A multi-objective optimization method for refinery production planning

Abstract

The present application relates to the technical field of refinery production process optimization, and more particularly to a multi-objective optimization method for refinery production planning.The method comprises the following steps: step S1, fitting a real boiling point distillation curve of crude oil, reconstructing the properties of the crude oil by using the fitted curve of the real boiling point distillation curve of the crude oil, and characterizing the uncertainty of the properties of the crude oil; step S2, generating a simulation scenario; step S3, establishing a mechanism model of a process device, and reconstructing the input-output relationship of the process device in combination with the fitted curve of the real boiling point distillation curve of the crude oil; step S4, setting a target function and a constraint condition, and establishing a multi-objective model for refinery production planning; and step S5, solving the multi-objective model for refinery production planning to obtain an optimal production scheme for the refinery production planning.The present application takes economic benefits, carbon dioxide emissions and energy consumption as the target functions, realizes multi-objective optimization of the refinery production planning, and has high data processing efficiency and high accuracy.

Description

Technical Field

[0001] This invention relates to the field of refinery production process optimization technology, and more specifically, to a multi-objective optimization method for refinery production planning. Background Technology

[0002] Refineries are the main production and operating entities in the petroleum refining industry. They use petroleum as raw material and employ physical separation and chemical reactions to obtain various petroleum products. The diversification of raw materials and products ensures that optimizing refinery production planning remains a highly popular research topic. Maximizing refinery economic benefits through restrictions on feedstock and output, control of process quantities, and other constraints remains a consistent research direction.

[0003] Past refinery production planning has largely focused on maximizing economic benefits, with some considerations including energy consumption in the hope of achieving the greatest economic benefit with the least energy consumption. These relatively simplistic calculation models and planning schemes have been effective in addressing the economic benefits of specific refineries, but they often lack consideration for pollutant treatment and are difficult to calculate accurately.

[0004] However, with the national promotion of energy conservation and emission reduction and the advocacy of a low-carbon economy, the previously neglected areas of energy consumption and carbon dioxide emissions have received unprecedented attention. If refineries generate excessive amounts of gaseous pollutants or consume excessive energy during production, their economic efficiency will inevitably be affected. To meet national low-carbon environmental protection standards and the "dual reduction" standard, refineries need to sacrifice significant economic benefits in pollutant treatment, which undoubtedly contradicts the original intention of maximizing economic efficiency. Therefore, establishing a mathematical model for energy consumption and carbon dioxide emissions and seeking the optimal solution among the three has become the current research objective.

[0005] Meanwhile, the uncertainty of crude oil properties has a significant impact on the profitability of refineries, and this uncertainty can be addressed through various methods. How to handle the uncertainty of crude oil properties is also a current research objective. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-objective optimization method for refinery production planning based on the reconstruction of the input-output relationship between crude oil and process equipment. Under the premise of considering the uncertainty of crude oil properties, this method solves the problems of single objective, low data processing efficiency and poor accuracy in existing refinery production planning.

[0007] To achieve the above objectives, the present invention provides a multi-objective optimization method for refinery production planning, comprising the following steps:

[0008] Step S1: Fit the crude oil true boiling point distillation curve, reconstruct the crude oil properties using the fitted curve, and characterize the uncertainty of the crude oil properties.

[0009] Step S2: Generate a simulation scene;

[0010] Step S3: Establish a process unit mechanism model and reconstruct the input-output relationship of the process unit by combining the fitting curve of the crude oil actual boiling point distillation curve.

[0011] Step S4: Set the objective function and constraints to establish a multi-objective model for refinery production planning;

[0012] Step S5: Solve the multi-objective model of the refinery production plan to obtain the optimal production scheme of the refinery production plan.

[0013] In one embodiment, fitting the crude oil true boiling point distillation curve in step S1 further includes:

[0014] The beta function was used to fit the actual boiling point distillation curve of crude oil. The expression corresponding to the beta function is as follows:

[0015]

[0016] Where α and β are positive parameters controlling the shape of the distribution function, A and B are the lower and upper bounds of the distribution function, x is the normalized recovery temperature, and Γ is the standard gamma function.

[0017] In one embodiment, step S2 further includes the following steps:

[0018] The moment estimation method is used to obtain the candidate points of the scene;

[0019] The scene is constructed by sampling random vectors;

[0020] The constructed scene is normalized to obtain the normalized probability of the scene.

[0021] In one embodiment, step S3 further includes the following steps:

[0022] For atmospheric and vacuum distillation units, a mechanism model of atmospheric and vacuum distillation units is established, and the input-output relationship of atmospheric and vacuum distillation units is reconstructed by combining the mechanism model of atmospheric and vacuum distillation units with the fitting curve of crude oil actual boiling point distillation curve.

[0023] For secondary processing units, a mechanism model of the secondary processing unit is established and corrected. The input-output relationship of the secondary processing unit is reconstructed by combining the corrected mechanism model with the fitting curve of the crude oil true boiling point distillation curve.

[0024] In one embodiment, the establishment and correction of the mechanism model of the secondary processing device further includes using a neural network to correct the mechanism model;

[0025] The step of using a neural network to correct the mechanistic model further includes the following steps:

[0026] Preprocess the actual working condition data to determine the training set and test set;

[0027] The neural network model is trained using training and testing sets.

[0028] The trained neural network model outputs a predicted product yield, which is then used to correct the product yield output of the mechanistic model.

