第38卷3 期第.pdfVIP

第 38 卷 　第 3 期 西　安　交　通　大　学　学　报 Vol. 38 　№3 2004 年 3 月 　 Mar. 2004 J OU RNAL OF XI′AN J IAO TON G UN IV ERSIT Y 反问题与神经网络相结合的混流泵叶片优化设计 卢金铃 , 席　光 , 祁大同 (西安交通大学能源与动力工程学院 , 710049 , 西安) 摘要 : 针对三元扭曲叶片优化设计过程中设计变量较多的问题 ,采用动量矩为设计变量 ,再通过反问题计算 得到叶片来间接对叶片进行参数化 ;针对评价函数计算量太大的问题 ,根据试验设计理论安排训练样本 ,采 用神经网络建立设计变量与目标函数间的复杂的响应关系 ,并且详细研究了反向传播和径向基函数 2 种网 络在对评价函数进行预测过程中的应用 ,建立了一种新的叶片优化设计方法. 与传统的优化方法相比 ,其设 计变量数目较少 , 以叶轮内的三维粘性流动分析为基础且大大缩短了计算时间. 利用此方法对一台混流泵的 扬程和效率进行优化 ,所得叶片性能良好 ,从而验证了此方法的有效性. 关键词 : 反问题 ;神经网络 ;优化 ( ) 中图分类号: O357 　文献标识码 : A 　文章编号 : 0253 - 987X 2004 03 - 0308 - 05 Blade Optimization of MixedFlow Pump Using Inverse Design Method and Neural Net work L u Jinlin g , Xi Guang , Qi Daton g ( School of Energy and Power Engineering , Xi ′an Jiaotong University , Xi ′an 710049 , China) Abstract : The ordinary parameterization pro grams of threedimensional blade contain a great deal of design vari ables , therefore an indirect parameterization method was proposed , where the angular momentum was treated as design variables , the blade was calculated by inverse design method , and neural networks were adopted to con struct the response relation between the design variable and the objective function. The sample data used to train neural networks were schemed according to design of experiment theory , the application of two kinds of neural networksback propagation network and radial basic function network were investigated in detail , and a new op timization method was proposed. Compared with