ArticlePDF Available

Planning for sustainable cities by estimating building occupancy with mobile phones


Abstract and Figures

Accurate occupancy is crucial for planning for sustainable buildings. Using massive, passively-collected mobile phone data, we introduce a novel framework to estimate building occupancy at unprecedented scale. We show that, at urban-scale, occupancy differs widely from current estimates based on building types. For commercial buildings, we find typical occupancy rates are 5 times lower than current assumptions imply, while for residential buildings occupancy rates vary widely by neighborhood. Our mobile phone based occupancy estimates are integrated with a state-of-the-art urban building energy model to understand their impact on energy use predictions. Depending on the assumed relationship between occupancy and internal building loads, we find energy consumption which differs by +1% to -15% for residential buildings and by -4% to -21% for commercial buildings, compared to standard methods. This highlights a need for new occupancy-to-load models which can be applied at urban-scale to the diverse set of city building types.
This content is subject to copyright. Terms and conditions apply.
Planning for sustainable cities by estimating
building occupancy with mobile phones
Edward Barbour1,2,3, Carlos Cerezo Davila4, Siddharth Gupta1, Christoph Reinhart4, Jasleen Kaur5&
Marta C. González 1,3,6
Accurate occupancy is crucial for planning for sustainable buildings. Using massive,
passively-collected mobile phone data, we introduce a novel framework to estimate building
occupancy at unprecedented scale. We show that, at urban-scale, occupancy differs widely
from current estimates based on building types. For commercial buildings, we nd typical
occupancy rates are 5 times lower than current assumptions imply, while for residential
buildings occupancy rates vary widely by neighborhood. Our mobile phone based occupancy
estimates are integrated with a state-of-the-art urban building energy model to understand
their impact on energy use predictions. Depending on the assumed relationship between
occupancy and internal building loads, we nd energy consumption which differs by +1% to
15% for residential buildings and by 4% to 21% for commercial buildings, compared to
standard methods. This highlights a need for new occupancy-to-load models which can be
applied at urban-scale to the diverse set of city building types. OPEN
1Department of Civil and Environmental Engineering, MIT, Cambridge, MA, USA. 2Centre for Renewable Energy Systems Technology, Loughborough
University, Loughborough, LE, UK. 3Lawrence Berkeley National Laboratory, Berkeley, CA, USA. 4Sustainable Design Lab, MIT, Cambridge, MA, USA.
5Signify Research North America (formerly Philips Lighting), Cambridge, MA, USA. 6Department of City and Regional Planning, UC, Berkeley, CA, USA.
Correspondence and requests for materials should be addressed to M.C.G. (email:
NATURE COMMUNICATIONS | (2019) 10:3736 | | 1
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Energy use in buildings accounts for over 40% of total pri-
mary energy use in the U.S and E.U.1,2, and is particularly
concentrated in urban areas, which are set to rapidly expand
in the near future3.Efciency measures applied to urban build-
ings therefore constitute an unparalleled opportunity for global
energy use and emissions reduction4,5. Recognizing this, many
national and local governments have implemented or are con-
sidering policies to promote building efciency, for example, by
incentivizing increased insulation levels or high efciency appli-
ances. However, these initiatives typically operate with limited
resources, both in terms of budget and technical assistance.
Urban-scale building energy models have therefore been devel-
oped to support strategic decision-making by pinpointing policies
with the greatest saving potential.
Urban-scale energy models can be broadly split into two cate-
gories; those that rely on data-driven statistical or machine
learning models (black-box models) and those that include a
physics-based building simulation engine (white-box models). In
the rst category, statistical models typically reduce the energy
consumption of a large number of buildings to a small number of
explanatory variables68. These models can subsequently be used
for energy benchmarking and were the rst to estimate urban-
scale impacts of potential Energy Conservation Measures (ECMs).
However, while useful, these models have two notable limitations.
Firstly, the lack of detail regarding individual buildings can result
in poor out-of-sample predictions, particularly related to the
widespread impacts of many ECMs and feedback between ECMs
and building operation8, and secondly, they cannot be used in the
design of new, or densication of existing, urban districts required
to facilitate booming urban population growth9.
In the second category, a new class of Urban Building Energy
Models (UBEMs) has recently been developed, which includes
detailed physics-based simulations of thousands of individual
buildings in cities or districts1016. In an UBEM, each individual
building is represented by a dedicated physics-based engineering
model, the likes of which are commonly used worldwide for
engineering design, code-compliance demonstration, and
improved operation9. UBEMs rely on highly automated work-
ows to generate the detailed individual building models without
requiring time-intensive work by building modeling experts9,17,18,
often combining several datasets, including Geographic Infor-
mation Systems (GIS) databases, LiDAR and stock building
archetypes. The models are then calibrated to match real data
samples11,19. However, despite high versatility20, inaccuracies are
introduced in these models due to a lack of understanding
regarding building occupancy at an urban scale2022.
For most buildings, with the exception of certain industrial
facilities, occupant presence and behavior have a decisive impact
on building energy use. In recognition of this well-known fact, the
International Energy Agency (IEA) has funded two Annexes
dedicated to understanding occupant presence and behavior in
buildings and improving models, Annex 66 and its follow-up,
Annex 7923,24. While at the individual building level, several
occupancy models exist (i.e., refs. 25,26.), only a few studies have
considered occupancy in an urban context27. In absence of better
data, standardized deterministic space-based occupant presence
and behavior models are the only viable approach for UBEMs of
urban areas with mixed-use buildings20.
To understand building occupancy at an urban-scale requires
either a vast array of sensing infrastructure or knowledge of popu-
lation movements over an entire metropolitan regionscapturing the
daily movements of citizens between different buildings. To that end,
in this work, we propose using massive, passively-collected mobile
phone data to infer building occupancy on a city-level. This data has
already been used to extract locations where individuals stay (stay
points), as well as to infer their location-based activities28 and for
highly-aggregated electricity load predictions29,30. A recent modeling
framework (TimeGeo) synthesizes previous ndings in human
mobility31,32, demonstrating that sparse mobile phone data can be
used to model individual trajectories for entire urban populations33.
Several other passive data sources have already been used to
understand city dynamics34 and infer building occupancy3539,in
particular bluetooth, wi, cameras and electricity data, as well as
their combinations. However, in contrast with mobile phone data,
these sources are not available at sufcient scale for predicting
simultaneous occupancy for thousands of different buildings.
In this work, we develop a method for estimating building
occupancy at urban-scale by extending the TimeGeo frame-
work33. Our statistical method assigns occupants to buildings and
we demonstrate the proposed framework for 83,000 buildings in
the city of Boston, using the individual trajectories of 3.5 million
urban inhabitants. The building assignment is probabilistic and
analogous to route assignment models, successfully developed
with mobile phone data in vehicular trafc40,41. In order to
demonstrate the signicance of our results for energy modelling,
we develop an UBEM of a central, mixed-use neighborhood and
compare occupancy estimates between standard reference meth-
ods and the mobile-inferred building occupancy. Since occupancy
is a principal driver of building energy use but no model relating
occupancy and building loads on this scale exists, we develop
approximate upper and lower bound scenarios for the impact of
occupancy. Firstly, a low-impact of occupancy scenario wherein
we adjust the occupant-driven reference building loads according
to the relative occupancy, primarily changing the timing of loads
rather than the magnitude. Secondly, a high-impact scenario,
wherein we adjust the occupant-driven loads according to the
absolute mobile-inferred occupancy, signicantly changing both
load magnitude and timing. We nd that these scenarios imply
energy consumption that differs by +1to15% in residential
buildings and by 4to21% in commercial buildings. Finally,
we illustrate that the mobile occupancy implies the impact of
energy efciency measures may be mispredicted, nding savings
from insulation improvements up to 25% higher than predicted
with standard occupancy while savings from improving equip-
ment efciency were predicted up to 40% lower in our modelled
region. Our method therefore provides an improved alternative
estimate of the number of occupants in buildings, in contrast to
indirect space-based estimates which represent standard practice
in city-scale energy planning today.
Reference building occupancy. In the United States, the
Department of Energy (DOE) reference commercial buildings
dataset42 provides the standard set of templates for the US
building stock for the EnergyPlus simulation engine43,44. These
include sixteen different building categories in all US climate
zones with different construction periods. Each building category
has its own associated fractional occupancy and load schedules
based on previous studies by the DOE, ASHRAE and several US
National Research Labs. The schedules include estimates for
occupant density based on building type (i.e., small retail occu-
pancy: 0.141 occupants per m2of oor area) and peak load values
per unit oor area for different load types (i.e., ofce equipment
density: 17 Wm2)45. When used for individual buildings, the
space-based templates can be adjusted using local knowledge of
the occupants; however, this is not possible at an urban-scale20.
Hence, without any empirically based alternatives, city-scale
energy models typically use the reference building occupancy,
introducing uncertainty into energy-use predictions and, as a
result, into predicted savings from efciency measures20. This
barrier for urban energy policy development is an international
2NATURE COMMUNICATIONS | (2019) 10:3736 | |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
issue, since all national building stock models suffer from the
same lack of empirical occupancy data.
From mobile phone stay points to building occupancy. Using
the TimeGeo framework33 we simulate the individual mobility of
Bostons entire metropolitan area population at high resolution.
The framework starts with the Call Detail Records (CDRs) of 1.92
million anonymous mobile phone users of for the period 20
February till 30 March 2010 in the Greater Boston area (see
Methods). Stay points for these users are identied based on
consecutive mobile phone records within certain temporal and
spatial thresholds, namely 10 min and 300 m. Each stay point is
characterized as either home, work or other, depending on
whether it is estimated to occur at the individuals place of resi-
dence, work or some other location33, and includes a start time,
duration, latitude/longitude coordinate pair and user id. We nd
that nearly 200,000 users have more than 50 total stays and at
least 10 home stays (home stays only occur at an individuals
specic home-designated building) during the observation period.