[0029] In one embodiment, the objective function is to maximize economic benefits, minimize carbon dioxide emissions, and minimize energy consumption;

[0030] The constraints are supply and demand, processing capacity, and storage tank conditions.

[0031] In one embodiment, the expression corresponding to the economic benefit model is:

[0032]

[0033] Among them, P s C s and C u These represent the selling price of each final product, the cost of raw materials, and the operating costs associated with each processing unit, respectively.

[0034] MS s and MU u These represent the mass flow rate of the material flow s and the total inlet mass flow rate processed by the processing unit u, respectively.

[0035] In one embodiment, the expression corresponding to the carbon dioxide emission model is:

[0036]

[0037] Where, ξ i η is the energy consumption coefficient. i a is the production coefficient. fuel For fuel percentage, The conversion factor for standard tons of fuel, including coal, fuel oil, and natural gas. The conversion factor between electricity, steam, or water. This refers to the carbon dioxide emissions per ton of standard fuel.

[0038] In one embodiment, the energy consumption model is established using an artificial neural network, specifically including the following steps:

[0039] Collect relevant processing volume data and historical energy consumption data from the refinery;

[0040] Using refinery-related processing volume data as input samples and corresponding historical energy consumption data as output samples, an artificial neural network is trained to obtain a data-driven refinery energy consumption model.

[0041] In one embodiment, step S5 further includes solving the multi-objective model of the refinery production plan using the NSGA-II algorithm;

[0042] The process of solving the multi-objective model of refinery production planning using the NSGA-II algorithm further includes the following steps:

[0043] Randomly generate an initial population of size N;

[0044] After non-dominated sorting, the next generation is obtained through selection, crossover, and mutation operations using a genetic algorithm.

[0045] Starting from the second generation, the parent and offspring populations are merged and non-dominated sorting is performed. At the same time, crowding is calculated for individuals in each non-dominated layer.

[0046] A new parent population is obtained based on non-dominant relationships and the crowding of individuals;

[0047] The NSGA-II algorithm is used to evaluate the results of the rigorous model through constraints and objective function, and a new subpopulation is generated. The subpopulation is sorted according to non-dominated and crowded distance. Binary tournament selection is used to handle constraints until the maximum number of generations is satisfied, and the Pareto solution is output.

[0048] Otherwise, new populations will continue to be generated and used as input for rigorous simulation of the rigorous model.

[0049] This invention provides a multi-objective optimization method for refinery production planning, which uses economic benefits, carbon dioxide emissions, and energy consumption as objective functions to achieve multi-objective optimization of refinery production planning. It has high data processing efficiency and high accuracy. Attached Figure Description

[0050] The above and other features, properties and advantages of the present invention will become more apparent from the following description taken in conjunction with the accompanying drawings and embodiments, in which the same reference numerals always denote the same features, wherein:

[0051] Figure 1 A flowchart illustrating a multi-objective optimization method for refinery production planning according to an embodiment of the present invention is disclosed.

[0052] Figure 2 A simplified refinery flow chart according to an embodiment of the present invention is disclosed;

[0053] Figure 3 A crude oil true boiling point distillation curve according to an embodiment of the present invention is disclosed;

[0054] Figure 4 A schematic diagram of a catalytic cracking mechanism model according to an embodiment of the present invention is disclosed;

[0055] Figure 5 A flowchart of the algorithm solution according to an embodiment of the present invention is disclosed;

[0056] Figure 6 A schematic diagram illustrating the solution result of the algorithm according to an embodiment of the present invention is shown. Detailed Implementation

[0057] 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 for illustrative purposes only and are not intended to limit the invention.

[0058] Figure 1 A flowchart illustrating a multi-objective optimization method for refinery production planning according to an embodiment of the present invention is disclosed, as follows: Figure 1 As shown, the multi-objective optimization method for refinery production planning proposed in this invention includes the following steps:

[0059] Step S1: Fit the crude oil true boiling point distillation curve, reconstruct the crude oil properties using the fitted curve, and characterize the uncertainty of the crude oil properties.

[0060] Step S2: Generate a simulation scene;

[0061] Step S3: Establish a process unit mechanism model and reconstruct the input-output relationship of the process unit by combining the fitting curve of the crude oil actual boiling point distillation curve.

[0062] Step S4: Set the objective function and constraints to establish a multi-objective model for refinery production planning;

[0063] Step S5: Solve the multi-objective model of the refinery production plan to obtain the optimal production scheme of the refinery production plan.

[0064] This invention proposes a multi-objective optimization method for refinery production planning based on the reconstruction of crude oil and equipment input-output relationships. Considering the uncertainty of crude oil properties, the method reconstructs crude oil properties using the crude oil distillation true boiling point curve, and reconstructs the equipment input-output relationships by combining the crude oil distillation true boiling point curve and a corrected mechanistic model. Simultaneously, it comprehensively evaluates and calculates three objective functions (maximizing economic benefits, minimizing energy consumption, and minimizing carbon dioxide emissions) to establish the final multi-objective optimization model for refinery production planning. A multi-objective optimization algorithm is then used to solve the multi-objective optimization problem of refinery production planning, yielding the optimal production scheme for the refinery production plan.

[0065] Figure 2 A simplified refinery flow chart according to an embodiment of the present invention is disclosed, such as Figure 2 As shown, in the refinery process of this embodiment, the main raw materials of the refinery are mixed crude oil (MX1) as well as heavy aromatics (BHR), methanol (BME), hydrogen (BH2), reforming feedstock (BRF), MTBE (BMT), and ethylbenzene (BEB).

