摘要
提出了一种基于机器学习的高维混合模型用于离心泵的叶片优化方法.选用一台低比转速离心泵,以离心泵叶轮叶片为研究对象,通过对叶片型线拟合分离多变量参数,利用支持向量机的高维表示方法,结合计算流体动力学软件,经过对训练集的机器学习,构建了离心泵叶片型线优化的代理模型.依据遗传算法求解离心泵多变量代理模型,预测了离心泵效率最高点及在该点时的叶片型线几何参数.运用数值模拟和试验研究的方法验证了预测数据,结果表明:数值模拟性能曲线与试验结果大体相符;在设计工况点,经代理模型优化后的数值模拟效率值较原型泵提高了2.61%,扬程提升了0.82 m,试验效率值较原型泵提高了2.1%,扬程提升了0.75 m.
A high-dimensional hybrid model based on machine learning was proposed to optimize the centrifugal pump blade.A low specific speed centrifugal pump was selected,and the centrifugal pump impeller blade was taken as the research object.By fitting the blade profile,the multi-variable parameters were separated.The surrogate modelling of centrifugal pump blade profile optimization was constructed by using the support vector machine(SVM),high-dimensional model representation(HDMR)and computational fluid dynamics(CFD)software through machine learning of training set.The multi-variable surrogate model was solved by genetic algorithm(GA),and the highest efficiency point of the centrifugal pump and the geometric parameters of the blade profile were predicted.The prediction data was verified by numerical simulation and experimental study.Results show that the numerical simulation performance curve is in good agreement with the experimental results.At the design operating point,the numerical simulation efficiency value optimized by the surrogate model is 2.61%higher than that of the prototype pump,and the head is 0.82 m higher;the test efficiency value is 2.1%higher than that of the prototype pump,and the head is 0.75 m higher,proving the validity of the high-dimensional hybrid surrogate model.
作者
姜丙孝
杨军虎
白小榜
王晓晖
JIANG Bingxiao;YANG Junhui;BAI Xiaobang;WANG Xiaohui(School of Energy and Power Engineering,Lanzhou University of Technology,Lanzhou 730050,China;Key Laboratory of Fluid Machinery and Systems,Lanzhou University of Technology,Lanzhou 730050,China)
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2020年第7期128-132,共5页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(51569013)
甘肃省科技基金计划资助项目(17JR5RA110)。
关键词
支持向量机
高维表示方法
代理模型
遗传算法
离心泵
叶片优化
support vector machine
high-dimensional model representation
surrogate model
genetic algorithm
centrifugal pump
blade optimization
