Multi-level Hyper-Heuristic for Combinatorial Optimization Problems

Abstract

Hyper-heuristics are considered as one of the most popular search methods which can solve NP-hard problems. They aim to achieve level of generality of search techniques for solving a wide variety of problem domains. Hyper-heuristic framework involves two levels, high-level, and low-level heuristics. The high-level heuristic is responsible for selecting and applying an appropriate low-level heuristic to generate solutions and deciding whether to accept or reject the new solution. Low-level heuristics are a set of problem-specific heuristics. In this paper, we propose to improve the performance through adding a new level strategy to the hyper-heuristic framework. The highest-level strategy adopts the roulette wheel selection mechanism to select the appropriate hyper-heuristic according to its performance during the search process. The highest-level strategy selects the appropriate algorithm from a predefined set of hyper-heuristic algorithms to improve the generated solution. The performance of the proposed approach has been compared with one of the most recent methods as well as with other hyper-heuristics that used as components of the proposed approach. The results have been carried on six of commonly used benchmark datasets. The results show the effectiveness of the proposed framework. Hence it outperforms the other methods in five of the six benchmarks.