[0066] The processing unit comprises 28 units in total: two atmospheric and vacuum distillation units (CDU1, CDU2), one light hydrocarbon recovery unit (CUT), one reset pre-hydrogenation unit (TP1), one benzene extraction combined unit (AEX), two catalytic cracking units (FCC1, FCC2), one slurry tapping unit (SVTP), one high-pressure hydrocracking unit (HCU), one delayed coking unit (CK1), one solvent deasphalting unit (SAU), eight hydrotreating units (HF1-HF8), two gas separation units (GF), one polypropylene unit (SPPU), one MTBE unit, one sulfur recovery unit (PSA), one ethylbenzene unit (SEBN), one styrene unit (SSTR), and one alkylation unit (SALK).

[0067] Storage tanks include those for gasoline, diesel, aviation fuel, and naphtha.

[0068] A total of 13 final products were produced, namely:

[0069] 92-octane gasoline (W92), 95-octane gasoline (W95), jet fuel (JET), diesel (W00), ethylene naphtha (EEN), benzene (BNZ), xylene (XYL), commercial liquefied petroleum gas (PLG), polypropylene (PPP), petroleum coke (CCK), sulfur (SULH), liquid ammonia (NH3), and styrene (STR).

[0070] A total of 7 raw materials were purchased from external sources, namely:

[0071] Pipeline blend (MX1), heavy aromatics (BHR), methanol (BME), hydrogen (H2), reforming feedstock (BRF), MTBE (BMT), and ethylbenzene (BEB).

[0072] The following text is in the format of Figure 2 The simplified refinery shown is an example to illustrate these steps of the multi-objective optimization method for refinery production planning based on the reconstruction of crude oil and equipment input-output relationships proposed in this invention. It should be understood that, within the scope of this invention, the above-described technical features of this invention and the technical features specifically described below (such as in the examples) can be combined and correlated with each other to constitute preferred technical solutions.

[0073] Step S1: Fit the crude oil true boiling point distillation curve, and use the fitted curve of the crude oil true boiling point distillation curve to reconstruct the crude oil properties and characterize the uncertainty of the crude oil properties.

[0074] The crude oil true boiling point distillation curve is fitted using a probability density function, and the crude oil properties are reconstructed using the crude oil true boiling point distillation curve. At the same time, the uncertainty of the crude oil properties is characterized using the fitted curve.

[0075] Assuming the crude oil analysis data is known, its properties remain uncertain until it reaches the processing plant.

[0076] The crude oil true boiling point distillation curve is fitted, and the crude oil properties are reconstructed using the crude oil true boiling point distillation curve. In this embodiment, the beta function is used to fit the crude oil true boiling point distillation curve.

[0077] Given the known crude oil analysis data, the properties of crude oil can be reconstructed using the beta function. The expression for the beta function is:

[0078]

[0079] Where α and β are positive parameters controlling the shape of the distribution function, A and B are the lower and upper bounds of the distribution function, x is the normalized recovery temperature, and Γ is the standard gamma function.

[0080] Determine the optimal values ​​of the beta function parameters (A, B, α, β), and determine the normal distribution parameters of parameters α and β.

[0081] Figure 3 This is a crude oil true boiling point distillation curve diagram according to the first embodiment of the present invention, with... Figure 3 Taking crude oil CO1 as an example, the fitting process based on the least squares method is illustrated:

[0082] (1) Determination of parameters A, B, α, and β:

[0083] Since parameters A and B determine the upper and lower bounds of the distribution function, A and B are fixed at the optimal positions on the crude oil true boiling point distillation curve (TBP) of the target crude oil.

[0084] It is obtained that A=-0.0313 and B=0.9782. α=0.4554, β=0.3668.

[0085] (2) Determining x:

[0086] x represents the normalized temperature recovery value, and the upper and lower limits of the normalized temperature recovery are respectively:

[0087] =-300℉, =1800℉, the formula for analyzing crude oil data is obtained by fitting.

[0088] (3) Characterization of uncertainties in crude oil properties:

[0089] The description of crude oil properties using fitted curves involves using the uncertain positive parameters α and β in the beta function to characterize the uncertainty of crude oil properties. This assumes that the crude oil properties follow a normal distribution, in which case the parameters α and β also follow a normal distribution.

[0090] α and β determine the shape of TBP, i.e. the properties of crude oil. Since the properties of crude oil roughly follow a normal distribution, the parameters α and β also follow a normal distribution. The mean of the normal distribution is set as the optimal value of the uncertain parameters α and β for fitting the crude oil TBP curve.

[0091] In this embodiment, σ = 0.002 is a mitigating value that satisfies the assumption and will not change the quality of the crude oil. Therefore, the uncertainty of the crude oil properties is transformed into a probability distribution problem of the positive parameters α and β in the beta function.

[0092] Explanation of issues at both ends of the curve: Since the beta function still has some error in fitting, especially at both ends of the curve, in this embodiment, the initial boiling point (IBP) at the front end and the final boiling point (FBP) at the back end are set as constants.

[0093] Step S2: Generate a simulation scene.

[0094] To address the difficulty of solving numerous scenarios caused by Monte Carlo sampling, this embodiment utilizes the method of moments estimation to significantly reduce the number of scenarios and employs random vector sampling (RVS) to obtain the values ​​of the discrete points used, ensuring that each scenario can be obtained.

