Method for determining control variables in oil refining production

Abstract

The present application is applicable to the technical field of industrial variable optimization, and provides a method for determining control variables in oil refining production, wherein the key control variables are determined from the control variables by determining the target variables and the control variables; the steady-state data of the key control variables are extracted based on the multiple variation coefficient method; the initial parent population corresponding to the key control variables is generated; the individual mutations in the initial parent population are allowed to generate the intermediate generation mutation population; the dimension values ​​of the individuals in the intermediate generation mutation population that exceed the preset boundary are reset; the parent population of the intermediate generation mutation population and the individuals in the intermediate generation mutation population are crossed to obtain the first and second mutation populations; the intermediate generation parent population is generated according to the parent population of the intermediate generation mutation population, the first and second mutation populations; if the population evolution meets the evolution termination condition, the optimal individual in the population is used as the value of the key control variable. The present application can optimize the control variable values ​​in oil refining production.

Description

Technical Field

[0001] The present application belongs to the technical field of industrial variable optimization, and in particular, relates to a method for determining control variables in oil refining production. Background Art

[0002] Today's oil resources are becoming heavier and inferior, but the country's environmental protection requirements are becoming increasingly stringent, and the strategic goals of "carbon peak" and "carbon neutrality" have been proposed. As a pillar industry of the national economy, the "green" and "efficient" production of petrochemicals is of great significance to promoting energy conservation and emission reduction and serving the "dual carbon" goals. Therefore, petrochemical companies need to continuously improve production technology and improve the quality standards of petroleum products. Hydrocracking is an important process in oil refining production. It is one of the most effective process technologies for lightening and cleaning crude oil, and has many advantages in improving product quality and product solution flexibility. Hydrocracking is essentially the processing of heavy and inferior crude oil into high-quality diesel, aviation kerosene, naphtha and other terminal products required by the market. In actual production, hydrocracking process technology is usually affected by a variety of nonlinear and strongly correlated factors such as temperature, pressure, and flow in the reactor. These related factors constitute the key control variables of the hydrocracking production process and will affect the operation of the entire production process. When external force majeure factors such as crude oil supply, product production demand, and production environment change, it is necessary to adjust key control variables to optimize production operations in a timely manner to ensure that production efficiency and product quality are not drastically affected. Therefore, in order to ensure that the hydrocracking production process improves product yield and product quality while meeting safety and stability, and maximizes the benefits of oil refining enterprises, it is particularly important to adjust key control variables according to actual production needs and optimize the operation of the hydrocracking unit throughout the production process.

[0003] For global optimization operation, the accuracy of solving key control variables is the basis. Evolutionary optimization algorithms (evolutionary algorithms for short) have significant advantages in solving such optimization problems with complex nonlinear relationships and have been widely used and recognized. Evolutionary algorithms draw on the evolutionary laws of biological organisms in nature, and generally include basic parameter settings, population initialization, population individual variation and crossover updates, evaluation of population individuals and retaining excellent individuals to start the next round of evolution. The variation and crossover of population individuals is the most critical step in this type of algorithm, but traditional evolutionary algorithms are more likely to exceed the set boundaries when performing population individual variation and crossover, and after variation and crossover, a large variation population is usually generated (too much variation information and too little genetic information), which will limit the algorithm's search space and optimization ability, resulting in the control variable values ​​of the hydrocracking process being not good enough, which has an extremely adverse impact on the efficiency and stability of the production process, and cannot achieve its global optimization operation. Summary of the invention

[0004] The embodiment of the present application provides a method for determining control variables in oil refining production, which can solve the problem that the optimal value of the control variable cannot be obtained due to the small search space and insufficient optimization ability of the evolutionary algorithm in the oil refining production process.

[0005] In an embodiment of the present application, a method for determining a control variable in oil refining production is provided, comprising:

[0006] Step 1, determining the target variables and control variables to be optimized in oil refining production, and determining the key control variables from the control variables;

[0007] Step 2, extracting steady-state data of key control variables based on the multiple coefficient of variation method;

[0008] Step 3, obtaining an initial parent population corresponding to the key control variables based on an evolutionary algorithm; wherein each individual in the initial parent population includes a set of values ​​of all the determined key control variables, a set of values ​​is a value in the steady-state data, and multiple individuals in the initial parent population are different from each other;

[0009] Step 4, performing mutation operations on individuals in the initial parent population to generate an intermediate generation mutation population;

[0010] Step 5: for each individual in the intermediate generation mutation population, if the value of the individual in the jth dimension exceeds the preset boundary, the value of the individual in the jth dimension is reset to the average value of the values ​​of any two different individuals in the intermediate generation mutation population in the jth dimension;

[0011] Step 6, performing a crossover operation on the individuals in the intermediate generation mutation population and the individuals in the parent generation population before the intermediate generation mutation population mutates, to obtain a first mutation population and a second mutation population;

[0012] Step 7, generating an intermediate generation parent population according to the parent population before the intermediate generation mutation population mutation, the first mutation population, and the second mutation population;

[0013] Step 8, if the population evolution meets the preset evolution termination condition, the optimal individual in the intermediate generation parent population is used as the value of the key control variable; otherwise, the intermediate generation parent population is mutated to generate a new intermediate generation mutant population, and the new intermediate generation mutant population is used as the intermediate generation mutant population in step 5 and returned to execute step 5.

[0014] The above solution of the present application has the following beneficial effects:

[0015] In some embodiments of the present application, by determining whether the values ​​of each dimension of each individual in the intermediate generation mutant population exceed the preset boundaries, and resetting the dimension values ​​of the population individuals that exceed the preset boundaries, it is possible to avoid exceeding the preset boundaries when the population individuals mutate, thereby improving the optimization ability of the evolutionary algorithm; by controlling the correlation between the variable and the target variable, the key control variables of the target variable are screened, the number of control variables is reduced, and the optimization ability of the evolutionary algorithm is improved; in addition, after cross processing of the individuals in the parent population, not only the first mutant population but also the second mutant population can be obtained, which can expand the search space of the evolutionary algorithm and greatly improve the optimization ability of the evolutionary algorithm, thereby obtaining the optimal value of the control variable.

