Method and system for serving multi-target crude oil mixing production of a refining enterprise

Abstract

The present application relates to the technical field of production planning of refining enterprises, in particular to a selection method and system for multi-target crude oil mixing production of a refining enterprise, which comprises the following steps: creating a crude oil evaluation database containing various crude oil property data, and establishing a production optimization target system covering the target properties of mixed crude oil and the control targets of production cost; constructing a crude oil mixing and production process simulation model, which can simulate the properties of mixed crude oil through weighted calculation and account for the total cost; on the basis of the above, searching the database and optimizing the calculation under the premise of meeting the preset constraints, determining the candidate crude oil formula and mixing ratio that make the simulated properties closest to the target properties and the total cost closest to the control target; and performing robustness evaluation on the candidate scheme and screening the final crude oil mixing scheme that meets the anti-interference requirement. The present application realizes the collaborative optimization of the crude oil mixing scheme in terms of properties and cost, and enhances the practical stability of the scheme.

Description

Technical Field

[0001] This invention relates to the field of production planning technology for refining and chemical enterprises, and in particular to a selection method and system for multi-objective crude oil blending production in refining and chemical enterprises. Background Technology

[0002] In refining and chemical production planning, crude oil selection and formulation are crucial factors determining subsequent production efficiency. Existing technologies typically rely on operator experience or single-dimensional mathematical models to formulate blending schemes. The conventional approach is to establish a crude oil property database and, based on the property requirements of the target product, work backwards or use linear programming to find crude oil combinations with matching properties. This method often focuses on individual key property indicators of the blended crude oil, such as sulfur content or API gravity, striving to match them with target values. Another common approach is to optimize costs separately, prioritizing the lowest-priced crude oil variety while meeting basic process constraints. However, actual production is a multi-objective, collaborative process; achieving property compliance and cost control must be considered simultaneously.

[0003] Existing methods lack a unified framework for linking and balancing property matching with economic benefits, leading to potentially costly solutions or, while cost-effective, solutions with significant property deviations that negatively impact downstream plant operations. Furthermore, existing solution selection processes are often conducted under static, ideal conditions, failing to adequately consider the volatility of crude oil property data, measurement errors during production, and market supply uncertainties. This results in many theoretically sound optimization schemes failing in practice due to minor disturbances, causing critical properties to exceed limits or costs to spiral out of control. These schemes exhibit poor stability, require frequent adjustments, and fail to provide a reliable basis for stable production. Therefore, a method is needed to systematically and synergistically optimize the properties of blended crude oil against total cost targets, ensuring that the selected scheme possesses sufficient resilience in complex real-world environments. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a selection method and system for multi-objective crude oil blending production in refining and chemical enterprises.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a selection method for multi-objective crude oil blending production in refining and chemical enterprises, comprising:

[0006] Create a crude oil evaluation database, which contains a set of physical property data for various types of crude oil that are available.

[0007] Establish a production optimization target system for refining and chemical enterprises, wherein the production optimization target system for refining and chemical enterprises shall at least include the desired set of target properties of mixed crude oil and production cost control targets;

[0008] A crude oil blending and production process simulation model is constructed. The crude oil blending and production process simulation model can perform weighted calculations on the physical property data set of multiple crude oils according to the blending ratio to simulate the properties of the blended crude oil, and calculate the total cost based on raw material costs and processing paths.

[0009] Guided by the production optimization target system of the refining and chemical enterprise, and under the condition of meeting the preset constraints, the crude oil evaluation database is searched to determine at least one set of candidate crude oil formulas and their mixing ratios that make the properties of the mixed crude oil output by the crude oil mixing and production process simulation model closest to the expected set of mixed crude oil target properties and the total cost close to the production cost control target.

[0010] Robustness evaluation is performed on the at least one set of candidate crude oil formulations and their mixing ratios to screen out the final crude oil mixing scheme that meets the anti-interference requirements.

[0011] As a further aspect of the present invention, the creation of the crude oil evaluation database includes:

[0012] Multiple crude oil samples were collected, and standardized laboratory evaluations were performed on each crude oil sample to obtain the set of physical property data, including density, sulfur content, acid value, distillation range distribution, and content of key components.

[0013] The physical property data set is stored in association with the corresponding crude oil production location and crude oil variety name;

[0014] Receive actual processing yield data from the enterprise's historical production data, and associate the actual processing yield data with the corresponding crude oil variety name to supplement the crude oil evaluation database;

[0015] All data in the crude oil evaluation database are normalized to form a standardized data storage structure.

[0016] As a further aspect of the present invention, the construction of the crude oil blending and production process simulation model includes:

[0017] Define a mixing rule, which stipulates that for linearly additive properties in the physical property data set, the properties of the mixed crude oil are obtained by linearly weighting the properties of each component crude oil according to their mixing ratio;

[0018] Define transformation rules, which stipulate that for nonlinear properties or yields, calculations are performed by calling preset yield correlation formulas or empirical correlation models, with key parameters in the physical property data set as inputs;

[0019] An integrated cost calculation module is provided, which calculates the total raw material and processing cost per unit product based on the unit price of each component crude oil, the mixing ratio, and the processing consumption calculated by the conversion rules.

[0020] As a further aspect of the present invention, the search of the crude oil evaluation database includes:

[0021] Set optimization variables, which include combinations of crude oil types selected from the crude oil evaluation database and their corresponding mixing ratios;

[0022] Set constraints, including upper and lower limits for the blending ratio of a single crude oil, the total blending ratio, the range of key property indicators of the blended crude oil, and selectable limits on the total crude oil inventory.