[0095] In this embodiment, scene generation further includes moment estimation, random vector sampling, and normalization processing:

[0096] The moment estimation step involves using the moment estimation method to obtain the candidate points of the scene.

[0097] If the Monte Carlo method is used for sample average approximation, a large number of scenes need to be generated to ensure that even points with low probability can be captured, and the model size will inevitably be large.

[0098] In this embodiment, the method of moment estimation is used to discretize continuous variables. Based on Gaussian orthogonal moment estimation, (2N-1) center moments can be accurately estimated using N points. Therefore, only three points are needed to accurately estimate the mean, variance, skewness, and kurtosis. There are only three candidate points in the probability set, avoiding the large number of scenarios generated by the Monte Carlo method.

[0099] The random vector sampling step is used to construct the scene through random vector sampling.

[0100] Since there are only three points in the probability set, and we want all three points to be selected to construct these few scenarios, random vector sampling can meet the requirements.

[0101] For these three points, RVS constructs three scenarios each time, with one of the three points serving as the implementation of uncertain parameters.

[0102] The permutations of these three points (v1, v2, v3) constitute the scene tree, as shown in equation (1):

[0103] Ω(s1, s2, s3) = {(v1, v2, v3), (v1, v3, v2), (v2, v1, v3), (v2, v3, v1), (v3, v1, v2), (v3, v2, v1)} ​​(1)

[0104] Normalization steps: After all scenarios S are generated, for the k-th set of all scenarios, the probability that the uncertain parameter p is realized at the relevant value for scenario s is denoted as... The probability of each scenario is calculated using a probability normalization method. As shown in equation (2):

[0105]

[0106] Where, N r K is the normalization notation symbol, and K is the total number of sets.

[0107] The probability of scenario s in the k-th set of all cases is denoted as . The probabilities of the three cases in each set k are calculated using the formula shown in equation (3).

[0108]

[0109] Then, the three cases s = k, k+1, k+2 in the k-th set are normalized as shown in equation (4):

[0110]

[0111] Finally, the normalized probability of scenario s is shown in equation (5):

[0112]

[0113] Step S3: Establish a process unit mechanism model and reconstruct the input-output relationship of the process unit by combining the fitting curve of the crude oil actual boiling point distillation curve.

[0114] Since the operation of a process unit depends on the quality of the feed stream, such as crude oil as feed for an atmospheric and vacuum distillation unit (CDU) or distillate fractions as feed for a catalytic cracking unit (FCC), a mechanistic model of the process unit can be used to address the uncertainties of crude oil. Simultaneously, the properties of crude oil can be reconstructed using the actual boiling point curves of crude oil distillation.

[0115] In this embodiment, HYSYS software is used to establish the corresponding process unit mechanism model. The process unit mechanism model includes, but is not limited to, the mechanism model of atmospheric and vacuum distillation unit (CDU), catalytic cracking (FCC), hydrocracking (HCU) and other secondary processing units.

[0116] The uncertainty of crude oil properties manifests as changes in the crude oil distillation curve, which in turn leads to changes in the output of CDU products from atmospheric and vacuum distillation units.

[0117] Therefore, in this embodiment, the process unit mechanism model is combined with the fitting curve of the crude oil true boiling point distillation curve to reconstruct the input-output relationship of the process unit, so as to obtain more accurate product yield and unit input-output data.

[0118] Furthermore, for the atmospheric and vacuum distillation unit (CDU) mechanism model, its input-output relationship is reconstructed by combining the fitting curve of the crude oil actual boiling point distillation curve;

[0119] Mechanism models were established for secondary processing units such as catalytic cracking (FCC) and hydrocracking (HCU), and the mechanism models were corrected. The input-output relationship was reconstructed by combining the corrected mechanism models with the fitting curves of crude oil true boiling point distillation curves to obtain more accurate product yields.

[0120] Since the accuracy of the atmospheric and vacuum distillation unit (CDU) mechanism model is related to the accuracy of the intermediate flow production prediction, and thus affects the entire downstream processing, this embodiment reconstructs the input-output relationship of the atmospheric and vacuum distillation unit (CDU) by using the mechanism model and the fitting curve of the crude oil actual boiling point distillation curve considering the uncertainty of crude oil properties.

[0121] Meanwhile, in order to address the impact of changes in the model yield of the atmospheric and vacuum distillation unit (CDU), a corrected mechanistic model was used to reconstruct the input-output relationship of secondary processing units in refineries, such as the catalytic cracking (FCC) unit and the hydrocracking (HCU) unit.

[0122] The following section provides a detailed explanation of the establishment and calibration of the mechanistic model for a catalytic cracking (FCC) unit. Other process units are similar and will not be elaborated upon here.

[0123] Figure 4 A schematic diagram of a catalytic cracking mechanism model according to an embodiment of the present invention is disclosed, such as... Figure 4 The catalytic cracking (FCC) unit shown consists of a reaction-regeneration system, a separation system, and auxiliary equipment.

[0124] A mechanistic model of a catalytic cracking (FCC) unit was established using HYSYS software. The catalytic cracking module of HYSYS software has 21 built-in lumped kinetic models. The data such as feedstock properties, catalyst, equipment parameters, and operating parameters of the unit were input, and the default reaction kinetic parameters were used for initial simulation.

[0125] The reactor feeds 70% atmospheric residue oil and 30% atmospheric wax oil and coking wax oil.

