HyperILS: An Effective Iterated Local Search Hyper-heuristic for Combinatorial Optimisation

Two powerful ideas from search methodologies, iterated local search and hyperheuristics, are combined into a simple and effective framework to solve combinatorial optimisation problems (HyperILS). Iterated local search is a simple but successful algorithm. It operates by iteratively alternating between applying a move operator to the incumbent solution and restarting local search from the perturbed solution. This search principle has been rediscovered multiple times, within different research communities and with different names [2,12]. The term iterated local search (ILS) was proposed in [11]. Hyper-heuristics [4, 6,7] are a recent trend in search methodologies motivated (at least in part) by the goal of automating the design of heuristic methods to solve computational search problems. The aim is to develop more generally applicable methodologies. Metaheuristics are often used as the search methodology in a hyper-heuristic approach (i.e. a metaheuristic is used to search a space of heuristics). Machine learning approaches can and have also been used as the high-level strategy in hyper-heuristics such as reinforcement learning, case based reasoning, and learning classifier systems [4]. The ILS hyper-heuristic discussed here uses a form of reinforcement learning to adaptively select the best operator/heuristic to apply at each iteration (in either or both the perturbation and improvement stages) from an available pool of operators with different features. It differs from a standard ILS implementation which uses a