A product distribution prediction method and device

Abstract

The present invention provides a product distribution prediction method and device, comprising: inputting key parameters of a catalytic cracking reaction into a distribution prediction model to obtain product prediction results output by the distribution prediction model; the distribution prediction model is trained using sample key parameters and sample product distribution information corresponding to the sample key parameters; the key parameters and product prediction results are input into a target optimization model to determine product distribution information after product hydrogen distribution optimization. The product distribution prediction method and device provided by the present invention utilize the distribution prediction model and the target optimization model to perform target optimization and judgment on the key parameters and product prediction results, rapidly calculating product distribution information after hydrogen distribution optimization in a catalytic cracking reaction, and automatically calculating the optimal solution for product hydrogen content distribution.

Description

Technical Field

[0001] The present invention relates to the technical field of petroleum processing, and in particular to a product distribution prediction method and device. Background Art

[0002] The oil processing process is mainly a process of rebalancing carbon, hydrogen and other elements in crude oil, which can be divided into two cases: decarbonization and hydrogenation. The corresponding technical routes are decarbonization technology route and hydrogenation technology route.

[0003] Catalytic cracking is a key crude oil decarbonization process. Through cracking, hydrogen transfer, and isomerization, crude oil is converted into products such as dry gas, liquefied petroleum gas, gasoline, light cycle oil, slurry oil, and coke. Optimizing product distribution typically involves adjusting operating parameters and then sampling the resulting products to achieve the optimal yield of the target product.

[0004] Therefore, the above method focuses on the optimal yield after adjusting the operating parameters, rather than the optimal product hydrogen distribution in an essential sense. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the embodiments of the present invention provide a product distribution prediction method and device.

[0006] The present invention provides a product distribution prediction method, comprising:

[0007] Inputting key parameters of a catalytic cracking reaction into a distribution prediction model to obtain a product prediction result output by the distribution prediction model; the key parameters include basic device data and operating data of a catalytic cracking reaction unit; the distribution prediction model is trained using sample key parameters and sample product distribution information corresponding to the sample key parameters;

[0008] The key parameters and the product prediction results are input into a target optimization model to determine the product distribution information after the product hydrogen distribution is optimized; the target optimization model is constructed with the maximum output of liquefied gas and gasoline and the minimum hydrogen content in dry gas and coke as the optimization goals.

[0009] According to a product distribution prediction method provided by the present invention, the target optimization model includes an objective function and preset constraints; the determining of product distribution information includes:

[0010] Based on the objective function and the preset constraints, the key parameters are adjusted using the product prediction results to obtain new key parameters;

[0011] Inputting the new key parameters into the distribution prediction model to obtain new product prediction results corresponding to the new key parameters;

[0012] If the new key parameter and the new product prediction result meet the preset constraint conditions, the new product prediction result is determined to be the product distribution information.

[0013] According to a product distribution prediction method provided by the present invention, after adjusting the key parameters using the product prediction results to obtain new key parameters, the method further includes:

[0014] If the new key parameters and the new product prediction results do not meet the preset constraints, the key parameters are adjusted according to the objective function and the preset constraints until the new key parameters and the new product prediction results meet the preset constraints, and the new product prediction results are determined to be the product distribution information.

[0015] According to a product distribution prediction method provided by the present invention, the distribution prediction model includes a low-carbon hydrocarbon prediction model and a liquid oil prediction model; the product prediction results include a low-carbon hydrocarbon product prediction result and a liquid oil product prediction result;

[0016] Inputting the key parameters of the cracking catalytic reaction into the distribution prediction model and obtaining the product prediction results output by the distribution prediction model includes:

[0017] Inputting the key parameters into a low-carbon hydrocarbon prediction model to obtain the low-carbon hydrocarbon product prediction result output by the low-carbon hydrocarbon prediction model;

[0018] The key parameters are input into the liquid oil prediction model to obtain the liquid oil product prediction result output by the liquid oil prediction model.

[0019] According to a product distribution prediction method provided by the present invention, the distribution prediction model is constructed based on the following steps:

[0020] Build initial prediction models for low-carbon hydrocarbons and liquid oil based on data-driven models, and obtain key parameters for multiple samples;

[0021] A combination of a sample key parameter and sample light hydrocarbon product distribution information corresponding to the sample key parameter is used as a light hydrocarbon training sample to obtain a plurality of light hydrocarbon training samples;

[0022] Taking a combination of a sample key parameter and sample liquid oil product distribution information corresponding to the sample key parameter as a liquid oil training sample, and obtaining a plurality of liquid oil training samples;

[0023] Using the low-carbon hydrocarbon training samples to train the low-carbon hydrocarbon initial prediction model to obtain the low-carbon hydrocarbon prediction model;

[0024] The liquid-phase oil initial prediction model is trained using the liquid-phase oil training samples to obtain the liquid-phase oil prediction model.

[0025] According to a product distribution prediction method provided by the present invention, after obtaining the low-carbon hydrocarbon prediction model, the method further includes:

[0026] When the accuracy of the low-carbon hydrocarbon prediction model is less than an accuracy threshold, determining a supplementary parameter from a data warehouse, wherein a correlation between the supplementary parameter and the hydrogen content of the low-carbon hydrocarbon is greater than a correlation threshold;

[0027] Acquiring liquefied gas gas chromatography data corresponding to the supplementary parameter, and determining the hydrogen content of the liquefied gas corresponding to the supplementary parameter based on the liquefied gas gas chromatography data corresponding to the supplementary parameter;

[0028] Based on the supplementary parameters and the hydrogen content of the liquefied gas corresponding to the supplementary parameters, the low-carbon hydrocarbon prediction model is updated.

[0029] According to a product distribution prediction method provided by the present invention, the data warehouse is established based on the following steps:

[0030] Obtain historical raw material property parameters and historical operating parameters;

[0031] Data cleaning is performed on the historical raw material property parameters and the historical operation parameters, and the cleaned historical raw material property parameters and historical operation parameters are added to the data warehouse.

[0032] The present invention also provides a product distribution prediction device, comprising: an acquisition module for inputting key parameters of a cracking catalytic reaction into a distribution prediction model to obtain product prediction results output by the distribution prediction model; the key parameters include basic device data and operating data of a catalytic cracking reaction device; the distribution prediction model is trained using sample key parameters and sample product distribution information corresponding to the sample key parameters;

[0033] A determination module is used to input the key parameters and the product prediction results into a target optimization model to determine the product distribution information after the product hydrogen distribution is optimized; the target optimization model is constructed with the maximum output of liquefied gas and gasoline and the minimum hydrogen content in dry gas and coke as the optimization goals.

