Preliminary aerodynamic shape optimization using genetic algorithm and neural network

To reduce the expensive computational cost in genetic algorithm, approximation model is suggested to evaluate individual's fitness. However, for its approximation error the approximation model is likely to mislead the search in evolution. To avoid the danger, a fraction of the individuals should be evaluated with exact function or be controlled in other words. In this paper, a new method is proposed which combining generation based and individual based control method. To prevent the good schema from being lost during evolution, the exact function is used when good schema is found. The test cases show that this method is efficient and effective for high dimensional multimodal functions and aerodynamic shape optimization.