遺傳子 알고리듬의 效率性 增大에 關한 硏究
(A) Study on the Efficiency Enhancement of Genetic Algorithm(GA)
유전자 알고리듬 효율성 설계 최적화 설계민감도 이론;
- 원문 URL
Among the optimization methods, GA(Genetic Algorithm) is a very powerful searching method enough to compete with those based on design sensitivity analysis. GA is very easy to apply and simple to use. However, on the other hand, GA has a couple of disadvantages. It is computationally not efficient, and meanders near the optimum. For some problems, it converges to a wrong solution due to a genetic drifting. In this study, three ingredients are introduced to minimize the inherent disadvantages of GA. Firstly, parallel computation is adopted to improve computational efficiency. GA, by nature, has to perform a huge amount of repetitive computation. To improve the processing efficiency of GA, Paralleled GA(Paralleled GA) is introduced on a clustered Linux based personal computer system. Computation is distributed among the clustered computers, and can be finished within a relatively short period of time as a result. Secondly, gray coded encoding is introduced. Gray code is an encoding of numbers so that adjacent numbers have a single digit differing by one. A binary gray code with n digits corresponds to a hamiltonian path on an n-dimensional hypercube (including direction reversals). The term gray code is open used to refer to a reflected code, or more specifically still, the binary reflected gray code. This study proposes FGGA(Full Gray code GA) applying a gray code throughout all basic operation of GA, which has a good data processing ability to improve the slow convergence of binary code GA. GA meanders near the optimum, and at the same time, shows a phenomenon such as genetic drifting which converges to a wrong solution. These defects are the reasons why GA is not widely applied to real world problems. However, these problems can be overcame by introducing gray coded GA. Thirdly, ExpGA(Experience GA) is introduced. SGA(Standard GA) works fine on small to medium scale problems, but shows poor behavior for large-scale problems. This is the reason that large-scale problems with more than 500-bit of gene's have never been tested and published in papers. The searching ability of SGA fails to work on optimizing the problem that has 96 design valuables and 1536 bits of genetic information. So it converges to a solution which is not considered as a global optimum. This study proposes ExpGA which is a new genetic algorithm made by applying a new probability parameter called the experience value. Furthermore, this study finds the solution throughout the whole field searching. By applying ExpGA to large-scale optimization of a structure, optimization close to the best fitted value is obtained. The problem is known to have genetic drifting when the standard GA is used.