Oil reservoir well pattern optimization method

Abstract

This invention belongs to the field of oil and gas field development technology and discloses a reservoir well network optimization method, including the following steps: S1, preprocessing the geological model and establishing a microfacies lookup table; S2, defining decision variables and hyperparameters; S3, automatically screening and iterative evolution based on fitness function guidance and hyperparameters; S4, outputting the well placement scheme and performing geological verification. This invention transforms qualitative sedimentological understanding into computable optimization constraints and deeply integrates them with intelligent algorithms to generate well network deployment schemes that are both economically optimal and meet geological requirements. This fundamentally improves the credibility, interpretability, and computational efficiency of the optimization results, overcoming the shortcomings of existing intelligent optimization algorithms that are detached from geological mechanisms.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field development technology, and in particular to a method for optimizing reservoir well patterns. Background Technology

[0002] In oil and gas field development, optimizing the well network deployment is a core design element that directly determines the development effectiveness and economic benefits. Currently, well network optimization primarily relies on numerical simulation technology, mainly using simulations of production dynamics under different well placement schemes to select the optimal scheme for actual deployment. This traditional method depends on engineers manually designing regular well networks using methods such as the five-point method and the inverse nine-point method, followed by numerical simulation comparisons. This approach is not only inefficient but also heavily reliant on personal experience, making it difficult to handle complex heterogeneous reservoirs. Therefore, intelligent optimization algorithms, such as genetic algorithms and particle swarm optimization, have been introduced into this field to automatically optimize well location parameters and improve efficiency.

[0003] However, existing intelligent optimization algorithms are essentially "black box" operations, lacking clear guidance on geological mechanisms. This can lead to poor geological interpretability of the optimization results, especially for reservoirs with significant sedimentary control. These reservoirs are composed of sedimentary microfacies such as distributary channels, mouth bars, and sheet sands. The significant differences in properties and opposite vertical rhythms between these microfacies (e.g., distributary channels exhibit positive rhythms while mouth bars exhibit negative rhythms) result in drastically different development response mechanisms depending on well placement. Furthermore, existing optimization algorithms often perform unconstrained random searches across the entire spatial range. This "black box" operation ignores the strong spatial heterogeneity of sedimentary reservoirs, leading to numerical simulators expending significant computational resources to simulate geologically flawed development strategies. For example, a failed case might involve drilling a water injection well directly into the center of a high-permeability channel, resulting in explosive water breakthrough. Therefore, this search method is not only extremely inefficient, but even if the theoretically highest NPV (Net Present Value) is found, it often cannot be implemented in the oilfield due to its incompatibility with geological development principles.

[0004] Therefore, how to rationally integrate geological mechanism knowledge into well placement methods and embed optimization algorithms in the form of rules to achieve deep integration of geological knowledge and numerical simulation, and improve the reliability and optimization efficiency of well network schemes, is a problem that needs to be solved by researchers in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a method that transforms qualitative sedimentological understanding into computable optimization constraint rules and deeply integrates them with intelligent algorithms to generate optimal, economical well network deployment schemes that meet geological laws, thereby fundamentally improving the credibility and interpretability of the optimization schemes and optimizing computational efficiency.

[0006] To achieve this objective, the present invention adopts the following technical solution: The method for optimizing reservoir well patterns includes the following steps: S1. Preprocess the geological model and establish a microfacies lookup table; S2. Define decision variables and hyperparameters; S3. Automated screening and iterative evolution are performed based on the fitness function and the hyperparameter. S4. Output the well placement plan and conduct geological verification.

[0007] Optionally, the geological model in step S1 may be set to a three-dimensional geological model.

[0008] Optionally, step S1 includes: S1.1 Import the three-dimensional geological model containing the distribution of sedimentary microfacies into the numerical simulation software for preprocessing; S1.2. Use the partitioning function of the numerical simulation software to encode the attributes of the model mesh of each partition to form a microfacies type code, and establish a corresponding mapping relationship between the three-dimensional spatial coordinates (x, y, z) of the three-dimensional geological model and the microfacies type code. S1.3 Extract the center coordinates of each grid in the three-dimensional geological model and the corresponding microfacies type code as the microfacies partition data of the partition; S1.4 Summarize the microphase partition data of each partition to establish a microphase query table that can be queried in real time at the second level by the optimization algorithm.