[0023] Set an objective function, which is in a weighted form and includes a deviation term between the properties of the mixed crude oil and the desired set of target properties of the mixed crude oil, and a deviation term between the total cost and the production cost control target.

[0024] A multi-objective optimization algorithm is used to iteratively solve the objective function under the constraints to generate a Pareto front consisting of multiple non-dominated solutions. Each non-dominated solution corresponds to a set of candidate crude oil formulations and their mixing ratios.

[0025] As a further aspect of the present invention, the robustness evaluation of the at least one set of candidate crude oil formulations and their mixing ratios includes:

[0026] Identify the key physical property data of each crude oil in the candidate crude oil formulation, and set a reasonable fluctuation range for the key physical property data to simulate the property fluctuations of actual crude oil.

[0027] Within the fluctuation range, Monte Carlo random sampling is performed on the key physical property data;

[0028] Substitute the data obtained from each sampling into the crude oil blending and production process simulation model to recalculate the properties and total cost of the blended crude oil.

[0029] The probability that the properties of the mixed crude oil exceed the allowable range of the expected set of target properties of the mixed crude oil, and the probability that the total cost exceeds the preset risk threshold, are calculated among all sampling results.

[0030] Candidate crude oil formulations with a probability lower than a preset acceptable level are determined to meet the anti-interference requirements.

[0031] As a further aspect of the present invention, the key physical property data for identifying each crude oil in the candidate crude oil formulation includes:

[0032] Sensitivity analysis was performed on the crude oil blending and production process simulation model to calculate the partial derivatives of each property in the desired set of target properties of blended crude oil with each data in the set of physical property data of each component crude oil.

[0033] Physical property data with absolute values ​​of partial derivatives greater than a preset sensitivity threshold are selected and marked as key physical property data that have a significant impact on the target properties.

[0034] Based on the aforementioned key physical property data and combined with historical oil quality fluctuation data, a reasonable fluctuation range is determined.

[0035] As a further aspect of the present invention, the method also includes dynamic rolling optimization:

[0036] Periodically acquire the latest available crude oil inventory information, crude oil market price information, and updated production optimization target system for the aforementioned refining and chemical enterprises;

[0037] Update the constraints and objective function in the optimization algorithm search with the latest information;

[0038] Based on the updated constraints and objective function, the step of searching the crude oil evaluation database using the optimization algorithm is re-executed to generate a new Pareto frontier and candidate schemes.

[0039] The new candidate solution is compared with the currently implemented solution. If the improvement in benefits exceeds the switching cost, a solution adjustment recommendation is output.

[0040] As a further aspect of the present invention, the method also includes scheme tracing and modification:

[0041] Record the actual execution process of the final crude oil blending scheme, and collect the property data of each batch of crude oil actually used, the actual blending ratio, and the actual operating results and product yield of downstream production units;

[0042] The actual operating results are compared and analyzed with the results predicted by the crude oil blending and production process simulation model, and the prediction deviation of key indicators is calculated.

[0043] If the prediction deviation continues to exceed the allowable range, a model correction mechanism is triggered to calibrate the conversion rules or related model parameters in the crude oil blending and production process simulation model using the actual operating result data.

[0044] As a further aspect of the present invention, the calibration of the conversion rules or correlation model parameters in the crude oil blending and production process simulation model using the actual operating result data includes:

[0045] From the actual operating results data, extract the actual input properties, actual mixing ratio, and actual output yield and properties corresponding to the final crude oil mixing scheme;

[0046] Input the actual input properties and actual mixing ratios into the current version of the crude oil mixing and production process simulation model to obtain the model's predicted output yield and properties.

[0047] Construct a loss function that measures the difference between the output yield and properties predicted by the model and the actual output yield and properties;

[0048] The adjustable parameters in the crude oil mixing and production process simulation model are adjusted using either gradient descent or nonlinear least squares methods to minimize the loss function and complete model calibration.

[0049] As a further aspect of the present invention, the present invention also includes a selection system for multi-objective crude oil blending production in refining and chemical enterprises. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the selection method for multi-objective crude oil blending production in refining and chemical enterprises described above.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0051] By establishing a multi-objective optimization system that includes a set of target properties for mixed crude oil and a production cost control objective, property matching and economic benefits are incorporated into the same decision-making framework. The solution guides the search process, simultaneously seeking optimization in two directions—property fit and cost convergence—supported by a physical property database and a process simulation model. This avoids the one-sidedness caused by single-objective optimization; the generated candidate formulations are not simply the most property-matched or lowest-cost solutions, but rather comprehensive optimal solutions that achieve a balance between the two. This ensures that production planning directly links economic benefits while meeting product quality and process requirements, achieving integrated and synergistic optimization of technical and economic indicators.

[0052] After determining the candidate formulations, a robustness assessment process is introduced to quantitatively analyze the ability of the candidate schemes to resist disturbances. This assessment considers the potential fluctuation range of key physical properties, possible deviations in the mixing ratio, and changes in cost factors, examining the feasibility of the schemes under non-ideal conditions. Only those schemes that can maintain the properties of the mixed crude oil within acceptable ranges and whose costs do not fluctuate drastically under preset disturbance scenarios are selected as the final schemes. This ensures that the final crude oil blending schemes are not merely theoretically static optimal solutions, but possess resilience to cope with real-world complexity and uncertainty, significantly improving the reliability and stability of the production scheme in actual implementation and reducing the risk of unplanned shutdowns or quality accidents caused by external disturbances. Attached Figure Description

[0053] Figure 1 This is a flowchart of the selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in this invention;

