A parallel hyper-heuristic injection-production optimization method based on proxy model

Abstract

The application discloses a parallel hyper-heuristic injection-production optimization method based on a proxy model and belongs to the technical field of oil reservoir injection-production optimization, and comprises the following steps: step 1, constructing a three-dimensional geological model and an oil reservoir injection-production model; step 2, constructing a water drive oil reservoir injection-production optimization model and an initial database; step 3, constructing a proxy model; step 4, generating a subpopulation by using four algorithms for parallel evolution, evaluating the subpopulation by using the proxy model, and obtaining a candidate solution of each evolution algorithm; step 5, constructing a hyper-heuristic guided search strategy and an online parameter returning method, adjusting the search behavior of an evolutioner based on the candidate solution; step 6, repeating steps 3 to 5 until a preset maximum optimization iteration number is reached; and step 7, outputting an optimal Pareto front scheme and a corresponding economic net present value. The method can improve the injection-production optimization efficiency of the water drive oil reservoir, and dynamically adjusts the search direction by using the hyper-heuristic guided search strategy and the online parameter returning method.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of oil reservoir injection and production optimization, and particularly relates to a parallel hyper-heuristic injection and production optimization method based on a proxy model. BACKGROUND

[0002] There are many difficulties and challenges in the process of oil reservoir exploitation, especially in the high water cut or super high water cut stage, the rapid rise of water cut and the significant decline of oil production bring great pressure to the optimization of injection and production schemes and the control of water cut. In order to cope with these challenges, oil reservoir numerical simulation technology is widely used in production optimization to help predict reservoir dynamics and evaluate the effect of different well control schemes. However, the calculation cost of oil reservoir numerical simulation is extremely high, especially in the production optimization of large-scale oil reservoirs, a single simulation run may take tens of minutes or even hours. Therefore, it is urgent to develop an efficient algorithm to accelerate the optimization convergence and improve the production efficiency and economic benefits.

[0003] The proxy-assisted evolutionary algorithm combines evolutionary algorithms with proxy models, uses evolutionary algorithms with strong global search ability to quickly generate a large number of candidate Pareto solutions, uses low-cost proxy models to replace real calculations, and quickly evaluates new individuals to approximate target solutions with lower computational cost, significantly reducing the number of numerical simulation calls. This combination has shown good application potential and engineering value in dealing with oil reservoir injection and production optimization, and has become an important means to improve optimization efficiency.

[0004] Existing proxy-assisted evolutionary algorithms mostly use single evolutionary algorithms, which have limited search ability and adaptability, and also have limitations such as lack of algorithm diversity, rigid parameter configuration, and insufficient utilization of parallel resources, making it difficult to meet the growing demand for algorithm flexibility in oil reservoir injection and production optimization, thereby affecting the production efficiency and economic benefits of oil fields. Therefore, it is urgent to develop a method that can fully utilize the diversity of evolutionary algorithms and dynamically adjust and optimize production strategies to improve the exploitation efficiency and economic benefits of oil fields. SUMMARY

[0005] To solve the problem of insufficient search ability and weak adaptability of existing proxy-assisted evolutionary algorithms in water drive injection and production optimization, the application proposes a parallel hyper-heuristic injection and production optimization method based on a proxy model. The proposed parallel co-evolution framework improves the coverage ability of evolutionary algorithms on the search space and the diversity of solutions, enhances their ability to solve high-dimensional complex engineering problems, and the proposed hyper-heuristic guided search strategy and online parameter return method have both global and flexible characteristics, which can adjust the search behavior and direction of different evolutionary algorithms in real time during the optimization process, accelerate the advancement of the Pareto frontier, and realize stronger adaptability to changes in the oil reservoir environment during the optimization process, which is conducive to guiding the efficient design of water drive injection and production schemes for complex oil reservoirs.

[0006] The technical solutions of the present application are as follows:

[0007] A parallel hyper-heuristic injection-production optimization method based on a proxy model, comprising the following steps:

[0008] Step 1, obtaining oilfield water drive injection-production information, constructing a three-dimensional geological model and an oil reservoir injection-production model;

[0009] Step 2, constructing a water drive reservoir injection-production optimization model, using Latin hypercube sampling to obtain multiple injection-production schemes as an initial sample set, using a reservoir numerical simulator to calculate the economic net present value corresponding to each sample in the initial sample set, and constructing an initial database;

[0010] Step 3, screening the initial samples based on the economic net present value of the initial samples, and constructing a proxy model based on the economic net present value as an optimization objective;

[0011] Step 4, using genetic algorithm, particle swarm algorithm, annealing algorithm and tabu algorithm to generate a sub-population, using the proxy model to evaluate the sub-individuals, and obtaining the candidate solutions of each evolution algorithm;

[0012] Step 5, performing reservoir numerical simulation on the candidate solutions, constructing a hyper-heuristic guided search strategy and an online parameter return method, and adjusting the search behavior of the evolution based on the economic net present value of the candidate solutions;

[0013] Step 6, repeating steps 3 to 5 until a preset maximum optimization iteration number is reached, recording the samples evaluated using the reservoir numerical simulation and the reservoir economic net present value corresponding to the samples at each iteration optimization, and obtaining an updated database;

[0014] Step 7, after the iteration optimization is completed, the reservoir numerical simulator outputs the optimal Pareto front scheme and the corresponding economic net present value, and the water drive reservoir injection-production optimization is completed.

[0015] Further, in step 1, the oilfield water drive injection-production information includes oilfield geological data and oilfield production information; the oilfield geological data includes grid information of the oilfield, physical properties of rocks in the region where the oilfield is located, and fluid properties of oil and water; the oilfield production information includes types and positions of wells, and liquid production of each production well at each time step and water injection of each water injection well at each time step before optimization;

[0016] According to the oilfield geological data, a three-dimensional geological model is constructed by using an oilfield geological modeling software Petrol, the three-dimensional geological model is imported into a reservoir numerical simulator, and then an injection-production model of the reservoir is simulated by using a numerical simulation software Eclipse according to the oilfield production information, and is stored in the reservoir numerical simulator.

[0017] Further, the specific process of step 2 is:

[0018] Step 2.1: Construct the water-drive reservoir injection-production optimization model as follows:

[0019] (1);

[0020] in, For Find the maximum value of the objective function for the independent variable; The objective function is... Indicates the injection and extraction scheme; It is the economic net present value; For dimensions; for A real number of dimension; and They represent The lower and upper limits; the formula for calculating the economic net present value is as follows:

[0021] (2);

[0022] in, For time steps; For the block in the 1st Oil production rate at each time step; Indicates oil extraction revenue; For the block in the 1st Water production rate at each time step; Indicates the cost excluding water; For the block in the 1st The water injection rate at each time step; Indicates the cost of water injection; The annual discount rate; For time step;

[0023] Step 2.2: Use Latin hypercube sampling to obtain multiple samples within the upper and lower limits of the decision variables to construct an initial sample set. ,in, For the first The injection and collection scheme for each sample. This is the initial sample size;

[0024] Step 2.3: Initial sample set Import the data into the reservoir numerical simulator, and use the simulator to calculate the reservoir economic net present value for each sample, obtaining the result compared to the initial sample set. Corresponding reservoir economic net present value dataset , ,in, For the first The economic net present value of an oil reservoir for a sample;

[0025] Step 2.4, based on the initial sample set and its corresponding oil reservoir economic net present value data set , constructing an initial database .