[0009] Optionally, step S2 includes: S2.1 Using an intelligent optimization algorithm, the planar coordinates of the planned well are encoded as the decision variables; S2.2 Define at least two sets of hyperparameters to be optimized, including the global geological penalty coefficient α and the microfacies well type weight vector W.

[0010] Alternatively, the geological penalty coefficient α can be used to adjust the intensity of the intervention of geological laws in the optimization calculation process; Each element in the microfacies well type weight vector W represents the degree of preference for well placement of a specific sedimentary microfacies relative to a specific well type.

[0011] Alternatively, the fitness function in step S3 is: F = NPV[1 – α(Penalty)] facies / Penalty base )]; Where F is the fitness function value; NPV stands for Net Present Value in the numerical simulation scheme. α is the geological penalty coefficient; Penalty facies This is the geological penalty value; Penalty base This is the baseline penalty value.

[0012] Alternatively, the geological penalty value can be calculated as follows: The microfacies lookup table is queried based on the coordinates of all planned wells to obtain the microfacies type of each planned well location; Combining the current micro-phase well type weight vector W in the hyperparameter, calculate the (1-W) of all planned wells. i The sum of these values ​​is the geological penalty value. Among them, W i Let W be the value of the micro-phase well type weight vector corresponding to the i-th planned well.

[0013] Alternatively, the baseline penalty value may be a constant, and / or the baseline penalty value may be the average of the geological penalty values ​​of the initial population generated by the plane coordinate encoding initialization of the planned well.

[0014] Optionally, in step S3, the automated screening involves comparing the fitness function values ​​calculated based on the fitness function, retaining excellent individuals, eliminating inferior individuals, and generating a new population through crossover and mutation, thereby iteratively evolving repeatedly. And / or, the condition for iterative evolution is that the convergence condition is met or the number of iterations reaches the target value.

[0015] Optionally, in step S4, the geological verification is completed by overlaying the well locations of each planned well in the well layout scheme with the sedimentary microfacies plane distribution map.

[0016] The beneficial effects of this invention are: This invention preprocesses geological models based on relevant knowledge of sedimentology in geology, thereby forming a microfacies lookup table. This improves computational reliability when calculating fitness functions using hyperparameters, transforming qualitative sedimentological knowledge into computable optimization constraint rules. These rules are then deeply and rationally integrated with intelligent algorithms to generate well network deployment schemes that are both economically optimal and geologically sound. This fundamentally enhances the credibility, interpretability, and computational efficiency of the optimization results. Furthermore, this invention incorporates specific geological constraints into the reservoir well network optimization method, making well location optimization more reasonable and consistent with geological laws. Under these geological constraints, the optimization algorithm can find the optimal solution earlier and faster, rather than searching arbitrarily, thus avoiding wasted time in simulations and increased computational costs. This overcomes the shortcomings of existing intelligent optimization algorithms that are detached from geological mechanisms, achieving a leap from "manual experience-based parameter tuning" to "automatic algorithm-based optimization rules." Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of the reservoir well network optimization method according to an embodiment of the present invention; Figure 2 This is a planar distribution diagram of the depositional microphases in the Petrel software described in this embodiment of the invention; Figure 3 This is a porosity planar distribution map exported from the CMG software described in this embodiment of the invention; Figure 4 This is a permeability plane distribution map exported from the CMG software described in this embodiment of the invention; Figure 5 This is a schematic diagram of the underwater diversion channel (Sector 1) defined in the CMG software according to the embodiments of the present invention; Figure 6 This is a schematic diagram of the main body of the estuary dam (Sector2) defined in the CMG software according to the embodiments of the present invention; Figure 7 This is a schematic diagram of the inner edge of the estuary dam (Sector3) defined in the CMG software according to the embodiments of the present invention; Figure 8 This is a schematic diagram of the sheet-like sand (Sector4) defined in the CMG software described in the embodiments of the present invention; Figure 9 This is a convergence iteration curve of the inner particle swarm optimization process described in the embodiments of the present invention; Figure 10 This is a schematic diagram of the well location distribution of a standard five-point well network in the prior art as described in the embodiments of the present invention; Figure 11 This is a schematic diagram of the well location distribution of the intelligent optimized random well network applied to the reservoir well network optimization method described in this embodiment of the invention; Figure 12 This is a comparison curve of the recovery rate over time between the reservoir well network optimization method described in this embodiment of the invention and the method of the prior art. Detailed Implementation

