Lithologic oil reservoir variable credibility proxy optimization method based on multi-source multi-precision data fusion

Abstract

The invention relates to the technical field of lithologic reservoir scheme optimization, and discloses a multi-source multi-precision data fusion lithologic reservoir variable credibility proxy optimization method, which comprises the following steps of: constructing a low-precision carbonate reservoir numerical model only considering pores and a high-precision carbonate reservoir numerical model simultaneously considering pores and cracks; an optimization variable, a design space and a production scheme multi-objective optimization model with economic parameters and accumulated oil production as objective functions are set; sampling the optimization variables by adopting a Latin hypercube sampling method based on the design space so as to generate nl first production schemes and nh second production schemes; establishing a variable credibility agent model for fusing target functions calculated by the first production scheme and the second production scheme based on a scale function; and optimizing each parameter in the production scheme multi-objective optimization model based on a hybrid particle swarm algorithm. According to the method, the optimization efficiency and the quality of an optimization scheme are greatly improved.

Description

Technical Field

[0001] The invention relates to the technical field of lithologic reservoir scheme optimization, and in particular to a lithologic reservoir variable credibility proxy optimization method by multi-source and multi-precision data fusion. Background Art

[0002] In the commercial development of carbonate reservoirs, the optimization design of the development plan directly determines the benefits of its development and the final recovery rate. Due to its complex pore structure and permeability distribution, carbonate reservoirs are difficult to develop and the production behavior of the reservoir is difficult to predict. Therefore, it is crucial to formulate an efficient development plan. Traditional carbonate reservoir development plan optimization design often relies on the experience of reservoir engineers, which can provide preliminary development directions and strategies to a certain extent. However, relying solely on experience often cannot fully grasp the complex physical property changes of the reservoir and its impact on oil production.

[0003] In order to improve the accuracy and reliability of the design, reservoir numerical simulation technology has been widely used in the design of reservoir development plans. It uses computer simulation to predict the production behavior and recovery rate of reservoirs under different development plans by establishing geological and reservoir models. These models require a large amount of input data, such as rock physical parameters, fluid properties, well network layout, etc., and the simulation process requires a lot of computing resources and time. Although numerical simulation can provide relatively accurate prediction results, its high computing cost and long computing cycle limit the breadth and efficiency of its application to a certain extent.

[0004] In recent years, with the development of artificial intelligence technology, reservoir proxy model technology has been widely used in the oil and gas industry and has become a key tool in reservoir engineering. Reservoir proxy models use machine learning algorithms to learn and capture the complex behaviors and relationships of reservoirs from a large amount of numerical simulation data, thereby achieving rapid prediction and optimization of reservoir development plans. This technology can significantly reduce computing costs, improve the speed and flexibility of optimization design, and enable engineers to evaluate the potential effects of multiple development plans in a short period of time, and conduct sensitivity analysis and uncertainty assessment. With the introduction of advanced artificial intelligence technologies such as deep learning, the accuracy and adaptability of proxy models have been continuously improved, which has promoted the transformation of reservoir development from experience-driven to data-driven and intelligent decision-making, and opened up new prospects for improving oilfield development benefits and recovery. However, carbonate reservoirs have complex reservoir pore structures, strong heterogeneity of physical properties, and multi-scale flow mechanisms, which makes the existing proxy model modeling methods unable to accurately capture their flow characteristics. It is urgent to develop new proxy optimization methods and frameworks for carbonate reservoirs to achieve efficient optimization design of carbonate reservoir production plans. Summary of the invention

[0005] The present invention provides a lithologic reservoir variable credibility proxy optimization method with multi-source and multi-precision data fusion, aiming to solve the problem that the traditional carbonate reservoir production optimization requires huge computing resources and is difficult to find a global optimal solution.

[0006] The present invention is achieved through the following technical solutions:

[0007] A lithologic reservoir variable credibility proxy optimization method based on multi-source and multi-precision data fusion, comprising:

[0008] Construct a low-precision carbonate reservoir numerical model that only considers pores, and a high-precision carbonate reservoir numerical model that considers both pores and fractures;

[0009] Set optimization variables, design space, and a multi-objective optimization model for production plans with economic parameters and cumulative oil production as objective functions;

[0010] Based on the design space, the optimization variables are sampled using a Latin hypercube sampling method to generate n l and n h a first production scheme and a second production scheme, wherein the first production scheme is used for the low-precision carbonate reservoir numerical model to calculate the objective function, and the second production scheme is used for the low-precision carbonate reservoir numerical model and the high-precision carbonate reservoir numerical model to calculate the objective function;

[0011] Based on the scaling function, a variable credibility proxy model is established to fuse the objective functions respectively calculated from the first production plan and the second production plan;

[0012] The parameters in the multi-objective optimization model of the production plan are optimized based on a hybrid particle swarm algorithm, wherein during the optimization process, the variable credibility proxy model is updated based on the particle individuals and populations in the hybrid particle swarm algorithm.

[0013] As an optimization, the vector form of the optimization variable is expressed as u=(I j , P p ), where the optimization variable is I j and P p , I j represents the injection volume of injection well j, P p represents the bottom hole pressure of the production well p, and the design space includes the injection volume range of the water injection well and the bottom hole pressure level range of the production well, wherein the injection volume range of the water injection well is represented by I min ≤I j ≤I max The bottom hole pressure range of the production well is expressed as P min ≤P p ≤P max , where Imin ,I max Respectively represent the lower and upper limits of the injection volume of the injection well, P min , P max They respectively represent the lower and upper limits of the bottom hole pressure level of the production well.

