ASAGA

• DOI: 10.1145/1389095.1389289
• Publication Date: 2008-07-22T09:46:39Z
• Authors: Liang Shi; Khaled Rasheed
• Keywords: Fitness approximation; Surrogate model

Genetic algorithms (GAs) used in complex optimization domains usually need to perform a large number of fitness function evaluations order get near-optimal solutions. In real world application such as the engineering design problems, might be extremely expensive computationally. It is therefore common estimate or approximate using certain methods. A popular method construct so called surrogate meta-model original function, which can simulate behavior but evaluated much faster. difficult determine model should and/or what frequency usage be. The answer also varies depending on individual problem. To solve this problem, an adaptive approximation GA (ASAGA) presented. ASAGA adaptively chooses appropriate type; adjusts complexity and according time spent accuracy. introduces stochastic penalty handle constraints. Experiments show that outperforms non-adaptive surrogate-assisted GAs with statistical significance.