Method for detecting multiclass properties of oil product based on mode recognition and spectrogram mapping

Abstract

The invention relates to a method for detecting multiclass properties of an oil product based on mode recognition and spectrogram mapping. The method comprises the following steps: (S1) collecting a near-infrared spectrum of a sample of an unknown oil product, and mapping and projecting the near-infrared spectrum to a characteristic plane determined by virtue of a training set sample according to a principal component analytical method; (S2) classifying sample points of the unknown oil product into a certain sample class by virtue of a naive Bayesian classifier according to a region position, on which the sample of the unknown oil product is projected, on the characteristic plane of the training set sample; (S3) selecting and searching a radius range in a sample point class in which the sample points of the unknown oil product are located, and searching and finding adjacent points of the sample points of the unknown oil product; (S4) calculating mahalanobis distances from the adjacent points of the sample points to the sample points of the unknown oil product, and carrying out normalization on the mahalanobis distances to form a weight, calculating the properties of the sample points of the unknown oil product by carrying out weighting and summation on the properties of the adjacent points of the sample points of the unknown oil product, so as to obtain the multiclass properties of the sample points of the unknown oil product.

Description

technical field [0001] The invention belongs to the field of petroleum and petrochemicals, and relates to a method for detecting multi-category properties of oil products based on pattern recognition and spectrogram mapping. Background technique [0002] The near-infrared spectra of gasoline, diesel oil, and crude oil include a wealth of information on the composition of oil products, and the properties of oil products such as octane number, vapor pressure, cetane number, freezing point, cold filter point, flash point, distillation range, etc. It is the result of the comprehensive action of its family composition, and the spectrum corresponding to each oil product is uniquely determined. Therefore, the traditional method generally characterizes the near-infrared spectrum and directly correlates it with various properties of the oil, using multiple linear regression (MLR), stepwise regression (SMR), principal component analysis (PCA), principal component regression (PCR), par...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the existing technology, the single model of each property of the oil product has a narrow scope of application, large errors, and a large workload for establishing and maintaining the model. It is difficult to meet the long-term requirements of the industrial site and cannot meet the requirements of stable operation of many advanced petroleum and petrochemical companies. The bottleneck problem of control implementation, for this reason the purpose of the present invention is to provide a kind of detection method of oil product property accurate, fast detection, based on pattern recognition and spectrogram mapping oil product multi-category property

Images