[0018] In the description of this invention, unless otherwise explicitly specified and limited, the terms "connected," "linked," and "fixed" should be interpreted broadly. For example, they can refer to a fixed connection or a detachable connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0019] In the description of this invention, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0020] In oil and gas field development, optimizing the well network deployment is a core design element that directly determines the development effectiveness and economic benefits. Currently, well network optimization primarily relies on numerical simulation technology, mainly using simulations of production dynamics under different well placement schemes to select the optimal scheme for actual deployment. This traditional method depends on engineers manually designing regular well networks using methods such as the five-point method and the inverse nine-point method, followed by numerical simulation comparisons. This approach is not only inefficient but also heavily reliant on personal experience, making it difficult to handle complex heterogeneous reservoirs. Therefore, intelligent optimization algorithms, such as genetic algorithms and particle swarm optimization, have been introduced into this field to automatically optimize well location parameters and improve efficiency.

[0021] However, existing intelligent optimization algorithms are essentially "black box" operations, lacking clear guidance on geological mechanisms. This can lead to poor geological interpretability of the optimization results, especially for reservoirs with significant sedimentary control. These reservoirs consist of sedimentary microfacies such as distributary channels, mouth bars, and sheet sands. The significant differences in properties and opposite vertical rhythms between these microfacies (e.g., distributary channels exhibit positive rhythms while mouth bars exhibit negative rhythms) result in drastically different development response mechanisms depending on well placement. Furthermore, existing optimization algorithms often perform unconstrained random searches across the entire spatial range. This "black box" operation ignores the strong spatial heterogeneity of sedimentary reservoirs, leading to numerical simulators expending significant computational resources to simulate geologically flawed development strategies. For example, a failed case might involve drilling a water injection well directly into the center of a high-permeability channel, resulting in explosive water breakthrough. Therefore, this search method is not only extremely inefficient, but even if it finds the theoretically highest NPV, it often cannot be implemented in the oilfield due to its incompatibility with geological development principles.

[0022] Therefore, how to rationally integrate geological mechanism knowledge into well placement methods and embed optimization algorithms in the form of rules to achieve deep integration of geological knowledge and numerical simulation, and improve the reliability and optimization efficiency of well network schemes, is a problem that needs to be solved by researchers in this field.

[0023] The technical solution of this embodiment will be further described below with reference to the accompanying drawings and specific implementation methods.

[0024] like Figures 1 to 12 As shown, this embodiment provides a method for optimizing reservoir well patterns, including the following steps: S1. Preprocess the geological model and establish a microfacies lookup table; S2. Define decision variables and hyperparameters; S3. Automated screening and iterative evolution are performed based on fitness function guidance and hyperparameters; S4. Output the well placement plan and conduct geological verification.

[0025] Specifically, this embodiment preprocesses the geological model based on relevant knowledge of sedimentology in geology, thereby forming a microfacies lookup table. This improves computational reliability when using hyperparameters and other methods to calculate the fitness function. It also transforms qualitative sedimentological knowledge into computable optimization constraint rules, which are then deeply and rationally integrated with intelligent algorithms to generate well network deployment schemes that are both economically optimal and meet geological laws. This fundamentally improves the credibility, interpretability, and computational efficiency of the optimization process. Furthermore, this embodiment incorporates specific geological constraints into the reservoir well network optimization method, making well location optimization more reasonable and in line with geological laws. Under these geological constraints, the optimization algorithm can find the optimal solution earlier and faster, rather than searching arbitrarily, thus avoiding wasting time on simulations and increasing computational costs. This overcomes the shortcomings of existing intelligent optimization algorithms that are detached from geological mechanisms, achieving a leap from "manual experience-based parameter tuning" to "automatic algorithm-based optimization rules."

