论文部分内容阅读
柴油是重要的石油化工产品,凝点是确定其品质的主要理化指标,但传统测定方法周期长、费用高,且易受人为的影响。本文提出一种用近红外光谱(NIRS)分析技术的柴油凝点软测量方法。先用多项式卷积为原始的柴油MR光谱数据作光谱平滑、基线校正和标准归一化;然后,利用主元分析(PCA)组合近红外光谱数据集的高维特征并向低维空间投影,降低输入维数,提高各维特征的敏感性;最后,用最小二乘支持向量机(LS-SVM)回归算法建立凝点的软测量模型。用一个含120个样本的401维柴油近红外全光谱数据集建模和检验,通过PCA后,全光谱数据集的特征降到了6维,并保留了99.6%的信息。进一步的实验表明,采用PCA提取特征做软测量建模的性能,要普遍优于直接作用在光谱波长域的方法。与BP、PCA+BP及PCA +SVM方法相比,所提方法建立的柴油凝点软测量模型测量精度更高,且与标准方法测量的结果更为接近,因此,又为柴油凝点的在线测定提供一种新方法。
Diesel oil is an important petrochemical product. The freezing point is the main physical and chemical index to determine its quality. However, the traditional measurement method has a long period, high cost and is easily influenced by human. In this paper, a method of near infrared spectroscopy (NIRS) analysis is proposed for measuring the freezing point of diesel. First, polynomial convolution was used to original diesel MR spectroscopy data for spectral smoothing, baseline correction and standard normalization. Then PCA was used to combine the high-dimensional features of near-infrared spectral dataset and projected into low-dimensional space, Reduce the input dimension and improve the sensitivity of each feature. Finally, LS-SVM regression algorithm is used to establish the soft sensor model of freezing point. After modeling and verifying with a 401-dimensional diesel near infrared full-spectral dataset containing 120 samples, the characteristics of the full-spectrum dataset have been reduced to 6-dimensional with PCA and 99.6% of the information has been retained. Further experiments show that the performance of soft-sensing modeling using PCA extraction features is generally better than the direct effect in the spectral wavelength domain method. Compared with the BP, PCA + BP and PCA + SVM methods, the proposed method has the advantages of higher accuracy and accuracy compared with the standard method. Therefore, Determination provides a new method.