JouleSense: A Simulation based Platform for Proactive Feedback on
Building Occupants’ Energy Use
Georgios Lilis
, Shubham Bansal
and Maher Kayal
Electronics Laboratory,
´
Ecole Polytechnique F
´
ed
´
erale de Lausanne, Lausanne, Switzerland
Keywords:
Building Management System, Feedback Systems, Energy Awareness, Energy Efficiency, Behavioral Change,
Building Simulation.
Abstract:
A significant amount of energy in the buildings can be saved by inducing efficient occupant behavior. The
occupant’s awareness tools that have been shown to be effective in achieving energy efficiency gains depend
on various computational and estimation algorithms. This paper proposes an energy feedback scheme that
relies on a model based, building thermal simulation in order to identify the areas for efficiency improvement.
By leveraging the specific mathematical formulation of those models and a dedicated open-source solver,
improved computational speed, reduced cost and enhanced interoperability is obtained. This combined with
the integration into a building management system (BMS), permits real-time sensing and feedback. Unlike
similar studies, this work’s outcome allows the creation of the energy awareness tools that rely solely on
validated thermal model simulation, thus increasing their accuracy and potential in the future smart buildings.
1 INTRODUCTION
As the urban population has continued to increase in
the recent years, the sustainable urban development
initiatives have motivated the research efforts for re-
ducing the energy use in the building sector. Broadly,
there are two major approaches for reducing the en-
ergy use in buildings.
The most prominent one is the passive approach.
It focuses firstly on the improvement of the building
thermal envelop by incorporating improved isolation
and thermal storage material like the phase chang-
ing ones. Secondly, it promotes the wide adoption of
energy efficient appliances through citizen awareness
campaigns.
The second approach calls for active involvement
towards energy efficiency through the deployment of
smart building infrastructure. In this approach, ubiq-
uitous computing and embedded electronics play a
significant role by providing fine grained energy mon-
itoring and actuation capabilities. There are two
strategies in the active approach. In the first one, the
smart building automation technologies are deployed
to achieve optimal energy use while maintaining a
comfortable indoor environment without human in-
tervention. The other focuses on bringing user in the
*
The authors contributed equally to this work
loop for improving energy efficiency by facilitating
user awareness through targeted feedback on his con-
sumption (Mattern et al., 2010). This study focuses
on the development of a platform based on the latter
approach.
A significant amount of energy in the buildings
can be saved by inducing efficient occupant’s behav-
ior (Yu Zhun Jerry et al., 2011). Direct feedback to
building occupants through the use of in-house dis-
plays in the form of real time energy consumption
data has been shown to reduce energy use by up to
20% (Wood and Newborough, 2003). These savings
are a result of two major factors: the high dependence
of energy consumption on the usage patterns of the
occupants and the fact that decision to invest relies
solely on their willingness (Darby, 2010). Thus, ef-
ficiency gains can be realized by inducing behavioral
changes on people through appropriate feedback on
their energy consumption. In fact, studies have shown
that feedback to drive people towards energy efficient
behavior can be effective if it is frequent, uses an in-
teractive element and offers multiple options to the
occupant (Fischer, 2008).
At the heart of all the energy feedback based tools
lies a computational algorithm, the complexity of
which varies greatly depending on the approach uti-
lized, the nature of feedback provided and the avail-
able ubiquitous computing at the occupants’ living
Lilis, G., Bansal, S. and Kayal, M.
JouleSense: A Simulation based Platform for Proactive Feedback on Building Occupants’ Energy Use.
In Proceedings of the 5th International Conference on Smart Cities and Green ICT Systems (SMARTGREENS 2016), pages 279-285
ISBN: 978-989-758-184-7
Copyright
c
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
279
spaces (Spagnolli et al., 2011). Recent years have
seen the rise of data science based tools to deliver
personalized actionable energy reports for the occu-
pants in order to help them save energy (Zeifman,
2012; Chen and Cook, 2012; Birt et al., 2012). These
techniques mostly deploy disaggregation algorithms
on smart metering data in order to achieve appliance
level breakdown of energy consumption.