[0126] The high-temperature reaction oil and gas from the settling tank enters the main fractionation tower T101. After separation, the top product of the tower is a mixture of crude gasoline and rich gas, the side stripping tower T101-1 produces light diesel oil, and the bottom product is oil slurry.

[0127] The feed to the absorber T201 is crude gasoline, compressed rich gas, and supplementary absorbent from the stabilizer T204. The top output is lean gas, and the bottom output is rich absorbed gasoline, which is mixed with compressed rich gas and desorbed gas from the top of the desorption tower T202 and enters the flash tank.

[0128] The function of the desorption tower T202 is to desorb the condensed oil from the flash tank, and the output is deethaned gasoline which goes to the stabilizer tower.

[0129] The feed to the reabsorption tower T203 is lean gas and part of the light diesel oil from the main fractionation tower. The top product is dry gas and the bottom product is rich absorbent oil, which is returned to the main fractionation tower.

[0130] The feed to stabilizer T204 is deethaned gasoline, the top product is liquefied petroleum gas (LPG), and the bottom product is stabilized gasoline.

[0131] The entire separation system consists of four cycles: the bottom of the absorption tower is rich in absorbed gasoline, the bottom of the reabsorption tower is rich in absorbed oil, the top of the desorption tower is rich in desorbed gas, and the stabilizer tower uses part of the stabilized gasoline as a reabsorbent.

[0132] For the mechanistic model of a catalytic cracking (FCC) unit, a neural network is used to correct the mechanistic model, specifically including the following steps:

[0133] Preprocess the actual working condition data to determine the training set and test set;

[0134] The neural network model is trained using training and testing sets.

[0135] The error in the predicted product yield output by the trained neural network model is used to correct the product yield output by the mechanistic model.

[0136] The data preprocessing steps further include erroneous data removal, normalization, and determination of the training and test sets.

[0137] In this embodiment, actual factory data for a given year is collected monthly, with 20 sets per month, totaling 240 sets. Data preprocessing is performed as follows:

[0138] (1) Removal of erroneous data.

[0139] Due to various reasons, there may be some erroneous data in the dataset. In order to prevent it from affecting the training quality, the erroneous data is deleted first using prior theoretical knowledge of catalytic cracking reaction.

[0140] (2) Normalization process.

[0141] The difference in the span of each data point in the dataset will affect the convergence speed and accuracy of the model. Therefore, it is necessary to normalize the data in the dataset. The specific formula is shown in equation (6):

[0142]

[0143] In the formula, X represents the data before normalization. norm For the normalized data, X max ,X min These are the maximum and minimum values ​​in the data, respectively.

[0144] (3) Determine the training set and the test set.

[0145] The processed dataset is randomly shuffled, and then the training and test sets are divided in a 3:1 ratio.

[0146] In this embodiment, 16 input variables, including vanadium content, residual carbon content, and saturated hydrocarbon content, were selected, and the output variable was chosen as the yield of the main product.

[0147] The model was trained using actual working condition data. 180 of the 240 sets of data were used as training samples, and the remaining 60 sets were used as validation samples. The model was trained with 5 to 13 hidden layer neurons. When the number of hidden layer neurons was 5 to 7, the relative error of the model gradually decreased. When the number of hidden layer neurons was 7 to 13, the relative error of the model gradually increased. Obviously, the model accuracy was the highest when the number of hidden layer neurons was 7.

[0148] The neural network model was trained using training and testing sets, and the final neural network parameters are shown in Table 1.

[0149] Table 1 Parameters of Artificial Neural Networks

[0150] Training batch 64 Number of neurons in the input layer 16 Training times 1000 Number of neurons in hidden layer 7 Setting error 0.001 Number of neurons in the output layer 5

[0151] The step of using the predicted product yield error output by the trained neural network model to correct the product yield output by the mechanistic model further includes:

[0152] Actual operating data are input into the mechanistic model and the neural network model respectively to obtain the product yield distribution and the predicted product yield error. The product yield distribution output by the mechanistic model and the error predicted by the neural network model are superimposed to obtain the output of the calibrated model.

[0153] Table 2 shows the relative errors of the mechanistic model and the corrected model after neural network correction. As can be seen from Table 2, the average relative error decreased from 4.74 to 3.41, a decrease of 28%, which proves that the corrected model has a certain degree of effectiveness.

[0154] Table 2 Comparison of Model Product Yield Predictions

[0155]

[0156] Step S4: Set the objective function and constraints to establish a multi-objective model for refinery production planning;

[0157] With the objective functions of maximizing economic benefits, minimizing carbon dioxide emissions, and minimizing energy consumption, and with supply and demand, processing capacity, and storage tank conditions as constraints, a multi-objective model for refinery production planning is established.

[0158] Profit equals total revenue from the final product minus total cost of raw material procurement and total operating cost of the processing unit. Throughout the refinery, U represents a unit, B represents a tank, and S represents streams, including raw materials, intermediate streams, and finished products. The inlet streams of processing units and tanks are represented by sets UI = {(u,s)} and BI = {(b,s)}, respectively; similarly, the outlet streams are represented by sets UO and BO. The set E represents the critical properties of the final critical product.

[0159] In this embodiment, the economic benefit model of the established refinery production plan multi-objective model is shown in the following formula (7).

