Application of the artificial memory approach to multicriteria scheduling problems

This paper is devoted to the development of a knowledge-based system (KBS) called ‘Artificial Memory’, The goal of this KBS is to ‘solve’ multicriteria job-shop scheduling problems. Since job-shop scheduling problems are NP-hard, it is extremely difficult to obtain optimal solutions for industrial problems. Thus, a host of heuristic algorithms, most of which are based on priority rules, have been proposed in the literature. The efficiency of these algorithms strongly depends on the criteria to be optimized as well as the values of the parameters associated with the particular instance of the scheduling problem. The basic hypothesis of the artificial memory approach is a continuity assumption: we assume that identical decisions applied to similar instances lead to similar values of the criteria. This assumption is fundamental to validate this knowledge-based system. For each criterion, the artificial memory contains a synthesis of the performances of different algorithms upon sets of ‘similar’ instances. These performances are acquired using simulation. When the artificial memory is employed, the characteristic values of a new instance are computed and examined by the artificial memory system. The performances of the different algorithms for the considered criterion are estimated for the new instance and an appropriate algorithm is chosen accordingly. In order to build this KBS and to estimate the performances of algorithms upon a new instance, we use a mathematical approach. Some difficulties arose in the development of this KBS and had to be overcome: the corresponding proposed solutions are developed. The paper also presents a number of numerical experimental applications.