SHORE: A Model-driven Approach That Combines Goal, Semantic and

Variability Models for Smart HOme self-REconﬁguration

Denisse Mu

nante

1 a

, Bruno Traverson

, Sophie Chabridon

3 b

and Amel Bouzeghoub

3 c

ENSIIE & SAMOVAR,

Evry, France

EDF R&D, Saclay, France

SAMOVAR, T

ecom SudParis, Institut Polytechnique de Paris, Evry, France

Keywords:

Goal-orientation, Ontology, Feature Models, Smart Home, Self-reconﬁguration.

Abstract:

Smart Homes, and more generally the Internet of Things (IoT), are gaining more and more audience, but this

kind of environment is still challenging on several aspects. Semantic representations, mainly ontologies, have

been used to cope with the complexity and lack of interoperability due to the wide variety of objects and

services. However, these semantic representations do not permit a ﬁne description of variability and evolution

to achieve particular objectives. Therefore, in this paper, we introduce a model-driven approach, so-called

SHORE for Smart HOme self-REconﬁguration, which addresses these limitations and covers the complexity

and variability of Smart Homes allowing automated reconﬁguration guided by their goals. Thus, SHORE

includes artifacts, based on goal, semantic and feature models, that are used at design-time to conceptualise

Smart Homes, their variants and goals. Whilst, at runtime, SHORE exploits these artifacts to detect situations

that require reconﬁguration, i.e., detect deviations from their goals, and then to calculate optimal reconﬁgu-

ration plans to cope with these situations.

1 INTRODUCTION

A Smart Home relies on information and communi-

cation technologies for controlling household appli-

ances and other domestic equipments remotely. It is

becoming possible to adapt the behavior of a house

to the habits of its residents. Dynamic adaptation can

take place as soon as some change is detected, like

a decrease of the external temperature requiring to

switch on the heating system in order to keep indoor

temperature comfortable. The automation degree of

a Smart Home can be very high by deﬁning elabo-

rated rules and targeted goals, related to the expected

temperature of the house for instance, in order to

provide added-value services to the residents. Thus,

self-regulation of the Smart Home services could be

a good means when any deviation from the Smart

Home goals is detected.

The dynamic adaptation of a system is con-

ceived as a way to avoid or correct the degrada-

tion of its quality of service during its execution,

https://orcid.org/0000-0003-2621-8342

https://orcid.org/0000-0002-1591-6754

https://orcid.org/0000-0003-4890-9005

which may be due to changes in environmental con-

ditions. A substantial research effort has been done

towards deﬁning methods and techniques for manag-

ing self-adaptive systems (SASs). Model-driven ap-

proaches (C

amara et al., 2017) and in particular mo-

del@run.time methods have been proposed to spe-

cify how self-adaptive systems should react in dy-

namically changing runtime environments (Bencomo

et al., 2019; Erazo-Garz

on et al., 2021). Regarding

Smart Homes, self-adaptation is guided by ontolo-

gies, e.g., (Seydoux et al., 2016; Lork et al., 2019).

Ontologies allow to address the interoperability of the

wide variety of Smart Homes appliances and services.

However, challenges remain to reach a high ex-

pressiveness and ease of use of dynamic reconﬁgu-

ration both at design and execution times. Adaptation

rules should be expressed at a high level of abstrac-

tion to be understood by inhabitants and should be

managed dynamically to evolve with regards to mo-

diﬁcations in environment properties and user prefer-

ences. Also, even more importantly, in case of multi-

ple services operating in a same area of a Smart Home

and thus sharing a set of sensors and actuators, poten-

tially conﬂicting goals should be dealt with when the

Smart Home requires adaptation. One optimal solu-

328

Muñante, D., Traverson, B., Chabridon, S. and Bouzeghoub, A.

SHORE: A Model-driven Approach That Combines Goal, Semantic and Variability Models for Smart HOme self-REconﬁguration.

DOI: 10.5220/0010907300003119

In Proceedings of the 10th International Conference on Model-Driven Engineering and Software Development (MODELSWARD 2022), pages 328-335

ISBN: 978-989-758-550-0; ISSN: 2184-4348

tion should be elected among several alternatives in a

manageable and responsive way.

Therefore, our research objective concerns the

building of a tool-supported approach for the

Smart HOme services REconﬁguration, so-called the

SHORE approach. More speciﬁcally, in this paper we

focus on the following two research questions: RQ1

- How to deﬁne Smart Home variants at design-time

allowing reconﬁguration at runtime? RQ2 - How to

select an optimal Smart Home variant to correct devi-

ations from goals?

