Residual oil hydrotreating catalyst grading filling optimization method and device

Abstract

The invention provides a residual oil hydrotreating catalyst grading filling optimization method and device, and the method comprises the steps: carrying out the collection of residual oil hydrogenation reaction data, and constructing a first data set for machine learning; based on the constructed first data set, establishing an initial residual oil hydrogenation reaction prediction model by using a deep learning technology; collecting production process data of the industrial residual oil hydrogenation device to be tested, and constructing a second data set; based on the second data set, optimizing parameters of the screened initial residual oil hydrogenation reaction prediction model to obtain a target residual oil hydrogenation reaction prediction model; and carrying out catalyst grading filling optimization on the industrial residual oil hydrogenation device to be tested by utilizing the target residual oil hydrogenation reaction prediction model. The method is helpful for improving the prediction precision and the application range of the residual oil hydrogenation prediction model.

Description

Technical Field

[0001] The invention relates to the technical field of catalyst materials, and in particular to a method and a device for optimizing the graded loading of a residual oil hydroprocessing catalyst. Background Art

[0002] At present, with the increasing heaviness and inferiority of oil resources worldwide, as well as the urgent requirements for upgrading the quality of domestic oil products and increasingly stringent environmental regulations, refining companies 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 impurities such as 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 squeeze out the crude oil in the process of oil refining. At present, the total capacity of fixed-bed residue oil hydrogenation in my country exceeds 70 million tons / year, which has become the core technology for refining companies to process heavy and inferior raw materials and improve economic benefits.

[0003] Catalysts are the core of fixed-bed residue oil hydroprocessing technology. According to different functions, fixed-bed residue oil hydroprocessing catalysts are usually divided into four categories: protective agents, demetallization catalysts, desulfurization catalysts, and carbon removal and denitrification catalysts, with up to more than ten grades. Industrial residue oil hydroprocessing units must grade and load catalysts of different performances into the reactor according to the requirements of the reaction, so as to achieve the purpose of step-by-step removal or accommodation of different impurities in the residue oil, optimize the synergistic effect between different catalysts, make the hydrogenation products meet the index requirements, and achieve the synchronous deactivation of different types of catalysts, so as to maximize the service life of the catalyst and the operation cycle of the unit. Therefore, graded loading of catalysts is one of the core technologies of fixed-bed residue oil hydroprocessing, which has a key impact on the decontamination effect and operation cycle of the unit. The design of catalyst grading or loading ratio is a key link in the formulation of fixed-bed residue oil hydroprocessing catalyst application plan.

[0004] In the process of residual oil hydrogenation technology application, in order to design the technical scheme well, provide high-quality catalyst solutions to refineries, and ensure the best effect of technical application, it is generally necessary to use the real raw oil of the intended refinery to carry out raw oil adaptability evaluation test before the preparation of the technical application plan, obtain evaluation data, and guide the preparation of the technical application plan. The traditional method of obtaining data using fixed bed evaluation test equipment has the disadvantages of long time, high cost, low efficiency, and poor flexibility. It cannot meet the needs of large-scale promotion and application of residual oil hydrogenation technology, and has become a key link restricting the efficient application of technology in this field.

[0005] Therefore, for industrial hydrogenation units, in order to guide production operations flexibly, efficiently and at low cost, people use methods based on reaction kinetics to establish mathematical models to predict product properties, optimize reaction conditions, predict unit operation cycles and optimize catalyst loading schemes. People have developed a variety of kinetic models to simulate the residue hydrogenation process. For example, Jiang Lijing (Research on Residue Hydrogenation Reaction Kinetics and Comprehensive Process [D], Dalian University of Technology, 2011) used the measured data of residue hydrogenation industrial units of FZC series catalysts, applied empirical kinetic models and catalyst time-varying deactivation models to simulate and analyze the measured operation data of three units in Maoming, Qilu and Hainan, and established a residue hydrogenation deactivation kinetic model. Li Xinyuan (Research on Residue Hydrogenation Reaction Kinetic Model [D], Xiamen University, 2019) constructed a residue catalytic hydrodesulfurization (HDS) kinetic model to describe the quantitative relationship between reaction conditions, catalyst deactivation and feedstock properties and residue HDS conversion rate.

[0006] However, residue oil is the part of crude oil with the highest boiling point, largest molecular weight, most complex composition and highest impurity content. During the hydrogenation process, residue oil undergoes many reactions such as hydrodesulfurization, demetallization, denitrogenation, carbon removal and cracking. The impurity content, molecular structure characteristics and hydrogenation reaction characteristics of residue oil from different origins and different base crude oils vary greatly. In addition, during the hydrogenation process of residue oil, the catalyst usually deactivates much faster than the distillate oil hydrogenation catalyst due to the deposition of metal impurities and carbon deposition. Therefore, it is very difficult to simulate the residue oil hydrogenation process using kinetic equations based on reaction mechanisms. It is difficult to accurately reflect the effects of changes in factors such as raw material properties, catalyst performance and grading, process conditions and catalyst deactivation on the reaction results. Kinetic models for residue oil hydrogenation generally have problems such as narrow adaptability, low precision, poor flexibility, and difficulty in fitting and solving model parameters.

[0007] In order to overcome the defects of the mechanism-based modeling method, machine learning methods have also been used to establish a prediction model for the residue oil hydrogenation process to predict product properties and optimize reaction conditions. For example, Pan Shengping (Application of Artificial Neural Networks in Residue Oil Hydrogenation Process Research [J]. Fushun Hydrocarbon Processing Technology, 1999 (2): 12-16.) used the evaluation data of the 200ml small residue oil hydrogenation unit of Fushun Petrochemical Research Institute, applied artificial neural networks to simulate and analyze the fixed bed residue oil hydrogenation process, and established a product property prediction model. Sheng Mingmin (Research on Modeling of Residue Oil Hydrogenation Unit Based on Deep Recurrent Neural Network [D], Zhejiang University, 2019) used long short-term memory neural networks and gated recurrent unit neural networks to establish new hydrogen flow prediction models, which are used to achieve stable and accurate prediction of the new hydrogen flow of the residue oil hydrogenation unit.