[0026] Further, the specific process of step 3 is as follows:

[0027] Step 3.1, setting the maximum number of iterations of reservoir numerical simulation optimization , sorting the samples in the initial database according to the oil reservoir economic net present value from high to low, and selecting the top samples with the highest economic net present value as the training set for the agent model;

[0028] Step 3.2, constructing a radial basis function agent model based on the samples in the training set; the radial basis function agent model is constructed based on a Gaussian kernel function, and the formula is:

[0029] (3)

[0030] wherein, is the radial basis function; is the serial number of the sample in the agent model training set; is the number of samples in the agent model training set; is the weight of the th sample in the training set; is the injection-production scheme of the th sample in the agent model training set; is the Euclidean distance between samples; is the kernel function, and the kernel function is a Gaussian function, and the expression is:

[0031] (4)

[0032] wherein, is the exponential function; is the shape parameter of the Gaussian function, and the expression is:

[0033] (5)

[0034] wherein, is the maximum distance between all sample pairs in the agent model training set.

[0035] Further, the specific process of step 4 is as follows:

[0036] Step 4.1, setting the number of individuals in the population and the number of built-in evolutionary iterations , four parallel evolution evolutioners are set, which are genetic evolutioner containing genetic algorithm, particle swarm evolutioner containing particle swarm algorithm, annealing evolutioner containing annealing algorithm and tabu evolutioner containing tabu algorithm; the genetic evolutioner, the particle swarm evolutioner, the annealing evolutioner and the tabu evolutioner are used for parallel evolution, and four sub-populations are generated simultaneously;

[0037] Step 4.2, the economic net present value of the reservoir corresponding to the individual of the sub-population is predicted by using the agent model constructed in step 3; the specific prediction process is to calculate by using formula (3), to replace the time-consuming reservoir simulation process by using the mapping process of the radial basis function, so as to use the calculation result of the radial basis function as the predicted economic net present value of the reservoir, and four candidate solutions are selected from the four sub-populations, which are the optimal sub-individuals of the genetic evolutioner ,the optimal sub-individual of the particle swarm evolutioner ,the optimal sub-individual of the annealing evolutioner , and the optimal sub-individual of the tabu evolutioner .

[0038] Further, in step 4.1, the process of generating the sub-population by the genetic algorithm includes parent selection, gene crossover and offspring selection; one gene of the individual in the genetic algorithm corresponds to the production data of a single well at a single time step in the injection-production scheme ;

[0039] The parent selection process is as follows:

[0040] (6);

[0041] Wherein, , are the first parent and the second parent respectively; and are the randomly selected integers in the population; the individual is a one-dimensional array, and each value in the numerical value corresponds to a gene, and the definition is , , are two different individuals in the population; is the first gene of the first individual in the population; is the first gene of the second individual in the population; , are the first gene of the first individual and the second individual in the population, respectively, and the total number of genes is ; corresponds to the dimension ;

[0042] ​​​When the crossover method is two-point crossover, the gene crossover process is as follows:

[0043] (7);

[0044] in, and First paternal parent Second father The first and second offspring individuals generated by crossover; and for The intersection of two randomly selected integers;

[0045] When the crossover method is heuristic crossover, the gene crossover process is as follows:

[0046] (8);

[0047] in, For offspring individuals; A decimal number randomly selected from [0,1].

[0048] When the crossover method is dispersed crossover, the gene crossover process is as follows:

[0049] (9);

[0050] in, For the individual's first One gene; For the first Boolean mask for each gene;

[0051] The offspring selection process is as follows:

[0052] (10);

[0053] In the formula, For the first The first generation of offspring population Individual; A decimal number randomly selected from [0,1]. Crossover rate; The father is the parent.

[0054] Furthermore, in step 4.1, the process of generating the offspring population using the particle swarm optimization algorithm includes velocity updates and position updates; in the particle swarm optimization algorithm, the position of one particle corresponds to a complete injection and sampling scheme. ;

[0055] The speed update process is as follows:

[0056] (11);

[0057] The location update process is as follows:

[0058] (12);

[0059] in, For the first The particle in the first The speed of the generation, the first The particle corresponds to the first in the offspring population. Individual; Inertial weights; For the first The particle in the first The speed of generation; The weight of the particle itself; As a group's social weight; and A random number between [0, 1]; For the first Before each particle The optimal position of the generation; For the first The particle in the first The position of the generation; For the entire population The global optimal position of the generation; For the first The particle in the first The position of the generation.

[0060] Furthermore, in step 4.1, the annealing algorithm's process of generating the offspring population includes generating neighborhood solutions and accepting new solutions. A cooling process is also required after generating the offspring population. In the annealing algorithm, one solution corresponds to a complete injection / collection scheme. ;

[0061] The process of generating the domain solution is as follows:

[0062] (13);

[0063] (14);

[0064] in, For the first The generation of the iteration The new solution is in the first The value of the i-th dimension, the i-th The dimension corresponds to the first dimension in the genetic algorithm. One gene; For the front Global optimal solution of the generation In the Values ​​for each dimension; For the first Disturbance in each dimension; For interval Uniform distribution on; Step size; Each dimension Composition of the first A new interpretation , No. The new solution corresponds to the first generation in the offspring population. Individual;

[0065] The process of accepting a new solution is as follows:

[0066] (15);

[0067] in, The probability of accepting a new solution; For the front Global optimal solution of the generation With the The first generation A new interpretation The difference in fitness evaluation function; For the first Temperature parameters for the next iteration;

[0068] If the new solution is accepted, then the global optimal solution is updated:

[0069] (16);

[0070] in, For the front The global optimal solution of the generation;

[0071] The cooling process is as follows:

[0072] (17);

[0073] In the formula, For the first Temperature parameters for the next iteration; The initial temperature at the start of the iteration; , and All are constant parameters; , and There are three cooling functions;

[0074] The tabu algorithm's process of generating offspring population includes generating neighborhood solutions and selecting the optimal solution. After generating the offspring population, the tabu table also needs to be updated. Each offspring individual in the tabu algorithm corresponds to a new solution in the annealing algorithm, i.e., a complete injection-collection scheme. ;

[0075] The generating field solution process is shown in formula (13), The first generation of the first offspring individual in the first dimension, The first dimension of the first offspring individual in the first generation, , The first offspring individual in the first generation of the offspring population ;

[0076] The optimal solution selection process is: select the offspring individual in the offspring population which does not exist in the tabu table and has the highest fitness value as the global optimal solution of the previous generation:

[0077] (18);

[0078] Wherein, is the offspring individual which does not exist in the tabu table; is the fitness evaluation function, and the radial basis function is used to evaluate the fitness of the offspring individual in the iteration process; The updating tabu table process is: check whether the tabu table is full, when the tabu table is full, eliminate the solution added first in the tabu table, and then add the new optimal solution

[0079] to the tabu table .