To answer the ﬁrst research question, we start

from the state of the art of standardised se-

mantics (i.e., ontologies) for Smart Homes, e.g.,

SAREF (Smart Appliances, 2017). Then, we inves-

tigate how to extend these standards to add notions

that allow the self-reconﬁguration of Smart Home

systems, i.e., notions to deﬁne Smart Home goals,

e.g., i* (Yu, 1997), and variants using variability mo-

dels, e.g., feature models (Kang et al., 1990). Con-

cretely, starting from a goal-orientation viewpoint, we

elicit the SHORE artifacts, which are used at design-

time. Thus, SAREF

and FM

were proposed as ex-

tensions of SAREF and feature models respectively.

SAREF

allows to describe Smart Homes by adding

notions around their goals. Based on this descrip-

tion, automated rules to detect situations that require

adaptation can be formulated. Whilst FM

comple-

ments this description by adding Smart Home va-

riants. Based on these variants, we can ﬁnd correct

adaptation plans to cope with detected situations.

Concerning the second research question, we ex-

plore existing techniques to exploit gathered data

when taking reconﬁguration decisions in order to deal

with conﬂicting goals. In particular, we use the op-

timisation approach based on grammar guided ge-

netic programming that was introduced in a previous

work (Kifetew et al., 2017a; Mu

nante et al., 2018).

This paper is organised as follows: Section 2

introduces the problem formulation for the self-

reconﬁguration of Smart Homes services. Section 3

gives background. Section 4 describes the SHORE

approach to support this reconﬁguration. Section 5

discusses related works and Section 6 identiﬁes next

steps and concludes the paper.

2 PROBLEM FORMULATION

Roughly speaking, Smart Homes are composed of de-

vices (actuators) that offer services to make the life of

residents more pleasant. Thus, one crucial goal is to

keep user comfort high. However, Smart Homes have

been also created to accomplish other goals, for in-

stance to keep energy consumption low or at least un-

der control. These goals might be in conﬂict when

they cannot be reached at the same time. The po-

tential goal conﬂicts could be solved using priorities

and variants for services’ settings. These variants

may reach the goals at different levels of accomplish-

ment, being in the range from no compliance to full

compliance. While many techniques might be used

to manage conﬂicts, we focus on priorities as they

have widely shown their efﬁciency in other research

domains such as security (Breaux and Ant

on, 2008).

In this paper, we address the problem of Smart

Home self-reconﬁguration at runtime, taking into

consideration priorities for Smart Homes goals and

variants for their services’ settings. This problem

involves choosing the conﬁguration (variant) for the

Smart Home’s services that best respects the goals

and their priorities. Hence, the problem of Smart

Homes services conﬁguration can be deﬁned as fol-

lows. Given:

M goals for the Smart Home SmartHome

goals

{goal

,goal

,...,goal

M priorities for these goals (one priority per goal)

SmartHome

priorities

= {prio

, prio

,..., prio

R actuators that change the state of the Smart Home

SmartHome

actuators

= {act

,act

,...,act

P services offered by the Smart Home actu-

ators to achieve their goals (an actuator can

offer 1 or n services) SmartHome

services

{service

,service

,...,service

Services dependencies are twofold: i) inclusion de-

pendency: service

includes service

and ii) exclu-

sion dependency: service

excludes service

Each service contains a set of conﬁguration va-

riants. For instance: X variants for service A:

serviceA

variants

= {variant

,variant

,...,variant

and we complete the conﬁguration variants for the

other P services,

In order to select a good combination of variants

of the Smart Home services, we should deﬁne met-

rics that will be associated to these variants. Thus, the

above problem formulation is updated to add the men-

tioned metrics, henceforth called quality attributes

(QAs), as follows. Given:

S QAs associated to each conﬁguration variant

SmartHome

QualityAttributes

= {qa

,qa

,...,qa

2 operators for evaluating the QAs {min,max}, i.e.,

minimise and maximise the values of QAs,

Then, the problem formulation of the Smart

Home services conﬁguration is deﬁned as the

selection of the P variants (one per service) that

best respect the M Smart Home goals and their

priorities with respect to the evaluation (min or

max) of the S QAs.