[0160] max∑ s∈FPS P s MS s -∑ s∈RMS C s MS s -∑ u∈U C u ·MU u (7);

[0161] Among them, P s C s and C u These represent the selling price of each final product, the procurement cost of raw materials, and the operating costs associated with each processing unit, respectively; FPS represents product flow, and RMS represents raw material flow.

[0162] MS s and MU u These represent the mass flow rate of the material flow s and the total inlet mass flow rate processed by the processing unit u, respectively.

[0163] In this embodiment, the carbon dioxide emission model in the multi-objective model of refinery production planning is established by combining two analysis methods: material flow and energy flow.

[0164] The material flow analysis method specifically includes the following steps:

[0165] Because the fuel refining process is relatively complex, a combination of stationary observation and follow-up observation methods was used for the study. Under normal operating conditions, the analytical boundary included the production process and operating procedures. Carbon was selected as the object of observation.

[0166] Gasoline and diesel production units were selected as observation areas to analyze changes in material flow in these areas. Material consumption during the production process depends not only on the total amount of material inflow but also on the demand of upstream products on downstream production units.

[0167] Therefore, material consumption takes into account both the raw material consumption of the production unit and the impact of downstream production units.

[0168] In this embodiment, there are 6 kinds of material flows in the refinery production unit, and the material balance of the production unit is shown in Equation (8).

[0169] F i +A i =P ie +P i +W i +R i(8)

[0170] Among them, F i For the flow rate of other substances at the inlet of the device, A i P is the feed flow rate at the inlet of the device. ie P represents the product flow rate entering downstream equipment. i For the product flow rate of the device, W i For the export waste flow, R i This refers to the circulating material flow rate of the device.

[0171] The production coefficient of production unit i (the ratio of product output to raw material consumption of production unit i) is expressed by equation (9).

[0172]

[0173] Where, η i Let be the production coefficient of production unit i.

[0174] The energy flow analysis method specifically includes the following steps:

[0175] Due to the complexity of raw materials and products, the chemical reactions within the production unit are also complex. In conventional reaction engineering methods, it is difficult to establish suitable energy conservation equations for the production facility. However, according to systems theory, the production unit is considered a complete reactor system; only in this way can energy balance be theoretically established.

[0176] In this embodiment, energy flow is divided into 7 types according to the source, destination and function of energy.

[0177] The energy balance of production unit i is represented by equation (10), with the energy flow unit being tons of standard fuel oil.

[0178] B i +E i +E ri =E ni +E i+1 +L i +S i (10)

[0179] Among them, B i E represents the energy flowing in from other devices. i E represents the energy E of the product from device i-1 entering device i. ri To recover energy, E ni E represents the total energy flowing into the device. i+1 L represents the energy entering device i from device i+1. i To lose energy, S i Energy flowing into other devices.

[0180] Energy consumption in a production unit during the refining process typically refers to the difference between the energy flowing in and the energy flowing out, such as electricity, steam, and water.

[0181] Energy consumption coefficient ξ of the production unit i It is represented by equation (11).

[0182] ξ i =B i -S i =E ni +E i+1 +L i -E i -E ri (11)

[0183] Where, ξ i B is the energy consumption coefficient of the production unit. i E represents the energy flowing in from other devices. i E represents the energy E of the product from device i-1 entering device i. ri To recover energy, E ni E represents the total energy flowing into the device. i+1 L represents the energy entering device i from device i+1. i To lose energy, S i Energy flowing into other devices.

[0184] Carbon dioxide emissions from oil refineries are influenced by factors such as energy structure and consumption, fuel heat output and carbon content, carbon dioxide emissions per unit of fuel, and pollution control levels.

[0185] In the carbon dioxide emission model, the purchased energy takes into account both fuel and electricity, assuming that the fuel is converted into carbon dioxide through combustion and eventually released into the environment.

[0186] Total carbon dioxide emissions Equation (12) represents:

[0187]

[0188] Direct emissions of carbon dioxide V Direct Equation (13) represents:

[0189]

[0190] Among them, a fuel For fuel percentage, The conversion factor between electricity, steam, or water. ξ represents the carbon dioxide emissions per ton of standard fuel. i η is the energy consumption coefficient. i This is the production coefficient.

[0191] Indirect carbon dioxide emissions V Indirect Equation (14) represents:

[0192]

[0193] Among them, a fuel For fuel percentage, v is the conversion factor between electricity, steam, or water. CO2 ξ represents the carbon dioxide emissions per ton of standard fuel. i η is the energy consumption coefficient. i This is the production coefficient.

[0194] The total carbon dioxide emissions of production unit i are represented by equation (15), that is, the expression of the carbon dioxide emission model is as follows:

[0195]

[0196] Where, ξ i η is the energy consumption coefficient. i a is the production coefficient. fuel For fuel percentage, The conversion factor for standard tons of fuel, including coal, fuel oil, and natural gas. v is the conversion factor between electricity, steam, and water. CO2 Carbon dioxide emissions per ton of standard fuel (t).

[0197] In this embodiment, the energy consumption model of the established refinery production plan multi-objective model is established using an artificial neural network, specifically including the following steps:

[0198] Collect relevant processing volume data and historical energy consumption data from oil refineries;

[0199] Using artificial neural network technology, with refinery processing volume data as input samples and corresponding energy consumption data as output samples, the artificial neural network is trained to obtain a data-driven refinery energy consumption model.

[0200] In this embodiment, a BP neural network is selected to fit the relevant data. The BP neural network mainly consists of an input layer, a hidden layer, and an output layer.

