CN116843052A

CN116843052A - Multi-objective optimization method for refinery production plan

Info

Publication number: CN116843052A
Application number: CN202310575267.2A
Authority: CN
Inventors: 杜文莉; 隆建; 钱锋; 钟伟民; 杨明磊
Original assignee: East China University of Science and Technology
Current assignee: East China University of Science and Technology
Priority date: 2023-05-22
Filing date: 2023-05-22
Publication date: 2023-10-03

Abstract

The application relates to the technical field of optimization of refinery production processes, in particular to a multi-objective optimization method of a refinery production plan. The method comprises the following steps: step S1, fitting a crude oil true boiling point distillation curve, reconstructing crude oil properties by adopting the fitted curve of the crude oil true boiling point distillation curve, and carrying out uncertainty characterization on the crude oil properties; s2, generating a simulation scene; s3, establishing a mechanism model of the process device, and reconstructing the input-output relation of the process device by combining a fitting curve of a crude oil true boiling point distillation curve; s4, setting an objective function and constraint conditions, and establishing a multi-objective model of the refinery production plan; and S5, solving the multi-objective model of the refinery production plan to obtain an optimal production scheme of the refinery production plan. The application takes economic benefit, carbon dioxide emission and energy consumption as objective functions to realize multi-objective optimization of the refinery production plan, and has higher data processing efficiency and higher accuracy.

Description

Multi-objective optimization method for refinery production plan

Technical Field

The application relates to the technical field of optimization of refinery production processes, in particular to a multi-objective optimization method of a refinery production plan.

Background

The refinery is the main production management body in petroleum refining industry, and uses petroleum as raw material, and adopts the modes of physical separation and chemical reaction to obtain various petroleum products. The problem of optimizing the production plan of a refinery remains extremely high research heat throughout the year due to the diversity of raw materials and products. The economic benefit of the refinery is always the same research direction through the limitation of the feeding and discharging, the control of the process quantity and other constraints.

In past refinery production programs, most of the concerns have focused on maximization of economic efficiency, partly taking into account energy consumption, and it is desirable to produce the greatest economic efficiency with the smallest energy consumption. These relatively single, simplified modes of operation and planning schemes provide good results in dealing with economic benefits of a particular refinery, but often lack consideration in the handling of contaminants and are difficult to calculate accurately.

However, with the popularization of energy conservation and emission reduction and the advocation of low-carbon economy in China, the field of energy consumption and carbon dioxide emission which are always lack of attention is paid to unprecedented importance. Once the refinery is in production, gas pollutants are excessively generated or energy is excessively consumed, the economic benefit of the refinery is inevitably affected. In order to meet the national low-carbon and environmental-friendly standards and the "double-abatement" standards, refineries need to sacrifice a large amount of economic benefit in the treatment of pollutants, which is certainly not in line with the original purpose of production with maximized economic benefit. Therefore, how to build a mathematical model of energy consumption and carbon dioxide emission and find an optimal solution among the three is a current research goal.

Meanwhile, uncertainty of crude oil properties has a great influence on benefits of refineries, and the uncertainty of crude oil properties can be solved by different methods. Uncertainty in how to deal with crude oil properties is also a current research goal.

Disclosure of Invention

The application aims to provide a multi-objective optimization method for a refinery production plan based on reconstruction of input-output relation of crude oil and a process device, which solves the problems of single objective, low data processing efficiency, poor accuracy and the like of the refinery production plan in the prior art on the premise of considering uncertainty of crude oil properties.

In order to achieve the above object, the present application provides a multi-objective optimization method for a refinery production plan, comprising the steps of:

step S1, fitting a crude oil true boiling point distillation curve, reconstructing crude oil properties by adopting the fitted curve of the crude oil true boiling point distillation curve, and carrying out uncertainty characterization on the crude oil properties;

s2, generating a simulation scene;

s3, establishing a mechanism model of the process device, and reconstructing the input-output relation of the process device by combining a fitting curve of a crude oil true boiling point distillation curve;

s4, setting an objective function and constraint conditions, and establishing a multi-objective model of the refinery production plan;

and S5, solving the multi-objective model of the refinery production plan to obtain an optimal production scheme of the refinery production plan.

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

fitting a crude oil true boiling point distillation curve by using a beta function, wherein the beta function corresponds to the expression:

wherein 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.

In an embodiment, the step S2 further includes the steps of:

acquiring a point to be selected of a scene by adopting a moment estimation method;

constructing a scene through random vector sampling;

and carrying out normalization processing on the constructed scene to obtain the normalization probability of the scene.

