Product property analysis method and device for residual oil hydrogenation reaction

Abstract

The invention discloses a product property analysis method and device for residual oil hydrogenation reaction, and relates to the technical field of heavy oil hydrogenation and artificial intelligence, and the method comprises the following steps: obtaining high-flux experimental data of the residual oil hydrogenation reaction and historical data of a specified residual oil hydrogenation reaction device; training the plurality of different types of initial neural network models by using the high-throughput experimental data to obtain a plurality of product property analysis models; determining the product property analysis model with the minimum error as a specified product property analysis model; optimizing parameters of the specified product property analysis model by utilizing historical data of the specified residual oil hydrogenation reaction device to obtain a product property analysis model corresponding to the specified residual oil hydrogenation reaction device; and inputting the to-be-analyzed reaction data into a product property analysis model corresponding to the specified residual oil hydrogenation reaction device, and determining product property data. According to the method, the problems of narrow application range and low analysis precision of the established analysis model in the prior art can be solved.

Description

Technical Field

[0001] The invention relates to the fields of heavy oil hydrogenation and artificial intelligence technology, and in particular to a method and device for analyzing properties of products of residual oil hydrogenation reaction. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No admission is made that the description herein is prior art by inclusion in this section.

[0003] At present, with the increasing heaviness and inferiority of oil resources, as well as the urgent requirements for upgrading the quality of oil products and the increasingly stringent environmental protection regulations, refining enterprises are in urgent need of clean and efficient processing technology for heavy oil. Residue oil hydrogenation is the most effective processing technology for heavy and residual oil raw materials. Through residue oil hydrogenation, most of the metals, sulfur and nitrogen in the residue oil are removed, and the residual carbon value is reduced, which is conducive to its further efficient and clean processing and conversion. The technical combination of fixed-bed residue oil hydrogenation treatment and residue oil catalytic cracking can not only maximize the conversion of residual oil with low utilization value and easy to cause environmental pollution, greatly improve the yield of light oil, but also obtain clean oil products with high added value and high quality. In a sense, the crude oil has been converted 100%, realizing the desire to use up the crude oil in the oil refining process.

[0004] At present, machine learning methods are used to establish analytical models for residue oil hydrogenation processes to analyze product properties and optimize reaction conditions. Machine learning methods are to deeply mine the rules and relationships hidden in the existing residue oil hydrogenation reaction data, and then establish a relationship model between residue oil hydrogenation reaction conditions and reaction results. The effect of the established model is very dependent on the quality of the data set. However, the existing analytical methods usually only use data from specific residue oil hydrogenation reaction units in the modeling process. In order to maintain stable operation, the raw material properties and operating conditions of the residue oil hydrogenation reaction unit usually do not change much, and the catalyst loading and grading ratio are fixed. The universality of the model established by the data of a specific unit is restricted. If it is necessary to establish an analytical model for different residue oil hydrogenation reaction units, it is necessary to use the data of the new unit to start from scratch to establish the analytical model, which is labor-intensive and inflexible. Therefore, the existing analytical models have the problems of narrow adaptability and low analytical accuracy. Summary of the invention

[0005] The embodiment of the present invention provides a method for analyzing the properties of products of a residue oil hydrogenation reaction, which is used to solve the problems of narrow adaptability and low analysis accuracy of the established analysis model in the prior art. The method comprises:

[0006] Obtaining high-throughput experimental data of residue oil hydrogenation reaction and historical data of a designated residue oil hydrogenation reaction unit; the high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data, and product property data;

[0007] Constructing a first sample data set based on high-throughput experimental data of residue oil hydrogenation reaction;

[0008] constructing a second sample data set according to historical data of a designated residue oil hydroprocessing reaction unit;

[0009] Using the first sample data set to train a plurality of preset initial neural network models of different types, respectively, to obtain a plurality of residual oil hydrogenation reaction product property analysis models;

[0010] Calculating the errors of multiple residue oil hydrogenation reaction product property analysis models, and determining the residue oil hydrogenation reaction product property analysis model with the smallest error from the multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model;

[0011] Optimizing the parameters of the specified residue oil hydrogenation reaction product property analysis model using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit;

[0012] The reaction data to be analyzed of the designated residue oil hydrogenation reaction unit are input into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed.

[0013] The embodiment of the present invention further provides a product property analysis device for residual oil hydrogenation reaction, which is used to solve the problems of narrow application range and low analysis accuracy of the established analysis model in the prior art. The device comprises:

[0014] An acquisition module is used to acquire high-throughput experimental data of residue oil hydrogenation reaction and historical data of a designated residue oil hydrogenation reaction unit; the high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data and product property data;

[0015] A first sample construction module is used to construct a first sample data set according to high-throughput experimental data of residual oil hydrogenation reaction;

[0016] A second sample construction module is used to construct a second sample data set according to historical data of a designated residue oil hydrogenation reaction unit;

[0017] A model training module, used to train a plurality of preset initial neural network models of different types using the first sample data set to obtain a plurality of residual oil hydrogenation reaction product property analysis models;

[0018] A model screening module is used to calculate the errors of multiple residue oil hydrogenation reaction product property analysis models, and determine the residue oil hydrogenation reaction product property analysis model with the smallest error from the multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model;

[0019] A model optimization module, used to optimize the parameters of the specified residue oil hydrogenation reaction product property analysis model using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit;

[0020] The analysis module is used to input the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed.

[0021] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned product property analysis method of the residue oil hydrogenation reaction when executing the computer program.

[0022] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the product property analysis method of the residue oil hydrogenation reaction is implemented.

[0023] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the method for analyzing the properties of products of the residual oil hydrogenation reaction is implemented.

[0024] In the embodiment of the present invention, high-throughput experimental data of residue oil hydrogenation reaction and historical data of a designated residue oil hydrogenation reaction device are obtained; the high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data and product property data; a first sample data set is constructed based on the high-throughput experimental data of residue oil hydrogenation reaction; a second sample data set is constructed based on the historical data of the designated residue oil hydrogenation reaction device; a plurality of different types of preset initial neural network models are trained respectively using the first sample data set to obtain a plurality of residue oil hydrogenation reaction product property analysis models; a plurality of residue oil hydrogenation reaction product property analysis models are calculated; The error of the residue oil hydrogenation reaction product property analysis model is determined, and the residue oil hydrogenation reaction product property analysis model with the smallest error is determined from multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model; the parameters of the designated residue oil hydrogenation reaction product property analysis model are optimized using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit; the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit is input into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed. Compared with the existing technical solutions for analyzing the properties of products of residue oil hydrogenation reaction, firstly, multiple different types of neural network models are trained by high-throughput experimental data of residue oil hydrogenation reaction to obtain multiple residue oil hydrogenation reaction product property analysis models, and the residue oil hydrogenation reaction product property analysis model with the smallest error is screened out, so as to establish a basic model with a wide range of adaptability. Then, the parameters of the basic model are adaptively optimized by using the historical data of the specified residue oil hydrogenation reaction unit to obtain the residue oil hydrogenation reaction product property analysis model of the specified residue oil hydrogenation reaction unit. In this way, the product property analysis model of the residue oil hydrogenation reaction unit is used to analyze the product property data of the reaction to be analyzed of the specified residue oil hydrogenation reaction unit, so as to reduce the prediction error of the model on the specified residue oil hydrogenation reaction unit. The method of the embodiment of the present invention is helpful to expand the scope of application of the model and improve the prediction accuracy of the model. At the same time, the demand for the data of the specified residue oil hydrogenation reaction unit when building the model is reduced, the difficulty of modeling is reduced, and the work efficiency is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0026] Figure 1This is a flow chart of a method for analyzing properties of products of a residual oil hydrogenation reaction provided in an embodiment of the present invention;

