Aggregated Performance and Qualitative Modeling Based

Smart Thermal Control

Afef Denguir

1,2

, François Trousset

and Jacky Montmain

LGI2P, Ecole des Mines d’Ales, 69 avenue Parc scientifique Georges Besse, Nîmes, France

LIRMM, Université de Montpellier 2, 161 rue Ada, Montpellier, France

Keywords: Energy Optimization, Smart Thermal Control, Thermal Comfort, Preference Model, Choquet Integral, Multi

Attribute Utility Theory, Utility Functions, Qualitative Modeling, Approximate Reasoning, Online

Learning, Preference Learning.

Abstract: In order to ensure thermal energy efficiency and follow government’s thermal guidance, more flexible and

efficient buildings’ thermal controls are required. This paper focuses on proposing an efficient, scalable,

reusable, and data weak dependent smart thermal control approach based on an aggregated performance and

imprecise knowledge of buildings’ thermal specificities. Its main principle is to bypass data unavailability

and quantitative models identification issues and to ensure an immediate thermal enhancement. For this, we

propose, first, an aggregated performance based smart thermal control in order to identify relevant thermal

setpoints. An extended thermal qualitative model is then introduced to guarantee an efficient achievement of

the identified thermal setpoints. Uncertainty about how relevant a thermal control is for a given thermal

situation is thus reduced using online and preference based learnings.

1 INTRODUCTION

Buildings’ energy efficiency has been widely

discussed in literature and supported by industrial

applications. In fact, buildings are responsible for

more than 40% of total energy consumption in

Europe (EP&C, 2012) which has led to restricted

energy policies for buildings’ energy control. As

47% of buildings’ energy consumption is used for

space heating and cooling (PNNL, 2012), buildings’

thermal control has particularly become an

important target in order to reduce buildings’ energy

consumption. Studies on smart thermal control are

thus relevant and are facing, nowadays, new

industrial challenges. The RIDER (Research for IT

Driven Energy Efficiency) project is one of recent

researches on smart thermal control which focuses

on the final solution deployment properties. It

considers the scalability and reusability of the

control solution which lead to a large application

area (i.e., form buildings to neighbor-hoods) and

deployment costs saving (i.e., neither specific

studies, nor particular information, are required for

the deployment) of the final solution. Moreover, the

RIDER project deals with data availability issues.

Indeed, it has to ensure as efficient as possible

thermal control whenever sufficient data are

available or not. For instance, a new-build that has

no historical data can immediately get advantage

from the RIDER solution without proceeding by the

new-build thermal behavior’ study or gathering

learning data. This work is part of the RIDER

project and proposes a new complete thermal

enhancement approach satisfying RIDER’s

deployment expectations. The proposed thermal

enhancement approach provides recommendations

starting from thermal setpoints to controls achieving

them. This paper explains our overall smart thermal

control and is organized as follows: first, some smart

thermal control related works are discussed. Section

3 explains our thermal enhancement approach

denoted by RIDER STC (RIDER’s Smart Thermal

Control) which fulfills RIDER’s deployment

expectations. Conclusions are finally presented.

2 RELATED WORKS

Most buildings’ thermal controls are simple and aim

to maintain the overall thermal state around a

conventional operational point (i.e., indoor tempera-

Denguir A., Trousset F. and Montmain J..

Aggregated Performance and Qualitative Modeling Based Smart Thermal Control.

DOI: 10.5220/0005063300630076

In Proceedings of the 11th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2014), pages 63-76

ISBN: 978-989-758-039-0

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

ture setpoints). Among well-known thermal

regulator, we name TOR (Tout-Ou-Rien), P

(Proportional action), PI (Proportional Integral

action), PID (Proportional Integral Derivative

action) control based regulators. P, PI, and PID

based regulators implement manual, automatic,

adaptive, and identification techniques in order to

define the regulator’s parameters. For instance,

(Zaheer-uddin, 2004) proposes a neural network

based approach to update PID control’ parameters.

Yet, adaptive and individualized thermal control

remains not supported by conventional thermal

regulators which requires an extra thermal control

level (Nassif, 2005); commonly denoted by the

smart thermal control. As buildings’ thermal

processes are usually characterized by a slow

dynamics and a high inertia, anticipating measurable

disturbances related to building’s usage, weather

conditions, and energy price variations, may lead to

a significant thermal control’ efficiency

enhancement (i.e., decreasing thermal energy

consumption, increasing occupants’ thermal

comfort). Hence, the smart thermal control is more

than a simple room’s temperature control. In order to

ensure the smart thermal control, additional

information on building’s usage (e.g., the requested

thermal comfort, the building’s occupancy profile,

etc.), climatic conditions (e.g., weather forecasts,

sunshine, etc.), and energy prices variations, need to

be introduced in thermal behavior models which

may quickly transform the smart thermal control

into a complex problem. To solve the complex

problem that the smart thermal control is, several

approaches based either on predictive control

techniques or advanced control techniques have

been proposed. For instance, we name Homes,

OptiControl (Oldewurtel, 2010), and the IC-

Berkeley’s energy storage (Ma, 2009) projects that

propose a predictive control based smart thermal

control. Applied to the thermal context, the

predictive control considers socio-economic

objectives such as minimizing energy consumption

and maximizing thermal comfort (Ma, 2009). It is

based on a mathematical thermal modeling which

considers precise quantitative physical aspects of the

thermal behavior and explains why mathematical

models are also called by white type models. In the

thermal context, mathematical modeling is

supported by software such as TRNSYS,

EnergyPlus, SPARK, and SIMBAD which offer a

convenient environment for detailed mathematical

thermal modeling and thermal simulation. Thus,

they can be directly deployed to ensure building’s

smart thermal control. Obviously, the more detailed

the thermal mathematical model is, the more

efficient the smart control would be. However,

detailed mathematical design requires expertise, as

well as, specific and precise quantitative data and

knowledge on buildings’ thermal behavior which

makes it intricate. Moreover these specific and

precise quantitative data are not commonly available

for most of buildings which has led to the grey type

modeling. This last tries to reduce the complexity of

mathematical modeling by introducing identification

(respectively estimation) techniques in order to

complete lacking parameters (respectively input

data). For instance, in order to build his monthly

energy consumption forecasting model, (White,

1996) has based his thermal modeling on monthly

average predicted outdoor temperature rather than

detailed forecasts. Identification techniques have

been also applied in order to automatically learn

thermal parameters such as (Wang, 2006) that has

used a genetic algorithm to identify thermal envelop’