These are designated active users and their phone records are
used to extract the mobility parameters for an empirically based
population-wide mobility model, reliant only on measurable
parameters in the data. In each census tract in Bostons metro
area, TimeGeo expands the active phone users to the population
(i.e., simulating 3.54 million people including 2.10 million
workers and 1.44 million non-workers). The model encompasses
individuals with homes in the region shown in Fig. 1a. The colors
in Fig. 1a indicate the average expansion factor of commuters and
non-commuters for each tract, calculated as the ratio of the total
tract population (as dened by the 2010 census) to the number of
active mobile users with homes within that tract. The model has
been proven to be accurate at the census tract level33 in com-
parison with both the 2009 National Household Travel Survey46
and the 20102011 Massachusetts Travel Survey47 (see Supple-
mentary Fig. 1 and Supplementary Note 1). TimeGeo offers sig-
nicant improvements on current urban mobility models,
which typically involve expensive surveys and have very low
sampling rates. It should be noted that the model is primarily
limited by the extent, resolution and duration of the data. As
such, as more large-scale data with higher frequency (i.e., GPS
traces) and longer observation periods (i.e., years) become
available, the model resolution could be increased and hetero-
geneity improved33.
The ne-scale population mobility model has been shown to
agree with existing methods of quantifying population mobility at
the census tract level33. Hence, each stay can be thought of as a
person-visit to a tract, as implied by the empirical mobile phone
data and the Boston metro census population, and in agreement
with current best-in-class travel demand models. Figure 2a
illustrates the journeys for the people who visit the ve tracts
shown in Fig. 1c. The predicted tract-level occupancy is revealing,
along with the expected ux of people in-and-out-of the area, as
6 km0 20 60 km
Common type
Residence + Retail
Office + Retail
Hotel + Retail
Fig. 1 Extent of the data. aThe extent of the TimeGeo simulation in the greater Boston metropolitan area. Each census tract is colored by its expansion
factor. bThe extent of the Boston Buildings datasettracts are marked as either majority residential or majority commercial depending on the most
common building use. cThe buildings used in the Urban Building Energy Model, colored by type. These buildings occupy ve census tracts. Black boxes in
Fig. 1a. and Fig. 1b. illustrate the extent of the next panel. Maps created using Google Maps, 2019 Google
NATURE COMMUNICATIONS | (2019) 10:3736 | | 3
Content courtesy of Springer Nature, terms of use apply. Rights reserved
shown in Fig. 2b, c. Here, we see that the implied occupancy of
the ve tracts over the course of a typical weekday using DOE
reference building models ranges from 71,300122,500, whereas
the mobile-informed model ranges from 11,80020,700. Given
the census population for these ve tracts is 13,600, the mobile
model certainly seems more credible and the DOE reference
occupancy standard may be overestimating actual occupancy
numbers by over 600%.
Each tract typically encompasses hundreds to thousands of
buildings, for example, the ve tracts used in the energy model
(Fig. 1c) contain 1330 buildings. Therefore, to assign each stay to
a building we assume that each stay point represents a visit to a
building in the same tract and probabilistically map each stay to a
building. Buildings data is provided by the city of Boston and
includes footprint, geometry, height and a tax assessment type
(see Supplementary Fig. 2 and Supplementary Note 2). We re-
classify the buildings into three broad classes; residential,
commercial and industrial (see Methods) and stipulate that
home type stays can only be assigned to residences, while work
stays can be assigned to all building types except residential and
other stays can be assigned to commercial buildings only.
Therefore, we assume that an individual stay point can potentially
be located in any building within the same tract, provided the
building is open for the stay duration and the building
functionality matches the stay type. To probabilistically select a
particular building, we use nominal building capacities, and
assume that, in general, buildings with higher capacity attract
more stays. To estimate the nominal capacities we use typical Per
Capita Area (PCA) values for each functional building type (see
Methods, Supplementary Table 1 and Supplementary Note 3).
We assume residential buildings are always open and have a PCA
of 40 m2, i.e. a house with a oor area of 200 m2would have a
nominal capacity of ve occupants. For non-residential buildings,
we use Places Of Interest (POIs) available in digital maps where
the exact building functional use is unknown (Fig. 3b, see
Methods, Supplementary Table 2 and Supplementary Note 4).
Furthermore, we hypothesize that at a city-block level there
may be a rich-get-richer effect, which results in popular areas
attracting more people per-unit-oor-area. Therefore, in our
model, stays are preferentially attracted to areas with larger
capacities for that type of stay. This is based on the observation
that shops and industries agglomerate due to clustering of
economic activities48for example, resulting in the formation of
popular shopping, dining or entertainment districts. To model
this, we adopt a two stage process when assigning a stay to a
building within a particular tract. Firstly, we spatially group the
possible buildings by their centroid coordinates using hierarchical
agglomerative clustering and Wards minimal increase of variance
method49 to form city-blocks. Then, after probabilistically
selecting a block we select an individual building (Fig. 3c). The
preferential attraction at the cluster level is given as follows:
For a given cluster iwith a nominal capacity C
,P(i) is the
probability of that cluster being assigned a particular stay. The
total capacity of each cluster iis the sum of the capacity of the
buildings contained within-cluster i, and the total tract
capacity is the sum of the capacity of the Nclusters. Therefore,
j¼1Cij, where the buildings are indexed by j, and C
and A
are the PCA and total oor area of building jin
cluster i, respectively. The parameter μvaries the degree of
preferential attraction to areas with a high-capacity for stays of
that type (see next section). Once a cluster ihas been selected,
then within that cluster we assign the stay to a building jwith
probability P(j|i), proportional to its relative within-cluster
capacity, as described by Eq. (2).
PðjjiÞ¼ Cij
Figure 3illustrates the process of assigning stays to the
buildings. Once we have considered all the stay points within a
particular tract for the 24 h period, we assume that we have a
complete picture of building occupancy over the day.
Occupancy for Bostons buildings. We now study the expected
building occupancy for 0 μ1. The rationale for adopting these
bounds are, (1) we do not expect a rich-get-richer effect where the
0 20 60 km40
Distance (km)
DOE reference Census pop.
Population flux
4 a.m. 8 a.m. noon 4 p.m. 8 p.m.
Fig. 2 Residents and visitors in the Back Bay. aPredicted journeys from the TimeGeo for users who visit the ve tracts shown in Fig. 1c. Each line represents
approximately 50 users. bThe total predicted hourly occupancy for the area shown in Fig. 1c according to the TimeGeo model and as implied by the DOE
reference building model. The census population for the area is also shown. cThe predicted population ux as implied by the reference buildings and
TimeGeo model. Map created using Google Maps, 2019 Google
4NATURE COMMUNICATIONS | (2019) 10:3736 | |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
returns are diminishing (i.e., if μ< 0 then areas with more
opportunities (capacity) will attract proportionally less people per
unit oor space) and (2) we do not anticipate that μ> 1 because
above this value we nd that high-capacity areas start to dominate
the stay attraction to an unreasonable degree, leaving many
unoccupied buildings. The effect of the non-linear parameter μis
discussed in detail in Supplementary Note 5 (see also Supple-
mentary Fig. 3 and Supplementary Table 3).
The results are consistent with residences having high night
occupancy and commercial buildings having higher daytime
occupancy, as people generally leave their homes during the day
(Fig. 4a). Figure 4b shows that greater μresults in more highly
occupied spaces, shifting occupants from lower occupancy spaces
to higher ones. In the range 0 μ1 we see that shift in the
distributions change is slight and the median occupancy-over-
capacity ratio for residential buildings decreases from 1.4 to 1.2
while the median commercial occupancy decreases from 0.3 to
0.2. For the 83,000 buildings in the city of Boston, we nd that
most commercial buildings have nominal capacities that are
signicantly higher than their peak occupancy (Fig. 4b shows that
mobile-inferred peak occupancy is higher than the capacity in
only 5% of commercial buildings). Conversely, for residential
buildings, the occupancy-over-capacity distribution has a long tail
and implies a wide range of residential occupant densities. These
are highly neighborhood dependent and we observe high
numbers of occupants in student areas situated in close proximity
to universities, where higher-than-average residential occupancy
persists throughout the day (Fig. 4a and Supplementary Fig. 4).
For commercial buildings, our estimate of peak capacity is rarely
approached. We nd that when μ=0.5, the most common value
for residential peak-occupancy-over-capacity is 1 while the
median is 1.3. Therefore our assumed residential PCA appears
close to typical values for residential space in Boston.
Occupancy for urban energy predictions. At this point, we
create an urban energy model for the mixed-use district in the
Back-Bay region of Boston, covering ve census tracts with 1,330
buildings as shown in Fig. 1c. Out of these, we were able to model
1,266 buildings using the DOE reference buildings (see Supple-
mentary Fig. 5). The other buildings were considered in the
occupancy-assignment process (since they could still be open and
attract people); however, due to atypical uses (i.e,. re/police,
churches, etc.), no appropriate reference model existed and
therefore these were not included in the energy model. To gen-
erate the modeled buildings we used data regarding use per oor,
period of construction and geometry. We use the previously
developed UMI framework10 (see Methods) which is based on
EnergyPlus, a state-of-the-art whole building energy simulation
engine developed by the US DOE43. The building constructions
and systems were specied through archetype templates based on
the DOE Reference Buildings dataset (see Supplementary Note 6
and Supplementary Figs. 6 and 7). The templates have the same
nominal peak occupancy values as used for the 83,000 building
analysis and also contain 24-h occupancy proles dependent on
the building use. The occupancy is compared to our mobile-
informed occupancy results, as shown in Fig. 4d. Our results
suggest that in individual commercial buildings the reference
building occupancy can over-predict occupancy by an order of
magnitude or more at all times during the day, in our modeled
The reason for the major discrepancy between the reference
building occupancy and mobile-inferred occupancy is that the
mobile occupancy is normalized to the Boston metro area
population, whereas the reference building occupancy proles are
based on a combination of factors including ASHRAE ventilation
requirements, re safety codes, and generalized building-
manager-surveys. Importantly, these factors have no ability to
account for the city-specic population and its movements, and
therefore, as shown in Fig. 2b these are inconsistent with the
census population. Accordingly, for the total modeled neighbor-
hood we nd the mobile occupancy corresponds to only 1524%
of the reference building occupancy over the course of a typical
weekday. The population uxes also seem unreasonable whereas
the mobile-inferred people-uxes between different regions are
consistent with current transportation models33.