[0054] Figure 2 A flowchart for constructing a simulation model of crude oil blending and production processes;

[0055] Figure 3 Heatmap of crude oil mixing ratio for Pareto frontier candidate schemes;

[0056] Figure 4 Sensitivity heatmap of key physical properties to target properties of blended crude oil;

[0057] Figure 5 This is for comparing model parameters before and after calibration and for physical boundary constraints. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0059] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0060] See Figure 1The core of this invention lies in providing a selection method for multi-objective crude oil blending production in refining and chemical enterprises. This method achieves scientific decision-making on crude oil formulations through integrated data management, model simulation, optimization search, and robustness assessment. The overall implementation scheme of this invention is as follows: First, a crude oil evaluation database is created, which contains a set of physical property data for various available crude oils, providing basic data support for subsequent analysis. Second, a production optimization target system for refining and chemical enterprises is established, which includes at least the desired set of target properties for blended crude oils and production cost control targets, providing clear guidance for optimization. Third, a crude oil blending and production process simulation model is constructed. This model can perform weighted calculations on the physical property data sets of various crude oils based on the blending ratio to simulate the properties of blended crude oils and calculate the total cost based on raw material costs and processing paths. Guided by the production optimization target system for refining and chemical enterprises, and under preset constraints, the crude oil evaluation database is searched to determine at least one set of candidate crude oil formulations and their blending ratios that make the properties of the blended crude oil output by the crude oil blending and production process simulation model closest to the desired set of target properties for blended crude oils and the total cost approach the production cost control target. Robustness assessments are conducted on at least one set of candidate crude oil formulations and their mixing ratios to screen out the final crude oil mixing scheme that meets the anti-interference requirements.

[0061] In one embodiment of the present invention, the creation of a crude oil evaluation database involves a systematic data collection and processing process. Operators collect various representative crude oil samples from oil fields or trading channels in different production areas around the world. These crude oil samples cover crude oil varieties with different densities, sulfur contents, and distillation range characteristics. Each crude oil sample is sent to a laboratory that meets national or industry standards for comprehensive physical property analysis according to unified testing standards and specifications. The laboratory evaluation generates a series of standardized physical property data sets. The core indicators of this physical property data set include density, sulfur content, acid value, distillation range distribution data, and the content of key components such as wax content, carbon residue, and nickel and vanadium content. This set of physical property data directly generated by the laboratory is the most basic and authoritative data source in the database.

[0062] In some embodiments, after completing the laboratory evaluation, the physical property data set needs to be systematically associated and stored with the corresponding crude oil source information. Each physical property data record must be clearly bound to its corresponding crude oil origin information and crude oil variety name. The crude oil origin information can be refined to specific oil fields or blocks, and the crude oil variety name adopts the industry-standard naming rules. In addition to standard laboratory data, the implementation process also includes receiving and integrating actual processing yield data from the historical production records of refining and chemical enterprises. These actual processing yield data come from the enterprise's production execution system or database and record the actual output of each production unit when processing specific crude oil varieties in history. During implementation, data mapping technology is used to associate and map the actual processing yield data with the corresponding records in the crude oil evaluation database according to the crude oil variety name, and the actual processing yield data is entered into the database as key supplementary information.

[0063] Optionally, after the crude oil evaluation database compiles physical property data from laboratories and actual processing yield data from production history, in order to eliminate the influence of differences in dimensions and numerical ranges between data from different sources and to facilitate unified calculation and comparison of subsequent models, it is necessary to normalize all data in the crude oil evaluation database. Normalization can be achieved using a min-max scaling method to convert raw data with different dimensions to a unified numerical range. For a data matrix containing m types of crude oil, each with n property indicators, the normalization process can be described as follows: for the j-th property indicator value of the i-th crude oil... Its normalized value The result is obtained through calculation using the formula:

[0064]

[0065] in: and These represent the minimum and maximum values ​​of all m types of crude oil on the j-th property index, respectively. After the above processing, a unified and standardized data storage structure is ultimately formed. It can be understood that the creation of the entire crude oil evaluation database is a continuous process of maintenance and updating; new crude oil variety evaluation data and new production operation data will be periodically added to the database.

[0066] In practice, standardized laboratory evaluation is a complete process encompassing rigorous operating procedures and technical requirements. This process is based on adopting and implementing a series of recognized international and domestic standard methods. For example, in determining crude oil density, the laboratory strictly adheres to ASTM D4052, using a digital densitometer at a constant temperature of 15 degrees Celsius. Before measurement, the instrument is calibrated using standard materials, and the ambient temperature is ensured to be stable. For sulfur content analysis, the laboratory employs X-ray fluorescence spectrometry and follows all requirements of ASTM D4294, including sample cell pretreatment, establishment and validation of standard curves, and periodic calibration of the instrument's background to ensure that the test results fall within the accuracy and precision range validated by the method. The determination of distillation range distribution is strictly performed according to ASTM D86. During the experiment, the heating rate, condenser temperature, and receiver volume readings are monitored and recorded throughout the process to obtain accurate initial boiling point, final boiling point, and distillation volume percentage data for each temperature range. The content of all key components, such as nickel and vanadium, must be determined by inductively coupled plasma atomic emission spectrometry (ICP-AES) according to ASTM D5708 standards. The laboratory must regularly participate in proficiency testing programs to ensure the comparability and reliability of the test data. All raw data and calculations generated throughout the evaluation process are recorded in the laboratory information management system. Each crude oil evaluation report must be independently reviewed by another qualified analyst and approved by the technical lead before release. This results in a traceable and reproducible set of physical property data, providing a solid and consistent data foundation for subsequent hybrid simulations.