[0016] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flow chart of a method for determining control variables in oil refining production provided in one embodiment of the present application;

[0019] Figure 2 It is a comparison diagram of optimization effects of using the method for determining control variables in oil refining production provided by the present application and not using the method for determining control variables in oil refining production provided by the present application in the embodiments of the present application;

[0020] Figure 3a In the embodiment of the present application, the first convergence curve is independently run using the particle swarm optimization algorithm and the double mutation particle swarm optimization algorithm;

[0021] Figure 3b In the embodiment of the present application, the particle swarm optimization algorithm and the double mutation particle swarm optimization algorithm are used to independently run the convergence curve for the second time;

[0022] Figure 3c In the embodiment of the present application, the particle swarm optimization algorithm and the double mutation particle swarm optimization algorithm are used to independently run the convergence curve for the third time;

[0023] Figure 3d In the embodiment of the present application, the particle swarm optimization algorithm and the double mutation particle swarm optimization algorithm are used to independently run the fourth convergence curve;

[0024] Figure 4a In the embodiment of the present application, the differential evolution algorithm and the double mutation differential evolution algorithm are used to independently run the first convergence curve;

[0025] Figure 4b In the embodiment of the present application, the differential evolution algorithm and the double mutation differential evolution algorithm are used to independently run the convergence curve for the second time;

[0026] Figure 4c In the embodiment of the present application, the convergence curve of the third independent operation of the differential evolution algorithm and the double mutation differential evolution algorithm is shown;

[0027] Figure 4d In the embodiment of the present application, the differential evolution algorithm and the double mutation differential evolution algorithm are used to independently run the fourth convergence curve. DETAILED DESCRIPTION

[0028] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0029] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0030] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0031] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0032] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0033] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0034] Aiming at the problem that the optimal value of the control variable cannot be obtained due to the small search space and insufficient optimization ability of the evolutionary algorithm. The present application provides a method for determining the control variable in oil refining production. By judging whether the value of each dimension of each individual in the intermediate generation mutation population exceeds the preset boundary, and resetting the dimension value of the population individual that exceeds the preset boundary, it can avoid exceeding the preset boundary when the population individual mutates, and improve the optimization ability of the evolutionary algorithm; through the correlation between the control variable and the target variable, the key control variables of the target variable are screened, the number of control variables is reduced, and the optimization ability of the evolutionary algorithm is improved; in addition, after the individuals in the parent population are cross-processed, not only the first mutation population but also the second mutation population can be obtained, which can expand the search space of the evolutionary algorithm and greatly improve the optimization ability of the evolutionary algorithm, thereby obtaining the optimal value of the control variable.

[0035] like Figure 1 The method for determining the control variables in oil refining production provided by the present application comprises the following steps:

[0036] Step 1: determine the target variables and control variables to be optimized in oil refining production, and determine the key control variables from the control variables.

[0037] The above target variables are determined according to actual production needs. In the oil refining process, the target variable may be distillation temperature, etc.

[0038] There are a large number of control variables involved in the oil refining production process. Too many control variables will make the calculation cumbersome and the optimization process complicated. Screening the key control variables of the target variables from the control variables can reduce the number of control variables, screen out the control variables with a high degree of correlation with the target variables, and improve the response speed of the model.

[0039] It is worth mentioning that the technical solution provided in this application adopts the weighted correlation coefficient method (combining the linear correlation degree and nonlinear correlation degree between the control variable and the target variable), which is more conducive to the screening of key control variables.

[0040] Step 2: Extract steady-state data of key control variables based on the multiple coefficient of variation method.

[0041] The steady-state data of key control variables are extracted here to facilitate the construction of optimization models. Since the coefficient of variation has the advantage of objectively reflecting the absolute value of the degree of data dispersion without considering the dimension, this application uses the coefficient of variation method to extract the steady-state data of key control variables. It should be noted that the smaller the coefficient of variation, the lower the degree of data dispersion and the more stable the data.

[0042] It is worth mentioning that in the embodiments of the present application, a multiple coefficient of variation method (first coefficient of variation and second coefficient of variation) is used to extract steady-state data of key control variables, which greatly improves the stability of key control variables and enhances the optimization ability of the evolutionary algorithm.

[0043] Step 3: Obtain the initial parent population corresponding to the key control variables based on the evolutionary algorithm.

[0044] The purpose of generating the initial parent population corresponding to the key control variables is to use the optimization ability of the evolutionary algorithm to find the optimal value of the key control variables. Each individual in the initial parent population includes a set of values ​​of all the key control variables determined, a set of values ​​is the value in the steady-state data, and multiple individuals in the initial parent population are different from each other.

[0045] For example, let's assume that the key control variables include temperature x1 and pressure x2, and the steady-state data for temperature is x1={23℃, 36℃, 52℃, 54℃, 53℃, 56℃}, and the steady-state data for pressure is x2={100Pa, 101Pa, 110Pa, 111Pa, 121Pa, 130Pa}, then the individuals can be {(23℃, 100Pa)}, {(36℃, 101Pa)}, {(52℃, 110Pa)}, {(54℃, 111Pa)}, {(53℃, 121Pa)}, {(56℃, 130Pa)}.

[0046] Step 4: Perform mutation operations on the individuals in the initial parent population to generate an intermediate generation mutant population.

[0047] The mutation operation of population individuals is a key step in the evolutionary algorithm, which can increase the probability of excellent individuals in the population.

[0048] Step 5: For each individual in the intermediate generation mutation population, if the value of the individual in the jth dimension exceeds the preset boundary, the value of the individual in the jth dimension is reset to the mean value of any two different individuals in the intermediate generation mutation population in the jth dimension.

[0049] Traditional evolutionary algorithms are prone to the problem of population individuals exceeding the set boundaries after population individuals mutate. Once this happens, it will greatly affect the optimization ability of the evolutionary algorithm.

[0050] It is worth mentioning that in the case where the latitude value of an individual exceeds the set boundary, the latitude value of the individual that exceeds the set boundary is reset with the mean value of the values ​​of two different individuals in the population on this dimension, thereby effectively avoiding the negative impact of the latitude value exceeding the set boundary on the evolutionary algorithm and improving the optimization ability of the evolutionary algorithm.

[0051] Step 6, performing a crossover operation on the individuals in the intermediate generation mutation population and the individuals in the parent generation population before the intermediate generation mutation population mutates, to obtain a first mutation population and a second mutation population.

[0052] Traditional evolutionary algorithms usually only form a mutant population after crossover operations on population individuals, which will cause the population to have too much mutation information and too little genetic information.

[0053] It is worth mentioning that this application generates two mutation populations (a first mutation population and a second mutation population) through different mutation strategies, which expands the search space of the evolutionary algorithm and improves the optimization ability of the evolutionary algorithm.

[0054] Step 7, generating an intermediate generation parent population according to the parent population before the intermediate generation mutation population mutation, the first mutation population and the second mutation population.

[0055] Step 8, if the population evolution meets the preset evolution termination condition, the optimal individual in the intermediate generation parent population is used as the value of the key control variable; otherwise, the intermediate generation parent population is mutated to generate a new intermediate generation mutant population, and the new intermediate generation mutant population is used as the intermediate generation mutant population in step 5 and returned to execute step 5.

