The Weather Impact on Efﬁcient Home Heating
with Smart Thermostats
British Columbia Institute of Technology, Department of Computing, Vancouver, Canada
Abstract—In this paper, we compare the currently imple-
mented smart heating algorithms with our proposed method -
Weather Based Heating Algorithm. Our simulations are based on
the data from years 2015-2018, provided by Ecobee Inc., which
is one of the major manufacturers of smart thermostats. The
main problem that we focus on is the study of the overheated
homes. Our results showed that the problem of overheated
households during winter affects North America. The signiﬁcant
impact is visible during sunny days. On the other hand, the
implementation of adaptive strategies, such as Weather Based
Heating algorithm, results in three times reduction of overheated
time and 20% reduction in energy costs.
Keywords—energy, environment, home heating
Over the last few years, we have seen an increased interest
in the Internet of Things solutions making their way into
residential homes. Nowadays, smart devices are inevitable in
our lives , . Smart thermostats are one of them. In contrast
to regular thermostats, the smart ones allow to control the
climate at home remotely, show the energy consumption in
real-time, and can even adjust themselves based on ambient
conditions like humidity or temperature outside –. These
are the crucial features as residential heating uses more energy
than any other home expenditure including air conditioning,
humidiﬁcation, and water heating . It is essential to accom-
modate efﬁcient techniques to increase the efﬁciency of HVAC
systems and decrease energy costs. Moreover, studies have
shown that more than half of the installed smart thermostats
are used in manual modes, due to the inefﬁciency of automatic
In this paper, we introduce a weather-based algorithm to
increase the efﬁciency of the heating using smart thermostats.
The standard algorithms work using very simple if −else
statements. More advanced ones, use historical heating and
cooling run-times to control the HVAC system. Our proposed
algorithm uses an innovative approach that takes into an
account live weather forecasts to compute the impact on indoor
Several works have studied the problem for automatic
thermostat control. The majority of them were focusing on
the occupancy prediction. The authors of  introduced
occupancy prediction algorithms based on GPS-enabled smart-
phones. In , the thermostat control system is based on the
mobility prediction of users, using contextual information ob-
tained by mobile phones. Other papers – are using the
data analytics techniques to predict user behavior in terms of
home occupancy. Some authors are also focusing on proposing
energy scheduling methods, with both, the smart pricing tariffs
and the user comfort implication –. While occupancy
is an important factor to enable smart thermostat control, the
latest studies shown that people spend more than 75% of time
indoors , which makes it less signiﬁcant for the efﬁcient
home heating then the weather impact.
The remainder of the paper is organized as follows. In the
next Section, we introduce an optimization problem. Next, we
present the algorithms design. Finally, we provide simulation
setup and results discussion, followed by a conclusion.
II. OP TI MI ZATI ON P ROB LE M
Datasets used in the simulations were provided by Ecobee
Inc., Canada. The company is the manufacturer of smart ther-
mostat devices. They established a research program named
Donate Your Data . The data is shared from the home
devices in North America, from the years 2015-2018 and
consists of the following metadata ﬁelds: location; number of
occupants; heat/cool temperature settings; occupancy sched-
ules; indoor/outdoor temperature/humidity, HVAC runtimes.
Data from each thermostat is contained in an individual
.csv ﬁle, organized by a month and a year, using 5 minutes
intervals. The following symbols are used to describe it:
•compHeat: heating runtime in seconds
•dateTime: date and time that the reading was taken
•schedule: three defaults states of the thermostat -
sleep/away/home - that are used to deﬁne the comfort
settings (preferred temperature)
•Tctrl: current average temperature in the household
•Tout: outdoor temperate
•TstpHeat : indoor heat setpoint in the current schedule
•t: time period
Moreover, we introduce the following variables:
•heat: time to gain 1°C in particular household
•lef t: time left before the change of the schedule
•over: overheated time
B. Optimization problem
To introduce the optimization problem, let us as look at
the Figure 1. The data was obtained from the Ecobee device,
located in the Vancouver, Canada, during a sunny day. The
temperature setting for the day was 24°C and 21°C for the
night. At night, the temperature in the home was equal to
21°C. Around 1:00, the system turned on the heating to achieve
the set point for the day preference, which was set for 9:00.
The temperature in the house reached the desired value. At
11:35 the temperature was 0.5°C higher than preferred. Around
the 16:30 the indoor temperature reached the day high 26°C,
which was 2.5°C higher than the comfortable setting for this
household, due to the weather impact. The time that the
temperature was higher than the preference setting was an
example of overheated time - over.
