¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
Modélisation déclarative et sémantique, ontologies, assemblage
et intégration de modèles, génération de code
Declarative and semantic modelling, ontologies, model
linking and integration, code generation
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
Introduction ( Ù )
Ce dossier traite de la représentation des interfaces des
modèles dans une ontologie. Il est question des bénéfices retirés de cette
approche lorsqu’il s’agit d’assembler et d’intégrer des modèles (partage,
réutilisation, qualité, fiabilité, robustesse). Il est présenté l’ontologie MIO
(Model Interface Ontology) et une mise en
pratique de l’approche sur le projet Seamless-IP par la communauté APES
(Agricultural Production Externalities Simulator).
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
Source ( Ù )
Ce dossier repose sur un article qui est paru lors de la conférence de l’iEMSs 2006 (International Environmental Modelling and Software Society) :
|
« Enriching
software model interfaces using
ontology-based tools » I.N. Athanasiadis a, A.E. Rizzoli a, M. Donatelli b,
and L. Carlini b a Dalle Molle Institute for Artificial
Intelligence, Lugano, Switzerland b CRA – Research Institute for Industrial Crops,
Bologna, Italy http://www.iemss.org/iemss2006 Chemin d’accès complet -
URL
http://www.iemss.org/iemss2006 , -
sous
la rubrique « Sessions »
, -
puis « S5. Integrated software solutions for environmental
problems - architecture, frameworks and data structures (ISESS) (Dave Swayne
and Rob Argent) » , -
puis « Theme 4: Knowledge Engineered Modelling ». |
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
Relevé de quelques éléments ( Ù )
Il est repris ci-dessous – sans aucune complétude - quelques éléments de cet article, dont il est fait des citations (texte entre guillemets).
Différentes
utilisations des ontologies
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
Context, situation ( Ù )
Le contexte, la situation
From the modelled system to the software form
of the model
Du système modélisé à la forme informatique du modèle
Simulating a system includes :
à Firstly, the modelled system : the very
complete and complex real system.
à Secondly, the model : « an
abstraction of the real world processes », « using a given formalism »
; « particular assumptions and hypotheses about the phenomena involved are
made ».
à Thirdly, the software implementation of the
model : « a poor realization of the original formalisation » ; « more
assumptions, more limitations » are introduced (discretization…).
Models are seldom
reused
Les modèles sont rarement réutilisés
« Common practice has proven that software
implementations of environmental models are seldom reused by broader
communities or in different modelling frameworks. One of the reasons for this
situation is the poor semantics of model interfaces. Model interfaces
describe a critical amount of modellers’ knowledge, but their software
implementations fail to represent the complexity of model assumptions in
software terms. »
What « integrating models across scales
and disciplines » implies
Ce qu’implique le fait d’intégrer des modèles de différentes échelles
et disciplines
« Software integration is not the sole
necessary condition for a proper assemblage of […] models. »
« If a set of (good) software model
implementations are working together, this is not at all a sign that the
compound model makes any sense from a modelling point of view and generates
credible results. »
« Sound integration of […] models also
requires automated coupling of the knowledge hidden behind each software
implementation. »
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
« A
declarative approach for describing model interface to facilitate model linking
and integration using ontologies » ( Ù )
Adopter une approche déclarative pour
décrire l’interface des modèles, de sorte à faciliter l’assemblage et
l’intégration des modèles grâce aux ontologies
« Sound model linking and
integration »
Assembler et intégrer correctement des modèles (pertinence, cohérence,
exactitude)
« A software component implementing a
model will consist of two parts, the interface and the implementation. The
interface defines the inputs, outputs and parameters of a model, while the
implementation defines the model equations. »
« Simple integration in software terms is
not enough for sound model integration ». In usual approaches « a software
implementation of a […] model does not take into account the semantics of the
software interface ». « The information associated with the inputs,
states, outputs and parameters is limited to their data type », which is
« not enough for encapsulating the full knowledge of the model
interface ». As a result, « someone has to read the documentation in
order to understand how to reuse [a] model properly ».
The problem (the lack) is that « the model’s
knowledge related to its interface is not encapsulated in the actual interface
of the model implementation in a self-explained fashion ».
A solution would be « to express all the
knowledge related to the model interface in a declarative way, using an
ontology » (expressed concept, characteristic times, units, pre- and post-
conditions, temporal or spatial dimensions, sampling rates…). « Ontologies
provide a formal support to express conceptualisations […]. Furthermore, model
knowledge stored in the ontology can be used both for formal documentation and
provide functionalities which go beyond the computation of model variables ».