[0027] Figure 2 A flow chart of a method for constructing a first sample data set based on high-throughput experimental data of residual oil hydrogenation reaction provided in an embodiment of the present invention;

[0028] Figure 3 A schematic diagram of the structure of an initial neural network model provided in an embodiment of the present invention;

[0029] Figure 4 A flow chart of a method for optimizing the parameters of a specified residue oil hydrogenation reaction product property analysis model using a second sample data set to obtain a residue oil hydrogenation reaction product property analysis model corresponding to a specified residue oil hydrogenation reaction unit provided in an embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of a product property analysis device for a residual oil hydrogenation reaction provided in an embodiment of the present invention;

[0031] Figure 6 A schematic diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0033] In the description of this specification, the terms "include", "including", "have", "contain", etc. are all open terms, which mean including but not limited to. The descriptions with reference to the terms "one embodiment", "a specific embodiment", "some embodiments", "for example", etc. mean that the specific features, structures or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. The order of steps involved in each embodiment is used to schematically illustrate the implementation of the present application, and the order of steps therein is not limited and can be appropriately adjusted as needed.

[0034] Research has found that with the development of machine learning algorithms, there are already considerations of using machine learning methods to establish a residual oil hydrogenation process analysis model for product property analysis, reaction condition optimization, etc. For example, using data from a small residual oil hydrogenation unit, artificial neural networks were used to simulate and analyze the residual oil hydrogenation process, and a product property prediction model was established. For another example, a new hydrogen flow prediction model was established using long short-term memory neural networks and gated recurrent unit neural networks to achieve stable and accurate prediction of the new hydrogen flow of the residual oil hydrogenation unit.

[0035] However, the above analysis methods usually only use data from a specific residue oil hydroprocessing reaction unit in the modeling process. In order to maintain stable operation, the raw material properties and operating conditions of the residue oil hydroprocessing reaction unit usually do not change much, and the catalyst loading and grading ratio are fixed. The universality of the model established by the data of a specific unit is restricted. If it is necessary to establish an analysis model for different residue oil hydroprocessing reaction units, it is necessary to use the data of the new unit to start from scratch to establish the analysis model, which is labor-intensive and inflexible. Therefore, the existing analysis model has the problems of narrow adaptability and low analysis accuracy.

[0036] In view of the deficiencies of the prior art, the embodiments of the present invention provide a method and device for analyzing the properties of products of a residue oil hydrogenation reaction, which can solve the problems of narrow adaptability and low analysis accuracy of the established analysis model in the prior art.

[0037] Figure 1 A flow chart of a method for analyzing properties of products of a residual oil hydrogenation reaction provided in an embodiment of the present invention, the method comprising the following steps:

[0038] Step 101, obtaining high-throughput experimental data of residue oil hydrogenation reaction and historical data of a designated residue oil hydrogenation reaction unit; the high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data and product property data;

[0039] Step 102, constructing a first sample data set according to high-throughput experimental data of residual oil hydrogenation reaction;

[0040] Step 103, constructing a second sample data set according to the historical data of the designated residue oil hydrogenation reaction unit;

[0041] Step 104, using the first sample data set to train a plurality of different types of preset initial neural network models respectively, to obtain a plurality of residual oil hydrogenation reaction product property analysis models;

[0042] Step 105, calculating the errors of a plurality of residue oil hydrogenation reaction product property analysis models, and determining a residue oil hydrogenation reaction product property analysis model with the smallest error from the plurality of residue oil hydrogenation reaction product property analysis models as a designated residue oil hydrogenation reaction product property analysis model;

[0043] Step 106, optimizing the parameters of the specified residue oil hydrogenation reaction product property analysis model using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit;

[0044] Step 107, inputting the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit, and determining the product property data of the reaction data to be analyzed.

[0045] In the embodiment of the present invention, high-throughput experimental data of residue oil hydrogenation reaction and historical data of a designated residue oil hydrogenation reaction device are obtained; the high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data and product property data; a first sample data set is constructed based on the high-throughput experimental data of residue oil hydrogenation reaction; a second sample data set is constructed based on the historical data of the designated residue oil hydrogenation reaction device; a plurality of different types of preset initial neural network models are trained respectively using the first sample data set to obtain a plurality of residue oil hydrogenation reaction product property analysis models; a plurality of residue oil hydrogenation reaction product property analysis models are calculated; The error of the residue oil hydrogenation reaction product property analysis model is determined, and the residue oil hydrogenation reaction product property analysis model with the smallest error is determined from multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model; the parameters of the designated residue oil hydrogenation reaction product property analysis model are optimized using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit; the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit is input into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed. Compared with the existing technical solutions for analyzing the properties of products of residue oil hydrogenation reaction, firstly, multiple different types of neural network models are trained by high-throughput experimental data of residue oil hydrogenation reaction to obtain multiple residue oil hydrogenation reaction product property analysis models, and the residue oil hydrogenation reaction product property analysis model with the smallest error is screened out, so as to establish a basic model with a wide range of adaptability. Then, the parameters of the basic model are adaptively optimized by using the historical data of the specified residue oil hydrogenation reaction unit to obtain the residue oil hydrogenation reaction product property analysis model of the specified residue oil hydrogenation reaction unit. In this way, the product property analysis model of the residue oil hydrogenation reaction unit is used to analyze the product property data of the reaction to be analyzed of the specified residue oil hydrogenation reaction unit, so as to reduce the prediction error of the model on the specified residue oil hydrogenation reaction unit. The method of the embodiment of the present invention is helpful to expand the scope of application of the model and improve the prediction accuracy of the model. At the same time, the demand for the data of the specified residue oil hydrogenation reaction unit when building the model is reduced, the difficulty of modeling is reduced, and the work efficiency is greatly improved.

[0046] Combine the following Figure 1 The method for analyzing the properties of the products of the residue oil hydrogenation reaction is described in detail.

[0047] In the above step 101, high-throughput experimental data of residue oil hydrogenation reaction and historical data of a designated residue oil hydrogenation reaction unit may be obtained.

[0048] The high-throughput experimental data and the historical data may include: crude oil property data, catalyst property data, reaction condition data and product property data.

[0049] In one embodiment, the crude oil property data may include: specific gravity, total nitrogen content, total sulfur content, residual carbon content, viscosity, carbon and hydrogen content (C / H content), and metal content of the crude oil.

