Method for establishing a prediction model for aviation kerosene properties using binary linear classifier

Abstract

The present invention provides a method for constructing a prediction model for the properties of aviation kerosene using a binary linear classifier, and belongs to the technical field of prediction of the properties of aviation kerosene. The method of the present invention comprises: obtaining the physical and chemical property data of a plurality of aviation kerosene samples; collecting near-infrared spectrum data of a plurality of aviation kerosene samples, and optimizing the near-infrared spectrum data at least twice; constructing a regression model associating the physical and chemical property data of aviation kerosene with the near-infrared spectrum data using a binary linear classifier; substituting the optimized data into the regression model to obtain a plurality of physical and chemical properties of the sample to be tested. The present invention constructs a regression model for predicting aviation kerosene based on a binary linear classifier, and a single operation can realize the detection of a variety of physical and chemical properties of aviation kerosene. The processing of the original spectrum, the establishment and update of the model do not need to be performed offline, and the regression model can be updated timely and effectively to improve the prediction accuracy.

Description

Technical Field

[0001] The invention belongs to the technical field of aviation kerosene property detection, and in particular relates to a method for constructing an aviation kerosene property prediction model by utilizing a binary linear classifier. Background Art

[0002] With the rapid development of the aviation industry, aviation kerosene, as the main energy source of the aviation industry, plays a vital role in the quality and safety of the entire industry chain. At present, the transportation methods of aviation kerosene in China are mainly pipeline transportation, railway tank car and road transportation. Because some fuel transportation media also transport other types of fuel, quality problems such as aviation kerosene pollution often occur.

[0003] Traditional methods for testing the properties of aviation kerosene are complex to operate, time-consuming, and limited by professional laboratory space. It is difficult to meet the requirements for real-time monitoring of aviation kerosene property data, and it is not easy to detect aviation kerosene quality problems in a timely manner. The currently developing chemometrics technology can analyze a large amount of data in a short period of time, making it possible to quickly evaluate the quality of aviation kerosene.

[0004] Among them, near-infrared spectroscopy is one of the most widely used rapid analysis technologies in the field of industrial production. This technology has the advantages of fast spectral information acquisition, high sensitivity, and low price. Combined with appropriate modeling methods, rapid detection of the physical and chemical properties of aviation kerosene can be achieved. However, traditional modeling methods require offline processing of raw spectra, model establishment and updating, but the industrial application site is complex and changeable, so the model needs to be updated in a timely and effective manner.

[0005] Therefore, based on the above technical problems, the present invention proposes a method for constructing an aviation kerosene property prediction model using a binary linear classifier. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art and provides a method for constructing an aviation kerosene property prediction model using a binary linear classifier.

[0007] The present invention provides a method for constructing an aviation kerosene property prediction model using a binary linear classifier, comprising the following steps:

[0008] Obtain physical and chemical property data of multiple aviation kerosene samples;

[0009] Collecting near infrared spectrum data of the plurality of aviation kerosene samples, and performing at least two optimization processes on the near infrared spectrum data;

[0010] Using a binary linear classifier to establish a regression model that associates the physicochemical property data of the aviation kerosene sample with the near infrared spectrum data;

[0011] Near infrared spectrum data of the sample to be tested is collected and optimized at least twice, and the optimized data is substituted into the regression model to obtain multiple physical and chemical properties of the sample to be tested.

[0012] Optionally, the physical and chemical properties include density at 20°C, freezing point, flash point, initial distillation point, 50% recovery temperature and final distillation point.

[0013] Optionally, a transmission near-infrared spectrometer is used to collect near-infrared spectrum data of the aviation kerosene sample; wherein,

[0014] The acquisition range is 4000cm -1 ~12000cm -1 , the scanning times range from 5 to 120 times.

[0015] Optionally, the near infrared spectrum data is optimized at least twice, including:

[0016] Performing a first optimization process on the near infrared spectrum data using first order differential and vector normalization method;

[0017] The PCA-MD method was used for the second optimization process to eliminate abnormal aviation kerosene samples.

[0018] Optionally, the vector normalization formula is as follows:

[0019] x′=(x-Min(x)) / (Max(x)-Min(x))

[0020] Where Max(x) is the maximum value of the sample data, Min(x) is the minimum value of the sample data, x is the sample data to be normalized, and x' is the normalized sample data.