[0067] Data mapping technology refers to a data processing method that accurately correlates actual processing yield data from historical production records with standardized records in a crude oil evaluation database. This technology requires establishing a standardized dictionary of crude oil product names within the system. The dictionary includes all possible aliases, trade names, and abbreviations for each crude oil, and assigns a unique primary name identifier to each. When an actual processing yield data record containing a crude oil product name field is received by the system, the data mapping engine attempts to perform an exact string match, searching for a completely matching record in the "Crude Oil Product Name" field of the crude oil evaluation database.

[0068] If an exact match fails, the engine initiates a fuzzy matching algorithm. This algorithm calculates the similarity between the input name and all known names in the database based on edit distance, and combines auxiliary fields such as crude oil origin code and supplier information for comprehensive judgment, filtering out one or more candidate names whose similarity exceeds a preset threshold. The system will push a mapping confirmation request interface to production planning or technical management personnel. The interface clearly lists the historical data records to be mapped, the system-recommended candidate matches and their matching criteria, and allows for manual final confirmation or specification of the correct mapping relationship. All confirmed mapping relationships are stored in a dedicated mapping relationship table, forming a persistent link between historical data fields and standard database fields. Subsequent data import processes can directly call this mapping table to achieve automated association. For newly emerging crude oil names that cannot be mapped, the system will generate an alert and trigger the evaluation and database entry process for the new crude oil variety, thereby ensuring that actual processing yield data can be continuously and accurately integrated into the data ecosystem of the crude oil evaluation database.

[0069] In one embodiment of the present invention, see [reference] Figure 2 In practical implementation, constructing a crude oil blending and production process simulation model requires clearly defining data processing rules and cost accounting logic. The core function of the model is to perform weighted calculations on the physical property data sets of multiple crude oils based on the blending ratio to simulate the properties of the blended crude oil, and to calculate the total cost based on raw material costs and processing paths. To achieve this function, the model needs to define blending rules. These rules stipulate that for properties identified as linearly additive in the physical property data set, such as density and sulfur content, the corresponding property values ​​of the blended crude oil are obtained by linearly weighting the property values ​​of each component crude oil according to their blending ratio. For non-linearly additive properties, other rules need to be applied. In some embodiments, the model needs to further define transformation rules to handle the calculation of non-linear properties or yields. These transformation rules stipulate that for non-linear properties or yields, calculations are performed by calling preset yield correlation formulas or empirical correlation models. These yield correlation formulas or empirical correlation models take key parameters in the physical property data set as inputs. For example, key parameters can be the yield of the middle distillate of crude oil, carbon residue, or metal content. A specific yield correlation formula can be expressed as:

[0070]

[0071] in: Represents the yield of a specific product. Represents the selection from the physical property data set One key physical property parameter, This represents a functional relationship determined through fitting historical production data or industry experience. In some embodiments, yield correlation or empirical correlation models can be subdivided and configured according to different processing units or processes, enabling the model to simulate the entire chain from mixed crude oil feedstock to final product output. The accuracy of the conversion rules directly affects the model's prediction accuracy of production results.

[0072] Optionally, based on the defined mixing and conversion rules, the crude oil blending and production process simulation model needs to integrate a cost calculation module to complete total cost accounting. This module calculates the total raw material and processing costs per unit product based on the real-time or contract unit price of each component crude oil, the optimized mixing ratio, and the material and energy consumption during processing calculated through the conversion rules. Optionally, the cost calculation module's logic can cover all cost elements from crude oil procurement to product delivery. The raw material cost is summed based on the unit price of each component crude oil and the mixing ratio, while the processing cost is calculated based on the processing path, consumption indicators, and corresponding energy unit price and catalyst loss simulated by the conversion rules. It can be understood that the crude oil blending and production process simulation model, by integrating mixing rules, conversion rules, and the cost calculation module, constitutes a complete digital mapping reflecting the entire process from raw material formulation to product yield and cost. The model's output is the prediction of the properties of the blended crude oil and the total cost estimate, providing a reliable evaluation basis for subsequent multi-objective optimization searches. It can also be understood that the yield correlation or empirical correlation model parameters in the model need to be periodically calibrated based on the company's actual production data to maintain consistency between their predictive ability and actual production conditions.

[0073] In practical implementation, the empirical correlation model refers to a mathematical model built using mathematical statistical methods based on historical production data and the fundamental properties of crude oil, used to predict nonlinear processing yields or product properties. The model's construction begins with compiling historical datasets. These datasets need to cover complete input and output information for different types of crude oil processed at different times. Input information comes from a crude oil evaluation database, including physical property data such as density, sulfur content, distillation range distribution, and key component content of the processed crude oil. Output information comes from the production execution system, recording the actual product yield distribution and key product quality indicators of major units such as atmospheric and vacuum distillation units, catalytic cracking units, and hydrotreating units within the corresponding production cycle. The construction process begins with process engineers and data analysts jointly determining the target variable for modeling, such as the light oil yield of the catalytic cracking unit or the cetane number index of diesel products. Based on process knowledge, a batch of possible influencing variables is initially selected as the candidate set of independent variables for the model. Subsequently, feature selection methods such as stepwise regression and LASSO regression are used to screen out key physical property parameters that significantly explain the target variable from the candidate set, eliminating parameters with strong collinearity or low contribution. The model form is selected based on the preliminary analysis of the relationships between variables. When the relationships exhibit significant nonlinearity, multinomial regression, support vector regression, or a simplified neural network structure will be used. Model parameters are estimated using nonlinear least squares or maximum likelihood estimation, and historical datasets are used for fitting. After the initial model is established, it needs to be validated using another set of historical data not used in the modeling (test set). Statistical indicators such as the mean absolute percentage error between predicted and actual values ​​and the coefficient of determination are calculated. Only models that pass validation and whose errors are within acceptable limits will be formally integrated into the crude oil blending and production process simulation model for predicting nonlinear properties or yields.

