基于动态XGBoost与MaxLIPO置信域算法的减重涡轮叶型气动优化设计  

Aerodynamic Optimization Design of Weight Reduction Turbine Blade Based on Dynamic XGBoost and MaxLIPO Trust Region Algorithm

作　　者：张子卿 宋宇宽 王名扬 卢新根[1,2] 张燕峰 ZHANG Zi-Qing;SONG Yu-Kuan;WANG Ming-Yang;LU Xin-Gen;ZHANG Yan-Feng(Key Laboratory of Light-duty Gas-turbine,Institute of Engineering Thermophysics,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院工程热物理研究所/轻型动力重点实验室,北京100190 [2]中国科学院大学,北京100049

出　　处：《工程热物理学报》2021年第11期2804-2815,共12页Journal of Engineering Thermophysics

基　　金：国家自然科学基金资助项目(No.51876202,No.51836008)。

摘　　要：本文首次提出使用机器学习XGBoost算法作为气动性能的回归模型,并使用全局寻优算法MaxLIPO置信域方法在回归模型上进行寻优的气动优化方法。为保证回归模型在最优值附近的准确性,采用动态识别加点法构造动态更新的回归模型,并使用双重收敛准则判断优化流程的收敛。构建优化流程后,对一种新型减重高压涡轮叶型进行优化。结果表明,该优化流程相较传统优化方法能实现气动性能的快速有效寻优,在最优值附近的回归预测精度达到与CFD结果误差极小的水平,并能分析各几何参数对气动性能的影响权重,最终实现了对减重涡轮叶型的有效气动优化设计。This research first proposed a new aerodynamic optimization method that using machine learning XGBoost algorithm as the regression model,and global optimization algorithm MaxLIPO Trust Region method as the optimization method on regression model.In order to ensure the accuracy of the regression model near the optimal value,a dynamic recognition add point method is used to construct a dynamically updated regression model,and a double convergence criterion is used to determine the convergence of the optimization process.After constructing the optimization process,a new type of weight reduction high-pressure turbine blade was optimized.The results show that the optimization process can achieve rapid and effective optimization of aerodynamic performance compared to traditional optimization methods,and the regression prediction accuracy near the optimal value reaches a level with a minimum error from the CFD result.Meanwhile,the influence weight factor of geometrical parameters on the aerodynamic performance could be analyzed.Finally,the effective aerodynamic optimization design of the weight reduction turbine blade is achieved.

关 键 词：减重涡轮叶片 优化算法 XGBoost 机器学习 B样条 

分 类 号：TK14[动力工程及工程热物理—热能工程]