基于高维机器学习方法的离心泵扭曲叶片优化

提出将高维表示方法与机器学习中的支持向量机方法结合对离心泵扭曲叶片优化。选用1台中比转速离心泵为研究对象，通过对扭曲叶片3条叶片型线的参数化，分离控制变量并确定代理模型的训练空间。经过对离心泵扭曲叶片水力模型的算法学习，得到以叶片型线参数为自变量，效率为目标函数的离心泵代理模型，运用遗传算法求解并反馈叶片型线参数。采用数值模拟及试验的方法验证了代理模型的预测结果，并从动能方程的角度分析了优化前后扭曲叶片叶轮内流场的变化。结果表明:在设计工况点，优化后的扭曲叶片离心泵数值模拟效率值比原型泵提高了1.72%，扬程提升了0.41 m，试验效率值比原型泵提高了1.5%，扬程提升了0.35 m。 

Optimization of twisted blade of centrifugal pump based on high dimensional machine learning method

The combination of high dimensional model representation （HDMR） and support vector machine （SVM） in machine learning was proposed to optimize the twisted blade of centrifugal pump．A centrifugal pump with medium specific speed was selected as the research object．The three blade profiles of twisted blade were parameterized，the control variables were separated and the training space of surrogate model was determined.After algorithm learning of the hydraulic models of the twisted blade of the centrifugal pump，the surrogate model of the centrifugal pump with the blade profile parameters as the independent variables and the efficiency as the objective function was obtained．The prediction results of the surrogate model were verified by numerical simulation and experiment.The change of flow field in twisted blade impeller before and after optimization was analyzed from the view of kinetic energy equation．The results showed that at the design operating point，the numerical simulation efficiency of the optimized twisted blade centrifugal pump was 1.72% higher than that of the prototype pump，and the head was 0.41 m higher；the test efficiency was 1.5% higher than that of the prototype pump，and the head was 0.35 m higher than that of the prototype pump.