[0056] It should be noted that when returning to execute step 5, the new intermediate generation mutant population is used as the intermediate generation mutant population in the original step 5. Specifically, for each individual in the new intermediate generation mutant population, if the value of the individual in the jth dimension exceeds the preset boundary, the value of the individual in the jth dimension is reset to the average value of the values ​​of any two different individuals in the new intermediate generation mutant population in the jth dimension.

[0057] Here, the iteration of population individuals is completed through population evolution until the optimal population individual is found.

[0058] It is worth mentioning that the method for determining control variables in oil refining production provided in the present application can avoid exceeding the preset boundaries when the individuals in the population mutate, by judging whether the values ​​of each dimension of each individual in the intermediate generation mutation population exceed the preset boundaries, and resetting the dimension values ​​of the individuals in the population that exceed the preset boundaries, thereby improving the optimization ability of the evolutionary algorithm; through the degree of correlation between the control variables and the target variables, the key control variables of the target variables are screened, the number of control variables is reduced, and the optimization ability of the evolutionary algorithm is improved; in addition, after cross processing of the individuals in the parent population, not only the first mutation population but also the second mutation population can be obtained, which can expand the search space of the evolutionary algorithm and greatly improve the optimization ability of the evolutionary algorithm, thereby obtaining the optimal value of the control variable.

[0059] In some embodiments of the present application, the above-mentioned evolution termination condition includes the population evolution generation. Specifically, the generation of population evolution can be indirectly represented by counting the number of times the intermediate generation parent population is generated. When the population evolution generation reaches the preset maximum population evolution generation, it is assumed that the population reaches the evolution termination condition at this time, and the value of the control variable in the optimal individual of the population is used as the optimal value of the control variable. Otherwise, the population continues to evolve until the maximum evolution generation is reached or the predicted value of the target variable corresponding to the individual in the population is closest to the preset threshold of the target variable. If the predicted value of the target variable corresponding to the optimal individual generated by the e-th population evolution is farther from the preset threshold of the target variable than its parent generation, the value of the control variable in the optimal individual of its parent generation is used as the optimal value of the control variable.

[0060] The specific process of determining the key control variables from the control variables in step 1 (determining the target variables and control variables to be optimized in the oil refining production, and determining the key control variables from the control variables) is exemplified below.

[0061] Step 1.1, by calculating the formula

[0062]

[0063] Get the linear correlation degree P between each control variable and the target variable i , where P i represents the i-th control variable x i Pearson Correlation Coefficient for the target variable y, represents the i-th control variable x i The standard deviation of (each control variable has several groups of data, and the standard deviation of several groups of data for each control variable is calculated here), Represents the standard deviation of the target variable y (each set of data for each control variable corresponds to a value of the target variable, and several sets of data correspond to several values ​​of the target variable. What is calculated here is the standard deviation of several values ​​of the target variable), 1≤i≤m, m represents the number of control variables related to the target variable.

[0064] Step 1.2, by calculating the formula

[0065]

[0066] Get the nonlinear correlation degree K between each control variable and the target variable i , where K i represents the i-th control variable x i Kendall Correlation Coefficient for the target variable y, C i It means (x i ,y) has the same number of pairs of elements, (x i ,y) represents the control variable x i The jth element x i,j and the jth element y of the target variable y j The set of elements composed of i It means (x i ,y), S represents the number of inconsistent elements in i The number of the first set of identical elements in P s represents the number of elements in the sth first set, and T represents the target variable y t The number of the second set of identical elements in Q t Indicates the number of elements in the t-th second set, the control variable x i The number of elements of the target variable y is n, t=1,2,...,T, s=1,2,...,S.

[0067] It should be noted that the above consistency means: for variable x i and y, the number of their elements is n, and the j-th element of the two variables is x i,j and j Indicates that they form an element set (x i ,y). (x i Any two elements (x,y) i,j ,y j ) and (x i,k ,y k ), if x i,j >x i,k And y j >yk , or x i,j <x i,k And y j <y k , then the two elements are consistent. If x i,j >x i,k And y j <y k , or x i,j <x i,k And y j >y k , then the two elements are inconsistent. If x i,j =x i,k or j =y k In this case, the two elements are neither consistent nor inconsistent.

[0068] Step 1.3, by calculating the formula W i =α1P i +α2K i Get the weighted correlation coefficient W between each control variable and the target variable i , W i represents the i-th control variable x i The weighted correlation coefficient of the target variable y, where α1 represents P i The weight parameter, α2 represents K i The weight parameter of .

[0069] Step 1.4: Arrange the control variables in the steady-state data in descending order of weighted correlation coefficients, and take the first m c control variables as the key control variable x; among them, m c =mρ, ρ represents the proportionality coefficient, Each is a control variable in the key control variable group.

[0070] The specific process of step 2 (extracting steady-state data of key control variables based on the multiple coefficient of variation method) is exemplified below.

[0071] Step 2.1, obtain historical data of key control variables, obtain a historical data set X, and divide the historical data set into N data blocks; each data block has L groups of data, and each group of data includes a set of values ​​of all key control variables.

[0072] For example, the length of the sliding window is first set to L, and then the historical data of the key control variables are obtained to obtain the historical data set X, and then the historical data set X is divided into N data blocks with the length of the sliding window as the unit length, and finally N data blocks are obtained, which can be recorded as X1, X2, ..., X N, each data block includes L groups of data for all key control variables.

[0073] Specifically, assume that there are m key control variables to be acquired. After the above steps, each data block X n All key control variables (x1, x2, ..., x m ), n=1,2,...,N. And each control variable has L sets of data.

[0074] Exemplarily, there are two key control variables obtained, including temperature x1 and pressure x2. The historical data set corresponding to the key control variables is X={x1{23℃,36℃,52℃,54℃,53℃,56℃}, x2{100Pa,101Pa,110Pa,111Pa,121Pa,130Pa}}. After dividing the historical data set, three data blocks (X1, X2, X3) are obtained. Each data block has two groups of data. At this time, L=2. Correspondingly, X1={(56℃,100Pa),(53℃,101Pa)}, x2={(36℃,130Pa),(23℃,121Pa)}, X3={(54℃,110Pa),(52℃,111Pa)}.

[0075] Step 2.2, for each data block, calculate the first coefficient of variation of each key control variable in the data block.

[0076] Specifically, the calculation formula of the coefficient of variation is as follows:

[0077]

[0078] Among them, s i represents the standard deviation of the ith variable, μ i represents the mean of the i-th variable, σ i represents the variance of the ith variable.