SHORE: A Model-driven Approach That Combines Goal, Semantic and Variability Models for Smart HOme self-REconﬁguration

329

To tackle the problem of the Smart Home

services reconﬁguration, we add the following

variables. Given: the current conﬁguration

of the P Smart Home’s services Con f ig

cur

{variantA

,variantB

,...,variantP

}, and the alert

that contains information on the deviation of any

Smart Home goal.

The decision-making for the Smart

Home reconﬁguration is to ﬁnd

a new conﬁguration Con f ig

new

{variantA

,variantB

,...,variantP

} of

its services, such that: Con f ig

new

is better than

Con f ig

cur

. The deﬁnition of being “better” is

based on a ﬁtness function which computes

an aggregated value of the QAs that reﬂects a

conﬁguration of the Smart Home that increases

the level of achievement of its goals. Since QAs

could have different natures, a normalization

technique is employed as shown in Equations 1

and 2 ([lv,rv] represents the interval from the

minimal to maximal values that could take a

QA). The minimisation problem is translated

in a maximisation one. Then, the aggregated

ﬁtness value of all the variants in Con f ig

new

calculated.

max(var

) =

∑

i=1

prio

∗ (qa

− lv)/(rv − lv) (1)

min(var

) =

∑

i=1

prio

∗ [1 − (qa

− lv)/(rv − lv)] (2)

3 BACKGROUND

3.1 Ontology-based Approaches for

Smart Home

In the domain of Smart Home, several ontologies

have been proposed see the study presented by Ba-

jaj et al. in (Bajaj et al., 2018) for an in-depth sur-

vey of IoT ontologies. Among them, the Smart Ap-

pliance REFerence SAREF

is a reference ontology

for smart appliances. It was developed with the sup-

port of the European Commission and is published by

ETSI (Smart Appliances, 2017). SAREF mainly fo-

cuses on devices that are “tangible objects designed

to accomplish a particular task in households, com-

mon public buildings or ofﬁces” and may be further

reﬁned as sensors or actuators. From a user perspec-

tive, a device is designed to achieve a task (Task)

https://forge.etsi.org/rep/SAREF

related to a feature of interest (Commodity) using

a speciﬁc strategy (Profile). In case of sensors,

devices make measurements (Measurement) related

to some properties (Property) and in speciﬁc units

(Unit of measure). Devices are accessed via an in-

terface (Service) implemented on top of function-

alities (Function) that are composed of commands

(Command) and states (State).

Some limitations of SAREF can be identiﬁed. For

instance, devices are only described as atomic, i.e.,

they are not composite, not allowing to include com-

plex devices. Also, services are not composable and

are tightly coupled to devices preventing a higher

level of abstraction. Moreover, services cannot be

customised, this means that services cannot be con-

ﬁgured/regulated according to some parameters. Fi-

nally, potential conﬂicts between Smart Home goals

cannot be managed.

3.2 Variability Models in the Software

Product Line

Software Product Line Engineering (SPL) follows the

principles of product lines to enable product mass

customisation. The selection of a set of suitable fea-

tures when a product is conﬁgured is a crucial activ-

ity. Such selection is typically made by exploring the

space of trade-offs along different attributes of inter-

est, for instance cost and value.

Feature models (FMs) (Kang et al., 1990) are of-

ten used in SPL to describe the commonalities and

variability of product families. In FMs, features are

hierarchically organised by means of parent-child and

tree constraint relations. Tree constraint relations are:

Mandatory, Optional, Alternative, and Or. If a fea-

ture has a Mandatory relation with its parent, it must

be included in all software conﬁgurations in which its

parent appears. If a feature has an Optional relation

with its parent, it can be optionally included in pro-

ducts in which its parent appears. A set of features

are grouped as Alternatives if exactly one of these

features has to be included when their parent appears

in the conﬁguration. A set of features are grouped

using an Or relation if one or more of these features

can be included when their parent appears in the con-

ﬁguration. An overview of it is shown in Figure 2 (a).

In addition to the previous relations, a FM may

also include cross-tree constraints between features:

Implies and Excludes. If featureA implies featureB,

and featureA appears in a product, then featureB must

be selected for the product. If featureA excludes fea-

tureB, and featureA appears in a product, then fea-

tureB must not be selected for the product.

