A case-based reasoning system that avoids the problem of the case identification

The subject matter of this paper is the prediction of evolution of complex physical phenomena. The method proposed is based on observed “behaviors”. We utilize case-based reasoning. A case describes a behavior thanks to the succession of the values of the different parameters from which it is derived. The weak knowledge about the domain does not make it possible to represent cases by a set of features as it is usually done. The mechanism that traditionally allows to select the best case in memory has thus been completely reconsidered. Case representation by means of successions of events and matching thanks to a string matching algorithm make up the main originality of our work