[0080] Further, the specific process of step 5 is:

[0081] Step 5.1, put the four candidate solutions , , and selected from the four evolutioners into the reservoir numerical simulator for numerical simulation, respectively, to obtain the economic net present values , , and corresponding to the reservoir numerical simulation;

[0082] ​​​Step 5.2, judging the performance of the evolutioner in this round of iteration based on the economic net present value of the four candidate solutions after reservoir numerical simulation, the evolutioner evolution performance being the economic net present value calculated by the reservoir numerical simulator, and the candidate solution with the highest economic net present value indicating that the evolutioner performs best in this round of iteration;

[0083] Step 5.3, constructing a super-heuristic guiding strategy according to the performance of the evolutioner, the super-heuristic guiding strategy including guiding of the evolutioner performing best and the three evolutioners performing poorly, the guiding of the evolutioner performing best being retaining the evolution parameters in this round of iteration, and the guiding of the three evolutioners performing poorly being updating the evolution parameters of the corresponding evolutioners by using the online parameter return method;

[0084] Step 5.4, constructing an online parameter return method, the online parameter return method including four cases of genetic evolutioner parameter updating, particle swarm evolutioner parameter updating, annealing evolutioner parameter updating and tabu evolutioner parameter updating.

[0085] Further, the specific process of step 5.4 is as follows:

[0086] The genetic evolutioner parameter updating process includes modifying the crossover rate or the crossover function number, defining as the crossover rate, as the minimum value of the crossover rate, as the step length of the crossover rate change, as the crossover function number, as the maximum value of the crossover rate;

[0087] When , the is reduced by , obtaining the updated crossover rate ;

[0088] When , the is increased by 1, obtaining the updated crossover function number ;

[0089] When , the is reset to 1 and the is reset to ;

[0090] The particle swarm evolutioner parameter updating process includes modifying the inertia weight, modifying the particle self weight and modifying the group social weight; defining as the inertia weight range; as the particle self weight; as the maximum value of the particle self weight; as the step length of the particle self weight change; is the group social weight; is the maximum value of the group social weight; is the step of the group social weight change; is the minimum value of the particle self weight; is the minimum value of the group social weight;

[0091] When , is increased by 1 to get the updated inertia weight range :

[0092] When , is increased by to get the updated particle self weight :

[0093] When , is increased by to get the updated group social weight :

[0094] When or or , is reset to 1, is reset to and is reset to : :

[0095] The annealing evolver parameter update process includes modifying the cooling function or modifying the initial temperature: define as the cooling function number; as the initial temperature at the beginning of iteration; as the maximum value of temperature; as the step of temperature change; as the minimum value of temperature;

[0096] When , is increased by 1 to get the updated cooling function number :

[0097] When , is increased by to get the updated temperature :

[0098] When or , is reset to 1 and is reset to : :​​​

[0099] The tabu-evolver parameter updating process includes modifying the tabu list length or modifying the field search step: define As the minimum value of the field search step; As the field search step; As the step of field search step change; As the minimum value of the tabu list length; As the tabu list length; As the step of tabu list length change; As the maximum value of the field search step; As the maximum value of the tabu list length;

[0100] When , the is reduced , obtaining the updated field search step :

[0101] When , the is reduced , obtaining the updated tabu list length :

[0102] When or , reset to or reset to .

[0103] The beneficial technical effects brought by the present application: the present application adopts four evolution algorithms with different properties, genetic algorithm, particle swarm optimization algorithm, annealing algorithm and tabu algorithm, as parallel evolution, parallel evolution produces a large number of new individuals and uses the proxy model to predict the economic net present value of the new individuals, and the parallel evolution can explore the solution space using different strategies at the same time, producing more new individuals, and different evolution algorithms share the same proxy model for evaluation of new individuals, which enhances the richness and prediction accuracy of the training sample of the proxy model, and thus improves the overall optimization efficiency; at the same time, a super-heuristic guided search strategy is designed as a top-level scheduling mechanism to dynamically adjust the search behavior of the four evolution algorithms, and an online parameter returning strategy is used to update the evolution parameters of the evolution algorithm that performs poorly in real time, and the cooperation of the two strategies forms a feedback and adaptive adjustment mechanism, providing diverse and reliable decision-making basis for the formulation of water injection and production programs, and showing high engineering application value and broad development prospects. BRIEF DESCRIPTION OF DRAWINGS

[0104] Figure 1 The flowchart of the parallel super-heuristic injection and production optimization method based on the proxy model of the present application.

[0105] Figure 2 This is a permeability field diagram of the reservoir model in the study block in this embodiment of the invention.

[0106] Figure 3 This is an optimization convergence graph comparing the method of this invention with two other methods, using net present value as the objective function.

[0107] Figure 4 This is a schematic diagram of the production situation of the method of the present invention and two other methods; wherein, (a), (b), and (c) are the curves of cumulative oil production, cumulative water injection, and cumulative water production as a function of production time, respectively.

[0108] Figure 5 This is a schematic diagram illustrating how each evolver obtains the optimal solution during the first 50 optimization processes of the method of the present invention. Detailed Implementation

[0109] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0110] like Figure 1 As shown, a parallel hyperheuristic injection-collection optimization method based on a surrogate model specifically includes the following steps:

[0111] Step 1: Obtain water-drive injection and production information from the oilfield, and construct a three-dimensional geological model and a reservoir injection and production model;

[0112] Oilfield water drive injection and production information includes oilfield geological data and oilfield production information. Oilfield geological data includes grid information of the oilfield, physical properties of rocks in the oilfield area, fluid properties of oil and water, etc. Oilfield production information includes well type and location, as well as the production volume of each production well and the injection volume of each injection well in each time step before optimization.

[0113] Based on oilfield geological data, a three-dimensional geological model was constructed using the oilfield geological modeling software Petrol. The three-dimensional geological model was then imported into the reservoir numerical simulator. Based on oilfield production information, the reservoir injection and production model was simulated using the numerical simulation software Eclipse and stored in the reservoir numerical simulator.

[0114] Step 2: Construct an injection-production optimization model for water-drive reservoirs. Multiple injection-production schemes are obtained using Latin hypercube sampling as an initial sample set. The economic net present value corresponding to each sample in the initial sample set is calculated using a reservoir numerical simulator. An initial database is then constructed to provide basic data support for the subsequent construction of proxy models. The specific process is as follows:

[0115] Step 2.1, for the injection-production optimization problem of water drive reservoir, a model of injection-production optimization of water drive reservoir is constructed with the maximum economic net present value of reservoir as objective function, development control parameters of production well and injection well as decision variables, and variable boundary constraint conditions.

[0116] The injection-production optimization model of water drive reservoir is:

[0117] (1);

[0118] wherein, is the maximum value of objective function with as independent variable; is the objective function; represents injection-production scheme, i.e. production system of each time step, including input of injection well and liquid production of production well; is economic net present value; is dimension; is dimensional real number; and respectively represent lower limit and upper limit of ; the calculation formula of economic net present value is as follows:

[0119] (2);

[0120] wherein, is time step number; is oil production rate of block in the th time step; represents oil production income; is water production rate of block in the th time step; represents water cost except water; is injection rate of block in the th time step; represents injection cost; is annual discount rate; is time step length;

[0121] Step 2.2, Latin hypercube sampling is adopted to obtain multiple samples within lower limit and upper limit of decision variable, and an initial sample set is constructed, wherein, is injection-production scheme of the th sample, is initial sample amount;

[0122] Step 2.3, the initial sample set is imported into reservoir numerical simulator, and reservoir economic net present value of each sample is calculated by using reservoir numerical simulator, so that reservoir economic net present value of each sample is obtained corresponding economic net present value data set of the oil reservoirs , wherein, is the economic net present value of the i-th sample of the oil reservoirs;

[0123] Step 2.4, based on the initial sample set and its corresponding economic net present value data set of the oil reservoirs , an initial database is constructed , .