[0050] In one embodiment, the catalyst property data may include: constituent elements of the catalyst and the mass fraction of each constituent element, bulk density, specific surface area, pore size, pore volume and active metal content.

[0051] In one embodiment, the above reaction condition data may include: feedstock oil space velocity, hydrogen pressure, temperature, ratio of hydrogen to oil, and catalyst grading ratio.

[0052] In one embodiment, the product property data may include: density, viscosity, nitrogen content, sulfur content, carbon residue value, carbon and hydrogen content and metal impurity content of the hydrogenated product.

[0053] In specific implementation, high-throughput experimental data can be collected through high-throughput experiments of residual oil hydrogenation reactions. Specifically, high-throughput experimental data can be obtained through the following steps:

[0054] Step 1.1: prepare m types (m>=3) of residual oil hydrogenation catalysts and n types (n>=3) of residual oil raw materials, measure the properties of each catalyst and the raw oil properties and save the measurement results. Wherein, the residual oil hydrogenation catalyst comprises at least one demetallizing agent, one desulfurizing agent, and one decarbonizing or denitrifying catalyst, the catalyst properties measured above include the constituent elements of the catalyst and their mass fraction, bulk density, specific surface area, pore size, pore volume and active metal content, and the residual oil properties measured above include density, viscosity, average molecular weight, nitrogen content, sulfur content, residual carbon value, carbon and hydrogen content (C / H content) and metal impurity (Ni, V, Fe, Ca, etc.) content.

[0055] Step 1.2: Setting high-throughput reaction conditions, wherein the reaction conditions include feed space velocity of raw oil, hydrogen pressure, temperature, ratio of hydrogen to oil (hydrogen / oil ratio) and catalyst grading ratio. For example, the feed space velocity of raw material can be selected in the range of 0.1 to 1.0 h -1 The optional pressure range is 8-22MPa, the optional temperature range is 330-420℃, the optional hydrogen / oil ratio range is 300-1500, the optional demetallization catalyst grading ratio range is 20-70%, the optional desulfurization catalyst grading ratio range is 20-60%, and the optional carbon removal or denitrification catalyst grading ratio range is 10-60%.

[0056] Step 1.3: Perform residual oil hydrogenation reaction analysis. Select a residual oil feedstock and a residual oil hydrogenation catalyst combination prepared in step 1.1, select a set of reaction conditions (feedstock oil feed space velocity, hydrogen pressure, hydrogen / oil ratio and catalyst grading ratio parameters) within the range set in step 1.2, load, dry, sulfide the catalyst and switch the residual oil feedstock according to conventional methods, and set the initial reaction temperature to 340°C. During the reaction, keep other process conditions except the reaction temperature unchanged, regularly analyze the density, viscosity, nitrogen content, sulfur content, residual carbon value and metal impurity (Ni, V, Fe, Ca, etc.) content of the hydrogenation product, and calculate the cumulative deposition of metal Ni and V per unit volume of catalyst based on the comparison of the content of metal Ni and V in the feedstock oil and the hydrogenation product. For every 0.01g / mL increase in the amount of metal deposited per unit volume of the catalyst, the reaction temperature is increased by 5°C. At the same time, the properties of the hydrogenation products when the reaction temperature increases by 3°C, increases by 6°C, decreases by 3°C, and decreases by 6°C are investigated respectively. The reaction temperature and analysis results are recorded. The analysis results include the density, viscosity, nitrogen content, sulfur content, residual carbon value, and metal impurity (Ni, V, Fe, Ca, etc.) content of the hydrogenation products; the cumulative deposition of Ni and V per unit volume of catalyst metals is calculated based on the analysis results. The reaction temperature is continuously increased with the increase in the amount of catalyst metal deposited. When the reaction temperature reaches 420°C and the properties of the hydrogenation products are investigated, the experiment is terminated to obtain the product property data:

[0057] Step 1.4: Reset the reaction conditions, and then repeat step 1.3 until the collected experimental data meets the preset requirements, which include the number of experimental data, the coverage of operating conditions included in the experimental data, and the number of groups.

[0058] In step 102, a first sample data set may be constructed based on high-throughput experimental data of residual oil hydrogenation reaction.

[0059] During specific implementation, it is necessary to perform data preprocessing on the collected high-throughput experimental data to construct a first sample data set for machine learning.

[0060] In one embodiment, Figure 2 As shown, step 102 may specifically include:

[0061] Step 201, unifying the units of various data of the high-throughput experimental data to obtain first processed high-throughput experimental data;

[0062] Step 202, correcting the erroneous data and inconsistent data in the first processed high-throughput experimental data to obtain second processed high-throughput experimental data;

[0063] Step 203, removing abnormal data from the second processed high-throughput experimental data to obtain third processed high-throughput experimental data;

[0064] Step 204, completing the missing data in the third processing high-throughput experimental data, deleting the data that cannot be completed, and obtaining the fourth processing high-throughput experimental data;

[0065] Step 205, encoding the non-numerical data in the fourth processed high-throughput experimental data to obtain fifth processed high-throughput experimental data;

[0066] Step 206 , perform feature scaling on the data in the fifth processed high-throughput experimental data to complete the construction of the first sample data set.

[0067] During specific implementation, in step 201, the units of various data of the high-throughput experimental data may be unified to obtain the first processed high-throughput experimental data.

[0068] During specific implementation, in step 202, erroneous data and inconsistent data in the first processed high-throughput experimental data may be corrected to obtain second processed high-throughput experimental data.

[0069] In specific implementation, in step 203, abnormal data that may exist in the first processed high-throughput experimental data can be detected and removed. The abnormal data detection method includes one or more combinations of the box plot method, DBSCAN clustering method, isolation forest method and Z-score method.

[0070] In specific implementation, in step 204, the missing data in the third processed high-throughput experimental data can be completed, and the data that cannot be completed can be deleted to obtain the third processed high-throughput experimental data. Among them, the method of completing the missing data can include one or more combinations of mean filling, median filling, mode filling, interpolation, forward filling, and backward filling.

[0071] In specific implementation, in step 205, the non-numerical data in the fourth processed high-throughput experimental data can be encoded to obtain the fifth processed high-throughput experimental data, wherein the encoding method can be one-hot encoding; the non-numerical data can be, for example, data such as catalyst components that cannot be directly represented by numerical values.

[0072] In specific implementation, in step 206, feature scaling can be performed on the data in the fifth processing high-throughput experimental data, wherein the feature scaling method includes one or more combinations of normalization, standardization and interval scaling, thereby completing the construction of the first sample data set.

[0073] Specifically, after the high-throughput experimental data are processed as above, new features that are useful for analyzing the residue oil hydrogenation reaction process can be constructed based on the feedstock oil property data, catalyst property data and reaction condition data in the processed high-throughput experimental data; the new features are added to the first sample data set, and key feature data that affect the residue oil hydrogenation reaction results are screened therefrom, and the key feature data are marked, wherein the screening method can include one or more combinations of filtering method, packaging method and embedding method.

[0074] In step 103, second sample data may be constructed based on historical data of a designated residue oil hydrogenation reaction unit.