However, these techniques necessitate the exis-
tence of a usually large input dataset. Moreover, they
frequently involve complex calculations in order to
extract various patterns from this data, thereby re-
quiring considerable computational power and time,
nowhere near the capabilities of current embedded
electronics. Therefore, a key limitation of these so-
lutions is that such information is provided after a
significant gap of time and after the energy has been
consumed.
This paper presents an integrated simulation based
platform for providing proactive energy savings rec-
ommendations to building occupants with regard to
their heating equipment by leveraging the power of
Internet of Things (IoT) enabled sensing technolo-
gies, validated thermal models and a custom, opti-
mized simulation engine.
2 BACKGROUND THEORY
2.1 Choice of Modeling Approach
The ability of our integrated tool to deliver accurate
feedback rests on the underlying thermal model. The
proposed system needs to be tailored to the physi-
cal properties of the building in question and hence
should be able to capture the interactions between
physically connected spaces in the building. This pa-
per represents such a thermal model of the building
using a network of resistances and capacitances. A
typical building is made up of ceilings, floors, fa-
cade and internal walls as well as windows. All these
different elements can both store heat and transfer it
through various mechanisms. Apart from these ele-
ments, room air and other mass (ex. furniture) also
participate in the above-mentioned processes. So a
useful representation is to model the heat storage us-
ing capacitors and the heat transmission using resis-
tors. This work is built on the well-studied and proved
lumped capacitance method. (Maasoumy et al., 2011;
Fraisse et al., 2002). The choice of this modeling ap-
proach has been motivated by the following consider-
ations:
1. The resulting model should be descriptive enough
to capture all the relevant dynamics to give reli-
able and accurate results. For this, it was neces-
sary to model each room and wall with at least one
node.
2. It should have reasonable data needs and be com-
putationally efficient to allow for near real time
applications.
3. Finally, it should be dynamically customizable for
various buildings with minimal overhead.
2.2 Developing an Electrical Network
An equivalent electrical network has been developed
in order to represent the thermal processes in the
building. For this, each node is assigned to every
room and wall (if the wall has multiple layers then an
equal number of nodes can be assigned to the wall),
which is then connected to the ground via a capacitor,
C.
Heat transfer in a typical building takes place
through the three processes: conduction, convection
and radiation. Heat conduction across walls under
steady state condition can be described by
Q
cond
=
k · A · (T
2
T
1
)
L
(1)
where Q
cond
is the conductive heat transfer rate, k is
the thermal conductivity, A and L is the area and the
thickness of the wall accordingly, with T
1
& T
2
the
temperatures on the two sides of the wall. Convective
heat exchange also takes place from the surface of the
walls and the room air. This rate of heat transfer is
given by
Q
conv
= h · A · (T
s
T
air
) (2)
where Q
conv
is the conductive heat transfer rate, h is
the convective heat transfer coefficient, T
s
is the sur-
face temperature and T
air
is the temperature of the
surrounding air . In addition to these, heat transfer
also takes place via radiation exchange that occurs
between the internal surfaces of the wall, between fa-
cades surfaces and the sky and irradiation from the
sun. The heat exchange between the internal sur-
faces of the walls is neglected. This is justified since
walls of rooms are almost at the same temperature and
therefore net heat exchange between them can be ne-
glected. Further, long-wave radiation exchange with
the sky can be modeled using a combined convective
and radiative heat transfer coefficients for the exter-
nal surfaces as has been proposed in (Gyalistras and
Gwerder, 2009). Heat gain from solar radiation can
be modeled as direct heat inputs to room air and wall
surfaces.
All the above mentioned heat transfer mecha-
nisms, can now be represented using an electric anal-
ogy. In such a model, voltage source plays the role
SMARTGREENS 2016 - 5th International Conference on Smart Cities and Green ICT Systems
280
Figure 1: Electrical equivalent circuit to represent thermal
processes in an internal wall.
of the temperature of a building element or room,
whereas current represents the heat flow. Since re-
sistance is defined as the ratio of potential difference
over current, the resistance associated with conduc-
tion is
R
cond
=
(T
2
T
1
)
Q
cond
=
L
k · A
(3)
and the one associated with convection
R
conv
=
(T
s
T
air
)
Q
conv
=
1
h · A
(4)
Furthermore, heat storage capacity of walls and
rooms can be represented using heat capacitance of
capacity
C = m · c
p
(5)
where m is the mass and c
p
is the specific heat capac-
ity.