Fig. 1: The example scenario
Thus, we want to optimize the following problem:
over(t), where over(t)∈Tctr l(t)≥TstpHeat(t)(1)
Tctrl ≤TstpHeat no longer than 2 hours (2)
over(t)/∈sleep ∪away (3)
The formula (1) deﬁnes that we are minimizing the time
when the home is overheated (Tctrl(t)≥TstpHeat(t)), subject
to two constraints:
(2) - we cannot wait to heat up the household to reach the
comfortable temperature (TstpHeat ) for longer than 2
hours (system requirement)
(3) - the overheated time is not considered during sleep
or away schedules (after the analysis we can observe
that during these two stages homes are in 85% of
time not heating, just ”cooling down”)
A. Standard Heating Algorithm
The logic in the Standard Heating (SH) algorithm is
straightforward - it turns the heating pump and the fan on when
Tctrl < TstpHeat . Please note that the measurement of data in
Ecobee Smart Thermostats is performed every 5 minutes; thus,
the compHeat time is set to 300 seconds.
B. History Based Heating Algorithm
The History Based Heating (HBH) algorithm calculates the
average time to heat up and cool down the house. It then
uses this information to determine the best way to use the
HVAC system to achieve the desired indoor temperature (see
the pseudo-code in Algorithm 1).
The procedure of home heating in case of Home sched-
ule is similar to the SH algorithm. Whenever the preferred
temperature is higher than the current one, the system turns
on the heating. The main difference is during Sleep and
Algorithm 1: History Based Heating Algorithm (HBH)
Data: Tctrl,TstpHeat ,schedule,lef t,heat
1Calculate the average heating time heat to gain 1°C;
2switch Schedule do
3case Home do
4while Tctrl < TstpHeat do
5Turn the fan and heating pump on;
6Set compHeat = 300s;
10 case Sleep/Away do
11 Read TstpHeat from the next Home schedule;
12 if TstpHeat −Tctrl ≥left −heat then
13 Turn the fan and heating pump on;
14 Set compHeat = 300s;
19 return compHeat;
Away schedules, where the algorithm calculates the average
household heating time to gain 1°C. It is then used to estimate
the correct time to start the heating before the next Home
schedule. It is called SmartRecovery - the algorithm prioritizes
the time when household members are awake, and start heating
earlier, so they can wake up in the comfortable temperature.
As we can observe in the lines 10-17 (case Sleep/Away), we
compare the difference between the two: the subtraction of
preferred temperature and the current one with the subtraction
of time remaining to the next Home schedule (left) and the
time needed to gain 1°C (heat). If the condition is fulﬁlled,
we then turn the heating on.
C. Weather Based Heating Algorithm
Our proposed solution, the Weather Based Heating (WBH)
algorithm, calculates the average time to heat up and cool down
the house in correlation to the weather based on the nearest
weather station (see Algorithm 2).
The idea behind the Away/Sleep schedules is the same as
for the HBH algorithm. The difference is during the Home
schedule. During this stage, we check if the current tempera-
ture is lower than preferred, but instead of turning the system
on, we run the predictHeating() method, shown in Algorithm 3.
First, we run a linear regression method on historical data to
calculate the average weatherImpact for every household. It
is a relation on how speciﬁc weather affects the temperature
inside, measured in °C per 5 minutes (e.g., 0.21°C per 5
minutes). It depends on multiple criteria: the current weather,
the temperature outside, the size of the household, the season
of the year and the time of the day. Next, we obtain the weather
forecast from DarkSky API. In the ﬁnal step, we check the if-
else condition, using the average weather impact for the next
three hours (averageImpact), based on the obtained weather
forecast. If the condition from line 3 is met, we turn the fan
and heating pump on; otherwise, we set compHeat = 0;
Algorithm 2: Weather Based Heating Algorithm (WBH)
Data: Tctrl,TstpHeat ,schedule,lef t,heat,
1Calculate the average heating time heat to gain 1°C;
2switch Schedule do
3case Home do
4while Tctrl < TstpHeat do
5predictHeating() and set compH eat;
9case Sleep/Away do
10 Read TstpHeat from the next Home schedule;
11 if TstpHeat −Tctrl ≥left −heat then
12 Turn the fan and heating pump on;
13 Set compHeat = 300s;
18 return compHeat;
Algorithm 3: The predictHeating() method
Data: Tctrl,TstpHeat ,schedule,lef t,heat
1For every thermostat calculate weatherImpact;
2Using Dark Sky API check weather forecast;
3if TstpHeat −Tctrl +averageImpact ≥lef t −heat
4Turn the fan and heating pump on;
5Set compHeat = 300s;
9return compHeat = 0;
IV. SIMULATION SETUP
Simulations were performed using data from three winters
(December-March), i.e., 2015-2018. We used data only from
Ecobee 3, Ecobee 3 Lite and Ecobee 4 devices, as they have
the same sensors to measure the temperature. The average
user preferences that we obtained from the data were as
follows: Tctrl = 20.6°C and TstpHeat = 21.7°C. The number
of households was equal 54000.
