GRASP with path-relinking (GRASP+PR) is a metaheuristic for finding optimal or near-optimal solutions of combinatorial optimization problems. This paper proposes a new automatic parameter tuning procedure for GRASP+PR heuristics based on a biased random-key genetic algorithm (BRKGA). Given a GRASP+PR heuristic with N input parameters, the tuning procedure makes use of a BRKGA in a first phase to explore the parameter space and set the parameters with which the GRASP+PR heuristic will run in a second phase. The procedure is illustrated with a GRASP+PR for the generalized quadratic assignment problem with N=30 parameters. Computational results show that the resulting hybrid heuristic is robust.
Citation
AT&T Labs Research Technical Report, AT&T Labs Research, Florham Park, NJ 07932, Feb, 2010.