[0075] In specific implementation, the historical data of the designated residue oil hydrotreating reaction unit is preprocessed, and the preprocessing includes: unifying data units, correcting erroneous and inconsistent data, eliminating abnormal data, filling or deleting missing data, encoding non-numeric data, feature scaling, etc.

[0076] In the above step 104, the first sample data set may be used to train a plurality of preset initial neural network models of different types to obtain a plurality of residual oil hydrogenation reaction product property analysis models.

[0077] In specific implementations, multiple different types of initial neural network models may include convolutional neural networks, graph neural networks, recurrent neural networks, feedforward neural networks, and Transformers.

[0078] In one embodiment, before the above step 104, the following steps may also be included:

[0079] Build the initial neural network model as follows:

[0080] Setting multiple different types of training hyperparameter types of neural networks, wherein the training hyperparameter types include one or any combination of the number of neural network layers, the number of neurons in each layer, the batch size, the learning rate, the number of training rounds, the weight decay coefficient, and the probability of randomly discarding neurons;

[0081] Using a random search algorithm, randomly sampling the value of each training hyperparameter type of each type of neural network from the preset value distribution information of each training hyperparameter type as a training hyperparameter combination, and randomly sampling multiple sets of training hyperparameter combinations of each type of neural network;

[0082] For each set of training hyperparameter combinations of each type of neural network, an initial neural network model is built, where the initial neural network model consists of two parts: a shallow neural network and a deep neural network. The shallow neural network is used to input feature data, and the deep neural network is used to input full data. The output data of each initial neural network model is the linear weighted sum of the output data of the shallow neural network and the output data of the deep neural network.

[0083] In one embodiment, the above step 104 may specifically include:

[0084] Dividing the first sample data set into a training set, a validation set, and a test set according to a preset ratio;

[0085] Initial neural network models built with multiple sets of training hyperparameter combinations for each type of neural network:

[0086] Use the training set to train the initial neural network model built with each set of training hyperparameter combinations;

[0087] Use the validation set to determine the error of the initial neural network model built with each set of training hyperparameter combinations, and stop model training in advance when the error of the initial neural network model no longer decreases;

[0088] After the initial neural network model constructed by multiple sets of training hyperparameter combinations of each type of neural network is trained, the generalization performance of the trained initial neural network model is tested using the test set, and the optimal training hyperparameter combination of each type of neural network is selected according to the test results;

[0089] When the generalization performance of the initial neural network model constructed by the optimal training hyperparameter combination of each type of neural network does not meet the preset conditions, the training hyperparameter combination of this type of neural network is re-collected, and the initial neural network model constructed by the re-collected training hyperparameter combination of this type of neural network is trained using the training set until the generalization performance of the initial neural network model meets the preset conditions, thereby obtaining a residue oil hydrogenation reaction product property analysis model for each type of neural network.

[0090] During specific implementation, the process of building an initial neural network model can be: first, determine a plurality of different types of neural networks, such as convolutional neural networks, graph neural networks, recurrent neural networks, feedforward neural networks, Transformers, and the like. For each type of neural network, it is also necessary to set the corresponding training hyperparameter type. Then, using a random search algorithm, randomly select the value of each training hyperparameter type from the value distribution information (pre-set) of each training hyperparameter type as a set of training hyperparameter combinations. In the implementation of the present invention, for each type of neural network, it is necessary to determine multiple sets of training hyperparameter combinations according to a preset number of groups. Finally, for each set of training hyperparameter combinations, build an initial neural network model. In other words, for each type of neural network, build multiple initial neural network models, and the values ​​of the training hyperparameters of each initial neural network model are different.

[0091] In specific implementation, the structure of each initial neural network model constructed above includes a shallow neural network and a deep neural network. The shallow neural network is mainly used to input feature data (i.e., the key feature data in the first sample data set above), and the deep neural network is used to input the full amount of data (i.e., the crude oil property data, catalyst property data, and reaction condition data in the first sample data set); the output data of each neural network model is the linear weighted sum of the output data of the shallow neural network and the output data of the deep neural network. In this way, the information loss caused by the layer-by-layer propagation of the deep neural network can be avoided. And the number of neurons in the input layer of each initial neural network model matches the number of input data, and the number of neurons in the output layer is the same as the number of product property data characterizing the residual oil hydrogenation reaction, and the network parameters are randomly initialized. Among them, the parameter random initialization method is one of normal initialization, uniform distribution initialization, Xavier initialization, and He initialization.

[0092] For example, Figure 3 This is a structural diagram of the initial neural network model provided in an embodiment of the present invention. Feature data is input into a shallow neural network. Preferably, the feature data may be total nitrogen content, total sulfur content, metal content, hydrogen pressure, temperature, catalyst grading ratio, catalyst pore size, specific surface area, etc. of the crude oil; data on crude oil properties, catalyst properties, reaction conditions, etc. are input into a deep neural network; the output data of the initial neural network model is the linear weighted sum of the output data of the shallow neural network and the output data of the deep neural network, i.e., product property data.

[0093] In specific implementation, the training of the above-mentioned initial neural network models can be to train the initial neural network models constructed by combining multiple sets of training hyperparameters for each type of neural network. The training process can be specifically: first, the first sample data set is divided into three parts: training set, validation set and test set according to a certain ratio, wherein the proportion of the training set data volume to the total data volume is not less than 60%, the training set is used for model training, the validation set is used to check the state of the model during the training process, and the test set is used to evaluate the generalization performance of the model.

[0094] Then, each initial neural network model is trained using the training set. Specifically, during model training, batch-sized crude oil property data, catalyst property data, reaction condition data, and corresponding product property data, as well as marked feature data, are randomly extracted from the training set each time, and the marked feature data are input into a shallow neural network to obtain predicted product property data output by the shallow neural network, and crude oil property data, catalyst property data, and reaction condition data are input into a deep neural network to obtain predicted product property data output by the deep neural network through network forward propagation, and the two predicted product property data are linearly weighted summed to obtain the final model prediction value, and the gap between the model prediction value and the actual product property data is calculated, and the model parameters are updated through the batch gradient descent method, and the training is completed when the set number of training rounds is reached.

[0095] Among them, the mean squared error (MSE) loss function can be used to calculate the error between the model prediction value and the true value. The calculation formula is shown as follows:

[0096]

[0097] Where: y i is the true value, is the model prediction value; MSE is the error, and n is the total number of model prediction values ​​or true values.

[0098] In specific implementation, the validation set can also be used to monitor the training process of the initial neural network model. Specifically, the error of the initial neural network model is calculated based on the validation set. When the error no longer decreases, even if the number of training times of the initial neural network model has not reached the number of training rounds, the training of the initial neural network model needs to be stopped in advance. In this way, overfitting of the initial neural network model can be avoided.