In an embodiment, the step S3 further includes the steps of:

for an atmospheric and vacuum device, establishing an atmospheric and vacuum device mechanism model, and reconstructing the input-output relationship of the atmospheric and vacuum device by adopting the atmospheric and vacuum device mechanism model and a fitting curve of a crude oil true boiling point distillation curve;

and (3) for the secondary processing device, a mechanism model of the secondary processing device is established and corrected, and the input-output relation of the secondary processing device is reconstructed by adopting the corrected mechanism model and combining a fitting curve of a crude oil true boiling point distillation curve.

In one embodiment, the modeling and correcting the secondary processing device mechanism further includes correcting the mechanism model using a neural network;

the correction of the mechanism model using the neural network further comprises the steps of:

preprocessing actual working condition data to determine a training set and a testing set;

training a neural network model by adopting a training set and a testing set;

and outputting the predicted product yield of the trained neural network model, and correcting the output product yield of the mechanism model.

In one embodiment, the objective function is economic benefit maximization, carbon dioxide emission minimization, and energy consumption minimization;

the constraint conditions are supply and demand relation, processing capacity and storage tank conditions.

In an embodiment, the economic benefit model corresponds to the expression:

wherein P is _s 、C _s And C _u Representing sales price of each final product, purchasing cost of raw materials, and operation cost associated with each processing unit, respectively;

MS _s and MU (multi-user) _u The mass flow of the material flow s and the total inlet mass flow handled by the processing unit u are indicated, respectively.

In an embodiment, the expression corresponding to the carbon dioxide emission model is:

wherein, xi _i As energy consumption coefficient, eta _i For the production factor, a _fuel Is the fuel ratio of the fuel to be used,conversion coefficient of standard fuel of coal, fuel oil and natural gas ton>Conversion coefficient for electricity, steam or water, < >>Carbon dioxide emissions per ton of standard fuel.

In an embodiment, the energy consumption model is built by adopting an artificial neural network, and specifically includes the following steps:

collecting refinery-related process volume data and energy consumption historical data;

and training the artificial neural network by taking the relevant processing amount data of the refinery as an input sample and the corresponding energy consumption historical data as an output sample to obtain a data-driven-based refinery energy consumption model.

In one embodiment, the step S5 further includes solving the refinery production planning multi-objective model using NSGA-II algorithm;

the method for solving the multi-objective model of the refinery production plan by adopting the NSGA-II algorithm further comprises the following steps:

randomly generating an initial population of size N;

the next generation is obtained through the selection, crossing and mutation operations of a genetic algorithm after non-dominant sorting;

combining parent and offspring populations from the second generation, performing non-dominant ranking, and simultaneously performing crowding degree calculation on individuals of each non-dominant layer;

obtaining a new parent population according to the non-dominant relationship and the crowding degree of the individuals;

adopting NSGA-II algorithm to evaluate the result of the strict model through constraint condition and objective function, generating new sub-population, sorting according to non-inferior and crowded distance, adopting binary tournament selection to process constraint until the maximum genetic algebra is satisfied, and outputting Pareto solution;

otherwise, new populations will continue to be generated and rigorous simulations will be performed as inputs to the rigorous model.

The multi-objective optimization method for the refinery production plan provided by the application takes the economic benefit, the carbon dioxide emission and the energy consumption as objective functions, realizes multi-objective optimization of the refinery production plan, and has higher data processing efficiency and higher accuracy.

Drawings

The above and other features, properties and advantages of the present application will become more apparent from the following description of embodiments taken in conjunction with the accompanying drawings in which like reference characters designate like features throughout the drawings, and in which:

FIG. 1 discloses a multi-objective optimization method step diagram of a refinery production plan according to one embodiment of the present application;

FIG. 2 discloses a simplified refinery flow diagram according to one embodiment of the present application;

FIG. 3 discloses a graph of crude oil true boiling distillation according to one embodiment of the present application;

FIG. 4 discloses a schematic view of a catalytic cracking mechanism model according to an embodiment of the application;

FIG. 5 discloses an algorithm solving flowchart according to an embodiment of the application;

FIG. 6 discloses a schematic diagram of an algorithm solution according to an embodiment of the present application.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

FIG. 1 is a step diagram of a multi-objective optimization method for a refinery production plan according to one embodiment of the present application, as shown in FIG. 1, the multi-objective optimization method for a refinery production plan according to the present application comprises the steps of:

s2, generating a simulation scene;

According to the refinery production plan multi-objective optimization method based on the reconstruction of the crude oil and device input-output relation, the crude oil property is reconstructed by utilizing the crude oil distillation true boiling point curve under the premise of considering the uncertainty of the crude oil property, the device input-output relation is reconstructed by combining the crude oil distillation true boiling point curve and the corrected mechanism model, and meanwhile, three objective functions (economic benefit maximization, energy consumption minimization and carbon dioxide emission minimization) are used for comprehensive evaluation and calculation, a final refinery production plan multi-objective optimization model is established, and a multi-objective optimization algorithm is adopted to solve the refinery production plan multi-objective optimization problem, so that an optimal production scheme of a refinery production plan is obtained.