[0014] As an optimization, the objective function is:

[0015]

[0016] The constraints are:

[0017] stg(u)≤0,lb≤u≤ub

[0018] Among them, f1(u) represents the economic parameter, f2(u) represents the cumulative oil production, g(u) is the constraint condition, lb and ub are the lower and upper limits of the optimization variable, and u is the vector form of the optimization variable;

[0019]

[0020] Among them, r o is the oil price; r w is the water treatment price; r g is the gas treatment price; c wi is the water injection price; c gi is the gas injection price, and is the oil production, water production and gas production of the jth well in the nth time period; and is the water injection volume and gas injection volume of the jth well in the nth time period; t n is the number of production time steps; b is the discount rate; P is the number of production wells; I is the number of injection wells; N t is the production time step, Δt n is the length of the nth time step.

[0021] As an optimization, the optimization variables are sampled based on the design space using a Latin hypercube sampling method to generate n l and n h The specific process of calculating the objective function by using the low-precision carbonate reservoir numerical model and the low-precision carbonate reservoir numerical model and the high-precision carbonate reservoir numerical model is as follows:

[0022] A1. Generate n l First production plan:

[0023] A1.1. Divide the injection volume and bottom hole pressure into n equal parts based on the corresponding design spacel small intervals, so that the design space is evenly divided into (n l ) 2 small interval;

[0024] A1.2, in (n l ) 2 n l The first production plan formed by each sampling includes an injection volume and bottom hole pressure, so that n l The first production plan represents the first production plan of the i-th order, i∈[1,n l ];

[0025] A2. Generate n h The second production plan is:

[0026] A2.1. Divide the injection volume and bottom hole pressure into n equal parts based on the corresponding design space. h small intervals, so that the design space is evenly divided into (n h ) 2 small interval;

[0027] A2.2, in (n h ) 2 n h The second production plan formed by each sampling includes an injection volume and bottom hole pressure, so that n h The second production plan represents the i-th second production plan, i∈[1,n h ];

[0028] A3. Using the low-precision carbonate reservoir numerical model to l The first production plan is used to calculate the objective function to obtain n l The first objective function response value, n l The first objective function response value is expressed as

[0029] A4. Using the high-precision carbonate reservoir numerical model to h The second production plan calculates the objective function to obtain n h The second objective function response value, n h The second objective function response value is expressed as

[0030] A5. Using the low-precision carbonate reservoir numerical model to hThe second production plan calculates the objective function to obtain n h The third objective function response value, n h The third objective function response value is expressed as

[0031] As an optimization, the variable credibility proxy model is expressed as:

[0032]

[0033] in, represents a variable credibility proxy model, where a set of sampled data x includes an injection volume and bottom hole pressure, represents the scaling function C(x h ),

[0034] As an optimization, The specific expression is:

[0035]

[0036] r h is the correlation vector R h is the correlation matrix, q is filled with 1 and has a length of n h Column vector of .

[0037] As an optimization, the specific process of optimizing each parameter in the multi-objective optimization model of the production plan based on the hybrid particle swarm algorithm is as follows:

[0038] B1, initializing parameters, setting the maximum number of iterations, the maximum speed of individual particles, the population size N and the termination condition, and initializing the positions and speeds of the individual particles to form a population, wherein the population is composed of a number of the optimization variables;

[0039] B2. Calculate the fitness value of each individual particle;

[0040] B3. Select the global optimal value u from the population according to the fitness value best and the individual optimal value u′ best ;

[0041] B4. Randomly select half of the individual particles in the population and use the DE / rand-to-best operator to perform mutation operation. At the same time, randomly select individual particles in the population and perform crossover operation according to the set crossover operation operator. The crossover operation operator is specifically:

[0042]

[0043] Among them, ui,j (g+1) represents the position of the jth dimension of the i-th particle in the (g+1)th iteration, j∈{1,2}, j=1 represents the injection volume, j=2 represents the bottom hole pressure, V i,j (g+1) represents the position of the jth dimension of the ith particle in the (g+1)th iteration after the mutation operation using the DE / rand-to-best operator, V i,j (g+1)=u i,j (g)+F·(u best，j -u i,j )+F·(u p1，j -u p2,j ), where F is the scaling factor, p1 and p2 are random numbers, p1, p2∈[1, N], N is a positive integer, and u i,j (g) represents the position of the jth dimension of the i-th particle individual in the g-th iteration, and CR is the crossover probability;

[0044] At the same time, the other half of the individual particles in the population are updated according to the following formula:

[0045]

[0046] B5. Calculate the fitness value of each individual particle in the new population, and use the selection operator to update the population, wherein the selection operator is specifically:

[0047]

[0048] Among them, u i (g+1) represents the particle individual after the (g+1)th iteration, u i (g) represents the particle individual after the gth iteration, f(u i (g+1)) represents the fitness value of the individual particle after the (g+1)th iteration, f(u i (g)) represents the fitness value of the individual particle after the g-th iteration; wherein the fitness value of the individual particle is the objective function of the individual particle;

[0049] B6. Determine whether the termination condition is met. If so, output the global optimal value u best , otherwise, jump to B7;

[0050] B7, determine whether a predefined fixed number of iterations is reached, wherein the fixed number of iterations is less than the maximum number of iterations. If so, jump to B8, otherwise, return to B4;

[0051] B8. The minimum Euclidean distance between the individual particles in the current population and the samples in the high-precision sample set composed of the second production scheme other than the population is used as a new individual particle to update the population, and then jump to B6.