[0099] In specific implementation, after the training of the initial neural network model constructed by multiple sets of training hyperparameter combinations of each type of neural network is completed, the generalization performance of each trained initial neural network model can be tested using a test set, and the optimal training hyperparameter combination for the initial neural network model of each neural network type is screened according to the test results. When the generalization performance of the initial neural network model under the optimal training hyperparameter combination does not meet the preset requirements, the training hyperparameter combination search is performed again, and the initial neural network model is retrained using the training set. When the generalization performance meets the preset requirements, a residue oil hydrogenation reaction product property analysis model of each neural network type can be obtained.

[0100] That is to say, for neural network types such as convolutional neural network, graph neural network, recurrent neural network, feedforward neural network, Transformer, etc., multiple convolutional neural network models, multiple graph neural network models, multiple recurrent neural network models, multiple feedforward neural network models, and multiple Transforme models are built according to different training hyperparameter values, and the model with the optimal training hyperparameter values ​​is selected from the multiple models of each type as the property analysis model of the residue oil hydrotreating reaction product of each neural network type.

[0101] In the above step 105, the errors of multiple residue oil hydrogenation reaction product property analysis models are calculated, and the residue oil hydrogenation reaction product property analysis model with the smallest error is determined from the multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model.

[0102] During specific implementation, a group of data can be selected and input into each residue oil hydrogenation reaction product property analysis model respectively, the error of each residue oil hydrogenation reaction product property analysis model can be calculated, and the residue oil hydrogenation reaction product property analysis model with the smallest error can be selected as the designated residue oil hydrogenation reaction product property analysis model.

[0103] In the above step 106, the second sample data set may be used to optimize the parameters of the specified residue oil hydrogenation reaction product property analysis model to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit.

[0104] In specific implementation, since the second sample data set is constructed based on historical data of a specified residue oil hydrogenation reaction unit, the specified residue oil hydrogenation reaction product property analysis model is optimized using the second sample data set to obtain a residue oil hydrogenation reaction product property analysis model suitable for the specified residue oil hydrogenation reaction unit.

[0105] In one embodiment, Figure 4 As shown, step 106 may specifically include:

[0106] Step 401, replacing the output layer of the specified residue oil hydrogenation reaction product property analysis model according to the product property data in the second sample data set;

[0107] Step 402: Use the second sample data set to train the replaced output layer to obtain a residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit.

[0108] In specific implementation, the parameters of the neural network layers except the output layer in the specified residue oil hydrogenation reaction product property analysis model can be frozen, and the output layer can be replaced according to the product property data in the second sample data set; then, a small amount of optimization learning is performed on the replaced output layer using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit.

[0109] In this way, by using the second sample data set constructed by the historical data of the specified residue oil hydrogenation reaction unit to optimize the parameters of the specified residue oil hydrogenation reaction product property analysis model, only a small amount of device data needs to be collected to adaptively optimize the model parameters, and the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit can be obtained, thereby reducing the prediction error of the model on the specified residue oil hydrogenation reaction unit.

[0110] In the above step 106, the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit is input into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed.

[0111] In specific implementation, the reaction data to be analyzed are processed to meet the input layer requirements of the model, including unit conversion, data scaling and other measures; the processed reaction data to be analyzed are input into the optimization model, and the model prediction results, i.e., the product property data of the reaction data to be analyzed, are read.

[0112] In summary, the above-mentioned product property analysis method of residue oil hydrogenation reaction has the following beneficial effects:

[0113] 1. It helps to improve the analytical accuracy and adaptability of the product property analysis model of residue oil hydrogenation reaction. Specifically, a large amount of high-throughput experimental data of residue oil hydrogenation reaction is obtained by using a high-throughput experimental device to build a basic model, and then the historical data of the specified residue oil hydrogenation reaction device is used to optimize the model, so that only a small amount of device data is needed to obtain an analytical model with a wide range of applicability and good analytical performance.

[0114] 2. It can quickly build analysis models for residue oil hydrogenation reactions for different purposes. Specifically, when building analysis models to solve different problems, it is only necessary to replace the output layer of the basic model according to the characteristics of the problem, and then use the dataset of the specific problem to make a small adjustment to the model. It is simple and convenient to implement, and can be flexibly adjusted according to actual conditions.

[0115] 3. It helps to reduce the workload of residual oil hydrogenation reaction analysis and save a lot of manpower and material resources. Specifically, only a small amount of historical data information is needed to analyze the reaction performance of new catalysts and the reaction characteristics of unknown raw oil, thereby reducing the workload of manual analysis experiments and reducing R&D costs.

[0116] In order to more clearly understand the above-mentioned method for analyzing the properties of the products of the residue oil hydrogenation reaction, a specific embodiment and the comparison between the embodiment and the prior art are described below.

[0117] Embodiment 1:

[0118] Step 1: Determine the properties of the catalyst and the feedstock oil. In this embodiment, a total of 3 catalysts and 10 feedstock oils were used. The catalyst properties measured included the constituent elements, and the feedstock oil properties measured included specific gravity, nitrogen content, sulfur content, residual carbon value, viscosity, C / H content, and metal impurity content.

[0119] Step 2: Set the reaction conditions for the high-throughput experiment and perform product property data analysis of the residue oil hydrogenation reaction. In this embodiment, the hydrogen pressure, temperature, feed space velocity of the crude oil, hydrogen / oil ratio, and catalyst grading ratio of the high-throughput experiment are set respectively. For each reaction condition, 3 different options are set, and 243 groups of different reaction conditions are finally generated through permutations and combinations. The reaction performance of the three catalysts with different crude oils is analyzed under 243 reaction conditions using a high-throughput evaluation device.

[0120] Step 3: Regularly sample and analyze the reaction products in the high-throughput experimental device. Set the total reaction time to 1440h, sample and analyze the reaction products of each reactor every 24 hours, record the nitrogen content, sulfur content, and carbon residue value of the product, and record the analysis results, and finally obtain 243×60×3×10=437400 reaction data.

[0121] Step 4: Clean the collected high-throughput experimental data and construct a residue oil hydrogenation high-throughput experimental reaction data set (i.e., the first sample data set). First, unify the data units in the experimental data, then correct the abnormal data in the evaluation data, remove the data that cannot be corrected, and then complete the missing data in the experimental data. Finally, 430,847 data are obtained for model construction.

[0122] Step 5: Based on the constructed residue oil hydrogenation high-throughput experimental reaction data set, a deep learning algorithm is used to construct a residue oil hydrogenation reaction product property analysis model. The data set is divided into three parts: training, validation and testing, where the training set contains 330,847 data, the validation set contains 50,000 data, and the test set contains 50,000 data. This embodiment uses a deep residual network to construct an analysis model. The constructed analysis model uses product property data as model output and other data as model input. The training set is used to train models with different hyperparameters. The mean square error is used as the loss function during training, and the stochastic gradient descent algorithm is used to update the model parameters. After the training is completed, the test set is used to test the effects of each model, and the model with the lowest prediction error is saved.

[0123] Step 6: Collect the actual industrial production process data set of residue oil hydrogenation (specify the historical data of the residue oil hydrogenation reaction unit) and fine-tune the model with the lowest prediction error. In this embodiment, 442 fine-tuning data were collected from the refinery residue oil hydrogenation analysis unit. When fine-tuning the model, the main structure of the model was frozen, the model output layer was replaced, and a small amount of training was performed on the model output layer using the collected industrial production process data set.

