基于特征提取和极限学习机的软测量方法  

Soft Sensor Approach based on Feature Extraction and Extreme Learning Machines

作　　者：严东[1] 汤健[1,2] 赵立杰[2] 

机构地区：[1]中国人民解放军92941部队 [2]东北大学自动化研究中心

出　　处：《控制工程》2013年第1期55-58,共4页Control Engineering of China

基　　金：中国博士后自然科学基金(20100471464);国家自然科学基金重点项目(60534010)

摘　　要：针对建模数据存在的高维、共线性等特征,以及常用的基于人工智能的建模方法存在的模型结构难以确定、学习速度慢等缺点,提出了由基于主元分析(PCA)的特征提取和基于优化极限学习机(OELM)的建模算法两部分组成的软测量方法。采用PCA消除输入变量间的共线性并降低输入变量维数,以提取的线性无关的独立变量作为软测量模型的输入,从而简化模型结构。采用集成极限学习机(ELM)与支持向量机(SVM)算法优点的OELM方法作为建模算法,避免了ELM模型的随机性和SVM模型求解的复杂性。将特征提取方法与OELM方法结合后,提高了软测量模型的训练速度和预测性能。采用所述方法,对混凝土抗压强度的软测量问题进行了实验研究,验证了所提方法的有效性。该方法同时可以应用于基于雷达、光电等高维数据的目标识别,具有广阔的应用前景。The modelling data have the characteristics of high dimension, colinearity etc, and the existing artificial intelligence-based soft sensor approaches have deficiencies such as difficulty in determine model＇ s structure, low learning speed etc, Aim at these prob- lems, a novel modelling approach is proposed, which consists of principal component analysis （PCA） based on feature extraction and optimize extreme learning machine （OELM） based modelling algorithm. The PCA is used to eliminate collinearity among the input vari- ables and reduce the dimension. The extracted independent invariables are represented as the input of the soft sensor models, which simplifies the structure of the models. The OELM algorithm is a new coming modelling approach based on article intelligence, which in- tegrates the advantage of extreme learning machine （ELM） and support vector machine （SVM）. It has overcome the randomize disad- vantage of the ELM and the complexity solution process of the SVM. With the integration of the PCA and 0ELM, the training time of modelling process is reduced, and the prediction performance is improved. Base on the proposed method, the soft sensor problem of the concrete compression strength is constructed, which verifies the validity of the proposed method. This approach can also be used to the reorganization problem based on the high dimension Radar and photoelectric data, which has a bright future.

关 键 词：优化极限学习机 主元分析 特征提取 软测量 

分 类 号：TP27[自动化与计算机技术—检测技术与自动化装置]