基于神经网络的换热器翅片参数灵敏度分析  

Parameter sensitivity analysis of fin based on neural network in heat exchanger

作　　者：肖宝兰[1,2] 俞小莉[1] 韩松[1] 陆国栋 夏立峰 

机构地区：[1]浙江大学动力机械及车辆工程研究所,浙江杭州310027 [2]浙江大学城市学院机械工程系,浙江杭州310015 [3]浙江银轮机械股份有限公司,浙江天台317200

出　　处：《浙江大学学报（工学版）》2011年第1期122-125,145,共5页Journal of Zhejiang University：Engineering Science

摘　　要：为了探索将人工神经网络技术应用于翅片参数对换热器(HX)性能影响研究的可行性,建立2个结构不同的3层反向传播(BP)神经网络进行训练及优化.分别对流动阻力特性和传热特性进行性能预测,根据预测结果进行翅片参数的灵敏度分析.训练和测试样本数据来源于大量的风洞实验和数值仿真结果.经过优化后的预测传热和流动阻力的网络隐层神经元个数分别为2和6,隐层和输出层的传递函数分别为tansig和purelin函数,采用基于Levenberg-Marquardt(L-M)算法的训练函数.网络性能测试结果表明,人工神经网络以优越的非线性映射能力,能够很好地预测翅片参数变化对换热器性能的影响.翅片参数灵敏度分析结果与实践工程经验比较吻合.Two different 3-layer back propagation（BP） neural network models were established in order to explore the feasibility of applying neural network technology to analyze the parametric effect on performance in heat exchanger（HX）.The models were trained and optimized to predict the heat transfer performance and flow resistance of a plate-fin HX.The parameter sensitivity analysis of fin was conducted according to the prediction results.The data of train and test samples were from many wind tunnel tests and simulation results.The models had two and six hidden layer neurons irrespectively after optimized.The transfer functions of hidden and output layers were tansig and purelin function irrespectively,and the train function was based on Levenberg-Marquardt（L-M） method.Results show that neural network can predict the effect of fin parameter on HX performance.The results of fin parameter sensitivity analysis accorded well with the engineering experience.

关 键 词：神经网络 翅片参数 热力性能 灵敏度分析 

分 类 号：TK402[动力工程及工程热物理—动力机械及工程]