Fig. 2 shows a simplified refinery flow chart according to an embodiment of the present application, as shown in fig. 2, in the refinery process of this embodiment, the main raw materials of the refinery are mixed crude oil (MX 1) and heavy aromatics (BHR), methanol (BME), hydrogen (BH 2), reforming raw materials (BRF), MTBE (BMT), ethylbenzene (BEB).

The treatment unit comprises 28 sets of devices including two atmospheric and vacuum units (CDU 1 and CDU 2), one set of light hydrocarbon recovery unit (CUT), one set of reset pre-hydrogenation unit (TP 1), one set of benzene extraction combined unit (AEX), two sets of catalytic cracking units (FCC 1 and FCC 2), one set of slurry head (SVTP), one set of high-pressure hydrocracking unit (HCU), one set of delayed coking unit (CK 1), one set of solvent deasphalting unit (SAU), 8 sets of hydrogenation units (HF 1-HF 8), two sets of gas separation units (GF), one set of polypropylene unit (SPPU), one set of MTBE unit, one set of sulfur recovery unit (PSA), one set of ethylbenzene unit (SEBN), one set of styrene unit (SSTR) and one set of alkylation unit (SALK).

The storage tanks include gasoline, diesel, aviation fuel and naphtha storage tanks.

A total of 13 final products were produced, respectively:

92 # gasoline (W92), 95 # gasoline (W95), JET, diesel (W00), ethylene naphtha (EEN), benzene (BNZ), xylene (XYL), commodity liquefied gas (PLG), polypropylene (PPP), petroleum coke (CCK), sulfur (SULH), liquid ammonia (NH 3), styrene (STR).

Totally outsourcing 7 raw materials, namely:

tube blends (MX 1), heavy aromatics (BHR), methanol (BME), hydrogen (H2), reforming feedstock (BRF), MTBE (BMT), ethylbenzene (BEB).

These steps of the proposed multi-objective optimization method for refinery production planning based on crude oil and unit input-output relationship reconstruction of the present application are described in detail below using the simplified refinery shown in fig. 2 as an example. It is understood that within the scope of the present application, the above-described technical features of the present application and technical features specifically described below (e.g., in the examples) may be combined with each other and associated with each other, thereby constituting a preferred technical solution.

And S1, fitting a crude oil true boiling point distillation curve, reconstructing crude oil properties by adopting the fitted curve of the crude oil true boiling point distillation curve, and carrying out uncertainty characterization on the crude oil properties.

Fitting a crude oil true boiling point distillation curve by using a probability density function, reconstructing crude oil properties by using the crude oil true boiling point distillation curve, and simultaneously characterizing uncertainty of the crude oil properties by using the fitted curve.

It is assumed that the crude oil analysis data is known, but the properties are still uncertain before the crude oil reaches the factory.

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, which in this example is fitted using a beta function.

Under the condition that the analysis data of crude oil are known, the property of the crude oil is reconstructed by utilizing a beta function, and the expression corresponding to the beta function is as follows:

The optimal values of the beta function parameters (A, B, alpha, beta) are determined, and the normal distribution parameters of the parameters alpha and beta are determined.

Fig. 3 is a graph of crude oil true boiling point distillation according to a first embodiment of the present application, and taking crude oil CO1 shown in fig. 3 as an example, a fitting process based on a least square method is described:

(1) Determination of parameters A, B, α, β:

since the parameter A, B determines the upper and lower bounds of the distribution function, A, B is fixed at the optimal location to fit the crude true boiling point distillation curve (TBP) of the target crude.

A= -0.0313, b= 0.9782 results. α= 0.4554, β= 0.3668.

(2) Determination of x:

x represents a normalized temperature recovery value, and the upper limit and the lower limit of the normalized temperature recovery are respectively as follows

= -300°f, = 1800°f, fitting to obtain a formula for analyzing crude oil data.

(3) Uncertain characterization of crude oil properties:

the fitted curve is used for describing the crude oil property, namely uncertainty of the crude oil property is characterized by using uncertain positive parameters alpha and beta in a beta function, namely the parameters alpha and beta have normal distribution on the premise that the crude oil property is in normal distribution.

The alpha and beta determine the shape of TBP, namely the property of crude oil, and since the property of crude oil approximately follows normal distribution, the parameters alpha and beta also follow normal distribution, and the average value of the normal distribution is set as the optimal value of the uncertain parameters alpha and beta fitting the TBP curve of the crude oil.

In this example, σ=0.002 is a relaxed value that meets the assumption, and does not change the quality of crude oil. The uncertainty in the nature of the crude oil then translates into a problem of probability distribution of the positive parameters α, β in the beta function.