parameters. Once all thermal parameters/inputs are

well identified/estimated, the grey type modeling

can lead to an efficient and accurate smart thermal

control. However, building’s thermal parameters can

only be identified through specific thermal tests also

known as building’s thermal excitement tests. These

tests add an extra deployment constraint to the smart

thermal control solution. In fact, depending on

building’s usage, some thermal tests are not

conservable which makes building’s thermal

parameters identification impossible. Although white

and grey models are efficient and accurate for the

smart thermal control, they do not fulfill RIDER’s

deployment expectations. In fact, focusing on the

accuracy of physical behavior decreases the

scalability of white and grey modeling. Collecting/

operating specific thermal information/tests entails

extra costs each time that the smart thermal control

solution needs to be deployed which means that the

thermal enhancement solution is not reusable and

cannot be immediately operated unless these

information/tests are available/allowed. Based on

those observations, Artificial Intelligence (AI)

techniques have been introduced 20 years ago in

order to ensure the advanced thermal control. They

provide a simple, efficient and adaptive smart

thermal control without requiring any a priori know-

ledge on thermal physical behavior. AI’ learning

techniques have been massively applied in order to

learn quantitative thermal models also known as

black type thermal models. (Kalogirous, 2000) have

proposed an ANN (Ant Neural Network) based

approach in order to learn building’s thermal

behavior. Therefore, ANN based smart thermal

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

control becomes possible and can answer specific

questions such as when is the best moment to restart

a heating system after an inoccupation period

(Yang, 2003). Thus, a well-trained ANN thermal

model could lead to similar accuracy as white and

grey thermal models. (Aydinalp, 2002) has shown

that for cooling’ energy cost prediction ANN based

modeling is more efficient than the grey based one.

The supervised leaning technique SVM (Support

Vector Machine) has been recently applied in the

thermal control area. It has been used for large scale

smart thermal control (Dong, 2005) and HVAC

(Heating Ventilation and Air-Conditioning) systems

control (Li, 2009). (Li, 2009) has proven that the

SVM based smart thermal control is more efficient

than the ANN based one. Yet, the main advantage of

SVM based smart thermal control is that only little

information are required for the model training

compared to the ANN based one. Training-data are

usually collected through onsite measurements,

surveys, and available documentations. Data pre-

treatment and post-treatment are, hence, requested in

order to improve the model efficiency (Li, 2009).

Therefore, significant computation loads and

efficient training-data are required to learn a black

type quantitative thermal model. However, under

real application conditions, thermal data are subject

to uncertainty (e.g., whether windows were opened

or not), imprecision (e.g., depends on sensors’

precision) and incompleteness. Therefore, a black

model based smart thermal control does not totally

meet the RIDER’s deployment expectations. In fact,

the availability of sufficient and efficient training-

data is an important factor to determine the smart

thermal control solution deployment potential.

Considering real application conditions, qualitative

modeling has been introduced in the thermal control.

It is based on the relevance of physical behavior

representation which makes it useful to understand

complicated physical phenomena. In fact, variables’

variation ranges are usually reduced to symbolic sets

(i.e., {negative, null, positive}) and qualitative

simulations are operated to compute the

recommended control option (Kupiers, 1986). In

literature, fuzzy thermal rules have been applied for

heating systems’ control and building’s temperature

regulation (Dounis, 1995). A review on thermal

fuzzy controller can be found in (Singh, 2006).

Qualitative model based smart thermal control has

particularly focused on thermal comfort regulation

(Calvio, 2004). Fuzzy predictive control has been

also introduced in the smart thermal control

(Terziyska, 2006). By focusing on the relevance

rather than precision, the complexity of the smart

thermal control is reduced through the qualitative

thermal modeling. Indeed, simple thermal control

rules expressed by the Energy Manager may appear

sufficient to ensure a smart thermal control which

entails no constraint on thermal data availability.

Thanks to its data weak dependency property, the

qualitative modeling meets all RIDER deployment

expectations in order to develop a scalable, reusable,

and data weak dependent smart thermal control

solution. However, ambiguities and lack of accuracy

may negatively affect the qualitative modeling

efficiency and longevity for a continuous thermal

control enhancement purpose.

3 RIDER STC APPROACH

In order to ensure an efficient, scalable, reusable,

and data weak dependent smart thermal control, we

first propose to focus our thermal enhancement

approach on thermal setpoints optimization rather

than efficiently reaching them. For this, we propose

an aggregated performance based reasoning which,

unlike the behavioral based reasoning, fulfills

RIDER deployment expectations. An aggregated

performance allows having an overall assessment on

the process. It is built through an aggregation

function defined over elementary performances.

These last come from the process’ output evaluation.

An aggregated performance corresponds to the

analytic formalization of user’s preferences.

Occupants’ social and preferential lines are then

considered once an aggregated performance is used

in the thermal enhancement reasoning. Moreover,

unlike behavioral models, the deployment of the

aggregated performance based enhancement solution

does not require any beforehand knowledge on the

thermal process which guarantees an immediate

improvement of the thermal control. In order to

evaluate the expected gain of thermal setpoints’

optimization, it becomes unavoidable to deal with

thermal behavioral models. For this, we propose an

extended qualitative model based smart thermal

control approach to efficiently reach the optimized

setpoints. Well-known qualitative enhancement

techniques have been used in our extended thermal

qualitative model. These techniques were proposed a

long time ago by (Williams, 1989), (Kuipers, 1986)

and others, such as (Dubois, 1989), in order to

improve qualitative models efficiency and reduce

their ambiguities. A survey is proposed in (MQ&D,

1995). For a better understanding, the RIDER STC

is explained on one building thermal scale but still

easily adaptable for larger thermal scales. For this,

AggregatedPerformanceandQualitativeModelingBasedSmartThermalControl

the thermal comfort is considered as the aggregated

performance used for setpoints optimization. To

simplify the optimization problem solving, we

introduced a new thermal comfort model denoted by

CIPPD

(Choquet Integral Predicted Percentage

Dissatisfied) which is a MAUT (Multi Attribute

Utility Theory) version of the

PPD

thermal comfort

standard (NF EN ISO 7730, 2006). The reason

behind using such formalism is explained in the next

section. Then, in order to efficiently reach the

optimized thermal setpoints, we introduced a

building scale’ Extended Qualitative Model (EQM).

Time-related information and available quantitative

observations have been used in order to improve the

EQM reliability and accuracy. Moreover, simplified

and generalized thermal behaviors have been

considered for the thermal control qualitative

modeling which is, also, recognized as a substantial

qualitative enhancement technique. Hence, the EQM

allows the abstraction of thermal specificities while

maintaining a sufficiently relevant representation for

thermal enhancement purposes. An EQM based

approximate reasoning can thus be generalized for

larger and various thermal scales and specificities.

Furthermore, the RIDER STC approach does not

either requires any particular setting data or

important computation loads to be deployed.

