Research on soft measuring for freezing point of diesel fuel based on NIR spec-trometry

Wu Dehui,Wang Xiaohong

DOI: https://doi.org/10.3969/j.issn.1001-4160.2007.10.004

2007-01-01

Abstract:Freezing point was an important physical-chemical property of diesel fuel. The traditional method for measuring freezing point was time consuming and expensive, and the results were largely dependent on the operator. A soft sensor modeling method for freezing point of diesel fuel with near-infrared spectrometry (NIRS) was presented. Firstly, the polynomial convolution algorithm was used in spectrogram smoothness, baseline correction and standardization. Secondly, using principal component analysis( PCA) to associate the NIRS data sets in higher dimensional space and project them to lower dimensional feature space, the method can reduce the number of input dimension and increase their sensitivity to output. Finally, The soft sensor model for freezing point was built using LS-SVM regression algorithm, 401 dimensional NIRS data sets, including 120 samples, were used to model and test. Through PCA, the data sets had been reduced to 6 dimensions and contain 99. 6% messages. Experiment results also show that capabilities of soft sensor modeling method using PCA to extract features are generally better than those used directly in spectrum wavelength domain. Compared with BP, PCA + BP and PCA + SVM methods, the proposed method can obtain higher precision, so it has considerable applied value in fast measuring of freezing point.