[0034] The present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described product distribution prediction methods are implemented.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described product distribution prediction methods.

[0036] The product distribution prediction method and device provided by the present invention utilize a distribution prediction model and a target optimization model to perform target optimization and judgment on key parameters and product prediction results, rapidly measure the product distribution information after the product hydrogen distribution in the catalytic cracking reaction is optimized, and realize automatic calculation of the optimal solution for the product hydrogen content distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 It is a flow chart of the product distribution prediction method provided by the present invention;

[0039] Figure 2 This is a schematic diagram of the structure of the low-carbon hydrocarbon product flow rate and hydrogen content prediction system provided by the present invention;

[0040] Figure 3 It is a structural schematic diagram of the liquid oil product flow rate and hydrogen content prediction system provided by the present invention;

[0041] Figure 4 It is a structural schematic diagram of the product distribution prediction device provided by the present invention;

[0042] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] It should be noted that, in the description of the embodiments of the present invention, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising 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 "include a ..." do not exclude the presence of other identical elements in the process, method, article or device comprising the elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0045] The existing method for calculating the hydrogen content of products from a commonly used catalytic cracking reaction unit is to collect a certain amount of sample and then test its hydrogen content using an elemental analysis instrument.

[0046] Hydrogen balance calculation and analysis can be used to evaluate the rationality of product distribution and hydrogen utilization efficiency in catalytic cracking units. In actual production, composition analysis of dry gas and liquefied gas is widely used, and their hydrogen content can be calculated from their composition. The hydrogen content of coke can be calculated from the flue gas composition. The hydrogen content of liquid oil products is primarily measured using elemental analyzers, empirical formulas, and simulation software.

[0047] Regarding the hydrogen content of catalytic cracking reaction unit products, the existing calculation methods have the following main characteristics:

[0048] (1) It relies on sampling and cannot predict the hydrogen content of the device product without sampling. Taking the calculation of gasoline hydrogen content as an example, after sampling a certain amount of gasoline sample, its hydrogen content can be tested by elemental analysis instrument, or the key physical and chemical properties of the gasoline sample can be analyzed and tested first, and then its hydrogen content can be calculated by empirical formula. Therefore, the existing analysis method cannot predict the hydrogen content of gasoline without sampling;

[0049] (2) It is highly dependent on instrumental analysis. Dry gas, liquefied gas, flue gas, etc. need to be analyzed by experimental instruments to obtain component distribution before their hydrogen content can be calculated. Liquid oil products also need to be analyzed by professional elemental analysis equipment to calculate their hydrogen content; (3) The hydrogen content measurement process is only related to the basic properties of the sample and is not related to the main operating parameters of the device. The hydrogen content or hydrogen balance of the device obtained by the existing method can only indirectly reflect the operation of the device.

[0050] The present invention provides a product distribution acquisition method and device suitable for catalytic cracking product distribution prediction and optimization, which is used to analyze and evaluate the product distribution and operation rationality of the device, and can also guide the optimization of the device operation.

[0051] The following combination Figures 1 to 5The product distribution prediction method and device provided by the embodiments of the present invention are described.

[0052] Figure 1 It is a flow chart of the product distribution prediction method provided by the present invention, such as Figure 1 As shown, including but not limited to the following steps:

[0053] First, in step S1, the key parameters of the cracking catalytic reaction are input into the distribution prediction model to obtain the product prediction results output by the distribution prediction model; the key parameters include the basic device data and operation data of the catalytic cracking reaction unit; the distribution prediction model is trained with sample key parameters and sample product distribution information corresponding to the sample key parameters.

[0054] The distribution prediction model is constructed based on a data-driven model, and the distribution prediction model is trained based on sample key parameters and sample product distribution information corresponding to the sample key parameters.

[0055] The key parameters include basic device data and operation data of the catalytic cracking reaction device, wherein the operation data may include: raw material data, product data and device operating parameters of the catalytic cracking reaction device.

[0056] The product prediction results may include hydrogen content distribution data in the products of the catalytic cracking reactor and product flow data, wherein the product flow data must satisfy the material balance.

[0057] Specifically, the key parameters of the catalytic cracking reaction are input into the distribution prediction model, which then makes predictions based on the basic data and operating data of the device to obtain product prediction results corresponding to the key parameters. The product prediction results include predictions of the flow rate and hydrogen content of low-carbon hydrocarbon products, the flow rate and hydrogen content of liquid oil products, and the flow rate and hydrogen content of coke products. Low-carbon hydrocarbons include liquefied gas and dry gas, and liquid oil includes gasoline, light cycle oil, and slurry oil. After the distribution prediction model predicts the product flow rate and hydrogen content of low-carbon hydrocarbons and liquid oil, the hydrogen balance and material balance are used to subtract the coke product flow rate and hydrogen content.

[0058] Furthermore, in step S2, the key parameters and the product prediction results are input into a target optimization model to determine product distribution information after product hydrogen distribution optimization; the target optimization model is constructed with the maximum output of liquefied gas and gasoline and the minimum hydrogen content in dry gas and coke as the optimization goals.

[0059] The target optimization model includes: objective function and preset constraints; the objective function may include: maximum production function of liquefied gas and gasoline, and minimum hydrogen content function in dry gas and coke; the preset constraints may include: balance constraints and other constraints, and other constraints may include: content constraints, temperature constraints, flow constraints and product hydrogen content constraints.

[0060] Specifically, the preset constraints are used to judge the key parameters and product prediction results. When the key parameters and product prediction results meet the preset constraints, the objective function is used to calculate the target values ​​corresponding to the key parameters. The target values ​​include: the maximum output of liquefied gas and gasoline, and the minimum hydrogen content in dry gas and coke.

[0061] The key parameters can be adjusted within the preset constraints and multiple target values ​​can be obtained. Based on the multiple target values, the product prediction results corresponding to the target values ​​with the maximum liquefied gas and gasoline production and the minimum hydrogen content in dry gas and coke are determined as the final product distribution information.

[0062] The product distribution prediction method provided by the present invention utilizes a distribution prediction model and a target optimization model to perform target optimization and judgment on key parameters and product prediction results, rapidly measures product distribution information after hydrogen distribution optimization in catalytic cracking reactions, and realizes automatic calculation of the optimal solution for product hydrogen content distribution.

