燃气轮机高压透平进口温度场受到燃烧室结构、冷却系统及燃烧状态的综合作用，分布极其不均匀。燃烧室出口燃气高温核心区域被称为“热斑”，其温度明显高于其他区域。热斑进入燃气透平通道后，不仅会与叶栅通道内二次流相互作用带来额外损失并降低叶栅效率，而且会引起透平叶片表面局部过热，从而降低叶片使用寿命。但是目前透平叶片优化研究工作多采用均匀进口条件，因此，需要建立考虑进口热斑条件的透平叶片优化设计系统，优化透平进口导叶，在保证效率前提下，降低热斑带来的影响。 本文开发了透平叶片多目标优化设计系统，包括参数化造型模块、网格生成与变形模块、气动评估模块、优化算法模块。优化系统的参数化造型模块采用了自由变形方法，网格生成与变形模块采用基于径向基函数的插值方法，优化算法模块采用第二代非支配排序遗传算法。将优化系统应用于单级跨音透平叶片VKI-LS59的优化设计，优化后叶片尾缘附近的激波强度减弱，绝热效率提高了0.916%，从而验证了优化设计系统。 本文构建了人工神经网络代理模型以替代CFD方法进行叶片气动评估，提高优化系统效率。本文采用3000个训练样本分步训练神经网络，最终训练误差达10-5，样本预测误差控制在1.5%以内，模型预测精度较高，可以在优化早期有效替代CFD方法，加速优化设计进程。本文研究了热斑条件下目标函数的敏感性，并设计了叶型控制点参数自适应程序，增强了优化系统的准确性与稳定性。本文先求取控制点参数与目标函数的相关性，再分析高相关性控制点参数与目标函数的敏感性。控制点参数的相关性大小决定控制点变化幅度大小，而控制点参数的敏感性大小决定控制点变化增量大小。 最后基于本文构建的透平叶片多目标优化设计系统，在热斑条件下对GE-E3进口导叶进行了优化设计，优化后叶片绝热效率提高0.52%，叶片表面最高温度降低15.96K，叶片表面平均温度降低12.19K。 综上所述，本文发展了一套考虑进口热斑条件的透平叶片优化设计系统，为燃气轮机透平叶片优化设计提供了有效的指导。

The temperature profiles of the inlet of the high pressure gas turbine are non-uniform for the structure, cooling and combustion mechanism of the gas turbine combustor. The core of the high temperature is the hot streak. The migration of hot streaks on the one hand would interact with the secondary flow of the cascade, on the other hand would excessively heat blade surfaces. Obviously, hot streaks lead the extraneous loss; reduce the stable operation range and shorten the life of blades. However, optimization researches of turbine blades always ignore the non-uniform of turbine inlet. So, it is necessary to develop the multi-objective optimization system of turbine vane with inlet hot streak to weaken damages of hot streaks. This paper developed a multi-objective optimization system of turbine vane, including the Free-Form Deformation parametric modeling method, the Radial Basis Function grid modification method and the Non-dominated Sorting Genetic Algorithm Ⅱ. The validation was conducted on the transonic cascades VKI-LS59. The optimized cascades improved the adiabatic efficiency by more than 0.916%. In order to improve the efficiency of the optimization system, this paper developed an artificial neural network surrogate model to evaluate the aerodynamic performance of turbine vanes. A total of 3000 samples were used to train the neural network, which made the prediction error less than 1.5%. The artificial neural network surrogate model effectively increased the speed of the optimization system. The new sensitivity analysis method was conducted to reduce geometric variables. This paper combined the correlation analysis and the sensitivity analysis. Before the sensitivity analysis, the correlation analysis was carried out after the optimization. High correlation control points were selected for the sensitivity analysis. The correlation coefficient was proportional to the amplitude of the design variable, while the sensitivity coefficient was inversely proportional to the increment of the design variable. Based on the multi-objective optimization system, the subsonic turbine vane GE-E3 was optimized with inlet hot streak. Both of the efficiency and thermal load were improved. The optimized vane improved the adiabatic efficiency by 0.916%, reduced the maximum surface temperature by 15.96 K, and reduced the average surface temperature by 12.19 K. In summary, this paper developed a multi-objective optimization system of turbine vane with inlet hot streaks, which would provide guidance for the design of turbine vanes.