variable population models in a neural network- augmented genetic algorithm for shape optimization

The optimization process for an airfoil using genetic algorithm has been an increasingly popular problem in recent years. in recent years, the role of the population model in genetic algorithm has been underlined. in many of the recently proposed models, the convergence time was adversely increased or the elitism or mutation operators failed to work properly due to the inherent oscillations in the oncoming generations. in this paper, the idea of continuous variable population size has been introduced to optimize the airfoil shape. this scheme has been shown to converge to higher performance airfoils and can decrease the convergence time, without any oscillatory behavior. furthermore, to reduce the run time to evaluate the fitness value, a generalized regression neural network has been developed and trained by the numerical data to evaluate the lift to drag ratio for a vast range of naca four digits airfoils. the values predicted by this neural network have been proved to be in good agreement with the other experimental and numerical data and were then used to calculate the lift-to-drag ratios as the fitness value for various airfoils generated during the optimization process. the idea can ever be more effective in similar problems with a huge amount of computational time to calculate the fitness values and converge to the most efficient airfoil in a reasonable time.