[0074] Yield correlations are a specific mathematical expression of empirical correlation models, typically represented by one or more explicit algebraic equations. Their establishment is closely dependent on the process mechanism and operational data of a specific production unit. Taking the prediction of distillate oil yield from an atmospheric and vacuum distillation unit as an example, establishing the correlation first requires collecting operational records and product yield data of the unit when processing crude oils of various properties. Process engineers, based on the principles of mass conservation and material balance, and combined with the separation characteristics of the unit's towers, propose a basic functional framework for the correlation. For example, they might assume a strong correlation between the yield of light naphtha and the content of light fractions (such as fractions from the initial boiling point to 180 degrees Celsius) in crude oil, while the yield of vacuum wax oil is related to the content of fractions above 350 degrees Celsius and the residual carbon value in the crude oil.

[0075] Within this framework, data analysts use relevant physical properties (such as yield, density, and acid value within a specific distillation range) from crude oil evaluation data as independent variables and the actual measured yields of each distillate fraction as dependent variables. They then employ multiple linear regression or nonlinear fitting methods to estimate the parameters, resulting in a specific mathematical expression. For example, a simplified yield correlation for vacuum gas oil might be expressed as: Vacuum gas oil yield = a × (content of fractions greater than 350 degrees Celsius) + b × (carbon residue) + c, where coefficients a, b, and c are determined through regression analysis. Each such yield correlation must undergo rigorous statistical testing and engineering feasibility assessment. Its predicted trend must conform to common technological knowledge, and its prediction error must meet pre-defined accuracy requirements in historical data testing. These validated yield correlations are then encoded as callable functions in the conversion rules of the crude oil blending and production process simulation model. When the model needs to calculate the product distribution of blended crude oil after passing through a specific processing path, it calls the corresponding yield correlation, inputs the weighted average physical properties of the blended crude oil, and thus calculates the predicted product yield.

[0076] In one embodiment of the present invention, in specific implementation, the optimization search guided by the production optimization target system of refining and chemical enterprises requires the construction of a complete mathematical optimization framework. The optimization search process aims to determine candidate formulations from the crude oil evaluation database that satisfy the target and have similar costs when the crude oil blending and production process simulation model outputs. During implementation, optimization variables are set, including combinations of crude oil types selected from the crude oil evaluation database and their corresponding blending ratios. The blending ratios are expressed as decimals or percentages, and the sum of the blending ratios of all selected crude oils is 1. In some embodiments, constraints must be set to ensure that the optimization results conform to actual production and safety regulations. These constraints include upper and lower limits for the blending ratio of a single crude oil, a total blending ratio sum of 1, a range of key property indicators for the blended crude oil, and a limit on the total available crude oil inventory. The range of key property indicators for the blended crude oil is set based on the target property set and the processing capacity of the equipment, while the limit on the total available crude oil inventory is set based on real-time data from the enterprise's storage system. In some embodiments, an objective function needs to be set to quantitatively evaluate the merits of different crude oil formulation schemes. The objective function is in a weighted form, including a deviation term between the properties of the blended crude oil and the desired set of target properties of the blended crude oil, and a deviation term between the total cost and the production cost control target. The deviation term is usually constructed in the form of squared error or absolute error. A specific objective function... The format is:

[0077]

[0078] in: The total number of terms representing the target property. The first result is obtained by calculation using a crude oil blending and production process simulation model. The value of the term property, Representing the The expected target value of the property. Representing the Weighting coefficients for the deviation of the item properties This represents the total cost of model computation. Represents the production cost control target. The weighting coefficients represent cost deviations. A multi-objective optimization algorithm is used to iteratively solve the objective function under constraints. This algorithm can be NSGA-II or MOEA / D. The algorithm searches to generate a Pareto front consisting of multiple non-dominated solutions. Each non-dominated solution on the Pareto front corresponds to a set of candidate crude oil formulations and their mixing ratios.

[0079] Optionally, to address changes in market and production conditions, the implementation process includes a dynamic rolling optimization mechanism. This mechanism periodically acquires the latest available crude oil inventory information, crude oil market price information, and updated production optimization target systems for refining enterprises. This information is automatically or manually imported through data interfaces with the enterprise resource planning (ERP) system and production management system. The constraints and objective functions in the optimization algorithm search are updated with the latest information. For example, the latest crude oil inventory data updates the total inventory limit in the constraints, the latest market prices update the unit price parameters in the cost calculation module, and the latest production optimization target systems update the expected target values ​​and weighting coefficients in the objective function. Based on the updated constraints and objective functions, the step of searching the crude oil evaluation database using the optimization algorithm is re-executed to generate new Pareto fronts and candidate solutions. In essence, dynamic rolling optimization is a continuous cyclical process, and its execution cycle can be set daily or weekly according to the enterprise's planning and scheduling rhythm. Optionally, after generating a new candidate solution, it needs to be compared with the currently implemented solution. The comparison includes property satisfaction, expected total cost, and the operational and time costs required to switch from the current solution to the new one. If the expected overall benefit improvement of the new solution exceeds the estimated switching cost, a solution adjustment suggestion is provided to the production planner. It can be understood that dynamic rolling optimization ensures that the crude oil blending production solution can respond promptly to changes in the external environment and internal demand.