Description of the problem at the two ends of the curve: since there is still some error in the beta function to fit, especially at the two 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 to be constants.

And S2, generating a simulation scene.

In order to solve the problem of difficulty in solving a large number of scenes caused by Monte Carlo sampling, in this embodiment, the number of scenes is greatly reduced by using a method of moment estimation, and a Random Vector Sampling (RVS) is used to obtain the values of the discrete points used, so as to ensure that each scene can be obtained.

In this embodiment, the scene generation further includes moment estimation, random vector sampling, normalization processing:

and a moment estimation step, namely acquiring a point to be selected of the scene by adopting a moment estimation method.

If the Monte Carlo method is adopted to perform sample average approximation, a large number of scenes are required to be generated so as to ensure that points with low probability can be obtained, and the scale of the model is necessarily not small.

In the embodiment, the discretization of continuous variables is performed by adopting a moment estimation method, and N points can be used for accurately estimating (2N-1) center distances based on Gaussian orthogonal moment estimation, so that the mean value, the variance, the skewness and the kurtosis can be accurately estimated by only three points, and only three points to be selected are in a probability set, thereby avoiding a large number of scenes generated by a Monte Carlo method.

A random vector sampling step, namely constructing a scene through random vector sampling.

Since there are only three points in the probability set, which are all expected to be selected to construct the few scenarios, random vector sampling may suffice.

For these three points, the RVS builds three scenes at a time, with one of the three points being an implementation of the uncertainty parameter.

The full arrangement of these three points (v 1, v2, v 3) forms a scene tree, as shown in equation (1):

Ω(s ₁ ，s ₂ ，s ₃ )＝{(v ₁ ，v ₂ ，v ₃ )，(v ₁ ，v ₃ ，v ₂ )，(v ₂ ，v ₁ ，v ₃ )，(v ₂ ，v ₃ ，v ₁ )，(v ₃ ，v ₁ ，v ₂ )，(v ₃ ，v ₂ ，v ₁ )}(1)

normalization processing: after all scenes S are generated, the probability that the uncertainty parameter p is implemented at the correlation value for a scene S in the kth set of all scenes is recorded asCalculating probability of each scene by probability normalization methodAs shown in formula (2):

wherein N is _r For normalized sign, K is the total number of sets.

The probability of scene s in the kth set of all cases is noted asThe probabilities of three cases in each set k are calculated by a formula as shown in formula (3).

Then, three cases s=k, k+1, k+2 in the kth set are normalized as shown in formula (4):

finally, the normalized probability of scene s is shown as equation (5):

and S3, establishing a mechanism model of the process device, and reconstructing the input-output relation of the process device by combining a fitting curve of the crude oil true boiling point distillation curve.

Since the operation of a process unit depends on the quality of the feed stream, such as crude oil as the feed to a atmospheric and vacuum unit (CDU), and distillate fraction as the feed to a catalytic cracking unit (FCC), the uncertainty of the crude oil is handled by a model of the mechanism of the process unit. And simultaneously, the crude oil property can be reconstructed by utilizing the crude oil distillation true boiling point curve.

In this embodiment, the HYSYS software is used to build a corresponding process plant mechanism model, which includes, but is not limited to, a common pressure reduction unit (CDU) mechanism model, a catalytic cracking (FCC), a Hydrocracking (HCU) secondary processing unit mechanism model, and the like.

Uncertainty in crude oil properties manifests itself as variations in crude oil distillation curves, which in turn lead to variations in the CDU product yield of the atmospheric and vacuum unit.

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

Further, reconstructing the input-output relationship of a mechanism model of an atmospheric and vacuum unit (CDU) by combining a fitting curve of a crude oil true boiling point distillation curve;

and a mechanism model is established for secondary processing devices such as catalytic cracking (FCC), hydrocracking (HCU) and the like, the mechanism model is corrected, and the corrected mechanism model is combined with a fitting curve of a crude oil true boiling point distillation curve to reconstruct the input-output relationship, so that more accurate product yield is obtained.

Because accuracy of a mechanism model of a atmospheric and vacuum unit (CDU) is related to prediction accuracy of yield of an intermediate stream, and thus the whole downstream processing process is affected, in this embodiment, a fitting curve of the mechanism model and a crude oil true boiling point distillation curve under consideration of uncertainty of crude oil properties is adopted for reconstructing the input-output relationship of the atmospheric and vacuum unit (CDU) device.

Meanwhile, in order to cope with the influence of the change in yield of the atmospheric and vacuum unit (CDU) model, a corrected mechanism model is used for reconstructing the input-output relationship of refinery secondary processing units such as a catalytic cracking (FCC) unit and a Hydrocracking (HCU) unit.

The establishment and correction of the mechanism model of the catalytic cracking (FCC) unit will be explained in detail, and other process units are similar to the establishment and correction, and will not be described herein.