[0063] Optionally, the key parameters are retrieved from a plant data center, which is established based on the following steps: acquiring basic data of the device and operating data of the catalytic cracking reaction device under different operating conditions, and building the plant data center;

[0064] The basic data of the device include technical design data of the catalytic cracking reaction device;

[0065] The operation data includes: raw material data, product data and device operating parameters of the catalytic cracking reaction device.

[0066] Collect basic data on the catalytic cracking reactor and its corresponding operating data under different operating conditions to establish a plant data center. Different operating conditions can be divided according to different processed raw materials, reactor temperatures, and reactor outlet pressures.

[0067] Among them, the basic data of the device may include basic design data of the device, such as device processing scale, reactor structural parameters, basic properties of the catalyst, tower equipment data, heat exchanger equipment data, etc.

[0068] Operation data may include the device operating parameters of the catalytic cracking reaction unit, such as temperature, pressure, residence time, etc.; it may also include raw material data and product data. Raw material data includes the laboratory analysis data of the raw materials, and product data includes the laboratory analysis data of the products, such as density, distillation range, sulfur content, nitrogen content, hydrogen content, etc.

[0069] Taking the refinery's 2.0Mt / a catalytic cracking reactor as an example, the reaction and regeneration type of the unit is a high-low parallel type; the reaction part adopts the clean gasoline production technology (MIP-CGP) process technology of increasing the production of propylene and producing more isomerized alkanes through the internal riser, and the regeneration part adopts a parallel two-vessel regeneration technology; the first regenerator adopts incomplete regeneration technology and is equipped with two sets of external heat exchangers; the second regenerator adopts complete regeneration technology and is equipped with a regenerated catalyst degassing tank.

[0070] During the operation of the device, the flue gases from the first regenerator and the second regenerator are mixed in the flue and supplemented with air to cause the CO to burn. The high-temperature flue gas is heated by the high-temperature heat collector and then sent to the third-stage cyclone separator after cooling.

[0071] Furthermore, historical data was organized and categorized into several operating conditions based on the unit's raw material processing capacity, slag blending amount, recycled oil volume, reactor outlet temperature, and reactor outlet pressure. Key parameters related to one of these operating conditions are described below, while the main operating parameters for the separation process are omitted. Table 1 shows the main properties of the feedstock oil, which is included in the raw material data within the operational data.

[0072] Table 1 Main properties of crude oil

[0073] project Numerical project Numerical <![CDATA[Density (20 °C), kg / m 3 > 903.7 Distillation range, ℃ Residual carbon value, % 4.08 Initial distillation point 180.9 <![CDATA[Viscosity (80 °C), mm 2 / s]]> 34.32 5％ 271.5 Relative molecular mass 509 10％ 357.5 Element mass fraction, % 20％ 412.9 C 86.78 30％ 446.6 H 12.91 40％ 494.9 S 0.28 50％ 540.4 N 0.18 60％ 577.8

[0074] Table 2 shows the basic properties of the catalyst, which are part of the basic data of the device. Table 3 shows the operating parameters of the device.

[0075] Table 2 Basic properties of catalysts

[0076] project Fresh catalyst Regeneration catalyst Chemical composition, % <![CDATA[(La2O3 / CeO2)RE2O3]]> 0.75 / 1.5 2.7 <![CDATA[Al2O3]]> 56.1 52.7 <![CDATA[Na2O]]> 0.19 0.33 <![CDATA[SiO2]]> 37.0 38.9 <![CDATA[P2O5]]> 0.418 0.462 <![CDATA[SO3]]> 2.01 0.127 <![CDATA[K2O]]> 0.312 0.269 <![CDATA[TiO2]]> 0.133 0.174 Physical properties <![CDATA[Total specific surface area, m 2 / g]]> 258 91 <![CDATA[Specific surface area of the substrate, m 2 / g]]> 81 45 <![CDATA[Micropore specific surface area, m 2 / g]]> 177 47 Total pore volume, mL / g 0.179 0.158 Micropore volume, mL / g 0.082 0.022 <![CDATA[Bulk density, g / cm 3 > 0.74 0.8 Carbon content, % - 0.03

[0077] Table 3 Device operating parameters

[0078]

[0079]

[0080] Optionally, the data warehouse is established based on the following steps:

[0081] Obtain historical raw material property parameters and historical operating parameters;

[0082] Data cleaning is performed on the historical raw material property parameters and the historical operation parameters, and the cleaned historical raw material property parameters and historical operation parameters are added to the data warehouse.

[0083] Optionally, the distribution prediction model includes a low-carbon hydrocarbon prediction model and a liquid oil prediction model; the product prediction results include a low-carbon hydrocarbon product prediction result and a liquid oil product prediction result;

[0084] Inputting the key parameters of the cracking catalytic reaction into the distribution prediction model and obtaining the product prediction results output by the distribution prediction model includes:

[0085] Inputting the key parameters into a low-carbon hydrocarbon prediction model to obtain the low-carbon hydrocarbon product prediction result output by the low-carbon hydrocarbon prediction model;

[0086] The key parameters are input into the liquid oil prediction model to obtain the liquid oil product prediction result output by the liquid oil prediction model.

[0087] Optionally, the distribution prediction model is constructed based on the following steps:

[0088] Build initial prediction models for low-carbon hydrocarbons and liquid oil based on data-driven models, and obtain key parameters for multiple samples;

[0089] A combination of a sample key parameter and sample light hydrocarbon product distribution information corresponding to the sample key parameter is used as a light hydrocarbon training sample to obtain a plurality of light hydrocarbon training samples;

[0090] Taking a combination of a sample key parameter and sample liquid oil product distribution information corresponding to the sample key parameter as a liquid oil training sample, and obtaining a plurality of liquid oil training samples;

[0091] Using the low-carbon hydrocarbon training samples to train the low-carbon hydrocarbon initial prediction model to obtain the low-carbon hydrocarbon prediction model;

[0092] The liquid-phase oil initial prediction model is trained using the liquid-phase oil training samples to obtain the liquid-phase oil prediction model.

[0093] Optionally, after obtaining the low-carbon hydrocarbon prediction model, the method further includes:

[0094] When the accuracy of the low-carbon hydrocarbon prediction model is less than an accuracy threshold, determining a supplementary parameter from a data warehouse, wherein a correlation between the supplementary parameter and the hydrogen content of the low-carbon hydrocarbon is greater than a correlation threshold;

[0095] Acquiring liquefied gas gas chromatography data corresponding to the supplementary parameter, and determining the hydrogen content of the liquefied gas corresponding to the supplementary parameter based on the liquefied gas gas chromatography data corresponding to the supplementary parameter;

[0096] Based on the supplementary parameters and the hydrogen content of the liquefied gas corresponding to the supplementary parameters, the low-carbon hydrocarbon prediction model is updated.