[0080] See Figure 3The graph visually illustrates the mixing proportions of six crude oils (vertical axis) across 10 candidate schemes (horizontal axis) on the Pareto front. The color gradient represents the magnitude of the mixing proportions, ranging from light yellow (low proportion) to deep burgundy (high proportion), clearly reflecting the differences in the proportions of various crude oils under different schemes. For example, in candidate scheme 0, Middle Eastern light crude oil has the highest proportion (31.6%), while Daqing crude oil has the lowest (5.2%); candidate scheme 9 is characterized by the highest proportion of Middle Eastern heavy crude oil (28.8%). Overall, Middle Eastern light crude oil has a relatively high proportion in multiple schemes, reflecting its core role in meeting the target properties and cost control objectives of the mixed crude oil; while the proportions of Bohai crude oil and Russian crude oil fluctuate significantly across different schemes, reflecting the dynamic adjustment process of the multi-objective optimization algorithm when balancing property deviations and cost deviations. The heatmap not only visualizes the non-dominated solutions of the Pareto front but also provides intuitive data support for subsequent robustness assessments and scheme traceability.

[0081] In one embodiment of the present invention, a robustness assessment of at least one set of candidate crude oil formulations and their mixing ratios is a step in screening the final scheme. The robustness assessment aims to simulate the impact of actual oil property fluctuations on the stability of the scheme. The implementation process requires identifying key physical property data for each crude oil in the candidate formulations and setting reasonable fluctuation ranges for these key physical property data. Key physical property data refers to crude oil property parameters that significantly affect the target properties of the blended crude oil. In some embodiments, determining the key physical property data and their reasonable fluctuation range requires combining model analysis and historical data to perform sensitivity analysis on the crude oil blending and production process simulation model. This involves calculating the partial derivatives of each property in the desired set of target blended crude oil properties with respect to each data in the set of physical property data for each component crude oil. The partial derivatives characterize the amount of change in the target property caused by a unit change in physical property data. Physical property data with absolute values ​​of partial derivatives greater than a preset sensitivity threshold are selected and marked as key physical property data that significantly affect the target properties. For these identified key physical property data, it is necessary to combine them with historical oil quality fluctuation data to determine their reasonable fluctuation range. For example, by analyzing the test records of a certain property index of crude oil of the same origin and type in the past year, the average value plus or minus three times the standard deviation can be taken as the upper and lower limits of the fluctuation range.

[0082] In practical implementation, to intuitively display the identified key physical property data and their fluctuation range, the following table can be constructed. Referring to Table 1, it lists some of the key physical property data identified for a candidate formulation (including crude oil A, crude oil B, and crude oil C), their baseline values, and the set fluctuation ranges.

[0083] Table 1: Fluctuation Range of Key Physical Properties of Candidate Formulations

[0084]

[0085] Optionally, after defining the key physical property data and their reasonable fluctuation range, the robustness assessment enters the simulation calculation stage. Monte Carlo random sampling is performed on the key physical property data within the fluctuation range. Monte Carlo random sampling means that for each simulation experiment, a value is independently generated randomly from the fluctuation range of each key physical property data according to a specific distribution (such as a uniform distribution). The data obtained from each sampling is substituted into the crude oil blending and production process simulation model to recalculate the properties and total cost of the blended crude oil. This process is typically repeated thousands of times to obtain stable statistical results. The probability that the properties of the blended crude oil exceed the allowable range of the expected target properties set of the blended crude oil, and the probability that the total cost exceeds a preset risk threshold, are calculated from all sampling results. The probability of exceeding the threshold can be calculated by statistically analyzing the ratio of the number of unsafe samples to the total number of samplings, expressed by the formula:

[0086]

[0087] in: This represents the probability of exceeding the permissible range or risk threshold. This represents the total number of samples that do not meet the requirements in the sampling results. This represents the total number of Monte Carlo random samplings. Candidate crude oil formulations with probabilities below a preset acceptable level are deemed to meet the anti-interference requirements. It is understood that robustness assessment places candidate formulations from static ideal conditions into a dynamic, fluctuating environment for testing. It is also understood that the sensitivity threshold and acceptable probability level need to be set based on the actual stability requirements of production.

[0088] See Figure 4 This paper presents the sensitivity analysis results of key physical properties of each crude oil to the target properties of the blended crude oil. In this analysis, sensitivity is expressed as the absolute value of the partial derivative; the larger the value, the more significant the influence of the physical property on the target property. From the color and numerical distribution in the figure, it can be seen that the carbon residue value of crude oil A has the highest sensitivity to the blended carbon residue value (partial derivative of 0.95) and also a relatively high sensitivity to the blended sulfur content (0.88), indicating that fluctuations in these values ​​significantly affect these two target properties. The sulfur content of crude oil A has a sensitivity of 0.92 to the blended sulfur content, making it a core parameter affecting this target property, but its influence on the blended carbon residue value is minimal (0.15). The acid value and nickel content of crude oil B, and the density of crude oil C, all have sensitivities below 0.3 to both target properties, indicating they are relatively weak physical property parameters. The color gradient design of the heatmap intuitively reflects the magnitude of the partial derivatives. The red area represents high sensitivity (partial derivative ≥ 0.8), and the blue area represents low sensitivity (partial derivative ≤ 0.3). It can be directly used to screen the key physical property data required for robustness assessment and provides the core basis for setting the fluctuation range of subsequent Monte Carlo simulations.