First, we want to present the computed values of
weatherI mpact, in Table I. As expected, the largest weather
impact on the household temperature was during sunny or
windy days. Sun and high temperature during the day caused
the increase of a maximum of 1.41°C per 5 minutes, whereas
the snowy weather during cold day impacted particular house-
hold with a maximum of -3.3°C per 5 minutes. Please note
that we calculated the impact during the moments where the
thermostats were in the Away schedule, which means that no
heating was involved. Moreover, we checked only the houses
without cooling systems. That allowed us to calculate the direct
weather impact on every single household.
TABLE I: The average ranges of weather impact and the
weather type categories
Category Impact Category Impact
Sunny (≤-25°C) -0.7°C:0.01°C Sunny (-25°C:-10°C) -0.3°C:0.31°C
Sunny (-10°C:0°C) 0.12°C:0.36°C Sunny (0°C:15°C) 0.28°C:0.43°C
Cloudy (≤-25°C) -1.25°C:-1.12°C Cloudy (-25°C:-10°C) -1.09°C:-0.81°C
Cloudy (-10°C:0°C) -1.01°C:-0.56°C Cloudy (0°C:15°C) -0.83°C:0.12°C
Snowy (≤-25°C) -3.24°C:-1.22°C Snowy (-25°C:-10°C) -2.3°C:-0.61°C
Snowy (-10°C:0°C) -0.88°C:-0.25°C Snowy (0°C:15°C) -0.55°C:-0.11°C
Windy (≤-25°C) -3.3°C:-2.1°C Windy (-25°C:-10°C) -2.71°C:-1.18°C
Windy (-10°C:0°C) -1.67°C:-0.6°C Windy (0°C:15°C) -0.81°C:0.32°C
Rainy (-25°C:-10°C) N/A Rainy (-10°C:0°C) -0.92°C:-0.02°C
Rainy (0°C:15°C) 0.20°C:0.73°C Rainy (≥15°C) 0.53°C:0.99°C
Another important observation was that during cloudy
days, the temperature did not increase much, even on hot days.
The direct sunlight, even with the lower outside temperature,
had more effect on the households than the weather during
the cloudy day with the higher temperature outside. Finally,
the snowy weather caused reduction of the indoor temperature,
even if the temperature outside was above 0°C.
Next, we present the maps related to the average temper-
ature outside, and the average overheated time of households,
during winters of 2015-2018, shown in Figure 2 and Figure 3,
Fig. 2: The average temperature outside in North America in
winters of 2015/2016-2017/2018
Fig. 3: Overheated time for the households in North America
during winters of 2015/2016-2017/2018
There is a correlation between the average temperature
outside and the overheated time. In the regions with higher
temperature, even during the winter, the overheated time was
in a range of 20-30% of overall time. Moreover, we can notice
that houses in the most populated areas on the West and East
Coasts of North America were signiﬁcantly overheated. The
gray area in the Northern regions indicates that we do not
have enough data.
Finally, we present the comparison of the algorithms that
were designed to reduce the overheated time, in Table II. We
compared three algorithms, presented in this paper, namely
Standard Heating (SH), History Based Heating (HBH) and
Weather Based Heating (WBH). The best algorithm was
TABLE II: Algorithms comparison for the overheated time.
Years SH HBH WBH
2015/2016 31.59% 20.83% 12.83%
2016/2017 33.22% 16.93% 12.86%
2017/2018 30.12% 14.02% 10.83%
WBH. It signiﬁcantly reduced the overheated time of the
households to the range of 10-12% of the time, where the
worst, SH algorithm, resulted in around 30% overheated time.
The next observation is that History Based Heating algorithm
(HBH) improved in each subsequent winter, from 20.83% of
the overheated time to 14.02%. The ability to ”learn” about
the particular household bring the positive effect on the overall
performance in time.
We also analyzed the performance of the algorithms based
on the country, shown in Table III.
TABLE III: Algorithms comparison for the overheated time.
Country SH HBH WBH
Canada 29.36% 22.11% 12.44%
USA 23.31% 15.88% 12.94%
The trend was similar. The best algorithm was WBH, the
second best HBH, and the worst performance was shown
using SH algorithm. The interesting observation is that SH
and HBH were able to achieve better results in the USA than
in Canada. It is correlated to the results presented in Figure 3.
The overheated times of the houses in Montana, North Dakota
and Minnesota were much lower than in the other parts of the
United States, thus, it explains a slightly better performance
than in Canada.
The reduction of heating time in comparison to Standard
Heating algorithm is shown in Table IV. The usage of the
HBH and WBH algorithms allowed to save the time of the
heating in about 12-15%, and 19-21%, respectively. It means
that we also were able to save the costs of gas, electricity or
any energy sources.