(Batory, 2005) introduces a mapping between

MODELSWARD 2022 - 10th International Conference on Model-Driven Engineering and Software Development

330

FMs and grammars allowing to apply formal reaso-

ning. Our mapping from tree constraints in FMs to

grammars is summarised as follows:

Optional: OPTIONAL(F) → <F_opt> ::= <F> | λ

Alternative: ALTERNATIVE(F1,F2) →

Or: OR(F1, F2) →

<F_or> ::= <F> <F_or> | <F> ;

And (*): AND(F1,F2) → <F> ::= <F1> <F2>

(*) And was included to capture grouped features with

the same parent but without alternative/or relations.

Non-terminals are represented by names in angle

brackets while terminals are represented by quoted

strings. For the sake of simplicity, λ rules, which

mean empty, are represented by a |.

Cross-tree constraints in FMs could be sufﬁxed

to the grammar so that they can be read and applied

by a program. In our case, we adopt the propositional

logic notation proposed by (Batory, 2005).

Attributes are associated to each feature so that

optimal product conﬁgurations can be derived using

the values of the attributes.

4 SHORE: AN APPROACH FOR

SMART HOME

SELF-RECONFIGURATION

(Weyns, 2019) introduces a conceptual model for

building SASs. It is mainly composed of the environ-

ment, managed system, managing system and adap-

tation goals components. We envision the SHORE

approach by leveraging the high level components of

Weyns’ model. The environment refers to the ex-

ternal world with which the SAS interacts. For in-

stance, the environment of a Smart Home includes the

weather, which can inﬂuence the levels of humidity,

temperature and illuminance. The managed System

comprises the application code that realises the sys-

tem domain functionality. For instance, in the case of

Smart Homes, services that switch on/off actuators or

conﬁgure their settings in order to accomplish a spe-

ciﬁc purpose such as the regulation of levels of tem-

perature, humidity or carbon emissions inside build-

ing spaces. The adaptation Goals are concerns of the

managing system over the managed system to pre-

serve certain conditions or qualities. For instance, an

adaptation goal of a Smart Home could be to keep re-

sidents comfort high. It can be translated as reaching

the acceptable values for the temperature, humidity

or level of carbon in a room. These acceptable values

could be established using standardised regulations or

users’ preferences. The managing System conducts

the managed system, it comprises the adaptation logic

that deals with one or more adaptation goals. Diffe-

rent terms may be used to refer to a managing system,

e.g., autonomic manager, adaptation engine, reﬂective

subsystem or controller.

The architecture of managing systems is usu-

ally based on the MAPE-K (Monitor, Analyse, Plan,

Execute, over shared Knowledge) loop (Kephart and

Chess, 2003). Monitor allows to collect the sys-

tem environmental context, besides components of

the managed system. In a Smart Home, sensors col-

lect data about current levels of temperature, humidity

and carbon concentration in a room. Using the mon-

itored data, Analysis determines if the adaptation of

the managed system is needed or not. For instance,

if the speciﬁed goals (see the adaptation goals com-

ponent) are violated, an alert is sent. Next, Plan eva-

luates a set of system variants in order to choose the

most adequate one to address the situation alerted to

by the analysis. Finally, Execute takes into account

the architecture model of the managed system in or-

der to deploy the adaptation plan at runtime. In the

Smart Home context, we should take into account the

actuators present in the house, for instance an HVAC

(Heating, Ventilation, Air-Conditioning) system.

Figure 1 introduces a goal-oriented model; it

helps us to identify agents, goals, qualities, tasks

and resources involved in the Smart Home self-

reconﬁguration context. This model was designed

using the SAS elements detailed above. A goal-

oriented model highlights the interaction between

agents to achieve their goals and/or qualities. For

instance, the resident agent has goals such as G1:

change temperature settings and G2: change air ex-

traction settings to Q1: keep her/his comfort high

without impacting Q2: the saving of energy consump-

tion. To accomplish G1 and G2, resident changes

actuators settings. Notice that the tasks executed by

the resident contribute to Q1 and Q2 in a posi-

tive or negative way. Hence, the resident relies

on the managing system to keep the accomplish-

ment of Q1 and Q2 stable. To do that, the managing

system ﬁnds a new conﬁguration (G3, G4) that al-

lows this balance by ﬁrst analysing changes caused

by residents or the environment, next planning the

new conﬁguration, and ﬁnally enacting the plan. To

plan new conﬁgurations, the managing system uses

a space of conﬁguration variants and the current con-

ﬁguration for the Smart Home actuators. To enact

a new conﬁguration, the managing system relies on

the Smart Home actuators.

