A training method and a prediction method and device of a diesel cetane number prediction model

Abstract

This invention discloses a training method, prediction method, and apparatus for a diesel cetane number prediction model, belonging to the field of petrochemical technology. The training method for the diesel cetane number prediction model includes: acquiring a training dataset, which includes explanatory variables and corresponding target variables. The explanatory variables include physicochemical index data of different diesel samples, and the target variables include the cetane numbers of different diesel samples. Based on the training dataset, a deep learning regression model is trained to obtain the diesel cetane number prediction model. The diesel cetane number prediction model constructed using this invention can quickly and accurately predict the cetane number of diesel fuel.

Description

Technical Field

[0001] This invention belongs to the field of petrochemical technology, and particularly relates to a training method, prediction method and device for a diesel cetane number prediction model. Background Technology

[0002] Cetane number is a key indicator of diesel fuel combustion performance. It refers to the volume percentage of n-hexadecane in diesel fuel compared to that in a standard fuel under specific conditions. This indicator represents the auto-ignition capability of diesel fuel, that is, its ability to spontaneously ignite in an engine. Diesel fuel with a high cetane number has better auto-ignition capability, making it easier to start and burn, thus significantly improving engine performance and efficiency. Therefore, accurate cetane number measurement is crucial for ensuring diesel fuel quality, as it directly affects engine performance and emissions levels. Inaccurate cetane number measurement may lead to unstable engine operation, increased maintenance costs, and even negative environmental impacts.

[0003] To determine cetane number, specialized cetane number testing instruments can be used. These instruments determine the number through dielectric properties or other physical methods. The testing process typically includes diesel sample preparation, instrument connection, parameter setting, and automated analysis. However, cetane number testing instruments are generally expensive, complex to operate, and time-consuming, making it impossible to comprehensively configure standard cetane number testing engines to test all diesel samples.

[0004] Another lower-cost and simpler method is to use the cetane index, calculated using an empirical formula, to characterize the cetane number. However, it's important to note that the cetane index cannot be arbitrarily used to replace the cetane number measured by standard engine testing equipment. In practical use, the cetane index has certain limitations. The correlation between the cetane index and cetane number varies depending on the source of crude oil and the blending scheme used to produce diesel fuel. Especially at high cetane numbers, the correlation decreases significantly. Therefore, the cetane index should only be used as an auxiliary means to determine the cetane number of diesel fuel, and cannot completely replace standard cetane number determination methods.

[0005] In conclusion, designing and developing a method that can quickly and accurately predict the cetane number of diesel fuel is of great significance. Summary of the Invention

[0006] The purpose of this invention is to meet practical needs by providing a training method, prediction method, and apparatus for a diesel cetane number prediction model. The diesel cetane number prediction model constructed using this method can quickly and accurately predict the cetane number of diesel.

[0007] To achieve the above-mentioned objectives, the first objective of this invention is to provide a training method for a diesel cetane number prediction model, comprising the following steps:

[0008] Obtain a training dataset, which includes explanatory variables and corresponding target variables. The explanatory variables include physicochemical index data of different diesel samples, and the target variables include cetane numbers of different diesel samples.

[0009] Based on the training dataset, a deep learning regression model is trained to obtain a diesel cetane number prediction model.

[0010] Preferably, obtaining the training dataset includes:

[0011] Diesel fuel samples were taken and tested to obtain physicochemical data of the diesel fuel samples and the cetane number of each diesel fuel sample.

[0012] A training dataset was constructed using the physicochemical index data of diesel samples and the cetane number corresponding to each diesel sample.

[0013] Preferably, the cetane number acquisition step for each diesel sample includes:

[0014] Under the standard operating conditions of the cetane number tester, the diesel sample to be tested was mixed with different standard fuels with known cetane numbers;

[0015] The ignition properties of the mixture were determined under the same conditions;

[0016] The cetane number of the diesel sample to be tested was obtained by comparison.

[0017] Preferably, the method for obtaining the deep learning regression model includes: establishing a regression model using the h2o package in R software.

[0018] Preferably, after obtaining the diesel cetane number prediction model, the fitting coefficient R is used. 2 Measuring the predictive power of the diesel cetane number prediction model; specifically including:

[0019] When R 2 When the value is ≤0.3, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is poor.

[0020] When 0.3 < R 2 When the value is ≤0.4, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is weak.

[0021] When 0.4 < R 2 When the value is ≤0.6, the predicted value fits the observed value moderately, and the prediction ability of the prediction model is moderate.

[0022] When 0.6 < R2 When the value is ≤1.0, the predicted value fits the observed value well, and the prediction model has strong predictive ability.

[0023] Preferably, the physicochemical index data include: density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20°C, cloud point, and total contaminant content.

[0024] A second objective of this invention is to provide a method for predicting the cetane number of diesel fuel, comprising:

[0025] Obtain the physicochemical index data of the diesel sample to be tested;

[0026] The physicochemical index data of the diesel sample to be tested are input into the diesel cetane number prediction model mentioned above to obtain the cetane number of the diesel sample to be tested.