Occupant-driven energy loads. Since occupancy is often a large
driver for building energy use8,20,37,50, the differences between the
reference building and the mobile-inferred occupancy imply
different energy consumption patterns. Unfortunately, while
methodologies exist that can predict appliance use for individual
buildings based on the maximum number of occupants26,no
satisfactory method exists that can function at urban-scale20,50,51.
Therefore, we rely on the occupant-driven loads prescribed in the
DOE reference buildings, which are empirically related to the
reference occupancy. We develop two rule-based scenarios, which
broadly represent high-impact-of-occupancy and low-impact-of-
occupancy case studies, and compare the results of each with a
base model using the DOE reference buildings. The resulting
three scenarios are described as follows. First, the DOE reference
0 100 200 300 m
Possible buildings
Other tract buildings Stay
0 100 200 300 m
Selected building
Other cluster buildings
0 100 200 300m
Get possible buildings Cluster into 'walkable' groups
For each stay:
Select a cluster and then a building
Probabilistic occupancy assignment
Update building occupancy
6 a.m. Noon 6 p.m.
Call detail records
Identify active users and extract
mobility parameters
Urban population movements:
Stay types, times, locations and
Building outline, height, class
Use google maps API to get POIs
From POIs types estimate PCA and
opening hours for non-residential
Building data
Fig. 3 CDR data and building data to building occupancy method. aThe
TimeGeo framework going from call detail records to population-wide stay
locations33.bThe building data includes building type, PCA and opening
hours. cThe set of possible buildings for a stay depends on building type
and opening hours (for the stay type other as shown, possible buildings are
open commercial buildings). Buildings are clustered into localized spatial
groups using Wards minimum increase of variance method and the
dendrogram is truncated at a Ward distance of 500 m, which yields city-
block shaped clusters. Then, a cluster is probabilistically selected followed
by a building, and the stay is added to that buildings occupancy prole.
Maps created using Google Maps, 2019 Google
NATURE COMMUNICATIONS | (2019) 10:3736 | | 5
Content courtesy of Springer Nature, terms of use apply. Rights reserved
scenario is the control scenario and presents simulations using
occupancy and load schedules unaltered from DOE reference
building templates. Second, the low-impact scenario develops
occupancy-driven load schedules considering the ratio between
the relative mobile occupancy and relative reference building
occupancy, as described in Eq. (3). Third, the high-impact sce-
nario develops occupancy-driven load schedules considering the
ratio between the absolute mobile occupancy and reference
building occupancy, as described in Eq. (4).
LmobðhÞ¼bþðLDRBðhÞbÞOmob ðhÞ
In Eqs (3) and (4), Lmob(h) is the fractional schedule for each
load at hour hassociated with the mobile occupancy, Omob(h).
LDRB(h) is the DOE Reference Building load schedule associated
with the reference building occupancy prole ODRB(h) and bis
the baseloadthe proportion of load independent of occupancy
which we assume is given by the minimum value of LDRB(h).
The purpose of the scenarios is to create relationships between
occupancy and the resulting use of lights, equipment and hot
water. In both, we dene a constant baseload which includes non-
occupant related energy use, such as refrigerators etc. For the
remaining loads we assume upper and lower bounds as follows:
For the lower bound (low-impact), we assume that loads vary by
the relative ratios between mobile and reference building
occupancy, so if OmobðhÞ
MAXðODRBÞthe predicted load increases.
This primarily corrects time-of-day effects, i.e., if occupants arrive
earlier loads are shifted to be earlier. In contrast, the upper bound
(high-impact) assumes non-baseload loads change as the absolute
ratio between the mobile and reference building occupancies, i.e.,
if the mobile occupancy predicts half the number of people at a
certain time the (non-baseload) loads will be halved. The high-
impact scenario works best in a compartmentalized building,
such as an apartment block, where light and equipment scale with
the number of occupied rooms at any moment in time.
Conversely, the low-impact scenario is suitable for open plan
ofces or retail, where occupants drive when lighting, fans etc. are
switched on, but the exact consumption is insensitive to the
absolute occupancy. However, Eq. (3) leads to implausible results
if the occupant number predicted by the mobile occupancy
schedule at a particular time is much lower than the reference
building schedule and simultaneously corresponds to a larger
relative mobile occupancy. Therefore, we stipulate that the
predicted mobile load can only be a factor of Omob ðhÞ
ODRBðhÞhigher than
the reference building load. Supplementary Fig. 8 illustrates the
occupancy-driven load schedules for different types of building
under each scenario (see also Supplementary Note 7).
Heating and cooling may also be occupant-drivenoccupants
may change temperature set-points according to personal thermal
comfort and reduce heating/cooling in unused space. EnergyPlus
species heating and cooling set-points and requires a schedule
for ON/OFF operation. In all scenarios, we use the consistent set
of heating and cooling set-points provided in the DOE reference
buildings dataset. The differences in heating and cooling loads
between the scenarios is then driven by the number of occupants
and their appliance usages.
Urban energy prediction with mobile occupancy. Using the
previously described UBEM we now run simulations for the 1,266
modeled buildings under each of the three scenarios (DOE
reference, low-impact and high-impact), each with occupancy
distributions for μ=0, μ=0.5, and μ=1. We simulate one
Noon: residential 9 p.m.: residential
Noon: commercial 9 p.m.: commercial
Estimated number of people by building type
No data
0.5 Residential, = 0.0
Res. = 0.0
Comm. = 0.0
Res. = 0.5
Res. = 1.0
Comm. = 0.5
Comm. = 1.0
Commercial, = 0.0
Mixed, = 0.0
Residential, = 0.5
Commercial, = 0.5
Mixed, = 0.5
Residential, = 1.0
Commercial, = 1.0
Mixed, = 1.0
–10 0 10 100 1000
Mean hourly occ. difference (persons)
P(Mean hourly occ. difference)
Max. occupancy
Max. capacity
P( )
Max. occupancy
Max. capacity
Fig. 4 Building Occupancy results. aEstimated number of building occupants by building type for the mobile-inferred model for residential and commercial
buildings. bDistributions of the ratio of maximum observed occupancy to nominal building capacity for ~83,000 buildings in Boston. cDistribution of
average hourly occupancy difference between DOE reference and mobile-inferred occupancy for the 1,266 buildings in the energy model. Positive means
the DOE occupancy is greater while negative implies the opposite. Maps created using Google Maps, 2019 Google
6NATURE COMMUNICATIONS | (2019) 10:3736 | |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
representative day in each season using the Typical Meteor-
ological Year (see Methods) to get a predicted daily energy con-
sumption value for each building in each season. For the
reference model we also compare the predicted annual Energy
Use Intensity (EUI)the energy consumption per unit oor area
for each building to the national average value for the
equivalent building types in the Commercial Building Energy
Consumption Survey and the Residential Building Energy Con-
sumption Survey. Here, we nd the average difference of the
modeled EUI compared to the CBECS/RBECS averages for
the specic geographic region ranged between 5% and 20% for
the modeled building types with the simulated value always being
higher than the CBECS ref. 19.
Figure 5a, b shows the predicted distributions of daily
Energy Use Intensity (EUI) averaged over all seasons, grouped
into residential-only and commercial buildings, for each μ
value. Consumption is further split into gas (heating and hot
water) and electricity (cooling, equipment, and lighting)
estimates. We nd that the low-impact scenario predicts
increased gas use and decreased electricity use compared to the
DOE reference scenario, with both effects being larger in
commercial buildings. The high-impact scenario exaggerates
both of these effects in commercial buildings, although for
residential buildings gas use is decreased. The uncertainty from
the different occupancy distributions (as generated by the
range of μvalues) is also illustrated. We see that the differences
between the occupancy distributions are insignicant com-
pared to the difference between the high-impact and low-
impact scenarios.
In comparison to the DOE reference scenario, the low-impact
scenario predicts a median EUI increase of 0.6 ± 0.1% for
residential buildings and a decrease of 4.2 ± 0.1% for commercial
buildings. The high-impact scenario predicts median EUI
decreases of 14.9 ± 0.9% and 21.0 ± 0.4% for residential buildings
and commercial buildings, respectively. Figure 5c, d illustrates
hourly electricity consumption predictions for all scenarios
aggregated for all modeled buildings. Electrical load proles and
total winter gas usage are particularly important to grid operators
and utilities to plan peak electrical generation capacity and gas
storage respectively. We see that the prominence of the winter
electric peak is reduced by 4 MW in the low-impact scenario and
almost completely removed in the high-impact scenario. The
summer peak electricity loads are also predicted signicantly
lower, due both to reduced numbers of occupants and
heterogeneous occupancy. These effects are overemphasized in
the high-impact scenario. Total winter gas use is predicted ~9.9%
greater in the low-impact scenario while it is 10.4% greater in the
high-impact scenario (see Supplementary Tables 4 and 5,
Supplementary Note 8 and Supplementary Figs. 9 and 10).