[0008] The modeling method based on machine learning is to deeply mine the rules and relationships hidden in the existing residue oil hydrogenation reaction data, and then establish a relationship model between the residue oil raw material properties, catalysts, hydrogenation process conditions and reaction results. The effect of the built model is very dependent on the quality of the data set. The existing prediction methods usually only use specific industrial device data in the modeling process. In order to maintain stable operation, the raw material properties and operating conditions of industrial devices usually do not change much, and the catalyst loading and grading ratio are fixed. The universality of the model established by the specific device data is restricted. If it is necessary to establish a prediction model for different industrial devices, it is necessary to use the data of the new device to start from scratch to establish the prediction model, which is labor-intensive, difficult to collect data, and poor in flexibility. Therefore, there is an urgent need in this field to develop a residue oil hydrogenation reaction process prediction method that has low dependence on industrial device data, is flexible and efficient, has a wide range of adaptability, and has high prediction accuracy.

[0009] Among them, patent CN 106202910A provides a real-time prediction method for the yield of a residue oil hydroprocessing unit and its application. It uses data collected from the actual residue oil hydroprocessing process. On the basis of the hydrogenation reaction principle, based on the mechanism and operating characteristics of the hydrogenation unit, the reaction kinetics and reactor model of the residue oil hydroprocessing unit are established based on the mathematical equation model. According to the actual industrial unit data, the improved differential algorithm is used to perform real-time correction on the residue oil hydrogenation mechanism model. Based on this, the operating characteristics analysis is carried out for key operating / process conditions. Combined with neural network technology, an agent model is established to associate the operating conditions with Delta-Base data to achieve real-time prediction of the yield of the residue oil hydroprocessing unit. The defect of this technology is that it trains the neural network real-time prediction model by calculating the analysis data obtained by the correction model. The data source used is the simulation data generated by the mechanism model, which has a certain deviation from the real data, affecting the prediction effect of the neural network model on the real device. In addition, the established model is only applicable to specific devices and has poor adaptability. If a prediction model for different devices needs to be established, it is necessary to re-collect data to calibrate the mechanism model, and then establish a real-time neural network proxy model based on the corrected mechanism model.

[0010] Patent CN 110070923 A provides a semi-supervised deep GRU-based residual oil hydrogenation model and establishment method. The technology first performs time series processing on the original data to obtain an unlabeled sample training set A and a labeled sample training set A*, and then uses the min-max standardization method to normalize the sample data to eliminate the influence of different dimensions. The normalized sample data sets A and A* are both input data for the modeling process. The processed unlabeled sample data set A is used to build a deep GRU model, and the GRU autoencoder is used to perform unsupervised pre-training on the deep GRU model layer by layer according to the greedy principle. Finally, the parameters obtained from the pre-training are used to initialize the deep GRU model of the same structure, and the labeled data set A* is used for secondary training. After supervised learning fine-tuning, the final residual oil hydrogenation model based on the semi-supervised deep GRU is obtained. However, the defect of this technology is that it only uses specific industrial device data to establish the model during the modeling process, and the model has a small range of adaptability and is difficult to expand and apply to other devices.

[0011] The modeling method of the residue oil hydrogenation reaction process provided by the above-mentioned comparative documents mainly uses BP neural network and recurrent neural network to establish the reaction prediction model. During the model establishment process, only the data of specific industrial devices are used to optimize the model parameters. Since the industrial device operates relatively smoothly and the data changes little, the established model is usually only applicable to the residue oil hydrogenation reaction process under specific conditions, and the scope of application is relatively narrow. Summary of the invention

[0012] In view of the shortcomings of the prior art, the present invention proposes a method and device for optimizing the graded loading of a residue oil hydroprocessing catalyst, which is helpful to improve the prediction accuracy and adaptability of a residue oil hydroprocessing prediction model.

[0013] In order to achieve the above object, the present invention provides, on one hand, a method for optimizing the graded loading of a residue hydroprocessing catalyst, comprising:

[0014] Collecting residual oil hydrogenation reaction data and constructing the first data set for machine learning;

[0015] Based on the first constructed data set, a prediction model for initial residue oil hydrogenation reaction is established using deep learning technology;

[0016] Collecting production process data of the industrial residue oil hydrogenation unit to be tested to construct a second data set;

[0017] Based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation unit to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation unit to be tested;

[0018] The target residue oil hydrogenation reaction prediction model is used to optimize the catalyst graded loading for the industrial residue oil hydrogenation unit to be tested.

[0019] Optionally, a high throughput evaluation device is used to collect residual oil hydrogenation reaction data, including:

[0020] Prepare m kinds of residue oil hydrogenation catalysts and n kinds of residue oil raw materials respectively, and measure the properties of each catalyst and residue oil raw material, wherein m>=3, n>=3;

[0021] A residual oil feedstock and a residual oil hydrogenation catalyst combination are selected, and the residual oil hydrogenation reaction is evaluated under the preset high-throughput reaction conditions, and experimental data that meets the preset requirements are collected.

[0022] Optionally, the method of establishing an initial residue oil hydrogenation reaction prediction model based on the constructed first data set using deep learning technology also includes:

[0023] Based on the first constructed data set, multiple residue oil hydrogenation reaction prediction models are established using different neural networks, each neural network corresponding to a residue oil hydrogenation reaction prediction model;

[0024] The optimal prediction model is selected from the plurality of residue oil hydrogenation reaction prediction models as the initial residue oil hydrogenation reaction prediction model.

[0025] Optionally, based on the constructed first data set, multiple residue oil hydrogenation reaction prediction models are established using different neural networks, including:

[0026] Dividing the first data set into three parts: a first training set, a first validation set and a first test set according to a certain ratio;

[0027] A neural network is selected and a corresponding neural network prediction model is constructed as a residue oil hydrogenation reaction prediction model, wherein the key features affecting the residue oil hydrogenation reaction results screened out after preprocessing the residue oil hydrogenation reaction data are used as input features for constructing the prediction model, the number of neurons in the input layer of the network matches the number of features of the key features, the number of neurons in the output layer of the network is the same as the number of indicators characterizing the residue oil hydrogenation reaction results, and the network parameters are randomly initialized;

[0028] Preset network training hyperparameters, use the first training set to train the constructed residue oil hydrogenation reaction prediction model, and monitor the model training process through the first validation set, and stop model training when the model error on the first validation set no longer decreases;

[0029] The prediction performance of the trained model is tested using the first test set. If the test results do not meet the set standards, the model structure and network training hyperparameters are adjusted until the model test results meet the set standards.