[0027] A third objective of this invention is to provide a training apparatus for a diesel cetane number prediction model, comprising:

[0028] The data acquisition module acquires a training dataset, which includes explanatory variables and corresponding target variables. The explanatory variables include physicochemical index data of different diesel samples, and the target variables include cetane numbers of different diesel samples.

[0029] The model training module trains a deep learning regression model based on the training dataset to obtain a diesel cetane number prediction model.

[0030] Preferably, obtaining the training dataset includes:

[0031] Diesel fuel samples were taken and tested to obtain physicochemical data of the diesel fuel samples and the cetane number of each diesel fuel sample.

[0032] A training dataset was constructed using the physicochemical index data of diesel samples and the cetane number corresponding to each diesel sample;

[0033] The steps for obtaining the cetane number of each diesel sample include:

[0034] Under the standard operating conditions of the cetane number tester, the diesel sample to be tested was mixed with different standard fuels with known cetane numbers;

[0035] The ignition properties of the mixture were determined under the same conditions;

[0036] The cetane number of the diesel sample to be tested was obtained by comparison.

[0037] Preferably, the method for obtaining the deep learning regression model includes: establishing a regression model using the h2o package in R software.

[0038] Preferably, after obtaining the diesel cetane number prediction model, the fitting coefficient R is used. 2 Measuring the predictive power of the diesel cetane number prediction model; specifically including:

[0039] When R 2 When the value is ≤0.3, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is poor.

[0040] When 0.3 < R 2 When the value is ≤0.4, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is weak.

[0041] When 0.4 < R 2 When the value is ≤0.6, the predicted value fits the observed value moderately, and the prediction ability of the prediction model is moderate.

[0042] When 0.6 < R 2 When the value is ≤1.0, the predicted value fits the observed value well, and the prediction model has strong predictive ability.

[0043] Preferably, the physicochemical index data include: density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20°C, cloud point, and total contaminant content.

[0044] A fourth objective of the present invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting the cetane number of diesel fuel.

[0045] A fifth objective of the present invention is to provide a computer program product comprising a computer program that, when executed by a processor, implements the aforementioned method for predicting the cetane number of diesel fuel.

[0046] The advantages and positive effects of this application are:

[0047] The core of this invention lies in pre-acquiring physicochemical index data of diesel samples and constructing a dataset. This dataset is then used for an automated machine learning process to build a predictive model capable of predicting the cetane number of the diesel sample to be tested. Specifically, this invention selects common physicochemical indicators from diesel sample testing as explanatory variables. These indicators include, but are not limited to, density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity (measured at 20°C), cloud point, and total contaminant content. These indicators serve as input variables, while the cetane number is the target variable. By employing a deep learning algorithm, this invention successfully establishes an efficient and accurate predictive model.

[0048] A series of verification experiments showed that the prediction model constructed using the method provided in this invention has a strong fitting effect. This means that the cetane number can be accurately predicted simply by analyzing the physicochemical indicators of diesel samples. This method not only effectively solves the complexity and inability to monitor in real time in traditional indicator testing processes, thus saving significant time and human resources and avoiding cumbersome testing procedures, but also allows for in-depth analysis of the intrinsic response relationship between the physicochemical indicators of diesel and the cetane number based on interpretable machine learning methods. In this way, researchers and engineers can better understand the performance characteristics of diesel and thus optimize the production and use of diesel. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart of the training method provided by a preferred embodiment of the present invention is shown;

[0051] Figure 2 A flowchart illustrating the construction of the training dataset in a preferred embodiment of the present invention is shown;

[0052] Figure 3 A flowchart illustrating the process of obtaining the cetane number of each diesel sample in a preferred embodiment of the present invention is shown.

[0053] Figure 4 A flowchart of the prediction method provided by a preferred embodiment of the present invention is shown;

[0054] Figure 5The neural network diagram used in a preferred embodiment of the present invention is shown;

[0055] Figure 6 A graph showing the fitting relationship between the detected value and the predicted value in a preferred embodiment of the present invention is shown.

[0056] Figure 7 The diagram illustrates a partial dependency relationship between the target variable and each explanatory variable in the prediction model of a preferred embodiment of the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figure 1 ;

[0059] The first embodiment provides a training method for a diesel cetane number prediction model, comprising the following steps:

[0060] Obtaining the training dataset is crucial, as it forms the core foundation for model construction and significantly impacts the model's output. To construct an ideal predictive model, the training dataset used in this embodiment primarily comprises explanatory variables and their corresponding target variables. The explanatory variables cover the physicochemical properties of different diesel samples, while the target variables include the cetane number of the corresponding diesel samples.

[0061] Based on the provided training dataset, a deep learning regression model is trained to obtain a model for predicting the cetane number of diesel fuel.

