基于主元分析与RBFNN的真空玻璃传热过程预测  

RBF neural networks modeling of heat conduction process of vacuum glazing based on principal component analysis

作　　者：张亮[1] 王磊 王元麒[2] 

机构地区：[1]海南大学信息科学技术学院南海海洋资源利用国家重点实验室,海口570228 [2]米兰理工大学,意大利米兰20133

出　　处：《真空》2017年第3期66-70,共5页Vacuum

基　　金：国家自然科学基金(61463011);国家重点研发计划课题(2016YFC0700804)

摘　　要：提出一种基于主元分析和RBF神经网络相结合的真空玻璃传热过程预测模型,研究真空玻璃的保温性能。采用主元分析对原始多维输入变量进行预处理,选择输入变量的主成分作为RBF神经网络的输入,既可以减少输入变量的维数,也可以消除各变量之间的相关性,同时又包含原变量的大部分信息。基于主元分析,可对真空玻璃非热源一侧温度进行智能软侧量,提高网络的收敛性和稳定性,仿真结果表明证明此模型的正确性和理论分析的合理性,该方法对具备不同传热系数(U值)的真空玻璃具有良好的自适应性,对接下来研究真空玻璃保温性能智能软测量提供一定的理论基础。Based on principal component analysis and RBF neural network, a prediction model for insulation performance of vacuum glazing was put forward. Principal component analysis （PCA） was used to pre-treat the original multidimensional input variables. The principal components of input variables were selected as the input of RBF neural network, containing most of the information of the original variables. It can reduce the dimension of input variables and eliminate the correlation among variables. Based on the principal component analysis, the off heat source temperature of the vacuum glass can be measured at an intelligent and soft way so as to improve the convergence and stability of the network. The simulation results show that the model is correct and the theoretical analysis of the vacuum glass with different heat transfer coefficient （U value） shows good adaptability. Our work provides certain theoretical basis for the vacuum glass thermal insulation performance with intelligent soft measurement.

关 键 词：主元分析 RBF神经网络 保温性能 智能软测量 

分 类 号：TB771[一般工业技术—真空技术]