[0124] Step 3, based on the economic net present value of the initial samples, the initial samples are screened, and a surrogate model based on the economic net present value as the optimization objective is constructed. The specific process is as follows:

[0125] Step 3.1, set the maximum number of iterations of reservoir numerical simulation optimization , the samples in the initial database are sorted according to the economic net present value from high to low, and the top samples with the highest economic net present value are selected as the training set of the training surrogate model;

[0126] Step 3.2, based on the samples in the training set, a radial basis function surrogate model is constructed; the radial basis function surrogate model is constructed based on a Gaussian kernel function, and the formula is as follows:

[0127] (3);

[0128] wherein, is the radial basis function; is the serial number of the sample in the training set of the surrogate model; is the number of samples in the training set of the surrogate model; is the weight of the i-th sample in the training set; is the injection-production scheme of the i-th sample in the training set of the surrogate model; is the Euclidean distance between samples; is the kernel function, and the kernel function is a Gaussian function, and the expression is as follows: (4);

[0129] wherein, is the exponential function;

[0130] is the shape parameter of the Gaussian function, and the expression is as follows: (5);

[0131] wherein,

[0132] ​​​Maximize the distance between all sample pairs in the training set of the surrogate model.

[0133] Step 4, use genetic algorithm, particle swarm optimization algorithm, annealing algorithm and tabu search algorithm to generate offspring population in parallel evolution, evaluate the offspring individuals using the surrogate model, and obtain the candidate solutions of each evolution algorithm. The specific process is as follows:

[0134] Step 4.1, set the number of individuals in the population and the number of built-in evolution iterations , set four parallel evolution evolutioners, which are genetic evolutioner containing genetic algorithm, particle swarm evolutioner containing particle swarm optimization algorithm, annealing evolutioner containing annealing algorithm and tabu evolutioner containing tabu search algorithm; use genetic evolutioner, particle swarm evolutioner, annealing evolutioner and tabu evolutioner to evolve in parallel, and generate four offspring populations simultaneously; the injection-production scheme to be optimized in the present application corresponds to the concept of "individual" in genetic algorithm, the concept of "particle position" in particle swarm optimization algorithm, the concept of "solution" in annealing algorithm, and the concept of "offspring individual" in tabu search algorithm.

[0135] The process of generating offspring population by genetic algorithm includes parent selection, gene crossover and offspring selection; one gene of an individual in genetic algorithm corresponds to the production data of a single well at a single time step in injection-production scheme , and one individual corresponds to a complete injection-production scheme ;

[0136] The parent selection process is as follows:

[0137] (6)

[0138] wherein, , are the first parent and the second parent, respectively; and are randomly selected integers; an individual is a one-dimensional array, and each value in the numerical value corresponds to a gene, and the definition is , are two different individuals in the population; is the first gene of the th individual in the population, i.e. the first gene in the first parent ; is the first gene of the th individual in the population, i.e. the first gene in the second parent ; , are the first gene of the th individual and the first gene of the th individual in the population, respectively; total number of genes corresponding dimension

[0139] The gene crossover process is to generate offspring by utilizing the existing parent gene crossover; wherein, when the crossover mode is two-point crossover, the expression of the gene crossover is:

[0140] (7)

[0141] wherein, and are the first parent and the second parent crossover to generate the first and second offspring individuals; and are two randomly selected integer crossover points in

[0142] When the crossover mode is heuristic crossover, the expression of the gene crossover is:

[0143] (8)

[0144] wherein, is the offspring individual; is a randomly selected decimal number in [0, 1];

[0145] When the crossover mode is dispersion crossover, the expression of the gene crossover is:

[0146] (9)

[0147] wherein, is the serial number of the gene in the individual; is the gene of the individual; is a Boolean mask, which is a random vector composed of equal probability 0 and 1 with a length of , and the length corresponds to the dimension ; is the Boolean mask of the gene;

[0148] The offspring selection process is:

[0149] (10)

[0150] wherein, is the serial number of the individual in the population; is the offspring population of the ​An individual contains a number of genes; A decimal number randomly selected from [0,1]. Crossover rate; As the father;

[0151] The particle swarm optimization (PSO) algorithm generates offspring populations by including velocity updates and position updates; in the PSO algorithm, the position of a particle corresponds to a complete injection / collection scheme. ;

[0152] The speed update process is as follows:

[0153] (11);

[0154] The location update process is as follows:

[0155] (12);

[0156] in, The index of the particle in the population; This refers to the number of built-in evolution iterations. For the first The particle in the first The speed of the generation, the first The particle corresponds to the first in the offspring population. Individual; Inertial weights; For the first The particle in the first The speed of generation; The weight of the particle itself; As a group's social weight; and A random number between [0, 1]; For the first Before each particle The optimal position of the generation; For the first The particle in the first The position of the generation; For the entire population The global optimal position of the generation; For the first The particle in the first The position of the generation;

[0157] The annealing algorithm's process of generating offspring population includes generating neighborhood solutions and accepting new solutions. A cooling process is also required after generating the offspring population. In the annealing algorithm, one solution corresponds to a complete injection / collection scheme. ;

[0158] The process of generating the domain solution is as follows:

[0159] (13);

[0160] (14);

[0161] in, For the first The generation of the iteration The new solution is in the first The value of the i-th dimension, the i-th The dimension corresponds to the first dimension in the genetic algorithm. One gene; For the front Global optimal solution of the generation In the Values ​​for each dimension; For the first Disturbance in each dimension; For interval Uniform distribution on; Step size; Each dimension Composition of the first A new interpretation , No. The new solution corresponds to the first generation in the offspring population. Individual;

[0162] The process of accepting the new solution is as follows:

[0163] (15);

[0164] in, The probability of accepting a new solution; For the front Global optimal solution of the generation With the The first generation A new interpretation The difference in fitness evaluation function; For the front Global optimal solution of the generation The fitness evaluation function value; For the first The first generation A new interpretation The fitness evaluation function value; For the first Temperature parameters for the next iteration;

[0165] If the new solution is accepted, the global optimal solution is updated. The process of updating the global optimal solution is as follows:

[0166] (16);

[0167] in, For the front The global optimal solution of the generation;

[0168] The cooling process is as follows:

[0169] (17);

[0170] In the formula, For the first Temperature parameters for the next iteration; The initial temperature at the start of the iteration; , and All are constant parameters; , and There are three cooling functions;

[0171] The tabu algorithm's process of generating offspring population includes generating neighborhood solutions and selecting the optimal solution. After generating the offspring population, the tabu list also needs to be updated. Each offspring individual in the tabu algorithm corresponds to a new solution in the annealing algorithm, i.e., a complete injection-collection scheme. ;

[0172] The process of generating the domain solution is the same as that in formula (13). For the first The generation of the first generation The offspring individuals in the first Values ​​of each dimension Each dimension Composition of the first The first generation individual offspring , individual offspring Composition of the first Offspring population ;

[0173] The process of selecting the optimal solution is as follows: selecting the offspring population. It is not present in the taboo list. Among them, the offspring with the highest fitness value is selected as the first generation. The global optimal solution of the generation:

[0174] (18);

[0175] in, For the front The global optimal solution of the generation; for Offspring individuals not listed in the taboo list; In order to adapt to the fitness evaluation function, the offspring individuals are evaluated in the iteration process by using the radial basis function;

[0176] The updating tabu list process is: checking whether the tabu list is full, when the tabu list is full, eliminating the solution added first in the tabu list, and then adding the new optimal solution Into the tabu list .