In addition, external and internal heat gains can be
easily added to our model. Internal heat gains and ra-
diators can be modeled as direct power inputs to the
room node. This basically translates into an appropri-
ate current source at the room nodes. For a detailed
description for modeling other heat gains, the reader
can refer to (Lehmann et al., 2013).
It is important to note that in deriving this model,
the following assumptions were made:
1. Heat transfer across the walls has been assumed
to take place perpendicular to the surface. Thus,
there is no variation in temperature over a surface.
2. Spatial variations in the temperature of the room
have been ignored; therefore, one node is suffi-
cient to represent a complete room.
3. The heat capacity of room air has been assumed
to constant at 1.007
kJ
/kg·K. This is a justified
assumption since this value is 1.006
kJ
/kg·K and
1.0007
kJ
/kg·K at 250 K and 300K accordingly.
We demonstrate the modeling procedure using a
test case shown in Fig. 1. In this example, there is an
internal wall of area A, thickness L and thermal con-
ductivity k. The heat transfer coefficient on the side
of room 1 is h
1
and that on the side of room 2 is h
2
. In
an equivalent thermal circuit, there are three different
nodes with potentials T
1
, T
2
and T
w
that correspond to
the temperatures of air in both the rooms and the wall
respectively. Note that the node for the wall temper-
ature has been assigned to the centerline of the walls.
These nodes are connected to the ground via the ca-
pacitors
C
1,2
= ρ
a
· v
1,2
· c
a
(6)
C
w
= ρ
w
· A · L · c
w
(7)
where ρ
w
represents the density and c
w
the specific
heat capacity of the wall, ρ
a
, c
a
are the respective ones
for air and v
1,2
the volume of each room. Eq. 6 rep-
resents the heat storage capacity of air in both rooms
and Eq. 7 corresponds to the heat storage capacity of
the wall. The heat transfer across the wall has been
modeled using the resistances
R
h;1,2
=
1
h
1,2
· A
(8)
R
w
=
L
k · A
(9)
The Eq. 8 represents the convective thermal resis-
tance and the Eq. 9 corresponds to the conductive
thermal resistance. This resistance of the wall has
been split across the centerline resulting into two ther-
mal resistances of
R
w
/2.
An important advantage of using the aforemen-
tioned approach is that all its parameters have phys-
ical interpretations. This direct relation allows us to
examine the effect of changing any parameter in a
physical building by changing the relevant parameter
in the electrical network. Additionally, this permits
our proposed tool to provide energy savings strategies
to the building occupants.
3 SYSTEM ARCHITECTURE
However, the aforementioned simulation is computa-
tionally heavy and the simulators commonly used in
the literature are built for use in specialized mathe-
matical software like Matlab
®
. As a result, the con-
trol and automation modules built around those simu-
lation cores are also written in the same software lan-
guage (Sturzenegger et al., 2014). The problem with
the use of such expensive proprietary software pack-
ages is that the tools developed in such environments
get restricted to just lab level research projects and do
not achieve commercial adoption.
JouleSense: A Simulation based Platform for Proactive Feedback on Building Occupants’ Energy Use
281
Figure 2: Architecture of the platform developed in this work.
This study presents an alternative approach for
providing the same thermal simulation outputs in
terms of accuracy but with increased performance and
usability. The dedicated numerical solver deployed in
the literature, is replaced by a circuit simulation en-
gine. Since the problem and model formulation scru-
tinized in Section 2 fits excellently with the purpose
of a circuit simulation engine, an increase in perfor-
mance in the numerical operations required for those
models is expected along-with a reduction in overall
cost due to the free availability of the software.
The integrated tool presented in this paper is com-
posed mainly of the following four discrete agents:
the weather related estimation agent, the building
management system (BMS), the simulation engine
and the feedback agent. Fig. 2 shows the complete
architecture of the platform with all the agents inter-
connected.
