Nogood Recording with Tabu Search for CSP (Application to FAP)

In order to increase the effectiveness and the adaptability of methods, the constraints programming has gradually evolved to hybrid algorithms which combine exact and approximated methods in order to get the advantages of each of them. In this paper we propose a deterministic hybrid method which combines the backtracking, the constraints propagation and the Tabu Search. The method is based on finding and storing the variable/value associations that lead to infeasible solutions. It allows the algorithm to dynamically manage the cuts of branches and the access to the decision variables. The work is thus about a learning system which progressively extends the algorithm knowledge on the problem and increases its effectiveness. Benchmarks on frequency assignment problem, CELAR and GRAPH, were used for test. The results are at the level of the best ones currently known and they improve the success percentages on most of these problems.