A FRAMEWORK FOR THE IMPROVEMENT OF COMBINATORIAL OPTIMIZATION: AN EXPERIENCE FEEDBACK APPROCH

The problem of selection of project scenarios in a multi-objective context is difficult to solve in an optimal manner. The idea proposed in this paper consists in using previous experiences in order to accelerate the search process for solutions and to limit a too important combinatorial explosion. A framework which allows re-using knowledge resulting from experience feedback is proposed in this paper. This knowledge is capitalized from experiences about previously planned projects in order to guide and to refine the search for new solutions. The method for selection of scenarios is based on an evolutionary algorithm. This algorithm is modified in order to allow the reuse of capitalized knowledge. The knowledge is generated and updated starting from the results of the evaluation of the scenarios by the evolutionary algorithm and/or from the analysis of the projects previously carried out. They are gathered in an influence diagram (extension of the Bayesian networks for the decision-making) allowing their reuse by the algorithm.