[0030] Optionally, the neural network used includes one or a combination of convolutional neural network, graph neural network, recurrent neural network, feedforward neural network, and Transformer.

[0031] Optionally, the method further comprises:

[0032] Before constructing the first data set, the collected residual oil hydrogenation reaction data is further preprocessed, including:

[0033] The units of the various residual oil hydrogenation reaction data collected and evaluated by high throughput are unified;

[0034] Detecting possible abnormal data in the residual oil hydrogenation reaction data, and manually correcting or eliminating the abnormal data;

[0035] Fill in the missing data in the residue oil hydrogenation reaction data, and delete the data that cannot be filled;

[0036] One-hot encoding the categorical features in the residue oil hydrogenation reaction data;

[0037] Converting the numerical data of the residual oil hydrogenation reaction data;

[0038] Construct new features that are useful for predicting the process of residue hydrotreating;

[0039] The key features that affect the results of the residue hydrogenation reaction are screened as input features of the constructed residue hydrogenation reaction prediction model.

[0040] Optionally, the method further comprises:

[0041] Before constructing the second data set, the collected production process data is further preprocessed, including:

[0042] Unifying the units of the collected production process data;

[0043] Detecting possible abnormal data in the production process data, and manually correcting or eliminating the abnormal data;

[0044] Fill in the missing data in the production process data, and delete the data that cannot be filled;

[0045] The data value range of the production process data is scaled.

[0046] Optionally, the use of the target residue oil hydrogenation reaction prediction model to optimize catalyst graded loading for the industrial residue oil hydrogenation unit to be tested includes:

[0047] Preprocessing the production process data to be optimized to meet the input layer requirements of the target residue oil hydrogenation reaction prediction model;

[0048] Randomly select multiple groups of catalyst graded loading ratios, load the target residue oil hydrogenation reaction prediction model once at a certain temperature interval within a preset reaction temperature range, input the pre-treated production process data into the model, read the model prediction results, and predict the properties of the hydrogenation product;

[0049] The predicted deviation of hydrogenation product properties under each set of catalyst graded loading ratios is calculated, and the graded loading ratio with the smallest deviation of single hydrogenation product properties and the graded loading ratio with the smallest comprehensive deviation are output respectively.

[0050] Optionally, the production process data includes raw material properties, catalyst properties, product requirements and process conditions;

[0051] The raw material properties include density, viscosity, average molecular weight, nitrogen content, sulfur content, carbon residue value, C / H content, metal impurity content, and the composition of saturates, aromatics, colloids and asphaltenes;

[0052] The catalyst properties include the constituent elements of the catalyst and their mass fraction, bulk density, specific surface area, pore size, pore volume and active metal content;

[0053] The product requirements include density, viscosity, nitrogen content, sulfur content, residual carbon value and metal impurity content of the hydrogenated product; the process conditions include reaction hydrogen pressure, raw material feed space velocity, and hydrogen / oil ratio.

[0054] Another aspect of the present invention provides a residual oil hydroprocessing catalyst graded loading optimization device, using the residual oil hydroprocessing catalyst graded loading optimization method, the device at least comprises:

[0055] Dataset building blocks for:

[0056] Collecting residual oil hydrogenation reaction data to construct a first data set for machine learning; and

[0057] Collecting production process data of the industrial residue oil hydrogenation unit to be tested to construct a second data set;

[0058] Predictive model building blocks for:

[0059] Based on the first constructed data set, using deep learning technology to establish an initial residue oil hydrogenation reaction prediction model; and

[0060] Based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation unit to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation unit to be tested;

[0061] Catalyst graded loading optimization module for:

[0062] The target residue oil hydrogenation reaction prediction model is used to optimize the catalyst graded loading for the industrial residue oil hydrogenation unit to be tested.

[0063] It can be seen from the above scheme that the advantages of the present invention are:

[0064] The present invention provides a method for optimizing the graded loading of residue oil hydroprocessing catalysts, using a high-throughput evaluation device to collect residue oil hydrogenation reaction data, constructing a first data set for machine learning, and establishing an initial residue oil hydrogenation reaction prediction model based on the constructed first data set using deep learning technology. At the same time, the production process data of the industrial residue oil hydrogenation device to be tested is collected to construct a second data set, and based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation device to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation device to be tested; and finally, the target residue oil hydrogenation reaction prediction model is used to optimize the graded loading of catalysts for the industrial residue oil hydrogenation device to be tested. The method obtains a large amount of residue oil hydrogenation evaluation data by using a high-throughput evaluation device to construct a basic model, and then optimizes the model using the production process data of the industrial residue oil hydrogenation device to be tested, so that only a small amount of industrial data is required to obtain a prediction model with a wide range of applications and good prediction performance, which helps to improve the prediction accuracy and adaptability of the residue oil hydrogenation prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A schematic flow chart of a method for optimizing the graded loading of a residue oil hydroprocessing catalyst provided in an embodiment of the present invention;

[0066] Figure 2 for Figure 1 Specific flow chart of step S2;

[0067] Figure 3 for Figure 1 The specific flow chart of step S5;

[0068] Figure 4 It is a framework diagram of the device for optimizing the graded loading of the residual oil hydroprocessing catalyst of the present invention;

[0069] in:

[0070] 400- Residue hydroprocessing catalyst graded loading optimization device;

[0071] 401-dataset construction module;

[0072] 402-prediction model building module;

[0073] 403-Catalyst graded loading optimization module. DETAILED DESCRIPTION

[0074] In order to make the above features and effects of the present invention more clearly understood, embodiments are given below and described in detail with reference to the accompanying drawings.