[0079] Step 2.3, calculate the second coefficient of variation of the first coefficient of variation of each key control variable.

[0080] Specifically, the first coefficient of variation refers to the calculation of each key control variable x in each data block. i The first coefficient of variation of all L groups of data. The second coefficient of variation refers to the calculation of the key control variable x in all N data blocks. i The second coefficient of variation corresponding to the first coefficient of variation of all L groups of data.

[0081] Step 2.4, for each data block, if the first coefficient of variation of a key control variable in the data block is less than the preset threshold of the first coefficient of variation, and the second coefficient of variation corresponding to the key control variable is less than the preset threshold of the second coefficient of variation, then the data of the key control variable in the data block is used as the steady-state data of the key control variable; wherein the preset threshold of the first coefficient of variation is greater than the preset threshold of the second coefficient of variation.

[0082] Specifically, the preset threshold of the first coefficient of variation is recorded as The preset threshold of the second coefficient of variation is recorded as Compare The first coefficient of variation, and the size of the second coefficient of variation, where

[0083] If there is a coefficient of variation value greater than the threshold (the first coefficient of variation is greater than the preset threshold of the first coefficient of variation or the second coefficient of variation is greater than the preset threshold of the second coefficient of variation), it is considered that the L group of data of the key control variable in the data block should not be regarded as the steady-state data of the key control variable. If both coefficient of variation values ​​are smaller than the threshold (the first coefficient of variation is less than the preset threshold of the first coefficient of variation and the second coefficient of variation is less than the preset threshold of the second coefficient of variation), it means that the L group of data of the key control variable in the data block has a small degree of discreteness and high stability, and can be regarded as the steady-state data of the key control variable.

[0084] It is worth mentioning that the standard of the second coefficient of variation is stricter than that of the first coefficient of variation, which can improve the stability of the data after screening.

[0085] The specific process of step 3 (obtaining the initial parent population corresponding to the key control variables based on the evolutionary algorithm) is exemplified below.

[0086] The first step is to determine the control parameters of the evolutionary algorithm; the control parameters include the population size N, the dimension D of each individual in the population, and the maximum evolutionary generation E of the population.

[0087] The second step is to calculate the formula

[0088]

[0089] Get the initial parent population corresponding to the key control variables, where x i,j (0) represents the value of the jth dimension of the i-th individual in the initial parent population, rand(0,1) represents a random number that follows a uniform distribution in the interval (0,1), represents the upper bound of the value of the jth dimension of the ith individual, It represents the lower bound of the value of the jth individual on the jth dimension. Each individual in the initial parent population includes a set of values ​​of all key control variables determined, a set of values ​​is the value in the steady-state data, and multiple individuals in the initial parent population are different from each other.

[0090] It should be noted that the principle of obtaining the initial parent population based on the evolutionary algorithm is common knowledge and will not be elaborated here.

[0091] The following is an exemplary explanation of the specific process of step 5 (for each individual in the intermediate generation variant population, if the value of the individual in the j-th dimension exceeds the preset boundary, the value of the individual in the j-th dimension is reset to the average of the values ​​of any two different individuals in the intermediate generation variant population in the j-th dimension).

[0092] By calculating the formula

[0093]

[0094] For individuals in the intermediate generation mutant population whose values ​​on the jth dimension exceed the preset boundary, the values ​​on this dimension are reset, where v i,j (e) represents the value of the jth dimension of the ith individual in the intermediate generation mutation population of the eth generation. r1 and r2 are two different random numbers in [1,2,...,N]. v r1,j (e) and v r2,j (e) represent the values ​​of two individuals randomly selected from the intermediate generation mutation population in the jth dimension, e = 1, 2, ..., E.

[0095] The specific process of step 6 (performing a crossover operation on the individuals in the intermediate generation mutation population and the individuals in the parent generation population before the mutation of the intermediate generation mutation population to obtain the first mutation population and the second mutation population) is exemplified below.

[0096] Step 6.1, by calculating the formula

[0097]

[0098] Get the first mutation population; where x i,j (e) represents the value of the jth dimension of the i-th individual in the e-th generation parent population, v i,j (e) represents the value of the jth dimension of the i-th individual in the e-th intermediate generation mutation population, u i,j (e) represents the first mutant population u i (e) The value of the i-th individual on the j-th dimension, R represents the crossover probability, j rand is a random number in [1, 2, …, D].

[0099] Step 6.2, by calculating the formula

[0100]

[0101] Get the second mutant population; among them, w i,j (e) represents the value of the jth dimension of the i-th individual in the second mutant population.

[0102] The specific process of step 7 (generating the intermediate generation parent population according to the parent population before the intermediate generation mutation population mutation, the first mutation population and the second mutation population) is exemplified below.

[0103] By calculating the formula

[0104]

[0105] Get the intermediate generation parent population {x i (e+1)},i=1,...,N; where x i (e) represents the i-th individual in the e-th generation parent population, u i (e) represents the i-th individual in the first mutation population of the e-th generation, w i (e) represents the i-th individual in the second variant population of the e-th generation, and f(x) represents the individual x (x is x i (e) and u i (e) w i (e))The corresponding objective function value.

[0106] When building a prediction model, since different historical data sets are obtained, their sample similarity relationships may be different. Therefore, a single similarity measure is not enough, and sometimes even fails to capture the true similarity structure between data samples. Based on this, this application proposes an integrated local weighted partial least squares method, which is specifically implemented before taking the optimal individual in the parent population as the value of the key control variable, including the following steps:

[0107] Step a, for each individual in the intermediate parent population, calculate the formula

[0108]

[0109] Get the hth historical sample x h Based on the first similarity measurement method, the similarity weight S1,h with the query sample is calculated, where x h represents the hth historical sample, and the historical sample represents several values ​​of all key control variables obtained, x q represents the query sample, which represents an individual in the intermediate parent population, and the individual includes a set of values ​​of all key control variables; d hrepresents the Euclidean distance between the hth historical sample and the query sample; p1 represents an adjustable parameter used to adjust the decay rate of the similarity weight with distance.

[0110] It should be noted that the first similarity measurement method is defined based on the most widely used Euclidean distance.

[0111] Step b, by calculating the formula

[0112]

[0113] d h =||x q -x h ||2

[0114]

[0115] Get the hth historical sample x h The similarity weight S2,h with the query sample calculated based on the second similarity measurement method, where λ represents the trade-off parameter, θ h Represents x q and x h The angle between them, p2 is an adjustable parameter used to adjust the decay rate of the similarity weight with distance; the second similarity measurement value measures the additional angle information beyond the distance. If two historical samples have the same distance from the query sample, the angle θ h Small samples will be preferentially selected as local modeling samples.

[0116] It can be seen that the similarity calculation method in the embodiment of the present application adopts two similarity measurement methods (the first similarity measurement method combines the Euclidean distance, and the second similarity measurement method combines the Euclidean distance and the angle), which can better capture the similarity structure between data samples.

[0117] Step c, by calculating the formula

[0118]

[0119] Get query sample x q The predicted value of the corresponding target variable in, represents the query sample x calculated by the kth similarity metric method q The corresponding predicted value of the target variable, γ k represents the integrated weight of the k-th similarity measurement method, R k Indicates the k-th similarity metric method in the cross-validation dataset {(X C ,Y C )}, {(XC ,Y C )}={(x1,y1),...,(x C ,y C )},x c (c=1,…,C) represents the value of a set of key control variables, y c Indicates the corresponding target variable value, K = 2, y c represents the control variable x of group c c The actual value of the corresponding target variable, represents the cth group of control variables x obtained by the kth similarity measurement method c The predicted value of the corresponding target variable.

[0120] Calculate the objective function value of the predicted value corresponding to all individuals in the intermediate generation parent population, and take the individual corresponding to the objective function value closest to the preset threshold of the objective function as the optimal individual.

[0121] Specifically, the objective function value γ of the predicted value y can be calculated by the calculation formula γ=ψ(y). When the objective function value corresponding to the predicted value is closest to the preset threshold of the objective function, the individual corresponding to the predicted value is selected as the optimal individual. ψ represents the objective function, which can be a commonly used linear function, that is, the objective function value γ is linearly related to the predicted value y.

[0122] Specifically, when a query sample x q When it arrives, each similarity measurement method is used to calculate the similarity weight with all historical samples. By rearranging the obtained similarity weights in descending order, the locally relevant samples can be determined. Here, it is assumed that the relevant input samples and output samples are The weights corresponding to these samples are S k Indicates that the kth similarity is used. And using {(X k ,Y k )}and An online local model y=f can be constructed for the query sample k (x), assuming that x is obtained using the k-th similarity measurement method q The predicted output value is Integrate the prediction values ​​calculated by each similarity measurement method to obtain the final prediction value

[0123] R k It can reflect the prediction performance of the prediction model when the kth similarity measurement method is selected, R k The smaller it is, the higher the prediction accuracy of the model is. Accordingly, a larger weight should be assigned in the integration process.

[0124] In order to better understand the technical solution provided by the present application, the method for determining the control variables in the oil refining production provided by the present application is exemplified below in combination with specific embodiments.

[0125] Embodiment 1

[0126] The double variation strategy based Differential Evolution Algorithm (DV-DE) algorithm is used to find the minimum positive deviation between the 95% distillation temperature of diesel in the hydrocracking process of a refinery and the maximum value of 365°C of the 95% distillation temperature of diesel required by the National V standard.

[0127] The 95% distillation temperature of diesel to be optimized is taken as the target variable y. In this example, the minimum positive deviation between the 95% distillation temperature of diesel and 365°C is sought, that is, the minimum value of (365-y). Therefore, the objective function of the double mutation differential evolution algorithm can be determined as follows:

[0128] γ=min(365-y)

[0129] In the actual production process, 89 control variables related to y were collected. The present invention combines the Pearson correlation coefficient and the Kendall correlation coefficient to perform correlation screening on the historical data of the hydrocracking process collected under the background of stable working conditions. The i-th control variable x in the historical data i The linear correlation between the target variable y and the control variable y is expressed by the Pearson correlation coefficient P i The calculation formula is as follows:

[0130]

[0131] Where i = 1,…,89, and Respectively represent x i and the standard deviation of y.

[0132] Kendall correlation coefficient K i The nonlinear correlation can be measured as follows:

[0133]

[0134]

[0135] Among them, C i It means (x i ,u) has the same number of pairs of elements, D i It means (x i,y) has inconsistent pairs of elements, (x i ,y) represents the control variable x i The jth element x i,j and the jth element y of the target variable y j The set of elements composed of i It means (x i ,y) has inconsistent element pairs, S represents the control variable x i The number of the first set of identical elements in P s represents the number of elements in the sth first set, T represents the number of second sets consisting of the same elements in the target variable y, Q t Indicates the number of elements in the t-th second set, the control variable x i The number of elements of the target variable y is n, t=1,2,...,T, s=1,2,...,S.

[0136] The calculation formula of the weighted correlation coefficient is as follows:

[0137] W i =α1P i +α2K i

[0138] Among them, α1 and α2 represent P i , K i The weight parameter of .

[0139] The control variable is W i Arrange from large to small and set a proportional coefficient ρ based on empirical knowledge

[0140] m c =mρ

[0141] Keep the first m c The control variables are used as the key control variables x:

[0142]

[0143] is a control variable in the key control variable group, and the target variable and the control variable meet

[0144]

[0145] in Represents a mapping relationship of x→y. The present invention selects the local partial least squares method to express this mapping, constructs a prediction model, and realizes the prediction of x to y (control variable predicts target variable).

[0146] Through variable correlation analysis, 40 relevant control variables x that have a greater impact on the 95% distillation temperature y of diesel in the actual hydrocracking process are finally selected, which can be recorded as x1, x2, ..., x 40 , a control variable corresponds to an individual in the population of the evolutionary algorithm. The weighted correlation coefficient can be used to screen key control variables and reduce the complexity of subsequent prediction model establishment.

[0147] The coefficient of variation method is used as the basis for determining steady-state data, and its calculation formula is as follows:

[0148]

[0149] Among them, s i represents the standard deviation of the ith variable, μ i represents the mean of the i-th variable, σ i represents the variance of the ith variable. (Coefficient of variation) can objectively reflect the absolute value of the degree of data dispersion without considering the dimension. The smaller its value is, the smaller the degree of data dispersion is.

[0150] This application uses a two-coefficient of variation (first coefficient of variation and second coefficient of variation) steady-state discrimination method to extract steady-state data (steady-state data of key control variables) in the oil refining production process. Among them, the first coefficient of variation refers to the coefficient of variation of each key variable in the current sliding window area; the second coefficient of variation refers to the coefficient of variation of the first coefficient of variation. Finally, a total of 995 groups of steady-state data of key control variables were extracted through the two-coefficient of variation steady-state discrimination method.

[0151] This application uses two representative measurement methods when selecting relevant samples to calculate similarity. The first similarity measurement method is based on Euclidean distance, which is defined as

[0152]

[0153] Among them, x h With x q denote historical samples and query samples respectively; S1,h is the hth historical sample x h Calculate the similarity weight with the query sample based on the first similarity measurement method; d h represents the Euclidean distance between the hth historical sample and the query sample; p1 is an adjustable parameter used to adjust the decay rate of the similarity weight with distance.

[0154] The second similarity measure is based on distance and angle information and is defined as:

[0155]

[0156] dh =||x q -x h ||2

[0157]

[0158] Among them, λ is a trade-off parameter between 0 and 1; θ h is x q (query sample) and x h (historical samples). That is, the second similarity measures the additional angle information besides the distance. If two historical samples have the same distance from the query sample, the angle x h Small samples will be preferentially selected as local modeling samples.

[0159] Using S1 and S2 to get x q The predicted output values ​​are and Integrate the prediction values ​​calculated by each similarity measurement method to obtain the final prediction value

[0160]

[0161] Among them, γ k is the integrated weight of the k-th similarity.

[0162] Calculate each S k (kth similarity) in the cross validation dataset {(X C ,Y C )}={(x1,y1),…,(x C ,y C )} on the root mean square error R k

[0163]

[0164] Among them, y c and Represent the true output value of the data set and the output prediction value of the local partial least squares method respectively.

[0165] R k It can reflect the selection of similarity S k The prediction performance of the local partial least squares model when R k The smaller the value, the higher the prediction accuracy of the method. Accordingly, a larger weight should be assigned in the integration process. Therefore, the integration weight function is designed as follows:

[0166]

[0167] In summary, the local partial least squares method not only reflects the absolute difference in distance but also takes into account the relative difference in direction. Therefore, the online prediction model of distillation temperature based on it has better performance.

[0168] Determine the basic parameters of the double mutation differential evolution algorithm: population size N = 40, dimension of each individual D = 40, maximum evolutionary generations E = 1000.

[0169] Generate the initial parent population:

[0170]

[0171] Among them, x i,j (0) represents the value of the jth dimension of the i-th individual in the initial parent population, rand(0,1) represents a random number that follows a uniform distribution in the interval (0,1), represents the upper bound of the value of the jth dimension of the ith individual, represents the lower bound of the value of the i-th individual on the j-th dimension, i = 1, 2, ..., 40, j = 1, 2, ..., 40. Each individual in the initial parent population corresponds to a set of values ​​of all key control variables.

[0172] Use local weighted partial least squares to predict each individual x in the initial population i The corresponding diesel 95% distillation temperature y i , and calculate the objective function f i =365-y i The value of the objective function f i The smaller the y i The smaller the positive deviation from 365℃, the smaller the corresponding population individual x i The better.

[0173] Parent population individual x i Perform mutation operation to obtain the intermediate mutation population v i The specific process is: randomly select two different individuals from the parent population, scale them by vector difference, and then perform vector synthesis with the individual to be mutated to obtain the intermediate generation mutant population individual:

[0174] v i (e) = x i (e)+F×[x r1 (e)-x r2 (e)],i≠r1≠r2

[0175] Where 0≤e≤1000, F is the scaling factor, x i (e) represents the i-th individual in the parent population of the e-th generation, v i(e) represents the u-th individual in the e-th intermediate generation mutation population.

[0176] Judgement i Whether the value of each dimension is within the set range, for dimension values ​​that are out of range, the mean of any two different individuals in the population on that dimension is used to regenerate, that is, or hour:

[0177]

[0178] Among them, 0≤e≤1000, v i,j (e) represents the value of the jth dimension of the ith individual in the intermediate mutation population of the eth generation, r1 and r2 are two different random numbers in [1,2,...,40], and v r1,j (e) and (v) r2,j (e) represent the values ​​of two different individuals r1 and r2 randomly selected from the e-th intermediate generation mutation population on the j-th dimension.

[0179] Cross the parent population individuals with the intermediate mutation population individuals.

[0180] The specific process of generating the first mutant population is as follows:

[0181]

[0182] The specific process of generating the second mutant population is as follows:

[0183]

[0184] Among them, 0≤e≤1000, x i,j (e) represents the value of the jth dimension of the i-th individual in the parent population of the e-th generation, v i,j (e) represents the value of the jth dimension of the ith individual in the intermediate mutation population of the eth generation, u i,j (e) represents the value of the jth dimension of the i-th individual in the first variant population, w i,j (e) represents the value of the jth dimension of the i-th individual in the second variant population, R is the crossover probability, j rand is a random integer in [1,2,...,40].

[0185] In the parent population of generation e, {x i (e)}, the first variant population {u i (e)} and the second variant population {w i (e)}, and obtain the optimal solution of the e-th generation:

[0186]

[0187] Among them, 0≤e≤1000, x i (e+1) represents the e+1th generation parent population, and f(x) represents the objective function value corresponding to individual x.

[0188] If the evolution termination condition or the maximum evolution generation is reached, min{f(x i (e+1))} is the global optimal solution obtained by the optimization algorithm; if it is not reached, x i (e+1) is regarded as the parent population of the e+1th generation and continues to evolve until the global optimal solution is obtained.

[0189] Embodiment 2

[0190] The double variation strategy based particle swarm optimization algorithm (DV-PSO) is used to find the minimum positive deviation between the 95% distillation temperature of diesel and the maximum value of 365°C of the 95% distillation temperature of diesel required by the National V standard.

[0191] The method of screening out key control variables, extracting steady-state data of key control variables, and constructing a prediction model of control variables to targets is the same as that in the first embodiment.

[0192] The basic parameter settings of the DV-PSO algorithm are as follows:

[0193] The particle swarm size N = 40, the dimension of each particle D = 40, and the maximum evolutionary generation E = 1000.

[0194] The generation method of the initial particle group is similar to the generation method of the initial population in Example 1. However, particles have two basic properties, speed and position.

[0195] The position of the particle is the search space of the particle, so the particle x i The position interval is:

[0196] The initial position of the i-th particle in the initial particle group is x i (0):

[0197]

[0198] The speed of a particle indicates how fast it moves. The speed range of a particle group is [v min ,v max ]:

[0199]

[0200] The initial velocity of the i-th particle in the initial particle group is v i (0):

[0201]

[0202] Among them, the number of particles in the particle swarm is 40, so 1≤i≤40, rand(0,1) represents a random number that obeys a uniform distribution in the interval (0,1), and the position information of all particles constitutes the search space of the entire particle swarm. The speed and position of the particles are randomly initialized in the search space and speed interval.

[0203] The method of calculating the individual objective function value of the initial particle swarm and evaluating the initial particle swarm is the same as that in the first embodiment.

[0204] Update the velocity and position of the particle group:

[0205] Updated speed v' i (e)

[0206] v′ i (e) = w × v i (e)+c1×rand(0,1)×(pbest i -x i (e))+c2×rand(0,1)

[0207] ×(gbest i -x i (e))

[0208] Updated position x' i (e)

[0209] x′ i (e) = x i (e)+v′ i (e)

[0210] Among them, 1≤u≤40; v i (e), x i (e) is the velocity and position of the e-th generation particle before updating; w is the inertia factor, which is non-negative; rand(0,1) is a random number uniformly distributed in the interval (0,1); c1 and c2 are learning factors, c1=c2=1; pbest i (e) is the optimal position of the i-th particle in the current e-th generation particle swarm, and gbest(e) is the global optimal position of the current e-th generation particle.

[0211] Determine whether the updated speed and dimensional position of each particle are within the set speed range and position range. For particles whose speed exceeds the range, if v' i (e) <v min, then v' i (e) = v min ; if v' i (e)>v max , then v' i (e) = v max If the position of the ith particle in the jth dimension exceeds the interval range, it is regenerated using the mean of the positions of any two different particles in the particle swarm in this dimension, that is, when or

[0212]

[0213] Among them, 0≤e≤1000, 1≤i,j≤40, x' i,j (e) is the new position of the ith particle in the jth dimension, x' i1,j (e) and x' i2,j (e) represent the positions of the j-th dimension of two different particles randomly selected from the updated particle swarm.

[0214] Generate the first mutant particle swarm:

[0215]

[0216] Generate the second mutant particle swarm:

[0217]

[0218] Among them, 0≤e≤1000, x i,j (e) represents the position of the i-th individual in the parent particle swarm of the e-th generation in the j-th dimension, x' i,j (e) represents the position of the i-th individual in the j-th dimension after the parent particle swarm performs the update operation, u i,j (e) represents the position of the jth dimension of the ith individual in the first mutation particle swarm of the eth generation particle swarm, w i,j (e) represents the position of the jth dimension of the i-th individual in the second mutation particle swarm of the e-th generation particle swarm, R is the crossover probability, j rand is a random integer in the range [1,2,…,40].

[0219] In the parent particle swarm of the e-th generation {x i (e)}, the first mutant particle swarm {u i (e)} and the second mutant particle swarm {w i (e)}, and obtain the optimal solution of the e-th generation:

[0220]

[0221] Among them, 0≤e≤1000, xi (e+1) represents the (e+1)th generation parent population, and f(x) represents the objective function value corresponding to individual x.

[0222] If the evolution termination condition or the maximum evolution generation is reached, min{f(x i (e+1))} is the global optimal solution obtained by the optimization algorithm; if the evolution termination condition or the maximum evolution generation is not reached, x i (e+1) is regarded as the parent particle swarm of the e+1th generation and continues to evolve until the global optimal solution is obtained.

[0223] The parameters used in the above modeling process have been experimentally adjusted to the optimal values. In order to verify the effectiveness of the proposed method, the particle swarm optimization algorithm (PSO) and the DV-PSO algorithm are used to seek the minimum positive deviation between the 95% distillation temperature of diesel and 365°C. After modeling and verification, the final comparison results are as follows: Figure 2 As shown, after the two algorithms use the method provided by this application, the algorithm optimization time is greatly shortened compared with not using the method provided by this application, which verifies the effectiveness of the method provided by this application. Figure 2 The vertical axis is the objective function value, which represents the optimization ability. Figure 3a , Figure 3b , Figure 3c , Figure 3d They represent the convergence curves of the particle swarm optimization algorithm and the double mutation particle swarm optimization algorithm running independently. Figure 3a , Figure 3b , Figure 3c , Figure 3d )The horizontal axis represents the evolutionary generation, and the vertical axis represents the optimal solution. Figure 4a , Figure 4b , Figure 4c , Figure 4d They represent the convergence curves of the differential evolution algorithm and the double mutation differential evolution algorithm running independently. Figure 4a , Figure 4b , Figure 4c , Figure 4d )The horizontal axis represents the evolutionary generation, and the vertical axis represents the optimal solution. Figure 2 and Figure 3( Figure 3a , Figure 3b , Figure 3c , Figure 3d ) respectively show that the DV-PSO algorithm has stronger optimization ability and faster convergence speed in actual industrial production operation, which verifies that the DV-PSO algorithm has better performance. Combined with Figure 4 ( Figure 4a , Figure 4b , Figure 4c , Figure 4d) It can be seen that the DV-DE algorithm has a stronger optimization ability in actual industrial production operation than the DE algorithm, and the convergence speed is also improved, which verifies that the DV-DE algorithm has better performance than the DE algorithm.