[0124] Step 7: Use the fine-tuned and learned product property analysis model for residue oil hydrogenation reaction to analyze the product property data of the actual residue oil hydrogenation process.

[0125] Result verification: In order to test the effect of the fine-tuning model (the property analysis model of the residue oil hydrogenation reaction product corresponding to the specified residue oil hydrogenation reaction unit), this embodiment collected an additional 100 data to test the effect of the fine-tuning model, and recorded the prediction index error as shown in Table 1.

[0126] Comparative Example 1:

[0127] Comparative Example 1 uses the same deep residual network as Example 1 to construct a prediction model. The model is trained using only the industrial data collected in Example 1. High-throughput experimental data is not used to pre-train the model during model construction. The optimal model structure is selected based on the effect of the model on the test set. The test results of the optimal model on 100 data are shown in Table 1.

[0128] Comparative Example 2:

[0129] Comparative Example 2 is a residue oil hydrogenation model and establishment method based on semi-supervised deep GRU, the method comprising: performing time series processing on the original data to obtain an unlabeled sample training set A and a labeled sample training set A*; using the min-max normalization method to normalize the sample data to eliminate the impact of different dimensions; the normalized sample data sets A and A* are both input data for the modeling process; using the unlabeled sample data set A processed as above to construct a deep GRU model, and using the GRU autoencoder to perform unsupervised pre-training on the deep GRU model layer by layer according to the greedy principle; initializing a deep GRU model of the same structure with the parameters obtained from the pre-training, and performing secondary training using the labeled data set A*, and obtaining the final residue oil hydrogenation model based on the semi-supervised deep GRU after fine-tuning through supervised learning.

[0130] Using the same training set, validation set, and test set as in Example 1, the data was preprocessed and a prediction model was constructed in the same manner as in Comparative Example 2. During the model construction process, the model was not pre-trained with high-throughput experimental data, and the optimal model structure was screened based on the effect of the model on the test set. The test results of the optimal model on 100 pieces of data are shown in Table 1.

[0131] Table 1 Prediction index error

[0132] Evaluation indicators Example 1 Comparative Example 1 Comparative Example 2 Mean Square Error (MSE) 2.373 4.865 3.763 Mean absolute error (MAE) 1.037 1.526 1.216 <![CDATA[Coefficient of determination R 2 > 0.933 0.873 0.895

[0133] As can be seen from Table 1, the difference between the predicted value and the actual value of the fine-tuning model combined with high-throughput experimental data is small, and the linear correlation is the best. Comparative Example 1, which is not combined with high-throughput experimental data, has a 105% increase in mean square error and a 47% increase in mean absolute error compared to the embodiment, and the model prediction model error increases, and the linear correlation deteriorates. The prediction model established according to the method described in Comparative Example 2 has a 53% increase in mean square error and a 17% increase in mean absolute error compared to the embodiment, and the model prediction effect is worse than the prediction model established according to this method.

[0134] The present invention also provides a product property analysis device for residual oil hydrogenation reaction, as described in the following embodiments. Since the principle of the device to solve the problem is similar to the product property analysis method for residual oil hydrogenation reaction, the implementation of the device can refer to the implementation of the product property analysis method for residual oil hydrogenation reaction, and the repeated parts will not be repeated.

[0135] like Figure 5 FIG. 1 is a schematic diagram of a product property analysis device for a residual oil hydrogenation reaction provided by an embodiment of the present invention, and the device may include:

[0136] The acquisition module 501 is used to acquire high-throughput experimental data of the residue oil hydrogenation reaction and historical data of the designated residue oil hydrogenation reaction device; the high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data and product property data;

[0137] A first sample construction module 502 is used to construct a first sample data set according to high-throughput experimental data of residual oil hydrogenation reaction;

[0138] The second sample construction module 503 is used to construct a second sample data set according to the historical data of the designated residue oil hydrogenation reaction unit;

[0139] A model training module 504 is used to train a plurality of different types of preset initial neural network models using the first sample data set to obtain a plurality of residual oil hydrogenation reaction product property analysis models;

[0140] A model screening module 505 is used to calculate the errors of multiple residue oil hydrogenation reaction product property analysis models, and determine the residue oil hydrogenation reaction product property analysis model with the smallest error from the multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model;

[0141] The model optimization module 506 is used to optimize the parameters of the specified residue oil hydrogenation reaction product property analysis model using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit;

[0142] The analysis module 507 is used to input the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed.

[0143] In one embodiment, the catalyst property data may include: constituent elements of the catalyst and mass fraction of each constituent element, bulk density, specific surface area, pore size, pore volume and active metal content.

[0144] In one embodiment, the reaction condition data may include: feedstock oil space velocity, hydrogen pressure, temperature, ratio of hydrogen to oil, and catalyst grading ratio.

[0145] In one embodiment, the product property data may include: density, viscosity, nitrogen content, sulfur content, carbon residue value, carbon and hydrogen content and metal impurity content of the hydrogenated product.

[0146] In one embodiment, the first sample construction module may be specifically used for:

[0147] Unifying the units of various data of the high-throughput experimental data to obtain first processed high-throughput experimental data;

[0148] Correcting erroneous data and inconsistent data in the first processing high-throughput experimental data to obtain second processing high-throughput experimental data;

[0149] Eliminating abnormal data in the second processing high-throughput experimental data to obtain third processing high-throughput experimental data;

[0150] Completing the missing data in the third processing high-throughput experimental data, deleting the data that cannot be completed, and obtaining the fourth processing high-throughput experimental data;

[0151] Encoding the non-numerical data in the fourth processing high-throughput experimental data to obtain fifth processing high-throughput experimental data;

[0152] The data in the fifth processing high-throughput experimental data are feature scaled to complete the construction of the first sample data set.

[0153] In one embodiment, a model building module may also be included, which is used before the model training module uses the first sample data set to train a plurality of preset initial neural network models of different types to obtain a plurality of residual oil hydrogenation reaction product property analysis models:

[0154] Setting multiple different types of training hyperparameter types of neural networks, wherein the training hyperparameter types include one or any combination of the number of neural network layers, the number of neurons in each layer, the batch size, the learning rate, the number of training rounds, the weight decay coefficient, and the probability of randomly discarding neurons;

[0155] Using a random search algorithm, randomly sampling the value of each training hyperparameter type of each type of neural network from the preset value distribution information of each training hyperparameter type as a training hyperparameter combination, and randomly sampling multiple sets of training hyperparameter combinations of each type of neural network;

[0156] For each set of training hyperparameter combinations of each type of neural network, an initial neural network model is built, where the initial neural network model consists of two parts: a shallow neural network and a deep neural network. The shallow neural network is used to input feature data, and the deep neural network is used to input full data. The output data of each initial neural network model is the linear weighted sum of the output data of the shallow neural network and the output data of the deep neural network.