[0062] To gain a deeper understanding of the technical solution of this invention, non-limiting examples are given below:

[0063] Please see Figure 2 In the process of obtaining the training dataset, diesel samples must first be collected and tested to obtain the physicochemical properties of the samples and the corresponding cetane number; then, based on these physicochemical properties and the corresponding cetane number, a training dataset is constructed.

[0064] Please see Figure 3 The steps for obtaining the cetane number of each diesel sample include:

[0065] First, under the standard operating conditions of the cetane number tester, the diesel sample to be tested is mixed with different standard fuels with known cetane numbers;

[0066] Subsequently, the ignition properties of the mixture were determined under the same conditions;

[0067] Finally, the cetane number of the diesel sample to be tested was obtained through comparative analysis.

[0068] Please see Figure 5 In this embodiment, the method for obtaining the deep learning regression model includes the following steps: constructing the required regression model using the h2o package in R software. Specifically, R software is a widely used statistical analysis and graphical representation tool that provides a rich set of packages and functions to handle various data analysis tasks. The h2o package is a package specifically designed for big data analysis and automated machine learning, supporting a variety of machine learning algorithms, including deep learning regression models.

[0069] Using the h2o package, users can easily train deep learning regression models. First, users need to install and load the h2o package, and then import the dataset to be processed. Next, users can leverage automated machine learning to define and optimize the model and its parameters, such as the number and size of hidden layers, activation functions, loss functions, and optimization algorithms. After defining the model and its parameters, users can use the functions provided by the h2o package to train the model and evaluate its performance through methods such as cross-validation.

[0070] Ultimately, in this way, users can successfully obtain a deep learning regression model that can be used to predict and analyze various complex data relationships. This method is not only concise and efficient, but also leverages the power of R software and the h2o package, making the construction of deep learning regression models easier and more reliable.

[0071] After obtaining the diesel cetane number prediction model, the fitting coefficient R is used... 2 To evaluate the predictive performance of this model, the specific evaluation criteria are as follows:

[0072] If R 2 If the value is ≤0.3, it indicates that the fitting effect between the predicted value and the actual observed value is poor, and the prediction efficiency of the prediction model is low.

[0073] If 0.3 < R 2 If the value is ≤0.4, it indicates that the fit between the predicted value and the actual observed value is weak, and the prediction efficiency of the prediction model is also weak.

[0074] If 0.4 < R 2 If the value is ≤0.6, it indicates that the fit between the predicted value and the actual observed value is at a moderate level, and the predictive effectiveness of the prediction model is moderate.

[0075] If 0.6 < R 2A value ≤1.0 indicates a strong fit between the predicted and observed values, meaning the prediction model has high predictive efficiency.

[0076] The physicochemical data include: density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20°C, cloud point, and total contaminant content.

[0077] Second embodiment, please refer to Figure 4 A method for predicting the cetane number of diesel fuel, comprising:

[0078] Obtain the physicochemical index data of the diesel sample to be tested;

[0079] The physicochemical index data of the diesel sample to be tested are input into the diesel cetane number prediction model of the first embodiment to obtain the cetane number of the diesel sample to be tested.

[0080] A third embodiment provides a training device for a diesel cetane number prediction model, comprising:

[0081] The data acquisition module, particularly the training dataset, forms the core foundation for model construction, and its selection has a decisive impact on the model's output. To construct an ideal prediction model, the training dataset used in this embodiment primarily includes explanatory variables and their corresponding target variables. The explanatory variables cover the physicochemical index data of different diesel samples, while the target variables include the cetane number of the corresponding diesel samples.

[0082] The model training module trains the deep learning regression model based on the provided training dataset to obtain a model for predicting the cetane number of diesel fuel.

[0083] To gain a deeper understanding of the technical solution of this invention, non-limiting examples are given below:

[0084] To construct the training dataset, the following steps were performed: First, diesel samples were collected and their physicochemical properties were tested to obtain physicochemical index data; second, the cetane number of each sample was recorded. Based on these physicochemical index data and corresponding cetane numbers, a training dataset for subsequent analysis was constructed.

[0085] The steps for obtaining the cetane number of each diesel sample are as follows:

[0086] First, under the standard operating conditions of the cetane number tester, the diesel sample to be tested is mixed with a series of standard fuels with known cetane numbers.

[0087] Subsequently, the ignition characteristics of the mixture were analyzed under the same experimental conditions;

[0088] Finally, through comparative analysis, the cetane number of the diesel sample to be tested was determined.

[0089] In this embodiment, the method for obtaining the deep learning regression model involves using the h2o package in R software to construct the regression model. Specifically, the h2o package needs to be installed and loaded in the R environment first. After installation, a deep learning regression model can be created by calling the relevant functions in the h2o package. This model can be used to handle various regression problems, such as predicting continuous numerical results. By setting appropriate parameters, such as the number of layers, the number of neurons, and the activation function, the model's performance can be further optimized. Ultimately, this deep learning regression model built using the h2o package can be used to analyze and predict target variables in a dataset.

