基于BP神经网络代理模型的翼型优化及领域自适应研究  

Research on Airfoil Optimization and Domain Adaptation Based on BP Neural Network Surrogate Model

作　　者：陈晨铭 郭雪岩[1] 常林森 CHEN Chenming;GUO Xueyan;CHANG Linsen(School of Energy and Power Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学能源与动力工程学院,上海200093

出　　处：《动力工程学报》2022年第7期657-663,共7页Journal of Chinese Society of Power Engineering

摘　　要：采用神经网络代理模型和遗传算法相结合的方法对NACA64(3)-618风力机翼型进行了气动优化。针对青藏高原风场条件下某一工况进行优化时,利用拉丁超立方采样生成参数样本集、通过B样条曲线对翼型进行光滑化处理、采用基于深度前馈网络的代理模型预测了升、阻力系数,并结合遗传算法实现了气动优化选型,利用CFD方法验证了优化结果。结果表明:优化翼型的升阻比和升力系数分别提高了4.52%和4.05%,阻力系数降低了0.42%;优化流程能用低维参数表达比较完整的翼型,代理模型能在严苛条件下得到较好的翼型;阻力系数代理模型的精度较高,明显优于升力系数代理模型,而且阻力系数代理模型在领域自适应方面表现良好。The aerodynamic optimization of NACA64(3)-618 wind turbine airfoil was carried out by using the combination of neural network surrogate model and genetic algorithm.In the optimization for a certain working condition of wind field conditions in the Qinghai-Tibet Plateau,the parameter sample set was generated with Latin hypercube sampling,the airfoil was smoothed by B-spline curve,the lift coefficient and drag coefficient were predicted by surrogate model based on deep feedforward network,and the aerodynamic optimization selection was realized by combining with genetic algorithm.The optimization results were finally verified by using CFD simulation method.Results show that the lift drag ratio and lift coefficient of the optimized airfoil are increased by 4.52%and 4.05%respectively,and the drag coefficient is reduced by 0.42%.The optimization process can express a relatively complete airfoil with low dimensional parameters,and the surrogate model can get a better airfoil under severe conditions.The drag coefficient surrogate model has a higher accuracy and better performance in the field of domain adaptation,and is obviously better than the lift coefficient surrogate model.

关 键 词：翼型优化 深度学习 代理模型 深度前馈网络 领域自适应 

分 类 号：TK8[动力工程及工程热物理—流体机械及工程]