[0157] In one embodiment, the model training module can be specifically used to:

[0158] Dividing the first sample data set into a training set, a validation set, and a test set according to a preset ratio;

[0159] Initial neural network models built with multiple sets of training hyperparameter combinations for each type of neural network:

[0160] Use the training set to train the initial neural network model built with each set of training hyperparameter combinations;

[0161] Use the validation set to determine the error of the initial neural network model built with each set of training hyperparameter combinations, and stop model training in advance when the error of the initial neural network model no longer decreases;

[0162] After the initial neural network model constructed by multiple sets of training hyperparameter combinations of each type of neural network is trained, the generalization performance of the trained initial neural network model is tested using the test set, and the optimal training hyperparameter combination of each type of neural network is selected according to the test results;

[0163] When the generalization performance of the initial neural network model constructed by the optimal training hyperparameter combination of each type of neural network does not meet the preset conditions, the training hyperparameter combination of this type of neural network is re-collected, and the initial neural network model constructed by the re-collected training hyperparameter combination of this type of neural network is trained using the training set until the generalization performance of the initial neural network model meets the preset conditions, thereby obtaining a residue oil hydrogenation reaction product property analysis model for each type of neural network.

[0164] In one embodiment, the model optimization module can be specifically used to:

[0165] replacing the output layer of the specified residue oil hydrogenation reaction product property analysis model according to the product property data in the second sample data set;

[0166] The replaced output layer is trained using the second sample data set to obtain a residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit.

[0167] The embodiment of the present invention also provides a computer device, such as Figure 6 As shown, it is a schematic diagram of a computer device in an embodiment of the present invention. The computer device 600 includes a memory 610, a processor 620, and a computer program 630 stored in the memory 610 and executable on the processor 620. When the processor 620 executes the computer program 630, the product property analysis method of the above-mentioned residue oil hydrogenation reaction is implemented.

[0168] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the product property analysis method of the residue oil hydrogenation reaction is implemented.

[0169] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the method for analyzing the properties of products of the residual oil hydrogenation reaction is implemented.

[0170] In the embodiment of the present invention, high-throughput experimental data of residue oil hydrogenation reaction and historical data of a designated residue oil hydrogenation reaction device are obtained; the high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data and product property data; a first sample data set is constructed based on the high-throughput experimental data of residue oil hydrogenation reaction; a second sample data set is constructed based on the historical data of the designated residue oil hydrogenation reaction device; a plurality of different types of preset initial neural network models are trained respectively using the first sample data set to obtain a plurality of residue oil hydrogenation reaction product property analysis models; a plurality of residue oil hydrogenation reaction product property analysis models are calculated; The error of the residue oil hydrogenation reaction product property analysis model is determined, and the residue oil hydrogenation reaction product property analysis model with the smallest error is determined from multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model; the parameters of the designated residue oil hydrogenation reaction product property analysis model are optimized using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit; the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit is input into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed. Compared with the existing technical solutions for analyzing the properties of products of residue oil hydrogenation reaction, firstly, multiple different types of neural network models are trained by high-throughput experimental data of residue oil hydrogenation reaction to obtain multiple residue oil hydrogenation reaction product property analysis models, and the residue oil hydrogenation reaction product property analysis model with the smallest error is screened out, so as to establish a basic model with a wide range of adaptability. Then, the parameters of the basic model are adaptively optimized by using the historical data of the specified residue oil hydrogenation reaction unit to obtain the residue oil hydrogenation reaction product property analysis model of the specified residue oil hydrogenation reaction unit. In this way, the product property analysis model of the residue oil hydrogenation reaction unit is used to analyze the product property data of the reaction to be analyzed of the specified residue oil hydrogenation reaction unit, so as to reduce the prediction error of the model on the specified residue oil hydrogenation reaction unit. The method of the embodiment of the present invention is helpful to expand the scope of application of the model and improve the prediction accuracy of the model. At the same time, the demand for the data of the specified residue oil hydrogenation reaction unit when building the model is reduced, the difficulty of modeling is reduced, and the work efficiency is greatly improved.

[0171] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0173] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0174] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0175] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing the properties of products of residual oil hydrogenation reaction, characterized in that: include: Obtain high-throughput experimental data for residue hydroprocessing and historical data for a specified residue hydroprocessing unit; The high-throughput experimental data and the historical data both include: feed oil property data, catalyst property data, reaction condition data and product property data; Constructing a first sample data set based on high-throughput experimental data of residue oil hydrogenation reaction; constructing a second sample data set according to historical data of a designated residue oil hydroprocessing reaction unit; Using the first sample data set to train a plurality of preset initial neural network models of different types, respectively, to obtain a plurality of residual oil hydrogenation reaction product property analysis models; Calculating the errors of multiple residue oil hydrogenation reaction product property analysis models, and determining the residue oil hydrogenation reaction product property analysis model with the smallest error from the multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model; Optimizing the parameters of the specified residue oil hydrogenation reaction product property analysis model using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit; The reaction data to be analyzed of the designated residue oil hydrogenation reaction unit are input into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed.

2. The method according to claim 1, characterized in that The catalyst property data include: constituent elements of the catalyst and the mass fraction of each constituent element, bulk density, specific surface area, pore size, pore volume and active metal content.

3. The method according to claim 1, characterized in that The reaction condition data include: feed space velocity of crude oil, hydrogen pressure, temperature, ratio of hydrogen to oil, and catalyst grading ratio.

4. The method according to claim 1, characterized in that The product property data include: density, viscosity, nitrogen content, sulfur content, residual carbon value, carbon and hydrogen content and metal impurity content of the hydrogenated product.

5. The method according to claim 1, characterized in that Based on the high-throughput experimental data of residue hydrotreating reaction, the first sample data set was constructed, including: Unifying the units of various data of the high-throughput experimental data to obtain first processed high-throughput experimental data; Correcting erroneous data and inconsistent data in the first processing high-throughput experimental data to obtain second processing high-throughput experimental data; Eliminating abnormal data in the second processing high-throughput experimental data to obtain third processing high-throughput experimental data; Completing the missing data in the third processing high-throughput experimental data, deleting the data that cannot be completed, and obtaining the fourth processing high-throughput experimental data; Encoding the non-numerical data in the fourth processing high-throughput experimental data to obtain fifth processing high-throughput experimental data; The data in the fifth processing high-throughput experimental data are feature scaled to complete the construction of the first sample data set.

6. The method according to claim 1, characterized in that Before respectively training a plurality of preset initial neural network models of different types using the first sample data set to obtain a plurality of residual oil hydrogenation reaction product property analysis models, the method further includes: Setting multiple different types of training hyperparameter types of neural networks, wherein the training hyperparameter types include one or any combination of the number of neural network layers, the number of neurons in each layer, the batch size, the learning rate, the number of training rounds, the weight decay coefficient, and the probability of randomly discarding neurons; Using a random search algorithm, randomly sampling the value of each training hyperparameter type of each type of neural network from the preset value distribution information of each training hyperparameter type as a training hyperparameter combination, and randomly sampling multiple sets of training hyperparameter combinations of each type of neural network; For each set of training hyperparameter combinations of each type of neural network, an initial neural network model is built, where the initial neural network model consists of two parts: a shallow neural network and a deep neural network. The shallow neural network is used to input feature data, and the deep neural network is used to input full data. The output data of each initial neural network model is the linear weighted sum of the output data of the shallow neural network and the output data of the deep neural network.