[0177] Step 4.2, using the proxy model constructed in step 3 to predict the economic net present value of the reservoir corresponding to the individual of the offspring population, the prediction process is to put into formula (3) for calculation, using the mapping process of the radial basis function to replace the time-consuming reservoir simulation process, so as to use the calculation result of the radial basis function as the predicted economic net present value of the reservoir, and four candidate solutions are selected from the four offspring populations respectively, which are the optimal sub-individuals of the genetic evolution device , the optimal sub-individuals of the particle swarm evolution device , the optimal sub-individuals of the annealing evolution device And the optimal sub-individuals of the tabu evolution device .

[0178] Step 5, carrying out reservoir numerical simulation on the candidate solutions, constructing a super-heuristic guided search strategy and an online parameter returning method, and adjusting the search behavior of the evolution device based on the economic net present value of the candidate solutions. The specific process is:

[0179] Step 5.1, putting the four candidate solutions selected from the four evolution devices , , and Into the reservoir numerical simulator for numerical simulation, respectively obtaining the economic net present value corresponding to the reservoir numerical simulation , , and ;

[0180] Step 5.2, judging the performance of the evolution device in this round based on the economic net present value of the four candidate solutions after the reservoir numerical simulation, the performance of the evolution device is the high and low of the economic net present value calculated by the reservoir numerical simulator, and the candidate solution with the highest economic net present value represents the best performance of the evolution device in this round of iteration;

[0181] The process of judging the performance of the evolution device in this round is:

[0182] When The highest, the candidate solution generated by the genetic algorithm in this round Is the best, the genetic evolution device performs best in this round, and the particle swarm, annealing and tabu evolution devices perform poorly in this round;

[0183] When Best, the candidate solution generated by the genetic algorithm in this round Best, the candidate solution generated by the genetic algorithm in this round

[0184] When Best, the candidate solution generated by the genetic algorithm in this round Best, the candidate solution generated by the genetic algorithm in this round

[0185] When Best, the candidate solution generated by the genetic algorithm in this round Best, the candidate solution generated by the genetic algorithm in this round

[0186] Step 5.3, constructing a super-heuristic guiding strategy according to the performance of the evolution algorithm, the super-heuristic guiding strategy including guiding the best one evolution algorithm and guiding the three poor evolution algorithms, the guiding of the best one evolution algorithm is keeping the evolution parameters in this round iteration, and the guiding of the three poor evolution algorithms is adjusting the evolution parameters of the evolution algorithm;

[0187] The super-heuristic guiding strategy is:

[0188] When the genetic evolution algorithm performs best, the genetic evolution algorithm parameters remain unchanged, and the particle swarm, annealing, and taboo evolution algorithms use the online parameter return method to update the parameters of their evolution algorithms;

[0189] When the particle swarm evolution algorithm performs best, the particle swarm evolution algorithm parameters remain unchanged, and the genetic, annealing, and taboo evolution algorithms use the online parameter return method to update the parameters of their evolution algorithms;

[0190] When the annealing evolution algorithm performs best, the annealing evolution algorithm parameters remain unchanged, and the genetic, particle swarm, and taboo evolution algorithms use the online parameter return method to update the parameters of their evolution algorithms;

[0191] When the taboo evolution algorithm performs best, the taboo evolution algorithm parameters remain unchanged, and the genetic, particle swarm, and annealing evolution algorithms use the online parameter return method to update the parameters of their evolution algorithms;

[0192] Step 5.4, constructing an online parameter return method, the online parameter return method including four cases of genetic evolution algorithm parameter update, particle swarm evolution algorithm parameter update, annealing evolution algorithm parameter update, and taboo evolution algorithm parameter update;

[0193] The genetic evolution algorithm parameter update process includes modifying the crossover rate or the crossover function number, defining is the crossover rate, is the minimum value of the crossover rate, a step size for the crossover rate, a crossover function number, a maximum value for the crossover rate:

[0194] when , the crossover rate is reduced by , obtaining an updated crossover rate :

[0195] (19);

[0196] when , the crossover function number is increased by 1, obtaining an updated crossover function number :

[0197] (20);

[0198] when , the crossover function number and the crossover rate are reset:

[0199] (21);

[0200] in addition, when the crossover mode is two-point crossover, when the crossover mode is heuristic crossover, when the crossover mode is scattered crossover;

[0201] the particle swarm evolutioner parameter updating process comprises modifying the inertia weight, modifying the particle self weight, modifying the group social weight; defining as the inertia weight range, when the range is [0.8, 1.2], when the range is [0.9, 1.3], when the range is [1, 1.4]; defining as the particle self weight; as the maximum value of the particle self weight; as the step size for the particle self weight change; as the group social weight; as the maximum value of the group social weight; as the step size for the group social weight change; as the minimum value of the particle self weight; as the minimum value of the group social weight;

[0202] when , the inertia weight range is increased, obtaining an updated inertia weight range :

[0203] (22);

[0204] When , the particle self weight is increased by , obtaining the updated particle self weight :

[0205] (23);

[0206] When , the group social weight is increased by , obtaining the updated group social weight :

[0207] (24);

[0208] When or or , the inertia weight range, the particle self weight and the group social weight are reset:

[0209] (25);

[0210] The annealing evolutioner parameter updating process comprises modifying a cooling function or modifying an initial temperature: defining as a cooling function number, is a cooling function in formula (17), is a cooling function in formula (17), is a cooling function in formula (17); defining as an initial temperature at the beginning of iteration; as a maximum temperature; as a step length of temperature change; as a minimum temperature;

[0211] When , the cooling function number is increased by 1, obtaining the updated cooling function number :

[0212] (26);

[0213] When , the initial temperature is increased by , obtaining the updated temperature :

[0214] (27);

[0215] When or Reset the cooling function and initial temperature when

[0216] (28);

[0217] The tabu-evolver parameter updating process includes modifying the tabu list length or modifying the field search step: define as the minimum value of the field search step; as the field search step; as the step of field search step change; as the minimum value of the tabu list length; as the tabu list length; as the step of tabu list length change; as the maximum value of the field search step; as the maximum value of the tabu list length;

[0218] When , the field search step is reduced by , and the updated field search step is obtained:

[0219] (29);

[0220] When , the tabu list length is reduced by , and the updated tabu list length is obtained:

[0221] (30);

[0222] When or , reset the tabu list length or the field search step:

[0223] (31);

[0224] Step 6, repeat steps 3 to 5 until the preset maximum optimization iteration number is reached, and record the sample evaluated using the reservoir numerical simulation and the economic net present value corresponding to the sample in each iteration optimization to obtain an updated database.

[0225] Step 7, after the iteration optimization is completed, the reservoir numerical simulator outputs the optimal Pareto front scheme and the corresponding economic net present value, and the injection-production optimization of the water drive reservoir is completed.

