Application des techniques de l'intelligence artificielle à la géoprospective: le projet Expect - Application of Artificial Intelligence techniques to géoprospective: the Expect project

International audienceDifficulties in the géoprospective approach arise from the diversity and nature of accumulated phenomenological data, the necessity of guaranteeing compatibility between hypotheses introduced into scenarios of future evolution, and finally from the necessity of substituting, for the system considered, a train of discontinuous states by a description of its progressive evolution . These specific constraints have led BRGM to consider using techniques of Artificial Intelligence and qualitative reasoning in the field of géoprospective; this is the aim of the Expect project (EXpert system applied to the prediction of the site Evolution Context) being carried out in collaboration with the Laboratoire de Recherche en Informatique (Informatics Research Laboratory) of the University of Orsay. Several specificities in the field of géoprospective have guided the selection of these techniques: - A scenario can be considered as a chain of causality beginning with likely hypotheses concerning the occurrence of phenomena, founded on knowledge of past climatic and geological history. It is a multi-expert problem. - Géoprospective is based, above all, on a qualitative vision of phenomena; this is related to the type of data manipulated (often empirical and interpretative) and the complexity of the field. Although it is possible to establish that one phenomenon has an influence on another, and in what way, this knowledge cannot generally be translated into an exactly corresponding numerical model. - The quantitative data used are often marred by uncertainties that increase with time. - Numerous parameters have notable values (which may be expressed as intervals) that, associated with the direction of parameter variation, provide pertinent information on the evolution of the system. This poster presents the main concepts associated with this technique, a taxonomy of the knowledge that is manipulated in géoprospective, the formalism developed and the modalities for using such a system. A few examples are given to illustrate these points