The impact of energy efciency measures. Finally, we predict
the impact of two different generic energy efciency interven-
tions, that could be encouraged by energy policy (see Supple-
mentary Note 9). Firstly, we improve wall and roof insulation by
10% in comparison to the 2010 building code requirements for all
residential spaces, and secondly, we implement a 10% efciency
gain for equipment in commercial spaces (plug loads not
including light and heating or cooling). Figure 5e, f shows the
cd e f
P(Daily EUI)
Daily energy use intensity (kWhm–2)
Energy use (×103 kWh)
Energy use (×103 kWh)
Δ(Energy use) (%)
Δ (Energy use) (%)
0.05 0.15 0.25 0.35 0.45
High-impact: = 0 = 0.5
= 0.5 = 0
= 1
= 1
DOE ref:
0.05 0.15 0.25 0.35 0.45
30 Winter electric
DOE ref. High
Summer electric
Insultation upgrade Equipment upgrade
Elec. Gas Total
06 a.m. Noon 6 p.m. 6 a.m. Noon 6 p.m. Ref Low High Ref Low High
40 4.0
0.15 0.25 0.35 0.45
P(Daily EUI)
Daily energy use intensity (kWhm–2)
0.2 0.4 0.6
0.2 0.4 0.6
0.5 1.0 1.5
Fig. 5 Scenario energy consumption predictions. aDistribution of the daily EUI for all residential-only buildings for μ=0, μ=0.5, and μ=1 occupancy.
bDistribution of the daily EUI for all non-residential-only buildings. cHourly electricity use results for winter in each scenario (shading denotes the range
from the different μvalues). dHourly electricity use results for summer in each scenario. eMedian change in energy use (original minus upgrade) from
upgrading the insulation effectiveness in residential spaces under each scenario. Error bars illustrate the range of predictions for different μvalues.
fMedian change in energy use from upgrading the equipment efciency in commercial spaces under each scenario
NATURE COMMUNICATIONS | (2019) 10:3736 | | 7
Content courtesy of Springer Nature, terms of use apply. Rights reserved
median reduction for each efciency measure in each scenario.
The DOE reference scenario predicts upgrading insulation
decreases the median gas use by around 3.8% but increases
electricity by 0.1%, which results in a median decrease in total
energy use for residential buildings of ~2.3%. While the effects are
similar for the low-impact scenario, the high-impact gas saving
increases to 4.2%. Interestingly, electricity consumption decreases
by 0.1% in the high-impact scenario, having increased by 0.1% in
the DOE reference and low-impact scenarios. With upgraded
equipment efciency, Fig. 5f shows that electricity use is
decreased while gas use increases to compensate for the reduction
in the heat generated by equipment. The DOE reference scenario
predicts the median commercial building saves approximately
2.9% in electricity consumption, which is improved to 3.2% in the
low-impact scenario and to 4.4% in the high-impact scenario.
However, due to increased gas use, the median total energy
savings for commercial buildings are 1.2, 0.9, and 0.7% in the
DOE reference, low-impact and high-impact scenarios, respec-
tively (see Supplementary Tables 6 and 7). These results show
that knowledge of occupancy is crucial to accurately predict the
results of large-scale efciency improvements in building envel-
opes or systems.
In this paper, we have presented a framework for improving
urban scale building occupancy estimates. Our method extends
proven techniques for modeling urban mobility based on mobile
phone and census data and is the rst to estimate urban-scale
building occupancy with mobile phone data. We nd that typical
maximum daily occupancy in individual commercial buildings is
likely to be only 2030% of assumed capacity by building type.
Conversely, residential occupancy is highly neighborhood
dependent, with certain areas experiencing much higher occu-
pancy per unit of oor space than others. The differences between
our mobile phone based occupancy estimates and current occu-
pancy assumptions based on building types arise because current
methods treat buildings in isolation whereas our estimates con-
sider that occupants can visit multiple buildings, i.e., people leave
residential buildings to go to ofce buildings to work. Therefore
our estimates are normalized against the total urban population.
While the exact results are specic to the modeled region, the
method is easily portable to other locations due to the ubiquitous
nature of the data sources used.
These differences between the estimated mobile occupancy and
current building-based assumptions have strong implications for
urban energy modelling and predictions. In a state-of-the-art
UBEM of a mixed-use urban neighborhood in Boston, the mobile-
estimated occupancy predicted energy consumption that differed by
+1to15% for residential buildings and was reduced by 421%
for commercial buildings compared with reference building model.
This is highly consequential for urban energy policy development,
since policy makers rely on urban energy models to inform their
decisions and these decisions have widespread and long-term
impacts. For example, in the US, entire new neighborhoods are
designed with the US Green Building CouncilsLEED(Leadership
in Energy and Environmental Design) certication, which acts as a
de facto building standard in certain jurisdictions and for which
compliance is demonstrated through a simulated comparison
against a code compliant base version45. Furthermore, urban energy
models will be required to develop future sustainable solutions
specic to individual cities or neighborhoods52.
Given the large range between the low-impact and high-impact
of occupancy scenarios, our work underscores an urgent need to
understand how occupancy drives different load classes53,54
across the diverse set of building types found in cities. Our results
suggest that building heating and electrical loads may be sys-
tematically mispredicted due to incorrect occupancy assumptions,
especially in large commercial buildings. However, while our
results imply heating loads may be under-predicted in most
buildings due to lower-than-design occupancy, the opposite may
also be true if signicant portions of unoccupied indoor space are
allowed to cool below comfortable temperature levels. Deter-
mining whether or not this is the case will require new large-scale
datasets relating building occupancy and energy use. Further
uncertainty analysis of building parameters (i.e., range of heating/
cooling set-points, max hot water temperatures etc.) may also
prove informative in this regard, however without empirical data
on both the most important parameters (each building model
contains many thousands of parameters) and the typical ranges
encountered for the set of city building types modelled, it is
unclear how informative this may be. Accordingly, collating data
on the most inuential parameters and their corresponding
ranges for different building types would be a useful exercise. Our
analysis also highlights how the different occupancy patterns and
their assumed relationship with building loads implies different
effectiveness for building efciency measures. This is of particular
importance, since energy efciency incentives are vital in the real
world of sustainable urban planning with limited budgets.
Through the inclusion of heterogeneous mobile phone inferred
occupancy patterns, this work represents a signicant improve-
ment on current best practice occupancy estimates and an
important step towards bespoke urban building energy models. It
also has its set of limitations, including the lack of available real
occupancy data on a sufciently large scale for comparison.
Therefore, we expect that as higher resolution data relating to
urban mobility, building occupancy and building energy use
become available, the presented modeling framework may be
Finally, while our work has focused on improving urban
building energy models with the mobile-inferred occupancy
proles, our results are also relevant for building utilization. The
large discrepancy between estimated building occupancy and
actual capacity implies that building efciency may also be
improved through better space utilization, for example by tran-
sitioning residential spaces into commercial spaces at times of
low-residential occupancy. This is an interesting area of future
urban planning and is especially important due to the high pre-
dicted expansion rates of global urban centers, although it has
associated legislative and social barriers.
Mobile phone data and TimeGeo. The TimeGeo framework33 starts with the
CDRs of 1.92 million people in the Greater Boston Area. The data were collected by
AirSage for operational purposes for two mobile phone carriers. The location
coordinates are estimated by the data provider using standard triangulation algo-
rithms and have higher resolution (accuracy of 200300 meters) compared to
tower-based CDRs55.
TimeGeo extracts stay points from each individuals sequence of consecutive
mobile phone records, and separates commuters from non-commuters by
identifying users with work stay points. For each tract in the Boston metropolitan
area expansion factors are calculated for commuters and non-commuters using
census data. Based on the distributions of empirical individual mobility parameters
extracted from the active user data (0.78 million active users with homes inside the
census regionFig. 1a), the entire urban population movement is simulated. See33.
Building data for occupancy model. Building data are obtained from the city of
Boston, for 82,542 buildings. The geographic extent of this data is shown in Fig. 1b.
Buildings are classied into several tax classes including residential classes, com-
mercial classes, an industrial class and a mixed residential-commercial class (see
Supplementary Fig. 2 and Supplementary Note 2). For mixed-use residential-
commercial buildings, if the building is multi-oor we assume the rst oor is
commercial and if single-oored then we assume the oor area is split equally
between residential and commercial classes.
8NATURE COMMUNICATIONS | (2019) 10:3736 | |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Building opening hours and Per Capita Areas. We collect all the stays in each
census tract using outlines provided by the Massachusetts 2010 census (and the
latitude/longitude for each stay). For each building, we use the centroid and collect
all buildings in each tract. We calculate the building total oor area, assuming each
oor has equal area and the number of oors is the greatest integer of (building
height)/(oor height). We assume all oor heights are 10 feet. We also add a broad
classier marking each building as residential, commercial, and industrial, which
we use for stay compatibility.
To get building capacity, we need further information and we rely on a mapping
between the functional use of buildings and the PCA required for that use casesee
Supplementary Notes 24 and Supplementary Tables 1 and 2. Unfortunately, the set
of functional uses for which PCAs are available is not coincident with the Boston
buildings classication, nor is a credible translation possible, except for residential
uses. Therefore we obtain additional use information at high spatial resolution from
digital maps. We use the Google Places API Service and query for Points of Interest
(POIs) located throughout Boston in the proximity of each non-residential building.
We aggregate the results to a list of POIs near non-residential buildings,
discounting duplicate information. The information associated with each POI is
variable and can include category (i.e., restaurant, bar, bank etc), place name, location,
address, hours of operation, ratings, and reviews amongst others. Not all information
elds are always available. We then use the building outlines to nd POIs contained
within each building and subsequently assign opening hours and PCAs.
Residential buildings are always open. For each non-residential building, the
opening hours is the superset of the opening hours for all contained POIs. If there
is no opening hour information then we assign opening hours of 7am9pm
(commercial) or 9am5pm (industrial).
To assign PCAs, we map the different POI categories to a functional use for
which we have an estimate of PCA (see Supplementary Note 4). Residential
buildings are assigned 40 m2person1. For non-residential buildings, we rst assign
PCAs to the set of buildings containing POIsassigning the highest PCA for all
contained POIs. Second, for each building without a PCA assigned, a PCA is
randomly selected from the distribution of non-residential PCAs in that tract
(inferred by POIs). In this way, if there are a high number of restaurants in a tract,
an unmarked building would have a relatively higher probability of being allocated
the PCA of a restaurant.