[0226] To verify the effect of the method of the present application applied to injection-production optimization of water drive reservoirs, the parallel hyper-heuristic injection-production optimization method based on the agent model as described above is applied to a certain research block to perform injection-production optimization of the water drive reservoir in the research block. In this embodiment, the reservoir model is a two-dimensional reservoir model, the grid system of the reservoir model is 25 25 1, the grid size is 2500ft 2500ft 20ft, the permeability field of the reservoir model is as shown in the figure; the well location coordinates of the reservoir to be optimized include 9 production wells and 4 injection wells, wherein the production well numbers are P1-P9, the injection well numbers are I1-I4, the well spacing constraint is based on the grid distance, and the spacing between adjacent well locations is not less than 10m. Figure 2

[0227] The method as described above is used to perform water drive injection-production optimization of the reservoir, the maximum number of optimization iterations is set to 370, and the specific process is as follows:

[0228] According to the process of step 1, a three-dimensional geological model of a certain research block is constructed, then the three-dimensional geological model is imported into a reservoir numerical simulator, and then an injection-production model of the reservoir is simulated according to the oilfield production information using the numerical simulation software Eclipse, and stored in the reservoir numerical simulator.

[0229] According to the process of step 2, the injection-production optimization model of the water drive reservoir is constructed, the lower limit of the injection rate of the injection well is set to 0 STB·d -1 , the upper limit is 500 STB·d -1 , the lower limit of the liquid production of the production well is 0 STB·d -1 , and the upper limit is 200 STB·d -1 ; 130 initial samples are obtained by Latin hypercube sampling, the production time of the reservoir is set to 1800 days, the time step is 5, the time step is 360 days, the annual discount rate is 0, the oil production income is 80 USD / STB, the water cost is 5 USD / STB, and the injection cost is 5 USD / STB. The economic net present value of the reservoir is calculated by using the reservoir simulation software Eclipse, the initial injection-production scheme and the calculated economic net present value are constructed to obtain an initial database and stored in the database, which provides basic data support for subsequent agent model construction.

[0230] According to the process of step 3, the samples in the initial database are sorted in descending order of the economic net present value of the reservoir, and the number of samples in the training set of the agent model is set to 100.​​ The value is 130. The top 130 samples with the highest economic net present value are selected as the training set for training the agent model, and an agent model based on economic net present value as the optimization objective is constructed.

[0231] Following the process in step 4, a progeny population is generated using a surrogate-assisted parallel co-evolution method. The progeny individuals are then evaluated using a surrogate model to obtain candidate solutions for each evolutionary algorithm.

[0232] In this embodiment, the number of individuals in the population is first set. The number is 50, with a built-in number of evolution iterations. Set the crossover function number to 50 in the initial parameters of the genetic algorithm. The crossover rate is 1. The inertia weight range in the initial parameters of the particle swarm optimization algorithm is 0.8. The particle's own weight is 1. The social weight of the group is 1.49. The value is 1.49, which is the number of the cooling function in the initial parameters of the annealing algorithm. The initial temperature at the start of the iteration is 1. The value is 100, representing the length of the tabu list in the initial parameters of the tabu algorithm. The domain search step size is 20. The value is set to 1; then, a progeny population is generated through parallel evolution using genetic algorithms, particle swarm optimization, annealing, and tabu algorithms. A surrogate model is used to predict the reservoir economic net present value corresponding to each individual in the progeny population. The optimal progeny of the genetic evolutionary unit is selected from each of the four progeny populations. The optimal sub-individual of the particle swarm evolutioner The optimal sub-individual of the annealing evolver The optimal sub-individual of the forbidden evolutionary device .

[0233] Following step 5, numerical simulations are performed on the candidate solutions to construct a hyperheuristic guided search strategy and an online parameter return method. The search behavior of the evolver is adjusted based on the economic net present value of the candidate solutions. In this embodiment, the four candidate solutions selected from the four evolvers are first... , , and Numerical simulations of the reservoir were performed, and the corresponding net present value of the reservoir was obtained. , , and , the economic net present value of reservoir simulation based on four candidate solutions is used to judge the performance of the current evolutioner, then, according to the performance of the evolutioner, the super heuristic guided search strategy is used to guide the evolutioner with the best performance and the evolutioner with poor performance respectively, the online parameter return method is used to adjust the evolution parameters of the three evolutioners with poor performance, and the search behavior of the evolutioner is adjusted by updating the parameters of the evolutioner; in the online parameter return method, the minimum value of the crossover rate in the genetic algorithm is 0.6, the maximum value of the crossover rate is 0.8, the step length of the change of the crossover rate is 0.05; the minimum value of the particle weight of the particle swarm algorithm is 1.49, the maximum value of the particle weight is 1.99, and the step length of the change of the particle weight is 0.1; the minimum value of the group social weight is 1.49, the maximum value of the group social weight is 1.99, and the step length of the change of the group social weight is 0.1; the minimum value of the temperature in the annealing algorithm is 100, the maximum value of the temperature is 500, and the step length of the change of the temperature is 100; the minimum value of the field search step length in the tabu algorithm is 0.6, the maximum value of the field search step length is 1, and the step length of the change of the field search step length is 0.05; the minimum value of the length of the tabu table is 10, the maximum value of the length of the tabu table is 20, and the step length of the change of the length of the tabu table is 2.

[0234]

[0235]

[0236] ​​​​​​​​​​​​​​​​​​​​​​​The SAPCH method is defined as a parallel hyper-heuristic injection-production optimization method based on an agent model proposed in the application.

[0237] The other two comparison methods are the SAPGA method and the SAPCH-NPR method; the SAPGA method is a single-species evolutionary agent-assisted evolutionary algorithm using four genetic populations for parallel evolution, and the SAPCH-NPR method is a multi-evolution parallel injection-production optimization method based on an agent model.

[0238] The above three methods are set with the same parameters, the initial sampling sample number is set to 130, the initial population individual number is set to 50, the built-in evolution iteration number is set to 50, the maximum optimization iteration number is 370, the evolution parameters in the evolutioner of the SAPGA method and the SAPCH-NPR method remain unchanged during the iteration process, and the sample number for constructing the agent model is set to 130.

[0239] Figure 3 The convergence curves of the economic net present value of the method of the application and the other two methods with the optimization iteration number are compared, from which Figure 3 It can be seen that under the condition of the same optimization iteration number, the economic net present value obtained by the Pareto frontier generated by the SAPCH method of the application in the target space is optimal, which shows that the SAPCH method proposed in the application can effectively improve the economic benefit.

[0240] Figure 4 The production situation diagram of the method of the application and the other two methods is shown in the figure, from which it can be found that the SAPCH method of the application uses different evolution methods to explore the solution space using different strategies, and at the same time uses a hyper-heuristic guiding strategy and an online parameter returning strategy to adjust the evolutioner, modify its poor search behavior, and can obtain the highest cumulative oil production, although the cumulative water injection is also the highest, and the cumulative water production is at a medium level, but since the treatment cost of water injection and water production is low, higher economic benefit is finally obtained.