[0026] The specific details of the reservoir well network optimization method in this embodiment are explained below.

[0027] The method for optimizing reservoir well patterns specifically includes the following steps: S1. Preprocess the geological model and establish a microfacies lookup table; S2. Define decision variables and hyperparameters; S3. Automated screening and iterative evolution are performed based on fitness function guidance and hyperparameters; S4. Output the well placement plan and conduct geological verification.

[0028] Step S1 includes: S1.1 Import the three-dimensional geological model containing the distribution of sedimentary microfacies into the numerical simulation software for preprocessing; S1.2. Use the partitioning function of the numerical simulation software to encode the attributes of the model mesh of each partition to form microfacies type codes, and establish a corresponding mapping relationship between the three-dimensional spatial coordinates (x, y, z) of the three-dimensional geological model and the microfacies type codes. S1.3 Extract the center coordinates of each model grid and the corresponding microfacies type code from the three-dimensional geological model as the microfacies partition data for that partition; S1.4 Summarize the microphase partition data of each partition to establish a microphase query table that can be queried in real time at the second level by the optimization algorithm.

[0029] Optionally, in step S1, the geological model is set to a three-dimensional geological model. This allows the three-dimensional geological model containing the distribution of sedimentary microfacies to be imported into the numerical simulation software. The software's own partitioning function can be used to partition the three-dimensional geological model grid according to the microfacies type. Then, the microfacies partitioning data of each partition's grid model can be extracted from each model, thus establishing a microfacies lookup table that can be queried in real time by the optimization algorithm, serving as a database for subsequent calculations.

[0030] Further, step S2 includes: S2.1 Using intelligent optimization algorithms, the planar coordinate codes of the planned wells are set as decision variables; S2.2 Define at least two sets of hyperparameters to be optimized, including the global geological penalty coefficient α and the microfacies well type weight vector W.

[0031] In step S2.1, when the planar coordinate code of the planned well is set as a decision variable, the particle swarm algorithm can be used as an intelligent optimization algorithm, and the planar coordinate code of the planned well is set as the particle position in the particle swarm algorithm.

[0032] Specifically, the total number of wells in the well network to be optimized is set to N (including production wells and injection wells). The location of a single well is determined by two-dimensional coordinates. Therefore, a complete well layout scheme corresponds to a 2N-dimensional decision vector containing 2N coordinate parameters, i.e., the particle position is X = (x1, y1, x2, y2, ..., x...). i y i , ..., x N y N ), where (x i y i Let be the plane coordinates of the i-th well. The search space for the entire set of parameters is limited by the reservoir boundary. A single particle represents a complete well placement scheme, and the population size M represents the number of particles in each generation, i.e., the total number of well placement schemes evaluated simultaneously.

[0033] The particle swarm optimization (PSO) algorithm uses position variables X and velocity variables V to iteratively optimize well locations. X represents the actual coordinates of the wells and is the final solution parameter, while V represents the update step size of the particle positions. The algorithm first updates the velocity and then refreshes the well location coordinates based on the updated velocity. The update formula for the velocity variable V is as follows: V kd (t+1) = ωV kd (t) + c1r1(pbest) kd -X kd (t))+c2r2(gbest d -X kd (t)); Where k is the particle index, corresponding to the kth well layout scheme; d is the dimension index, corresponding to a single coordinate within the scheme; t is the current iteration generation, and t+1 is the iteration generation after the update; ω is the inertia weight, controlling the degree to which the particle maintains its previous motion inertia; c1 is the individual learning factor, representing the step size of the particle learning towards its own historical best position (pbest); c2 is the social learning factor, representing the step size of the particle learning towards the global historical best position (gbest); r1 and r2 are independent random numbers in the interval [0, 1], used to increase the randomness of the search; pbest kd The k-th scheme has the historically optimal coordinates; gbest d These are the coordinates corresponding to the optimal solution for the entire population.