The BMS developed in (Lilis et al., 2015) is a
hybrid platform which integrates the sensing and ac-
tuation end devices with building structure data in
a common environment. A Representational State
Transfer (REST) Application Programming Interface
(API) provides accessibility interface to all the dy-
namic (real-time sensing) and static (building struc-
ture) data. In addition, the API provides the means
of action through the installed actuators. The sens-
ing capabilities include among other, temperature, hu-
midity, luminosity, presence and individual powers of
electrical loads. By leveraging the advancements in
embedded electronics and networks they provide in-
creased granularity in indoor living space sensing and
acting.
The weather estimator provides the necessary fu-
ture environmental data of the neighboring simulated
zones for use in the thermal simulation model. In
order to generate the relevant outputs it takes into
consideration the historic values of the zones and the
weather prediction from external sources. Using an
experimental solar heat gains model, it is possible to
estimate within an acceptable error the future temper-
ature data-points of the neighbor zones.
The feedback agent generates intelligent insights
and recommendations for the occupant by analyzing
the data provided by the simulation engine. As an in-
put, it receives the predicted time series temperature
data from the simulation engine, and runs appropriate
analysis on it. Its task is to provide the high abstrac-
tion level outputs that could be leveraged by the user
devices and in-house displays in order to provide the
desired user awareness. The feedback agent is com-
pletely decoupled from the inner-workings of the sim-
ulation engine and any intermediate steps. In addi-
tion, it is totally independent of the particular struc-
ture of the building and the smart sensing and actu-
ation infrastructure present in it. This is highly im-
portant since it facilitates the creation or reuse of uni-
versal energy awareness applications that have been
already proposed or implemented in the literature.
Therefore, by keeping the compatibility of the sim-
ulation engines output with current literature solution
the potential impact grows beyond the limits of this
study.
Finally, the core of this work focuses on the de-
sign and validation of the simulation engine as seen
in Fig. 2. This simulation engine comprises of a ded-
icated circuit simulator SPICE, a coordinator agent
and a software utility to convert output of the solver
into usable format. The coordinator agent firstly re-
trieves data from BMS structure database (DB) in or-
der to create the resistance-capacitance thermal model
of the building and automatically synthesizes all its
parameters from the building structure data as has
been described in Section 2. Subsequently, it gath-
ers the environmental data from the weather estima-
tor and real-time DB of BMS and creates the dynamic
controllable sources of the circuit. The output at this
stage is a netlist file compatible with the spice based
solvers which describes the circuit to be simulated in
full detail. In the next stage, the coordinator agent
SMARTGREENS 2016 - 5th International Conference on Smart Cities and Green ICT Systems
282
Figure 3: Schematic of the hypothetical room used for val-
idation along with the equivalent R-C network representa-
tion.
launches an LTspice (Linear Technology, 2015) con-
sole instance and supplies the generated netlist file to
perform circuit simulation. Once the simulation ter-
minates, the coordinator agent invokes the conversion
utility to convert back the circuit related raw outputs
to the physically meaningful ones. Eventually, the
feedback agent pulls these outputs to run its analysis
and generate insights for the occupants.
Therefore, with the help of the four agents devel-
oped in this work, our platform, JouleSense, is able
to provide proactive energy savings recommendations
to the occupants. To use the platform in a new build-
ing, the building structure has to be provided through
the BMS interface together with the necessary sen-
sors and actuators. Once the system is ready, the oc-
cupant only interacts with the platform through the
mobile interface by using the feedback agent and by
receiving the suggestions. By the time a new request
is generated for the feedback agent for the updated
predicted thermal dynamics, the simulator engine is
invoked and the entire aforementioned process is ex-
ecuted.
4 VALIDATION
In order to validate our model, the authors performed
a comparison of the results of the model with that
of state-of-the-art Matlab based building resistance-
capacitance modeling (BRCM) tool (Sturzenegger
et al., 2014). This tool has been experimentally vali-
dated extensively over several months for model pre-
dictive control on a real and fully operating office
building in the OptiControl-II project (Gwerder and
Gyalistras, 2013). Thus, it provides a good basis to
validate the accuracy of the proposed tool.