[0052] As an optimization, the fitness value of the individual particle is the objective function of the individual particle. When calculating the fitness value of the individual particle, if the dominant state of the individual particle is not changed due to the prediction uncertainty of the variable credibility proxy model, the fitness value of the individual particle is calculated by the variable credibility proxy model, otherwise it is calculated by the high-precision carbonate reservoir numerical model.

[0053] As an optimization, the prediction uncertainty of the individual particles is characterized as follows:

[0054]

[0055] Among them, u i represents the i-th particle individual, if Then determine the particle individual x i The dominant state of is not changed by the prediction uncertainty of the variable credibility proxy model, where x i represents the dominance point, x k represents a non-dominated point, (i≠k) and i, k∈[1,M], M is the number of individual particles in the population, MMD is the minimum value of the minimum distance between all dominated points and all non-dominated points, is the variance; r c is the correlation vector of the scaling function; R c is the correlation matrix of the scaling function, σ c (·) indicates that the prediction of individual particles is uncertain.

[0056] As an optimization, in B8, the formula corresponding to population update is:

[0057]

[0058] Among them, U p is a high-precision sample set, l is the number of samples in the high-precision sample set, d(u i ,u m ) represents the Euclidean distance between the particle individual i of the current population and the mth sample in the high-precision sample set.

[0059] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0060] The present invention aims at the reservoir physical properties and pore deconstruction characteristics of carbonate reservoirs, respectively establishes carbonate reservoir numerical models of different precisions that only consider pores and pore fractures at the same time, and samples multi-source data based on this to obtain data samples of different precisions, and applies the variable credibility proxy model modeling method based on the scaling function to realize the fusion proxy modeling of multi-source and multi-precision data, give play to the characteristics and advantages of data samples of different precisions, and improve the prediction accuracy of the proxy model. The method is based on multi-source and multi-precision data, variable credibility modeling technology based on scaling functions, and model iteration update strategy. In the process of development scheme optimization iteration, the individual-based and population-based model update strategies are considered to realize the efficient and accurate construction of carbonate reservoir proxy models. On this basis, combined with the hybrid multi-objective particle swarm algorithm, an intelligent optimization framework based on the variable credibility proxy model and the multi-objective optimization algorithm is built to realize the efficient multi-objective optimization of the carbonate reservoir production plan. This optimization method and framework can complete the production plan prediction of carbonate oil reservoirs within seconds, achieve rapid response of the objective function in the iterative process of production plan optimization, replace time-consuming numerical simulation calculations, and overcome the problem that traditional proxy models cannot adapt to the complex reservoir characteristics of carbonate oil reservoirs, greatly improving the optimization efficiency and the quality of the optimization plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0062] Figure 1 is a flow chart of the optimization method of the present invention;

[0063] Figure 2 This is the interface diagram of the low-precision numerical simulation model;

[0064] Figure 3 This is the interface diagram of the high-precision numerical simulation model;

[0065] Figure 4 This is the NPV prediction accuracy verification diagram of the variable credibility model;

[0066] Figure 5 This is the accuracy verification diagram of cumulative oil production prediction of variable credibility model;

[0067] Figure 6 A schematic diagram of a multi-objective optimization solution set in an embodiment;

[0068] Figure 7 This is a production plan display diagram in which both NPV and cumulative oil production are relatively large in the embodiment; Figure 8 Schematic diagram of the calculation process of MMD in the embodiment; Fig. 9It is a representation diagram of the relationship between the two MMDs and the prediction deviation in the embodiment. DETAILED DESCRIPTION

[0069] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0070] This embodiment 1 provides a lithologic reservoir variable credibility proxy optimization method using multi-source and multi-precision data fusion, including:

[0071] S1. Construct a low-precision carbonate reservoir numerical model that only considers pores and a high-precision carbonate reservoir numerical model that considers both pores and fractures;

[0072] That is, this step is to build a sampling numerical model. Construct numerical models of carbonate reservoirs with different precisions, marked as A and B, considering only pores and pore fractures at the same time. The number of grids n and the number of fractures n f and the reservoir scale x×y×z are set according to the actual reservoir conditions.

[0073] That is, in this embodiment, model A is a low-precision carbonate reservoir numerical model, and model B is a high-precision carbonate reservoir numerical model.

[0074] S2, setting optimization variables, design space and a multi-objective optimization model for production plans with economic parameters and cumulative oil production as objective functions;

[0075] That is, this step is to design a multi-objective optimization model of optimization variables, design space, objective function and production plan.

[0076] 1. Optimize variable definition and set the injection volume I of the injection well j , the bottom hole pressure of the production well P p , forming the optimization variable vector u=(I j ,P p ).

[0077] That is, the vector form of the optimization variable is expressed as u=(I j , P p ), where the optimization variable is I j and P p , I j represents the injection volume of injection well j, P p It represents the bottom hole pressure of production well p, j is the number of injection well, and p is the number of production well.

[0078] 2. Define the design space of the optimization variables and set the injection volume range of the injection well to I min To I maxm 3 / day, the bottom hole pressure of the production well is P min To P max MPa.

[0079] That is, the design space includes the injection volume range of the water injection well and the bottom hole pressure range of the production well, wherein the injection volume range of the water injection well is represented by I min ≤I j ≤I max The bottom hole pressure range of the production well is expressed as P min ≤P p ≤P max , where I min ,I max Respectively represent the lower and upper limits of the injection volume of the injection well, P min , P max They respectively represent the lower and upper limits of the bottom hole pressure level of the production well.

