基于离散粒子群的气液二相流型特征选择  

Selection method for gas-liquid two-phase flow regime feature based on discrete binary version of particle swarm optimization

作　　者：孙斌[1] 王虹[1] 

机构地区：[1]东北电力大学能源与动力工程学院,吉林吉林132012

出　　处：《化学工程》2011年第5期67-71,共5页Chemical Engineering(China)

基　　金：国家自然科学基金资助项目(50706006)

摘　　要：针对气液二相流型识别中存在的大量无关或冗余的特征会降低分类器性能的缺陷,提出了离散粒子群算法(BPSO)和最小二乘支持向量机(LS-SVM)封装模式的流型特征选择方法。该方法分别采用小波包和经验模式分解方法(EMD)对原始压差波动信号进行分解,分别提取原始信号和各分解信号的时域量纲一指标组成融合特征。然后采用BPSO进行寻优搜索,以类内类间方差和正确分类率作为BPSO的适应度函数,并应用LS-SVM对选出特征子集的识别能力进行评估。实验结果表明:以正确分类率为适应度函数的BPSO特征选择方法更能有效滤除无关特征,提高流型识别的正确率,从而证明了其有效性。Considering that many irrelevant or redundant features in gas-liquid two-phase flow regime identification seriously reduce the classifier performance,a flow regime features selection method based on discrete binary version of particle swarm optimization and least squares support vector machines（BPSO-LS-SVM） was proposed.The original signals were decomposed via wavelet packet and empirical mode decomposition（EMD） respectively;the dimension-one indicators in time domain were extracted from the original signals and each decomposed signal to construct the combined features.BPSO was used to search the feature subsets in combined features with in-inter classes variance and accuracy to be fitness function of the BPSO.The least square support vector machine（LS-SVM） was used as an evaluator to estimate the performance of the selected feature subsets.The experimental results show that the accuracy fitness function not only prunes the irrelevant features but also improves the accuracy of classifiers,which indicates that the proposed method is effective.

关 键 词：气液二相流 流型识别 特征选择 离散粒子群算法 最小二乘支持向量机 

分 类 号：O359.1[理学—流体力学]