[0090] After obtaining the diesel cetane number prediction model, the fitting coefficient R is used. 2 Measuring the predictive power of the diesel cetane number prediction model; specifically including:

[0091] When R 2 When the value is ≤0.3, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is poor.

[0092] When 0.3 < R 2 When the value is ≤0.4, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is weak.

[0093] When 0.4 < R 2 When the value is ≤0.6, the predicted value fits the observed value moderately, and the prediction ability of the prediction model is moderate.

[0094] When 0.6 < R 2 When the value is ≤1.0, the predicted value fits the observed value well, and the prediction model has strong predictive ability.

[0095] The physicochemical data include: density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20°C, cloud point, and total contaminant content.

[0096] Fourth embodiment: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting the cetane number of diesel fuel.

[0097] Fifth embodiment, a computer program product, includes a computer program that, when executed by a processor, implements the above-described method for predicting the cetane number of diesel fuel.

[0098] The following section uses the physicochemical index data of the above seventeen parameters as examples for detailed explanation:

[0099] Diesel fuel samples were taken and tested to obtain physicochemical data and cetane number; specifically:

[0100] First, a number of diesel samples were obtained, and their physicochemical properties were acquired. The cetane index was then calculated using these physicochemical properties. Next, the cetane number was calculated using a specialized measuring machine. The physicochemical properties included density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity (20°C), cloud point, and total contaminant content.

[0101] The basis for selecting the above 17 physicochemical indicators in this embodiment includes:

[0102] Relationship between cetane number and composition: Among various diesel fuel hydrocarbon components with the same number of carbon atoms, n-alkanes have the highest cetane number; polycyclic aromatic hydrocarbons have the lowest cetane number; alkenes and cycloalkanes fall between alkanes and aromatic hydrocarbons. The higher the degree of isomerism of a hydrocarbon and the more rings it has, the lower its cetane number; for cycloalkanes and aromatic hydrocarbons, the cetane number increases with the increase of side chain length; while the cetane number decreases with the increase of side chain branching.

[0103] The relationship between cetane index and cetane number: The cetane number and cetane index of diesel fractions cut from intermediate-based crude oil are relatively consistent; the cetane number of diesel fractions cut from naphthenic crude oil is mostly lower than the cetane index; the cetane number of diesel fractions cut from paraffinic crude oil is mostly higher than the cetane index. The correlation between cetane number and cetane index is best for straight-run diesel, followed by hydrorefined diesel, and worst for hydrocracking diesel.

[0104] Relationship between cloud point, pour point, cold filter plugging point (CPP) and cetane number: Cloud point, pour point, and CPP are the main indicators for evaluating the low-temperature fluidity of diesel fuel. The more cycloalkanes and aromatics a diesel fuel contains, the better its low-temperature fluidity, and the lower the test results for indicators such as cloud point, pour point, and CPP. Conversely, the more n-alkanes a diesel fuel contains, the worse its low-temperature fluidity. Therefore, without the addition of low-temperature fluidity improvers, the low-temperature fluidity indicators of diesel fuel should be negatively correlated with the cetane number.

[0105] Relationship between corrected wear scar diameter and cetane number: Studies show that the content of polar components in diesel fuel is inversely proportional to the content of saturated hydrocarbons; that is, the lower the content of polar components and the higher the content of saturated hydrocarbons, the worse the lubricity. This is manifested in the fact that the higher the content of n-alkane, the larger the measured value of the corrected wear scar diameter, indicating a positive correlation between the corrected wear scar diameter and the cetane number.

[0106] The relationship between density and cetane number: Diesel fuel with high polycyclic aromatic hydrocarbon and aromatic hydrocarbon content has a high density, while diesel fuel with high straight-chain alkane content has a low density. Therefore, a higher density results in a lower cetane number, and a lower density results in a relatively higher cetane number; density and cetane number are negatively correlated.

[0107] Relationship between kinematic viscosity and cetane number: The kinematic viscosity of diesel fuel is mainly affected by the molecular weight, chemical structure, and type of hydrocarbon molecules. Generally, hydrocarbons with larger molecular weights or more complex chemical structures lead to increased diesel viscosity. Among all hydrocarbons, alkanes have the lowest viscosity because their molecular structures are relatively simple and the intermolecular forces are relatively weak. In contrast, cycloalkanes and aromatic hydrocarbons have higher viscosity due to stronger intermolecular forces. In summary, as diesel fuel viscosity increases, the alkane content decreases, the cycloalkanes and aromatic hydrocarbons content increases, and the cetane number decreases.

