Font Size: a A A

Application Of Artificial Neural Network In NIR Analysis And Dark-colored Oil Analysis

Posted on:2001-05-29Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y B WangFull Text:PDF
GTID:1101360062975603Subject:Applied Chemistry
Abstract/Summary:PDF Full Text Request
The Near Infrared Spectroscopy (NIR) analysis method, as a new highly efficient analytical technique, was used in many fields of petrochemical industxy, with its advantage of better accuracy and repeatability, shorter time of testing, simpler operation, no need of solvent and easy actualizing on-line analysis. But few literatures had been published for the application of NJR technique in determination of the properties of dark colored petroleum and petroleum products. In this paper, analyses of dark-colored petroleum and petroleum products by NIR method were studied, to meet the need of quick analyses of dark colored oil and broaden the application area of NIR analysis method. At the same time, the artificial neural network was applied in MR analysis for calibration.The BP neural network was used to construct the mathematical models, which correlated the NTR spectral features of samples with their properties. Gasoline and disel oil samples of wide variation, i.e., gasoline including catalytic cracking gasoline, reforming gasoline and some commercial gasoline and diesel oil including first-cut, second-cut and third-cut diesel oil, were studied for their RON and flash point, respectively. Compared with other multivariate calibration methods such as locally weighted regression (LWR), partial least squares (PLS) and principal component regression (PCR), the artificial neural network could produce more accurate and rugged calibration. To select the training set, the KennardStone design method was applied. The results demonstrated that the Kennard-Stone~ design method could get a training set with representative and uniform distribution, so the calibration result and training speed were improved. Principle components from both PLS and PCA were used to reduce noise and the number of input of artificial neural network. It was shown that the artificial neural network with the principle components from PLS could get better calibration results with less input numbers. In addition, the artificial neural network was applied to establish the correlation models,which correlate the NIR spectral features of crude oil and lube base oils with their boiling range distribution and viscosity, respectively. Compared with PLS, the artificial neural network gave more accurate results.MR analysis method for measuring the boiling range distribution of crude oil was studied, based on the results from gas chromatography simulation distillation method. A special method of sample handling was developed in order to obtain spectrum of crude oil simply and conveniently. A spectrum pretreatment for dark colored oil was used, which included second-differentiation and Max-Mm normalization. Through the pretreatment, repeatible spectra were obtained. Twelve calibration models (including the initial boiling point, 5%, 10%, 20%, 30%, 40%, 50%, 60% off temperature, and the yield of gasoline fraction, diesel fraction, lube ~oil fraction and residue fraction) were established by artificial neural network. These models were used to predict the boiling range distribution of validation sample set. The difference between the property tested by MR method and that by the gas chromatography method was no greater than the reproducibility limit of ASTM D5307. Paired t test showed that the result of NIR method and that of gas chromatography method were coincident. Determining the boiling range distribution of crude oil by NIR technique was proved to be practical and fast.NJR analytical methods for determining the chemical composition and viscosity index of lube base oils were developed on samples of various kinds. Partial least squares (PLS) technique was applied to construct the mathematical models, which correlate the MR spectral features with the chemical composition and viscosity index analyzed by reference method or standard method. There was no evident difference between the result of validation set tested by MR method and that by other reference methods. Determining viscosity of the lube base oils by MR method was also discussed. Because of...
Keywords/Search Tags:near-infrared spectrosc opy, chemometric s, artificial neural network~ crude oil, boiling range distribution, lube base oil, chemical composition: viscosity index: viscosity, bitumen, wax content, carbon residue
PDF Full Text Request
Related items