[0034] The update formula for the position variable X is as follows: X kd (t+1) = X kd (t) + V kd (t+1); The latest moving speed, calculated using the update formula for the velocity variable V, is superimposed on the displacement on the original well location coordinates to obtain the new generation of well location parameters.

[0035] With this setup, the particle swarm optimization algorithm uses well location coordinates as decision variables and iteratively searches within the given reservoir space boundary to find the optimal well placement scheme.

[0036] Furthermore, the geological penalty coefficient α is used to adjust the intensity of geological intervention in the optimization calculation process; each element in the microfacies well type weight vector W represents the degree of preference for a specific sedimentary microfacies relative to a specific well type. For example, by defining two sets of hyperparameters to be optimized, the geological penalty coefficient α and the weight vector W, outer-layer optimization can be achieved to facilitate the calculation of the fitness function in the next step, thereby achieving iteration. For example, the specific well type corresponding to the microfacies well type weight vector W includes, but is not limited to, production wells or injection wells.

[0037] Specifically, the fitness function in step S3 is: F=NPV[1–α(Penalty facies / Penalty base )]; Where F is the fitness function value; NPV stands for Net Present Value in the numerical simulation scheme. α is the geological penalty coefficient; Penalty facies This is the geological penalty value; Penalty base This is the baseline penalty value.

[0038] Optionally, the geological penalty value Penalty facies The calculation method is as follows: Based on the coordinates of all planned wells, query the microfacies lookup table to obtain the microfacies type of each planned well's location; combine this with the microfacies well type weight vector W in the current hyperparameters to calculate the (1-W) of all planned wells. i The sum of W and W is the geological penalty value. i Let W be the value of the microfacies well type weight vector corresponding to the i-th planned well. Further, let Penalty be the baseline penalty value. base For constants, and / or, the baseline penalty value is the average of the geological penalty values ​​of the initial population generated by the plane coordinate encoding initialization of the planned well.

[0039] Specifically, this embodiment uses a fitness function as a guide for iterative evolution, and the fitness function value F is the product of the net present value (NPV) of the scheme and the geological penalty ratio term. The above formula is cleverly used for calculation to achieve inner-layer optimization with well location coordinates as decision variables and fitness function value F as the objective, thereby achieving the effect of well network optimization search.

[0040] Accordingly, the automated screening in step S3 specifically involves: comparing the fitness function values ​​F calculated based on the fitness function, tracking and recording the individual historical best position (pbest) of each particle and the global historical best position (gbest) of the entire particle swarm, and then guiding each particle towards pbest and gbest directions according to the velocity update formula and position update formula of the particle swarm optimization algorithm, thereby generating a new generation of particle swarms, and iterating and evolving repeatedly; and / or, the condition for iterative evolution is reaching the convergence condition or reaching the target number of iterations. Specifically, in each generation of evolution, the NPV of candidate schemes calculated on the numerical simulation software can be automatically called, and the fitness function value F can be calculated using the calculation formula in step S3. Then, the optimization algorithm is used to retain excellent individuals and eliminate inferior individuals based on the value of the fitness function value F, and new populations are generated using crossover and mutation operations, so as to reach the convergence condition or the target number of iterations, i.e., the maximum number of iterations. Furthermore, by reducing the need for numerical simulations of numerous invalid geological schemes, the algorithm can more effectively focus on finding optimal well locations in areas with "high weight and high potential," thus greatly reducing the search space.

[0041] In this embodiment, the specific process of "using an optimization algorithm to retain excellent individuals and eliminate inferior individuals based on the fitness function value F, and using crossover and mutation operations to generate a new population, so as to achieve the convergence condition or the target number of iterations" is as follows: The fitness function value F of each particle position calculated according to the fitness function is compared. Each particle compares its current position fitness value with the fitness value of its individual historical best position (pbest). If the fitness value of the current position is higher, then pbest is updated to the current position. At the same time, each particle's pbest is compared with the fitness value of the global historical best position (gbest). If the fitness value of a particle's pbest is higher than gbest, then gbest is updated to the pbest. Then, the velocity and position of all particles are updated according to the aforementioned velocity update formula and position update formula to generate a new generation of particle swarm. This process is repeated iteratively until the convergence condition is reached or the number of iterations reaches the preset maximum value.