FIG. 4 discloses a schematic view of a catalytic cracking mechanism model according to an embodiment of the present application, and the catalytic cracking (FCC) unit mechanism shown in FIG. 4 is composed of a reaction-regeneration system, a separation system, and auxiliary equipment.

The mechanism model of a catalytic cracking (FCC) device is built by adopting HYSYS software, a 21 lumped kinetic model is built in a catalytic cracking module of the HYSYS software, data such as raw material properties, catalysts, equipment parameters, operation parameters and the like of the device are input, and initialization simulation is carried out by adopting default reaction kinetic parameters.

The reaction feed of the device is 70% of atmospheric residuum mixed with 30% of atmospheric wax oil and coking wax oil.

The high-temperature reaction oil gas from the settler enters a main fractionating tower T101, the top discharge is a mixture of crude gasoline and rich gas after separation, the side stripping tower T101-1 is used for extracting light diesel oil, and the bottom product is slurry oil;

the feeding of the absorption tower T201 is the rough gasoline, the compressed rich gas and the supplementary absorbent from the stabilizing tower T204, the discharging of the tower top is lean gas, the discharging of the tower bottom is rich absorption gasoline, and the rich absorption gasoline, the compressed rich gas and the desorption gas from the top of the desorption tower T202 are mixed and enter a flash tank;

the desorber T202 is used for desorbing condensed oil from the flash tank, and the discharged material is a deethanized gasoline destabilizer;

the feed of the reabsorption tower T203 is lean gas and part of light diesel oil from the main fractionating tower, the top product is dry gas, and the bottom product is rich absorption oil and returns to the main fractionating tower;

the feed of the stabilizing tower T204 is deethanized gasoline, the top product is liquefied gas, and the bottom product is stabilized gasoline.

The whole separation system has 4 cycles of absorbing the rich gasoline at the bottom of the absorption tower, re-absorbing the rich oil at the bottom of the absorption tower, desorbing the gas at the top of the desorption tower and stabilizing the partial stable gasoline of the stabilizer.

For a catalytic cracking (FCC) device mechanism model, correcting the mechanism model by using a neural network, and specifically comprises the following steps of:

and training the neural network model by adopting the training set and the testing set.

And outputting the predicted product yield error of the trained neural network model, and correcting the output product yield of the mechanism model.

The data preprocessing step further comprises error data elimination, normalization processing, training set determination and test set determination.

In this example, actual data of a plant of a certain year is taken monthly, and a total of 240 groups of 20 groups per month are subjected to data preprocessing in the following manner:

(1) And (5) error data rejection.

For various reasons, there may be some erroneous data in the dataset, which is not prevented from affecting the training quality, and is first deleted using a priori theoretical knowledge of the catalytic cracking reaction.

(2) And (5) normalization treatment.

The difference of the span of each data in the data set affects the convergence speed and accuracy of the model, so that normalization operation is required to be performed on the data in the data set, and a specific formula is shown in formula (6):

wherein X is data before normalization, X _norm For normalized data, X _max ,X _min Respectively, maximum and minimum values in the data.

(3) A training set and a test set are determined.

The processed data sets are randomly disordered, and then the training set and the testing set are divided in a ratio of 3:1.

In this example, 16 input variables such as vanadium content, carbon residue content, saturated hydrocarbon content, etc. were selected, and the output variable was selected as the main product yield.

Inputting actual working condition data for model training, taking 180 groups of 240 groups of data as training samples, taking the rest 60 groups of data as verification samples, respectively selecting 5 to 13 hidden layer neurons for model training, gradually reducing the relative error of the model when the number of the hidden layer neurons is 5 to 7, gradually increasing the relative error of the model when the number of the hidden layer neurons is 7 to 13, and obviously, the model precision is highest when the number of the hidden layer neurons is 7.

The neural network is model trained using the training set and the test set, and the final neural network parameters are shown in table 1.

TABLE 1 parameters of artificial neural networks

Parameters (parameters)	Numerical value	Parameters (parameters)	Numerical value
				Training batch	64	Number of input layer neurons	16
Training times	1000	Hidden layer neuron number	7
				Setting error	0.001	Output layer neuron number	5

The neural network model output prediction product yield error after training is used for correcting the output product yield of the mechanism model, and the method further comprises the following steps:

and respectively inputting the actual working condition data into the mechanism model and the neural network model to respectively obtain product yield distribution and predicted product yield errors, and superposing the product yield distribution output by the mechanism model and the errors predicted by the neural network model to output as a corrected model.

Table 2 shows that the relative error of the mechanism model and the correction model corrected by the neural network is reduced from 4.74 to 3.41, and the average relative error is reduced to 28%, which proves that the correction model has certain effectiveness.

Table 2 model product yield prediction comparison

and (3) establishing a multi-objective model of the refinery production plan by taking the maximum economic benefit, the minimum carbon dioxide emission and the minimum energy consumption as objective functions and taking the supply-demand relationship, the processing capacity and the storage tank condition as constraints.

Profit is equal to the total revenue of the final product minus the total raw material procurement cost and the total operating cost of the processing unit. Throughout the refinery, U denotes the unit, B denotes the storage tank, and S denotes the stream, including feedstock, intermediate streams and products. The inlet flows of the process units and tanks are represented by sets UI = { (u, s) } and BI = { (b, s) }, as well as the outlet flows by sets UO and BO, and the key property attributes of the final key product by set E.

In this embodiment, a multi-objective model of the refinery production plan is built, wherein the economic benefit model is shown in the following formula (7).

max∑ _s∈FPS P _s ·MS _s -∑ _s∈RMS C _s ·MS _s -∑ _u∈U C _u ·MU _u (7)；

Wherein P is _s 、C _s And C _u Representing sales price of each final product, purchasing cost of raw materials, and operation cost associated with each processing unit, respectively; FPS represents product stream, RMS represents feed stream.

In this embodiment, the established multi-objective model of the refinery production plan, wherein the carbon dioxide emission model is established by combining two analysis methods of material flow and energy flow.

The material flow analysis method specifically comprises the following steps:

because the production process of fuel refining is complex, a method combining standing observation method and following observation method is adopted for research. Under normal operating conditions, the analysis boundaries include production flows and operating protocols. Carbon was selected as the observation object.

The gasoline and diesel production devices are selected as observation areas, and the material flow changes of the areas are analyzed. The consumption of material in the production process depends not only on the total amount of material flowing in, but also on the demand of the upstream product for the downstream production units.

Thus, the material consumption takes into account both the raw material consumption of the production unit and the influence of the downstream production units.

In this example, the refinery process unit has 6 material flows and the material balance of the process unit is shown in formula (8).

F _i +A _i ＝P _ie +P _i +W _i +R _i (8)

Wherein F is _i For the inlet of the device other material flows, A _i For the inlet feed flow to the apparatus, P _ie For product flow into downstream apparatus, P _i To set the product flow, W _i For outlet waste flow, R _i The material flow is circulated for the device.

The production coefficient of the production unit i (ratio of the product yield to the raw material consumption of the production unit i) is represented by formula (9).

Wherein eta _i Is the production coefficient of the production unit i.

The energy flow analysis method specifically comprises the following steps:

the chemical reactions of the production units are also complicated due to the complexity of the raw materials and products. In conventional reaction engineering methods, it is difficult to establish a suitable production plant energy conservation equation. However, according to the methodology of the system theory, the production unit is considered as a complete reactor system, only so that an energy balance can theoretically be established.

In this embodiment, the energy flows are classified into 7 types according to the source, destination, and effect of the energy.

The energy balance of the production unit i is represented by formula (10), and is expressed in terms of ton of standard fuel oil as an energy flow unit.

B _i +E _i +E _ri ＝E _ni +E _i+1 +L _i +S _i (10)

Wherein B is _i E for energy flowing in from other devices _i Energy, E, of the product entering the device i for the device i-1 _ri E for recovering energy _ni For total energy flowing into the device E _i+1 For the energy of device i into device i+1, L _i To lose energy, S _i Is the energy flowing into other devices.

The energy consumption of a production unit during refining is generally referred to as the difference between the energy in-flow and the energy out-flow, such as electricity, steam and water.

Energy consumption coefficient ζ of production unit _i Represented by formula (11).

ξ _i ＝B _i -S _i ＝E _ni +E _i+1 +L _i -E _i -E _ri (11)

Wherein, xi _i For the energy consumption coefficient of the production unit, B _i Is the slaveEnergy flowing in by other devices, E _i Energy, E, of the product entering the device i for the device i-1 _ri E for recovering energy _ni For total energy flowing into the device E _i+1 For the energy of device i into device i+1, L _i To lose energy, S _i Is the energy flowing into other devices.

The carbon dioxide emissions from refineries are affected by factors such as energy structure and consumption, heat output and carbon content of the fuel, carbon dioxide emissions per unit fuel, pollution control levels, etc.

In the carbon dioxide emissions model, the purchased energy takes into account fuel and electricity, assuming that the fuel is converted to carbon dioxide by combustion and eventually released into the environment.

Total carbon dioxide emissionsRepresented by formula (12):

direct emission of carbon dioxide V _Direct Represented by formula (13):

wherein a is _fuel Is the fuel ratio of the fuel to be used,conversion coefficient for electricity, steam or water, < >>ζ is the carbon dioxide emissions per ton of standard fuel _i As energy consumption coefficient, eta _i Is the production coefficient.

Indirect carbon dioxide emission V _Indirect Represented by formula (14):

wherein a is _fuel Is the fuel ratio of the fuel to be used,conversion coefficient for electricity, steam or water, v _CO2 ζ is the carbon dioxide emissions per ton of standard fuel _i As energy consumption coefficient, eta _i Is the production coefficient.

The total carbon dioxide emission of the production unit i is represented by the formula (15), that is, the expression of the carbon dioxide emission model is as follows:

wherein, xi _i As energy consumption coefficient, eta _i For the production factor, a _fuel Is the fuel ratio of the fuel to be used,conversion coefficient of standard fuel of coal, fuel oil and natural gas ton>Conversion coefficient for electricity, steam or water, v _CO2 Carbon dioxide emissions (t) per ton of standard fuel.

In this embodiment, the energy consumption model of the established multi-objective model of the refinery production plan is established by using an artificial neural network, and specifically includes the following steps:

collecting refinery-related process data and energy consumption history data;

and training the artificial neural network by taking refinery processing data as an input sample and corresponding energy consumption data as an output sample by using an artificial neural network technology to obtain a data-driven refinery energy consumption model.

In this embodiment, a BP neural network is selected to fit the relevant data, and the BP neural network mainly comprises an input layer, a hidden layer and an output layer.

The learning process of the BP neural network consists of forward propagation of signals and backward propagation of errors, and when the model obtained by training reaches the minimum root mean square error (RMS), the network training is completed.

Specifically, the RMS value is:

where i=1, 2, …, m, m is the number of training times, j=,for the estimated value of model output, O _ij The actual value output for the process model.

In this embodiment, a multi-objective model of the refinery production plan is built, wherein constraints include supply and demand relationships, process capacity, and storage tank conditions.

The expressions corresponding to the constraint conditions of the supply-demand relationship are formula (17) and formula (18):

equation (17) describes limiting the supply of each raw material during planning;

wherein, the liquid crystal display device comprises a liquid crystal display device,and->Respectively representing the minimum value and the maximum value of the outsourced raw materials of the refinery;

equation (18) describes the market demand for various products.

Wherein, the liquid crystal display device comprises a liquid crystal display device,and->Representing minimum and maximum market demand for the product, respectively.

The expression corresponding to the constraint condition of the processing ability is formula (19) -formula (23).

Equation (19) describes the total mass flow of the inlet into each processing unit;

equation (20) describes the restriction of the feed flow rate processed by the processing unit;

wherein, the liquid crystal display device comprises a liquid crystal display device,and->Representing the maximum and minimum throughput of the device, respectively.

Formulas (21) and (22) calculate the output mass flow rate of each product stream using variable throughput methods, and there are multiple modes of operation of the processing units, such as FCC, HCU, etc., represented by the aggregated MMU, corresponding to different product yields.

Binary variables are introduced to represent the selection of different modes of operation for each cell, aggregate MO represents the modes of operation, and equation (23) forces operation of only one production mode during the planning period.

Wherein x is _u,m Indicating the selection of different modes of operation for each device.

The expression corresponding to the constraint of the tank condition is equation (24) -equation (28).

Equation (24) calculates the flow rates of the various end products:

wherein MS is _s′ Representing the inlet flow of each tank, MS _s Representing the flow rate of the various end products;

equation (25) represents the relationship between mass flow and volumetric flow,

wherein SPG is an abbreviation for specific gravity;

the mass characteristics assume a linear mixture on a mass or volume basis, as shown in equations (26) - (28),

wherein, the liquid crystal display device comprises a liquid crystal display device,and->Representing the worst and best properties of the product s, respectively.

In the embodiment, an NSGA-II algorithm is adopted to solve a multi-objective optimization model of a refinery production plan, and an optimal production scheme is obtained.

The NSGA-II algorithm selects an initial variable to be assigned to HYSYS software for calculation, and continuously corrects the optimized searching direction according to the simulation calculation result and the optimized solution until convergence conditions are met to obtain a pareto optimal solution, so that an optimal production scheme is obtained.

The NSGA-II algorithm is one of the most popular multi-objective genetic algorithms, reduces the complexity of non-inferior sorting genetic algorithms, has the advantages of high running speed and good convergence of solution sets, and becomes a benchmark for the performance of other multi-objective optimization algorithms.

In this embodiment, MATLAB sends the variable of each individual in the population to HYSYS one by one for running simulation calculations, and the individual will be discarded with the help of the HYSYS running state parameters if there is no convergence or error for HYSYS.

In this embodiment, the known conditions of the production plan optimization model include the following conditions:

a production planning period; mixing crude oil and the available amount of intermediate feedstock; maximum and minimum throughput of the processor and its mode of operation; requirements and performance specifications of the final product; the sales price of each final product, the procurement cost of each raw material, the operating cost of each processing unit.

