Current state-of-the-art methodologies are mostly developed for stationary optimization problems. However, many real-world problems are dynamic in nature, where different types of changes may occur over time. Population-based approaches, such as evolutionary algorithms, are frequently used for solving dynamic environment problems. Selection hyper-heuristics are highly adaptive search methodologies that aim to raise the level of generality by providing solutions to a diverse set of problems having different characteristics. In this study, the performances of 35 single-point-search-based selection hyper-heuristics are investigated on continuous dynamic environments exhibiting various change dynamics, produced by the Moving Peaks Benchmark generator. Even though there are many successful applications of selection hyper-heuristics to discrete optimization problems, to the best of our knowledge, this study is one of the initial applications of selection hyper-heuristics to real-valued optimization as well as being among the very few which address dynamic optimization issues using these techniques. The empirical results indicate that learning selection hyper-heuristics incorporating compatible components can react to different types of changes in the environment and are capable of tracking them. This study shows the suitability of selection hyper-heuristics as solvers in dynamic environments.
Abstract. Hyper-heuristics are high level methodologies that perform search over the space of heuristics rather than solutions for solving computationally difficult problems. A selection hyper-heuristic framework provides means to exploit the strength of multiple low level heuristics where each heuristic can be useful at different stages of the search. In this study, the behavior of a range of selection hyper-heuristics is investigated in dynamic environments. The results show that hyper-heuristics embedding learning heuristic selection methods are sufficiently adaptive and can respond to different types of changes in a dynamic environment.
Abstract. Dynamic environment problems require adaptive solution methodologies which can deal with the changes in the environment during the solution process for a given problem. A selection hyper-heuristic manages a set of low level heuristics (operators) and decides which one to apply at each iterative step. Recent studies show that selection hyperheuristic methodologies are indeed suitable for solving dynamic environment problems with their ability of tracking the change dynamics in a given environment. The choice function based selection hyper-heuristic is reported to be the best hyper-heuristic on a set of benchmark problems. In this study, we investigate the performance of a new learning hyper-heuristic and its variants which are inspired from the ant colony optimisation algorithm components. The proposed hyper-heuristic maintains a matrix of pheromone intensities (utility values) between all pairs of low level heuristics. A heuristic is selected based on the utility values between the previously invoked heuristic and each heuristic from the set of low level heuristics. The ant-based hyper-heuristic performs better than the choice function and even its improved version across a variety of dynamic environments produced by the Moving Peaks Benchmark generator.
The generalized assignment problem is a well-known NP-complete problem whose objective is to find a minimum cost assignment of a set of jobs to a set of agents by considering the resource constraints. Dynamic instances of the generalized assignment problem can be created by changing the resource consumptions, capacity constraints and costs of jobs. Memorybased approaches are among a set of evolutionary techniques that are proposed for dynamic optimization problems. On the other hand, a hyper-heuristic is a high-level method which decides an appropriate low-level heuristic to apply on a given problem without using problem-specific information. In this paper, we present the applicability of hyper-heuristic methods for the dynamic generalized assignment problem. Our technique extends a memory-based approach by integrating it with various hyperheuristics for the search population. Experimental evaluation performed on various benchmark instances indicates that our hyper-heuristic based approaches outperform the memory-based technique with respect to quality of solutions.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.