摘要
针对应用遗传算法进行气动优化需要巨大计算量和计算时间的问题,采用将级联前向神经网络作为流场计算的代理模型的方法,能够减少计算量和计算时间.采用类别形状函数(CST)参数化方法,对翼型进行参数化,在限定的范围内随机生成翼型样本,应用样本对级联前向神经网络进行训练,用训练后精度达到要求的级联前向网络作为翼型流场数值计算的代理模型.采用单目标的遗传算法,将级联前向网络和流场数值计算的升阻比作为目标函数,将翼型的CST参数作为单位个体的所有基因,对标准翼型进行优化.数值试验表明,用级联前向网络计算出的升阻比可以达到进行气动优化所需要的精度要求,对于给定的优化目标可以节约大量的计算时间.
The cascade feedforward neural network was applied as a surrogate model in order to solve the problem that aerodynamic optimization based on the genetic algorithm consumed larger amounts of computing resource and time.The computing resource and time can be reduced.The class-shape function transformation(CST)method was used to parameterize an airfoil.The samples of airfoils were randomly generated in a constrained space,and the samples were employed to train the cascade feedforward network.The trained cascade feedforward network with required accuracy was served as the surrogate model to replace the fluid dynamics solver for computing fluid field around an airfoil.The lift-drag ratio that was correspondingly calculated by the cascade feedforward network and fluid dynamics solver was employed as the objective function by using the single objective genetic algorithm.The CST parameters of the sharp function of the airfoil were selected as the genes of an individual in order to optimize the original airfoil.The numerical experiments showed that the cascade feedforward network provided the required accuracy with a significant reduction of computing time under a specific optimization objective.
出处
《浙江大学学报(工学版)》
EI
CAS
CSCD
北大核心
2017年第7期1405-1411,共7页
Journal of Zhejiang University:Engineering Science
基金
国家自然科学基金青年资助项目(61501399)
关键词
级联前向神经网络
遗传算法
气动优化设计
代理模型
cascade feedforward neural network
genetic algorithm
aerodynamic optimization design
surrogate model