[0080] Define the objective function, considering the economic parameter NPV (f1) and the cumulative oil production COP (f2) as the objective function of multi-objective optimization:

[0081]

[0082] Among them, r o is the oil price; r w is the water treatment price; r g is the gas treatment price; c wi is the water injection price; c gi is the gas injection price, and is the oil production, water production and gas production of the jth well in the nth time period; and is the water injection volume and gas injection volume of the jth well in the nth time period; t n is the number of production time steps; b is the discount rate; P is the number of production wells; I is the number of injection wells; N t is the production time step, Δt n is the length of the nth time step.

[0083] 4. Multi-objective optimization model of production plan. Based on the optimization variables, design space and objective function settings, the multi-objective optimization model of production plan is established as follows:

[0084]

[0085] The constraints are:

[0086] s .t. g(u)≤0 ,lb≤u≤ub (4)

[0087] Where f1(u) represents the economic parameter, f2(u) represents the cumulative oil production, g(u) is the constraint condition, lb and ub are the lower and upper limits of the optimization variable, and u is the vector form of the optimization variable; lb = {I min , P min},ub={I max , P max}.

[0088] S3, based on the design space, the optimization variables are sampled using a Latin hypercube sampling method to generate n l and n h a first production scheme and a second production scheme, wherein the first production scheme is used for the low-precision carbonate reservoir numerical model to calculate the objective function, and the second production scheme is used for the low-precision carbonate reservoir numerical model and the high-precision carbonate reservoir numerical model to calculate the objective function;

[0089] For example, the production plan is equivalent to x, the objective function is equivalent to y, and the carbonate reservoir numerical model is equivalent to the function of the relationship between x and y. The production method is input into the carbonate reservoir numerical model, and simulation calculation is performed to obtain the objective function value.

[0090] The Latin Hypercube Sampling (LHS) method is used to sample multi-source and multi-precision data. In the design space of the optimization variables, the LHS method is used to produce n l and n h The basic principle of a production plan is as follows: Assume that n data samples are generated in a k-dimensional design variable space, that is, n samplings are performed. The k design variables need to be equally divided into n small intervals within their set range, so that the entire design space is equally divided into (n) k Then, in the design space, we perform n samplings from 0, 1, ..., n-1 that meet the following conditions: (1) the sample points of each sampling are randomly distributed in the equally divided interval; (2) after sampling, the projection of all data sample points on any dimension is unique. And use model A to l The objective function is calculated for each production plan using the A and B models. h The objective function is calculated for each production plan.

[0091] In some embodiments, the optimization variables are sampled based on the design space using a Latin hypercube sampling method to generate n l and n h The specific process of calculating the objective function by using the low-precision carbonate reservoir numerical model and the low-precision carbonate reservoir numerical model and the high-precision carbonate reservoir numerical model is as follows:

[0092] A1. Generate n l First production plan:

[0093] A1.1. Divide the injection volume and bottom hole pressure into n equal parts based on the corresponding design space l small intervals, so that the design space is evenly divided into (n l ) 2 small interval;

[0094] A1.2, in (n l ) 2 n l The first production plan formed by each sampling includes an injection volume and bottom hole pressure, so that n l The first production plan represents the first production plan of the i-th order, i∈[1,n l ];

[0095] A2. Generate n h The second production plan is:

[0096] A2.1. Divide the injection volume and bottom hole pressure into n equal parts based on the corresponding design space. h small intervals, so that the design space is evenly divided into (n h ) 2 small interval;

[0097] A2.2, in (n h ) 2 n h The second production plan formed by each sampling includes an injection volume and bottom hole pressure, so that n h The second production plan represents the i-th second production plan, i∈[1,n h ];

[0098] A3. Using the low-precision carbonate reservoir numerical model to l The first production plan is used to calculate the objective function to obtain n l The first objective function response value, n l The first objective function response value is expressed as

[0099] A4. Using the high-precision carbonate reservoir numerical model to h The second production plan calculates the objective function to obtain n h The second objective function response value, n hThe second objective function response value is expressed as

[0100] A5. Using the low-precision carbonate reservoir numerical model to h The second production plan calculates the objective function to obtain n h The third objective function response value, n h The third objective function response value is expressed as

[0101] S4, establishing a variable credibility proxy model based on a scaling function to fuse the objective functions respectively calculated from the first production plan and the second production plan;

[0102] The variable credibility proxy model is used to accelerate the calculation of the objective function in the optimization process, replacing the numerical simulation process. The purpose of variable credibility is to achieve a dynamic balance between the modeling computational burden and model performance by fusing data of different accuracies, and to improve the performance of the proxy model under limited computing resources.

[0103] That is, multi-source and multi-precision data fusion variable credibility modeling based on scaling function. The scaling function is expressed as the gap between the high-precision model and the low-precision model, with the high-precision data sample set As the benchmark, the high-precision objective function response is The scaling function sample set is The scaling function in the variable credibility proxy model is expressed as follows:

[0104]

[0105] The Gaussian process regression model is used to characterize the scaling function as follows:

[0106]

[0107] For detailed Gaussian process regression model, please refer to relevant literature.

[0108] On this basis, the variable credibility proxy model based on scaling function is expressed as follows:

[0109]

[0110] Based on the above process, in this embodiment, the variable credibility proxy model is expressed as:

[0111]

[0112] in, represents a variable credibility proxy model, where a set of sampled data x includes an injection volume and bottom hole pressure, represents the scaling function C(x h ),

[0113] The specific expression is:

[0114]

[0115] r h is the correlation vector R h is the correlation matrix, q is filled with 1 and has a length of n h Column vector of .