In the next subsections, we present the SHORE

artifacts. It is worth to mention that SHORE mainly

SHORE: A Model-driven Approach That Combines Goal, Semantic and Variability Models for Smart HOme self-REconﬁguration

331

Q1: keep user

comfort high

Q2: save energy

consumption

G1: change

temperature

settings

G2: change

air extraction

settings

increase

temperature

decrease

temperature

increase

air extraction

decrease

air extraction

G5: change

temperature sensation

G6: change gas

concentration

Q2: save energy

consumption

plan

reconﬁguration

G3: regulate

temperature

G4: regulate

air quality

enact

reconﬁguration

Q1: keep user

comfort high

analyse

changes

quality

goal

contribution

AND/OR

decomposition

dependency

LEGEND

task

resource

current

conﬁguration

deploy new

conﬁguration

space of

variants

actor

resident

managing

system

environ-

ment

Smart

Home

actuators

Figure 1: Goal-oriented model for the SHORE (Smart HOme self-REconﬁguration) approach.

supports the analyse and plan steps of the MAPE-K.

4.1 RQ1:

How to deﬁne Smart Home variants at design-time

allowing reconﬁguration at runtime?

To answer RQ1, we use the variables identiﬁed

in the problem formulation (see Section 2): goals,

priorities of these goals, actuators, services offered

by actuators, variants of services settings, quality at-

tributes associated to these variants, the values for

quality attributes and the running/current Smart Home

services conﬁguration. Notice that for this paper, we

only consider the reconﬁguration for services of actu-

ators and not for services of sensors. It is because we

consider that actuators are the only ones to be directly

implied in the changing of the state of Smart Homes.

Whilst sensors are required to evaluate Smart Homes

states gathering environmental contextual data.

Table 1 shows how to represent the identiﬁed vari-

ables using SAREF and/or FMs formalisms. As ob-

served, not all of the variables can be expressed by

one of the two formalisms. To ﬁll this gap, we intro-

duce an extension of SAREF, called SAREF

(Baghli

et al., 2018), and an extension of FMs, called

(Mu

nante et al., 2018), and we integrate them.

SAREF

is mainly dedicated to deﬁne semantically

the Smart Home, whilst FM

complements this def-

inition adding variants for the service conﬁguration.

In SAREF

, actuators are explicitly deﬁned.

The concept of service is central and is highly ﬂex-

ible. It may be composed, a service can be a com-

position of multiple services, and even dedicated to

composition, e.g., an orchestrator service that arbi-

trates conﬂict among concurrent services. It may

be loosely coupled to objects (or composition of ob-

jects) enabling the implementation of a service on

a variety of objects matching the service speciﬁca-

tion and dynamic replacement in case of failure or

other context evolution. On the other hand, in FM

actuators and services are represented as fea-

tures. Thus, services are child features of actuators.

Moreover, dependencies between services can be

expressed through the cross-tree constraints: imply

and exclude. Notice that, FMs tend to represent eve-

rything as features. In order to keep the domain spe-

ciﬁc terminology of Smart Homes, we need to use of

an ontology such as SAREF.

In SAREF

, a service is speciﬁed not only by

its interface but also by its dynamic behaviour using

SWRL rules. Static and dynamic aspects of services

may be customized using proﬁle (goal) and proﬁle

measurements, for instance priorities. A proﬁle

measurement describes the metric used to evaluate the

achievement level of a certain proﬁle. For instance,

in a proﬁle such as energy efﬁciency or renewable

energy, proﬁle measurements could be energy con-

sumption or type of energy source. In FM

, goals

are not explicitly represented but their achievement

are interpreted as the selection of an optimal conﬁgu-

ration that will be “better” than the current conﬁgura-

tion. To do that, a space of variants should be evalua-

ted. Variants are represented as features, and they

represent the alternatives for services conﬁgurations.

Thus, variants are child features of services. To se-

lect “better” conﬁgurations, we use quality attributes

(QAs) that are associated to variants. QAs and proﬁle

measurements accomplish the same objective, how-

ever they are not applied at the same level of analysis.

Proﬁle measurements are related to proﬁle (see (*) in

Table 1) (i.e., goals), while QAs are related to va-

riants. The priorities of goals in FM

are ex-

pressed as the weights for QAs, and the values for

QAs are deﬁned using a default value of QAs. More-

over, the minimum range value and maximum range

MODELSWARD 2022 - 10th International Conference on Model-Driven Engineering and Software Development

332

Table 1: Variables needed for the Smart Home services self-reconﬁguration. Where: min = minimisation, max = maximi-

sation, - means that the variable is not supported by the formalism, + means in addition to its predecessor, and ≡ means

equivalent to its predecessor. (*) Notice that proﬁle measurement is related to proﬁle and not to variants.

