Research On The Method And Algorithm Of Detecting The Refined Oil's Properties Based On Near Infrared Spectroscopy

The wide application of near infrared spectral analysis technology has promoted the rapid development of the detection technology of refined oil's properties.It is the primary goal of near infrared spectral detection technology to establish a detection model of high prediction accuracy,high reliability and good stability.The method of detecting refined oil's properties based on near infrared spectroscopy is designed in this thesis,and the further research of prediction accuracy and reliability of the model is carried out.The research background is first presented in this thesis and the current research status of the detection of refined oil's properties are reviewed.In chapter 2,the basic principle of partial least squares method is introduced,and the process of the detection of refined oil's properties is shown based on partial least squares method.It mainly includes seven parts:collecting near infrared spectrum data,selecting feature wavelength,spectral data preprocessing,selecting similar samples,establishing partial least squares model,predicting and analyzing the results.Finally,the problems in the detection process are discussed.The third chapter designs a method for eliminating abnormal samples in the calibration set based on principal component analysis and correlation analysis between properties.The influence of abnormal samples on the prediction accuracy of the model is analyzed,and the basic principle of principal component analysis is introduced.Then,the detailed steps of the method for eliminating abnormal samples in the calibration set is presented.At last,the detection of 93#gasoline's octane number in one refinery enterprise is taken as a case,and the causes of abnormal samples are analyzed in detail.In chapter 4,the factors affecting the prediction accuracy of the model detecting refined oil's properties are analyzed,including temperature and noise.Then,the method based on the spectral temperature correction which improves the prediction accuracy is proposed,and the construction of temperature transfer function based on piecewise direct standardization algorithm is introduced.As a case,the detection of 95#gasoline's octane number in one refinery enterprise is analyzed concretely.In addition,the method based on discrete wavelet transform and fast Fourier transform algorithm which improves the prediction accuracy is designed,and the basic principles of discrete wavelet transform and fast Fourier transform are introduced,as well as the detailed steps of the detection method based on discrete wavelet transform and fast Fourier transform algorithm are given.A detailed analysis is given about 95#gasoline research octane number in one refinery enterprise at last.The reliability evaluation method of the model detecting refined oil's properties based on The concentration of sample distribution and the predictive capability of model is designed in chapter 5.The evaluation index commonly used of the detection model is given,and the detailed steps of the reliability evaluation method are presented.A case concerning 95#gasoline research octane number is analyzed in the end.In this thesis,the detection method of refined oil's properties based on near infrared spectroscopy can meet the demand of high prediction accuracy and high reliability,and improve the efficiency of the refinery enterprises.