[0097] Figure 2 This is a schematic diagram of the structure of the low-carbon hydrocarbon product flow and hydrogen content prediction system provided by the present invention. Figure 2 As shown, the low-carbon hydrocarbon product flow rate and hydrogen content prediction system is constructed based on a mathematical model for predicting low-carbon hydrocarbon product flow rate and hydrogen content. The low-carbon hydrocarbon prediction model is a mathematical model for predicting low-carbon hydrocarbon product flow rate and hydrogen content based on catalytic dry gas and liquefied petroleum gas. The system includes a data warehouse module, a low-carbon hydrocarbon data extraction module, a low-carbon hydrocarbon feature selection module, a low-carbon hydrocarbon model training module, and a low-carbon hydrocarbon prediction module.

[0098] First, the data warehouse module is used to build a data warehouse. That is, from the historical working conditions, unit operating parameters, physical and chemical properties analysis reports of crude oil and main products of large-scale oil refining enterprises, historical raw material data and historical unit operating parameters are extracted from the plant principle laboratory physical properties and plant distributed control system (DCS) monitoring data. After data cleaning, corresponding MySQL databases and data tables are established for each production unit, and a CSV cache mechanism is implemented to achieve unified storage, management and call of large-scale data.

[0099] Then, the low-carbon hydrocarbon feature selection module performs feature selection on the parameters in the data warehouse, selects the key parameters that affect the product flow and hydrogen content, and calculates the low-carbon hydrocarbon product flow data and hydrogen content distribution data corresponding to the key parameters based on the low-carbon hydrocarbon gas chromatography data through the low-carbon hydrocarbon data extraction module, so that the low-carbon hydrocarbon model training module can be based on the key parameters and the corresponding key parameters.

[0100] The low-carbon hydrocarbon prediction model is obtained by training the low-carbon hydrocarbon product flow data and hydrogen content distribution data. After obtaining the low-carbon hydrocarbon prediction model, the low-carbon hydrocarbon prediction module can predict the low-carbon hydrocarbon product flow data and hydrogen content data based on the low-carbon hydrocarbon prediction model. Among them, the low-carbon hydrocarbon feature selection module can be a data-driven model. Based on the chemical reaction mechanism related to the petroleum refining process, the operating experience of on-site personnel, and the main control loop of the device, the low-carbon hydrocarbon feature selection module can extract the main characteristic variables for constructing the low-carbon hydrocarbon prediction model, and can also screen the position number according to the process, and can also recommend the characteristic position number by algorithm.

[0101] Among them, the initial main characteristic variables can be manually selected from the model database based on experience. When the modeling accuracy of the low-carbon hydrocarbon prediction model is insufficient, the low-carbon hydrocarbon feature selection module can select supplementary characteristic variables with a high correlation with product distribution information from the data warehouse.

[0102] The low-carbon hydrocarbon model training module uses the hydrogen content samples obtained from the routine monitoring reports of catalytic cracking dry gas and liquefied gas as the dependent variable for model training, and constructs a low-carbon hydrocarbon prediction model for the hydrogen content of dry gas and liquefied gas.

[0103] The low-carbon hydrocarbon product flow and hydrogen content prediction system can also include a low-carbon hydrocarbon model evaluation module. The low-carbon hydrocarbon model evaluation module can select hyperparameters of the eXtreme Gradient Boosting (XGBoost) model based on the cross-validation results, evaluate the model's predictive ability based on its performance on the validation set, and achieve model parameter optimization and model evaluation.

[0104] XGBoost was used to build an initial prediction model for low-carbon hydrocarbons, and the grid search method was used to adjust the original parameters of the model. The principle and model structure of XGBoost are as follows:

[0105] XGBoost is a boosting method, which is an additive model composed of k base models. Assume that the tree model to be trained in the tth iteration is f t (x), then:

[0106]

[0107] in, represents the prediction result of sample i after the tth iteration, is the result of the first t-1 trees, f t (x i ) is the model of the t-th tree.

[0108] The objective function of the XGBoost model is as follows:

[0109]

[0110] Among them, l′ is the loss function, Ω(f t ) is the regularization term, and constant is the constant term. After Taylor expansion approximation and omitting the constant loss function in each iteration, the objective function can be simplified to:

[0111]

[0112] in They represent the first-order partial derivative and second-order partial derivative of the loss function in the previous iteration respectively.

[0113] In addition, the embodiment of the present invention can select the hyperparameters of the XGBoost model based on the cross-validation results; and can evaluate the predictive ability of the model based on the performance of the model on the validation set.

[0114] The cross-validation script designed based on Python can directly isolate 20% of the original data as a test set. Model selection and parameter adjustment are performed in the training set, and then the k-fold cross-validation method is used to divide the training set and the validation set.

[0115] During the XGBoost training process, the grid search method is used. Based on the principle of minimizing the objective function Obj, each parameter to be adjusted is adjusted within a certain range. The optimal model parameters are determined by the principle of rough adjustment first and then fine adjustment.

[0116] The Python-based model evaluation script can evaluate the quality of the model based on its performance on the validation set. The main model evaluation indicators used include but are not limited to:

[0117] Root mean square error:

[0118] Mean absolute error:

[0119] Coefficient of determination:

[0120] The final low-carbon hydrocarbon prediction model can only be output when the model evaluation indicators meet the requirements. Otherwise, the hyperparameters of the XGBoost model will be updated and iterated again until the evaluation indicators meet the requirements.

[0121] Figure 3 This is a schematic diagram of the structure of the liquid oil product flow rate and hydrogen content prediction system provided by the present invention. Figure 3 As shown in Figure 1, the liquid oil product flow rate and hydrogen content prediction system is built based on a mathematical model for predicting liquid oil product flow rate and hydrogen content. The liquid oil prediction model is a mathematical model for predicting liquid oil product flow rate and hydrogen content for gasoline, light cycle oil, and slurry oil. The system includes a plant data center, a low-carbon hydrocarbon data extraction module, a liquid oil feature selection module, a liquid oil model training module, and a liquid oil prediction module.

