基于LS-SVM的特征提取及在凝点软测量中的应用  

Feature Extraction Method Based on LS-SVM and Its Application to Soft Sensor for Freezing Point

作　　者：吴德会[1] 

机构地区：[1]九江学院电子工程系

出　　处：《系统仿真学报》2008年第4期917-920,925,共5页Journal of System Simulation

基　　金：国家自然科学基金(70272032);江西省教育厅科技项目(2007328)资助

摘　　要：提出了一种基于最小二乘支持向量机(LS-SVM)回归算法的特征提取新方法,并将其成功应用于柴油凝点近红外(NIR)光谱软测量建模。在该方法中,将特征提取公式表达成与LS-SVM回归算法相同的形式,这样就能通过LS-SVM求取最优的特征投影向量。用一个含120个样本的401维柴油近红外光谱数据集进行测试,通过该方法提取后,原始光谱数据集的特征被降到了6维并保留了原有99.58%的信息。同时,用该数据建立的软测量模型具有更快的学习速度和更高的测量精度。实验结果验证了所提的特征提取新方法应用于近红外光谱特征提取的可行性和有效性。A novel feature extraction method was proposed based on least squares support vector machine （LS-SVM） regression algorithm and it was applied to soft sensor modeling for freezing point of diesel fuel with near-infrared （NIR） spectrometry successfully. In this method, the formulation of feature extraction was made in the same fashion as that in the LS-SVM linear regression algorithm. So the optimal projection vectors could be obtained by LS-SVM. 401 dimensional NIR spectrum data sets, including 120 samples, were used to test. Through presented method, the data sets were reduced to 6 dimensions and contain 99.58% messages. The soft sensor model for freezing point with the extracted characters could obtain faster learning speed and higher accuracy. The results indicate that the presented method is not only feasible but effective in NIR spectrometry feature extraction.

关 键 词：特征提取 主元分析 最小二乘支持向量机 软测量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]