Creating EnergyPlus building templates. EnergyPlus building models were cre-
ated using the UMI framework, as developed in ref. 10. Using this tool building
footprints were imported into Rhinoceros 3D CAD and extruded to the height
specied by the data. Constructions, systems, and load characteristics were assigned
to buildings using an appropriate template from the DOE reference buildings
dataset, and based on the use and age of the structure as reported in Bostons tax
assessment dataset (see Supplementary Note 6). EnergyPlus takes hourly occu-
pancy as an input, dened by a combination of a maximum occupant density per
unit oor area and a fractional hourly schedule. Additionally, inputs for occupant-
driven loads are also required (i.e., lighting, equipment and hot water), again
specied by peak values per unit area and a fractional schedule.
Manipulating EnergyPlus input les and running simulations. All adjustments
to the generated building models, including changes to the occupancy patterns
and implementing energy efciency measures, were made in Python using
Eppy56, a python package for scripting EnergyPlus input les. For the 1,266
modeled buildings EnergyPlus was run for one representative day for each
season. Weather data for the simulations was extracted from BostonsTypical
Meteorological Year (TMY) data climate le provided by the US DOE57,and
developed from weather information gathered at Bostons Logan Airport station
(~3 miles from the simulated site). The TMY data dene a typical yearsweather
in the region and are commonly used for annual energy simulations. However,
since this data is typical rather than extreme, it is not suitable for designing
building systems for worst-case scenarios. Hence, the energyplus weather les
(EPW les) also contain designations for the most extreme and most typical
weather-weeks in each season. Since we are interested in understanding the
average effect of occupancy on energy consumption and we only have a single
day of mobile-inferred occupancy, we select the mid-range day from each typical
week of weather data by season for our simulations. Therefore, each building was
simulated for the days 30-Jan (winter), 1-Apr (spring), 30-Jul (summer), 23-Oct
(autumn) under all scenarios.
Data availability
Data to run the whole analysis for a demonstration census tract are provided along with
the codesee Code Availability. Boston buildings data are available from the city of
Bostons data portal Analyze Boston ( Larger samples of the
data are available from the corresponding author upon reasonable request.
Code availability
Python scripts to run the analysis are available at
Mob_Building_Occ_Energy_Use.git, with stay point and building data provided for a
demonstration census tract. This includes the source code used to generate the analysis
Figures in the manuscript and supplementary material. EnergyPlus (also required) is
freely available from the US DOE ( The UMI urban modelling
interface used to create the energyplus building models is available from the sustainable
design lab at MIT (
Received: 22 October 2018 Accepted: 24 July 2019
1. U.S Energy Information Administration. Monthly energy review. https://www. (2017).
2. Cao, X., Dai, X. & Liu, J. Building energy-consumption status worldwide and
the state-of-the-art technologies for zero-energy buildings during the past
decade. Energy Build. 128, 198213 (2016).
3. Güneralp, B. et al. Global scenarios of urban density and its impacts on
building energy use through 2050. Proc. Natl Acad. Sci. USA 114, 89458950
4. International Energy Agency. Transition to sustainable buildings: Strategies
and opportunities to 2050. Tech. Rep., IEA.
freepublications/publication/Building2013_free.pdf (2013).
5. Kammen, D. M. & Sunter, D. A. City-integrated renewable energy for urban
sustainability. Science 352, 922928 (2016).
6. Qomi, M. J. A. et al. Data analytics for simplifying thermal efciency planning
in cities. J. R. Soc. Interface 13, 20150971 (2016).
7. Kontokosta, C. E. & Tull, C. A data-driven predictive model of city-scale
energy use in buildings. Appl. Energ. 197, 303317 (2017).
8. Hsu, D. How much information disclosure of building energy performance is
necessary? Energ. Policy 64, 263272 (2014).
9. Reinhart, C. F. & Davila, C. C. Urban building energy modelinga review of a
nascent eld. Build. Environ. 97, 196202 (2016).
10. Reinhart, C., Dogan, T., Jakubiec, J. A., Rakha, T. & Sang, A. Umi-an urban
simulation environment for building energy use, daylighting and walkability.
In Proc13th Conference of International Building Performance Simulation
Association,Chambery,France, 2013.
11. Hong, T., Chen, Y., Lee, S. H. & Piette, M. A. Citybes: A web-based platform
to support city-scale building energy efciency. Urban Computing.https:// (2016).
12. Walter, E., & Kämpf, J. H. A verication of CitySim results using the BESTEST
and monitored consumption values. Proceedings of the 2nd Building
Simulation Applications conference (No. CONF, pp. 215222). (Bozen-Bolzano
University Press 2015).
13. Kazas, G., Fabrizio, E. & Perino, M. Energy demand prole generation with
detailed time resolution at an urban district scale: A reference building
approach and case study. Appl. Energ. 193, 243262 (2017).
14. Nageler, P. et al. Novel validated method for gis based automated dynamic
urban building energy simulations. Energy 139, 142154 (2017).
15. Sousa, G., Jones, B. M., Mirzaei, P. A. & Robinson, D. An open-source
simulation platform to support the formulation of housing stock
decarbonisation strategies. Energy Build. 172, 459477 (2018).
16. Quan, S. J., Li, Q., Augenbroe, G., Brown, J., & Yang, P. P. J. Urban data and
building energy modeling: A GIS-based urban building energy modeling
system using the urban-EPC engine. Planning Support Systems and Smart
Cities (pp. 447469). (Springer 2015).
17. Chen, Y., Hong, T. & Piette, M. A. Automatic generation and simulation of
urban building energy models based on city datasets for city-scale building
retrot analysis. Appl. Energ. 205, 323335 (2017).
18. Frayssinet, L. et al. Modeling the heating and cooling energy demand of
urban buildings at city scale. Renew. Sustain. Energy Rev.81, 23182327
19. Davila, C. C., Reinhart, C. F. & Bemis, J. L. Modeling boston: A workow for
the efcient generation and maintenance of urban building energy models
from existing geospatial datasets. Energy 117, 237250 (2016).
20. Happle, G., Fonseca, J. A. & Schlueter, A. A review on occupant behavior in
urban building energy models. Energy Build. 174, 276292 (2018).
21. Francisco, A., Truong, H., Khosrowpour, A., Taylor, J. E. & Mohammadi, N.
Occupant perceptions of building information model-based energy
visualizations in eco-feedback systems. Appl. Energ. 221, 220228 (2018).
22. Baetens, R. & Saelens, D. Modelling uncertainty in district energy simulations
by stochastic residential occupant behaviour. J. Build. Perform. Simul. 9,
431447 (2016).
23. International Energy Agency. IEA-EBC Annex 66: Denition and Simulation
of Occupant Behavior in Buildings. (2018).
24. International Energy Agency. IEA-EBC Annex 79: Occupant-Centric Building
Design and Operation. (2018).
NATURE COMMUNICATIONS | (2019) 10:3736 | | 9
Content courtesy of Springer Nature, terms of use apply. Rights reserved
25. Page, J., Robinson, D., Morel, N. & Scartezzini, J.-L. A generalised stochastic
model for the simulation of occupant presence. Energy Build. 40,8398
26. Richardson, I., Thomson, M., Ineld, D. & Clifford, C. Domestic electricity
use: a high-resolution energy demand model. Energy Build. 42, 18781887
27. Chen, Y., Hong, T., & Luo, X. An agent-based stochastic occupancy simulator.
Building Simulation (vol. 11, 3749). (Tsinghua University Press 2018).
28. Jiang, S. et al. review of urban computing for mobile phone traces: current
methods, challenges and opportunities. Proceedings of the 2nd ACM SIGKDD
international workshop on Urban Computing (p. 2). (ACM 2013).
29. Wheatman, B., Noriega, A. & Pentland, A. Electricity Demand and Population
Dynamics Prediction from Mobile Phone Metadata. International Conference
on Social Computing, Behavioral-Cultural Modeling and Prediction and
Behavior Representation in Modeling and Simulation (pp. 196205). (Springer
30. Bogomolov, A. et al. Energy consumption prediction using people dynamics
derived from cellular network data. EPJ Data Sci. 5, 13 (2016).
31. Toole, J. L. et al. The path most traveled: travel demand estimation using big
data resources. Transp. Res. Part C Emerg. Technol. 58, 162177 (2015).
32. Alexander, L., Jiang, S., Murga, M. & González, M. C. Origindestination trips
by purpose and time of day inferred from mobile phone data. Transp. Res.
part c Emerg. Technol. 58, 240250 (2015).
33. Jiang, S. et al. The timegeo modeling framework for urban mobility without
travel surveys. Proc. Natl Acad. Sci. USA 113, E5370E5378 (2016).
34. Miranda, F. et al. Urban pulse: capturing the rhythm of cities. IEEE. Trans.
Vis. Comput. Graph. 23, 791800 (2017).
35. Corna, A., Fontana, L., Nacci, A. A. & Sciuto, D. Occupancy detection via
iBeacon on Android devices for smart building management. Proceedings of
the 2015 Design, Automation & Test in Europe Conference & Exhibition
(pp. 629-632). (EDA Consortium 2015).
36. Akkaya, K., Guvenc, I., Aygun, R., Pala, N. & Kadri, A. IoT-based occupancy
monitoring techniques for energy-efcient smart buildings. Wireless
Communications and Networking Conference Workshops (WCNCW), New
Orleans, LA, 2015, pp. 58-63. (IEEE 2015).
37. Chen, D., Barker, S., Subbaswamy, A., Irwin, D. & Shenoy, P. Non-intrusive
occupancy monitoring using smart meters. Proceedings of the 5th ACM
Workshop on Embedded Systems For Energy-Efcient Buildings (pp. 1-8).
(ACM 2013).
38. Kim, Y.-S., Heidarinejad, M., Dahlhausen, M. & Srebric, J. Building energy
model calibration with schedules derived from electricity use data. Appl.
Energ. 190, 9971007 (2017).
39. Shen, W. & Newsham, G. Smart phone based occupancy detection in ofce
buildings. 2016 IEEE 20th International Conference on Computer Supported
Cooperative Work in Design (CSCWD), Nanchang, 2016, pp. 632636. (IEEE
40. Wu, C., Thai, J., Yadlowsky, S., Pozdnoukhov, A. & Bayen, A. Cellpath: Fusion
of cellular and trafc sensor data for route ow estimation via convex
optimization. Transp. Res. Part C: Emerg. Technol. 59, 111128 (2015).