In order to ensure a continuous thermal control

enhancement, RIDER STC is operated for every

new thermal control situation. This last is triggered

for every new thermal context and objective. For

instance, whenever a thermal context variable (i.e.,

sunshine, humidity, etc.) significantly changes, a

new thermal control situation is created in order to

adapt the current thermal control. The same goes

true when occupants’ thermal comfort objectives

change (i.e., vacancy periods, personnel change,

preference change, etc.). The RIDER STC is then an

iterative approach which tries to improve

continuously the thermal performances. Figure 1

displays an overview of one enhancement iteration

of RIDER STC approach. According to thermal

context and thermal comfort objectives, the

CIPPD

based control compute new optimized setpoints that

are considered by the EQM based control as the new

thermal objectives. Depending on the available

quantitative data (i.e., historical data), the EQM

based control suggests quantitative

recommendations to the existing thermal control

system (e.g., an HVAC system in Figure 1) which

computes thermal equipment’s command laws. The

operated thermal control is then evaluated by the

RIDER STC (i.e., check how much thermal

expectations have been satisfied by the thermal

control) and saved in RIDER’s Database (DB).

Figure 1: Overview of RIDER STC iterative approach.

Figure 1 highlights, as well, the middleware

function of the RIDER STC. In fact, the RIDER

STC does not substitute any of the existing thermal

regulations but tries to continuously point the correct

energy amount (the minimum energy ensuring

thermal constraints) from the energy providers to

thermal regulators. For this, relevant

recommendations on thermal setpoints, heating/

cooling starting time, heating/cooling loads, and so,

ensure a personalized and efficient energy usage

depending on building/room’s thermal specificities.

3.1 CIPPD based Thermal Control

As discussed in section 2, today’s smart thermal

control is only conceived through behavioral thermal

process modeling. Considering the thermal comfort

performance rather than building’s thermal behavior

was inspired by bioclimatic architectures. In fact, in

such buildings’ architecture the economic (energy

consumption) and social (comfort) lines are

intimately related. The building’s location and

orientation are considered in the architecture design

in order to take advantage of naturally existing

climate. For instance, solar radiation and natural air

flow are used to, respectively, provide a natural

space heating and cooling. Thus, these natural

climate elements contribute to maintain occupants’

thermal comfort and also reduce thermal energy

consumption. Since the thermal comfort is a

complex multidimensional concept defined mainly

by the indoor temperature but also humidity, radiant

temperature, and air flow, we assume that thermal

comfort’ achievement could be delegated to

attributes other than the indoor temperature.

Therefore, achieving the expected thermal comfort

may be less costly once most of the thermal comfort

attributes are considered in the smart thermal

control. For instance, in winter time, adapting the

thermal control w.r.t. significant solar radiation

fluctuations contributes not only to maintain the

requested thermal comfort but also to make the most

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

of solar radiation in order to ensure some free

heating needs. Hence, during the day, the indoor

temperature setpoint could be adjusted depending on

solar radiation prediction which may potentially

contribute to reduce buildings’ energy consumption.

Basing our smart thermal control on the thermal

comfort performance consists then in identifying,

from that last, thermal setpoints satisfying the

expected occupants’ comfort and leading to as less

costly as possible thermal energy consumption. For

this, we introduce the

CIPPD

thermal model which

matches RIDER’s deployment requirements and

simplifies its optimization purpose.

3.1.1 CIPPD Model Motivations

Despite thermal comfort inherent subjectivity,

standards have attempted to evaluate thermal

sensations. The most popular standards are

ASHRAE standard 55 and ISO 7730 that,

respectively, implement Gagge’s and Fanger’s

thermal comfort models. Gagge’s model defines

thermal conditions for which at least 80% of

individuals would feel comfortable from a thermal

point of view (Gagge, 1986). Fanger’s model

introduces the

PMV

(Predicted Mean Vote) and

PPD

(Predicted Percentage Dissatisfied) indexes

(Fanger, 1967). The

PMV

index corresponds to the

mean thermal sensation vote expressed on Fanger’s

scale (i.e., the scale range is defined in [-3,3] which

corresponds to human thermal sensation from cold

to hot and where the null value refers to the neutral

thermal sensation). It is defined through 4 thermal

condition variables (i.e., indoor air temperature Ta,

air humidity Hy, air velocity Va, and mean radiant

air temperature Tr) and 2 human parameters (i.e.,

metabolic rate Me and cloth insolation index Ci).

The

PPD

index is based on the

PMV

one and

indicates the percentage of thermal dissatisfied

persons (1). From a thermal point of view, a person

is considered not satisfied when his/her

PMV

index

belongs to [-3,-2]



[2,3]. Both of the thermal

comfort standards are statistical models. They are

evaluated in different world parts by asking

thousands of people about their thermal sensation in

different thermal conditions.



0.03353* 0.2179*

100 95

PMV PMV

PPD e





(1)

Using statistical thermal comfort models for the

smart thermal control meets RIDER’s deployment

expectations. In fact, since these models are the

result of surveys, they do not require any adaptation

to ensure the smart thermal control of any building

that belongs to the survey’s scope. However, the

statistical based thermal comfort is usually complex

and does not help the optimization purpose that is

meant for in our RIDER STC. The complexity of a

statistical thermal comfort based control may then

significantly increase which entails efficiency

problems. Moreover, Gagge’s and Fanger’s models

deal with the thermal comfort concept as a physical

phenomenon. Although the body’s temperature is

related to thermodynamic interactions (i.e.,

convection, radiation, evaporation, conduction),

thermal sensation depends on people preferences

and their sociocultural backgrounds. Therefore

interaction between thermal comfort’ attributes are

preferential ones rather than physical which justifies

a preference based comfort model. Preference

modeling



is a central topic in the Multi Criteria

Decision Analysis (MCDA) and measurement

theory. Usually, it comes down to find a real-valued

overall utility function

:uX 

that verifies (2)

where

corresponds to the alternative set. For

multidimensional alternatives, the Krantz’s

decomposable model (Krantz, 1971) is widely

studied. It implements the MAUT theory which

consists on assessing any measurement as a

satisfaction degree in the [0,1] scale where 0 refers

to the worst alternative and 1 to the best one.

Measurements are thus made commensurate and

interpretable (Fishburn, 1970). Accordingly, the

Krantz’s preference model is built through utility

functions

uX [0,1] defined over each attribute

, where

X corresponds to the attribute

measurement scale, and an aggregation function



where



refers to this latter’s parameters (3).

,()()

yXxy ux uy





(2)

Krantz’s preference model seems quite

interesting for the RIDER STC optimization

purpose. In fact, it is commonly known that for



verifying both the independence and weak

separability properties,



is strictly increasing.

Thus, comonotony between

and

u s holds on

X .

This property is useful to identify simple and

interpretable thermal comfort adjustment rules and

simplifies the RIDER STC’s thermal comfort

adjustment. For instance, thermal comfort may be

improved when humidity rate increases for one

given ambient temperature, whereas it can be

disturbed for another one. The coexistence of such

rules makes difficult for the Energy Manager to

AggregatedPerformanceandQualitativeModelingBasedSmartThermalControl

decide about attribute variations in order to adjust

occupants’ thermal control. Hence, Krantz’s

preference model would greatly simplify the thermal

comfort rules design.