[0075] As mentioned above, in view of the deficiencies of the prior art, the present invention provides a method and device for optimizing the graded loading of a residue oil hydroprocessing catalyst. Different from the prior art, the present invention first needs to conduct a high-throughput evaluation experiment in the modeling process, collect reaction data under different raw materials, different process conditions, different catalyst properties and grading ratios, and pre-train the model through high-throughput evaluation data, so as to establish a basic model with a wide range of adaptability, high precision and large capacity. Then, when predicting the residue oil hydrogenation reaction process of a specific industrial device, only a small amount of device data needs to be collected to optimize the model parameters adaptively, reduce the prediction error of the model on the specific industrial device, and use the optimized model for prediction. Relying on the constructed high-precision and high-adaptability residue oil hydrogenation reaction model, a catalyst graded loading adaptability optimization method is established for the raw material properties, product requirements and operating processes of a specific industrial residue oil hydrogenation device, providing guidance and support for the design of industrial application solutions. Constructing the target residue oil hydrogenation reaction model by combining pre-training with adaptive optimization helps to expand the scope of application of the model and improve the prediction accuracy of the model. At the same time, it reduces the demand for industrial device data when building the model, reduces the difficulty of modeling, and improves work efficiency.

[0076] refer to Figure 1 As shown, Figure 1 Detailed description of the process flow of the residual oil hydroprocessing catalyst graded loading optimization method provided by an embodiment of the present invention is shown in FIG.

[0077] A method for optimizing the graded loading of a residue hydroprocessing catalyst, specifically comprising:

[0078] S1. Collect residual oil hydrogenation reaction data and construct the first data set for machine learning.

[0079] In the specific implementation, this embodiment specifically uses a high-throughput evaluation device to collect residual oil hydrogenation reaction data. First, it is necessary to prepare m kinds of residual oil hydrogenation catalysts and n kinds of residual oil raw materials respectively, measure the properties of each catalyst and residual oil raw material and save the measurement results, where m>=3, n>=3. Among them, the residual oil hydrogenation catalyst contains at least one demetallizing agent, one desulfurizing agent and one decarbonization or denitrification catalyst. The measured catalyst properties include the constituent elements of the catalyst and their mass fraction, bulk density, specific surface area, pore size, pore volume and active metal content. The measured residual oil properties include density, viscosity, average molecular weight, nitrogen content, sulfur content, residual carbon value, C / H content, metal impurity (Ni, V, Fe, Ca, etc.) content and four components (saturated fraction, aromatic fraction, colloid and asphaltene), etc. At the same time, set high-throughput reaction conditions, including feed space velocity of raw oil, hydrogen pressure, temperature, hydrogen / oil ratio and catalyst grading ratio. The feed space velocity of the raw material can be selected in the range of 0.1 to 0.6h -1, the pressure range is 10-22MPa, the temperature range is 330-420℃, the hydrogen / oil ratio range is 300-1500, the demetallization catalyst grading ratio in the catalyst grading ratio is 20-70%, the desulfurization catalyst grading ratio is 20-60%, and the decarbonization or denitrification catalyst grading ratio is 10-60%. Then, the residual oil hydrogenation reaction evaluation experiment is carried out, specifically by selecting a residual oil raw material and a residual oil hydrogenation catalyst combination, under the above-mentioned preset high-throughput reaction conditions, the residual oil hydrogenation reaction evaluation is carried out, and the experimental data that meets the preset requirements are collected. That is: a set of raw oil feed space velocity, hydrogen pressure, hydrogen / oil ratio and catalyst grading ratio parameters are selected. The catalyst is loaded, dried, sulfurized and the residual oil raw material is switched according to the conventional method, and the initial reaction temperature is set to 340℃. During the reaction, other process conditions except the reaction temperature are kept unchanged, and the density, viscosity, nitrogen content, sulfur content, residual carbon value and metal impurity (Ni, V, Fe, Ca, etc.) content of the hydrogenated product are regularly analyzed. Based on the comparison of the content of metal Ni and V in the feedstock oil and the hydrogenated product, the cumulative deposition of metal Ni and V per unit volume of the catalyst MOC is calculated. When MOC is lower than or equal to 0.01g / mL, the reaction temperature is increased by 4°C for every 0.002g / mL increase in MOC, and the properties of the hydrogenated product when the reaction temperature increases by 2°C, increases by 4°C, decreases by 2°C, and decreases by 4°C are respectively examined; when MOC is higher than 0.01g / mL, the reaction temperature is increased by 4°C for every 0.01g / mL increase in MOC, and the properties of the hydrogenated product when the reaction temperature increases by 2°C, increases by 4°C, decreases by 2°C, and decreases by 4°C are respectively examined. Record the reaction temperature and analysis results, including the density, viscosity, nitrogen content, sulfur content, residual carbon value and metal impurity (Ni, V, Fe, Ca, etc.) content of the hydrogenated product; calculate the MOC value based on the analysis results. Continuously increase the reaction temperature as the amount of catalyst metal deposition increases. When the reaction temperature reaches 420°C and the properties of the hydrogenated product are investigated, end the evaluation experiment. Finally, reset the high-throughput reaction conditions, and then proceed with the above-mentioned residual oil hydrogenation reaction evaluation experiment process until the collected experimental data meets the preset requirements, including the number of experimental data requirements, the range of operating conditions covered by the experimental data, and the number of groups required.

[0080] S2. Based on the first constructed data set, a deep learning technology is used to establish an initial residue oil hydrogenation reaction prediction model.

[0081] In a specific implementation, before constructing the first data set, the collected residue oil hydrogenation reaction data is further preprocessed. The specific method used for preprocessing in this embodiment includes: first, unifying the units of the collected residue oil hydrogenation reaction data after high-throughput evaluation; then, detecting the abnormal data that may exist in the residue oil hydrogenation reaction data, and manually correcting or eliminating the abnormal data; wherein, 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. At the same time, during the detection process, the vacant data in the residue oil hydrogenation reaction data is completed, and the data that cannot be completed is deleted. Among them, the missing value filling method includes one or more combinations of mean filling, median filling, mode filling, interpolation method, forward filling method, and backward filling method. In addition, the classification features in the residue oil hydrogenation reaction data are further uniquely encoded, and the numerical data of the residue oil hydrogenation reaction data is converted. The conversion method includes one or more combinations of normalization, standardization and interval scaling. Then, new features that are beneficial to predicting the residue oil hydrogenation reaction process are constructed based on the existing data, and key features that affect the results of the residue oil hydrogenation reaction are screened as input features of the constructed residue oil hydrogenation reaction prediction model.