[0116] S5. Optimizing each parameter in the multi-objective optimization model of the production plan based on a hybrid particle swarm algorithm, wherein during the optimization process, the variable credibility proxy model is updated based on the particle individuals and populations in the hybrid particle swarm algorithm.

[0117] That is, this step is a multi-objective optimization based on the hybrid particle swarm algorithm DE-PSO algorithm combined with the variable credibility agent model.

[0118] The pseudo code of the conventional DE-PSO algorithm is as follows:

[0119] Initialize population P and velocity V

[0120] Calculate the fitness of the population P and set the global best position g_best and the individual best position p_best

[0121] While the termination condition is not met:

[0122] For each individual iin population P:

[0123] If i belongs to the differential evolution part:

[0124] Generate new individuals u_i through DE mutation and crossover

[0125] else:

[0126] Generate new position x_i through PSO velocity and position update

[0127] Calculate the fitness of the new individual and update p_best and g_best according to the fitness

[0128] Update the population using a selection operation

[0129] Output the global best solution g_best

[0130] DE-PSO can effectively balance the global search and local search capabilities by combining the advantages of DE and PSO.

[0131] In the present invention, a new strategy is set for updating individual particles and populations. Next, step S5 is specifically introduced.

[0132] 1. Initialize:

[0133] B1, initializing parameters, setting the maximum number of iterations, the maximum speed of individual particles, the population size N and the termination condition, and initializing the positions and speeds of the individual particles to form a population, wherein the population is composed of a number of the optimization variables;

[0134] The first and second production plans are used to construct a variable credibility proxy model. Therefore, the initialization of individual particles here is a random optimization production plan, which is randomly generated and selected, which is different from the first and second production plans mentioned above. If the injection and production parameters are continuous variables, they are directly initialized using the following formula: Among them, lb and ub are the lower and upper bounds of each dimension of the optimization variable respectively; Represents a random number in the interval [0,1].

[0135] B2. Calculate the fitness value of each individual particle;

[0136] B3. Select the global optimal value u from the population according to the fitness value best and the individual optimal value u′ best ;

[0137] 2. Perform mixing operations:

[0138] B4. Randomly select half of the individual particles in the population and use the DE / rand-to-best operator to perform mutation operation. At the same time, randomly select individual particles in the population and perform crossover operation according to the set crossover operation operator. The crossover operation operator is specifically:

[0139]

[0140] Among them, u i,j (g+1) represents the position of the jth dimension of the i-th particle in the (g+1)th iteration, j∈{1,2}, j=1 represents the injection volume, j=2 represents the bottom hole pressure, V i,j (g+1) represents the position of the jth dimension of the ith particle in the (g+1)th iteration after the mutation operation using the DE / rand-to-best operator. Vi ,j (g+1)=u i,j (g)+F·(u best，j -u i,j )+F·(u p1，j -u p2,j), where F is the scaling factor, p1 and p2 are random numbers, p1, p2∈[1, N], N is a positive integer, and u i,j (g) represents the position of the jth dimension of the i-th particle individual in the g-th iteration, and CR is the crossover probability;

[0141] At the same time, the other half of the individual particles in the population are updated according to the following formula:

[0142]

[0143] That is, the hybrid operation mainly includes DE mutation and crossover operations and PSO speed and position update operations:

[0144] ①DE mutation and crossover operation:

[0145] Differential evolution (DE) achieves individual mutation through differential strategy. The common differential strategy is to randomly select two or more individuals in the population, calculate the vector difference, and then scale and synthesize the vector difference and further sum it with the individual to be mutated. Depending on the mutation operator, the differential evolution algorithm has a variety of mutation modes. DE / rand algorithms have strong global search capabilities, but slow convergence speed, and DE / best have strong local search capabilities and fast convergence speed, but are prone to fall into local optimality.

[0146] Randomly select half of the individuals in the population and use the DE / rand-to-best operator that combines DE / rand and DE / best to balance the global and local search capabilities of the algorithm: V i =u i +F·(u best -u i )+F·(u p1 -u p2 ). Among them, p1 and p2 are random numbers, and F = [0,2] is called the scaling factor, which is a certain constant. The purpose of the crossover operation is to randomly select individuals, because differential evolution is also a random algorithm. The crossover operator adopted by the differential evolution algorithm is: Among them, CR is the crossover probability; g represents the gth generation.

[0147] ②PSO speed and position update For another part of the individuals in the population, the PSO speed and position update formula is used to generate new individuals. In an n-dimensional space, there are m particles, and the position t of particle i at time t is i=1,2,…,m. The speed of particle i is At time t, the velocity and position of particle i in the jth dimension are updated as follows:

[0148]

[0149] B5. Calculate the fitness value of each individual particle in the new population, and use the selection operator to update the population, wherein the selection operator is specifically:

[0150]

[0151] Among them, u i (g+1) represents the particle individual after the (g+1)th iteration, u i (g) represents the particle individual after the gth iteration, f(u i (g+1)) represents the fitness value of the individual particle after the (g+1)th iteration, f(u i (g)) represents the fitness value of the particle individual after the g-th iteration; wherein the fitness value of the particle individual is the objective function of the particle individual; according to the greedy selection strategy in DE, that is, selecting the better individual as the new individual, for the maximization problem, the selection operation operator is:

[0152]

[0153] B6. Determine whether the termination condition is met. If so, output the global optimal value u best , otherwise, jump to B7;

[0154] B7, determine whether a predefined fixed number of iterations is reached, wherein the fixed number of iterations is less than the maximum number of iterations. If so, jump to B8, otherwise, return to B4;

[0155] B8. The minimum Euclidean distance between the individual particles in the current population and the samples in the high-precision sample set composed of the second production scheme other than the population is used as a new individual particle to update the population, and then jump to B6.

