrate the zone-level
OCC metric for the highest expected number of occupants in a zone.
We will now examine the OCC metric preferred thermal conditions using thermostat use data (as a proxy
ASHRAE JOURNAL

ashrae.org

M AY 2021

Ratio of Thermostat Overrides

Setpoint Decrease
0

0.75

FIGURE 7 Distribution of indoor temperatures at thermostat setpoint decrease and
increase instances.

0.05

0.5

Setpoint Increase

0.4
0.3
0.2

µ=21.3°C

µ=22.6°C

σ =1.2°C

σ =1.4°C

0.1
0

15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Indoor Temperature (°C)

for preferences) and concurrent indoor temperature
data. Because we did not have many thermostat overrides for each occupant, we could not examine preferred
conditions at the zone level (recall that there are emerging mobile and web applications to expedite occupant
preference data collection). There were 231 setpoint
decreases and 281 setpoint increases over one year by
the 37 occupants. (Note that hot/cold complaints can be
treated as a similar source of information about preferences at the building level.)
Figure 7 presents the distribution of indoor temperatures at the thermostat overrides, revealing the mean
indoor temperatures at setpoint increase (21.3°C
[70.3°F]) and decrease (22.6°C [72.7°F]). The preference
learning process should rely on the frequency of overrides at different indoor temperatures (or other IEQ
indicators not considered here for simplicity) rather
than on the absolute number of overrides. Given 20
occupied time steps spent below 18°C (64°F), if 15 of
them contain a setpoint increase event, then the probability of a setpoint increase below 18°C (64°F) is 0.75.
Logistic regression is the most common approach
to modeling the frequency of occupant actions such
as thermostat use.1 Figure 8 presents logistic regression models for setpoint increase and decrease events
as well as their superimposition, demonstrating that

http://www.ashrae.org

ASHRAE Journal - May 2021

Table of Contents for the Digital Edition of ASHRAE Journal - May 2021