摘要
针对软传感器建模过程中辅助变量通常是多因素的混杂信号,在原始特征空间很难进行原始特征约简的问题,提出一种结合独立成分分析(ICA)和虚假最近邻点法(FNN)的原始特征选择法。利用独立成分分析法(ICA)将原始特征空间的混杂信号映射到新的独立特征子空间;然后再利用FNN计算每个原始特征剔除前后在独立特征子空间里的相似性测度,进而判断它对主导变量的影响能力,由此选择出原始特征。仿真结果表明,该方法具有优秀的原始特征选择能力。因此,该研究为选择出软传感器模型的原始特征提供了新方法。
Aiming at the problems that the secondary variables in soft-sensor modeling process usually are mixed sig- nals with multi-factors, and it is difficult to achieve the original feature reduction in the original feature space, a new original feature selection method combining independent component analysis (ICA) and false nearest neighbors (FNN) is presented. By using the independent component analysis, the mixed signal in the original feature space could be mapped into a new independent feature subspace;then using FNN, the similarity measure of each original feature in the independent feature subspace is calculated when the original feature is either retained or eliminated, and the influence capability of the original feature on the dominant variables is determined, thus the original features could be selected. Simulation results show that the proposed method has good original feature selection capability. Therefore, the research provides a new method for the original feature selection of the soft-sensor model.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第4期736-742,共7页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金项目(61174015
51075418)
重庆市自然科学基金项目(CSTC2012JJA9011)
重庆市教委科学技术研究(KF121410)
重庆科技学院校内科研基金项目(CK2011B04
CK2011Z01)资助
关键词
软传感器
特征子空间
独立成分分析
虚假最近邻点法
特征选择
soft-sensor
feature subspace
independent component analysis(ICA)
false nearest neighbor(FNN)
fea-ture selection