For this, a hypothetical test room, Zone 4, has
been defined as shown in Fig. 3. Two other rooms
and a corridor surround this room. Further, for the
sake of simplicity floor and ceilings are assumed to
have adiabatic boundary conditions. However, they
can be also simulated if need be, by two additional
circuit branches of capacitances and resistances using
the building material and structure specs of the floor
and ceiling.
The temperature profiles of these rooms and cor-
ridor along-with the outside temperature have been
used as boundary conditions for the corresponding
walls. This time series temperature data was collected
from the sensors deployed in one of the EPFL cam-
pus buildings for the month of September 2015. Fur-
ther, the room is considered to be unoccupied, with no
furniture and no deliberate ventilation. The material
properties used for the simulation are described in the
Table 1. A window was placed on wall 4 which con-
nected the zone under consideration to the external
environment. The U-Value of the window was taken
to be 0.51
W
/K·m
2
with an area of 6.43 m
2
. Further-
more, a radiator has been assumed to be present in
the room which is modeled as a current source to the
room node.
The results of the simulation performed have been
shown in the Fig. 4 that demonstrates the comparable
accuracy of our platform with respect to the state-of-
the-art tool.
Table 1: Building material and structure specs.
Wall
W1 W2 W3 W4
Specific Heat
Capacity
J
/kg·K
1000 1000 1000 1000
Specific resis-
tivity
m·K
/W
4.76 4.76 4.76 1.49
Density
kg
/m
3
700 700 700 1600
Thickness m
0.1 0.1 0.1 0.27
Area m
2
13.40 13.40 9.19 9.19
The combination of the free circuit solver with
the integrated BMS provides a low cost, yet accurate
environment for developing third party applications
to realize energy savings through targeted occupant
JouleSense: A Simulation based Platform for Proactive Feedback on Building Occupants’ Energy Use
283
Figure 4: Comparative accuracy of this work’s solver vs
state-of-the-art Matlab based tool.
feedback. These applications can leverage the feed-
back agent analysis and the near-real time low com-
putationally requiring thermal simulation engine.
A specific agent, called time-to-target-
temperature has been developed. It leverages
the ability of this tool and demonstrates its usage.
This agent provides an estimate of the time taken
by a specific building zone to achieve the desired
temperature. Currently, a mobile application is used
as a frontend to a desired temperature set-point from
the user. The feedback agent invokes the simulator
engine through a RESTful API and receives a time
series of predicted temperature data. It then looks
up the time to reach the desired temperature in
the time series and displays the output to the user
through the application. A prototype screen-shot of
the application is visible in Fig. 5. The execution
time of the entire process is in the order of 1 sec
and hence fast enough for full scale implementation.
This agent can also be leveraged by third party
applications to generate various insights for the
user. For instance, the information of the time it
takes to achieve the desired temperature, can help
occupants save energy since they tend to set higher
set-point temperatures believing that it leads to faster
heating (Gupta et al., 2009). This inevitably leads to
energy waste if the temperature is not turned down.
Another application envisioned in this paper is an
energy savings recommendation engine. It invokes
the simulator agent multiple times and finds the most
optimum control set-points and actuator positions. It
achieves that through sequential thermal simulations
for all the different cases and by calculating the
energy consumption in each case with the help of
time-to-target-temperature agent. The most efficient
configuration is recommended to the occupant over
the mobile application.
Figure 5: The initial prototype of the time-to-temperature
user awareness application integrated with openBMS both
in steady state and in operation.
5 CONCLUSIONS
As future work, we envision the porting of the open
source of SPICE to ARM architecture in order to in-
tegrate our optimized awareness tool in low power
embedded electronics, thus, eliminating the need for
cloud computing.
In this work, an integrated simulation based
framework is presented. It allows the development
of applications for near real time feedback to build-
ing occupants. This approach is validated by com-
parison with state-of-the-art tools in accuracy, speed
and usability. Its capabilities are highlighted by the
development of the feedback agent: time-to-target-
temperature. We argue that the strength of our ap-
proach lies in the flexibility, affordability and usabil-
ity aspects of this platform; facilitating the creation of
an energy awareness application ecosystem encourag-
ing efficient occupant behavior.