111

( ,..., ) ( ( ),..., ( ))

nnn

ux x F u x u x





(3)

To identify

u and



through the one-

dimensional measurement scales

, the weak

difference separability property has to be fulfilled;

otherwise the multi-dimensional scale

has to be

considered. Hence, the MCDA has introduced

interview based approaches in order to identify

and



such as MACBETH which allows identifying

them over the decision maker’ preferences w.r.t.

alternatives defined on

. However, proceeding by

the interview based approaches to identify the

thermal comfort model adds an extra deployment

constraint to our RIDER STC solution. In fact,

occupants need to be interviewed about their thermal

preferences in every building which breaks RIDER’s

reusability requirement. Therefore, we introduce the

CIPPD

thermal comfort model where

and



are

identified from the

PD index. Basing the

u and



identification on a statistical thermal comfort

model enables the RIDER STC deployment in any

building that belongs to the statistical study scope.

The

PD choice is motivated by RIDER’s market

location which is the EU market. The ISO standard

has released an EU

PD version which fits well

with RIDER’s potential customers. For the

identification into Krantz’s decomposable model,

the Choquet fuzzy integral



aggregation function

seems to be the most appropriate model. Namely,

fuzzy integrals provide adequate models to capture

relative importance of attributes but also preferential

interactions among them (Grabisch, 1997). It then

allows emphasizing preference relationships among

PD attributes and their relative contribution to the

thermal comfort achievement. Since the

PD index

is built over a cardinal scale and has a symmetry

property regarding the neutral thermal sensation (

PMV



), the

PPD

representation on a cardinal

positive scale seems to be sufficient which again

justifies the Choquet Integral. Moreover, the



has

linearity property by simplex which goes

accordingly with RIDER STC thermal comfort

based control.

3.1.2 CIPPD Identification

In order to identify the

CIPPD

model, (Labreuche,

2011) approach has been adapted to the

context. In fact, when



is a Choquet Integral,

Labreuche has proposed an original approach to

compute both

u and



without any

commensurateness assumption. Thus, before

proceeding by the identification process, MAUT and

Labreuche assumptions have to be checked. Let

consider

the

CIPPD

thermal attributes’ set built

from

PD ’s thermal condition variables (

, and

) and human parameters (

and

Independency, weak separability and monotony

assumption have then to be verified among

‘s

attributes. In next points, we develop and explain

these latter assumptions verification in order to

approximate the

PD into a Choquet Integral C



 Independency Verification

Structural interactions, such as physical ones,

that may jointly influence the comfort overall utility

are not tolerated in our

CIPPD

model. Yet, physical

interactions entailed by

and

parameters

exist. In fact,

and

do not convey a

preferential point w.r.t. thermal conditions. They

rather illustrate the body energy contribution and

clothing insolation in the convective and radiative

physical phenomena which make physical

interactions with

, and Tr attributes

obvious. Moreover, for simplification reasons, the

radiative phenomena, entailed by

Tr , has been

considered as convective ones in the

PD model.

This simplification leads to physical interaction

between

Tr and

. Therefore, only

, and

y attributes can be considered in the

CIPPD

preference model. Hence,

{, , }NTaVaHy

corresponds to the

CIPPD

attributes’ set. The

CIPPD

can thus be identified for given values of

and Tr .

Since it is not conceivable to identify a

CIPPD

model for every possible

and Tr values,

we restricted our study to administrative buildings

where occupants have a sedentary activity rate (

1, 2

emet



) and similar clothing habits: pant and

shirt (

0,7Ci clo



). However, radiant temperature

Tr is necessary to evaluate thermal comfort and

adjust it accordingly to solar radiation variations.

The

CIPPD

model is then defined as (4) where

1, 2

emet



and

0,7Ci clo



To deal with

Tr variations, a fuzzy interpolation

has been considered. It is defined on five

CIPPD

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

models over

r 

where

{1 5

,

25,30}

and

()

Tr Tr





refer to the

*Tr Tr

CIPPD



membership

degrees for

[10, 40Tr

]. This decomposition gives

the best compromise between the

approximation efficiency and the

CIPPD

complexity.

(,, ) ( (), (), ( ))

Tr Ta Va Hy

CIPPD Ta Va Hy C u Ta u Va u Hy





(4)

 Weak Separability Verification

For every

CIPPD





model, The weak

separability property (5) has also to be verified for

each attribute



where

\{ }

(, )

iNi

refers to a

thermal alternative such as

corresponds to the

attribute

value and

\{ }Ni

to attribute’ values other

than

. To check the weak separability property, for

\{}





, we have studied the

PPD

order

relationship w.r.t. the attribute

values. For

instance, Figure 2 shows same iso-temperature

shapes which may be interpreted that the

order

relationship w.r.t.

is invariant. However, the

minimum value of

PD is reached for slightly

different thermal conditions and, consequently,

entails order relationship variation (bordered by the

2 planes). Thus, the

PPD

does not fulfill the weak

separability property on the

Ta Va Hy

XXX

scale.

Yet, the weak separability property can be showed

on partitions of the

PD scale.

Figure 2:

( , , 50)

PPD Ta Va Hy



\{ } \{ }

\{ }

''''

\{ } \{ } \{ } \{ }

,,, ,

(, )(, ) (, )(, )

ii i Ni Ni j

jN i

iNiiNi iNiiNi

xx X y y X

xy xy xy xy



 







(5)

 Labreuche’s Assumption Verification

In order to avoid saturation problems, Labreuche

has supposed strictly increasing

functions. In fact,

saturation problems are commonly known in

interview based

u identification and consists on

similar thermal sensation assessments for slightly

different thermal alternatives:

\{ }

(, )

iNi

and

\{ }

(', )

iNi

, where ,'

ii i

xX and

\{ }

Ni j

jN i





Figure 2 shows saturation phenomenon w.r.t.

variations. Moreover, iso-temperature functions

(Figure 2) show a

PPD

gradient sign variation w.r.t.

. This observation can also be intuitively noticed.

In fact, in winter time, it is obvious that an

increasing

is appreciated until an upper

threshold, above it people get hot and their thermal

sensation progressively decreases which implies at

least one monotony variation of the

function. In

the next section, we explain how the weak

separability and monotony property have been

verified to identify

CIPPD

3.1.3 CIPPD Thermal Comfort Model

Since the

PD does not fulfill the weak separability

and monotony properties on the

Ta Va Hy

XXX

scale, one Choquet Integral cannot be identified for

each

CIPPD

model,





. Yet, the continuous

and regular

PD variation w.r.t. its attributes let

think that

CIPPD

can be identified on

PPD

’s

scale partition

'''

Ta Va Hy



that satisfies MAUT

and Labreuche assumptions. According to Figure 2,

2 partitions seems sufficient to cover most of the

PPD

scale for each

CIPPD





. Therefore,

the Labreuche approach has been applied in order to

compute

u and



for each

CIPPD

partition

without any commensurateness assumption.