[0201] The learning process of a BP neural network consists of forward propagation of the signal and backward propagation of the error. The network training is complete when the trained model reaches the minimum root mean square error (RMS).

[0202] Specifically, the value of RMS is:

[0203]

[0204] Where i = 1, 2, ..., m, m is the number of training iterations, j = ... For the estimated value output by the model, O ij The actual values ​​output by the process model.

[0205] In this embodiment, the established multi-objective model for refinery production planning includes constraints such as supply and demand, processing capacity, and storage tank conditions.

[0206] The constraints of the supply and demand relationship are expressed by formulas (17) and (18):

[0207] Formula (17) describes the restriction of the supply of each raw material during the planning period;

[0208]

[0209] in, and These represent the minimum and maximum values ​​of raw materials purchased by the refinery, respectively.

[0210] Formula (18) describes the market demand for various products.

[0211]

[0212] in, and These represent the minimum and maximum market demand for the product, respectively.

[0213] The expressions corresponding to the constraints on processing capacity are Equations (19) to (23).

[0214] Formula (19) describes the total inlet material flow into each processing unit;

[0215]

[0216] Formula (20) describes the limitation on the feed flow rate of the processing unit;

[0217]

[0218] in, and These represent the maximum and minimum processing capacity of the device, respectively.

[0219] Formulas (21) and (22) use the variable yield method to calculate the output mass flow rate of each product stream. Processing units such as FCC and HCU, represented by the MMU, have multiple working modes, corresponding to different product yields.

[0220]

[0221]

[0222] A binary variable is introduced to represent the choice of different operation modes for each unit. The set MO represents the operation mode. Formula (23) forces only one production mode to be operated within the planning cycle.

[0223]

[0224] Where, x u,m This indicates the selection of different operating modes for each device.

[0225] The expressions corresponding to the constraints of the storage tank conditions are Equations (24)-(28).

[0226] Formula (24) calculates the flow rates of various final products:

[0227]

[0228] Among them, MS s′ MS represents the inlet flow rate of each storage tank. s This represents the flow of various final products;

[0229] Formula (25) represents the relationship between mass flow rate and volumetric flow rate.

[0230]

[0231] SPG is an abbreviation for proportion.

[0232] The mass characteristics are assumed to be linearly mixed on a mass or volume basis, as shown in equations (26)-(28).

[0233]

[0234]

[0235]

[0236] in, and These represent the worst and best properties of product s, respectively.

[0237] Step S5: Solve the multi-objective model of the refinery production plan to obtain the optimal production scheme of the refinery production plan.

[0238] In this embodiment, the NSGA-II algorithm is used to solve the multi-objective optimization model of the refinery production plan to obtain the optimal production scheme.

[0239] The NSGA-II algorithm selects initial variables and assigns them to the HYSYS software for calculation. Based on the simulation results and the optimized solution, it continuously corrects the optimization search direction until the convergence condition is met and the Pareto optimal solution is obtained, thus obtaining the optimal production plan.

[0240] The NSGA-II algorithm is one of the most popular multi-objective genetic algorithms. It reduces the complexity of non-dominated sorting genetic algorithms and has the advantages of fast running speed and good convergence of solution sets, making it the benchmark for the performance of other multi-objective optimization algorithms.

[0241] In this embodiment, MATLAB sends the variables of each individual in the population to HYSYS one by one for simulation calculation. If HYSYS fails to converge or malfunctions, the individual will be discarded with the help of HYSYS running status parameters.

[0242] In this embodiment, the known conditions of the production planning optimization model include the following conditions:

[0243] Production planning cycle; availability of blended crude oil and intermediate feedstocks; maximum and minimum throughput of the processing unit and its operating mode; demand and performance specifications of the final product; sales price of each final product; procurement cost of each raw material; operating cost of each processing unit.

[0244] In this embodiment, the decision variables of the model include the following variables:

[0245] The quantity of raw materials purchased; the total feed rate of the processing equipment and the appropriate operating method; the flow rate of each intermediate; the quantity and properties of the product.

[0246] During the solution process, the population size, number of generations, crossover probability, and mutation probability are parameters that need to be set. The parameter settings are as follows:

[0247] Table 3 Solver parameter settings

[0248] numerical values 125 10000 0.8 0.1

[0249] Figure 5 A flowchart of the algorithm solution according to an embodiment of the present invention is disclosed, such as... Figure 5 As shown, the NSGA-II algorithm is used to solve the multi-objective model of refinery production planning, and the following steps are further included:

[0250] Randomly generate an initial population of size N;

[0251] After non-dominated sorting, the next generation is obtained through three basic operations of genetic algorithm: selection, crossover, and mutation.

[0252] Starting from the second generation, the parent and offspring populations are merged and non-dominated sorting is performed. At the same time, crowding is calculated for individuals in each non-dominated layer.

[0253] A new parent population is obtained based on non-dominant relationships and the crowding of individuals;

[0254] The NSGA-II algorithm evaluates the results of the HYSYS rigorous model by considering constraints and objective function, generates new subpopulations, sorts them according to non-dominated and crowded distances, uses binary tournament selection to handle constraints, and outputs the Pareto solution until the maximum number of generations is satisfied.

[0255] Otherwise, new populations will continue to be generated and used as input for rigorous simulation of the HYSYS rigorous model.