[0042] Specifically, in step S4, geological verification is completed by overlaying the well locations of each planned well in the well layout scheme with the sedimentary microfacies plane distribution map. For example, when outputting the well layout scheme with the highest fitness function value F in the final generation, it is necessary to overlay the well locations of the scheme with the sedimentary microfacies plane distribution map to perform intuitive geological rationality verification, thereby optimizing the distribution location of the reservoir well network.

[0043] In summary, this embodiment achieves automatic collaborative determination through a two-layer optimization framework using the geological penalty coefficient α and the weight vector W. The inner layer optimization involves using well location coordinates as the decision variable and the fitness function value F as the objective to perform well network optimization retrieval. The outer layer optimization involves using the geological penalty coefficient α and the weight vector W as decision variables and the true NPV of the scheme obtained from the inner layer optimization (i.e., the pure NPV value that does not consider the geological penalty term) as the objective function to automatically search for the optimal hyperparameter combination that can simultaneously maximize development benefits and geological rationality.

[0044] Furthermore, in the outer layer optimization, the range of values ​​for the weight vector W can be constrained by relative relationships based on sedimentology to ensure that the optimization results conform to relevant geological laws. For example, for production wells, the range of values ​​for W is: W(distributary channel) > W(main body of the mouth bar) > W(inner edge of the mouth bar) > W(sheet sand); for injection wells, the range of values ​​for W is: W(main body of the mouth bar) > W(inner edge of the mouth bar) > W(sheet sand), and W(distributary channel) takes the lower value.

[0045] In the normalized weighting system of this embodiment, the value range of the weight vector W is limited to the closed interval [0, 1], that is, the well placement weights corresponding to all microphases are constrained within the range of [0, 1]. Here, "high value" is defined as a value close to 1 (such as any value in the range of 0.8-1.0), representing a strong preference for well placement in this microphase, and the algorithm preferentially selects well locations in this region; "low value" is defined as a value close to 0 (such as any value in the range of 0-0.2), representing an extreme avoidance of well placement in this microphase, and the algorithm will actively avoid well placement in this region.

[0046] For the microfacies weighting rules of injection wells, the low value of W (distribution channel) is a constraint criterion determined based on sedimentological mechanisms. Distribution channel reservoirs have high permeability and exhibit positive rhythmic sedimentary characteristics. If injection wells are placed inside distribution channels, the injected water can easily migrate rapidly along high-permeability channels and surge unidirectionally towards production wells, causing development problems such as premature water breakthrough and single-layer water migration, resulting in a significant reduction in reservoir swept volume and oil recovery. Therefore, in the boundary constraints and population initialization process of outer layer optimization, the weight of the distribution channel corresponding to the injection well is forcibly limited to a very small positive number close to 0 (e.g., 0.01). This prevents the preferred distribution channel region for injection wells from being selected at the algorithmic level, ensuring that the optimization scheme conforms to the geological laws of reservoir development.

[0047] In other words, the reservoir well network optimization method in this embodiment mainly adopts a two-layer optimization approach to ensure a high degree of unity between geological logic and economic benefits. Furthermore, in this embodiment, geological rationality is the cornerstone of long-term economic value (NPV) in the two-layer optimization logic. The biggest problem with traditional single-layer black-box optimization in existing technologies is that it finds solutions with good short-term economic indicators but extremely unreasonable geological conditions. For example, a pseudo-optimal solution where injection wells directly enter high-permeability channels, leading to premature flooding, will inevitably perform poorly in long-term development, resulting in poor optimization effects. In this embodiment, by introducing two-layer optimization, the outer layer optimization is mainly responsible for automatically finding the geological constraint rules that best suit the current reservoir characteristics, i.e., automatically learning weights and penalty coefficients. The inner layer optimization, guided by these rules, finds the optimal well location. This nested architecture achieves a leap from "constraints given by human experience" to "adaptive evolutionary constraints of algorithms," ensuring that the final solution is a globally economically optimal solution guided by geological laws, resulting in better optimization effects.