41. Çolak, S., Lima, A. & González, M. C. Understanding congested travel in
urban areas. Nat. Commun. 7, 10793 (2016).
42. US Department of Energy. Commercial Reference Buildings. https://energy.
gov/eere/buildings/commercial-reference-buildings (2018).
43. Crawley, D. B. et al. Energyplus: creating a new-generation building energy
simulation program. Energy Build. 33, 319331 (2001).
44. US-DOE Ofce of Energy Efciency and Renewable Energy. EnergyPlus V8.4
[software]. (2018).
45. ASHRAE. Ansi/ashrae/ies standard 90.1-2016 energy standard for buildings
except low-rise residential buildings. Tech. Rep., ASHRAE. https://www. (2016).
46. US Department of Transportation Federal Highway Administration. 2009
National Household Travel Survey. (2009).
47. Massachusetts Department of Transportation. Massachusetts travel survey.
Reports/TravelSurvey.aspx (2011).
48. Fujita, M. & Thisse, J.-F. Economics of Agglomeration: Cities, Industrial
Location, and Regional Growth, 2nd Edition. (Cambridge, UK: Cambridge
University Press 2013)
49. Ward, J. Hierarchical grouping to optimize an objective function. J. Am. Stat.
Assoc. 58, 144236 (1963).
50. Hong, T., Yan, D., DOca, S. & Chen, C.-f Ten questions concerning
occupant behavior in buildings: the big picture. Build. Environ. 114, 518530
51. Yan, D. et al. Occupant behavior modeling for building performance
simulation: current state and future challenges. Energy Build. 107, 264278
52. Ramaswami, A., Russell, A. G., Culligan, P. J., Sharma, K. R. & Kumar, E.
Meta-principles for developing smart, sustainable, and healthy cities. Science
352, 940943 (2016).
53. Khosrowpour, A. et al. A review of occupant energy feedback research:
opportunities for methodological fusion at the intersection of
experimentation, analytics, surveys and simulation. Appl. Energ. 218, 304316
54. Kim, Y.-S. & Srebric, J. Impact of occupancy rates on the building electricity
consumption in commercial buildings. Energy Build. 138, 591600 (2017).
55. Song, C., Koren, T., Wang, P. & Barabási, A.-L. Modelling the scaling
properties of human mobility. Nat. Phys. 6, 818 (2010).
56. Santosh, P. Eppy 0.5.48 [software]. (2018).
57. US Department of Energy. EnergyPlus weather data.
weather (2018).
This work was supported by grants from the Centre for Complex Engineering Systems at
MIT-KACST, the MIT Energy Initiative and Philips Lighting.
Author contributions
E.B., M.G., S.G. and J.K. conceived the research. E.B. and S.G. wrote the code for the
occupancy analysis. C.C.D. and C.R. constructed the urban energy model. E.B., C.C.D.
and C.R. developed the occupancy scenarios. E.B. performed the energy analysis. M.C.G.,
J.K. and C.R. supervised the research and provided guidance. E.B., C.C.D., C.R. and
M.C.G. wrote the paper. All authors approved the manuscript.
Additional information
Supplementary Information accompanies this paper at
Competing interests: The authors declare no competing interests.
Reprints and permission information is available online at
Peer review information:Nature Communications thanks Tianzhen Hong, Jelena
Srebric and Deborah A Sunter for their contribution to the peer review of this work. Peer
reviewer reports are available.
Publishers note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional afliations.
Open Access This article is licensed under a Creative Commons
Attribution 4.0 International License, which permits use, sharing,
adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the articles Creative Commons license, unless
indicated otherwise in a credit line to the material. If material is not included in the
articles Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this license, visit
© The Author(s) 2019
10 NATURE COMMUNICATIONS | (2019) 10:3736 | |
Content courtesy of Springer Nature, terms of use apply. Rights reserved
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
... The estimation of occupants' presence and activities is one of the primary sources of uncertainty in urban energy modelling; thus, it is essential to introduce spatiotemporal modeling of occupant behavior to achieve a reliable and accurate prediction for future urban energy use (Reinhart and Davilav 2015;Hong et al. 2018). Occupant-centric data collected on an urban-scale can provides information to model the complex occupant behaviors in UBEMs, which are not easily captured following traditional modelling approaches (Barbour et al. 2019). Previous studies have investigated occupant behaviors and energy consumption at an urban scale using different data sources. ...
... Previous studies have investigated occupant behaviors and energy consumption at an urban scale using different data sources. Recently, Barbour et al. (Barbour et al. 2019) derived urban scale building occupancy from 1.92 million call detail records (CDR) data from anonymous mobile phone users in the Greater Boston area. Compared to standard methods, using the data-driven occupancy profiles in the building energy modelling resulted in varying energy consumption by +1% to -15% for residential buildings and -4% to -21% for commercial buildings (Barbour et al. 2019). ...
... Recently, Barbour et al. (Barbour et al. 2019) derived urban scale building occupancy from 1.92 million call detail records (CDR) data from anonymous mobile phone users in the Greater Boston area. Compared to standard methods, using the data-driven occupancy profiles in the building energy modelling resulted in varying energy consumption by +1% to -15% for residential buildings and -4% to -21% for commercial buildings (Barbour et al. 2019). Jeifan et al. (Jeifan et al. 2018) used mobile position information acquired from the Chinese social media company (Tencent) to derive an occupancy schedule database for 60 buildings in Shanghai, China. ...
Conference Paper
Urban-scale energy simulation relies on the understanding of occupants’ presence in buildings and consequently in cities. Therefore, occupancy profiles (i.e., the relative number of occupants in a specific hour of the day) are usually used in the energy simulation on the city level. However, available occupancy standard profiles are incapable of considering the dynamic nature of occupancy schedules and any changes that occurred due to contextual changes (such as the dramatic increase in remote working last year). Therefore, the need for a scalable method to generate dynamic occupancy profiles for buildings is crucial. Moreover, the targeted method should allow for tracking the changes that occur in occupancy profiles due to external disruption such as pandemics. In this context, this study aims at using the emerging mobile positioning data to generate context-specific data-driven occupancy profiles for commercial and institutional buildings in New York City. The generated profiles were then compared versus ASHRAE standard profiles for each building category. Then, the occupancy profiles were clustered for each building category, using K-means clustering algorithm. Finally, the effect of COVID-19 pandemic on the peak points and shape of occupancy profiles was investigated. The results showed a significant difference between the data-driven and ASHRAE standard profiles. Additionally, a considerable variation in the shape and peak hours of the generated occupancy profile clusters was detected for some building categories. These results can be used to improve the accuracy of the urban-scale simulation models. Furthermore, they can provide a more precise evaluation of the occupant’s schedules and consequently the urban scale energy consumption before field implementation of the operational strategies.
... The framework considers both temporal and spatial choices of agents, which are modeled by parameters such as the weekly home-based tour number, the travel circadian rhythm, the dwell rate, and the burst rate (for temporal choices); and a rank-based exploration and preferential return (r-EPR) mechanism (for spatial choices). The required data of the TimeGeo-based approach is mainly from CDR data [135,137]. Barbour et al. [137] employed CDRs of 1.92 million mobile phones and the TimeGeo framework to estimate urban-scale building occupancy in Boston (the U.S.). ...
... The required data of the TimeGeo-based approach is mainly from CDR data [135,137]. Barbour et al. [137] employed CDRs of 1.92 million mobile phones and the TimeGeo framework to estimate urban-scale building occupancy in Boston (the U.S.). ...
Urban Building Energy Modeling (UBEM) is essential for urban energy-related applications. Its generation mainly requires four data inputs, including geometric data, non-geometric data, weather data, and validation and calibration data. A reliable UBEM depends on the quantity and accuracy of the data inputs. However, the lack of available data and the difficulty in determining stochastic data are two of the main barriers in the development of UBEM. To bridge the research gaps, this paper reviews appropriate acquisition approaches for four data inputs, learning from both building science and other disciplines such as geography, transportation and computer science. In addition, detailed evaluations are also conducted in each part of the study, and the performance of the approaches are discussed, as well as the availability and cost of the implemented data. Systematic discussion, multidisciplinary analysis and comprehensive evaluation are the highlights of this review.
... The under-occupancy of commercial office buildings relative to their design intent is a welldocumented phenomenon (e.g., [48][49][50]). When peak occupancies are below the design capacity, the minimum OA damper position can be lowered correspondingly. ...
Full-text available
Partial occupancy of commercial offices has become the norm in the wake of the COVID-19 pandemic. Given this, occupant-centric control (OCC), which adapt building systems based on occupants’ presence or preferences, offer an alternative to traditional control that assumes full occupancy. However, poor sequences of operation can degrade the benefits of OCC. This paper explores this interaction by examining energy data from two buildings – one with two control logic faults corrected and an occupancy-based ventilation OCC implemented in 2020, and one with traditional ventilation – from 2019 to 2020. Sequences that impacted implementation in the first building are discussed. Then, a calibrated energy model of the second building is developed to evaluate how occupancy-based ventilation alongside changes to the sequences of operation – namely supply air temperature (SAT) reset and economizer high limits – impacted energy use. The inclusion of OCC and improved sequences in the second building saved 30.6% and 9.6% of annual heating and cooling energy, respectively. Without an SAT reset, OCC saved 4.4% and 3.9% of heating and cooling, respectively, compared to 15.7% and 5.7% when an SAT was present. These results begin to characterize the relationship sequences of operation and OCC implementations have with one another in commercial offices.
... For example, land use classes and night time lights, derived from remote sensing techniques, are a common set of information that are used in population estimations [1,4,[23][24][25]. Further examples are many: household counts [4,6], telecommunication data [10,26,27], tax parcel information [28], and social media [29,30]. The large number of disparate information and wide range of data sources used in the analyses are united in predicting the number of people living in an area, but they do not do much beyond that despite the diversity of input data. ...