[0021] Optionally, a second optimization process is performed using the PCA-MD method to remove abnormal aviation kerosene samples, including:

[0022] The PCA-MD method was used to calculate the Mahalanobis distance value and Mahalanobis distance standard deviation of the aviation kerosene sample after the first optimization treatment;

[0023] Obtaining a Mahalanobis distance threshold of the abnormal aviation kerosene sample according to the Mahalanobis distance value and the Mahalanobis distance standard deviation;

[0024] The Mahalanobis distance value of each aviation kerosene sample after the first optimization process is compared with the Mahalanobis distance threshold of the abnormal aviation kerosene sample, and the corresponding sample is eliminated in response to the Mahalanobis distance value being greater than the Mahalanobis distance threshold.

[0025] Optionally, the calculation formula of the Mahalanobis distance threshold is as follows:

[0026] W MD =mean(MD)+k*std(MD)

[0027] Where W MD is the Mahalanobis distance threshold of abnormal aviation kerosene samples, mean(MD) is the mean function of MD, std(MD) is the standard deviation function of MD, k is the threshold adjustment weight coefficient; and,

[0028] The calculation formula of the Mahalanobis distance value is:

[0029]

[0030] Where MD is the Mahalanobis distance value of aviation kerosene sample i, t i is the score vector of aviation kerosene sample i, C is the covariance matrix of the score matrix of all aviation kerosene samples, and T is the score matrix of all aviation kerosene samples.

[0031] Optionally, before using a binary linear classifier to establish a regression model associating the physicochemical property data of the aviation kerosene sample with the near infrared spectrum data, the method further includes:

[0032] The near-infrared spectral data are divided into n spectral intervals, local prediction models are established for the n spectral intervals respectively, and the root mean square error is used as an evaluation criterion to select a preset spectral interval.

[0033] Optionally, the calculation formula of the root mean square error is as follows:

[0034]

[0035] Where yi is the predicted value of aviation kerosene sample i, is the true value, and n is the number of samples.

[0036] Optionally, the use of a binary linear classifier to establish a regression model associating the aviation kerosene physicochemical property data with the near infrared spectrum data includes:

[0037] The near infrared spectrum data of the preset spectrum interval is mapped to a high-dimensional feature space and a correlation model is established with the physical and chemical property data, as follows:

[0038]

[0039]

[0040]

[0041] In the formula, is the mapping function that maps the near-infrared spectral input data to the high-dimensional feature space, γ is the regularization parameter, and e k is the error variable, the weight vector ω∈R n , bias value b∈R, output data y of the kth aviation kerosene sample k ∈R;

[0042] The Lagrange multiplier method is used to obtain the following formula:

[0043]

[0044] In the formula, the Lagrange multiplier α k ∈R, the input data x of the kth aviation kerosene sample k ∈R N ;

[0045]

[0046]

[0047]

[0048]

[0049] The binary linear classifier regression model is solved as follows:

[0050]

[0051] The present invention provides a method for constructing a prediction model for the properties of aviation kerosene using a binary linear classifier, comprising the following steps: obtaining the physicochemical property data of a plurality of aviation kerosene samples; collecting the near-infrared spectrum data of the plurality of aviation kerosene samples, and optimizing the near-infrared spectrum data at least twice; establishing a regression model associating the physicochemical property data of the aviation kerosene samples with the near-infrared spectrum data using a binary linear classifier; collecting the near-infrared spectrum data of the sample to be tested, and optimizing it at least twice, substituting the optimized data into the regression model to obtain a plurality of physicochemical properties of the sample to be tested. The present invention constructs a regression model for predicting aviation kerosene based on a binary linear classifier, and a single operation can detect a variety of physicochemical properties of aviation kerosene. The processing of the original spectrum, the establishment and update of the model do not need to be performed offline, and the regression model can be updated in a timely and effective manner to improve the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of a method for constructing an aviation kerosene property prediction model using a binary linear classifier according to an embodiment of the present invention;

[0053] Figure 2is the original near infrared spectrum of the aviation kerosene sample of the present invention;

[0054] Figure 3 This is the near infrared spectrum of the aviation kerosene sample after the optimization treatment of the present invention. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0056] like Figure 1 As shown, the present invention provides a method S110-S140 for constructing a prediction model for aviation kerosene properties using a binary linear classifier:

[0057] S110. Obtain physical and chemical property data of multiple aviation kerosene samples.