[0241] Figure 5The figure shows the optimal solutions obtained by each optimizer in the first 50 iterations of the SAPCH method for the case of maximizing the economic net present value (NPV). The results show that the curves of genetic algorithm (GA) and particle swarm optimization (PSO) exhibit significant fluctuations, indicating their exploratory nature in the global search process. In contrast, the curves of simulated annealing (SA) and tabu search (TS) are relatively stable, indicating their local optimization capabilities. GA and PSO belong to population-based optimizers, while SA and TS belong to trajectory-based optimizers. These two types of optimizers work together in the SAPCH framework: GA and PSO identify potential high-quality solutions, while SA and TS further refine these solutions, thereby improving the overall optimization performance. Additionally, the hyper-heuristic control and parameter return methods of SAPCH allow all optimizers to interact through competition and learning, constantly adjusting their search strategies to adapt to the changing conditions of the solution space. For example, the individuals optimized by the genetic algorithm initially have relatively low NPV, but SAPCH adjusts the crossover method and crossover rate of the genetic algorithm, modifying its search behavior to enable it to rediscover potential high-quality solutions. In the 11th iteration, the genetic algorithm surpasses the other three optimization algorithms, achieving the optimal economic net present value. Ultimately, SAPCH adjusts the search strategies of each optimizer in real-time based on their performance, ensuring that all optimizers achieve optimal performance during the search process. Therefore, SAPCH successfully integrates the strengths of the four optimizers, significantly improving the quality of the solutions.

[0242] Of course, the above description is not a limitation of the present application, and the present application is not limited to the above examples. Changes, modifications, additions or substitutions made by those skilled in the art within the spirit and scope of the present application should also be within the scope of protection of the present application.

Claims

1. A parallel hyperheuristic injection-collection optimization method based on a surrogate model, characterized in that, Includes the following steps: Step 1: Obtain water-drive injection and production information from the oilfield, and construct a three-dimensional geological model and a reservoir injection and production model; Step 2: Construct an injection-production optimization model for water-drive reservoirs. Use Latin hypercube sampling to obtain multiple injection-production schemes as an initial sample set. Calculate the economic net present value corresponding to each sample in the initial sample set using a reservoir numerical simulator to construct an initial database. Step 3: Select initial samples based on the economic net present value of the initial samples, and construct a proxy model based on the economic net present value as the optimization objective; Step 4: Generate offspring populations through parallel evolution using genetic algorithm, particle swarm optimization, annealing algorithm, and tabu algorithm. Use a surrogate model to evaluate offspring individuals and obtain candidate solutions for each evolutionary algorithm. Step 5: Perform reservoir numerical simulation on candidate solutions, construct a hyperheuristic guided search strategy and an online parameter return method, and adjust the search behavior of the evolver based on the economic net present value of the candidate solutions; The specific process is as follows: Step 5.1: Select four candidate solutions from the four evolvers. , , and The data were placed into a reservoir numerical simulator for numerical simulation, and the corresponding net present value of the reservoir numerical simulation was obtained. , , and ; Step 5.2: Determine the performance of the current round of the evolver based on the economic net present value of the four candidate solutions obtained through reservoir numerical simulation. The performance of the evolver is determined by the level of the economic net present value of its candidate solutions calculated by the reservoir numerical simulator. The candidate solution with the highest economic net present value indicates that the evolver performs best in this round of iteration. Step 5.3: Construct a superheuristic guidance strategy based on the performance of the evolvers. The superheuristic guidance strategy includes guidance for the best-performing evolver and the three poorly performing evolvers. The guidance for the best-performing evolver is to retain the evolution parameters in the current iteration. The guidance for the three poorly performing evolvers is to update the evolution parameters of the corresponding evolvers using an online parameter return method. Step 5.4: Construct an online parameter return method, which includes four cases: genetic evolutionary parameter update, particle swarm evolutionary parameter update, annealing evolutionary parameter update, and taboo evolutionary parameter update. Step 6: Repeat steps 3 to 5 until the preset maximum number of optimization iterations is reached. For each iteration, record the sample evaluated using reservoir numerical simulation and the corresponding reservoir economic net present value to obtain the updated database. Step 7: After the iterative optimization is completed, the reservoir numerical simulator outputs the optimal Pareto front scheme and the corresponding economic net present value, thus completing the water-drive reservoir injection-production optimization.

2. The parallel hyperheuristic injection-collection optimization method based on the surrogate model according to claim 1, characterized in that, In step 1, the oilfield water drive injection and production information includes oilfield geological data and oilfield production information; the oilfield geological data includes the grid information of the oilfield, the physical properties of the rocks in the oilfield area, and the fluid properties of oil and water; the oilfield production information includes the type and location of wells, as well as the production volume of each production well and the injection volume of each injection well in each time step before optimization. Based on oilfield geological data, a three-dimensional geological model was constructed using the oilfield geological modeling software Petrol. The three-dimensional geological model was then imported into the reservoir numerical simulator. Based on oilfield production information, the reservoir injection and production model was simulated using the numerical simulation software Eclipse and stored in the reservoir numerical simulator.

3. The parallel hyperheuristic injection-sampling optimization method based on the surrogate model according to claim 1, characterized in that, The specific process of step 2 is as follows: Step 2.1: Construct the water-drive reservoir injection-production optimization model as follows: （1）； in, For Find the maximum value of the objective function for the independent variable; The objective function is... Indicates the injection and extraction scheme; It is the economic net present value; For dimensions; for A real number of dimension; and They represent The lower and upper limits; the formula for calculating the economic net present value is as follows: （2）； in, For time steps; For the block in the Oil production rate at each time step; Indicates oil extraction revenue; For the block in the 1st Water production rate at each time step; Indicates the cost excluding water; For the block in the 1st The water injection rate at each time step; Indicates the cost of water injection; The annual discount rate; For time step; Step 2.2: Use Latin hypercube sampling to obtain multiple samples within the upper and lower limits of the decision variables to construct an initial sample set. ,in, For the first The injection and collection scheme for each sample. This is the initial sample size; Step 2.3: Initial sample set Import the data into the reservoir numerical simulator, and use the simulator to calculate the reservoir economic net present value for each sample, obtaining the result compared to the initial sample set. Corresponding reservoir economic net present value dataset , ,in, For the first The economic net present value of an oil reservoir for a sample; Step 2.4: Based on the initial sample set and its corresponding reservoir economic net present value dataset Build the initial database .

4. The parallel hyperheuristic injection-sampling optimization method based on the surrogate model according to claim 3, characterized in that, The specific process of step 3 is as follows: Step 3.1: Set the maximum number of optimization iterations for reservoir numerical simulation. The samples in the initial database are sorted from highest to lowest based on the reservoir's net economic present value (NPV), and the samples with the highest NPV are selected. Each sample is used as the training set for training the agent model; Step 3.2: Construct a radial basis function surrogate model based on samples in the training set; the radial basis function surrogate model is constructed based on the Gaussian kernel function, and the formula is: （3）； in, These are radial basis functions; The index of the sample in the training set of the surrogate model; The number of samples in the training set for the surrogate model; For the training set The weights of each sample; For the surrogate model training set, the first Injection and collection scheme for each sample; The Euclidean distance between the samples; Here is the kernel function, which is a Gaussian function, and its expression is: （4）； in, It is an exponential function; Let be the shape parameter of the Gaussian function, expressed as: （5）； in, This represents the maximum distance between all sample pairs in the training set of the surrogate model.