[0256] Figure 6 A schematic diagram of the algorithm solution result according to an embodiment of the present invention is disclosed. Under the premise of ensuring convergence, the genetic algorithm parameters are tuned. The process aims to minimize the number of generations, avoid local optima, and maximize the diversity of Pareto solutions to solve the algorithm optimization result.

[0257] The specific optimized production plan is shown in Table 4.

[0258] Table 4 Optimization values ​​of decision variables

[0259]

[0260] Compared with existing technologies, this invention provides a multi-objective optimization method for refinery production planning based on the reconstruction of the input-output relationship between crude oil and process equipment. It takes into account the uncertainty of crude oil properties and uses a corrected mechanistic model and crude oil true boiling point distillation curve to reconstruct the input-output relationship of the equipment, thereby achieving multi-objective optimization of refinery production planning and making the prediction of product yield more accurate.

[0261] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.

[0262] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0263] The above embodiments are provided for those skilled in the art to implement or use the present invention. Those skilled in the art can make various modifications or changes to the above embodiments without departing from the inventive concept of the present invention. Therefore, the protection scope of the present invention is not limited to the above embodiments, but should be the maximum scope that conforms to the innovative features mentioned in the claims.

Claims

1. A multi-objective optimization method for refinery production planning, characterized in that, Includes the following steps: Step S1: Fit the crude oil true boiling point distillation curve, reconstruct the crude oil properties using the fitted curve, and characterize the uncertainty of the crude oil properties. Step S2: Generate a simulation scene; Step S3: Establish a process unit mechanism model and reconstruct the input-output relationship of the process unit by combining the fitting curve of the crude oil actual boiling point distillation curve. Step S4: Set the objective function and constraints to establish a multi-objective model for refinery production planning; Step S5: Solve the multi-objective model of the refinery production plan to obtain the optimal production scheme of the refinery production plan; Step S2 further includes the following steps: The moment estimation method is used to obtain the candidate points of the scene; The scene is constructed by sampling random vectors; The constructed scene is normalized to obtain the normalized probability of the scene; Step S3 further includes the following steps: For atmospheric and vacuum distillation units, a mechanism model of atmospheric and vacuum distillation units is established, and the input-output relationship of atmospheric and vacuum distillation units is reconstructed by combining the mechanism model of atmospheric and vacuum distillation units with the fitting curve of crude oil actual boiling point distillation curve. For the secondary processing unit, a mechanism model of the secondary processing unit is established and corrected. The input-output relationship of the secondary processing unit is reconstructed by combining the corrected mechanism model with the fitting curve of the crude oil true boiling point distillation curve. The establishment and correction of the mechanism model of the secondary processing device further includes using a neural network to correct the mechanism model; The step of using a neural network to correct the mechanistic model further includes the following steps: Preprocess the actual working condition data to determine the training set and test set; The neural network model is trained using training and testing sets. The error in the predicted product yield output by the trained neural network model is used to correct the product yield output by the mechanistic model. The objective function is to maximize economic benefits, minimize carbon dioxide emissions, and minimize energy consumption. The constraints are supply and demand, processing capacity, and storage tank conditions; The expression corresponding to the economic benefit model is: ； in, , and These represent the selling price of each final product, the cost of raw materials, and the operating costs associated with each processing unit, respectively. and These represent the mass flow rate of the material flow s and the total inlet mass flow rate processed by the processing unit u, respectively. The expression corresponding to the carbon dioxide emission model is: ； in, Energy consumption coefficient For production coefficient, For fuel percentage, The conversion factor for standard tons of fuel, including coal, fuel oil, and natural gas. The conversion factor between electricity, steam, or water. This refers to the carbon dioxide emissions per ton of standard fuel.

2. The multi-objective optimization method for refinery production planning according to claim 1, characterized in that, The fitting of the crude oil true boiling point distillation curve in step S1 further includes: The beta function was used to fit the actual boiling point distillation curve of crude oil. The expression corresponding to the beta function is as follows: ； Where α and β are positive parameters controlling the shape of the distribution function, and A and B are the lower and upper bounds of the distribution function, respectively. x The normalized recovery temperature is given by Γ, which is the standard gamma function.

3. The multi-objective optimization method for refinery production planning according to claim 1, characterized in that, The energy consumption model is established using an artificial neural network, and specifically includes the following steps: Collect relevant processing volume data and historical energy consumption data from the refinery; Using refinery-related processing volume data as input samples and corresponding historical energy consumption data as output samples, an artificial neural network is trained to obtain a data-driven refinery energy consumption model.

4. The multi-objective optimization method for refinery production planning according to claim 1, characterized in that, Step S5 further includes solving the multi-objective model of the refinery production plan using the NSGA-II algorithm; The process of solving the multi-objective model of refinery production planning using the NSGA-II algorithm further includes the following steps: Randomly generate an initial population of size N; After non-dominated sorting, the next generation is obtained through selection, crossover, and mutation operations using a genetic algorithm. Starting from the second generation, the parent and offspring populations are merged and non-dominated sorting is performed. At the same time, crowding is calculated for individuals in each non-dominated layer. A new parent population is obtained based on non-dominant relationships and the crowding of individuals; The NSGA-II algorithm is used to evaluate the results of the rigorous model through constraints and objective function, and a new subpopulation is generated. The subpopulation is sorted according to non-dominated and crowded distance. Binary tournament selection is used to handle constraints until the maximum number of generations is satisfied, and the Pareto solution is output. Otherwise, new populations will continue to be generated and used as input for rigorous simulation of the rigorous model.

Images