[0082] In this embodiment, after completing the data preprocessing process, a first data set is further constructed based on the preprocessed residue oil hydrogenation reaction data, and then based on the constructed first data set, an initial residue oil hydrogenation reaction prediction model is established using deep learning technology. In order to improve the accuracy of the prediction model, this embodiment specifically adopts a variety of different neural network technologies to construct different residue oil hydrogenation reaction prediction models, and selects the model with the best performance as the initial residue oil hydrogenation reaction prediction model. Specifically, one or a combination of convolutional neural networks, graph neural networks, recurrent neural networks, feedforward neural networks, Transformers, etc. can be used, but it is not limited to the above types. Figure 2 As shown in Figure 2 Shows Figure 1 The specific flow chart of step S2 in FIG. 1 shows a process of using a variety of different neural network technologies to establish an initial residue oil hydrogenation reaction prediction model, which specifically includes:

[0083] S21. Based on the first constructed data set, multiple residue oil hydrogenation reaction prediction models are established using different neural networks, each neural network corresponds to a residue oil hydrogenation reaction prediction model, specifically:

[0084] First, the first data set is divided into three parts: a first training set, a first validation set and a first test set according to a certain ratio, wherein the data volume of the first training set accounts for no less than 60% of the total data volume, the first training set is used to build the model, the first validation set is used to verify the state of the model during the training process, and the first test set is used to evaluate the generalization performance of the model.

[0085] Then, for each neural network, a corresponding neural network prediction model is built as a residual oil hydrogenation reaction prediction model, for example, a convolutional neural network, a graph neural network, a recurrent neural network, a feedforward neural network, and a Transformer are selected respectively to build a corresponding neural network prediction model. Among them, the key features that affect the residual oil hydrogenation reaction results screened out after the preprocessing of the residual oil hydrogenation reaction data are used as input features for constructing the prediction model, the number of neurons in the input layer of the network matches the number of features of the key features, the number of neurons in the output layer of the network is the same as the number of indicators that characterize the residual oil hydrogenation reaction results, and the network parameters are randomly initialized. The parameter random initialization method can adopt one of normal initialization, uniform distribution initialization, Xavier initialization, and He initialization.

[0086] At the same time, network training hyperparameters are preset, including one or more combinations of batch size, learning rate, number of training rounds, weight decay coefficient, and probability of randomly discarding neurons. The constructed residual oil hydrotreating reaction prediction model is trained using the first training set, and the model training process is monitored through the first validation set. When the model error on the first validation set is no longer reduced, the model training is stopped and the model parameters are saved.

[0087] In the specific implementation, the MSE mean square error loss function can be used to calculate the gap between the model prediction value and the true value during model training. The calculation formula is shown as follows:

[0088]

[0089] Where: y i is the true value, is the model prediction value;

[0090] During model training, the model parameters are updated through the gradient descent algorithm, and the error size of the updated model on the validation set is calculated regularly.

[0091] Finally, the prediction performance of the trained model is tested using the first test set. If the test results do not meet the set standards, the model structure and network training hyperparameters are adjusted and the above model training process is repeated until the model test results meet the set standards.

[0092] S22, then, the optimal prediction model is selected from the plurality of residue oil hydrogenation reaction prediction models as the initial residue oil hydrogenation reaction prediction model. That is, by comparing the test results of the residue oil hydrogenation reaction prediction models constructed by each neural network on the test set, the optimal prediction model is selected as the initial residue oil hydrogenation reaction prediction model.

[0093] S3. Collect production process data of the industrial residue oil hydrogenation unit to be tested and construct a second data set.

[0094] In this embodiment, the collected production process data of the industrial residue oil hydrogenation device to be tested include raw oil properties, catalyst properties, catalyst graded loading ratio, reaction process conditions and corresponding product property data.

[0095] In addition, in a specific implementation, before constructing the second data set, the collected production process data is further preprocessed, specifically including: unifying the units of the collected production process data; detecting abnormal data that may exist in the production process data, and manually correcting or eliminating the abnormal data; completing missing data in the production process data, and deleting data that cannot be completed, etc.; scaling the data value range of the production process data, and other preprocessing processes.

[0096] S4. Based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation unit to be tested and the high-throughput evaluation device, so as to obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation unit to be tested.

[0097] In a specific implementation, the selected initial residue oil hydrogenation reaction prediction model is loaded, the parameters of each neural network layer except the output layer are frozen, and then a small amount of optimization learning is performed on the network output layer using the collected second data set to eliminate the systematic error factors between the industrial residue oil hydrogenation device to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation device to be tested. Therefore, in this embodiment, when constructing a prediction model for predicting and solving 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 data set of the specific problem to make a small amount of fine-tuning on the model, which is simple and convenient to implement, and can be flexibly adjusted according to actual conditions, so as to realize the rapid construction of residue oil hydrogenation prediction models for different purposes.

[0098] S5. Utilizing the target residue oil hydrogenation reaction prediction model, the catalyst graded loading optimization is performed on the industrial residue oil hydrogenation unit to be tested.

[0099] In a specific implementation, after the target residue oil hydrogenation reaction prediction model is constructed, the target residue oil hydrogenation reaction prediction model can be used to optimize the catalyst graded loading for the industrial residue oil hydrogenation unit to be tested. Figure 3 As shown in Figure 3 Shows Figure 1 The specific flow chart of step S5, that is, the catalyst graded loading optimization process specifically includes:

[0100] S51. Preprocess the production process data to be optimized to meet the input layer requirements of the target residue oil hydrogenation reaction prediction model. The production process data include raw material properties, catalyst properties, product requirements and process conditions. The raw material properties include density, viscosity, average molecular weight, nitrogen content, sulfur content, carbon residue value, C / H content, metal impurity content (Ni, V, Fe, Ca, etc.), and four components (saturated fraction, aromatic fraction, colloid and asphaltene); the catalyst properties include the constituent elements of the catalyst and their mass fraction, bulk density, specific surface area, pore size, pore volume and active metal content; the product requirements include hydrogenation product density, viscosity, nitrogen content, sulfur content, carbon residue value and metal impurity content (Ni, V, Fe, Ca, etc.); the process conditions include reaction hydrogen pressure, raw material feed space velocity, hydrogen / oil ratio, etc.