[0156] Specifically, the fitness value of the individual particle is the objective function of the individual particle. When calculating the fitness value of the individual particle, if the dominant state of the individual particle is not changed due to the prediction uncertainty of the variable credibility proxy model, the fitness value of the individual particle is calculated by the variable credibility proxy model, otherwise it is calculated by the high-precision carbonate reservoir numerical model.

[0157] The prediction uncertainty of the individual particles is characterized as follows:

[0158]

[0159] is the variance; r c is the correlation vector of the scaling function; R c is the correlation matrix of the scaling function;

[0160] Among them, x i represents the i-th particle individual, if (i≠k) and i, k∈[1,M], M is the number of individual particles in the population, then determine the individual particle x i The dominant state of is not changed by the prediction uncertainty of the variable credibility proxy model, where x i represents the dominance point, x k represents a non-dominated point, MMD is the minimum value of the minimum distance between all dominated points and all non-dominated points, σ c (·) indicates that the prediction of individual particles is uncertain.

[0161] It should be noted that dominated points and non-dominated points are basic concepts used to evaluate the relative merits of solution sets.

[0162] Domination: In multi-objective optimization, a solution x1 is said to dominate another solution x2 if x1 is not inferior to x2 in all objectives and is strictly superior to x2 in at least one objective. In other words, x1 is equivalent to x2 in all objectives but better than x2 in at least one objective.

[0163] Non-dominated: A solution is non-dominated if no solution dominates another solution. In simple terms, x1 is non-dominated if no solution is better than it in all objectives at the same time.

[0164] That is to say, the dominated point is inferior to the non-dominated point and is dominated by the non-dominated point.

[0165] x i (or called u i ), represents a solution in multi-objective optimization, then x k is not inferior to x i (Dominate x i ) is a solution to .

[0166] The formula corresponding to population update is:

[0167]

[0168] Among them, U p is a high-precision sample set, l is the number of samples in the high-precision sample set, d(u i ,u m ) represents the Euclidean distance between the particle individual i of the current population and the mth sample in the high-precision sample set.

[0169] Next, the specific process of updating the variable credibility proxy model based on the particle individuals and population in the hybrid particle swarm algorithm is specifically introduced.

[0170] Update strategy of individual-based variable credibility agent model;

[0171] Before a predefined fixed number of iterations, when a particle individual is a non-dominated point, the particle individual is calculated by a numerical simulation model to update the over-variable credibility proxy model;

[0172] After a predefined fixed number of iterations, the particle individual with the minimum Euclidean distance from the second production solution is selected and input into the variable credibility proxy model to update the variable credibility proxy model.

[0173] Based on the Gaussian process regression model theory, the prediction uncertainty at point x0 is characterized as follows:

[0174]

[0175] The prediction deviation I(x0) at point x0 (the prediction deviation is represented by the difference between the value of the second type of production plan predicted by the variable credibility model and the value calculated by the numerical simulation (the previous simulation model)) is within the 95.5% confidence interval of the variable credibility model. According to the 2σ principle, if the dominant state of a certain production plan individual is not changed by the prediction uncertainty of the variable credibility proxy model, its objective function is calculated by the variable credibility proxy model, otherwise it is calculated by model B. The selection criteria for its production plan individuals are: the relationship between the minimum distance minimum (MMD) and the prediction deviation (that is, the relationship between MMD and the prediction deviation, which is larger or smaller, that is, the relationship between MMD and the prediction deviation. If MMD is greater than the prediction deviation, the model does not need to be updated. If MMD is less than the prediction deviation, the model must be updated), such as Figure 8 As shown. In the selection process, the MMD of an individual is defined as the minimum distance between the dominant point and the non-dominated point. The specific calculation process of MMD is shown in the figure below. The distance d between the points is defined as the Euclidean distance between the two points.

[0176] At the same time, MMD and forecast bias have two different relationships: Fig. 9 As shown:

[0177] The first is (a), where the prediction bias does not affect the dominance relationship between individuals, and the second is (b), where the prediction bias affects the dominance relationship between individuals. When the first condition is met, there is no need to use model B to calculate the data sample, and when it belongs to the second condition, model B needs to be used to calculate the data sample.

[0178] The selection formula that needs to be used to calculate the data sample using model B is as follows:

[0179]

[0180] in:

[0181]

[0182] (2) Population-based model updating strategy.

[0183] In order to ensure the diversity of the solution search population during the optimization iteration process, the population-based update strategy is considered during the optimization iteration process. After a predefined fixed number of iterations, the samples with the minimum Euclidean distance between other HF samples (second production solutions) are selected for HF analysis. In summary, samples that meet the following conditions are selected:

[0184]

[0185] Where Xp is the current high-precision sample set, l is the number of samples in the current high-precision sample set, and d(x,x m ) is the Euclidean distance between x and the mth current high-precision sample set.

[0186] The above two model updating strategies continuously improve the accuracy of the variable credibility proxy model during the optimization process.

[0187] In each iteration step, new samples are added to the variable credibility proxy model to improve its prediction accuracy, which is based on the above-mentioned update strategy considering individuals and populations.

[0188] Therefore, the effects of the present invention are:

[0189] 1. The present invention realizes the sampling of multi-source and multi-precision data samples by establishing numerical simulation models of oil reservoirs with different precisions, and gives full play to the advantages of different data samples.

