| Some properties of petroleum products in refinery such as gasoline octane number and diesel cetane number are necessary to be online monitored. Laboratory analysis methods are usually not suitable for online monitoring because of their high cost and long time delay. Near-infrared spectroscopy (NIRS) is a non-destructive realtime analytical method, and it is preferable to online monitoring for the properties of petroleum products. Therefore, this thesis researched the application techniques of NIRS and developed an online NIR analyzer for properties of petroleum. The main contributions of this thesis are as follows:1. Introduce the concepts and the principles of NIR quantitative analysis, and then review the applications of NIRS in petroleum products analysis.2. For a set of gasoline samples, multivariate linear regression (MLR) and partial least squares (PLS) calibration models are built to predict research octane number (RON). Experimental results show that the PLS model gains higher prediction accuracy with the mean prediction error ≤ ±0.2 octane number.3. To overcome the disadvantages of the linear calibration methods such as MLR and PLS, least-squares support vector machine (LS-SVM) is introduced to NIR quantitative calibration in this thesis. For a set of diesel cetane number (CN) samples, the LS-SVM model obtains the best performance in the CN prediction, compared with the MLR and PLS model. To meet the need of online applications, a novel online modeling method has been proposed. Experimental results show that the online model exhibits a good self-learning ability and keeps high prediction accuracy even if great change of products quality occurs.4. Based on the above research results, an online NIR analyzing system has been developed to predict gasoline RON. This system has been installed in a catalytic reforming unit in a refinery, and industrial applications show its good performance in the prediction of RON.
|
PDF Full Text Request