[0048] Specifically, Figure 2 Medium red represents the distributary channel facies, the area with the best reservoir properties and the highest production well placement weight; orange / yellow represents the main body and inner edge of the mouth bar, with slightly lower properties and a sedimentary rhythm opposite to that of the distributary channel; light green / gray represents distal sandbars and sheet sands, with poorer properties, belonging to low-weight areas, and well placement will be automatically avoided in this type of area during the above algorithm. This is used to... Figure 2The planar distribution map of sedimentary microfacies provides a spatial "geological base map" for the algorithm, defining the boundaries of heterogeneity.

[0049] Accordingly, Figure 3 , Figure 4 and Figure 2 The displayed pattern is consistent. Figure 3 and Figure 4 This is primarily used to verify the accuracy of microfacies division, proving that the distribution of physical properties is strictly controlled by sedimentary facies, thus supporting the scientific validity of the weights in the formula in step S3 above. Warm colors (red / orange) represent high-value areas, i.e., high porosity and high permeability; cool colors (blue / green) represent low-value areas, i.e., low porosity and low permeability. Specifically, Figures 5-8 The pink grid points represent the selected grids in each partition, showing the logical affiliation of each microfacies zone in the numerical simulator, which is a direct manifestation of the "geological model preprocessing" in step S1 and serves as the index basis for the algorithm to query the microfacies weights.

[0050] Furthermore, Figure 9 The blue scatter plots represent the NPV (Net Promoter Value) distribution of all search individuals in each generation, i.e., the well placement scheme. The red solid line represents the global historical best return, indicating how the return of the currently optimal solution found by the algorithm changes with the number of iterations. The yellow star-shaped plots mark the globally optimal solution that was finally converged. Figure 9 The curve shows a rapid initial rise followed by a stable period at the optimal solution, demonstrating that the geological constraints successfully eliminated a large number of low-yield invalid regions, achieving the goal of "accelerated convergence".

[0051] Combination Figure 10 and Figure 11 As can be seen, the blue grid background map displays the three-dimensional structural morphology and permeability distribution of the reservoir; the gray / colored cylinders represent the locations of oil wells. Therefore, through the basic schemes in existing technologies, it is possible to... Figure 10 The well locations are arranged in a strictly equidistant geometric pattern, i.e., using the five-point method, without considering subsurface microfacies changes. However, through the optimization scheme of the reservoir well network optimization method in this embodiment, the well locations can be distributed irregularly, and the optimal well locations randomly found by the optimization algorithm can be optimized under geological constraints. This intuitively demonstrates the effect of the algorithm's transformation from "mathematical arrangement" to "mechanistic arrangement".

[0052] like Figure 12 As shown, the solid black line represents the trend of the recovery rate of the optimized well location scheme in this embodiment over time, such as the trend over 10 years, while the dashed red line represents the trend of the recovery rate of the basic well location in the prior art scheme. Comparing the two, the recovery rate after adopting the reservoir well network optimization method in this embodiment has been steadily increasing year by year, which is better than the scheme using single-layer calculation in the prior art.