[0058] Specifically, this embodiment collects aviation kerosene samples that are representative of the region, crude oil origin, and refining process across the country, with a total sample size of no less than 300. The physical and chemical properties of each sample are measured using the method required by the national standard GB6537-2018 for No. 3 jet fuel, and a physical and chemical properties database is established.

[0059] Furthermore, the physical and chemical properties of this embodiment include density at 20° C., freezing point, flash point, initial distillation point, 50% recovery temperature and final distillation point.

[0060] It should be noted that, at present, most of the tests on aviation kerosene are to analyze the concentration of free water and tiny impurities, or to analyze and detect the sulfur content in aviation kerosene, or to detect the concentration of aviation kerosene, but do not involve the detection of its physical and chemical properties, especially the detection of 20°C density, freezing point, flash point, initial distillation point, 50% recovery temperature and final distillation point. The above physical and chemical properties are very important for its storage, transportation and specific production applications. Therefore, this embodiment proposes a rapid detection method for the above multiple physical and chemical properties.

[0061] S120, collecting near-infrared spectrum data of a plurality of aviation kerosene samples, and performing at least two optimization processes on the near-infrared spectrum data.

[0062] This embodiment uses a low-cost transmission near-infrared spectrometer to collect near-infrared spectrum data of aviation kerosene samples. Specifically, each sample is scanned by a transmission near-infrared spectrometer, and the oil sample is always uniform during the measurement process. The collection range is 4000cm-1 ~12000cm -1 , the scanning times range from 5 to 120 times.

[0063] Specifically, the near-infrared spectrum data is optimized at least twice, including: using the first-order differential and vector normalization method to optimize the near-infrared spectrum data for the first time to eliminate baseline tilt, background interference and miscellaneous peaks, so as to eliminate noise information in the spectrum and improve the signal-to-noise ratio of the spectrum. And, using the principal component analysis-Mahalanobis distance method (Principal Components Analysis-Mahalanobis distance, PCA-MD) to optimize the second time to eliminate abnormal aviation kerosene samples.

[0064] Specifically, the vector normalization formula is as follows:

[0065] x′=(x-Min(x)) / (Max(x)-Min(x))

[0066] Where Max(x) is the maximum value of the sample data, Min(x) is the minimum value of the sample data, x is the sample data to be normalized, and x' is the normalized sample data.

[0067] Furthermore, the PCA-MD method is used to perform a second optimization process to eliminate abnormal aviation kerosene samples, including: using the PCA-MD method to calculate the Mahalanobis distance value and the Mahalanobis distance standard deviation of the aviation kerosene samples after the first optimization process; obtaining the Mahalanobis distance threshold of the abnormal aviation kerosene samples according to the Mahalanobis distance value and the Mahalanobis distance standard deviation; comparing the Mahalanobis distance value of each aviation kerosene sample after the first optimization process with the Mahalanobis distance threshold of the abnormal aviation kerosene sample, and eliminating the corresponding sample in response to the Mahalanobis distance value being greater than the Mahalanobis distance threshold. In other words, the Mahalanobis distance value of each optimized sample is compared with the Mahalanobis distance threshold of the abnormal sample to eliminate the sample greater than the Mahalanobis distance threshold.

[0068] The calculation formula of the Mahalanobis distance threshold is as follows:

[0069] W MD =mean(MD)+k*std(MD)

[0070] Where W MD is the Mahalanobis distance threshold of abnormal aviation kerosene samples, mean(MD) is the mean value function of MD, std(MD) is the standard deviation function of MD, k is the threshold adjustment weight coefficient; and the calculation formula of the Mahalanobis distance value is as follows:

[0071]

[0072] Where MD is the Mahalanobis distance value of aviation kerosene sample i, t i is the score vector of aviation kerosene sample i, C is the covariance matrix of the score matrix of all aviation kerosene samples, and T is the score matrix of all aviation kerosene samples.

[0073] It should be further explained that, in this embodiment, the near infrared spectrum data after optimization processing is also used to establish an aviation kerosene spectrum database.

[0074] S130. Using a binary linear classifier, a regression model is established to associate the physical and chemical property data of aviation kerosene with the near infrared spectrum data.

