الفهرس 
BACKGROUND AND LITERATURE REVIEW ThisOnly 14 pages are availabe for public view
 |  | from | 135 |  |  |
| | | from | 135 | | |
Abstract
Combinatorial optimization problems (COPs) are computationally challenging and
demand domain-specific knowledge-based methods that are not reusable across
different problem domains. In contrast, researchers are striving to develop more general
solution methods that exhibit effectiveness across several problem domains. Hyperheuristics are a widely adopted method for solving complex computational search
problems due to their capability to generalize across various problem domains.
Selection hyper-heuristics search through the space of heuristics by combining and
managing a set of low-level heuristics for tackling computationally difficult
combinatorial optimization problems. 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 problemspecific heuristics. This thesis presents a comprehensive Multi-level hyper-heuristic
(MHH) framework that facilitates the utilization and leverages the advantages of
different hyper-heuristic selection methods and multiple acceptance criteria throughout
the search process. This is achieved by incorporating an additional level strategy, known
as the highest-level strategy, into the hyper-heuristic framework This highest-level
strategy adapts by selecting the suitable hyper-heuristic based on its performance during
the search process. Within this strategy, an appropriate algorithm is chosen from a
predefined set of hyper-heuristic algorithms to enhance the generated solution. The
primary aim is to create a methodology that employs multiple hyper-heuristics to
achieve improved effectiveness, a higher level of generality, and enhanced efficiency
in overall performance when compared to the individual effects of each constituent
selection hyper-heuristic. Therefore, various multi-level hyper-heuristic frameworks
are deployed across six HyFlex problem domains and additionally evaluated on three
extended HyFlex problem domains: 0-1 Knapsack, Quadratic Assignment, and MaxCut. The empirical results indicate that the proposed framework exhibits robust
generalization capabilities, demonstrating strong performance not only within the
ABSTRACT
IV
standard HyFlex problem domains but also across the three extended ”unseen”
problems. When compared to a set of state-of-the-art hyper-heuristics, The proposed
framework has demonstrated excellence in its capacity for generalization, reusability,
and ease of implementation