The

CIPPD

models’ partitions have been

defined using Labreuche commensurateness

verification approach. In fact, to check

’s

attributes commensurateness and automatically

compute attributes’ commensurate values:

1 and

(i.e.,

() ( )

ii j j



and

() ( )

ii j j

uu00

,ij





and



) which refer, respectively, to good and

unacceptable satisfaction degrees w.r.t. the attribute

, Labreuche has proposed to study the gradient

function related to

w.r.t.

\{ }



variations. It

comes on studying function (6)’s shape: a constant

function means that no interaction exists between

attributes

and

; otherwise, preferential

interactions exist between attributes

and

, and

the commensurate value

, where

()

ux 

()

can, thus be computed. For this, at least one attribute

reference values

and

need to be given in order

to compute other attributes commensurate values.

For more information please refer to (Labreuche,

2011) (Denguir, 2012).

\{ } \{ }

,,,0

:(,)(,)

ij i Ni iNi

ij Ni j

fx PPDx x PPDxx









(6)

AggregatedPerformanceandQualitativeModelingBasedSmartThermalControl

For the

CIPPD

construction,

1 and

references have been considered. They come from

the

PD iso-temperature functions study where

have been chosen according to

PD ’s minimum

value which, consequently, entails the best

occupants’ thermal satisfaction. Therefore, basing on

(6),

1 ,

0 ,

and

commensurate values

have been atomically computed and the

CIPPD

partition scales identified. Utility functions

and

Choquet Integral parameters



can, then, be

identified for every partition

XX

, where



. For numerical details please refer to

(Labreuche, 2011).

3.1.4 Thermal Comfort Adjustment Rules

Our thermal comfort model relevance can

immediately be proven through its ability to auto-

generate thermal comfort adjustment rules. In fact,

thanks to the co-monotony between

CIPPD

and

u functions,





, interpretable thermal

adjustment rules can easily be identified as (7)

where

u

and

CIPPD

refer to the approximate

gradient of

u and

CIPPD

,,sng()sng( )

iTr

iNX X u CIPPD 

(7)

Moreover, in each

CIPPD

partition the

influence of each thermal attribute (

i.e.,



and



) on the

CIPPD

can easily be estimated

(8), where





refers to the Choquet Integral

approximate parameter. This result is all the more

useful than the Choquet integral is linear by simplex

(Denguir, 2012). Obviously, the efficiency of the

estimated gain can be discussed since it corresponds

to the fuzzy interpolation result; however, it remains

helpful for recommendation purposes. These thermal

comfort rules can immediately be applied by the

Energy Manager since they are interpretable as

satisfaction degrees which is different from the

PPD

where thermal attributes’ influences on the

PPD

are neither obvious to identify nor

interpretable for the Energy Manager.

1.2, 0.7,

Tr Tr

Me Ci Tr i i

CIPPD u i







 

(8)

3.1.5 CIPPD based Energy Consumption

Optimization

The

CIPPD

based smart thermal control consists on

adjusting thermal setpoints in order to reduce

thermal energy consumption while maintaining the

requested thermal comfort. For this, room

specificities (

i.e., occupant thermal comfort

requirement, solar radiation exposure) can be

considered to ensure a room customized thermal

control. For instance (9) corresponds to thermal

setpoints variation identification in order to ensure

as less as possible energy consuming thermal

comfort

c , where

belongs to the building room’

set

R . (9) assumes that the indoor temperature

control is the most energy consuming. Therefore,

minimizing



entails energy saving. It has to be

noted that thermal setpoint adjustment depends on

the equipment availability. For instance,



adjustment is only relevant when humidity control

kits are available in the building.

'' ' '

min

(,, )

'\{} (,,)

Tr k k k k k k k

Tr k k k k

CIPPD Ta Ta Va Va Hy Hy c

k R k CIPPD Ta Va Hy c



 













 



(9)

The optimization problem (10), where

'kk



is an

approximate thermal exchange rate between rooms

and

, considers the solar radiation exposure and

variation during the day

()

in order to adjust

thermal setpoints depending on radiant temperature.

Therefore, for every significant



, thermal set-

points are adjusted. Thus, the requested thermal

comfort is maintained and natural elements such as

the solar radiation is used in order to ensure parts of

heating necessities and then reduce the thermal

energy consumption.

()

min | |,

.. ,

(,, )

kkk kk

kR k R

Tr t k k k k k k k

sc k R

CIPPD Ta Ta Va Va Hy Hy c

  

 

























(10)

The

CIPPD

based smart thermal control can,

also, be used to adjust thermal setpoints according to

the context variation such as: raining days and

building’ occupancy variation. Note that these

optimization problems are simplified thanks to the

Choquet linearity by simplex property.

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

3.2 EQM based Thermal Control

The EQM based smart thermal control tries to

reproduce an approximate reasoning in order to

efficiently achieve the optimized thermal setpoints

computed in Section 3.1. In fact, when we are not

familiar with buildings’ thermal behavior, thermal

control of buildings may seem intricate. Uncertainty

about how relevant a thermal control is for a given

thermal situation, is then in its highest level. The

same reasoning remains true for the control of any

complex system. However, objective observations

(

i.e., vaguely identified physical behavior) and

subjective ones (

i.e., human preferences) may

contribute to reduce uncertainty about thermal

control. Therefore, we introduce our EQM which is

used to represent simplified thermal control rules. It,

also, defines how these thermal control rules should

be applied to ensure the control enhancement for

different thermal situations. The EQM design is

based on influence approximations relating thermal

control parameters to thermal performances. In order

to extend thermal qualitative modeling, the EQM’s

parameters and performances display time-related

information about thermal general behavior. The

influences, among parameters and performances, are

vaguely identified from thermal general behavior

and their accuracy is constantly improving through

online thermal quantitative observations. Therefore,

keeping track of predate thermal control, as well as,

their performances allow recalling them in similar

control situations. A Thermal Control Manager

(TCM) has then been conceived in order to maintain

thermal historical data. For each thermal control

attempt, the thermal situation, controls and

performances are, then, stored by the TCM. This last

is described by the following set

TCM



.. ,



(, , )}

kk k

S CMD PERF

where

is the number of

previous thermal experiences and

S ,

CMD and

PERF

are, respectively, the

k thermal situation

(

i.e., outdoor and indoor temperatures, etc.), controls

and performances. To support comparison over the

previous attempts and apply approximate reasoning,

AI techniques have been deployed. Figure 3 displays

the EQM based

smart thermal control general

approach;

new

S refers to a new thermal situation for

which an efficient thermal control needs to be

computed. It, mainly, involves indoor and outdoor

thermal current situations, as well as, thermal

setpoints, computed by the

CIPPD

, that need to be

reached before occupants show up.

Since the EQM uses quantitative thermal

experiences to improve its accuracy, TCM’s

quantitative knowledge need to be filtered before

proceeding by the EQM accuracy enhancement.

