Hybrid Genetic Algorithms

Hybrid Genetic Algorithms;Genetic Algorithms;Local Search;Evolutionary Algorithms

Partheepan, Rajasooriyar ; Dr. Ranji S. Ranjithan, Committee Chair,Dr. E. Downey Brill, Committee Member,G. (Kumar) Mahinthakumar, Committee Member,Partheepan, Rajasooriyar ; Dr. Ranji S. Ranjithan ; Committee Chair ; Dr. E. Downey Brill ; Committee Member ; G. (Kumar) Mahinthakumar ; Committee Member

University:North Carolina State University

关键词: Hybrid Genetic Algorithms;    Genetic Algorithms;    Local Search;    Evolutionary Algorithms;   

Others  :  https://repository.lib.ncsu.edu/bitstream/handle/1840.16/205/etd.pdf?sequence=1&isAllowed=y

美国|英语

来源: null

【 摘 要 】

Genetic algorithms (GAs), a class of evolutionary algorithms, emerging to be a promising procedure for solving engineering optimization problems. As GAs are able to conduct global search with minimal simplifying assumptions about the problem as well as the corresponding decision space, they offer a good alternative to the many gradient-based nonlinear local search procedures.While the underlying operators of a typical GA are designed for global search, their ability to search locally by exploiting information in the vicinity of apparently good solutions is relatively weak.This results in rapid convergence to a relatively good solution followed by slow improvements to that good solution, making GA computationally inefficient.To alleviate this deficiency, GAs can be integrated with local-search procedures such that the strengths of both global and local search approaches are embedded into a hybrid search procedure. One approach is to couple sequentially a GA-based global-search with a local-search procedure. Alternatively, local-search steps can be integrated within the GA operators to potentially refine the solutions throughout the global-search steps in the GA.This research investigates four hybrid search procedures, one based on a sequential approach and the others as local-search-based operators within a GA.These methods and a simple GA are evaluated using a set of test problems, and their performance (in terms of solution quality and computation time) is compared.These performance comparisons are conducted for multiple random trials.These methods are also applied and tested on a realistic urban runoff control problem.Compared to a simple GA, all methods perform well in terms of solution quality and number of fitness evaluations (used as a surrogate for computational resource needs).One of the local-search-based operator methods outperforms others consistently and exhibits a robust performance, indicating a promising hybrid search approach to solving real engineering optimization problems.