[0108] The relationship between flash point and cetane number: Flash point is related to the chemical composition and distillate composition of oils. Generally, oils with a higher alkanes content have a higher flash point than oils with the same viscosity but a higher content of cycloalkanes and aromatics. On the other hand, the higher the boiling point, the heavier the fraction, and the larger the relative molecular mass of an oil, the higher its flash point. Oils with a larger relative molecular mass generally have a higher aromatic hydrocarbon content than alkane content, so flash point and cetane number mainly show a negative correlation.

[0109] Relationship between Total Contaminants, Clogging Tendency, and Cetane Number: Total contaminants in diesel fuel refer to insoluble substances in a diesel sample, mainly including dust and particulate matter. These contaminants are primarily caused by pollution during diesel production and transportation, as well as by oxidation. Clogging tendency in diesel fuel refers to its ability to easily clog pipes or filters under specific conditions. This characteristic is mainly related to impurities, moisture, and insoluble substances in the diesel fuel. These factors can cause diesel fuel to clog pipes or filters during flow, affecting its normal use. As the depth of diesel fuel hydrorefining increases, the content of various impurities in the fuel decreases, while the content of saturated hydrocarbons in the diesel fuel relatively increases, resulting in an increase in cetane number. The total contaminant content and clogging tendency test results decrease, therefore these two indicators show a negative correlation with cetane number.

[0110] The relationship between acidity and cetane number: As diesel fuel becomes more refined, the amount of polar substances in the fuel decreases, resulting in poorer lubricity. In such cases, acidic lubrication improvers are often added to enhance lubricity. This leads to an increase in diesel fuel acidity, causing the cetane number to gradually rise with increasing acidity.

[0111] Relationship between 50%, 90%, and 95% recovery temperatures and cetane number: The distillation range of diesel fuel refers to the temperature range obtained by distillation under specific conditions. Diesel fuels with a lighter distillation range have better starting properties, are easier to evaporate and burn rapidly, but excessively light fractions may lead to a high auto-ignition point and a long ignition delay, potentially causing knocking. The 50% recovery temperature characterizes the content of light components in diesel fuel. Within a certain temperature range, the concentration of the diesel-air mixture is moderate, making it easy to burn and resulting in a higher cetane number. The 90% and 95% recovery temperatures represent the content of heavy components in diesel fuel. Diesel fuels with heavier fractions evaporate more slowly, leading to incomplete combustion and a poorer cetane number, showing a primarily negative correlation.

[0112] Relationship between sulfur content and cetane number: Sulfides in diesel fuel are mainly benzothiophene and alkylbenzothiophene, and dibenzothiophene and alkyldibenzothiophene sulfides. Thiophene and its derivatives are mainly distributed in fractions below 200℃, benzothiophene and its derivatives are mainly distributed in fractions between 200 and 300℃, while in fractions above 300℃, the sulfides are mainly dibenzothiophene and its derivatives. Among them, 4,6-DMDBT is found in fractions between 330 and 345℃, and more complex dibenzothiophenes with substituents at the 4 and 6 positions are mainly concentrated in fractions above 350℃. The straight-run diesel fuel fraction has a temperature range of approximately 200–365℃, concentrating most of the sulfur content in diesel fuel, and its cetane number is also relatively high. Therefore, as the sulfur content increases, the cetane number gradually increases.

[0113] The relationship between chlorine content and cetane number: The chlorine content in diesel fuel can originate from multiple sources, including the crude oil itself, contaminants during refining, pollution during transportation, and the use of additives. Upgrading refining processes and strengthening management can minimize the impact of the first three factors. Therefore, the chlorine content in diesel fuel mainly comes from additives used in the refining process. Refining processes can increase the cetane number of diesel fuel, thus there is a positive correlation between chlorine content and cetane number.

[0114] Relationship between oxidation stability and cetane number: Different hydrocarbons in fuel have varying resistance to oxidation in the liquid phase. Aromatic hydrocarbons, alkanes, and cycloalkanes exhibit good stability, while unsaturated hydrocarbons readily react with oxygen in the air at room temperature, which is the main cause of fuel oxidation and deterioration. Besides unsaturated hydrocarbons, sulfur- or nitrogen-containing compounds can also cause fuel discoloration or deterioration, reducing fuel stability. Diesel straight-run components have a higher sulfide content, making them more prone to oxidation and deterioration. Therefore, provided the diesel oxidation stability meets product standard requirements, a higher straight-run component content results in a higher oxidation stability test result and a higher cetane number.

[0115] Construct a training dataset, using physicochemical index data as explanatory variables and cetane number as the target variable;

[0116] The dataset includes explanatory variables and target variables; wherein, the explanatory variables include physicochemical index data of different diesel samples, and the target variables include cetane numbers of different diesel samples;

[0117] The dataset of diesel samples is randomly divided into a training set and a test set. The prediction model is built using the training set, and the prediction ability of the prediction model is verified using the test set. For example, 90% of the dataset can be used as the training set and 10% of the dataset can be used as the test set.

[0118] Based on the training dataset, a deep learning regression model is trained to obtain a diesel cetane number prediction model.