Variable SAREF FM SAREF

Goals proﬁle select new

conﬁguration

≡ + optimising

(min, max) QAs

Priorities of Goals - - priority for proﬁle weight of QAs

Actuator device feature actuator ≡

Service of Actuators service child features

of actuators

service / composite

service

≡

Dependency of services - imply, ex-

clude

- ≡

Variants for services - child features

of services

- ≡

QAs for variants - - proﬁle measurement

for proﬁle (*)

attribute where

attribute is QA

Values for QAs - - - default value of

QAs

Running conﬁguration - feature conf. - ≡

value are used to compare QAs of different nature.

Finally, the current conﬁguration is represented by

the feature conﬁguration derived from a FM

model.

4.2 RQ2:

How to select an optimal Smart Home variant to

correct deviations from goals?

As observed, several variables are involved in the

Smart Home services self-reconﬁguration. It quickly

makes the problem more complex to solve with the

exponential growth of variables when new actuators,

services and variants are added to the Smart Home.

Therefore, we address this problem as an optimi-

sation problem (an not an exact problem) allowing

the use of optimisation (meta-heuristic) techniques

widely adopted in the literature.

Changes originated by residents or the environ-

ment can provoke deviations from Smart Home goals

These deviations can be detected through SWRL

rules. If a rule antecedent holds true, it means that

an alert will be sent to the planning of the reconﬁgu-

ration. An alert contains information about the situ-

ation that we cope with, hence it contains constraints

to be respected. It is expressed as thresholds, for in-

stance an attribute value should be greater or less than

Xvalue. Therefore, SHORE not only ﬁnds good so-

lutions for the software conﬁguration, in addition it

ﬁnds constrained optimal solutions for it.

For the plan step, given a FM, the software pro-

duct conﬁguration problem involves the exploration

and selection of the set of products representing op-

timal trade-offs among the various attributes of the

FM. We propose to use the grammar guided genetic

programming (GGGP) (Kifetew et al., 2017b) for ex-

ploring the space of product conﬁgurations (see Fig-

ure 2 (a) and (b)). GGGP is a type of genetic pro-

gramming (GP) that provides a suitable representa-

tion of candidate solutions as trees that capture the re-

lationships among features imposed by the FM. Such

a representation facilitates the generation of candidate

solutions and evolving them to optimal solutions.

GP is part of Evolutionary Algorithms (EA). An

overview of a typical EA is shown in Figure 2 (a). An

EA starts by creating an initial population of individu-

als. It then evaluates each individual in the population

by means of an appropriate ﬁtness function and as-

signs it a ﬁtness value. Once individuals are assigned

ﬁtness values, the EA proceeds by selecting individ-

uals (to become parents) from the current population

and subjects them to the process of recombination or

crossover resulting in offspring. The selection pro-

cess could be performed using a variety of methods

based on the ﬁtness value. Commonly used selec-

tion procedures include ﬁtness proportional selection,

tournament selection, and rank based selection. Once

offspring are produced via recombination, they could

further be subjected to a process of mutation with the

aim of introducing diversity into the population. The

EA then selects, from the combined pool of parents

and offspring, the individuals that form the new popu-

lation in the next generation (survivors). Survivor se-

lection could also be performed in a number of ways

(e.g., based on ﬁtness of individuals or based on age).

This process continues to iterate until some stopping

condition is reached, in which case the EA terminates.

For the SHORE solution, as shown in Figure 2 (a),

the FM is ﬁrst serialised into a grammar (as de-

SHORE: A Model-driven Approach That Combines Goal, Semantic and Variability Models for Smart HOme self-REconﬁguration

333

Figure 2: The SHORE approach: (a) the optimisation process, and (b) a candidate evaluation.

scribed in Section 3), which is then fed to the GGGP-

based search module, which follows the evolution

cycle of a typical evolutionary algorithm, resulting

in a Pareto-optimal set of product conﬁgurations.