Therefore, step 1 of Figure 3 allows quantitative

linear reasoning around

new

S . The most favored

thermal experience

** *

(, , )S CMD PERF

(w.r.t. the

current situation) for the linear quantitative

reasoning is computed by step 2. The EQM thermal

enhancement rules are applied in step 3 in order to

compute a more likely

better command law

new

CMD

from

CMD . In fact, since the EQM influences have

been approximated using thermal objective and

subjective knowledge, thermal enhancement control

can be operated. Contradictory influences on thermal

performances can, simply, be resolved by

considering user’s priorities. For instance, building’s

occupants may be more demanding about their

thermal comfort. The EQM will, thus, give priority

to optimize thermal comfort related performances.

Hence, thanks to the EQM influences, it becomes

possible to recommend control parameters

increase/decrease.

new

CMD is finally applied and

evaluated in Figure 4’s step 4. In this paper, we

particularly focus on particular features used to

extend the qualitative thermal modeling. For this, we

explain the time-related, simple behavior analysis,

and quantitative observation used in order to extend

our EQM’s qualitative modeling.

EQM STC (

new

TCM

)

TCM





then call the energy manager else

1. Compute

TCM TCM

where,

(, , )S CMD PERF TCM

is similar to

new

TCM





then call the energy manager else

2. Find

** *

(, , )|S CMD PERF

(, , )S CMD PERF TCM

CMD

is most favored for

new

3. Compute

new

CMD

for

new

based on the EQM and the

quantitative information of

CMD

4. Apply

new

CMD

and update the

TCM

with the new attempt





new new new

S CMD PERF

end if

end

Figure 3: EQM based smart thermal control approach.

3.2.1 Time-Related Information

In order to ensure RIDER deployment expectations,

our EQM applies an event-based representation

(Montmain, 1991) for thermal control laws. This

latter is more relevant than a classical sampled time

representation in a qualitative approach. It is, also,

considered sufficient for the thermal control laws’

description since steps and ramps signals are usually

AggregatedPerformanceandQualitativeModelingBasedSmartThermalControl

used for the thermal regulations. In order to involve

time-related information, the EQM considers

thermal control starting time which is useful to

improve control delays. Therefore, for each thermal

control law

L( )

we associate a control parameter

vector

(

)

tpy

 

CMD

refers to the set of

control parameters vectors

applied on all

building’s actuators. These 3 control events are

described by the thermal example showed in Figure

4 and refer, respectively, to

L( )

delay (time-gap

between

L( )

starting time

and thermal control

starting time

), gradient (characterized by the time-

gap between

L( )

highest

and lowest

values)

and amplitude (height-gap between

L( )

highest and

lowest values). Moreover, time-related information

are considered in the EQM’ performances modeling.

In fact, rather than building’s thermal profiles,

thermal performances are considered to ensure

RIDER deployment expectations. Indeed, the

performance vector



(

,,cost com

ort

)

exi

ity

describing thermal energy consumption, stationary

thermal comfort and setpoints’ achievement delay,

ensures building’s thermal assessment.

exi

ity

shows time-related information which makes time-

related control enhancement possible.

PERF

corresponds then to the set of all building’s rooms

thermal performance vectors

Figure 4: EQM’s control events.

3.2.2 Simple Thermal Behavior Analysis

General and simple thermal behaviors have been

studied in order to identify how each control

parameter influences the considered thermal

performance. For instance, (11) describes one room

temperature profile

()

when applying the

command law

L( )

, where

e ,



and



refer,

respectively, to the outdoor temperature, room

passive resistance coefficient and thermal loss

coefficient. Hence, identifying the



and



coefficient does not concern the EQM modeling

since it implies buildings’ excitement scenarios

constraints, however, (11) is useful to study thermal

performance monotonies

w.r.t. command parameters

variations. The result of this study can, thus, be

immediately applied on any building without

requiring any extra information on building’s

thermal process.

()

(())(())0

dT t

tTt TeTt





L( )

(11)

Table 1 describes gradient directions computed

over each performance

w.r.t. each control parameter.

Considering gradient directions rather than precise

derivative values ensures the RIDER deployment

expectations. For each performance

, where



and

is the considered thermal

performance set (

e.g.,

{, , }

S cost comfort flexibility



), and control parameter

, where

iS

and

S is

the considered control parameter set (

e.g.,

{, , }

Stpy



 

), an influence function





{

}





is defined, where values of

thermal control parameters

c ,

cS , and

performances





, are, respectively,

defined in

and

indicates whether the

performance

increases (+) or decreases (-) w.r.t.

variations. A (0) valued

function indicates that

has no influence on

. The

qualitative gain can,

thus, be represented by the EQM and results from

studying gradient directions of simplified thermal

behaviors such as (11). For instance, it is commonly

known that, in winter time, thermal energy

consumption (

cos

) increases by increasing the

command law height (



). This is illustrated, in

Table 1, by a constant influence function describing

a gradual rule type on

cost





such as the greater

the heating step amplitude is, the greater the thermal

energy consumption would be. Therefore, regardless

of buildings thermal specificities,

can be

deduced from simplified physical behaviors (

e.g.,

ycost



Table 1: EQM influence modeling (0 means no influence).



p



cost





tcost t cost

Fcp







comfort



y comfort y comfort

Fcp



lexibility





Buildings’ special features can occasionally be

responsible of

’s sign variations (e.g.,

tcost



). In

this case, influence functions are subject to

uncertainty problems that are handled by

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

considering quantitative observations in the EQM

based reasoning.

3.2.3 Quantitative Observations

Uncertainty about some influence function may

negatively affect the EQM efficiency. However

considering thermal quantitative observations,

uncertainty about gradient directions is reduced

according to new thermal observations. Moreover,

the qualitative based reasoning may seems lacking

for continuous enhancement purposes. Nevertheless,

combining the qualitative reasoning to quantitative

observations makes the control enhancement more

accurate and seems to be sufficient for the EQM

based smart thermal control. In this section, we

explain how quantitative knowledge has been

considered to reduce uncertainty about influence

functions and how the EQM reasoning has been

made more accurate using the TCM quantitative

thermal experiences.

 Decreasing EQM’s Uncertainty

In order to reduce uncertainty about the EQM

influence functions, objective and subjective

knowledge has been considered. Objective

knowledge corresponds mainly to interpretable

physical phenomena. When uncertainty about

functions holds, simple learning techniques are

applied over TCM’s previous thermal experiences in

order to specifically identify each building’s bending

points. For instance,

tcost



depends on building

ventilation and insulation properties: starting the

heating process earlier or later impacts differently

the thermal energy consumption. Figure 6 shows

some possible shapes of the continuous function

relating



cost

values. The shape of this

function is obtained from the simplified thermal

behavior model (i.e., in some cases, the continuous

function relating



cost

displays a maximum.