[0075] It should be noted that based on the aviation kerosene spectral database, the training set samples and the validation set samples were randomly selected in a ratio of 5:1, that is, the screened spectral data were collected, 83% of the samples were randomly selected to form the training set, and the remaining 17% of the samples were used as the validation set. The regression model was established using the training set, and the validation set was used to verify the regression model.

[0076] Furthermore, before using a binary linear classifier to establish a regression model that associates the physicochemical property data of aviation kerosene samples with near-infrared spectral data, it also includes: dividing the near-infrared spectral data into n spectral intervals, establishing local prediction models for each of the n spectral intervals, using the root mean square error (RMSE) as an evaluation criterion, and increasing the spectral intervals participating in modeling starting from the interval with the smallest cross-validation root mean square error, and selecting 2-5 spectral intervals with the smallest root mean square error as the preset spectral intervals for constructing the regression model.

[0077] The calculation formula of the root mean square error is as follows:

[0078]

[0079] Where yi is the predicted value of aviation kerosene sample i, is the true value, and n is the number of samples.

[0080] Specifically, based on the preset spectral interval, a regression model associating the physicochemical property data of aviation kerosene with the near-infrared spectral data is established using a binary linear classifier, including:

[0081] The near-infrared spectral data of the preset spectral range are mapped to a high-dimensional feature space and an association model is established with the physical and chemical property data, as follows:

[0082]

[0083]

[0084]

[0085] In the formula, is the mapping function that maps the near-infrared spectral input data to the high-dimensional feature space, γ is the regularization parameter, and e k is the error variable, the weight vector ω∈R n , bias value b∈R, output data y of the kth aviation kerosene sample k ∈R;

[0086] The Lagrange multiplier method is used to obtain the following formula:

[0087]

[0088] In the formula, the Lagrange multiplier α k ∈R, the input data x of the kth aviation kerosene sample k ∈R N ;

[0089]

[0090]

[0091]

[0092]

[0093] The binary linear classifier regression model is solved as follows:

[0094]

[0095] S140, collecting near infrared spectrum data of the sample to be tested, and optimizing it at least twice, substituting the optimized data into a regression model to obtain a plurality of physical and chemical properties of the sample to be tested.

[0096] It should be noted that the near-infrared spectrum data of the aviation kerosene sample to be tested is substituted into the above regression model to obtain multiple physical and chemical properties of the sample to be tested, and the prediction process time is shorter and the accuracy is higher.

[0097] The following is a specific example of a method for constructing a prediction model for aviation kerosene properties using a binary linear classifier:

[0098] This example takes the prediction of 20℃ density as an example. The specific steps include:

[0099] Step 1: Establish a library of 300 aviation kerosene samples and measure the density data of all aviation kerosene samples using standard methods;

[0100] Step 2: Control the sample temperature at 20°C and use a near-infrared spectrometer to collect near-infrared spectrum data of the aviation kerosene sample with a wavelength range of 4000cm-1 ~12000cm -1 , the number of scans is 100;

[0101] Step 3: Perform the first optimization processing on the near-infrared spectrum data of aviation kerosene collected in step 2 using the first-order differential + vector normalization method to establish an aviation kerosene spectrum database;

[0102] Step 4: Perform PCA-MD principal component analysis-Mahalanobis distance analysis on the spectral data after the first optimization process to identify abnormal samples and eliminate samples with abnormal Mahalanobis distance values;

[0103] Step 5: Establish a spectral database of aviation kerosene and randomly select training set samples and validation set samples in a ratio of 5:1;

[0104] Step 6: Divide the scanning wavelength into 5 segments, establish binary linear classifier local prediction models respectively, and determine the best modeling interval as (4200cm -1 -4700cm -1 、6240cm -1 -6950cm -1 The characteristic modeling intervals are combined to participate in the final modeling to obtain the regression model.

[0105] Step 7: Collect near infrared spectrum data of the aviation kerosene sample to be tested, perform two optimization processes on it, and substitute the processed data into the above regression model to obtain the 20°C density data of the aviation kerosene sample to be tested.

[0106] It should be noted that the same method as above can be used to detect other different physical and chemical properties, such as freezing point, flash point, initial distillation point, 50% recovery temperature and final distillation point. By establishing a regression model, multiple physical and chemical properties can be obtained in one operation.

[0107] The present invention provides a method for constructing a prediction model of aviation kerosene properties using a binary linear classifier, which has the following beneficial effects compared to the prior art: the present invention constructs a regression model for predicting aviation kerosene based on a binary linear classifier, and a single operation can detect multiple physical and chemical properties of aviation kerosene, and the processing of the original spectrum, the establishment and update of the model do not need to be performed offline, and the regression model can be updated in a timely and effective manner to improve the prediction accuracy. In addition, the method of the present invention does not require any disposable consumables, has low cost, is easy to operate, has low technical threshold requirements for operators, and is suitable for large-scale promotion.

[0108] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for constructing a prediction model for aviation kerosene properties using a binary linear classifier, characterized in that: The steps include: Obtaining physical and chemical property data of multiple aviation kerosene samples, wherein the physical and chemical properties include density at 20°C, freezing point, flash point, initial distillation point, 50% recovery temperature and final distillation point; Collecting near infrared spectral data of the plurality of aviation kerosene samples, and performing at least two optimization processes on the near infrared spectral data, including: Performing a first optimization process on the near infrared spectrum data using first order differential and vector normalization method; The PCA-MD method was used for the second optimization process to eliminate abnormal aviation kerosene samples, including: The PCA-MD method was used to calculate the Mahalanobis distance value and Mahalanobis distance standard deviation of the aviation kerosene sample after the first optimization treatment; Obtaining a Mahalanobis distance threshold of the abnormal aviation kerosene sample according to the Mahalanobis distance value and the Mahalanobis distance standard deviation; The Mahalanobis distance value of each aviation kerosene sample after the first optimization process is compared with the Mahalanobis distance threshold of the abnormal aviation kerosene sample, and the corresponding sample is removed in response to the Mahalanobis distance value being greater than the Mahalanobis distance threshold; wherein, The calculation formula of the Mahalanobis distance threshold is as follows: W MD =mean(MD)+k*std(MD) Where W MD is the Mahalanobis distance threshold of abnormal aviation kerosene samples, mean(MD) is the mean function of MD, std(MD) is the standard deviation function of MD, k is the threshold adjustment weight coefficient; and, The calculation formula of the Mahalanobis distance value is: Where MD is the Mahalanobis distance value of aviation kerosene sample i, t i is the score vector of aviation kerosene sample i, C is the covariance matrix of the score matrix of all aviation kerosene samples, and T is the score matrix of all aviation kerosene samples; The near-infrared spectral data are divided into n spectral intervals, local prediction models are established for the n spectral intervals respectively, the root mean square error is used as an evaluation criterion, a preset spectral interval is selected, and the feature modeling intervals are combined to participate in the final modeling, so as to obtain a regression model that associates the physicochemical property data of the aviation kerosene sample with the near-infrared spectral data by using a supporting binary linear classifier, including: mapping the near-infrared spectral data of the preset spectral interval to a high-dimensional feature space and establishing and establishing an association model with the physicochemical property data, as follows: In the formula, is the mapping function that maps the near-infrared spectral input data to the high-dimensional feature space, γ is the regularization parameter, and e k is the error variable, the weight vector ω∈R n , bias value b∈R, output data y of the kth aviation kerosene sample k ∈R; The Lagrange multiplier method is used to obtain the following formula: In the formula, the Lagrange multiplier α k ∈R, the input data x of the kth aviation kerosene sample k ∈R N ; The binary linear classifier regression model is solved as follows: Near infrared spectrum data of the sample to be tested is collected and optimized at least twice, and the optimized data is substituted into the regression model to obtain multiple physical and chemical properties of the sample to be tested.

2. The method according to claim 1, characterized in that The near infrared spectrum data of the aviation kerosene sample is collected by using a transmission near infrared spectrometer; wherein, The acquisition range is 4000cm -1 ~12000cm -1 , the scanning times range from 5 to 120 times.

3. The method according to claim 1, characterized in that The vector normalization formula is as follows: x′=(x-Min(x)) / (Max(x)-Min(x)) Where Max(x) is the maximum value of the sample data, Min(x) is the minimum value of the sample data, x is the sample data to be normalized, and x' is the normalized sample data.

4. The method according to claim 1, characterized in that: The calculation formula of the root mean square error is as follows: Where yi is the predicted value of aviation kerosene sample i, is the true value, and n is the number of samples.

Images