压气机从设计到产品的发展过程，离不开设计修改与试验测量诊断的多次迭代。现在多级压气机的设计和流动诊断依然强烈地依赖于设计专家的经验与设计数据库。轴流压气机的级间试验数据诊断技术，在压气机设计迭代过程中具有关键的作用，会以非常小的代价，提升压气机的性能。本文提出了基于轴流压气机级间测量参数的子午流场反演方法，通过试验得到级间参数，求解子午面内的主控方程，反演得到了试验中难以直接观测的子午流场，可为多级压气机的匹配调试提供具备流场特征的诊断分析工具。基于数值模拟和试验数据结合的叶栅数据库，通过数据清洗、分类、降维等方法建立了数据驱动的轴流压气机静子展向分布损失模型，并充分考虑压气机试验数据来源可能多且复杂，存在传统的线性回归模型拟合误差大的问题，通过选择多种机器学习算法，岭回归、K近邻回归、支持向量回归、以及随机森林等算法对数据进行建模，提升了轴流压气机各工况下不同展向损失分布的预测精度。最后应用该方法和模型对多台典型压气机进行了流场反演，并详细分析了S2流面的流场和参数分布。

The evolution of a compressor from design to final product necessitates multiple iterations involving design modifications and experimental measurements for diagnostic purposes. Current design and flow diagnosis of multistage compressors still heavily depend on the expertise of design specialists and established design databases. Inter-stage experimental data diagnostic technology for axial compressors plays a critical role in the design iteration loop, offering a cost-effective means to enhance compressor performance. This paper proposes a method for inverting the meridional flow field based on inter-stage measurement parameters in axial compressors. By utilizing experimentally obtained inter-stage data to solve the governing equations within the meridional plane, the flow field—which is difficult to observe directly during experiments—is reconstructed. This provides a diagnostic analysis tool featuring flow field characteristics to assist in the matching and tuning of multistage compressors. Leveraging a cascade database combining numerical simulations and experimental data, a data-driven model for the spanwise loss distribution of axial compressor stators is established through data cleaning, classification, and dimensionality reduction. To address the challenge posed by the potentially diverse and complex sources of compressor experimental data, which can lead to significant fitting errors in traditional linear regression models, various machine learning algorithms—including Ridge Regression, K-Nearest Neighbors Regression, Support Vector Regression, and Random Forest—are employed for data modeling. This approach improves the prediction accuracy of spanwise loss distributions across different operating conditions of the axial compressor. Finally, the proposed method and model are applied to perform flow field inversion for several typical compressors, accompanied by a detailed analysis of the flow field and parameter distributions on the S2 stream surface