为解决重型燃气轮机压气机定制叶型设计参数多，传统设计方法效率低的问题，通过人工神经网络方法建立了快速预测重型燃机压气机定制叶型损失和落后角的代理模型。在该模型中，根据重型燃机压气机定制叶型造型方法和S1分析工具建立初始数据库。为进一步提高计算效率，对叶型设计参数进行了敏感度分析，识别叶型设计关键参数并对代理模型降维。结果显示：代理模型具有良好的预测精度，且预测时间相较传统设计方法大幅降低；针对关键参数对代理模型降维，可以降低训练成本，增强模型合理性，提高在叶型优化设计中的应用价值。

In order to solve the problem of low efficiency of traditional design methods due to the multiple design parameters for customized blade of gas turbine compressor, a surrogate model for quickly predicting the loss and deviation angle of customized blade of gas turbine compressors was established based on artificial neural networks. In this model, an initial database was established based on the customized blade shape method for gas turbine compressors and the S1 analysis tool. To further improve computational efficiency, sensitivity analysis was conducted on blade design parameters, identifying key parameters of blade design and reducing the dimensionality of the surrogate model. The results show that the surrogate models have good prediction accuracy, and the prediction time is significantly reduced compared to traditional design methods. Reducing the dimensionality of surrogate models based on key parameters can lower training costs, enhance model rationality, and improve the application value in blade design and optimization.

国家科技重大专项(J2019-II-0005-0025, J2019-II-0004-0024), 航空发动机及燃气轮机基础科学中心项目（P2022-B-V-004-001)