[0101] S52, randomly selecting multiple groups of catalyst graded loading ratios, loading the target residue oil hydrogenation reaction prediction model once at a certain temperature interval within a preset reaction temperature range, inputting the pre-treated production process data into the model, reading the model prediction results, and predicting the properties of the hydrogenation product.

[0102] In this embodiment, at least ten groups of catalyst graded loading ratios are selected, the preset reaction temperature is set within the range of 340-410°C, and the target residual oil hydrogenation reaction prediction model is loaded and saved once every 5°C within the preset reaction temperature range, and the pre-treated production process data is input into the optimization model, and the model prediction results are read to predict the properties of the hydrogenation product. The properties of the hydrogenation product include one or a combination of density, viscosity, nitrogen content, sulfur content, residual carbon value and metal impurity (Ni, V, Fe, Ca, etc.) content.

[0103] S53, calculating the predicted deviation of the properties of the hydrogenated products under each group of catalyst graded loading ratios, and outputting the graded loading ratio with the smallest deviation of the single hydrogenated product property and the graded loading ratio with the smallest comprehensive deviation, respectively. The deviation includes one or more of the mean absolute error, root mean square error, standard deviation, mean square error, and relative error. The comprehensive error can be the average predicted deviation of all properties of the hydrogenated products or a combination of several properties.

[0104] Therefore, through the above-mentioned catalyst graded loading optimization process, only a large amount of historical data information is needed to predict the reaction performance of new catalysts and the reaction characteristics of unknown raw oils, and to optimize the catalyst graded loading according to process conditions and product index requirements, thereby reducing the workload of manual evaluation experiments and reducing R&D costs.

[0105] The following are several specific experimental examples to illustrate the process of the residual oil hydroprocessing catalyst graded loading optimization method provided by the present invention, and to verify the specific effect. The present invention specifically sets two examples and two comparative examples for illustration, as follows:

[0106] Example 1

[0107] Step 1, measuring the properties of the catalyst and the crude oil. In this embodiment, 2 demetallization catalysts, 1 desulfurization catalyst and 1 carbon residue removal and denitrification catalyst are used, and a total of 10 crude oils are used. The measured catalyst properties include the types and contents of the constituent elements, bulk density, specific surface area and pore volume; the measured crude oil properties include specific gravity, average molecular weight, viscosity, nitrogen content, sulfur content, carbon residue value, C / H content, metal content and four-component content.

[0108] Step 2, set the reaction conditions of the high-throughput evaluation experiment and conduct the high-throughput evaluation experiment. This embodiment specifically sets the pressure, air velocity, hydrogen / oil ratio, and catalyst grading of the evaluation experiment. For each reaction condition, 4 different options are set. The pressures are 12MPa, 15MPa, 18MPa, and 21MPa, and the air velocity is 0.16h / min. -1 、0.18h -1 , 0.20h -1 , 0.25h -1 , hydrogen / oil ratios were 500, 800, 1200, 1500, and catalyst gradations were 20% / 10% / 40% / 30%, 15% / 25% / 30% / 35%, 30% / 15% / 30% / 25%, 35% / 20% / 25% / 20%. 256 different evaluation conditions were finally generated through permutations and combinations, and the reaction performance of 10 different raw oils was evaluated under 256 reaction conditions using a high-throughput evaluation device.

[0109] Step 3, each evaluation is carried out according to the corresponding grading ratio of the four catalysts loaded in the reactor, the catalyst is dried and sulfurized according to the conventional method, and the residual oil raw material is switched, and the initial reaction temperature is set to 340 ° C. During the reaction, other process conditions except the reaction temperature are kept unchanged, and the density, viscosity, nitrogen content, sulfur content, residual carbon value and metal impurity (Ni, V, Fe, Ca, etc.) content of the hydrogenated product are analyzed every 48 hours, and the cumulative deposition of metal Ni and V per unit volume of catalyst MOC is calculated based on the comparison of the content of metal Ni and V in the raw oil and the hydrogenated product. When MOC is lower than or equal to 0.01g / mL, the reaction temperature increases by 4°C for every 0.002g / mL increase in MOC, and the properties of the hydrogenated products at this reaction temperature and when the reaction temperature increases by 2°C, increases by 4°C, decreases by 2°C, and decreases by 4°C are investigated; when MOC is higher than 0.01g / mL, the reaction temperature increases by 4°C for every 0.01g / mL increase in MOC, and the properties of the hydrogenated products at this reaction temperature and when the reaction temperature increases by 2°C, increases by 4°C, decreases by 2°C, and decreases by 4°C are investigated. Each feedstock oil was evaluated under 256 reaction conditions, and more than 500,000 sets of reaction data were obtained.

[0110] Step 4: Clean the collected high-throughput evaluation experimental data and construct the first data set of residual oil hydrogenation evaluation experimental reaction. 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, and finally obtain more than 490,000 sets of data for model construction.

[0111] Step 5, based on the first data set of the constructed residue oil hydrogenation experimental reaction, use a deep learning algorithm to construct an initial residue oil hydrogenation reaction prediction model. The first data set is divided into three parts: training, verification and testing, wherein the first training set contains more than 390,000 groups of data, the first verification set contains 50,000 groups of data, and the first test set contains 50,000 groups of data. In this embodiment, a deep residual network is used to construct a prediction model. The constructed prediction model uses product properties as model output and other data as model input. The first training set is used to train models with different hyperparameters. The mean square error is used as the loss function during training, and the model parameters are updated using the stochastic gradient descent algorithm. After the training is completed, the first test set is used to test the effects of each model, and the model with the lowest prediction error is saved.

[0112] Step 6, collect the actual industrial production process data set of residue oil hydrogenation and perform model fine-tuning optimization learning to obtain the target residue oil hydrogenation reaction prediction model. In this embodiment, 442 sets of production process data are collected from an industrial residue oil hydrogenation unit A for model optimization. During model optimization, the main structure of the model is frozen, the model output layer is replaced, and a small amount of training is performed on the model output layer using the collected industrial data.