7. The method according to claim 6, characterized in that The first sample data set is used to train a plurality of preset different types of initial neural network models respectively, to obtain a plurality of residual oil hydrogenation reaction product property analysis models, including: Dividing the first sample data set into a training set, a validation set, and a test set according to a preset ratio; Initial neural network models built with multiple sets of training hyperparameter combinations for each type of neural network: Use the training set to train the initial neural network model built with each set of training hyperparameter combinations; Use the validation set to determine the error of the initial neural network model built with each set of training hyperparameter combinations, and stop model training in advance when the error of the initial neural network model no longer decreases; After the initial neural network model constructed by multiple sets of training hyperparameter combinations of each type of neural network is trained, the generalization performance of the trained initial neural network model is tested using the test set, and the optimal training hyperparameter combination of each type of neural network is selected according to the test results; When the generalization performance of the initial neural network model constructed by the optimal training hyperparameter combination of each type of neural network does not meet the preset conditions, the training hyperparameter combination of this type of neural network is re-collected, and the initial neural network model constructed by the re-collected training hyperparameter combination of this type of neural network is trained using the training set until the generalization performance of the initial neural network model meets the preset conditions, thereby obtaining a residue oil hydrogenation reaction product property analysis model for each type of neural network.

8. The method according to claim 1, characterized in that The parameters of the specified residue oil hydrogenation reaction product property analysis model are optimized using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit, including: replacing the output layer of the specified residue oil hydroprocessing reaction product property analysis model according to the product property data in the second sample data set; The replaced output layer is trained using the second sample data set to obtain a residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit.

9. A product property analysis device for residual oil hydrogenation reaction, characterized in that: include: An acquisition module is used to acquire high-throughput experimental data of residue hydroprocessing reaction and historical data of a specified residue hydroprocessing reaction unit; The high-throughput experimental data and the historical data both include: feedstock oil property data, catalyst property data, reaction condition data and product property data; A first sample construction module is used to construct a first sample data set according to high-throughput experimental data of residual oil hydrogenation reaction; A second sample construction module is used to construct a second sample data set according to historical data of a designated residue oil hydrogenation reaction unit; A model training module, used to train a plurality of preset initial neural network models of different types using the first sample data set to obtain a plurality of residual oil hydrogenation reaction product property analysis models; A model screening module is used to calculate the errors of multiple residue oil hydrogenation reaction product property analysis models, and determine the residue oil hydrogenation reaction product property analysis model with the smallest error from the multiple residue oil hydrogenation reaction product property analysis models as the designated residue oil hydrogenation reaction product property analysis model; A model optimization module, used to optimize the parameters of the specified residue oil hydrogenation reaction product property analysis model using the second sample data set to obtain the residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit; The analysis module is used to input the reaction data to be analyzed of the designated residue oil hydrogenation reaction unit into the residue oil hydrogenation reaction product property analysis model corresponding to the designated residue oil hydrogenation reaction unit to determine the product property data of the reaction data to be analyzed.

10. The device according to claim 9, characterized in that The catalyst property data include: constituent elements of the catalyst and the mass fraction of each constituent element, bulk density, specific surface area, pore size, pore volume and active metal content.

11. The device according to claim 9, characterized in that The reaction condition data include: feed space velocity of crude oil, hydrogen pressure, temperature, ratio of hydrogen to oil, and catalyst grading ratio.

12. The device according to claim 9, characterized in that The product property data include: density, viscosity, nitrogen content, sulfur content, residual carbon value, carbon and hydrogen content and metal impurity content of the hydrogenated product.

13. The device according to claim 9, characterized in that The first sample building module is specifically used to: Unifying the units of various data of the high-throughput experimental data to obtain first processed high-throughput experimental data; Correcting erroneous data and inconsistent data in the first processing high-throughput experimental data to obtain second processing high-throughput experimental data; Eliminating abnormal data in the second processing high-throughput experimental data to obtain third processing high-throughput experimental data; Completing the missing data in the third processing high-throughput experimental data, deleting the data that cannot be completed, and obtaining the fourth processing high-throughput experimental data; Encoding the non-numerical data in the fourth processing high-throughput experimental data to obtain fifth processing high-throughput experimental data; The data in the fifth processing high-throughput experimental data are feature scaled to complete the construction of the first sample data set.

14. The device according to claim 9, characterized in that The method further includes a model building module, which is used for training a plurality of preset initial neural network models of different types using the first sample data set before the model training module obtains a plurality of residual oil hydrogenation reaction product property analysis models: Setting multiple different types of training hyperparameter types of neural networks, wherein the training hyperparameter types include one or any combination of the number of neural network layers, the number of neurons in each layer, the batch size, the learning rate, the number of training rounds, the weight decay coefficient, and the probability of randomly discarding neurons; Using a random search algorithm, randomly sampling the value of each training hyperparameter type of each type of neural network from the preset value distribution information of each training hyperparameter type as a training hyperparameter combination, and randomly sampling multiple sets of training hyperparameter combinations of each type of neural network; For each set of training hyperparameter combinations of each type of neural network, an initial neural network model is built, where the initial neural network model consists of two parts: a shallow neural network and a deep neural network. The shallow neural network is used to input feature data, and the deep neural network is used to input full data. The output data of each initial neural network model is the linear weighted sum of the output data of the shallow neural network and the output data of the deep neural network.

15. The device according to claim 14, characterized in that Model training module, specifically used for: Dividing the first sample data set into a training set, a validation set, and a test set according to a preset ratio; Initial neural network models built with multiple sets of training hyperparameter combinations for each type of neural network: Use the training set to train the initial neural network model built with each set of training hyperparameter combinations; Use the validation set to determine the error of the initial neural network model built with each set of training hyperparameter combinations, and stop model training in advance when the error of the initial neural network model no longer decreases; After the initial neural network model constructed by multiple sets of training hyperparameter combinations of each type of neural network is trained, the generalization performance of the trained initial neural network model is tested using the test set, and the optimal training hyperparameter combination of each type of neural network is selected according to the test results; When the generalization performance of the initial neural network model constructed by the optimal training hyperparameter combination of each type of neural network does not meet the preset conditions, the training hyperparameter combination of this type of neural network is re-collected, and the initial neural network model constructed by the re-collected training hyperparameter combination of this type of neural network is trained using the training set until the generalization performance of the initial neural network model meets the preset conditions, thereby obtaining a residue oil hydrogenation reaction product property analysis model for each type of neural network.

16. The device according to claim 9, characterized in that Model optimization module, specifically used for: replacing the output layer of the specified residue oil hydroprocessing reaction product property analysis model according to the product property data in the second sample data set; The replaced output layer is trained using the second sample data set to obtain a residue oil hydrogenation reaction product property analysis model corresponding to the specified residue oil hydrogenation reaction unit.

17. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

18. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

19. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images