[0119] For example, firstly, the regression model is established using the h2o package in R software, and then a deep learning regression model is trained based on the training dataset to obtain a diesel cetane number prediction model.

[0120] A diesel cetane number prediction model is used to predict the cetane number of the diesel fuel under test; specifically, it includes:

[0121] First, the physicochemical index data of the diesel sample to be tested are obtained. The following parameters are used as explanatory variables: density at 20℃, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity (20℃), cloud point, and total contaminant content. Then, the explanatory variable data are substituted into the diesel cetane number prediction model, and the target variable value is the predicted cetane number of the diesel sample to be tested.

[0122] For any experimental steps or conditions not specified in the following embodiments, the procedures or conditions can be performed according to the conventional experimental steps or conditions described in the literature in this field. All raw materials or instruments used are commercially available conventional products, including but not limited to those used in the embodiments of this application.

[0123] Implementation Cases and the Establishment of a Diesel Cetane Number Prediction Model

[0124] This embodiment provides a method for establishing a diesel cetane number prediction model, and the specific steps are as follows:

[0125] (1) Diesel sample collection

[0126] Diesel samples that are stored and await testing during the production and operation process are selected. In accordance with the product standard GB / T 19147-2016 and the internal control indicators of externally sourced diesel, the testing work includes all items of the national standard and additional indicators (including odor, nitrate ester cetane number improver, cloud point, clogging tendency, water separation performance, chlorine content, and silicon content).

[0127] (2) Testing of the physical and chemical properties of diesel fuel

[0128] The following physicochemical properties of diesel fuel were tested: density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20°C, cloud point, and total contaminant content.

[0129] The 17 diesel fuel physicochemical properties used as explanatory variables include: density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane number, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20°C, cloud point, and total contaminant content. The formula for calculating the cetane number index (CI) is as follows:

[0130]

[0131] In the formula: T 10N =T 10 -215;

[0132] T 50N =T 50 -260;

[0133] T 90N =T 90 -310;

[0134] T 10 The temperature at which the sample recovers to 10% is indicated, in °C.

[0135] T 50 The temperature at which 50% of the sample is recovered is indicated in °C.

[0136] T 90 The temperature at which 90% of the sample is recovered is indicated in °C.

[0137] B = [exp(-0.0035D N )]-1;

[0138] D N =D-850;

[0139] D represents the density of the sample at 15℃, in kg / m³. 3 ;

[0140] Table 1. Detection methods for diesel fuel physicochemical properties.

[0141]

[0142]

[0143] (3) Cetane number detection

[0144] Under specified operating conditions, the combustion characteristics of a test fuel are determined by comparing it with those of a standard fuel with a known cetane number. Specifically, the engine compression ratio is adjusted (represented by a handwheel reading) to determine the "ignition lag period" of the test fuel, which is the time interval (in crankshaft angles) between the start of fuel injection and the start of combustion. Based on the compression ratio of the test fuel, two standard fuels with cetane numbers differing by no more than 5.6 units are selected, and their "ignition lag period" is determined using the same method. When the compression ratio of the test fuel falls between the selected compression ratios of the two standard fuels, the cetane number of the test fuel is calculated using interpolation based on the handwheel reading.

[0145] (4) Use deep learning algorithms to build a prediction model

[0146] In this embodiment, the AutoML algorithm provided by the H2O package in R is used for data analysis. The implementation of H2OAutoML is based on the JDK 17 environment and written in Java. The `h2o.automl` function, belonging to the H2O package, is used to perform the selection of an automated machine learning algorithm. To avoid evaluation bias when determining the range of model parameters, a comprehensive search of all built-in models and their parameter combinations is performed.

[0147] The original data, imported in CSV format, forms a general data frame structure. This data is then converted to an H2O data frame and input into the H2OAutoML model. The data is split using the `ratio` function built into the H2O package to ensure consistency of model results and prevent data leakage. To evaluate the model's optimization performance, this embodiment uses the five-fold cross-validation method provided with the H2O package's training model.

[0148] H2OAutoML has the ability to automatically select the optimal algorithm, with options including deep learning, random forest (RF), generalized linear model (GLM), gradient boosting machine (GBM), and XGBoost.

[0149] Ultimately, the optimal algorithm selected by H2O AutoML was h2o.deeplearning (a deep learning algorithm). Subsequently, this algorithm was used to build a preliminary regression model, and the R-values ​​of the model on the test set were used as a benchmark. 2 The generalization performance is evaluated using the value and MSE metric.

[0150] This embodiment involves 7304 sample data points, selecting 17 physicochemical parameters of oil products as explanatory variables, and using cetane number as the predictor variable. By comparing the prediction performance when the dataset split ratio (i.e., the ratio of the number of samples in the training set to the number of samples in the test set) is 0.7, 0.8, 0.85, and 0.9, the optimal prediction performance was finally determined to be 0.9. The study used 30 random seeds for cyclical experiments, running a total of 30 times. Each deep learning model was developed based on the training set and applied to the test set to verify its generalization ability.