[0122] In the factory data center, historical raw material data and historical device operating parameters are extracted from the factory principle laboratory physical properties and factory DCS monitoring data for data collection. Using database building scripts or Kettle tools, routine laboratory analysis data is screened and data is extracted to quickly establish calculation formulas for the hydrogen content of catalytic gasoline, light cycle oil, slurry oil, etc.

[0123] Then, the liquid oil feature selection module performs feature selection on the parameters in the data warehouse, selects key parameters that affect product flow and hydrogen content, and calculates the liquid oil product flow data and hydrogen content distribution data corresponding to the key parameters through the liquid oil data extraction module based on the conventional liquid oil test data. Thus, the liquid oil model training module can train the liquid oil prediction model based on the key parameters and the liquid oil product flow data and hydrogen content distribution data corresponding to the key parameters. After obtaining the liquid oil prediction model, the liquid oil prediction module can predict the liquid oil product flow data and hydrogen content data based on the liquid oil prediction model. Among them, the liquid oil feature selection module can be a data-driven model. Based on the chemical reaction mechanism related to the petroleum refining process, the operating experience of on-site personnel, and the main control loop of the device, the liquid oil feature selection module can extract the main characteristic variables for constructing the liquid oil prediction model.

[0124] Among them, a liquid oil feature selection module is constructed based on the rapid calculation of the hydrogen content of liquid oil products in catalytic cracking reactions to form sample points. It can select characteristic variables and key parameters related to the prediction of hydrogen content in catalytic cracking liquid oil products, including raw material properties, catalytic cracking liquid oil properties, and device operating parameters. It can also screen position numbers according to the process, and can also recommend characteristic position numbers through algorithms.

[0125] The liquid oil product flow and hydrogen content prediction system can also include a parameter optimization and model evaluation module, which can select hyperparameters of the Gradient Boosting Decision Tree (GBDT) model based on cross-validation results and evaluate the model's predictive ability based on its performance on the validation set.

[0126] GBDT is a boosting algorithm based on the Classification and Regression Trees (CART) model. GBDT requires minimizing the loss function of this round in each iteration.

[0127] For regression problems, it is an additive model for binary regression trees:

[0128] f M (x)=∑ M T(x;Θ m );

[0129] Where, T(x; Θ m ) represents a decision tree; Θ m is the parameter of the decision tree; M is the number of trees.

[0130] There is no weight set between boosted trees, and they are independent of each other. The forward distribution propagation process of the boosted tree is as follows:

[0131] First determine the initial boosting tree f0(x) = 0;

[0132] The model for step m is: f m (x) = f m-1 (x)+T(x;Θ m );

[0133] Among them, f m-1 (x) is the current model, and the parameter Θ of the next decision tree is determined by minimizing the empirical risk m :

[0134]

[0135] The training method of gradient boosting tree is as follows:

[0136] The input training data set is set as T = {(x1,y1),(x2,y2),…,(x N ,y N )};

[0137] in Maximum number of iterations M; loss function L(y,f(x)).

[0138] Regarding the output, the regression tree is obtained as follows.

[0139] First, initialize the model:

[0140]

[0141] Furthermore, for m=1,2,…,M, the calculation is as follows:

[0142] (1) For i = 1, 2, ..., N, calculate:

[0143]

[0144] (2) For r mi Fit the regression tree and get the leaf node area R of the mth tree mj j=1,2,…,J;

[0145] (3) For j = 1, 2, ..., J, calculate:

[0146]

[0147] (4) Update

[0148] Furthermore, the regression tree can be obtained:

[0149]

[0150] At this point, the GBDT model is built.

[0151] The cross-validation script designed based on Python can directly isolate 20% of the original data as a test set. Model selection and parameter adjustment are performed in the training set, and then the k-fold cross-validation method is used to divide the training set and the validation set.

[0152] During the GBDT training process, the grid search method is used. Based on the principle of minimizing the objective function Obj, each parameter to be adjusted is adjusted within a certain range. The optimal model parameters are determined by the principle of rough adjustment first and then fine adjustment.

[0153] The Python-based model evaluation script can evaluate the quality of the model based on its performance on the validation set. The main model evaluation indicators used include but are not limited to:

[0154] Root mean square error:

[0155] Mean absolute error:

[0156] Coefficient of determination:

[0157] Only when the model evaluation index meets the requirements can the final liquid oil prediction model be output. Otherwise, the hyperparameters of the GBDT model will be updated and iterated again until the evaluation index meets the requirements.

[0158] The trained low-carbon hydrocarbon prediction model can predict the product flow rate and hydrogen content distribution in low-carbon hydrocarbons, and the trained liquid oil prediction model can predict the product flow rate and hydrogen content distribution in liquid oil, thereby avoiding the problem of low efficiency caused by the need to collect low-carbon hydrocarbon samples or liquid oil samples for testing in traditional methods. Moreover, the embodiment of the present invention can accurately obtain product prediction results without high-precision instruments, that is, the embodiment of the present invention does not need to rely on analytical instruments for prediction. At the same time, after obtaining the product prediction results, if there are abnormalities in the product prediction results, the corresponding raw material property parameters and operating parameters can be directly queried, so as to facilitate the determination of the cause of the abnormality based on the raw material property parameters and operating parameters.

[0159] Optionally, the target optimization model includes an objective function and preset constraints;

[0160] The product distribution information after the product hydrogen distribution is optimized is determined, including:

[0161] Based on the objective function and the preset constraints, the key parameters are adjusted using the product prediction results to obtain new key parameters;

[0162] Inputting the new key parameters into the distribution prediction model to obtain new product prediction results corresponding to the new key parameters;

[0163] If the new key parameter and the new product prediction result meet the preset constraint conditions, the new product prediction result is determined to be the product distribution information.

[0164] Optionally, after adjusting the key parameters using the product prediction results to obtain new key parameters, the method further includes:

[0165] If the new key parameters and the new product prediction results do not meet the preset constraints, the key parameters are adjusted according to the objective function and the preset constraints until the new key parameters and the new product prediction results meet the preset constraints, and the new product prediction results are determined to be the product distribution information.

[0166] First, a multi-objective optimization mathematical model for the product distribution of a catalytic cracking reactor was established using Matlab software, aiming to optimize hydrogen distribution and maximize target product yields. This model served as the target optimization model. The optimization objectives included maximizing liquefied petroleum gas and gasoline production and minimizing hydrogen content in dry gas and coke. A multi-objective optimization algorithm was then applied, linking the target optimization model with a coupled mathematical model for the catalytic cracking reaction and separation. The target optimization model included an objective function and pre-set constraints.