Given the nature of the problem at hand, we propose

multi-objective optimisation, in particular we adopt

the dominance-based approach as implemented in the

NSGA algorithm (Deb et al., 2000). The steps execu-

ted by the SHORE solution are described as follows:

Candidate Encoding and Initialisation. Each can-

didate solution is a set of features, represented as a

derivation tree based on the given grammar (an ex-

ample is shown in Figure 2 (b)). Collecting the leaves

of the tree, we get the set of features represented by

the candidate solution (i.e., the combination of va-

riants for the Smart Home services settings accord-

ing to SHORE). The population is initialised by ran-

domly generating candidates from the grammar follo-

wing the process of derivation (Kifetew et al., 2017b).

Candidate solutions are by deﬁnition valid with res-

pect to the FM (grammar).

Fitness Evaluation. The ﬁtness function is computed

based on the values of QAs deﬁned a priori. Then,

given a candidate conﬁguration solution Con f ig =

{variant

1.n

,variant

P.m

} (see Figure 2 (b)), the ﬁtness

value of Con f ig is composed of the aggregate values

of each QA in Con f ig. Each attribute is an objec-

tive to be optimised. For example, the sum of the

cost of all features (i.e., variants for the Smart Home

services settings) in Con f ig represents one objective,

while the sum of value levels constitutes another ob-

jective. Notice that, candidate solutions whose ﬁtness

values or speciﬁc attributes do not respect restrictions

imposed by alerts (analysis step) are discarded.

Search Operators. Since candidate solutions are en-

coded as trees, the operators we employ are sub-

tree crossover and subtree mutation (Kifetew et al.,

2017b). Subtree crossover: given two parents, sub-

trees of the same type (rooted at the same gram-

mar non-terminal) are identiﬁed in each parent and

swapped, producing an offspring. Subtree mutation:

an individual is transferred by replacing a subtree by

a newly generated subtree from the grammar.

5 RELATED WORKS

IOT-O (Seydoux et al., 2016) supports self-

reconﬁgurable IOT systems by combining state-of-

the-art ontologies such as Semantic Sensor Network

(SSN). Moreover, in (Lork et al., 2019), an ontology-

based framework for the energy management in build-

ings is presented. We differ from these two re-

ferred works in the formulation of the reconﬁguration

planner. While these works focus on rule-based ap-

proaches without any strategy for managing conﬂicts

between the adaptation goals, we address this issue

by an heuristic algorithm that manages conﬂicts and

scales the problem formulation.

Regarding quality metrics for SASs, QoS-

MOS (Calinescu et al., 2011) deﬁnes a service-based

application as a Discrete Time Markov Chain to iden-

tify the service conﬁgurations that satisfy the quality

goals. Moreover, MARTAS (Weyns et al., 2018) com-

bines quality models at runtime and statistical tech-

niques to predict situations that require adaptation for

IoT systems. However, no details about the decision

making process for the planning adaptation were pro-

vided. The work presented in (De Sanctis et al., 2019)

is the closest to SHORE. It allows to deﬁne QAs asso-

ciated to the devices settings that guide the selection

of optimal conﬁgurations. However, nothing is men-

tioned about potential conﬂicts between adaptation

goals and the technique employed to implement the

optimisation problem. SHORE includes priorities to

manage conﬂicts between adaptation goals and imple-

ments a GGGP for the selection of optimal conﬁgura-

tions. GGGP generates valid candidates for the given

MODELSWARD 2022 - 10th International Conference on Model-Driven Engineering and Software Development

334

grammar, thus no validation steps is needed once can-

didates are generated.

6 CONCLUSION AND NEXT

STEPS

In this paper, we introduced a new approach called

SHORE. It is composed of a set of artifacts that al-

low the reconﬁguration of Smart Home services set-

tings. SHORE enhances Smart Home automation by

combining the ﬁne grain expressiveness of ontologies

together with feature models allowing to determine at

design-time different variants of a Smart Home be-

haviour by sharing and prioritising goals. At runtime,

when conﬂicting goals appear in candidate conﬁgu-

rations, genetic programming allows to automate fur-

ther the decision process. Our contributions include

extensions of the SAREF ontology and FMs, the for-

mulation of a genetic programming technique that ex-

ploits these formalisms, and their implementation for

a case study using a set of open-source tools.

We are currently working on complementing the

SHORE approach to determine at runtime deviations

from Smart Home goals as early as possible. and

also to translate a calculated conﬁguration to the tar-

get conﬁguration of the involved appliance. Further-

more, empirical evaluations are part of our next steps

to validate the performance and usability of SHORE.

REFERENCES

Baghli, R., Najm, E., and Traverson, B. (2018). Deﬁning

services and service orchestrators acting on shared

sensors and actuators. In 6th Int. Conf. MODEL-

SWARD, Funchal, Madeira.