[0089] In one embodiment of the present invention, scheme traceability and correction are crucial steps in ensuring the long-term accuracy of the crude oil blending and production process simulation model. The implementation process begins with the systematic recording of the actual execution of the final crude oil blending scheme. Operators collect property data, actual blending ratios, and actual operating results and product yields of each batch of crude oil used, as well as the actual operating results and product yields of downstream production units. This data is automatically aggregated into a database through on-site production instruments, laboratory analysis reports, and the Manufacturing Execution System (MES). The actual operating results are compared and analyzed with the results predicted by the crude oil blending and production process simulation model to calculate the prediction deviation of key indicators. The prediction deviation involves the difference between the predicted and measured values ​​of the blended crude oil properties, and the difference between the predicted and actual operating results of key product yields. If the prediction deviation continues to exceed the allowable range, a model correction mechanism is triggered. This mechanism automatically uses the accumulated actual operating result data to calibrate the transformation rules or related model parameters in the crude oil blending and production process simulation model.

[0090] In some embodiments, when calibrating a crude oil blending and production process simulation model using actual operational data, it is necessary to extract the actual input properties, actual mixing ratios, and actual output yields and properties corresponding to the final crude oil blending scheme from the actual operational data. Actual input properties refer to the set of property data obtained from the testing of each crude oil batch before it enters the plant or is blended. The actual input properties and actual mixing ratios are input into the current version of the crude oil blending and production process simulation model to obtain the model-predicted output yields and properties. These predicted output yields and properties are theoretical values ​​calculated based on the current model parameters. A loss function is constructed to measure the difference between the model-predicted output yields and properties and the actual output yields and properties. To unify the contribution of output terms with different dimensions to the loss function and ensure dimensional consistency, the difference is measured using a weighted form based on relative error. A specific loss function is described below. The construction is as follows:

[0091]

[0092] in: This represents the total number of output variables that need to be calibrated. Each output variable can be the yield of a specific product or a property of a mixed crude oil. The representative crude oil blending and production process simulation model for the first The predicted values ​​of the output variables. Representing the The actual execution result values ​​corresponding to each output variable Representing the The weighting coefficients of each output variable in the loss function are used. Gradient descent or nonlinear least squares methods are employed to adjust the adjustable parameters in the crude oil blending and production process simulation model to minimize the loss function, thus completing model calibration.

[0093] Optionally, the model calibration process can be set to be executed automatically periodically or manually triggered based on deviation warnings. The actual operating results data used for calibration are usually selected from historical data within a recent period to ensure that the model adapts to the latest production conditions. After calibration, a new version of the crude oil blending and production process simulation model is generated, and version change logs and parameter adjustments are recorded. It can be understood that scheme traceability and correction form a closed loop from production practice feedback to model optimization. In some embodiments, the specific form and weighting coefficients of the loss function can be differentiated according to the importance of different production units or data reliability. Optionally, for different types of correlation models in the transformation rules, the range of their adjustable parameters needs to be set with physical or empirical boundaries to prevent the calibration process from generating unrealistic parameter values. It can be understood that continuous model calibration is the foundation for maintaining the effectiveness of the entire crude oil blending production selection method.

[0094] See Figure 5 In the parameter calibration stage of the crude oil blending and production process simulation model, to ensure the physical rationality of the parameters, clear physical boundary constraints need to be applied to the adjustable parameters. Specifically, in the figure, the horizontal axis represents the five types of adjustable parameters of the model, including conversion rule parameter 1, yield correlation model parameter A, yield correlation model parameter B, property mixing parameter α, and property mixing parameter β; the vertical axis represents the numerical value of the parameters. The blue dashed line represents the lower physical bound of the parameters, the red dashed line represents the upper physical bound of the parameters, the green bars represent the initial parameter values ​​before calibration, the orange bars represent the parameter values ​​after calibration, and the error bars characterize the uncertainty range of the parameters. As can be seen from the figure, all calibrated parameter values ​​are within the preset upper and lower physical bounds, and no cases of exceeding the constraints have occurred, indicating that the calibration process effectively follows the physical boundary constraints. Among them, the yield correlation model parameter B, with a value of 0.35 before calibration, increased to 0.62 after calibration, still within the physical boundary of 0.04 to 0.80; the property mixing parameter α, with a value of 0.78 after calibration, is close to the physical upper limit of 0.82, indicating that this parameter has reached a value that is more in line with actual production data under physical constraints. During parameter configuration, the setting of physical upper and lower limits is based on industrial experience and physical property laws in crude oil processing. For example, the upper limit of the yield correlation model parameter is determined by the maximum conversion efficiency of crude oil components, while the lower limit is constrained by the minimum processing load of the equipment.

[0095] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A selection method for multi-objective crude oil blending production in refining and chemical enterprises, characterized in that, The method includes: Create a crude oil evaluation database, which contains a set of physical property data for various types of crude oil that are available. Establish a production optimization target system for refining and chemical enterprises, wherein the production optimization target system for refining and chemical enterprises shall at least include the expected set of target properties of mixed crude oil and production cost control targets; A crude oil blending and production process simulation model is constructed. The crude oil blending and production process simulation model can perform weighted calculations on the physical property data set of multiple crude oils according to the blending ratio to simulate the properties of the blended crude oil, and calculate the total cost based on raw material costs and processing paths. Guided by the production optimization target system of the refining and chemical enterprise, and under the condition of meeting the preset constraints, the crude oil evaluation database is searched to determine at least one set of candidate crude oil formulas and their mixing ratios that make the properties of the mixed crude oil output by the crude oil mixing and production process simulation model closest to the expected set of mixed crude oil target properties and the total cost close to the production cost control target. Robustness evaluation is performed on the at least one set of candidate crude oil formulations and their mixing ratios to screen out the final crude oil mixing scheme that meets the anti-interference requirements.

2. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 1, characterized in that, The creation of the crude oil evaluation database includes: Multiple crude oil samples were collected, and standardized laboratory evaluations were performed on each crude oil sample to obtain the set of physical property data, including density, sulfur content, acid value, distillation range distribution, and content of key components. The physical property data set is stored in association with the corresponding crude oil production location and crude oil variety name; Receive actual processing yield data from the enterprise's historical production data, and associate the actual processing yield data with the corresponding crude oil variety name to supplement the crude oil evaluation database; All data in the crude oil evaluation database are normalized to form a standardized data storage structure.

3. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 1, characterized in that, The construction of the crude oil blending and production process simulation model includes: Define a mixing rule, which stipulates that for linearly additive properties in the physical property data set, the properties of the mixed crude oil are obtained by linearly weighting the properties of each component crude oil according to their mixing ratio; Define transformation rules, which stipulate that for nonlinear properties or yields, calculations are performed by calling preset yield correlation formulas or empirical correlation models, with key parameters in the physical property data set as inputs; An integrated cost calculation module is provided, which calculates the total raw material and processing cost per unit product based on the unit price of each component crude oil, the mixing ratio, and the processing consumption calculated by the conversion rules.

4. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 1, characterized in that, The search of the crude oil evaluation database includes: Set optimization variables, which include combinations of crude oil types selected from the crude oil evaluation database and their corresponding mixing ratios; Set constraints, including upper and lower limits for the blending ratio of a single crude oil, the total blending ratio, the range of key property indicators of the blended crude oil, and selectable limits on the total crude oil inventory. Set an objective function, which is in a weighted form and includes a deviation term between the properties of the mixed crude oil and the desired set of target properties of the mixed crude oil, and a deviation term between the total cost and the production cost control target. A multi-objective optimization algorithm is used to iteratively solve the objective function under the constraints to generate a Pareto front consisting of multiple non-dominated solutions. Each non-dominated solution corresponds to a set of candidate crude oil formulations and their mixing ratios.

5. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 4, characterized in that, The robustness evaluation of the at least one set of candidate crude oil formulations and their mixing ratios includes: Identify the key physical property data of each crude oil in the candidate crude oil formulation, and set a reasonable fluctuation range for the key physical property data to simulate the property fluctuations of actual crude oil. Within the fluctuation range, Monte Carlo random sampling is performed on the key physical property data; Substitute the data obtained from each sampling into the crude oil blending and production process simulation model to recalculate the properties and total cost of the blended crude oil. The probability that the properties of the mixed crude oil exceed the allowable range of the expected set of target properties of the mixed crude oil, and the probability that the total cost exceeds the preset risk threshold, are calculated among all sampling results. Candidate crude oil formulations with a probability lower than a preset acceptable level are determined to meet the anti-interference requirements.

6. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 5, characterized in that, The key physical property data for identifying each crude oil in the candidate crude oil formulation includes: Sensitivity analysis was performed on the crude oil blending and production process simulation model to calculate the partial derivatives of each property in the desired set of target properties of blended crude oil with each data in the set of physical property data of each component crude oil. Physical property data with absolute values ​​of partial derivatives greater than a preset sensitivity threshold are selected and marked as key physical property data that have a significant impact on the target properties. Based on the aforementioned key physical property data and combined with historical oil quality fluctuation data, a reasonable fluctuation range is determined.

7. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 4, characterized in that, The method also includes dynamic scrolling optimization: Periodically acquire the latest available crude oil inventory information, crude oil market price information, and updated production optimization target system for the aforementioned refining and chemical enterprises; Update the constraints and objective function in the optimization algorithm search with the latest information; Based on the updated constraints and objective function, the step of searching the crude oil evaluation database using the optimization algorithm is re-executed to generate a new Pareto frontier and candidate schemes. The new candidate solution is compared with the currently implemented solution. If the improvement in benefits exceeds the switching cost, a solution adjustment recommendation is output.

8. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 5, characterized in that, The method also includes scheme tracing and correction: Record the actual execution process of the final crude oil blending scheme, and collect the property data of each batch of crude oil actually used, the actual blending ratio, and the actual operating results and product yield of downstream production units; The actual operating results are compared and analyzed with the results predicted by the crude oil blending and production process simulation model, and the prediction deviation of key indicators is calculated. If the prediction deviation continues to exceed the allowable range, a model correction mechanism is triggered to calibrate the conversion rules or related model parameters in the crude oil blending and production process simulation model using the actual operating result data.

9. The selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in claim 8, characterized in that, The calibration of the conversion rules or correlation model parameters in the crude oil blending and production process simulation model using the actual operational result data includes: From the actual operating results data, extract the actual input properties, actual mixing ratio, and actual output yield and properties corresponding to the final crude oil mixing scheme; Input the actual input properties and actual mixing ratios into the current version of the crude oil mixing and production process simulation model to obtain the model's predicted output yield and properties. Construct a loss function that measures the difference between the output yield and properties predicted by the model and the actual output yield and properties; The adjustable parameters in the crude oil mixing and production process simulation model are adjusted using either gradient descent or nonlinear least squares methods to minimize the loss function and complete model calibration.

10. A selection system for multi-objective crude oil blending production in refining and chemical enterprises, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the selection method for multi-objective crude oil blending production in refining and chemical enterprises as described in any one of claims 1 to 9.

Images