[0167] Among them, the optimization algorithm can be a fuzzy optimization algorithm, a multi-objective queue competition algorithm, a genetic algorithm, a neural network algorithm, etc.

[0168] The optimization objectives of the target optimization model are to optimize the hydrogen distribution of the catalytic cracking reactor products and maximize the target product yield. Optimal hydrogen distribution means minimizing the hydrogen content in dry gas and coke, while maximizing the target product yield means maximizing the yield of liquefied gas and gasoline.

[0169] The objective functions include: maximizing the output of liquefied gas and gasoline, and minimizing the hydrogen content in dry gas and coke.

[0170] Among them, the minimum function of hydrogen content in dry gas and coke is:

[0171] miny1=(F 干气 x 干气,H +F 焦炭 x 焦炭,H ) / (F 原料 x 原料,H );

[0172] The maximum function of liquefied gas and gasoline production is:

[0173] maxy2=(F液化气 +F 汽油 ) / F 原料 ;

[0174] The preset constraints include equilibrium constraints and other constraints.

[0175] Among them, the balance constraints include:

[0176] F 原料 =F 干气 +F 液化气 +F 汽油 +F 轻循环油 +F 焦炭 ;

[0177] F 原料 x 原料,H =F 干气 x 干气,H +F 液化气 x 液化气,H +F 汽油 x 汽油,H +

[0178] F 轻循环油 x 轻循环油,H +F 油浆 x 油浆,H +F 焦炭 x 焦炭,H ;

[0179] Other constraints include:

[0180] C3+ content constraint in dry gas, with the constraint condition being that the volume fraction of C3+ light hydrocarbons is ≤ 3%;

[0181] The C2 content in the liquefied gas is constrained, with the C2 volume fraction being ≤ 0.4%;

[0182] The C5 content in liquefied gas is constrained, with the C5 volume fraction being ≤1%;

[0183] Gasoline ASTM D86 dry point constraint, constraint conditions are 200~204℃;

[0184] Reactor outlet temperature constraint, 480℃≤t 反应器出口 ≤520℃;

[0185] Reaction pressure constraint, 0.25MPa≤P 反应 ≤0.40MPa;

[0186] Stable gasoline circulation rate, 25-45t / h;

[0187] Reabsorbent flow rate of reabsorption tower, 30-60t / h;

[0188] Reabsorbent temperature in reabsorption tower: 30-40℃;

[0189] Product hydrogen content constraints:

[0190] (a) The hydrogen content in dry gas and coke is less than that in liquid oil products:

[0191] F 干气 x 干气,H +F 焦炭 x 焦炭,H <F 液化气 x 液化气,H +F 汽油 x 汽油,H +

[0192] F 轻循环油 x 轻循环油,H +F 油浆 x 油浆,H ;

[0193] (b) The hydrogen content in the liquefied gas is greater than that in the slurry oil:

[0194] F 液化气 x 液化气,H >F 油浆 x 油浆,H ;

[0195] Among them, y1 is optimization target 1; y2 is optimization target 2; F 干气 is the dry gas flow rate, in t / h; x 干气,H is the mass fraction of hydrogen in dry gas, in %; F 液化气 is the liquefied gas flow rate, in t / h; x 液化气,H is the mass fraction of hydrogen in liquefied gas, in %; F 汽油 is the gasoline flow rate, in t / h; x 汽油,H is the mass fraction of hydrogen in gasoline, in %; F 轻循环油 is the light circulating oil flow rate, in t / h; x 轻循环油,H is the mass fraction of hydrogen in light cycle oil, in %; F 油浆 is the oil slurry flow rate, in t / h; x 油浆,H is the mass fraction of hydrogen in the oil slurry, in %; F 焦炭 is the coke flow rate, in t / h; x 焦炭 , H is the mass fraction of hydrogen in coke, in %; CT is the residual carbon in raw materials, in %; t 反应器出口 is the outlet temperature of the reactor (first reactor, second reactor), in °C; P 反应 is the reaction pressure, in MPa.

[0196] For key parameters, under certain raw material data and unit operating parameters, the distribution prediction model can simulate and calculate the product flow data, hydrogen content distribution data, and other calculation results of the catalytic cracking reaction unit, and output them as product prediction results. However, it cannot be concluded that the calculation results are optimal. These data need to be transmitted to the target optimization model. Through the target optimization model, calculations can be performed using the objective function and the formula of preset constraints:

[0197] If it is found that there is any calculation result that does not meet the preset constraints under the key parameters, then within the preset constraints, the reactor outlet temperature, pressure and other device operating parameters are adjusted to obtain new key parameters, and the new key parameters are returned to the distribution prediction model for simulation calculation to obtain new product flow data, hydrogen content distribution data, etc. This logic will eventually lead to the optimal distribution of operating parameter solutions within the constraints.

[0198] If the calculation results also meet the constraints, the reactor temperature, pressure and other operating parameters will still be adjusted, and the simulation calculation, optimization model calculation and other steps will be repeated. By calling MATLAB's own multi-objective optimization algorithm model, the best operating condition configuration plan that meets the objective function requirements can be searched, such as temperature, pressure, etc.

[0199] According to the hydrogen content distribution acquisition method provided by the present invention, by associating the distribution prediction model with the target optimization model, the hydrogen balance calculation and analysis in the target optimization model can evaluate the rationality of the device operation. The distribution prediction model can also be associated with the online monitoring system to realize real-time online measurement of the catalytic cracking product distribution.

[0200] Optionally, after determining the product distribution information, the method for obtaining hydrogen content distribution further includes:

[0201] A product optimization plan for the catalytic cracking reaction unit is determined based on the operating condition data and the product distribution information.

[0202] The product distribution optimization model of the catalytic cracking reaction unit developed by the application is used to calculate the main operating conditions under different operating conditions, obtain the optimal solution for the product distribution under the corresponding operating condition data, and use the combination of the operating condition data and the product distribution information corresponding to the operating condition data as the product optimization plan for the catalytic cracking reaction unit.

[0203] The product optimization plan includes: product flow data that satisfies material balance, such as flow rate; and hydrogen content distribution data, such as the hydrogen content of each product, which also satisfies the balance of hydrogen content entering and exiting the unit. Table 4 shows the calculated results of the product optimization plan. As shown in Table 4, when the feedstock is fixed, the target optimization model for catalytic cracking product distribution is solved to obtain the following product optimization plan. This means that under optimized unit operation, the unit product distribution is improved, and more hydrogen resources are transferred to the target products.