Otherwise it is decreasing for any



value). The

maximum remains to be identified. Figure 5’s

displayed maximums can be explained by the fact

that, when outdoor temperature is lower than the

indoor one, building’s ambient temperature

decreases until the control law is started. The



’s

interval for which

cost

increases refers to situations

where it is more costly to start heating for a short

time from a low temperature than heating the

building for a longer time but starting from a higher

temperature. The decreasing

cost

w.r.t.



refers to

the opposite behavior. Furthermore, the HVAC

system is responsible for the rapid decrease of

building’s ambient temperature when the heating

system is off. In fact, the HVAC continuously

injects a weak percentage of the outdoor air for

ventilation purposes. Therefore, we propose to use

TCM’s quantitative knowledge to capture, for each

building, the





value that entails sign

variation in the continuous function (Figure 5) and

finally online learn

tcost



influence function. We

have introduced a possibility based approach in

order to continually reduce uncertainty about EQM’s

influence functions. More complicated buildings’

thermal dependent influence functions have been

thus considered. For more information, please refer

to (Denguir, 2014).

Subjective knowledge can also be used in order

to reduce uncertainty about buildings thermal

control. For instance, the

CIPPD

thermal comfort

model can contribute to identify the

ycomfort



function (Table 1). Therefore, building’s occupants

thermal sensations and thermal context variations

(i.e., humidity and sunshine) are considered while

identifying

ycomfort



. In fact, depending on the

thermal context, an increasing

may either

improve or distract the occupant’s thermal comfort.

Hence,

ycomfort



acknowledge sign variations since

thermal command law height influences

. The

CIPPD

formalism helps

ycomfort



identification

since

sng( )

u

provides its values.

Figure 5:

cost

w.r.t.



variations from different

ventilation perspectives.

 Improving EQM’s Accuracy

Step 3 of Figure 3 shows that the EQM based

smart thermal control is built from the EQM

qualitative reasoning on the quantitative thermal

command

CMD which makes it more accurate. For

instant, (12) shows how one actuator

t

parameter (

new



) can be precisely evaluated considering predate

quantitative experience and particularly the



command parameter and

flexibility

performance.

Therefore, deciding about the most likely favored

TCM predate thermal experience, to be used in the

EQM reasoning, is an important issue and

corresponds to Figure 3’s step 2.

AggregatedPerformanceandQualitativeModelingBasedSmartThermalControl

**new

t t flexibility

CCP





(12)

Our EQM quantitative enrichment is decided

among the TCM prior experiences considering 3

decision criteria:

i. Similarity between previous situations

and

the new one

new

S : it allows overcoming non-

linearity problems related to thermal controls (step 1

of Figure 3) since maximizing the similarity allows

linear quantitative reasoning around a setting point.

Similarity between thermal situations is based on a

distance

(, )dist S S

, where

S and

S are two

thermal situations. The smaller

(, )dist S S

is, the

more similar

S and

S would be. Since thermal

situations are only defined by temperature

measurements, there are no commensurateness

problems in

(, )dist S S

definition.

ii. Thermal performances: Obviously, the better

the resulting thermal performances

PERF

are, the

more favored the control experience would be. For

this, MCDA techniques have been deployed. A

preference model over the considered performances

is identified. Firstly, utility functions (

cost

u ,

comfort

and

lexibility

) are defined for each

performance to ensure commensurability. They

allow the assessment of each performance over the

same scale which is the satisfaction degree or utility

scale [0,1]. Secondly, an aggregation function is

required in order to ensure the overall thermal

evaluation

for each room

rR

(

corresponds

to the building rooms’ set) and prior thermal control

experience

. These steps are related to the Energy

Manager

preferences which depend on his Energy

Policy

. Interview based approach such as

MACBETH can be applied in this case. We assume

that a weighted sum is sufficient to capture this

preference model. When thermal control is related to

a subset of rooms

'RR

, overall thermal

assessment has to consider all thermal performances

over

. Thus, our EQM proposes to proceed firstly

by aggregating all performances over

from the

energy consumption (

), thermal comfort (

min

)

and flexibility (

max

) points of view; secondly, the

preference model defined for one room is applied for

. We denote by

the overall building thermal

assessment associated to the

k (

PERF

) prior

thermal experience stored by the TCM.

iii. Previous enhancement results: predate

controls that have led to thermal enhancement

failures are penalized in future TCM evaluations.

Therefore, we associate a set

Bad to each

(, , )

kk k

S CMD PERF TCM



ad gathers prior

thermal experiences that were computed from

(, , )

kk k

S CMD PERF

and led to thermal performance

decreases.

Considering these 3 criteria, an overall score

core (13) can be computed for each TCM

experience in a limited neighborhood of

new

S (to

satisfy the thermal process linear quantitative

behavior expected property). The quantitative

information

(, ,SCMD

)

ERF

TCM

favored for

our EQM enrichment verifies:

core score

(, ,

SCMD



)

PERF

TCM



. Quantitative

knowledge can then be used to make more accurate

the EQM reasoning.

{1, . . , } , {1 ( , ) } . *

{1 ( , ) }

kknewk

knew k

kBad

knscore distSS

dist S S









(13)

4 CONCLUSIONS

In order to fulfil RIDER deployment expectations,

we have proposed the RIDER STC solution which

considers the CIPPD and EQM based reasoning in

one building scale. The CIPPD based reasoning

allows the identification of the most relevant thermal

setpoints in order to improve the thermal comfort

and reduce thermal energy consumption. Once the

optimized setpoints are provided, the EQM based

reasoning says how they can be efficiently achieved.

It implements an iterative approach that provides

thermal control recommendations as soon as it is

deployed without needing any

a priori learning or

identification. These control recommendations are

then refined thanks to quantitative observations and

qualitative physical aspects related to thermal

processes. When using the CIPPD based control, our

experimentations let expect about 10% of thermal

energy consumption decease. Combined to the EQM

based

smart thermal control, RIDER STC solution

reveals, for one room, about 7 to 31% of thermal

energy consumption decrease and 12 to 24% for

multi-room thermal enhancement. Average thermal

energy consumption decrease ensured by the RIDER

STC is evaluated to 16% which is significant

considering energy prices. How the RIDER STC can

bypass frequent thermal control deployment issues

such as quantitative data availability, can be

considered as an outstanding point compared to the

existent thermal control solutions. Just the CIPPD

based

smart thermal control can be considered as a

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics

remarkable shift in how smart thermal control has

been considered till these days. Comparing the EQM

based

smart thermal control efficiency with

commonly used approaches (based on white, grey or

black thermal models), is unbalanced considering

their different application conditions. In fact, trying

to operate an MPC (Model Predictive Control) in

few days on a completely unknown building is not

conceivable. It goes the same when asking the EQM

based control for the same efficiency as a MPC

based control in a fully identified building’s thermal

process. Yet, perspectives remain possible to

improve RIDER STC efficiency. For the CIPPD

based

smart thermal control, adaptive and dynamic

thermal comfort model could be considered in order

to ensure more personalized and individualized

thermal comfort adjustment. Yet, the PPD’s choice

satisfies the data unavailability issues. The transition

toward an adaptive and dynamic thermal comfort

model shall, thus, be supported by

online learning

techniques. The CIPPD identification could also be

improved by using bipolar utility scale wish gives

more expressivity to the thermal comfort models and

could lead to a better approximation of the PPD

function. The EQM based

smart thermal control

provides a methodology in order to improve the

qualitative based thermal control efficiency.