The Model Interface Ontology, « an
ontology for specifying model interfaces »
Model Interface Ontology, une ontologie pour spécifier les interfaces
des modèles
The Model Interface Ontology is an ontology « that
aims to encapsulate our knowledge on the model interface in a declarative
fashion ».
In order to represent « biophysical
agricultural models », two model types are identified in the Model
Interface Ontology :
-
Static models (« not required to be integrated
over time ») which have inputs and outputs.
-
Dynamic models (integrated over time) which have
inputs and outputs, states and rates (for stocks and flows).
Measurement : is a class of the Model Interface
Ontology. « The Measurement class is the key instrument for conceptualising the
model interface elements ». « All inputs, outputs, states and rates of
models are types of an abstract Measurement concept (ontology class), which is
used for defining their semantics in different contexts (space, time units, and
so on) ».
Measurement properties : « The Measurement class specifies the following properties
of a model interface element :
-
The
observed quantity.
-
The
spatial observation context.
-
The
temporal observation context.
-
The
sampling frequency.
-
Value
conditions (minimum, maximum and default value and default unit). »
For more details, see the
article in Section 3 « Towards an ontology for specifying model
interfaces », Figure 2 « The relations between the model type
concepts of the model interface ontology » and Figure 3 « The
relations of the Measurement concept ».
The Model Interface Ontology development
Le développement de l’ontologie Model Interface Ontology
The Model Interface Ontology has been developed
using the Web Ontology Language OWL
, through the Protégé ontology
editor (http://protege.stanford.edu/plugins/owl).
« The specifications of units and dimensions
were based on the SWEET ontologies »
(http://sweet.jpl.nasa.gov).
« The Model
Interface Ontology is available online » (http://seamless.idsia.ch/ontologies
, MIO ontology).
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
Tools ( Ù )
Outils
On the Seamless-IP project ( http://www.seamless-ip.org )
Sur le projet
Seamless-IP
The approach, consisting in « publishing
model interfaces in a declarative format » using « an ontology for
capturing the semantics of model interface elements », « was undertaken by the
Seamless-IP project and the community of Agricultural Production Externalities
Simulator (APES) modellers ».
A tool has been developed « to enable modellers
to share their knowledge related to environmental model components and their
interface variables » : AgrOntologies.
A tool has been developed « to enable modellers
to exploit the knowledge stored in the ontology by generating source code in an
automated fashion » : DCC.
« AgrOntologies : a Web-based tool for
communal ontology authoring »
AgrOntologies : un outil Web pour l’écriture collaborative d’une
ontologie
The AgrOntologies tool is « an easy to use
portal » for modellers to « register their models ». Modellers are
not « exposed to all the complexity of the internal ontology structure ».
« Through the AgrOntologies portal, a
modeller can » :
-
« Specify
model variables in detail, or even reuse existing variables defined by others
».
-
« Define
model interfaces ».
-
« Put
models together in components ».
« DCC : a tool for generating model source
code »
DCC : un outil pour générer un code source du modèle
« The application Domain Class Coder (DCC) is a
Windows application » which, from an input file extracted from the ontology
generates C# code.
See URL : http://www.isci.it/tools
(page « Windows XP utility applications »).
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾
Different ontologies uses ( Ù )
Différentes utilisations des ontologies
Ontologies
have been exploited in Environmental Management Information Systems with
different aims :
-
For
« seamless integration of environmental data repositories ».
-
«
More generic approaches for environmental data fusion ».
-
«
For efficient model integration » (a).
-
«
Focusing on extending the current framework by specifying model equations using
semantic modelling primitives » (b).
« Ontology representations of both model interfaces (a) and
equations (b) may
lead [us]
to a fully declarative modelling and simulation environment »
(a) Ontology representation of model
interfaces :
« For model linking and model component integration ». C’est le sujet de l’article.
(b) Ontology representation of model
equations :
« Specifying model equations using
semantic modelling primitives ». Voir
le dossier modelia de l’article « Declarative
modelling for architecture independence an data/model integration : a case
study » de Ferdinando Villa.
La page au format pdf (27/09/06)
-
mise en ligne
le 27/09/06 –
Plate-forme
INRA-ACTA-ICTA, Modelia http://www.modelia.org
¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾¾