Hybridization Of Deterministic And Metaheuristic Approaches In Global Optimization

• Main Author: Goh, Khang Wen
• Format: Thesis
• Language: English; English
• Published: 2019
• Subjects: Q Science (General); QA Mathematics
• Online Access: http://eprints.utem.edu.my/id/eprint/24514/1/Hybridization%20Of%20Deterministic%20And%20Metaheuristic%20Approaches%20In%20Global%20Optimization.pdf; http://eprints.utem.edu.my/id/eprint/24514/2/Hybridization%20Of%20Deterministic%20And%20Metaheuristic%20Approaches%20In%20Global%20Optimization.pdf; http://eprints.utem.edu.my/id/eprint/24514/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=117154

In solving general global optimization problems, various approaches methods have been developed since 1970’s which can be divided into two classes named deterministic and the probabilistic/metaheuristic approaches. Deterministic approaches provided a theoretical guarantee of locating the -global optimum solution. However, most of the time deterministic approaches required very high cost and time of computational to obtain the global optimum solution. The probabilistic/metaheuristic approaches are methods based on probability, genetic and evolution as its metaheuristic function for the guidance when solving the global optimization problem, and their accuracy of the solution obtained are not guaranteed. However, some time the metaheuristic approaches work very well in selected problems. The main objective of this research is to increase the accuracy of the solution obtained by Metaheuristic approaches by hybridization with some well-developed local deterministic approaches such as Steepest descent method, conjugate gradient methods and quasi-Newton’s methods. In the analysis of the literature, Artificial Bees Colony (ABC) Algorithm has been selected as the metaheuristic approach to be improved its capability and efficiency to solve the global optimization problems. Several enhancements have been done in this research. For derivative free, a new method called Simplexed ABC method hav an obtained a more accurate global optimum solution by using only 10 colony e been introduced. The numerical results show that Simplexed ABC c of bees with 10 cycle each compare to the 10,000 colony of bees with 100 cycles each in original ABC method. The successful of Simplexed ABC method leads this research to develop a mechanism to transform those well-developed gradient based local deterministic optimization approaches into solving global optimization approaches. These enhancements had produced methods called as ABCED Steepest Descent Method, five variants of ABCED Conjugate Gradient Methods and three variants of ABCED Quasi-Newton’s Methods. The numerical results prove that the enhanced ABCED Steepest Descent and two variants of ABCED Quasi-Newton Methods had perfectly solving all the selected benchmark global optimization problems. In another hand, numerical results of ABCED Conjugate Gradient Methods also achieved up to 80.95% of the selected benchmark global optimization been solved successfully. Besides that, the comparison results also indicated that the numerical performance of the new developed methods converges faster than the original ABC algorithm. The results reported are obtained by using standard benchmark test problems and all computation is done by using C++ programming language.