Tuning a Simulated Annealing metaheuristic for cross-domain search

Simulated Annealing is a well known local search metaheuristic used for solving computationally hard optimization problems. Cross-domain search poses a higher level issue where a single solution method is used with minor, preferably no modification for solving characteristically different optimisation problems. The performance of a metaheuristic is often dependant on its initial parameter settings, hence detecting the best configuration, i.e. parameter tuning is crucial, which becomes a further challenge for cross-domain search. In this paper, we investigate the cross-domain search performance of Simulated Annealing via tuning for solving six problems, ranging from personnel scheduling to vehicle routing under a stochastic local search framework. The empirical results show that Simulated Annealing is extremely sensitive to the initial parameter settings leading to sub-standard performance when used as a single solution method for cross-domain search. Moreover, we demonstrate that cross-domain parameter tuning is inferior to domain-level tuning highlighting the requirements for adaptive parameter configurations when dealing with cross-domain search