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

purchasing the quantity of raw materials; total feed of the processing device and proper operation mode; flow rate of each intermediate; the quantity and nature of the product.

The population scale, genetic algebra, crossover probability and mutation probability in the solving process are parameters to be set, and the parameters are set as follows:

table 3 solver parameter settings

Parameters (parameters)	Population size	Algebra of genetics	Crossover probability	Probability of variation
					Numerical value	125	10000	0.8	0.1

FIG. 5 discloses an algorithm solving flow chart according to an embodiment of the application, as shown in FIG. 5, for solving a multi-objective model of a refinery production plan using NSGA-II algorithm, further comprising the steps of:

randomly generating an initial population of size N;

the next generation is obtained through 3 basic operations of selection, crossing and mutation of a genetic algorithm after non-dominant sorting;

the NSGA-II algorithm evaluates the result of the HYSYS strict model through constraint conditions and objective functions, generates new sub-populations, sorts according to non-inferior and crowded distances, adopts binary tournament selection to process constraint until the maximum genetic algebra is met, and outputs Pareto solutions;

otherwise, new populations will continue to be generated and rigorous simulations will be performed as input to the HYSYS rigorous model.

FIG. 6 is a schematic diagram of an algorithm solution according to an embodiment of the present application, where genetic algorithm parameter setting is performed under the premise of ensuring convergence, and the process aims at minimizing genetic algebra, avoiding local optimization and diversity of Pareto solutions, and solving the algorithm optimization result.

The specific optimized production schemes are shown in table 4.

Table 4 decision variable optimization values

Compared with the prior art, the application provides the multi-objective optimization method for the refinery production plan based on the reconstruction of the input-output relationship between the crude oil and the process device, which considers the uncertainty of the crude oil property, reconstructs the input-output relationship of the device by using the corrected mechanism model and the crude oil true boiling point distillation curve, realizes the multi-objective optimization of the refinery production plan, and ensures that the prediction of the product yield has higher accuracy.

While, for purposes of simplicity of explanation, the methodologies are shown and described as a series of acts, it is to be understood and appreciated that the methodologies are not limited by the order of acts, as some acts may, in accordance with one or more embodiments, occur in different orders and/or concurrently with other acts from that shown and described herein or not shown and described herein, as would be understood and appreciated by those skilled in the art.

As used in the specification and in the claims, the terms "a," "an," "the," and/or "the" are not specific to a singular, but may include a plurality, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that the steps and elements are explicitly identified, and they do not constitute an exclusive list, as other steps or elements may be included in a method or apparatus.

The embodiments described above are intended to provide those skilled in the art with a full range of modifications and variations to the embodiments described above without departing from the inventive concept thereof, and therefore the scope of the application is not limited by the embodiments described above, but is to be accorded the broadest scope consistent with the innovative features recited in the claims.

Claims

1. A multi-objective optimization method for a refinery production plan, comprising the steps of:

s2, generating a simulation scene;

2. The method for multi-objective optimization of refinery production planning of claim 1, wherein said fitting of crude oil true boiling distillation curve in step S1 further comprises:

3. The method for multi-objective optimization of a refinery' S production schedule according to claim 1, wherein said step S2 further comprises the steps of:

constructing a scene through random vector sampling;

4. The method for multi-objective optimization of a refinery' S production schedule according to claim 1, wherein said step S3 further comprises the steps of:

5. The method of multi-objective optimization of a refinery's production plan of claim 4, wherein said modeling and correcting the secondary process plant mechanism further comprises correcting the mechanism model using a neural network;

training a neural network model by adopting a training set and a testing set;

6. The multi-objective optimization method of a refinery production planning of claim 1, wherein said objective functions are economic maximization, carbon dioxide emission minimization and energy consumption minimization;

7. The multi-objective optimization method of a refinery's production schedule according to claim 6, wherein the corresponding expression of the economic benefit model is:

max∑ _s∈FPS P _s ·MS _s -∑ _s∈RMS C _s ·MS _s -∑ _u∈U C _u ·MU _u ；

8. The multi-objective optimization method of a refinery's production schedule of claim 6, wherein the expression corresponding to the carbon dioxide emission model is:

wherein, xi _i As energy consumption coefficient, eta _i For the production factor, a _fuel Is the fuel ratio of the fuel to be used,conversion coefficient of standard fuel of coal, fuel oil and natural gas ton>Conversion coefficient for electricity, steam or water, v _CO2 Carbon dioxide emissions per ton of standard fuel.

9. The multi-objective optimization method of refinery production planning according to claim 6, wherein said energy consumption model is built by artificial neural network, comprising the following steps:

10. The method of multi-objective optimization of a refinery production planning according to claim 1, wherein said step S5 further comprises solving a multi-objective model of a refinery production planning using NSGA-II algorithm;

randomly generating an initial population of size N;