[0190] 2. The present invention utilizes a variable credibility proxy model modeling method based on a scaling function to achieve efficient fusion modeling of multi-source and multi-precision reservoir numerical simulation data.

[0191] 3. In the optimization iterative process, the present invention considers two different proxy model update strategies, namely, individual-based and population-based, to improve the prediction performance of the proxy model.

[0192] 4. Based on the integration of multi-source and multi-precision data sample variable credibility proxy modeling, the present invention combines the hybrid multi-objective example group algorithm to give full play to its ability to effectively balance global search and local search, and realize multi-objective optimization decision-making of carbonate oil reservoir production plans.

[0193] 5. The present invention proposes a complete carbonate reservoir variable credibility proxy optimization framework based on multi-source and multi-precision data fusion, from the definition of numerical models with different precisions, the setting of optimization models, the sampling of data samples, the training of variable credibility models, model updating strategies to hybrid multi-objective algorithms, to achieve rapid response of the objective function in the iterative process of production plan optimization, replace time-consuming numerical simulation calculations, and overcome the problem that traditional proxy models cannot adapt to the complex reservoir characteristics of carbonate reservoirs, greatly improving the optimization efficiency and the quality of the optimization plan.

[0194] Take a carbonate well with one injection and one production as an example:

[0195] 1. Establishment of numerical model of carbonate reservoir:

[0196] According to step S1, a carbonate reservoir model considering only pores with a grid number of 1000 and a reservoir size of 100×100×10 is established by a stochastic modeling method, and a carbonate reservoir model considering both pores and fractures with a grid number of 1000, a fracture number of 80 and a reservoir size of 100×100×10 is established. Figure 2 and 3 shown.

[0197] 2. Set the optimization variable space according to step S2, consider the injection pressure of the injection well to be 25-30MPa, the bottom hole pressure of the production well to be 10-25MPa, define NPV and oil production as two objective functions, and build a multi-objective optimization model for carbonate reservoir production plan

[0198] 3. Multi-source and multi-precision data sampling:

[0199] Based on step 2, the Latin hypercube sampling method is used to generate 1000 groups of samples (the first production plan) for sampling model A (low-precision numerical simulation model), and generate 100 groups of samples (the second production plan) for sampling model B (high-precision numerical simulation model).

[0200] 4. Establish a variable credibility proxy model for multi-source and multi-precision data fusion:

[0201] Based on the data sample set established in step 3, according to step S4, the fusion of multi-source and multi-precision data is realized, and the variable credibility proxy model is constructed and its accuracy is verified. The verification effect is as follows Figure 4 and 5 shown.

[0202] 5. Multi-objective optimization of production plans

[0203] According to step F, the advantages of the hybrid multi-objective particle swarm algorithm are brought into play, and the model update iteration strategy based on individuals and populations in step S5 is considered during the optimization iteration process to further improve the prediction performance of the proxy model during the optimization process, and obtain the optimal Pareto solution set as follows: Figure 6 As shown, the production scheme with relatively large NPV and oil production is selected as shown below Figure 7 shown.

[0204] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A lithologic reservoir variable credibility proxy optimization method based on multi-source and multi-precision data fusion, characterized in that: include: Construct a low-precision carbonate reservoir numerical model that only considers pores, and a high-precision carbonate reservoir numerical model that considers both pores and fractures; Set optimization variables, design space, and a multi-objective optimization model for production plans with economic parameters and cumulative oil production as objective functions; Based on the design space, the optimization variables are sampled using the Latin hypercube sampling method to generate n l and n h a first production scheme and a second production scheme, wherein the first production scheme is used for the low-precision carbonate reservoir numerical model to calculate the objective function, and the second production scheme is used for the low-precision carbonate reservoir numerical model and the high-precision carbonate reservoir numerical model to calculate the objective function; Based on the scaling function, a variable credibility proxy model is established to fuse the objective functions respectively calculated from the first production plan and the second production plan; The parameters in the multi-objective optimization model of the production plan are optimized based on a hybrid particle swarm algorithm, wherein during the optimization process, the variable credibility proxy model is updated based on the particle individuals and populations in the hybrid particle swarm algorithm.

2. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 1 is characterized in that: The vector form of the optimization variable is expressed as u=(I j , P p ), where the optimization variable is I j and P p , I j represents the injection volume of injection well j, P p represents the bottom hole pressure of the production well p, and the design space includes the injection volume range of the water injection well and the bottom hole pressure level range of the production well, wherein the injection volume range of the water injection well is represented by I min ≤I j ≤I max The bottom hole pressure range of the production well is expressed as P min ≤P p ≤P max , where I min ,I max Respectively represent the lower and upper limits of the injection volume of the injection well, P min , P max They respectively represent the lower and upper limits of the bottom hole pressure level of the production well.

3. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 2 is characterized in that: The objective function is: The constraints are: stg(u)≤0,lb≤u≤ub Among them, f1(u) represents the economic parameter, f2(u) represents the cumulative oil production, g(u) is the constraint condition, lb and ub are the lower and upper limits of the optimization variable, and u is the vector form of the optimization variable; Among them, r o is the oil price; r w is the water treatment price; r g is the gas treatment price; c wi is the water injection price; c gi is the gas injection price, and is the oil production, water production and gas production of the jth well in the nth time period; and is the water injection volume and gas injection volume of the jth well in the nth time period; t n is the number of production time steps; b is the discount rate; P is the number of production wells; I is the number of injection wells; N t is the production time step, Δt n is the length of the nth time step.

4. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 2 is characterized in that: Based on the design space, the optimization variables are sampled using the Latin hypercube sampling method to generate n l and n h The specific process of calculating the objective function by using the low-precision carbonate reservoir numerical model and the low-precision carbonate reservoir numerical model and the high-precision carbonate reservoir numerical model is as follows: A1. Generate n l First production plan: A1.

1. Divide the injection volume and bottom hole pressure into n equal parts based on the corresponding design space l small intervals, so that the design space is evenly divided into (n l ) 2 ; A1.2, in (n l ) 2 n l The first production plan formed by each sampling includes an injection volume and bottom hole pressure, so that n l The first production plan represents the first production plan of the ith one, i∈[1,n l ]; A2. Generate n h The second production plan is: A2.

1. Divide the injection volume and bottom hole pressure into n equal parts based on the corresponding design space. h small intervals, so that the design space is evenly divided into (b h ) 2 small interval; A2.2, in (n h ) 2 n h The second production plan formed by each sampling includes an injection volume and bottom hole pressure, so that n h The second production plan represents the i-th second production plan, i∈[1,n h ]; A3. Using the low-precision carbonate reservoir numerical model to l The first production plan is used to calculate the objective function to obtain n l The first objective function response value, n l The first objective function response value is expressed as A4. Using the high-precision carbonate reservoir numerical model to h The second production plan calculates the objective function to obtain n h The second objective function response value, n h The second objective function response value is expressed as A5. Using the low-precision carbonate reservoir numerical model to h The second production plan calculates the objective function to obtain n h The third objective function response value, n h The third objective function response value is expressed as 5. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 4 is characterized in that: The variable credibility proxy model is expressed as: in, represents a variable credibility proxy model, where a set of sampled data x includes an injection volume and bottom hole pressure, represents the scaling function C(x h ), cxih=Fhxih-Flxih, i∈1,nh.

6. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 5, characterized in that: The specific expression is: r h is the correlation vector R h is the correlation matrix, q is filled with 1 and has a length of n h Column vector of .

7. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 3 is characterized in that: The specific process of optimizing each parameter in the multi-objective optimization model of the production plan based on the hybrid particle swarm algorithm is as follows: B1, initializing parameters, setting the maximum number of iterations, the maximum speed of individual particles, the population size N and the termination condition, and initializing the positions and speeds of the individual particles to form a population, wherein the population is composed of a number of the optimization variables; B2. Calculate the fitness value of each individual particle; B3. Select the global optimal value u from the population according to the fitness value best and the individual optimal value u′ best ; B4. Randomly select half of the individual particles in the population and use the DE / rand-to-best operator to perform mutation operation. At the same time, randomly select individual particles in the population and perform crossover operation according to the set crossover operation operator. The crossover operation operator is specifically: Among them, u i,j (g+1) represents the position of the jth dimension of the i-th particle in the (g+1)th iteration, j∈{1,2}, j=1 represents the injection volume, j=2 represents the bottom hole pressure, V i,j (g+1) represents the position of the jth dimension of the ith particle in the (g+1)th iteration after the mutation operation using the DE / rand-to-best operator, V i,j (g+1)=u i,j (g)+F·(u best，j -u i,j )+F·(u p1，j -u p2,j ), where F is the scaling factor, p1 and p2 are random numbers, p1, p2∈[1, N], N is a positive integer, and u i,j (g) represents the position of the jth dimension of the i-th particle individual in the g-th iteration, and CR is the crossover probability; At the same time, the other half of the individual particles in the population are updated according to the following formula: B5. Calculate the fitness value of each individual particle in the new population, and use the selection operator to update the population, wherein the selection operator is specifically: Among them, u i (g+1) represents the particle individual after the (g+1)th iteration, u i (g) represents the particle individual after the gth iteration, f(u i (g+1)) represents the fitness value of the individual particle after the (g+1)th iteration, f(u i (g)) represents the fitness value of the individual particle after the gth iteration; B6, determine whether the termination condition is met, if so, output the global optimal value u best , otherwise, jump to B7; B7, determine whether a predefined fixed number of iterations is reached, wherein the fixed number of iterations is less than the maximum number of iterations. If so, jump to B8, otherwise, return to B4; B8. The minimum Euclidean distance between the individual particles in the current population and the samples in the high-precision sample set composed of the second production scheme other than the population is used as a new individual particle to update the population, and then jump to B6.

8. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 7, characterized in that: The fitness value of the individual particle is the objective function of the individual particle. When calculating the fitness value of the individual particle, if the dominant state of the individual particle is not changed due to the prediction uncertainty of the variable credibility proxy model, the fitness value of the individual particle is calculated by the variable credibility proxy model, otherwise it is calculated by the high-precision carbonate reservoir numerical model.

9. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 8, characterized in that: The prediction uncertainty of the individual particles is characterized as follows: in, is the variance; r c is the correlation vector of the scaling function; R c is the correlation matrix of the scaling function; x i represents the i-th particle individual, if (i≠k) and i, k∈[1,M], M is the number of individual particles in the population, then determine the individual particle x i The dominant state of is not changed by the prediction uncertainty of the variable credibility proxy model, where x i represents the dominance point, x k represents a non-dominated point, MMD is the minimum value of the minimum distance between all dominated points and all non-dominated points, σ c (·) represents the prediction uncertainty of individual particles.

10. The method for optimizing lithologic reservoir variable credibility proxy by multi-source and multi-precision data fusion according to claim 7, characterized in that: In B8, the formula corresponding to population update is: Among them, U p is a high-precision sample set, l is the number of samples in the high-precision sample set, d(u i ,u m ) represents the Euclidean distance between the particle individual i of the current population and the mth sample in the high-precision sample set.

Images