Full-text available
Mapping population distribution at a fine spatial scale is essential for urban studies and planning. Numerous studies, mainly supported by geospatial and statistical methods, have focused primarily on predicting population counts. However, estimating their socio-economic characteristics beyond population counts, such as average age, income, and gender ratio, remains unattended. We enhance traditional population estimation by predicting not only the number of residents in an area, but also their demographic characteristics: average age and the proportion of seniors. By implementing and comparing different machine learning techniques (Random Forest, Support Vector Machines, and Linear Regression) in administrative areas in Singapore, we investigate the use of point of interest (POI) and real estate data for this purpose. The developed regression model predicts the average age of residents in a neighbourhood with a mean error of about 1.5 years (the range of average resident age across Singaporean districts spans approx. 14 years). The results reveal that age patterns of residents can be predicted using real estate information rather than with amenities, which is in contrast to estimating population counts. Another contribution of our work in population estimation is the use of previously unexploited POI and real estate datasets for it, such as property transactions, year of construction, and flat types (number of rooms). Advancing the domain of population estimation, this study reveals the prospects of a small set of detailed and strong predictors that might have the potential of estimating other demographic characteristics such as income.
... 60 At the system level (i.e., AHU), maintaining an outdoor air damper at a constant minimum position of 10%-40% during the peak of the heating and cooling seasons is a common practice to maintain the minimum outdoor air requirement for acceptable IAQ for a fully occupied building. 61 While most buildings are operated based on the assumption of full occupancy, many field studies have provided evidence that actual occupancy of commercial buildings rarely exceeds 50% of the full occupancy (e.g., Refs [62][63][64]). This is similar to lighting, where several studies (e.g., Refs [65][66][67][68]) have shown that occupants' preferred illuminance setpoints (i.e., the threshold whereby occupants no longer desire additional lighting) are often below the thresholds specified in documents such as ASHRAE 90.1 69 or building codes, resulting in excessive artificial lighting and energy use. ...
Data-driven building operation and maintenance research such as metadata inference, fault detection and diagnosis, occupant-centric controls (OCCs), and non-invasive load monitoring have emerged (NILM) as independent domains of study. However, there are strong dependencies between these domains; for example, quality of metadata affects the usability of fault detection and diagnostics techniques. Further, faults in controls hardware and programs limit the performance of OCCs. To this end, a literature review was conducted to identify the dependencies between these domains of research. Additionally, real-world examples using operational data from three institutional buildings in Ottawa, Canada, were provided and discussed to demonstrate these dependencies. Finally, a holistic tool-agnostic workflow was introduced which suggested the implementation of operational energy efficiency measures in the following order to ensure their full potential: (1) improve metadata, (2) address faults, (3) implement OCCs, and (4) monitor enhanced key performance indicators (KPIs). The proposed workflow is intended to be comprehensive, reproducible, nonintrusive, and inexpensive to implement. Practical applications: Optimization of building operations has been emerging among energy management professionals as a relatively low-cost means to achieve energy efficiency and minimize occupants’ discomfort. To this end, this study introduces a tool-agnostic data-driven workflow to building energy management practitioners that can assist them in achieving increased energy efficiency. The proposed workflow recognizes the interdependency of the various domains of research which have historically been treated independently.
... As mapping features such as building footprints continues to be complex and time-consuming, they remain unmapped in much of the world. We postulate that GANmapper could be used as a solution to create maps that approximately reveal the urban form, which despite the synthetic nature of the data, may be found valuable by various use cases such as population estimation, urban morphology, energy, and climate simulations Wang et al., 2017;Yuan et al., 2019;Wang et al., 2020;Fleischmann et al., 2021;Barbour et al., 2019;Schug et al., 2021;Shang et al., 2021). For many of these applications, the exact geometry of each building is not essential, and such applications benefit from aggregated building data (e.g. total area covered by buildings), which our approach accomplishes well. ...
Full-text available
We present a new method to create spatial data using a generative adversarial network (GAN). Our contribution uses coarse and widely available geospatial data to create maps of less available features at the finer scale in the built environment , bypassing their traditional acquisition techniques (e.g. satellite imagery or land surveying). In the work, we employ land use data and road networks as input to generate building footprints and conduct experiments in 9 cities around the world. The method, which we implement in a tool we release openly, enables the translation of one geospatial dataset to another with high fidelity and morphological accuracy. It may be especially useful in locations missing detailed and high-resolution data and those that are mapped with uncertain or heterogeneous quality, such as much of OpenStreetMap. The quality of the results is influenced by the urban form and scale. In most cases, the experiments suggest promising performance as the method tends to truthfully indicate the locations, amount, and shape of buildings. The work has the potential to support several applications , such as energy, climate, and urban morphology studies in areas previously lacking required data or inpainting geospatial data in regions with incomplete data.
... However, the amount and availability of occupancy data depend on the source [43]. In particular, WiFi, Bluetooth, and GPS technologies have recently been used in mobile phone tracking methods [44][45][46]. Other works exploit the availability of environmental sensors inside buildings such as CO 2 , lighting or temperature, and BMS monitoring data [47,48]. ...
Full-text available
The energy consumption of cities is increasing fast due to growing global population and rapid urbanization. Urban Building Energy Models (UBEMs) are promising tools to simulate the energy demand of buildings under different urban scenarios. Nowadays, the major barriers to the effective use of UBEMs are the uncertainty related to their input parameters and the lack of good-quality, open energy consumption data. The latter make deterministic UBEM simulations unreliable, and calibration approaches unsuitable for most cities in the world. The present work proposes to combine physics-based UBEMs with Uncertainty and Sensitivity Analysis on the main input parameters using aggregated data on energy use from regional/national statistics. The proposed procedure selects the most influential input parameters and characterizes their uncertainty through Forward Uncertainty Analysis and Sensitivity Analysis to obtain stochastic load profiles for space heating and cooling. The method was first tested against hourly thermal power profiles metered on a heterogeneous sample of buildings in Verona (Italy). The average heating load profile obtained is significantly improved compared to deterministic, archetype-based simulations in terms of energy needs and peak loads. The overestimation of residential buildings peak load is reduced from 80% to 25%, and the deviation in the energy needs calculation drops from 18% to 10%. The proposed simulation procedure was then applied to a district of Milan (Italy), including more than 600 buildings, resulting in similar variations. Overall, the results demonstrate that considering the uncertainty of operational, geometrical and physical parameters is of the utmost importance to obtain reliable urban simulations.
... In other 2 studies, crowd positioning data were used to extract occupancy patterns in buildings [47,48]. However, in the research of building energy prediction, the state-of-the-art does not leverage any online data beyond manually collected and organized data on weekdays/weekends and national holidays. ...
Full-text available
In recent years, the availability of larger amounts of energy data and advanced machine learning algorithms has created a surge in building energy prediction research. However, one of the variables in energy prediction models, occupant behavior, is crucial for prediction performance but hard-to-measure or time-consuming to collect from each building. This study proposes an approach that utilizes the search volume of topics (e.g., education or Microsoft Excel) on the Google Trends platform as a proxy of occupant behavior and use of buildings. Linear correlations were first examined to explore the relationship between energy meter data and Google Trends search terms to infer building occupancy. Prediction errors before and after the inclusion of the trends of these terms were compared and analyzed based on the ASHRAE Great Energy Predictor III (GEPIII) competition dataset. The results show that highly correlated Google Trends data can effectively reduce the overall RMSLE error for a subset of the buildings to the level of the GEPIII competition’s top five winning teams’ performance. In particular, the RMSLE error reduction during public holidays and days with site-specific schedules are respectively reduced by 20–30% and 2–5%. These results show the potential of using Google Trends to improve energy prediction for a portion of the building stock by automatically identifying site-specific and holiday schedules.
Urban areas have become increasingly important for addressing carbon mitigation because of their significant contribution to global anthropogenic carbon emissions. Cities have been exploring effective actions to reduce their carbon impacts. These urban climate policies include addressing demand reduction (e.g., behavior nudges, and travel demand reduction) and reducing supply-side carbon impacts (e.g., waste-to-energy and local renewable energy adoption). During the past decade, the science of urban carbon accounting has advanced in response to these policy considerations. Four broad carbon accounting approaches (i.e., purely territorial carbon accounting, community-wide transboundary infrastructural supply chain carbon accounting, consumption-based carbon accounting, and total community-wide carbon accounting) have emerged at the city scale. This chapter summarizes the differences and overlaps of these accounting approaches, including their policy relevance and benchmarking metrics. It then discusses the value that each approach offers to urban climate policies, which has not been explained in previous literature. We emphasize that each approach is complete for its designed purposes and is valuable for different stakeholders (e.g., homes, businesses, industry, local government, etc.). In addition to explaining the value of four urban carbon approaches to individual cities, we discuss emerging efforts that quantify biogenic carbon from urban areas and account for all urban areas’ carbon emissions in a nation. This chapter provides a roadmap for urban policymakers choosing appropriate accounting approaches.
Due to rapid urbanisation and the significant contribution of cities to worldwide energy use and greenhouse gas emissions, urban energy system planning is growing more important. Urban building energy modelling (UBEM) draws increasing attention in the energy modelling field due to its inherent capacities for modelling entire cities or building stocks, and the potential of varying data inputs, approaches and applications. This review aims to identify best practices and improvements for UBEM applications by examining previous research, with a focus on the currently least established approaches. Different archetype development procedures are analysed for common problems, six main under-developed input approaches or parameters are identified, and applications for future scenario development are surveyed. By analysing previous studies in related fields, this paper provides an overview of gaps in the published research and possible additions to future UBEM projects that can help expanding the existing modelling procedures. Comprehensive human behaviour models with additional aspects beyond occupant presence are identified as a major point of interest. Further research on socio-economic parameters, such as household income and demographics, are also suggested to further improve modelling. This study also underlines the potential for utilising UBEM as a tool for evaluating future climate change scenarios.