[0113] Step 7, using the optimized and learned reaction prediction model of the residue oil hydrogenation process to optimize the catalyst graded loading of the industrial residue oil hydrogenation unit A. The comparison of the operating results of the industrial residue oil hydrogenation unit A before and after the optimization is shown in Table 1. The results in Table 1 show that after adopting the optimization method, the product properties are improved and the operation cycle is greatly extended.

[0114] Table 1 Comparison of catalyst grading before and after optimization in industrial residue oil hydrotreating unit A

[0115]

[0116]

[0117] *: Refers to the average relative error of hydroresidue S, CCR and Ni+V compared with the index.

[0118] Example 2

[0119] The same steps as steps 1-5 in Example 1 were used to establish a basic model for residue oil hydrogenation reaction, 328 sets of production process operation data were collected from a certain industrial residue oil hydrogenation unit B for model optimization, and a special model suitable for a certain industrial residue oil hydrogenation unit B was established. During model optimization, 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 production process data of the industrial residue oil hydrogenation unit B. When a certain industrial residue oil hydrogenation unit B changes the processed raw materials, this model is used to adaptively optimize the catalyst grading and loading, and the actual operation results are shown in Table 2. From the results in Table 2, it can be seen that after the raw material properties of a certain industrial residue oil hydrogenation unit B have changed significantly, the optimization method of the present invention is used to adaptively adjust the catalyst grading, and good results are achieved.

[0120] Table 2 Operation results of device B after catalyst grading optimization

[0121]

[0122]

[0123] Comparative Example 1

[0124] This comparative example uses the same deep residual network as Example 1 to construct a prediction model. The model is trained only using the production process operation data of the industrial residue oil hydrotreating unit A collected in Example 1. The high-throughput evaluation data is not used to pre-train the model during the model construction process. 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 3.

[0125] Comparative Example 2

[0126] This comparative example is a prediction model established according to the method described in patent CN110070923A.

[0127] The same training set, validation set, and test set as Example 1 were used to preprocess the data and build a prediction model according to the method described in patent CN110070923A. The high-throughput evaluation data was not used to pre-train the model during the model building process, and the optimal model structure was 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 3.

[0128] Table 3 Comparison of model accuracy using different methods

[0129]

[0130] As can be seen from Table 3, the difference between the predicted value and the actual value of the fine-tuning model combined with high-throughput evaluation data is small, and the linear correlation is the best. Comparative Example 1, which is not combined with high-throughput evaluation data, has a 105% increase in mean square error and a 47% increase in mean absolute error compared to the embodiment, an increase in the model prediction model error, and a deterioration in the linear correlation. The prediction model established according to the method described in patent CN110070923A 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 the method of the present invention.

[0131] In summary, the present invention provides a method for optimizing the graded loading of residue oil hydroprocessing catalysts, using a high-throughput evaluation device to collect residue oil hydrogenation reaction data, constructing a first data set for machine learning, and establishing an initial residue oil hydrogenation reaction prediction model based on the constructed first data set using deep learning technology. At the same time, the production process data of the industrial residue oil hydrogenation unit to be tested is collected to construct a second data set, and based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation unit to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation unit to be tested; and finally, the target residue oil hydrogenation reaction prediction model is used to optimize the graded loading of catalysts for the industrial residue oil hydrogenation unit to be tested. This method uses a high-throughput evaluation device to obtain a large amount of residue oil hydrogenation evaluation data to construct a basic model, and then uses the production process data of the industrial residue oil hydrogenation device to be tested to optimize the model, so that only a small amount of industrial data is needed to obtain a prediction model with a wide range of applications and good prediction performance, which helps to improve the prediction accuracy and adaptability of the residue oil hydrogenation prediction model.

[0132] In addition, the present invention specifically uses different neural networks to construct prediction models, and then selects the optimal prediction model, thereby improving the accuracy of the prediction model. When constructing prediction models for predicting and solving 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 data set of the specific problem to make a small fine-tuning of the model, which is simple and convenient to implement, and can be flexibly adjusted according to actual conditions, so as to realize the rapid construction of residue oil hydrogenation prediction models for different purposes.

[0133] In addition, the present invention only requires a large amount of historical data information to predict the reaction performance of new catalysts and the reaction characteristics of unknown raw oils, and optimizes the catalyst grading and loading according to process conditions and product index requirements, thereby reducing the workload of manual evaluation experiments and reducing R&D costs.

[0134] Reference Figure 4 , Figure 4 A residual oil hydroprocessing catalyst graded loading optimization device 400 is shown, which can be applied to personal terminals and host terminal equipment, and can be realized by Figure 1-Figure 3 The method for optimizing the graded loading of the catalyst for hydrotreating of residual oil shown in the embodiment of the present application can realize each process of the above method, and at least includes a data set construction module 401, a prediction model construction module 402, and a catalyst graded loading optimization module 403, which are specifically:

[0135] A residual oil hydroprocessing catalyst graded loading optimization device 400, the device at least comprising:

[0136] The data set construction module 401 is used to:

[0137] Using a high-throughput evaluation device to collect residual oil hydrogenation reaction data, and constructing a first data set for machine learning; and

[0138] Collecting production process data of the industrial residue oil hydrogenation unit to be tested to construct a second data set;

[0139] The prediction model building module 402 is used to:

[0140] Based on the first constructed data set, using deep learning technology to establish an initial residue oil hydrogenation reaction prediction model; and

[0141] Based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation unit to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation unit to be tested;

[0142] Catalyst graded loading optimization module 403 is used to:

[0143] The target residue oil hydrogenation reaction prediction model is used to optimize the catalyst graded loading for the industrial residue oil hydrogenation unit to be tested.

[0144] It should be understood that the various descriptions of the method are also applicable to the residue hydroprocessing catalyst graded loading optimization device 400 according to the embodiment of the present application, and will not be described in detail to avoid repetition.

[0145] In addition, it should be understood that in the residue hydroprocessing catalyst graded loading optimization device 400 according to the embodiment of the present application, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the device can be divided into functional modules different from the modules illustrated above to complete all or part of the functions described above.