5. The parallel hyperheuristic injection-sampling optimization method based on the surrogate model according to claim 4, characterized in that, The specific process of step 4 is as follows: Step 4.1: Set the number of individuals in the population. and built-in evolution iteration number Four parallel evolutionary generators are set up: a genetic evolutionary generator containing a genetic algorithm, a particle swarm evolutionary generator containing a particle swarm algorithm, an annealing evolutionary generator containing an annealing algorithm, and a tabu evolutionary generator containing a tabu algorithm. The genetic evolutionary generator, particle swarm evolutionary generator, annealing evolutionary generator, and tabu evolutionary generator are used to evolve in parallel, generating four offspring populations at the same time. Step 4.2: Use the surrogate model constructed in Step 3 to predict the reservoir economic net present value corresponding to the offspring population individuals; the specific prediction process is to use formula (3) for calculation, replace the time-consuming reservoir simulation process with the mapping process of the radial basis function, and use the calculation result of the radial basis function as the predicted reservoir economic net present value. Four candidate solutions are selected from the four offspring populations, which are the optimal offspring individuals of the genetic evolutionary unit. The optimal sub-individual of the particle swarm evolutioner The optimal sub-individual of the annealing evolver The optimal sub-individual of the forbidden evolutionary device .

6. The parallel hyperheuristic injection-sampling optimization method based on the surrogate model according to claim 5, characterized in that, In step 4.1, the process of generating the offspring population using the genetic algorithm includes parent selection, gene crossover, and offspring selection; one gene in an individual in the genetic algorithm corresponds to an injection / collection scheme. Production data for a single well at a single time step; each individual well corresponds to a complete injection-production plan. ; The parent selection process is as follows: （6）； in, , They are the first parent and the second parent, respectively. and for The integers are randomly selected from the data; each individual is a one-dimensional array where each value corresponds to a gene, defined as follows: , These are two distinct individuals within the population; The first in the population The first gene of an individual; The first in the population The first gene of an individual; , They are respectively the first in the population Individual, First The individual's first Genes, total number of genes Corresponding dimensions ; When the crossover method is two-point crossover, the gene crossover process is as follows: （7）； in, and First paternal parent Second father The first and second offspring individuals generated by crossover; and for The intersection of two randomly selected integers; When the crossover method is heuristic crossover, the gene crossover process is as follows: （8）； in, For offspring individuals; A decimal number randomly selected from [0,1]. When the crossover method is dispersed crossover, the gene crossover process is as follows: （9）； in, For the individual's first One gene; For the first Boolean mask for each gene; The offspring selection process is as follows: （10）； In the formula, For the first The first generation of offspring population Individual; A decimal number randomly selected from [0,1]. Crossover rate; The father is the parent.

7. The parallel hyperheuristic injection-sampling optimization method based on the surrogate model according to claim 6, characterized in that, In step 4.1, the process of generating the offspring population using the particle swarm optimization algorithm includes velocity updates and position updates; in the particle swarm optimization algorithm, the position of one particle corresponds to a complete injection and collection scheme. ; The speed update process is as follows: （11）； The location update process is as follows: （12）； in, For the first The particle in the first The speed of the generation, the first The particle corresponds to the first in the offspring population. Individual; Inertial weights; For the first The particle in the first The speed of generation; The weight of the particle itself; As a group's social weight; and A random number between [0, 1]; For the first Before each particle The optimal position of the generation; For the first The particle in the first The position of the generation; For the entire population The global optimal position of the generation; For the first The particle in the first The position of the generation.

8. The parallel hyperheuristic injection-collection optimization method based on the surrogate model according to claim 7, characterized in that, In step 4.1, the annealing algorithm's process of generating offspring population includes generating neighborhood solutions and accepting new solutions. A cooling process is also required after generating the offspring population. In the annealing algorithm, one solution corresponds to a complete injection / collection scheme. ; The process of generating the domain solution is as follows: （13）； （14）； in, For the first The generation of the iteration The new solution is in the first The value of the i-th dimension, the i-th The dimension corresponds to the first dimension in the genetic algorithm. One gene; For the front Global optimal solution of the generation In the Values ​​in each dimension; For the first Disturbance in each dimension; For interval Uniform distribution on; Step size; Each dimension Composition of the first A new interpretation , No. The new solution corresponds to the first generation in the offspring population. Individual; The process of accepting a new solution is as follows: （15）； in, The probability of accepting a new solution; For the front Global optimal solution of the generation With the The first generation A new interpretation The difference in fitness evaluation function; For the first Temperature parameters for the next iteration; If the new solution is accepted, then the global optimal solution is updated: （16）； in, For the front The global optimal solution of the generation; The cooling process is as follows: （17）； In the formula, For the first Temperature parameters for the next iteration; The initial temperature at the start of the iteration; , and All are constant parameters; , and There are three cooling functions; The tabu algorithm's process of generating offspring population includes generating neighborhood solutions and selecting the optimal solution. After generating the offspring population, the tabu table also needs to be updated. Each offspring individual in the tabu algorithm corresponds to a new solution in the annealing algorithm, i.e., a complete injection-collection scheme. ; The process of generating the domain solution is the same as that in formula (13). For the first The generation of the first generation The offspring individuals in the first Values ​​of each dimension Each dimension Composition of the first The first generation individual offspring , individual offspring Composition of the first Offspring population ; The process of selecting the optimal solution is as follows: Select the offspring population. It is not present in the taboo list. Among them, the offspring with the highest fitness value is selected as the first generation. Global optimal solution of the generation : （18）； in, for Offspring individuals not listed in the taboo list; The fitness evaluation function uses radial basis functions to evaluate the fitness of offspring individuals during the iteration process; The process of updating the tabu list is as follows: check if the tabu list is full. If the tabu list is full, discard the solution that was added to the tabu list first, and then add the new optimal solution. Add to the taboo list middle.

9. The parallel hyperheuristic injection-collection optimization method based on the surrogate model according to claim 8, characterized in that, The specific process of step 5.4 is as follows: The genetic evolutionary parameter update process includes modifying the crossover rate or crossover function number, and defining... Crossover rate, This represents the minimum crossover rate. The step size for changing the crossover rate. Number the cross function. This represents the maximum crossover rate. when At that time, reduce To obtain the updated crossover rate ; when At that time, Increment by 1 to obtain the updated cross function number. ; when Reset Set to 1 and reset for ; The particle swarm evolution parameter update process includes modifying inertia weights, modifying individual particle weights, and modifying swarm social weights; definition The range of inertial weights; The weight of the particle itself; This represents the maximum value of the particle's own weight. This is the step size for the change in the particle's own weight; As a group's social weight; This represents the maximum value of the group's social weight. The step size for changes in the social weight of a group; This is the minimum value of the particle's own weight; This represents the minimum social weight of the group. when hour, Increase by 1 to obtain the updated inertia weight range. : when At that time, Increase The updated particle weights are obtained. : when At that time, Increase The updated group social weight is obtained. : when or or Reset 1. Reset for and reset for : The annealing effector parameter update process includes modifying the cooling function or modifying the initial temperature: Definition Number the cooling function; The initial temperature at the start of the iteration; This represents the maximum temperature. This represents the step size for temperature changes; This is the minimum temperature value; when At that time, Increment by 1 to obtain the updated cooling function number. : when At that time, Increase The updated temperature was obtained. : when or Reset Set to 1 and reset for : The tabu evolver parameter update process includes modifying the tabu list length or modifying the domain search step size: Definition This represents the minimum step size for the neighborhood search. This is the step size for domain search; The step size for changing the step size of the domain search; This is the minimum value of the taboo list length; The length of the taboo table; The step size for changing the length of the taboo list; This represents the maximum value of the neighborhood search step size; This is the maximum value of the taboo list length; when At that time, reduce The updated domain search step size is obtained. : when At that time, reduce The updated taboo list length is obtained. : when or Reset for Or reset for .

Images