TABLE IV: Heating time improvement
Year HBH WBH
2015/2016 12.71% 19.31%
2016/2017 14.58% 21.32%
2017/2018 13.71% 20.99%
In this paper, we presented a study on the weather impact
on the households in North America. We compared three
algorithms, two used in current HVAC systems and one,
proposed in this paper. Results showed that the houses in North
America are overheated in around 20%-30% of the time. On
the other hand, the implementation of history based or weather-
related algorithms allows to increase the efﬁciency of home
heating, thus, reduce the costs of it.
We gratefully acknowledge the support and contribution of
Ecobee Ltd. and Ecobee customers to this research.
 H. Farhangi, “The path of the smart grid,” IEEE Power and Energy
Magazine, vol. 8, no. 1, pp. 18–28, 2010.
 G. Newe, “Delivering the internet of things,” Network Security, vol.
2015, no. 3, pp. 18–20, 2015.
 C. Perera, C. H. Liu, S. Jayawardena, and M. Chen, “A Survey on
Internet of Things from Industrial Market Perspective,” IEEE Access,
vol. 2, pp. 1660–1679, 2015.
 T. U. Daim and I. Iskin, “Smart thermostats: are we ready?” Interna-
tional Journal of Energy Sector Management, vol. 4, no. 2, pp. 146–151,
 J. Jacobson, “Info & Answers: Anatomy Lesson: Getting to Know Smart
 J. Torriti, “A review of time use models of residential electricity
demand,” pp. 265–272, 2014.
 M. Pritoni, A. K. Meier, C. Aragon, D. Perry, and T. Peffer, “Energy
efﬁciency and the misuse of programmable thermostats: The effective-
ness of crowdsourcing for understanding household behavior,” Energy
Research and Social Science, vol. 8, pp. 190–197, 2015.
 T. Peffer, M. Pritoni, A. Meier, C. Aragon, and D. Perry, “How people
use thermostats in homes: A review,” pp. 2529–2541, 2011.
 J. Ansei, “Smart vs. Connected Thermostats: What’s the Difference?”
Air Conditioning Heating & Refrigeration News, vol. 256, no. 7, pp.
 C. Koehler, B. Ziebart, J. Mankoff, and A. Dey, “TherML:
occupancy prediction for thermostat control,” Proceedings of
the 2013 . . . , pp. 103–112, 2013. [Online]. Available:
 S. Lee, Y. Chon, Y. Kim, R. Ha, and H. Cha, “Occupancy prediction
algorithms for thermostat control systems using mobile devices,” IEEE
Transactions on Smart Grid, vol. 4, no. 3, pp. 1332–1340, 2013.
 G. Hernandez, O. Arias, D. Buentello, and Y. Jin, “Smart Nest Ther-
mostat : A Smart Spy in Your Home,” Black Hat USA, pp. 1–8, 2014.
 J. Scott, A. Brush, J. Krumm, B. Meyers, M. Hazas, S. Hodges, and
N. Villar, “PreHeat: Controlling Home Heating Using Occupancy Pre-
diction,” UbiComp ’11 Proceedings of the 13th international conference
on Ubiquitous computing, no. UbiComp, pp. 281–290, 2011.
 J. Lu, T. Sookoor, V. Srinivasan, G. Gao, B. Holben, J. Stankovic,
E. Field, and K. Whitehouse, “The Smart Thermostat: Using Occupancy
Sensors to Save Energy in Homes,” Proceedings of ACM SenSys,
vol. 55, pp. 211–224, 2010.
 Y. Liang, D. I. Levine, and Z. J. M. Shen, “Thermostats for the smart
grid: Models, benchmarks, and insights,” Energy Journal, vol. 33, no. 4,
pp. 61–95, 2012.
 D. T. Nguyen and L. B. Le, “Joint optimization of electric vehicle and
home energy scheduling considering user comfort preference,” IEEE
Transactions on Smart Grid, vol. 5, no. 1, pp. 188–199, 2014.
 H. T. Nguyen, D. Nguyen, and L. B. Le, “Home energy management
with generic thermal dynamics and user temperature preference,” in
2013 IEEE International Conference on Smart Grid Communications,
SmartGridComm 2013, 2013, pp. 552–557.
 K. F. Fong, V. I. Hanby, and T. T. Chow, “HVAC system optimization
for energy management by evolutionary programming,” Energy and
Buildings, vol. 38, no. 3, pp. 220–231, 2006.
 S. Noh, J. Yun, and K. Kim, “An efﬁcient building air conditioning
system control under real-time pricing,” APAP 2011 - Proceedings:
2011 International Conference on Advanced Power System Automation
and Protection, vol. 2, pp. 1283–1286, 2011.
 T. F. Sanquist, H. Orr, B. Shui, and A. C. Bittner, “Lifestyle factors
in U.S. residential electricity consumption,” Energy Policy, vol. 42, pp.
 Ecobee Ltd., “Donate Your Data,” 2018. [Online]. Available: