A fuel property prediction method, device, apparatus and storage medium

Abstract

The application discloses a fuel property prediction method and device, equipment and a storage medium, and the method comprises the following steps: analyzing a to-be-tested fuel sample by means of a two-dimensional gas chromatograph to obtain hydrocarbon composition data of the to-be-tested fuel sample, inducing a plurality of key chemical component variables and constructing a component feature vector; performing standardization processing and principal component analysis on the component feature vector to obtain principal components meeting preset conditions and calculating a principal component score vector; inputting the principal component score vector into a trained nonlinear regression prediction model to obtain a predicted value of a physical and chemical property of the to-be-tested fuel sample; wherein the nonlinear regression prediction model comprises linear terms, square terms and interaction terms of the principal component score vector; wherein the square terms and the interaction terms are used to represent the nonlinear effect of the components on the physical and chemical property and the synergistic effect between the components. Thus, the problem of large fuel sample consumption and long test period can be solved, and the fuel property can be accurately predicted.

Description

Technical Field

[0001] This application relates to the field of aviation fuel detection and chemometrics technology, and in particular to a fuel property prediction method, apparatus, equipment and storage medium. Background Technology

[0002] Sustainable aviation fuel (SAF) can effectively reduce emissions of harmful substances such as particulate matter and sulfur oxides. It has multiple advantages, including carbon reduction and environmental protection, energy substitution, recycling, and compatibility with the existing civil aviation system. It has significant engineering application value and promising prospects for industrial promotion.

[0003] Currently, SAF (Self-Produced Air Fuel) is mainly prepared through processes such as hydrogenation of esters and fatty acids (HEFA) and Fischer-Tropsch synthesis of aviation kerosene (FT-SPK). Although this type of fuel shares some similarities with traditional aviation kerosene in terms of physicochemical properties such as density, viscosity, and calorific value, its hydrocarbon composition, molecular structure distribution, and component ratios at the microscopic level differ significantly from traditional aviation kerosene. According to international airworthiness standards such as ASTM D7566 and ASTM D4054, SAF must undergo a rigorous airworthiness certification process before being put into actual use, which typically involves the evaluation of multiple physicochemical properties. However, existing fuel property evaluation methods mainly rely on traditional physicochemical experiments to test fuel properties, but these methods suffer from problems such as large sample consumption and long testing cycles. Furthermore, some existing technologies predict fuel properties to assess them, but these prediction methods struggle to accurately characterize the patterns of property changes, leading to inaccurate fuel property predictions. Summary of the Invention

[0004] This application provides a fuel property prediction method, apparatus, equipment, and storage medium, which can solve the problems of large fuel sample consumption and long testing cycle, accurately characterize the property change law, and accurately predict fuel properties.

[0005] In a first aspect, embodiments of this application provide a fuel property prediction method, including:

[0006] The hydrocarbon composition data of the fuel sample to be tested is obtained by full two-dimensional gas chromatography analysis. Based on the preset chemical structure and carbon number range, the hydrocarbon composition data is summarized into several key chemical component variables, and a component feature vector is constructed based on the several key chemical component variables.

[0007] The component feature vectors are standardized, and principal component analysis is performed on the standardized component feature vectors to obtain principal components that meet preset conditions, and the principal component score vectors are calculated.

[0008] The principal component score vector is input into the trained nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested; wherein, the nonlinear regression prediction model includes linear terms, square terms, and interaction terms of the principal component score vector; wherein, the square terms and the interaction terms are used to characterize the nonlinear effects of components on physicochemical properties and the synergistic effects between components.

[0009] Secondly, embodiments of this application provide a fuel property prediction device, comprising:

[0010] The first processing module is used to analyze the fuel sample to be tested through full two-dimensional gas chromatography to obtain the hydrocarbon composition data of the fuel sample to be tested, and to summarize the hydrocarbon composition data into several key chemical component variables based on the preset chemical structure and carbon number range, and to construct a component feature vector based on the several key chemical component variables.

[0011] The second processing module is used to standardize the component feature vectors, perform principal component analysis on the standardized component feature vectors to obtain principal components that meet preset conditions, and calculate the principal component score vectors.

[0012] The third processing module is used to input the principal component score vector into the trained nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested; wherein, the nonlinear regression prediction model includes linear terms, square terms, and interaction terms of the principal component score vector; wherein, the square terms and the interaction terms are used to characterize the nonlinear effects of the components on the physicochemical properties and the synergistic effects between the components.

[0013] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method provided in embodiments of this application.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method provided in embodiments of this application.

[0015] The technical solution provided in this application, through full two-dimensional gas chromatography analysis of the fuel sample, can map and summarize the complex fuel system containing tens of millions of compounds into key chemical component variables characterizing the chemical composition. This reduces the amount of fuel sample used while achieving comprehensive characterization of the microscopic components. By standardizing the component feature vector constructed from the key chemical component variables and performing principal component analysis, and calculating the principal component score vector, multiple collinearity problems among the original key chemical component variables can be eliminated. This dimensionality reduction also preserves effective information reflecting the main chemical characteristics of the fuel sample. The principal component score vector is input into the nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested. The nonlinear regression prediction model includes the linear term, square term, and interaction term of the principal component score vector. By using the nonlinear regression prediction model containing the square term and interaction term, the law of physical property change can be accurately described, and the nonlinear response characteristics of fuel properties with component changes can be accurately described, thus achieving high-precision prediction of physicochemical properties. By using a combination of full two-dimensional gas chromatography analysis and principal component analysis, and using the nonlinear regression prediction model to predict the predicted values ​​of physical properties, the problems of long testing cycles and large consumption in fuel sample property evaluation can be solved. Attached Figure Description

[0016] Figure 1 A flowchart of a fuel prediction method provided for the implementation of this application;

[0017] Figure 2 A flowchart illustrating the training method for the nonlinear regression prediction model provided in this application embodiment;

[0018] Figure 3 Crustenite plots from principal component analysis;

[0019] Figure 4 This is a schematic diagram of the distribution of fuel samples in a two-dimensional principal component space consisting of the first principal component (PC1) and the second principal component (PC2);

[0020] Figure 5 This is a schematic diagram of the distribution of fuel samples in the three-dimensional principal component space consisting of the first principal component (PC1), the second principal component (PC2), and the third principal component (PC3);

[0021] Figures 6a-6f These are schematic diagrams showing the correlation analysis results between the density prediction, final boiling point prediction, lower heating value prediction, viscosity prediction at -20℃, viscosity prediction at -40℃, and initial boiling point prediction values ​​of the nonlinear regression prediction model and their corresponding measured values ​​on the training set.

[0022] Figures 7a-7fThese are schematic diagrams showing the correlation analysis results between the density prediction, final boiling point prediction, lower heating value prediction, viscosity prediction at -20℃, viscosity prediction at -40℃, and initial boiling point prediction values ​​of the nonlinear regression prediction model and their corresponding measured values ​​on the validation set.

[0023] Figure 8 A structural block diagram of a fuel prediction device provided in an embodiment of this application;

[0024] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Figure 1 This is a flowchart of a fuel prediction method provided in an embodiment of this application. The method can be executed by a fuel prediction device, which can be implemented by software and / or hardware, and can be configured in an electronic device such as a computer. Figure 1 As shown, the method provided in this application embodiment includes the following steps:

[0027] S110: The hydrocarbon composition data of the fuel sample to be tested is obtained by full two-dimensional gas chromatography analysis, and the hydrocarbon composition data is summarized into several key chemical component variables based on the preset chemical structure and carbon number range, and a component feature vector is constructed based on the several key chemical component variables.

[0028] In this embodiment, a two-dimensional gas chromatography (GC) method can be used to separate and analyze the fuel sample to obtain hydrocarbon composition data. For example, a two-dimensional GC equipped with a flame ionization detector can be used for analysis. By setting appropriate temperature programs and carrier gas flow rates, hundreds or even thousands of hydrocarbon compounds in the fuel sample can be separated and detected. Alternatively, a two-dimensional GC equipped with a mass spectrometer can be used to acquire mass spectrometric information of the components during separation, thereby aiding in the identification and quantification of hydrocarbon components. After obtaining the raw chromatographic data, chromatographic workstation software can be used to integrate and identify the chromatographic peaks to obtain the relative or absolute content of each hydrocarbon component, forming hydrocarbon composition data.

[0029] In this embodiment, based on a preset chemical structure and carbon number range, the hydrocarbon composition data is summarized into several key chemical component variables, and a component feature vector is constructed based on these key chemical component variables. This allows for dimensionality reduction and abstraction of high-dimensional hydrocarbon composition data, extracting component information that has a key impact on the properties of the fuel sample under test. Optionally, the number of key chemical component variables is not less than a preset threshold to cover different molecular structure types; wherein, the molecular structure types include at least chain structures, cyclic structures, and aromatic structures; the preset threshold can be 12, or it can be set as needed. The key chemical component variables include isoalkanes with a carbon number range of C5-C7, isoalkanes with a carbon number range of C8-C13, isoalkanes with a carbon number range of C14-C18, n-alkanes with a carbon number range of C5-C7, n-alkanes with a carbon number range of C8-C13, n-alkanes with a carbon number range of C14-C18, n-alkanes with a carbon number range of C19+, cycloalkanes with a carbon number range of C5-C7, cycloalkanes with a carbon number range of C8-C11, cycloalkanes with a carbon number range of C12+, monocyclic aromatic hydrocarbons, and polycyclic aromatic hydrocarbons. Optionally, the key chemical component variables can be sorted according to a set order to form a component feature vector.

[0030] S120: Standardize the component feature vectors, perform principal component analysis on the standardized component feature vectors to obtain principal components that meet preset conditions, and calculate the principal component score vectors.

[0031] In this embodiment, to eliminate the influence of differences in dimensions and numerical orientations among different key chemical component variables, the component feature vectors can be standardized. For example, the Z-score standardization method can be used, subtracting the mean from the data of each variable and then dividing by its standard deviation, so that the mean of all variables is 0 and the standard deviation is 1. After standardization, the processed component feature vectors are input into the principal component analysis algorithm. Principal component analysis can be implemented using singular value decomposition (SVD) or NIPALS algorithms. During principal component analysis, the number of principal components to be retained can be determined based on preset conditions such as cumulative variance contribution rate, eigenvalue magnitude, or scree plot. For example, principal components with a preset contribution rate of 85% can be selected. Once the principal components are determined, the projection values ​​of each fuel sample to be tested onto these principal components can be calculated, thereby obtaining the principal component score vector. This principal component score vector is a low-dimensional representation of the component feature vectors, which can effectively capture the main chemical composition characteristics of the fuel sample to be tested.

[0032] S130: Input the principal component score vector into the trained nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested; wherein, the nonlinear regression prediction model includes the linear term, the square term, and the interaction term of the principal component score vector; wherein, the square term and the interaction term are used to characterize the nonlinear effect of the components on the physicochemical properties and the synergistic effect between the components.

[0033] In this embodiment, optionally, the nonlinear regression prediction model is:

[0034]

[0035] in, These are predicted values ​​for physicochemical properties. For constant terms, The coefficients are linear. The coefficient of the quadratic term, The coefficients of the interaction terms; For the first Each principal component score vector; For the first Each principal component score vector; The number of principal components. The mathematical expression of the nonlinear regression prediction model employs a quadratic polynomial regression model, which includes a constant term, linear terms of the principal component score vectors, square terms, and interaction terms between different principal component score vectors. This explicit mathematical form can accurately capture the synergistic effects and nonlinear rheological characteristics between components, and more accurately fit complex component-property relationships. These are predicted values ​​of physicochemical properties, which are the output of a nonlinear regression prediction model. They represent the estimated values ​​of the fuel sample under test in terms of specific physicochemical properties, directly reflecting the performance indicators of the fuel. These can be predicted values ​​of specific physicochemical properties such as density, viscosity, lower heating value, and distillation range.

[0036] In this embodiment, optionally, the nonlinear regression prediction model includes a density prediction model, a first low-temperature viscosity prediction model, a second low-temperature viscosity prediction model, a lower heating value prediction model, a first boiling range prediction model, and a second boiling range prediction model. Inputting the principal component score vector into the trained nonlinear regression prediction model to obtain the predicted physicochemical properties of the fuel sample to be tested includes: inputting the principal component score vector into the trained density prediction model, the first low-temperature viscosity prediction model, the second low-temperature viscosity prediction model, the lower heating value prediction model, the first boiling range prediction model, and the second boiling range prediction model, respectively, to obtain the predicted density, the first low-temperature viscosity, the second low-temperature viscosity, the lower heating value, the initial boiling point, and the final boiling point of the fuel sample to be tested. Among them, the corresponding coefficients in the density prediction model, the first low-temperature viscosity prediction model, the second low-temperature viscosity prediction model, the lower heating value prediction model, the first boiling range prediction model, and the second boiling range prediction model are not the same. By inputting the principal component score vector into the nonlinear regression prediction model, the predicted values ​​of the physicochemical properties of the fuel sample to be tested, such as density, first low-temperature viscosity, second low-temperature viscosity, lower heating value, initial boiling point, and final boiling point, can be output.

[0037] In related technologies, statistical methods such as multiple linear regression are used to predict fuel properties. This application, through the introduction of a nonlinear regression prediction model, particularly including the squared term and interaction term of the principal component score vector, can more accurately characterize the nonlinear effects of components on physicochemical properties and the synergistic effects between components in complex fuel systems, thus accurately describing the laws governing property changes. Traditional linear models may fail to capture the volumetric non-additive effect caused by the mixing of light and heavy components, leading to prediction errors. However, the introduction of interaction terms in this application effectively simulates such complex interactions between components, making the predicted physicochemical properties closer to reality. Furthermore, this application uses full two-dimensional gas chromatography to obtain detailed hydrocarbon composition data and combines it with principal component analysis for data dimensionality reduction. This not only overcomes the problems of high sample consumption and long testing cycles in existing technologies (because full two-dimensional gas chromatography only requires trace amounts of fuel samples for analysis and is relatively fast), but also transforms high-dimensional component data into a few physically meaningful principal components, effectively solving the problems of excessively high original data dimensionality and strong collinearity between variables, thus improving the stability and interpretability of the nonlinear regression prediction model. More importantly, the embodiments of this application predict the structure-property relationship based on the microscopic chemical composition of the fuel, rather than relying solely on macroscopic parameters. This gives the method a stronger ability to analyze structure-property relationships. By analyzing the linear, quadratic, and interaction terms of the principal component scores in the nonlinear regression prediction model, a deeper understanding can be gained of how different chemical components (reflected through the principal components) affect the macroscopic physicochemical properties of the fuel, and how they interact with each other. This in-depth insight into structure-property relationships is difficult to provide by simple linear models in the prior art, and has important guiding significance for the research and development and formulation optimization of sustainable aviation fuels, which can significantly improve research and development efficiency and engineering application value.

[0038] In summary, the technical solution provided in this application, through full two-dimensional gas chromatography analysis of the fuel sample, can map and summarize the complex fuel system containing tens of millions of compounds into key chemical component variables characterizing its chemical composition. This reduces the amount of fuel sample required while achieving comprehensive characterization of the microscopic components. By standardizing the component feature vector constructed from the key chemical component variables and performing principal component analysis, and calculating the principal component score vector, multiple collinearity problems among the original key chemical component variables can be eliminated. This dimensionality reduction also preserves effective information reflecting the main chemical characteristics of the fuel sample. The principal component score vector is input into a nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested. The nonlinear regression prediction model includes linear terms, square terms, and interaction terms of the principal component score vector. By using a nonlinear regression prediction model that includes square terms and interaction terms, the law of change of physical properties can be accurately described, and the nonlinear response characteristics of fuel properties with changes in components can be accurately described, thus achieving high-precision prediction of physicochemical properties. By using a combination of full two-dimensional gas chromatography analysis and principal component analysis, and using a nonlinear regression prediction model to predict the predicted values ​​of physical properties, the problems of long testing cycles and large consumption in fuel sample property evaluation can be solved.

[0039] Based on the above embodiments, prior to S110, the method provided in this application embodiment may further include training a nonlinear regression prediction model. For example... Figure 2 As shown, the specific training process may include the following steps:

[0040] S210: Construct a training set using 100% sustainable aviation fuel, blended fuels containing different proportions of sustainable aviation fuel and aromatics, blended fuels containing the sustainable aviation fuel and conventional jet fuel, and 100% conventional jet fuel as fuel samples.

[0041] In this embodiment, 24 fuel samples can be prepared, of which 17 are used for training the nonlinear regression model and 7 are used for validating the nonlinear regression model. The samples in the training set cover different blending ratios of sustainable aviation fuels and control fuels to fully reflect the diversity of actual fuel usage scenarios: including 100% SAF, 92% SAF (SAF and aromatics blended at a volume ratio of 92:8), 85% SAF (independent validation set, SAF and aromatics blended at a volume ratio of 85:15), 50% SAF (92% SAF and No. 3 jet fuel blended at a volume ratio of 1:1), and pure No. 3 jet fuel. The aromatic components cover different fractions, including light, medium, and heavy fractions at 100℃, 150℃, and 200℃, to ensure that the samples are representative and complete in terms of chemical composition. No. 3 jet fuel is conventional jet fuel, i.e., No. 3 jet fuel, with the national standard code RP-3.

[0042] S220: The fuel samples in the training set are analyzed by full two-dimensional gas chromatography to obtain the hydrocarbon composition data of each fuel sample. Based on the preset chemical structure and carbon number range, the hydrocarbon composition data of each fuel sample is summarized into several key chemical component variables, and an independent variable matrix is ​​constructed based on the several key chemical component variables of each fuel sample.

[0043] In this embodiment, the fuel samples in the training set are analyzed by full two-dimensional gas chromatography to obtain the hydrocarbon composition data of each fuel sample. The specific method can refer to the full two-dimensional gas chromatography analysis method described in S110 above.

[0044] Table 1 contains detailed information on the key component variables of 24 fuel samples, including 17 fuel samples from the training set and 7 fuel samples from the validation set.

[0045] Table 1

[0046]

[0047] In this matrix, variables A–F represent paraffins with different carbon number ranges, variables G–H represent cycloalkanes (naphthenes), and variables I–L represent aromatics (including monocyclic and polycyclic aromatic hydrocarbons). Several key chemical component variables from each fuel sample are sorted according to a predetermined order to construct an independent variable matrix. For example, if there are 17 fuel samples, and each fuel sample is categorized into 12 key chemical component variables, a 17×12 independent variable matrix can be constructed.

[0048] S230: Standardize the independent variable matrix and perform principal component analysis on the standardized independent variable matrix to determine the principal component score vector for each fuel sample.

[0049] In this embodiment, the independent variable matrix, for example, a 17×12 matrix, was first standardized using Z-scores to eliminate the influence of differences in the dimensions of the variables on the analysis results. Then, the Kaiser-Meyer-Olkin (KMO) test and Bartlett's test of sphericity were used to verify whether the data in the independent variable matrix were suitable for principal component analysis. If KMO > 0.7 and P < 0.05, the prerequisites for principal component analysis were met. After meeting the prerequisites for principal component analysis, principal component analysis was performed on the independent variable matrix, and a scree plot was plotted. Figure 3 Among them, such as Figure 4 and Figure 5As shown, the principal component analysis results indicate that the cumulative variance contribution rate of the first three principal components (PC1, PC2, and PC3) reaches 91.112%, indicating that the dimensionality-reduced principal components can almost completely retain the information of the original data. Among them, the positive loading of PC1 (65.5%) mainly comes from isoalkanes, while the negative loading mainly corresponds to cycloalkanes and aromatics, reflecting the significant differences in fuel molecular structure types; the positive loading of PC2 (13.7%) corresponds to light alkanes, and the negative loading corresponds to heavy aromatics, revealing the molecular weight distribution (light / heavy) characteristics of the fuel; PC3 (11.9%) is mainly affected by extra-heavy alkanes (C19 and above) and bicyclic aromatics, indicating that the contribution of heavy tailing components in the fuel to the overall properties cannot be ignored.

[0050] In this embodiment, principal component analysis is performed on the independent variable matrix to obtain the top K principal components with a cumulative variance contribution rate greater than 85%. For example, by... Figure 3 It can be seen that the top three principal components have a cumulative variance contribution rate exceeding 85%. The score vectors of these top three principal components are calculated, resulting in the score vectors of the three principal components for each sample, denoted as follows: , and .

[0051] S240: Input the principal component score vector of each sample into the nonlinear regression prediction model to obtain the corresponding physicochemical property prediction values. Adjust the coefficients in the nonlinear regression prediction model based on the physicochemical property prediction values ​​and the corresponding measured physicochemical property values ​​until the training of the nonlinear regression prediction model is completed.

[0052] In this embodiment, the principal component score vectors of each fuel sample are input into a nonlinear regression prediction model to obtain the corresponding predicted physicochemical properties. The coefficients in the nonlinear regression prediction model are adjusted using these predicted and measured values ​​until a trained nonlinear regression prediction model is obtained. Specifically, the least squares method can be used to train the nonlinear regression prediction model and solve for each coefficient. The score vectors of each fuel sample in the training set are then input into the trained nonlinear regression prediction model to obtain the predicted physicochemical properties. The predicted values ​​are compared with the corresponding measured values ​​to determine the fitting of the training set samples. Figures 6a-6f The correlation results between the predicted density, first low-temperature viscosity, second low-temperature viscosity, lower heating value, initial boiling point, and final boiling point of the fuel samples in the training set and the corresponding measured values ​​show that the data points are evenly distributed on both sides of the y=x reference line, with no systematic error and high prediction accuracy.

[0053] Building upon the above embodiments, the analysis can further include examining the component information contained in the principal components based on the linear coefficients, quadratic coefficients, interaction coefficients, and the corresponding principal component score vectors. Specifically, the intrinsic relationships and nonlinear interactions implied by the principal components are analyzed using the linear coefficients, quadratic coefficients, interaction coefficients, and the corresponding principal component score vectors. The positive or negative sign of the linear coefficients, quadratic coefficients, and interaction coefficients determines the driving effect of the corresponding principal component vectors on the physicochemical properties. For example, if the coefficient is positive, the corresponding principal component vector has a positive driving effect on the physicochemical properties; if the coefficient is negative, the corresponding principal component vector has a negative driving effect on the physicochemical properties. This allows for a deeper understanding of the contribution patterns and mechanisms of different principal component score vectors on various physicochemical properties, enabling technicians to clearly understand how fuel components affect their macroscopic properties.

[0054] In this embodiment, the nonlinear regression prediction model includes a density prediction model, a first low-temperature viscosity prediction model, a second low-temperature viscosity prediction model, a lower heating value prediction model, a first distillation range prediction model, and a second distillation range prediction model. Each model is trained according to the aforementioned training method, and the density prediction model trained using the training set is as follows:

[0055] ;

[0056] in, This is the predicted density of the fuel at 15℃; the predicted fuel density is mainly affected by... The linear terms dominate. Meanwhile, The negative quadratic term and The negative quadratic term successfully captured the nonlinear synergistic effect between components and the fine-tuning effect of heavy components, confirming that the density does not change monotonically with the component ratio, but is controlled by complex interactions.

[0057] The first low-temperature viscosity prediction model that has been trained is:

[0058] ;

[0059] The second low-temperature viscosity prediction model, after training, is as follows:

[0060] ;

[0061] in, and These are the first and second predicted low-temperature viscosity values, respectively, representing the viscosities at -20℃ and -40℃; where Viscosity -20℃ and Viscosty -40℃It is a key physical indicator characterizing the low-temperature fluidity of fuels. Temperatures dropping from -20°C to -40°C significantly amplify the nonlinear effect of chemical components on viscosity. The significant jump in the linear coefficient confirms that the restricted movement of alkanes at low temperatures is the dominant factor in viscosity deterioration. It is worth noting that... The quadratic term has a large positive value at -40°C, indicating a sharp increase in the sensitivity of viscosity to changes in this component and a significant enhancement in intermolecular nonlinear interactions. Furthermore, at -40°C, The simultaneous increase of the quadratic term further confirms the decisive control effect of high-carbon heavy components on ultra-low temperature fluidity.

[0062] The trained low-temperature calorific value prediction model is as follows:

[0063] ;

[0064] in, This is a predicted value for the lower heating value (LHV) of a fuel; the LHV of a fuel is primarily controlled by its hydrogen-to-carbon ratio (H / C ratio). Under the same carbon number conditions, the H / C ratio of different hydrocarbons follows a decreasing pattern: alkanes > cycloalkanes > aromatics. Given that the calorific value of hydrogen is much higher than that of carbon, there is a significant positive correlation between the H / C ratio of a fuel and its LHV; that is, the higher the H / C ratio, the higher the LHV of the fuel. The LHV prediction model clearly quantifies the influence of chemical composition on LHV. and The consistent positive coefficients indicate that increasing the content of alkanes has a significant positive effect on the lower heating value; while increasing the content of cycloalkanes or aromatics leads to a decrease in the lower heating value. Further comparison of the coefficients reveals that... Sensitivity to LHV is The fact that the weighting difference is three times that of cycloalkanes indicates that aromatic hydrocarbons are the dominant factor causing the loss of lower heating values ​​compared to cycloalkanes.

[0065] The first distillation range prediction model that has been trained is as follows:

[0066] ;

[0067] The trained second distillation range prediction model is as follows:

[0068] ;

[0069] in, , These represent the predicted initial boiling point (IBP) and final boiling point (FBP) of the fuel, respectively. IBP and FBP characterize the volatility of the lightest and heaviest components of the fuel, respectively, and are key indicators for evaluating fuel start-up performance and combustion completeness. From a chemical composition perspective, enrichment of short-chain alkanes tends to lower IBP, which is beneficial for improving start-up performance; while an increase in the content of high-boiling-point heavy components and aromatic hydrocarbons significantly increases FBP, potentially leading to incomplete combustion. The first and second boiling point prediction models, after training, show that the increase in the score vectors of each principal component exhibits a dual effect of "significantly lowering IBP but greatly increasing FBP." In particular... The strong negative linear coefficient and quadratic term characteristics (square term and interaction term) accurately capture the nonlinear trend of "accelerated decline" of IBP caused by the enrichment of light alkanes. This seemingly contradictory phenomenon (decreasing IBP while increasing FBP) reveals that, compared with cycloalkanes and aromatics, certain alkanes have a wider distillation range distribution. This conclusion is consistent with the actual chemical composition of fuels.

[0070] After training, the density prediction model, the first low-temperature viscosity prediction model, the second low-temperature viscosity prediction model, the lower heating value prediction model, the first distillation range prediction model, and the second distillation range prediction model are obtained. The principal component score vectors of the sample to be tested can be input into each model to obtain the corresponding predicted physicochemical properties. These predicted values ​​can then be used to assess fuel properties and thus conduct airworthiness certification. It should be noted that the fuel samples used in the training set for the training models have similar compositions to the fuel samples to be tested, ensuring consistency in the number of principal components.

[0071] Based on the above embodiments, the method provided in this application may further include validating the trained nonlinear regression model. Specifically, samples not involved in model training, i.e., 18-24 (85% SAF), are selected as an independent validation set. Full two-dimensional gas chromatography analysis is performed on the samples in the validation set to obtain several key chemical component variables. An independent variable matrix is ​​constructed, and the independent variable matrix is ​​standardized. Principal component analysis is then performed to determine the principal component score vector. This principal component score vector is input into the trained density prediction model, the first low-temperature viscosity prediction model, the second low-temperature viscosity prediction model, the lower heating value prediction model, the first distillation range prediction model, and the second distillation range prediction model for validation, obtaining the corresponding physicochemical property prediction values. The validation results show that, as Figures 7a-7f As shown, the predicted values ​​of physicochemical properties obtained through the validation set are highly consistent with the corresponding measured values ​​of physicochemical properties, and their coefficient of determination is high. All remained above 0.9. Further error distribution analysis showed that the data points were evenly distributed across... No significant systematic bias was observed on either side of the diagonal, indicating that the model has good predictive stability and reliability.

[0072] The above verification results fully demonstrate that the nonlinear regression prediction model constructed in this application (specifically including a density prediction model, a first low-temperature viscosity prediction model, a second low-temperature viscosity prediction model, a lower heating value prediction model, a first distillation range prediction model, and a second distillation range prediction model) does not merely fit or memorize training samples, but effectively extracts and characterizes the intrinsic correlation between fuel chemical composition and macroscopic physicochemical properties, thus exhibiting excellent generalization ability. Based on this, the model can be applied to the rapid performance screening and evaluation of unknown sustainable aviation fuels, providing an efficient and reliable technical means for the research and development and engineering application of new aviation fuels.

[0073] Figure 8 This is a structural block diagram of a fuel property prediction device provided in an embodiment of this application, such as... Figure 8 As shown, the device includes:

[0074] The first processing module 810 is used to analyze the fuel sample to be tested through full two-dimensional gas chromatography, obtain the hydrocarbon composition data of the fuel sample to be tested, and summarize the hydrocarbon composition data into several key chemical component variables based on the preset chemical structure and carbon number range, and construct a component feature vector based on the several key chemical component variables.

[0075] The second processing module 820 is used to standardize the component feature vector, perform principal component analysis on the standardized component feature vector to obtain principal components that meet preset conditions, and calculate the principal component score vector.

[0076] The third processing module 830 is used to input the principal component score vector into the trained nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested; wherein, the nonlinear regression prediction model includes linear terms, square terms, and interaction terms of the principal component score vector; wherein, the square terms and the interaction terms are used to characterize the nonlinear effects of the components on the physicochemical properties and the synergistic effects between the components.

[0077] like Figure 9 As shown in the figure, this application provides an electronic device, including a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.

[0078] Memory 113 is used to store computer programs;

[0079] In one embodiment of this application, the processor 111, when executing a program stored in the memory 113, implements the method provided in any of the foregoing method embodiments.

[0080] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method provided in any of the foregoing method embodiments.

[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0082] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0083] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this application.

Claims

1. A method for predicting fuel properties, characterized in that, include: The hydrocarbon composition data of the fuel sample to be tested is obtained by full two-dimensional gas chromatography analysis. Based on the preset chemical structure and carbon number range, the hydrocarbon composition data is summarized into several key chemical component variables, and a component feature vector is constructed based on the several key chemical component variables. The component feature vectors are standardized, and principal component analysis is performed on the standardized component feature vectors to obtain principal components that meet preset conditions, and the principal component score vectors are calculated. The principal component score vector is input into the trained nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested; wherein, the nonlinear regression prediction model includes linear terms, square terms, and interaction terms of the principal component score vector; wherein, the square terms and the interaction terms are used to characterize the nonlinear effects of components on physicochemical properties and the synergistic effects between components.

2. The method according to claim 1, characterized in that, The nonlinear regression prediction model is as follows: ； in, These are predicted values ​​for physicochemical properties. For constant terms, The coefficients are linear. The coefficient of the quadratic term, The coefficients of the interaction terms; For the first Each principal component score vector; For the first Each principal component score vector; The number of main components.

3. The method according to claim 2, characterized in that, Also includes: The training set was constructed using 100% sustainable aviation fuel, blended fuels with different proportions of sustainable aviation fuel and aromatics, blended fuels with sustainable aviation fuel and conventional jet fuel, and pure conventional jet fuel as fuel samples. By analyzing the fuel samples in the training set using full two-dimensional gas chromatography, the hydrocarbon composition data of each fuel sample is obtained. Based on the preset chemical structure and carbon number range, the hydrocarbon composition data of each fuel sample is summarized into several key chemical component variables, and an independent variable matrix is ​​constructed based on the several key chemical component variables of each fuel sample. The independent variable matrix is ​​standardized, and principal component analysis is performed on the standardized independent variable matrix to determine the principal component score vector of each fuel sample. The principal component score vectors of each fuel sample are input into the nonlinear regression prediction model to obtain the corresponding physicochemical property prediction values. The coefficients in the nonlinear regression prediction model are adjusted based on the physicochemical property prediction values ​​and the corresponding measured physicochemical property values ​​until the training of the nonlinear regression prediction model is completed.

4. The method according to claim 2, characterized in that, The nonlinear regression prediction model includes a density prediction model, a first low-temperature viscosity prediction model, a second low-temperature viscosity prediction model, a lower heating value prediction model, a first distillation range prediction model, and a second distillation range prediction model. The principal component score vector is input into the trained nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested, including: The principal component score vectors are input into the trained density prediction model, the first low-temperature viscosity prediction model, the second low-temperature viscosity prediction model, the lower heating value prediction model, the first boiling range prediction model, and the second boiling range prediction model, respectively, to obtain the density prediction value, the first low-temperature viscosity prediction value, the second low-temperature viscosity prediction value, the lower heating value prediction value, the initial boiling point prediction value, and the final boiling point prediction value of the fuel sample to be tested.

5. The method according to claim 1, characterized in that, The number of key chemical component variables is no less than a preset threshold to cover different molecular structure types; wherein, the molecular structure types include at least chain structures, ring structures and aromatic structures; The key chemical component variables include isoalkanes with a carbon number range of C5-C7, isoalkanes with a carbon number range of C8-C13, isoalkanes with a carbon number range of C14-C18, n-alkanes with a carbon number range of C5-C7, n-alkanes with a carbon number range of C8-C13, n-alkanes with a carbon number range of C14-C18, n-alkanes with a carbon number range of C19+, cycloalkanes with a carbon number range of C5-C7, cycloalkanes with a carbon number range of C8-C11, cycloalkanes with a carbon number range of C12+, monocyclic aromatic hydrocarbons, and polycyclic aromatic hydrocarbons.

6. The method according to claim 4, characterized in that, Also includes: The component information contained in the principal components is analyzed based on the linear coefficients, quadratic coefficients, interaction coefficients, and the corresponding principal component score vectors.

7. The method according to claim 4, characterized in that, The principal components that meet the preset conditions are those whose cumulative variance contribution rate reaches a preset contribution rate; wherein, the preset contribution rate is 85%.

8. A fuel property prediction device, characterized in that, include: The first processing module is used to analyze the fuel sample to be tested through full two-dimensional gas chromatography to obtain the hydrocarbon composition data of the fuel sample to be tested, and to summarize the hydrocarbon composition data into several key chemical component variables based on the preset chemical structure and carbon number range, and to construct a component feature vector based on the several key chemical component variables. The second processing module is used to standardize the component feature vectors, perform principal component analysis on the standardized component feature vectors to obtain principal components that meet preset conditions, and calculate the principal component score vectors. The third processing module is used to input the principal component score vector into the trained nonlinear regression prediction model to obtain the predicted values ​​of the physicochemical properties of the fuel sample to be tested; wherein, the nonlinear regression prediction model includes linear terms, square terms, and interaction terms of the principal component score vector; wherein, the square terms and the interaction terms are used to characterize the nonlinear effects of the components on the physicochemical properties and the synergistic effects between the components.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-7.

Images