Bajaj, G., Agarwal, R., Singh, P., Georgantas, N., and Is-

sarny, V. (2018). 4W1H in IoT Semantics. IEEE Ac-

cess, 6.

Batory, D. S. (2005). Feature Models, Grammars, and

Propositional Formulas. In 9th Int. Conf. on Software

Product Lines (SPLC), Rennes, France, pages 7–20.

Bencomo, N., G

otz, S., and Song, H. (2019). Mo-

dels@run.time: a guided tour of the state of the art

and research challenges. Softw. Syst. Model., 18(5).

Breaux, T. D. and Ant

on, A. I. (2008). Analyzing regula-

tory rules for privacy and security requirements. IEEE

Trans. Software Eng., 34(1):5–20.

Calinescu, R., Grunske, L., Kwiatkowska, M. Z., Miran-

dola, R., and Tamburrelli, G. (2011). Dynamic QoS

Management and Optimization in Service-Based Sys-

tems. IEEE TSE, 37(3).

amara, J. et al. (2017). Self-aware Computing Systems:

Related Concepts and Research Areas. In Self-Aware

Computing Systems, pages 17–49. Springer.

De Sanctis, M., Spalazzese, R., and Trubiani, C. (2019).

Qos-based formation of software architectures in the

internet of things. In 13th ECSA, Paris, France, vol-

ume 11681 of LNCS. Springer.

Deb, K., Pratap, A., Agarwal, S., and Meyarivan, T.

(2000). A fast elitist multi-objective genetic algo-

rithm: NSGA-II. IEEE Trans. on Evolutionary Com-

putation, 6:182–197.

Erazo-Garz

on et al. (2021). Models@ runtime and internet

of things: A systematic literature review. In 2d Intl

Conf. ICI2ST. IEEE.

Kang, K., Cohen, S., Hess, J., Novak, W., and Peterson,

A. (1990). Feature-oriented domain analysis (FODA)

feasibility study. Technical Report CMU/SEI-90-TR-

21, CMU SEI.

Kephart, J. O. and Chess, D. M. (2003). The vision of auto-

nomic computing. Computer, 36(1):41–50.

Kifetew, F. M., Mu

nante, D., Gorro

nogoitia, J., Siena, A.,

Susi, A., and Perini, A. (2017a). Grammar based ge-

netic programming for software conﬁguration prob-

lem. In 9th Int. Symp. on Search Based Software Eng.

SSBSE, volume 10452 of LNCS. Springer.

Kifetew, F. M., Tiella, R., and Tonella, P. (2017b). Gen-

erating valid grammar-based test inputs by means of

genetic programming and annotated grammars. Em-

pirical SW Eng., 22(2):928–961.

Lork, C., Choudhary, V., Hassan, N. U., Tushar, W., Yuen,

C., Ng, B. K. K., Wang, X., and Liu, X. (2019).

An Ontology-Based Framework for Building Energy

Management with IoT. MDPI Electronics, 8(5).

nante, D., Kifetew, F. M., Gorro

nogoitia, J., Schaniel,

R., Perini, A., and Susi, A. (2018). Model Driven Soft-

ware Reconﬁguration by Exploiting Grammar Based

Genetic Programming. In 8th IEEE MoDRE@RE

Workshop, Banff, Canada.

Seydoux, N., Drira, K., Hernandez, N., and Monteil, T.

(2016). IoT-O, a Core-Domain IoT Ontology to Rep-

resent Connected Devices Networks. In 20th Intl.

Conf. EKAW, volume 10024 of Springer LNCS.

Smart Appliances (2017). SmartM2M; Smart Appliances;

Reference Ontology and oneM2M Mapping.

Weyns, D. (2019). Software Engineering of Self-adaptive

Systems. In Handbook of Software Engineering,

pages 399–443. Springer.

Weyns, D., Iftikhar, M. U., Hughes, D., and Matthys, N.

(2018). Applying architecture-based adaptation to au-

tomate the management of internet-of-things. In 12th

ECSA, volume 11048 of LNCS. Springer.

Yu, E. S. K. (1997). Towards modeling and reasoning sup-

port for early-phase requirements engineering. In 3rd

IEEE Intl Symp. on Requirements Engineering (RE),

Annapolis, MD, USA.

SHORE: A Model-driven Approach That Combines Goal, Semantic and Variability Models for Smart HOme self-REconﬁguration

335