http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
경계확장과 복합 유전알고리즘을 이용한 비선형 입체트러스의 최적설계
현승협(Hyun Seung-Hyup),김종범(Kim Jong-Bum),한상을(Han Sang-Eul) 대한건축학회 2003 大韓建築學會論文集 : 構造系 Vol.19 No.5
Genetic algorithms(GAs) are adaptable to discrete and constrained problems and do not require continuity, unimodality, and derivatives of function. But GAs have a demerit of late convergence, which means to be examined many individuals until find the best individual. And especially much time is required to compute the constraints of an individual in large scale or nonlinear structural problem. So it needs to be developed the genetic algorithm to converges fast in small population. Although the selected algorithm is superior, it's performance could be very fluctuate depending on search space. And it is possible to vary and adapt the algorithm to a placed circumstance. Therefore it is very important for user to create more efficient algorithm. In this paper, micro genetic algorithm effective in small population size is used to save computing time in size optimization of geometrically nonlinear space truss. But in practical structural size optimization problem, many design variables are needed and search space is very complicated. Also optimum solutions exist at the corner (boundaries between feasible and infeasible regions) of the search space. In this case, micro GAs exist so far are liable to fall into dead corner and some genetic operators of those do not work well because the outcomes of the operator are apt to be infeasible. To cope with these problem it is proposed to reduce the intensity of penalty in order to extend the boundaries of search space and to use the hybrid genetic algolithm by gradient-like-selector to converge fast and avoid the genetic drift as a result of copies.
경계확장과 복합 유전알고리즘을 이용한 비선형 입체트러스의 최적설계
현승협,이광순,한상을 대한건축학회 2001 대한건축학회 학술발표대회 논문집 - 계획계/구조계 Vol.21 No.2
A genetic algorith(GA) is adaptible to discrete and constrainted problem. But in structure optimization problem, optimum solutions exist at the comer(boundary between feasible and infeasible regions) of the search space and the boundary are much complecated. In this case some crossover operators do not work well because the outcomes of the operator are apt to be infeasible. and especially much time is repuired to estimate the constraints in large scale or nonlinear structures. In this paper, to cope with these problem we propose to reduce the intensity of penalty in order to extend the boundary of search space and to use the hybrid genetic algorithm by gradient-like-selection. Micro GA effective in small population size is used and design variables are selected from practically used standard steel sections. An incremental load approach with a modified Newton-Raphson iteration is used.