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

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

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,

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

cond

k · A · (T

− T

)

(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

& T

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

conv

= h · A · (T

− T

air

) (2)

where Q

conv

is the conductive heat transfer rate, h is

the convective heat transfer coefﬁcient, T

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

cond

− T

)

cond

k · A

(3)

and the one associated with convection

conv

− T

air

)

conv

h · A

(4)

Furthermore, heat storage capacity of walls and

rooms can be represented using heat capacitance of

capacity

C = m · c

(5)

where m is the mass and c

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

/kg·K. This is a justiﬁed

assumption since this value is 1.006

/kg·K and

1.0007

/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

and that on the side of room 2 is h

. In

an equivalent thermal circuit, there are three different

nodes with potentials T

, T

and T

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

1,2

= ρ

· v

1,2

· c

(6)

= ρ

· A · L · c

(7)

where ρ

represents the density and c

the speciﬁc

heat capacity of the wall, ρ

, c

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

h;1,2

1,2

· A

(8)

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

/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

/K·m

with an area of 6.43 m

. 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

/kg·K

1000 1000 1000 1000

Speciﬁc resis-

tivity

m·K

4.76 4.76 4.76 1.49

Density

700 700 700 1600

Thickness m

0.1 0.1 0.1 0.27

Area m

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

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