针对高速直升机旋翼翼型的气动优化问题，建立了一种基于深度学习的翼型多目标气动优化框架。首先，搭建深度神经网络预测旋翼翼型气动力系数的代理模型。以SC1095旋翼翼型作为基准翼型，采用CST方法对翼型进行参数化表示，采用拉丁超立方采样方法生成用于深度神经网络训练的翼型数据集。综合考虑直升机前飞、机动和悬停状态等多设计点的气动性能，开展了基于深度神经网络的代理模型结合多岛遗传算法的高速直升机旋翼翼型多目标气动优化设计。优化结果显示：相比于基准翼型，在没有损失悬停和机动性能的前提下，优化翼型显著改善了前飞性能。分别采用基准翼型和优化翼型建立共轴刚性旋翼，计算并分析前飞状态下基准旋翼和优化旋翼的气动性能，结果表明优化翼型可以显著地提升高速直升机旋翼的气动性能。

To optimize the aerodynamic shape of high-speed helicopter rotor airfoils， a multi-objective optimization framework is proposed based on deep learning. Firstly， a deep neural network is constructed as a surrogate model to predict the aerodynamic coefficients of rotor airfoils. The rotor airfoil SC1095 is selected as the baseline airfoil. The Class function/Shape function Transformation （CST） method is employed to parameterize the airfoil， and the Latin hypercube sampling method is used to generate the airfoil dataset for training deep neural networks. Then， comprehensively considering the aerodynamic performance of multiple design points such as forward flight， maneuvering and hover of the helicopter， a multi-objective aerodynamic shape optimization of the high-speed helicopter rotor airfoil is conducted by combining the deep neural network surrogate model with the multi-island genetic algorithm. The optimization results show that compared with the baseline airfoil， the optimized airfoil can significantly improve its forward flight performance without compromising hover and maneuvering performance. Finally， a rigid coaxial rotor is generated using the baseline and optimized airfoil respectively. The aerodynamic performance of these rotors in forward flight is computed and analyzed. The results indicate that the optimized airfoil significantly enhances the aerodynamic performance of the high-speed helicopter rotor.