Full-text available
Urban building energy models aspire to become key planning tools for the holistic optimization of buildings, urban design, and energy systems in neighborhoods and districts. The energy demand of buildings is largely influenced by the behavior of the occupants. The insufficient consideration of occupant behavior is one of the causes to the “performance gap” in buildings - the difference between the simulated and the actually observed energy consumption. On the urban-scale impacts of different occupant behavior modeling approaches onto the various purposes of urban building energy models are still largely unknown. Research shows that the inappropriate choice of occupant behavior model could result in oversized district energy systems, leading to over-investment and low operational efficiency. This work therefore reviews urban building energy models in terms of their occupant behavior modeling approaches. Three categories of approaches are established and discussed: (1) deterministic space-based approaches, (2) stochastic space-based approaches, and (3) stochastic person-based approaches. They are further assessed in terms of their strategy to consider diversity in occupant behavior. Stochastic models, especially stochastic person-based models, seem to be superior to deterministic models. However, there are no stochastic models available yet that can be used for case studies of mixed-districts, comprising buildings of various occupancy types. In the reviewed urban-scale approaches, only single-use type districts (residential or office) are modeled with stochastic techniques. However, people interact with various buildings on a daily basis. Their activities relate to their presence in different spaces at the urban-scale and to their use of appliances in those spaces. Their individual levels of comfort and behavioral patterns govern the control actions towards building systems. Therefore, a novel activity-based multi-agent approach for urban occupant behavior modeling is proposed as alternative to current approaches.
Full-text available
The modelling of whole urban districts requires an automated process to parameterize simulation tools. This paper presents a validated methodology for fully automated building modelling within urban districts based on publicly available data. Dynamic building models with detailed heating systems are created in the simulation environment IDA ICE. The method of data collecting and processing and result visualization in a geographical information system (GIS) and the data storage procedure in a PostgreSQL database is described in detail. The building simulation model is validated with consumption data available from 69 buildings of the town Gleisdorf (Austria). The results of the annual heating and domestic hot water demand show a good approximation to the measurement data with a mean deviation of -0.98 %. The urban simulation process was then extended to the whole community with its 1,945 buildings. This method helps to model and quantitatively describe current building stock in an efficient and timesaving way and enables to develop future smart energy systems, in which the buildings interact with the district heating networks, with limited effort.
Full-text available
Many computational approaches exist to estimate heating and cooling energy demand of buildings at city scale, but few existing models can explicitly consider every buildings of an urban area, and even less can address hourly -or less- energy demand. However, both aspects are critical for urban energy supply designers. Therefore, this paper gives an overview of city energy simulation models from the point of view of short energy dynamics, and reviews the related modeling techniques, which generally involve detailed approaches. Analysis highlights computational costs of such simulations as key issue to overcome towards reliable microsimulation of the power demand of urban areas. Relevant physical and mathematical simplifications as well as efficient numerical and computational techniques based on uncertainties analysis and error quantification should thus be implemented.
Housing Stock Energy Models (HSEMs) play a determinant role in the study of strategies to decarbonise the UK housing stock. Over the past three decades, a range of national HSEMs have been developed and deployed to estimate the energy demand of the 27 million dwellings that comprise the UK housing stock. However, despite ongoing improvements in the fidelity of both modelling strategies and calibration data, their longevity, usability and reliability have been compromised by a lack of modularity and openness in the underlying algorithms and calibration data sets. To address these shortfalls, a new open and modular platform for the dynamic simulation of national (in the first instance, the UK) housing stocks has been developed—the Energy Hub (EnHub). This paper describes EnHub's architecture, its underlying rationale, the datasets it employs, its current scope, examples of its application, and plans for its further development. In this we pay particular attention to the systematic identification of housing archetypes and their corresponding attributes to represent the stock. The scenarios we analyse in our initial applications of EnHub, based on these archetypes, focus on improvements to housing fabric, the efficiency of lights and appliances and of the related behavioural practices of their users. In this we consider a perfect uptake scenario and a conditional (partial) uptake scenario. Results from the disaggregation of energy use throughout the stock for the baseline case and for our scenarios indicate that improvements to solid wall and loft thermal performance are particularly effective, as are reductions in infiltration. Improvements in lights and appliances and reductions in the intensity of their use are largely counteracted by increases in heating demand. Housing archetypes that offer the greatest potential savings are apartments and detached dwellings, owing to their relatively high surface area to volume ratio; in particular for pre-1919 and inter-war epochs.
While technology advancements are improving the energy efficiency of buildings, occupant behavior remains a critical factor in ensuring the effectiveness of such enhancements. To this end, numerous eco-feedback systems have been developed to reduce building energy use through influencing occupants' behaviors during building operations. Information representation is a critical component in eco-feedback systems, affecting the users' interpretation, engagement, and motivation to reduce energy consumption. Many studies have focused on using different charts and technical units or abstract and artistic visualizations to represent energy consumption. However, the effectiveness of such techniques varies across studies. Recent research emphasizes the need to integrate information representation strategies that balance numeric and aesthetic appeal. Concurrently, studies have called for increased adoption of a Building Information Model (BIM) during a building's operations phase to improve facility management. In this paper, we introduce a new eco-feedback information representation method that combines numeric and aesthetic appeal through leveraging spatial and color-coding techniques in BIM. The BIM-integrated energy visualization approach developed in this paper uses the Revit Application Program Interface (API) and allows users to visualize and compare energy consumption values in 2D and 3D views of a multi-family building through a color-coding scheme in an as-built BIM. The method is validated through a user survey that quantitatively and qualitatively assesses the proposed 2D and 3D BIM eco-feedback compared to more traditional bar chart based eco-feedback. Our findings suggest that 2D spatial, color-coded eco-feedback provides the optimal information representation, as it is easy to understand, while evoking engaging and motivating responses from users. This study advances our understanding of eco-feedback information representation while expanding BIM applications during building operations. These are important steps to address the human dimension of energy efficiency in the built environment.
Occupants are integral elements of a building ecosystem and their behavior can have a substantial impact on energy consumption in buildings. A wide range of energy feedback programs have been developed to make energy consumption more visible and interpretable to occupants and help them learn how to control and save energy. In this paper, we conduct a critical review of the literature related to energy feedback and identify four key methodological approaches to designing and studying energy feedback programs: experiments, analytics, surveys and simulation. Our meta-analysis reveals five research gaps and opportunities for future methodological fusion at the intersection between such approaches, including the analytics-survey, experiments-analytics, experiments-analytics-surveys, simulation-experiments and analytics-simulation interfaces. Future research at these crucial interfaces could provide the deeper understanding necessary to develop energy feedback programs that yield substantial and persistent energy savings.
Occupancy has significant impacts on building performance. However, in current building performance simulation programs, occupancy inputs are static and lack diversity, contributing to discrepancies between the simulated and actual building performance. This paper presents an Occupancy Simulator that simulates the stochastic behavior of occupant presence and movement in buildings, capturing the spatial and temporal occupancy diversity. Each occupant and each space in the building are explicitly simulated as an agent with their profiles of stochastic behaviors. The occupancy behaviors are represented with three types of models: (1) the status transition events (e.g., first arrival in office) simulated with Reinhart’s LIGHTSWITCH-2002 model, (2) the random moving events (e.g., from one office to another) simulated with Wang’s homogeneous Markov chain model, and (3) the meeting events simulated with a new stochastic model. A hierarchical data model was developed for the Occupancy Simulator, which reduces the amount of data input by using the concepts of occupant types and space types. Finally, a case study of a small office building is presented to demonstrate the use of the Simulator to generate detailed annual sub-hourly occupant schedules for individual spaces and the whole building. The Simulator is a web application freely available to the public and capable of performing a detailed stochastic simulation of occupant presence and movement in buildings. Future work includes enhancements in the meeting event model, consideration of personal absent days, verification and validation of the simulated occupancy results, and expansion for use with residential buildings.
Many cities across the United States have turned to building energy disclosure (or benchmarking) laws to encourage transparency in energy efficiency markets and to support sustainability and carbon reduction plans. In addition to direct peer-to-peer comparisons, the benchmarking data published under these laws have been used as a tool by researchers and policy-makers to study the distribution and determinants of energy use in large buildings. However, these policies only cover a small subset of the building stock in a given city, and thus capture only a fraction of energy use at the urban scale. To overcome this limitation, we develop a predictive model of energy use at the building, district, and city scales using training data from energy disclosure policies and predictors from widely-available property and zoning information. We use statistical models to predict the energy use of 1.1 million buildings in New York City using the physical, spatial, and energy use attributes of a subset derived from 23,000 buildings required to report energy use data each year. Linear regression (OLS), random forest, and support vector regression (SVM) algorithms are fit to the city's energy benchmarking data and then used to predict electricity and natural gas use for every property in the city. Model accuracy is assessed and validated at the building level and zip code level using actual consumption data from calendar year 2014. We find the OLS model performs best when generalizing to the City as a whole, and SVM results in the lowest mean absolute error for predicting energy use within the LL84 sample. Our median predicted electric energy use intensity for office buildings is 71.2 kbtu/sf and for residential buildings is 31.2 kbtu/sf with mean absolute log accuracy ratio of 0.17. Building age is found to be a significant predictor of energy use, with newer buildings (particularly those built since 1991) found to have higher consumption levels than those constructed before 1930. We also find higher electric consumption in office and retail buildings, although the sign is reversed for natural gas. In general, larger buildings use less energy per square foot, while taller buildings with more stories, controlling for floor area, use more energy per square foot. Attached buildings – those with adjacent buildings and a shared party wall – are found to have lower natural gas use intensity. The results demonstrate that electricity consumption can be reliably predicted using actual data from a relatively small subset of buildings, while natural gas use presents a more complicated problem given the bimodal distribution of consumption and infrastructure availability.
The energy demand in urban areas has increased dramatically over the last few decades because of the intensive urbanization that has taken place. Because of this, the European Union has introduced directives pertaining to the energy performance of buildings and has identified demand side management as a significant tool for the optimization of the energy demand. Demand side management, together with thermal energy storage and renewable energy technologies, have mainly been studied so far at a building scale. In order to study and define potential demand side management strategies at an urban scale, an integrated urban scale assessment needs to be conducted.