[0204] Table 4 Calculation results of product optimization plan

[0205]

[0206] The hydrogen content distribution acquisition method provided by the present invention can calculate the distribution changes of the catalytic cracking reaction device products only through the device raw material properties and main operating parameters without relying on product sampling and analysis, and thus obtain the corresponding product optimization plan.

[0207] Figure 4 This is a schematic diagram of the structure of the product distribution prediction device provided by the present invention. Figure 4 As shown, including:

[0208] Acquisition module 401 is used to input key parameters of the catalytic cracking reaction into a distribution prediction model to obtain product prediction results output by the distribution prediction model; the key parameters include basic device data and operation data of the catalytic cracking reaction unit; the distribution prediction model is trained based on sample key parameters and sample product distribution information corresponding to the sample key parameters;

[0209] Determination module 402 is used to input the key parameters and the product prediction results into the target optimization model to determine the product distribution information after the product hydrogen distribution is optimized; the target optimization model is constructed with the maximum output of liquefied gas and gasoline and the minimum hydrogen content in dry gas and coke as the optimization goals.

[0210] First, the acquisition module 401 inputs the key parameters of the cracking catalytic reaction into the distribution prediction model to obtain the product prediction results output by the distribution prediction model; the key parameters include the basic device data and operation data of the catalytic cracking reaction unit; the distribution prediction model is trained with sample key parameters and sample product distribution information corresponding to the sample key parameters.

[0211] The distribution prediction model is constructed based on a data-driven model, and the distribution prediction model is trained based on sample key parameters and sample product distribution information corresponding to the sample key parameters.

[0212] The key parameters include basic device data and operation data of the catalytic cracking reaction device, wherein the operation data may include: raw material data, product data and device operating parameters of the catalytic cracking reaction device.

[0213] The product prediction results may include hydrogen content distribution data in the products of the catalytic cracking reactor and product flow data, wherein the product flow data must satisfy the material balance.

[0214] Specifically, the key parameters of the catalytic cracking reaction are input into the distribution prediction model, which then makes predictions based on the basic data and operating data of the device to obtain product prediction results corresponding to the key parameters. The product prediction results include predictions of the flow rate and hydrogen content of low-carbon hydrocarbon products, the flow rate and hydrogen content of liquid oil products, and the flow rate and hydrogen content of coke products. Low-carbon hydrocarbons include liquefied gas and dry gas, and liquid oil includes gasoline, light cycle oil, and slurry oil. After the distribution prediction model predicts the product flow rate and hydrogen content of low-carbon hydrocarbons and liquid oil, the hydrogen balance and material balance are used to subtract the coke product flow rate and hydrogen content.

[0215] Furthermore, the determination module 402 inputs the key parameters and the product prediction results into the target optimization model to determine the product distribution information after the product hydrogen distribution is optimized; the target optimization model is constructed with the maximum output of liquefied gas and gasoline and the minimum hydrogen content in dry gas and coke as the optimization goals.

[0216] The target optimization model includes: objective function and preset constraints; the objective function may include: maximum production function of liquefied gas and gasoline, and minimum hydrogen content function in dry gas and coke; the preset constraints may include: balance constraints and other constraints, and other constraints may include: content constraints, temperature constraints, flow constraints and product hydrogen content constraints.

[0217] Specifically, the preset constraints are used to judge the key parameters and product prediction results. When the key parameters and product prediction results meet the preset constraints, the objective function is used to calculate the target values ​​corresponding to the key parameters. The target values ​​include: the maximum output of liquefied gas and gasoline, and the minimum hydrogen content in dry gas and coke.

[0218] The key parameters can be adjusted within the preset constraints and multiple target values ​​can be obtained. Based on the multiple target values, the product prediction results corresponding to the target values ​​with the maximum liquefied gas and gasoline production and the minimum hydrogen content in dry gas and coke are determined as the final product distribution information.

[0219] The product distribution prediction device provided by the present invention uses a distribution prediction model and a target optimization model to perform target optimization and judgment on key parameters and product prediction results, quickly calculates the product distribution information after hydrogen distribution optimization in the catalytic cracking reaction, and realizes the automatic calculation of the optimal solution for the product hydrogen content distribution.

[0220] It should be noted that the product distribution prediction device provided in the embodiment of the present invention can be implemented based on the product distribution prediction method described in any of the above embodiments during specific execution, which will not be elaborated in this embodiment.

[0221] Figure 5 Schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 5 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communications bus 540. The processor 510 may call logic instructions in the memory 530 to execute a product distribution prediction method, which includes: inputting key parameters of a cracking catalytic reaction into a distribution prediction model to obtain product prediction results output by the distribution prediction model; the key parameters include basic device data and operating data of a catalytic cracking reaction unit; the distribution prediction model is trained using sample key parameters and sample product distribution information corresponding to the sample key parameters; the key parameters and the product prediction results are input into a target optimization model to determine product distribution information after product hydrogen distribution is optimized; the target optimization model is constructed with the optimization objectives of maximizing the output of liquefied gas and gasoline and minimizing the hydrogen content in dry gas and coke.

[0222] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0223] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the product distribution prediction method provided by the above methods, which method includes: inputting the key parameters of the cracking catalytic reaction into a distribution prediction model to obtain the product prediction results output by the distribution prediction model; the key parameters include the basic device data and operation data of the catalytic cracking reaction unit; the distribution prediction model is trained with sample key parameters and sample product distribution information corresponding to the sample key parameters; the key parameters and the product prediction results are input into a target optimization model to determine the product distribution information after the product hydrogen distribution is optimized; the target optimization model is constructed with the maximum output of liquefied gas and gasoline and the minimum hydrogen content in dry gas and coke as the optimization target.

[0224] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the product distribution prediction method provided in the above-mentioned embodiments, the method comprising: inputting key parameters of the cracking catalytic reaction into a distribution prediction model to obtain product prediction results output by the distribution prediction model; the key parameters include basic device data and operation data of the catalytic cracking reaction unit; the distribution prediction model is trained with sample key parameters and sample product distribution information corresponding to the sample key parameters; the key parameters and the product prediction results are input into a target optimization model to determine the product distribution information after the product hydrogen distribution is optimized; the target optimization model is constructed with the maximum output of liquefied gas and gasoline and the minimum hydrogen content in dry gas and coke as the optimization target.