[0151] In this embodiment, the model is used to make predictions using the training set, and the mean squared error (MSE) and the coefficient of fit (R²) are employed. 2 MSE is used as an evaluation metric. It is typically used to measure the performance of prediction results on the same dimension of an indicator. When predicting the same indicator, a lower MSE value is generally considered to indicate better prediction performance. 2 The higher the value, the more R is used. 2 As a key parameter for evaluating model performance. According to R 2 Different ranges of values ​​can be used to classify the degree of fit between predicted and actual observed values ​​into several levels: R 2 A value less than 0.3 indicates a poor fit, 0.3 to 0.4 indicates a weak fit, 0.4 to 0.6 indicates a moderate fit, and 0.6 to 1.0 indicates a strong fit. Furthermore, the h2o.deeplearning algorithm in the H2O package and the dataset partitioning are implemented using the "h2o" package and the "h2o.deeplearning" function in R software (version 4.3.3), respectively.

[0152] Validation of the prediction model

[0153] (1) Model training effect based on training set

[0154] Based on the training set, 17 diesel physicochemical indicators (as explanatory variables) were used for combined prediction. The fitting effect between the predicted and actual observed values ​​of the cetane number is as follows: Figure 6 As shown in Table 2, a comparison of the predicted cetane number data and the actual detected values ​​for a portion of the samples in the training set is presented.

[0155] Table 2. Training performance of the cetane number prediction model based on the training set.

[0156]

[0157]

[0158] As shown in Table 2 and Figure 6 As shown, the prediction model's cetane number predictions in the training set are highly consistent with the actual detected values, with a mean squared error (MSE) of 0.019 and a coefficient of determination (R²).2 The value reached 0.990, indicating a strong fit.

[0159] (2) Model prediction performance based on the test set

[0160] The prediction model obtained from the training set is applied to the test set to test the model's generalization performance, i.e., its predictive ability on new datasets, using mean squared error (MSE) and coefficient of determination (R²). 2 () is used as an evaluation indicator. The results of the prediction of the test set using the cetane number prediction model are presented as follows: Figure 6 The black scattered dots in the table show the predicted and measured cetane number data for some samples in the test set.

[0161] Table 3. Test results of the cetane number prediction model based on the test set.

[0162]

[0163]

[0164] As shown in Table 3 and Figure 6 As shown, the prediction model still achieved a high degree of fit for the cetane number prediction on the new dataset (test set), with a correlation coefficient R0. 2 The value reached 0.992, while the mean squared error (MSE) was 0.019.

[0165] In the cetane number prediction model, the partial dependency diagram between the target variable and each explanatory variable is shown below. Figure 7 .

[0166] Figure 7 The order of the variables reflects their importance, thus identifying the cetane index as a key feature for predicting the cetane number. A one-dimensional partial dependence plot between the cetane index and cetane number reveals a linear relationship, meaning that an increase in the cetane index is accompanied by a gradual increase in the cetane number. Furthermore, variables such as cloud point and sulfur content also show a positive correlation with the cetane number. However, variables such as the 90% recovery temperature and cloud point show a negative correlation with the cetane number. Additionally, variables such as the 95% recovery temperature and freezing point exhibit a non-linear relationship with the cetane number.

[0167] Therefore, the prediction model constructed by the method provided in this invention has excellent generalization ability and can accurately predict the cetane number of unknown diesel samples. The prediction model constructed in this invention uses a wide range of data, covering diesel of different qualities, and thus can be widely applied to the accurate prediction of diesel cetane number.

[0168] Cetane number prediction

[0169] Based on the previously constructed prediction model (which is based on training set data), the process of predicting the cetane number of the diesel sample to be tested is as follows: First, the physicochemical index data of the diesel sample to be tested are collected. These physicochemical indexes include density at 20°C, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20°C, cloud point, and total contaminant content. Subsequently, these physicochemical index data are input into the cetane number prediction model, and the target variable value calculated by the model is the predicted cetane number of the diesel sample to be tested.

[0170] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0171] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A training method for a diesel cetane number prediction model, characterized in that, include: A training dataset is obtained, comprising explanatory variables and corresponding target variables. The explanatory variables include physicochemical index data of different diesel samples, including: density at 20℃, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20℃, cloud point, and total contaminant content. The target variable includes the cetane number of different diesel samples. The formula for calculating the cetane number index (CI) is as follows: ； In the formula: ; ； ； The temperature at which the sample recovers 10% is indicated, in °C. The temperature at which 50% of the sample is recovered is indicated in °C. The temperature at which 90% of the sample is recovered is indicated in °C. ； ； D represents the density of the sample at 15℃, in kg / m³. 3 ; Based on the training dataset, a deep learning regression model is trained, and its generalization performance is evaluated based on the R² value and MSE index of the deep learning regression model on the test set, thus obtaining a diesel cetane number prediction model.