REFERENCES
Birt, B. J., Newsham, G. R., Beausoleil-Morrison, I., Arm-
strong, M. M., Saldanha, N., and Rowlands, I. H.
(2012). Disaggregating categories of electrical energy
end-use from whole-house hourly data. Energy and
Buildings, 50:93–102.
Chen, C. and Cook, D. J. (2012). Behavior-based home en-
ergy prediction. Proceedings - 8th International Con-
ference on Intelligent Environments, IE 2012, pages
57–63.
Darby, S. (2010). Smart metering: what potential for house-
holder engagement? Building Research & Informa-
tion, 38(5):442–457.
Fischer, C. (2008). Feedback on household electricity con-
sumption: A tool for saving energy? Energy Effi-
ciency, 1(1):79–104.
Fraisse, G., Viardot, C., Lafabrie, O., and Achard, G.
(2002). Development of a simplified and accurate
SMARTGREENS 2016 - 5th International Conference on Smart Cities and Green ICT Systems
284
building model based on electrical analogy. Energy
and Buildings, 34:1017–1031.
Gupta, M., Intille, S. S., and Larson, K. (2009). Adding
GPS-control to traditional thermostats: An explo-
ration of potential energy savings and design chal-
lenges. Lecture Notes in Computer Science (including
subseries Lecture Notes in Artificial Intelligence and
Lecture Notes in Bioinformatics), 5538 LNCS:95–
114.
Gwerder, M. and Gyalistras, D. (2013). Final Report :
Use of Weather And Occupancy Forecasts For Opti-
mal Building Climate Control Part II : Demonstration
( OptiControl-II ). Technical Report September, ETH
Z
¨
urich.
Gyalistras, D. and Gwerder, M. (2009). Use of weather and
occupancy forecasts for optimal building climate con-
trol (OptiControl): Two years progress report. Tech-
nical Report September, ETH, Zurich.
Lehmann, B., Gyalistras, D., Gwerder, M., Wirth, K., and
Carl, S. (2013). Intermediate complexity model for
Model Predictive Control of Integrated Room Au-
tomation. Energy and Buildings, 58:250–262.
Lilis, G., Conus, G., and Kayal, M. (2015). A Distributed,
Event-driven Building Management Platform on Web
Technologies. In 1st International Conference on
Event-Based Control, Communication, and Signal
Processing, Krakow.
Linear Technology (2015). LTspice.
Maasoumy, M., Pinto, A., and Sangiovanni-Vincentelli, A.
(2011). Model-Based Hierarchical Optimal Control
Design for HVAC Systems. ASME 2011 Dynamic Sys-
tems and Control Conference and Bath/ASME Sympo-
sium on Fluid Power and Motion Control, Volume 1,
pages 271–278.
Mattern, F., Staake, T., and Weiss, M. (2010). ICT for
green How Computers Can Help Us to Conserve En-
ergy. Proceedings of the 1st International Conference
on Energy-Efficient Computing and Networking - e-
Energy ’10, page 1.
Spagnolli, A., Corradi, N., Gamberini, L., Hoggan, E.,
Jacucci, G., Katzeff, C., Broms, L., and Jonsson, L.
(2011). Eco-feedback on the go: Motivating energy
awareness. Computer, 44(5):38–45.
Sturzenegger, D., Gyalistras, D., Semeraro, V., Morari, M.,
and Smith, R. S. (2014). BRCM Matlab Toolbox :
Model Generation for Model Predictive Building Con-
trol. American Control Conference.
Wood, G. and Newborough, M. (2003). Dynamic energy-
consumption indicators for domestic appliances: En-
vironment, behaviour and design. Energy and Build-
ings, 35(8):821–841.
Yu Zhun Jerry, Z. J., Haghighat, F., Fung, B. C. M., Morof-
sky, E., and Yoshino, H. (2011). A methodology for
identifying and improving occupant behavior in resi-
dential buildings. Energy, 36(11):6596–6608.
Zeifman, M. (2012). Disaggregation of home energy dis-
play data using probabilistic approach. IEEE Trans-
actions on Consumer Electronics, 58(1):23–31.
JouleSense: A Simulation based Platform for Proactive Feedback on Building Occupants’ Energy Use
285