[0225] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0226] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0227] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A product distribution prediction method, characterized in that: include: Inputting key parameters of the cracking catalytic reaction into a distribution prediction model to obtain product prediction results output by the distribution prediction model; The key parameters include basic data and operation data of the catalytic cracking reaction unit; the distribution prediction model is trained based on sample key parameters and sample product distribution information corresponding to the sample key parameters; Inputting the key parameters and the product prediction results into a target optimization model to determine product distribution information after product hydrogen distribution optimization; the target optimization model is constructed with maximizing the output of liquefied gas and gasoline and minimizing the hydrogen content in dry gas and coke as optimization goals; The target optimization model includes an objective function and preset constraints; the product distribution information after the product hydrogen distribution is optimized includes: Based on the objective function and the preset constraints, the key parameters are adjusted using the product prediction results to obtain new key parameters; Inputting the new key parameters into the distribution prediction model to obtain new product prediction results corresponding to the new key parameters; If the new key parameter and the new product prediction result satisfy the preset constraint condition, determining the new product prediction result as the product distribution information; Wherein, the distribution prediction model includes a low-carbon hydrocarbon prediction model and a liquid oil prediction model; the product prediction results include a low-carbon hydrocarbon product prediction result and a liquid oil product prediction result; Inputting the key parameters of the cracking catalytic reaction into the distribution prediction model and obtaining the product prediction results output by the distribution prediction model includes: Inputting the key parameters into a low-carbon hydrocarbon prediction model to obtain the low-carbon hydrocarbon product prediction result output by the low-carbon hydrocarbon prediction model; and inputting the key parameters into a liquid oil prediction model to obtain the liquid oil product prediction result output by the liquid oil prediction model; The distribution prediction model is constructed based on the following steps: Build initial prediction models for low-carbon hydrocarbons and liquid oil based on data-driven models, and obtain key parameters for multiple samples; A combination of a sample key parameter and sample light hydrocarbon product distribution information corresponding to the sample key parameter is used as a light hydrocarbon training sample to obtain a plurality of light hydrocarbon training samples; Taking a combination of a sample key parameter and sample liquid oil product distribution information corresponding to the sample key parameter as a liquid oil training sample, and obtaining a plurality of liquid oil training samples; Using the low-carbon hydrocarbon training samples to train the low-carbon hydrocarbon initial prediction model to obtain the low-carbon hydrocarbon prediction model; The liquid-phase oil initial prediction model is trained using the liquid-phase oil training samples to obtain the liquid-phase oil prediction model.

2. The product distribution prediction method according to claim 1, characterized in that: After adjusting the key parameters using the product prediction results to obtain new key parameters, the method further includes: If the new key parameters and the new product prediction results do not meet the preset constraints, the key parameters are adjusted according to the objective function and the preset constraints until the new key parameters and the new product prediction results meet the preset constraints, and the new product prediction results are determined to be the product distribution information.

3. The product distribution prediction method according to claim 1, characterized in that: After obtaining the low-carbon hydrocarbon prediction model, the method further includes: When the accuracy of the low-carbon hydrocarbon prediction model is less than an accuracy threshold, determining a supplementary parameter from a data warehouse, wherein a correlation between the supplementary parameter and the hydrogen content of the low-carbon hydrocarbon is greater than a correlation threshold; Acquiring liquefied gas gas chromatography data corresponding to the supplementary parameter, and determining the hydrogen content of the liquefied gas corresponding to the supplementary parameter based on the liquefied gas gas chromatography data corresponding to the supplementary parameter; Based on the supplementary parameters and the hydrogen content of the liquefied gas corresponding to the supplementary parameters, the low-carbon hydrocarbon prediction model is updated.

4. The product distribution prediction method according to claim 3, characterized in that: The data warehouse is established based on the following steps: Obtain historical raw material property parameters and historical operating parameters; Data cleaning is performed on the historical raw material property parameters and the historical operation parameters, and the cleaned historical raw material property parameters and historical operation parameters are added to the data warehouse.

5. A product distribution prediction device, characterized in that: include: an acquisition module, configured to input key parameters of the cracking catalytic reaction into a distribution prediction model and obtain a product prediction result output by the distribution prediction model; The key parameters include basic data and operation data of the catalytic cracking reaction unit; the distribution prediction model is trained based on sample key parameters and sample product distribution information corresponding to the sample key parameters; a determination module, configured to input the key parameters and the product prediction results into a target optimization model to determine product distribution information after product hydrogen distribution optimization; the target optimization model is constructed with maximizing the output of liquefied gas and gasoline and minimizing the hydrogen content in dry gas and coke as optimization objectives; The target optimization model includes an objective function and preset constraints; the product distribution information after the product hydrogen distribution is optimized includes: Based on the objective function and the preset constraints, the key parameters are adjusted using the product prediction results to obtain new key parameters; Inputting the new key parameters into the distribution prediction model to obtain new product prediction results corresponding to the new key parameters; If the new key parameter and the new product prediction result satisfy the preset constraint condition, determining the new product prediction result as the product distribution information; Wherein, the distribution prediction model includes a low-carbon hydrocarbon prediction model and a liquid oil prediction model; the product prediction results include a low-carbon hydrocarbon product prediction result and a liquid oil product prediction result; Inputting the key parameters of the cracking catalytic reaction into the distribution prediction model and obtaining the product prediction results output by the distribution prediction model includes: Inputting the key parameters into a low-carbon hydrocarbon prediction model to obtain the low-carbon hydrocarbon product prediction result output by the low-carbon hydrocarbon prediction model; and inputting the key parameters into a liquid oil prediction model to obtain the liquid oil product prediction result output by the liquid oil prediction model; The distribution prediction model is constructed based on the following steps: Build initial prediction models for low-carbon hydrocarbons and liquid oil based on data-driven models, and obtain key parameters for multiple samples; A combination of a sample key parameter and sample light hydrocarbon product distribution information corresponding to the sample key parameter is used as a light hydrocarbon training sample to obtain a plurality of light hydrocarbon training samples; Taking a combination of a sample key parameter and sample liquid oil product distribution information corresponding to the sample key parameter as a liquid oil training sample, and obtaining a plurality of liquid oil training samples; Using the low-carbon hydrocarbon training samples to train the low-carbon hydrocarbon initial prediction model to obtain the low-carbon hydrocarbon prediction model; The liquid-phase oil initial prediction model is trained using the liquid-phase oil training samples to obtain the liquid-phase oil prediction model.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the product distribution prediction method according to any one of claims 1 to 4 are implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the product distribution prediction method according to any one of claims 1 to 4 are implemented.

Images