Therefore, each step implementation technique

could be discussed. For instance, uncertainty

management in influence functions can be improved.

Ambiguous measurements coming from thermal

disturbances (

i.e., windows and door opening)

should complete this point. Sensors data precision

can be studied as well. Qualitative interactions

between the control enhancement parameters could

also be studied in order to compute enhancement

recommendations based on subsets of control

parameters variations instead of singletons. This will

warrantee the EQM control convergence to a global

improved control experience

rather than a local one.

The scalability of the RIDET STC solution could

also be discussed. In fact, we have shown in section

3.2.3 that multi-room transition needs some settings

and it goes the same for any scale transition. The

scalability could have been made automatic by

providing different scales templates in the RIDER

STC final solution. This task is simplified thanks to

the EQM modularity.

REFERENCES

Aydinalp, M., Ugursal, V.I., Fung, A.S., 2002. Modeling

of the appliance, lighting, and sapce-cooling enrgy

consumptions in the residential sector using neural

networks, Applied Energy, vol. 2, n°71, pp. 87-110.

Calvino, F., Gennusca, M.L., Rizzo, G., Scaccianoce, G.,

2004. The control of indoor thermal comfort

conditions: introduicing a fuzzy adaptive controller,

Energy and Buildings, vol. 36, pp. 97-102.

Denguir, A., Trousset, F., Montmain, J., 2012. Comfort as

a Multidimensional Preference Model for Energy

Efficiency Control Issues, In SUM 2012, pp. 486-499.

Denguir, A., Trousset, F., Montmain, J., 2014.

Approximate Reasoning for an Efficient, Scalable and

Simple Thermal Control Enhancement, IN IPMU,

Montpellier.

Dong, B., Cao, C., Lee, S.E., 2005. Applying support

vector machines to predict building energy

consumption in tropical region,Energy and Buildings,

vol. 5, n°37, pp. 545-553.

Dounis, A.I., Santamouris, M.J., Lefas, C.C,. Argirious,

A., 1995. Design of a fuzzy set environment comfort

system, Energy and Buildings, vol. 1, n°22, pp. 81-87.

Dubois, D., 1989. Order of magnitude reasoning with

fuzzy relations, Artificial Intelligence, vol. 3, n°14, pp.

69-94.

European Parliament and Council (EP&C)., 2012.

Directive 2012/27EU of European Parliament and

Council of 27 October 2012 on the energy

performance of buildings, Official Journal of the

European Union, pp. 1-56.

Fanger, P.O., 1967, Calculation of Thermal Comfort:

Introduction of a basic Comfort Equation. ASHRAE

Trans. Vol. 73.

Fishburn, P.C. 1970. Utility Theory for Decision-Making.

John Wiley & Sons, New York.

Gagge, A.P., Fobelets, A.P., Berglund, L.G., 1986. A

Standard Predictive Index of Human Response to the

Thermal Environment. ASHRAE Trans, vol. 92.

Grabisch, M. k-Ordered Discrete Fuzzy Measures and

Their Representation. In: Fuzzy sets and systems, vol.

92, pp. 167-189. (1997).

Kalogirou, S.A., Bojic, M., 2000. Artificial neural

networks for the prediction of the energy consumption

of a passive solar building, Energy, vol. 5, n°25, pp.

479-491.

Krantz, D.H., Luce, R.D., Suppes, P., Tversky, A. 1971.

Foundations of measurement, Additive and Poly-

nomial Representations, vol. 1. Academic Press.

Kuipers, B., 1986. Qualitative simulation, Artificial

Intelligence, vol. 29, pp. 289-388.

Labreuche, C.2011. Construction of a Choquet integral

and the value functions without any com-

mensurateness assumption in multi-criteria decision

making. In EUSFLAT-LFA, Aix-les-Bains, France.

Li, Q., Meng, Q.L., Cai, J.J., Hiroshi, Y., Akashi, M.,

2009. Applying support vector machine to predict

hourly cooling load in the building, Applied Energy,

vol. 10, n°86, pp. 2249-2256.

Ma, Y., Borrelli, F., Hencey, B., Packard, A., Bortoff, S.,

2009. Model predictive control of thermal energy

storage in building cooling systems. In Conference on

AggregatedPerformanceandQualitativeModelingBasedSmartThermalControl

Decision and Control and Chinese Control

Conference, pp. 392–397.

MQ&D coordinated by P.Dague, 1995. Qualitative

Reasoning: a survey of techniques and applications, AI

Communications The European Journal of AI, IOS

Press, vol. 8, pp. 119-192.

NF EN ISO 7730, 2006. Ergonomie des ambiances

thermiques : Détermination analytique et interprétation

du confort thermique à l'aide de calculs des indices

PMV et PPD et du confort thermique local, AFNOR.

Oldewurtel, A., Parisio, A., Jones, C.N., Morari, M.,

Gyalistras, D., Gwerder, M., 2010. Energy efficient

building climate control using stochastic model

predictive control and weather predictions, In

American Control Conference, Baltimore.

Pacific Northwest National Laboratory (PNNL)., March

2012. 2011 Building Energy, D&R International, Ltd.

Signh, J.N., Sharma, J.K., 2006. Fuzzy modelling and

control of HVAC systems - a review, Journal of

scientific and Industrial Research, vol. 65, n°6, pp.

470-476.

Terziyska, M., Todorov, Y., Petrov, M., 2006. Adaptive

supervisory tuning of nonlinear model predictive

controller for a heat exchanger. Energy saving control

in plants and buildings, pp. 93-98.

Wang, S., Xu, X., 2006. Simplified building model for

transient thermal performance estimation using GA-

based parameter identification, International Journal

of Thermal Sciences, vol. 4, n°45, pp. 419-432.

White, J.A., Reichmuth. R., 1996. Simplified method for

predicting building energy consumption using average

monthly temperatures, In Intersociety Energy

Conversion Engineering Conference, pp. 1834-1839.

Williams, B. C., 1989. Temporal qualitative analysis:

explaining how physical systems work, In Qualitative

reasoning about physical systems, Morgan Kauffmann

publishers, pp. 133-177.

Yang, I.H., Yeo, M.S., Kim, K.W., 2003. Application of

artificial neural network to predict the optimal start for

heating system in building, Energy conversion and

Management, vol. 17, n°44, pp. 2791-2809.

Zaheer-uddin, M., Tudoroiu, N., 2004. Neuro-PID

tracking control of a discharge air temperature system,

Energy Conversion and management, vol. 15-16,

n°45, pp. 2405-2415.

ICINCO2014-11thInternationalConferenceonInformaticsinControl,AutomationandRobotics