基于RW-GN的电容层析成像流型辩识算法  

Flow pattern identification algorithm for electrical capacitance tomography system based on RW-GN

作　　者：黄仲洋[1] 陈宇[1] 许莉薇 

机构地区：[1]东北林业大学信息与计算机工程学院,黑龙江哈尔滨150040

出　　处：《计算机工程与设计》2015年第4期1042-1045,1109,共5页Computer Engineering and Design

基　　金：国家948基金项目(2011-4-04);中央高校基本科研业务费专项基金项目(DL12CB02);黑龙江省教育厅科学技术研究基金项目(12513016);黑龙江省博士后基金项目(LBH-Z10273);黑龙江省自然科学基金项目(F201347);哈尔滨市科技创新人才专项基金项目(2013RFQXJ100)

摘　　要：为解决电容层析成像反问题流型识别困难的问题,提出一种基于修正加权高斯牛顿的神经网络(RW-GN)对电容层析成像流型辩识的算法。在研究高斯牛顿算法原理的基础上,对原算法进行优化,对优化之后算法的稳定性进行验证;在此基础上探讨将基于修正加权高斯牛顿的神经网络算法应用于电容层析成像流型辨识的可行性。实验结果表明,与BP神经网络、SVM、决策树识别算法相比,该算法对5种流型有较高的识别率。To solve the problem that flow pattern identification is more difficult in electrical capacitance tomography（ECT）system,based on revised and weighted Gauss Newton neural network（RW-GN）,flow pattern identification algorithm for electrical capacitance tomography system was presented.Based on the principle of Gauss Newton algorithm,the original algorithm was optimized,and the stability of optimization algorithm was proved.On this basis,the feasibility of applying the presented algorithm to flow pattern identification of electrical capacitance tomography was discussed.Experimental results show that the algorithm can obtain a higher identification rate of five flow patterns compared with BP neural network,SVM and decision tree recognition algorithm.

关 键 词：修正加权高斯牛顿 电容层析成像 稳定性 可行性 识别率 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]