[0146] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be applied, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0147] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for optimizing the graded loading of a residue hydroprocessing catalyst, characterized in that: include: Collecting residual oil hydrogenation reaction data and constructing the first data set for machine learning; Based on the first constructed data set, a prediction model for initial residue oil hydrogenation reaction is established using deep learning technology; Collecting production process data of the industrial residue oil hydrogenation unit to be tested to construct a second data set; Based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation unit to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation unit to be tested; The target residue oil hydrogenation reaction prediction model is used to optimize the catalyst graded loading for the industrial residue oil hydrogenation unit to be tested.

2. The method according to claim 1, characterized in that The high-throughput evaluation device is used to collect residual oil hydrogenation reaction data, including: Prepare m kinds of residue oil hydrogenation catalysts and n kinds of residue oil raw materials respectively, and measure the properties of each catalyst and residue oil raw material, wherein m>=3, n>=3; A residue oil feedstock and a residue oil hydrogenation catalyst combination are selected, and the residue oil hydrogenation reaction is evaluated under the preset high-throughput reaction conditions, and experimental data that meets the preset requirements are collected.

3. The method according to claim 1 or 2, characterized in that: The method of establishing an initial residue oil hydrogenation reaction prediction model based on the constructed first data set using deep learning technology also includes: Based on the first constructed data set, multiple residue oil hydrogenation reaction prediction models are established using different neural networks, each neural network corresponding to a residue oil hydrogenation reaction prediction model; The optimal prediction model is selected from the plurality of residue oil hydrogenation reaction prediction models as the initial residue oil hydrogenation reaction prediction model.

4. The method according to claim 3, characterized in that Based on the first data set constructed, multiple residue oil hydrogenation reaction prediction models are established using different neural networks, including: Dividing the first data set into three parts: a first training set, a first validation set and a first test set according to a certain ratio; A neural network is selected and a corresponding neural network prediction model is constructed as a residue oil hydrogenation reaction prediction model, wherein the key features affecting the residue oil hydrogenation reaction results screened out after preprocessing the residue oil hydrogenation reaction data are used as input features for constructing the prediction model, the number of neurons in the input layer of the network matches the number of features of the key features, the number of neurons in the output layer of the network is the same as the number of indicators characterizing the residue oil hydrogenation reaction results, and the network parameters are randomly initialized; Preset network training hyperparameters, use the first training set to train the constructed residue oil hydrogenation reaction prediction model, and monitor the model training process through the first validation set, and stop model training when the model error on the first validation set no longer decreases; The prediction performance of the trained model is tested using the first test set. If the test results do not meet the set standards, the model structure and network training hyperparameters are adjusted until the model test results meet the set standards.

5. The method according to claim 4, characterized in that The neural networks used include convolutional neural networks, graph neural networks, recurrent neural networks, feedforward neural networks, and Transformer, or a combination of several of them.

6. The method according to claim 4, characterized in that Also includes: Before constructing the first data set, the collected residual oil hydrogenation reaction data is further preprocessed, including: The units of the various residual oil hydrogenation reaction data collected and evaluated by high throughput are unified; Detecting possible abnormal data in the residual oil hydrogenation reaction data, and manually correcting or eliminating the abnormal data; Fill in the missing data in the residue oil hydrogenation reaction data, and delete the data that cannot be filled; One-hot encoding the categorical features in the residue oil hydrogenation reaction data; Converting the numerical data of the residual oil hydrogenation reaction data; Construct new features that are useful for predicting the residue hydrotreating process; The key features that affect the results of the residue hydrogenation reaction are screened as input features of the constructed residue hydrogenation reaction prediction model.

7. The method according to claim 1, characterized in that Also includes: Before constructing the second data set, the collected production process data is further preprocessed, including: Unifying the units of the collected production process data; Detecting possible abnormal data in the production process data, and manually correcting or eliminating the abnormal data; Fill in the missing data in the production process data, and delete the data that cannot be filled in; The data value range of the production process data is scaled.

8. The method according to claim 1, characterized in that The step of optimizing the catalyst graded loading for the industrial residue oil hydrogenation unit to be tested by using the target residue oil hydrogenation reaction prediction model comprises: Preprocessing the production process data to be optimized to meet the input layer requirements of the target residue oil hydrogenation reaction prediction model; Randomly select multiple groups of catalyst graded loading ratios, load the target residue oil hydrogenation reaction prediction model once at a certain temperature interval within a preset reaction temperature range, input the pre-treated production process data into the model, read the model prediction results, and predict the properties of the hydrogenation product; The predicted deviation of hydrogenation product properties under each set of catalyst graded loading ratios is calculated, and the graded loading ratio with the smallest deviation of single hydrogenation product properties and the graded loading ratio with the smallest comprehensive deviation are output respectively.

9. The method according to claim 8, characterized in that The production process data include raw material properties, catalyst properties, product requirements and process conditions; The raw material properties include density, viscosity, average molecular weight, nitrogen content, sulfur content, carbon residue value, C / H content, metal impurity content, and the composition of saturates, aromatics, colloids and asphaltenes; The catalyst properties include the constituent elements of the catalyst and their mass fraction, bulk density, specific surface area, pore size, pore volume and active metal content; The product requirements include density, viscosity, nitrogen content, sulfur content, residual carbon value and metal impurity content of the hydrogenated product; the process conditions include reaction hydrogen pressure, raw material feed space velocity, and hydrogen / oil ratio.

10. A residual oil hydroprocessing catalyst graded loading optimization device, characterized in that: The method for optimizing the graded loading of residue hydroprocessing catalyst according to any one of claims 1 to 9 is adopted, wherein the device at least comprises: Dataset building blocks for: Collecting residual oil hydrogenation reaction data to construct a first data set for machine learning; and Collecting production process data of the industrial residue oil hydrogenation unit to be tested to construct a second data set; Predictive model building blocks for: Based on the first constructed data set, using deep learning technology to establish an initial residue oil hydrogenation reaction prediction model; and Based on the second data set, the parameters of the screened initial residue oil hydrogenation reaction prediction model are optimized to eliminate the systematic error factors between the industrial residue oil hydrogenation unit to be tested and the high-throughput evaluation device, and obtain a target residue oil hydrogenation reaction prediction model suitable for the industrial residue oil hydrogenation unit to be tested; Catalyst graded loading optimization module for: The target residue oil hydrogenation reaction prediction model is used to optimize the catalyst graded loading for the industrial residue oil hydrogenation unit to be tested.

Images