[0224] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining a control variable in oil refining production, characterized in that: include: Step 1, determining the target variables and control variables to be optimized in oil refining production, and determining the key control variables from the control variables; Step 2, extracting steady-state data of the key control variables based on a multiple coefficient of variation method; Step 3, obtaining an initial parent population corresponding to the key control variables based on an evolutionary algorithm; wherein each individual in the initial parent population includes a set of values ​​of all the determined key control variables, the set of values ​​is the value in the steady-state data, and multiple individuals in the initial parent population are different from each other; Step 4, performing mutation operation on the individuals in the initial parent population to generate an intermediate generation mutation population; Step 5: for each individual in the intermediate generation variant population, if the value of the individual in the j-th dimension exceeds the preset boundary, the value of the individual in the j-th dimension is reset to the average value of the values ​​of any two different individuals in the intermediate generation variant population in the j-th dimension; Step 6, performing a crossover operation on the individuals in the intermediate generation mutation population and the individuals in the parent generation population before the intermediate generation mutation population mutates, to obtain a first mutation population and a second mutation population; Step 7, generating an intermediate generation parent population according to the parent population before the intermediate generation mutation population mutation, the first mutation population and the second mutation population; Step 8, if the population evolution meets the preset evolution termination condition, the optimal individual in the intermediate generation parent population is used as the value of the key control variable; otherwise, the intermediate generation parent population is mutated to generate a new intermediate generation mutant population, and the new intermediate generation mutant population is used as the intermediate generation mutant population in step 5 and returned to execute step 5; The method of extracting the steady-state data of the key control variables based on the multiple coefficient of variation method includes: Acquire historical data of the key control variables to obtain a historical data set X, and divide the historical data set into N data blocks; wherein each data block has L groups of data, and each group of data includes a set of values ​​of all key control variables; For each data block, respectively, calculating the first coefficient of variation of each key control variable in the data block; Calculate the second coefficient of variation of the first coefficient of variation of each key control variable; For each data block, if the first coefficient of variation of a key control variable in the data block is less than the preset threshold of the first coefficient of variation, and the second coefficient of variation corresponding to the key control variable is less than the preset threshold of the second coefficient of variation, the value of the key control variable in the data block is used as the steady-state data of the key control variable; wherein the preset threshold of the first coefficient of variation is greater than the preset threshold of the second coefficient of variation; Determining the key control variables from the control variables includes: By calculating the formula Get the linear correlation degree P between each control variable and the target variable i , where P i represents the i-th control variable x i Pearson correlation coefficient for the target variable y, represents the i-th control variable x i The standard deviation of represents the standard deviation of the target variable y, 1≤i≤m, and m represents the number of control variables related to the target variable; By calculating the formula Get the nonlinear correlation degree K between each control variable and the target variable i , where K i represents the i-th control variable x i Kendall correlation coefficient for the target variable y, C i It means (x i ,y) has the same number of pairs of elements, (x i ,y) represents the control variable x i The jth element x i,j and the jth element y of the target variable y i The set of elements composed of i It means (x i ,y) has inconsistent element pairs, S represents the control variable x i The number of the first set of identical elements in P s represents the number of elements in the sth first set, T represents the number of second sets consisting of the same elements in the target variable y, Q t Indicates the number of elements in the t-th second set, the control variable x i The number of elements of the target variable y is n, t=1,2,...,T, s=1,2,...,S; By calculating the formula W i =α1P i +α2K i Get the weighted correlation coefficient W between each control variable and the target variable i , where W i represents the i-th control variable x i The weighted correlation coefficient of the target variable y, α1 represents P i The weight parameter, α2 represents K i The weight parameter of Arrange all control variables in descending order according to the weighted correlation coefficient, and take the first m c control variables as the key control variable x; where m c =mρ, ρ represents the proportionality coefficient, For each individual in the intermediate generation variant population, if the value of the individual in the j-th dimension exceeds a preset boundary, resetting the value of the individual in the j-th dimension to the average value of the values ​​of any two different individuals in the intermediate generation variant population in the j-th dimension includes: By calculating the formula For individuals in the intermediate generation mutant population whose values ​​on the jth dimension exceed the preset boundary, the values ​​on this dimension are reset, where v i,j (e) represents the value of the jth dimension of the ith individual in the eth generation intermediate generation mutant population. r1 and r2 are two different random numbers in [1,2,...,N]. v r1,j (e) and v r2,j (e) represent the values ​​of two different individuals randomly selected from the e-th generation intermediate generation mutation population on the j-th dimension, e = 1, 2, ..., E.

2. The determination method according to claim 1, characterized in that: The initial parent population corresponding to the key control variable is obtained based on the evolutionary algorithm, including: Determine the control parameters of the evolutionary algorithm; wherein the control parameters include the population size N, the maximum dimension D of each individual in the population, and the maximum evolutionary number E of the population; By calculating the formula Get the initial parent population, where x i,j (0) represents the value of the jth dimension of the ith individual in the initial parent population, each individual includes a set of values ​​of all key control variables determined, rand(0,1) represents a random number that follows a uniform distribution in the interval (0,1), represents the upper bound of the value of the jth dimension of the ith individual, Represents the lower bound of the value of the jth dimension of the ith individual.

3. The determination method according to claim 2, characterized in that: The step of performing a crossover operation on the individuals in the intermediate generation mutation population and the individuals in the parent generation population before the intermediate generation mutation population mutates to obtain a first mutation population and a second mutation population comprises: By calculating the formula Get the first variant population u i,j (e); where x i,j (e) represents the value of the jth dimension of the i-th individual in the e-th generation parent population, v i,j (e) represents the value of the jth dimension of the i-th individual in the e-th intermediate generation mutation population, u i,j (e) represents the first mutant population u i (e) The value of the i-th individual on the j-th dimension, R represents the crossover probability, j arnd is a random number in [1,2,...,D]; By calculating the formula Get the second variant population w i,j (e); where w i,j (e) represents the second mutant population w i (e) The value of the jth dimension of the ith individual.

4. The determination method according to claim 3, characterized in that: The step of generating an intermediate generation parent population according to the parent population before the intermediate generation mutation population is mutated, the first mutation population, and the second mutation population comprises: By calculating the formula Get the intermediate generation parent population {x i (e+1)},i=1,...,N; where x i (e) represents the i-th individual in the e-th generation parent population, u i (e) represents the i-th individual in the first mutation population of the e-th generation, w i (e) represents the i-th individual of the second mutation population of the e-th generation, and f(x) represents the objective function value corresponding to individual x.

5. The determination method according to claim 1, characterized in that: Before taking the best individual in the intermediate generation parent population as the value of the key control variable, the determination method further includes: For each individual in the intermediate parent population, the formula Get the hth historical sample x h Based on the first similarity measurement method, the similarity weight S1,h with the query sample is calculated, where x h represents the hth historical sample, x q represents a query sample, the query sample is an individual in the intermediate generation parent population, and the individual includes a set of values ​​of all key control variables; d h represents the hth historical sample x h With query sample x q The Euclidean distance between them; p1 represents an adjustable parameter used to adjust the decay rate of the similarity weight with distance; By calculating the formula d h =||x q -x h ||2 Get the hth historical sample x h The similarity weight S2,h with the query sample calculated based on the second similarity measurement method, where λ represents the trade-off parameter, θ h Represents x q and x h The angle between them, p2 represents an adjustable parameter, which is used to adjust the decay rate of similarity weight with distance; By calculating the formula Get query sample x q The predicted value of the corresponding target variable in, represents the query sample x calculated by the kth similarity metric method q The corresponding predicted value of the target variable, γ k represents the integrated weight of the k-th similarity measurement method, R k Indicates the k-th similarity metric method in the cross-validation dataset {(X C ,Y C )}, {(X C ,Y C )}={(x1,y1),…,(x C ,y C )},x c (c=1,…,C) represents the value of a set of key control variables, y c represents the target variable value corresponding to the set of key control variables, K = 2, y c represents the control variable x of group c c The actual value of the corresponding target variable, represents the cth group of control variables x obtained by the kth similarity measurement method c The corresponding predicted value of the target variable; The objective function values ​​of the predicted values ​​corresponding to all individuals in the intermediate generation parent population are calculated, and the individual corresponding to the objective function value closest to the preset threshold of the objective function is taken as the optimal individual.

6. The determination method according to claim 5, characterized in that: Before step 8, the determination method further includes: Count the number of times the intermediate generation parent population is generated; If the generation times of the intermediate generation parent population reaches the preset maximum evolution times, it is determined that the population evolution meets the preset evolution termination condition; otherwise, it is determined that the population evolution does not meet the preset evolution termination condition.

Images