2. The training method for the diesel cetane number prediction model according to claim 1, characterized in that, Obtaining the training dataset includes: Diesel fuel samples were taken and tested to obtain physicochemical data of the diesel fuel samples and the cetane number of each diesel fuel sample. A training dataset was constructed using the physicochemical index data of diesel samples and the cetane number corresponding to each diesel sample.

3. The training method for the diesel cetane number prediction model according to claim 2, characterized in that, The steps for obtaining the cetane number of each diesel sample include: Under the standard operating conditions of the cetane number tester, the diesel sample to be tested was mixed with different standard fuels with known cetane numbers; The ignition properties of the mixture were determined under the same conditions; The cetane number of the diesel sample to be tested was obtained by comparison.

4. The training method for the diesel cetane number prediction model according to claim 1, characterized in that, The method for obtaining the deep learning regression model includes: establishing a regression model using the h2o package in R software.

5. The training method for the diesel cetane number prediction model according to claim 1, characterized in that, After obtaining the diesel cetane number prediction model, the fitting coefficient R is used. 2 Measuring the predictive power of the diesel cetane number prediction model; specifically including: When R 2 When the value is ≤0.3, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is poor. When 0.3 < R 2 When the value is ≤0.4, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is weak. When 0.4 < R 2 When the value is ≤0.6, the predicted value fits the observed value moderately, and the prediction ability of the prediction model is moderate. When 0.6 < R 2 When the value is ≤1.0, the predicted value fits the observed value well, and the prediction model has strong predictive ability.

6. A method for predicting the cetane number of diesel fuel, characterized in that, include: Obtain the physicochemical index data of the diesel sample to be tested; The physicochemical index data of the diesel sample to be tested are input into the diesel cetane number prediction model according to any one of claims 1 to 5 to obtain the cetane number of the diesel sample to be tested.

7. A training device for a diesel cetane number prediction model, characterized in that, include: The data acquisition module acquires a training dataset, which includes explanatory variables and corresponding target variables. The explanatory variables include physicochemical index data for different diesel samples, including: density at 20℃, 50% recovery temperature, 90% recovery temperature, 95% recovery temperature, clogging tendency, cold filter plugging point, sulfur content, chlorine content, pour point, flash point, cetane index, acidity, corrected wear track diameter, oxidation stability, kinematic viscosity at 20℃, cloud point, and total contaminant content. The target variable includes the cetane number (CI) of different diesel samples. The formula for calculating the cetane number index (CI) is as follows: ； T 10 The temperature at which the sample recovers 10% is indicated, in °C. T 50 The temperature at which 50% of the sample is recovered is indicated in °C. T 90 The temperature at which 90% of the sample is recovered is indicated in °C. B=[exp(-0.0035D N )]-1; D N =D-850; D represents the density of the sample at 15℃, in kg / m³. 3 ; The model training module trains a deep learning regression model based on the training dataset, and evaluates its generalization performance based on the R² value and MSE index of the deep learning regression model on the test set, thereby obtaining a diesel cetane number prediction model.

8. The training apparatus for the diesel cetane number prediction model according to claim 7, characterized in that, Obtaining the training dataset includes: Diesel fuel samples were taken and tested to obtain physicochemical data of the diesel fuel samples and the cetane number of each diesel fuel sample. A training dataset was constructed using the physicochemical index data of diesel samples and the cetane number corresponding to each diesel sample; The steps for obtaining the cetane number of each diesel sample include: Under the standard operating conditions of the cetane number tester, the diesel sample to be tested was mixed with different standard fuels with known cetane numbers; The ignition properties of the mixture were determined under the same conditions; The cetane number of the diesel sample to be tested was obtained by comparison.

9. The training apparatus for the diesel cetane number prediction model according to claim 7, characterized in that, The method for obtaining the deep learning regression model includes: establishing a regression model using the h2o package in R software.

10. The training apparatus for the diesel cetane number prediction model according to claim 7, characterized in that, After obtaining the diesel cetane number prediction model, the fitting coefficient R is used. 2 Measuring the predictive power of the diesel cetane number prediction model; specifically including: When R 2 When the value is ≤0.3, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is poor. When 0.3 < R 2 When the value is ≤0.4, the predicted value fits the observed value poorly, and the prediction ability of the prediction model is weak. When 0.4 < R 2 When the value is ≤0.6, the predicted value fits the observed value moderately, and the prediction ability of the prediction model is moderate. When 0.6 < R 2 When the value is ≤1.0, the predicted value fits the observed value well, and the prediction model has strong predictive ability.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method for predicting the cetane number of diesel fuel as described in claim 6.

12. A computer-readable storage medium storing a computer program, characterized in that, When executed by the processor, the program implements the method for predicting the cetane number of diesel fuel as described in claim 6.

Images