Reserch And Development Of Oil Modeling And Model Maintenance Technology Based On Near Infrared Spectroscopy

Rapid assay of oil properties plays significant role in production of refineries,helping to stabilize production,optimize resource utilization and improve the production efficiency.Current methods of rapid detection still need to be improved in the prediction accuracy and robustness.A dynamic modeling method based on near infrared(NIR)spectroscopy is proposed in this paper,and the optimization and regular maintenance of calibration set are studied to solve the problems above.The research background and the current research status are first presented in this paper,followed by summarizing the problems of NIR rapid analysis of oil properties.In chapter two,a dynamic model of oil properties based on partial least squares is established.After preprocessing the spectral data,principal component analysis is employed to search similar samples,and the calibration set is established to construct the property prediction model.Two industry cases concerning 93# gasoline research 50% evaporation temperature and saturated vapor pressure are analyzed in the end.Two methods are designed in the third chapter to distinguish outliers of calibration set.The first method based on the similarity analysis of multiple samples is employed to find samples with abnormal property values.The second method applies model prediction bias to judge the spectral or property abnormalities.Finally,both methods are applied to predict gasoline research octane number in the Szorb catalytic unit.In chapter four,a novel method for automatic maintenance is proposed based on similar samples to eliminate redundant samples and outliers.New samples are regularly screened and added to the calibration set to enrich the representativeness of calibration set samples,and the detailed procedure of automatic maintenance is elaborated.The industry case concerning the detection of 95# gasoline endpoint is implemented to illustrate the specific maintenance process.In the fifth chapter,the software based on JAVA EE is developed to realize the function of oil fast analysis.Data is stored in the MySQL database,while complex algorithms are implemented in MATLAB,and visualization is realized.Finally,the oil sample management,model configuration and model evaluation system are described in detail,including the functions and the main interfaces.The NIR modeling and model maintenance technology proposed in this paper can dynamically select calibration samples,and maintenance of calibration set is developed in face of the change of working conditions.The experimental results indicate that this method can effectively improve the model prediction accuracy and robustness,which is of great significance in controlling production in refinery plants.