Hybridization Approach Towards Improving the Performance of Evolutionary Algorithm
- PDF / 2,073,613 Bytes
- 22 Pages / 595.276 x 790.866 pts Page_size
- 108 Downloads / 225 Views
RESEARCH ARTICLE-COMPUTER ENGINEERING AND COMPUTER SCIENCE
Hybridization Approach Towards Improving the Performance of Evolutionary Algorithm Zainab Al Ani1 · Ashish M. Gujarathi1
· G. Reza Vakili-Nezhaad1 · Ala’a H. Al-Muhtaseb1
Received: 8 February 2020 / Accepted: 17 September 2020 © King Fahd University of Petroleum & Minerals 2020
Abstract Multi-objective differential evolution (MODE) algorithm has been widely used in solving multi-objective optimization problems. In this paper, a hybridization technique is proposed to improve the performance of MODE algorithm in terms of speed and convergence. The proposed hybrid MODE-dynamic-random local search (HMODE-DLS) algorithm combines MODE and dynamic-random local search (DLS) algorithm. To evaluate the proposed algorithm and validate its performance, benchmark test problems (both constrained and non-constrained) are considered to be solved using MODE and the proposed HMODE-DLS algorithms. To compare between the two algorithms, five performance metrices are calculated, which are convergence, spread, generational distance, spacing and hypervolume ratio. Mean and standard deviation values for the performance metrics are reported, and the best in each category is highlighted. The Conv metric results of the new hybrid MODE are compared with other reported ones. Additionally, the effect of local search probability is studied for selected problems. In general, HMODE-DLS performance outshines, in terms of convergence and robustness, compared with other tested algorithms. HMODE-DLS is, generally, faster, and its results are of improved quality compared to MODE algorithm. Keywords Differential evolution · Evolutionary algorithms · Hybrid algorithms · Multi-objective optimization
1 Introduction The problems having more than one objective with a trade-off nature among them are referenced as multi-objective optimization (MOO) problems. Due to the conflicting nature of these problems, they are largely found in both real and theoretical worlds. The usual expected outcome for a MOO problem is a set of non-dominated solutions known as Pareto front, which provides quite enough number of equally good solutions for the decision maker to select [1–3]. Multi-objective evolutionary algorithms (MOEA’s) are one of the famous effective approaches for solving MOO problems. In these algorithms, which are inspired by the nature surviving population, only the solutions that are not dominating each other stay in the population in each generation. As a working mechanism, MOEA’s begin working by generating random population based on considered decision
B 1
Ashish M. Gujarathi [email protected]; [email protected] Department of Petroleum and Chemical Engineering, College of Engineering, Sultan Qaboos University, Al-Khod, P.O. Box 33, 123 Muscat, Sultanate of Oman
variables’ ranges. Various reproduction-based approaches are used in different evolutionary algorithms to generate the offspring solutions from the parent solutions. Then, in each generation, the best non-dominated solutions a
Data Loading...