[0053] Therefore, this embodiment incorporates specific geological constraints into the reservoir well network optimization method, making well location optimization more reasonable and in line with geological laws. Under these geological constraints, the optimization algorithm can find the optimal solution earlier and faster, rather than searching arbitrarily, thus avoiding wasting time on simulations and increasing computational costs. Correspondingly, this embodiment achieves a shift from manual to intelligent computation in the reservoir well network optimization method. In existing technologies, in single-layer optimization, geological weights and penalty coefficients must be manually preset by engineers. However, due to the vastly different characteristics of different reservoirs, engineers need to preset values ​​based on experience. If the set weights are biased, it will not only fail to accelerate optimization but also mislead the algorithm. However, through the breakthrough of dual-layer optimization in this embodiment, an outer layer optimization can be introduced, allowing the algorithm to adaptively "learn" the optimal geological rules. Furthermore, the outer layer algorithm can continuously iterate and verify what weight configuration can enable the inner layer to find the most productive well location the fastest, achieving the effect of transitioning from "manual experience-based parameter tuning" to "algorithm-automatic optimization rules." Therefore, the inner layer acts as the execution layer, responsible for finding the well location, while the outer layer acts as the learning layer, responsible for finding the rules. The outer layer will optimize based on the simulation results fed back by the inner layer, so as to automatically iterate and find the weight vector W and geological penalty coefficient α that can best guide the algorithm to quickly find the optimal solution.

[0054] For example, the following detailed description is based on a braided river delta reservoir example from an oilfield: First, the geological model is prepared by importing the geological model built in Petrel software into CMG numerical simulation software. The model contains four sedimentary microfacies partitions: underwater distributary channel (Sector1), main body of the mouth bar (Sector2), inner edge of the mouth bar (Sector3), and sheet sand (Sector4). Then, the center point coordinates of each grid and its corresponding microfacies partition code are exported from the CMG model to generate a microfacies lookup table.

[0055] Secondly, geological mechanism constraints are defined. Based on sedimentological knowledge, boundary constraints are set for the search space of the outer layer optimization weight vector W, namely, production well weight: W(distributary channel) > W(main body of the mouth bar) > W(inner edge of the mouth bar) > W(sheet sand); injection well weight: W(main body of the mouth bar) > W(inner edge of the mouth bar) > W(sheet sand), and W(distributary channel) takes the lower value; for example, the specific value range of each weight vector W is set as follows: For production wells, the value range of W (diversion channel) is (0.8, 1]; the value range of W (main body of the estuary dam) is (0.6, 0.8]; the value range of W (inner edge of the estuary dam) is [0.4, 0.6]; and the value range of W (sheet sand) is [0.1, 0.2]. For injection wells, the value range of W (main body of the estuary dam) is (0.7, 0.9]; the value range of W (inner edge of the estuary dam) is [0.5, 0.7]; the value range of W (sheet sand) is (0.1, 0.2]; and the value range of W (diversion channel) is [0.01, 0.1].

[0056] Next, optimization parameters were set, using particle swarm optimization as the inner optimizer, with a population size of 50, a maximum number of iterations of 50, an inertia weight ω of 0.4, an individual learning factor c1 of 2, a social learning factor c2 of 2, and a well network size of 10 injection wells and 10 production wells. A Bayesian optimizer was used for the outer layer.

[0057] Then, a two-layer optimization is performed. The outer optimizer generates a set of candidate (α, W) parameters; these parameters are input into the inner layer, and the genetic algorithm is started. The inner algorithm generates a well location plan and then calls the CMG simulator to calculate NPV, and calculates the NPV according to the formula F=NPV[1–α(Penalty)]. facies / Penalty base The fitness function value is calculated and the evolution is performed. After the inner optimization is completed, the actual NPV of the optimal solution is fed back to the outer optimizer. The outer optimizer aims to maximize the actual NPV and iteratively searches for the optimal (α,W) combination.

[0058] Finally, the results are output, that is, the optimal well location scheme is output after optimization.

[0059] Therefore, this embodiment is the first to deeply embed geological knowledge into the core of the intelligent optimization algorithm, namely the fitness function, in the form of quantifiable rules, namely the weight vector W and the penalty function. Through a two-layer optimization mechanism, it automatically learns the optimal geological rule parameters, realizing a paradigm shift from "black box optimization" to "mechanism-guided optimization".

[0060] Compared with the prior art, this embodiment has at least the following significant beneficial effects: To enhance the credibility and interpretability of the optimization results, this embodiment innovatively transforms qualitative geological knowledge into a computable weight vector W and a penalty function, and embeds them into the optimization algorithm. This overcomes the shortcomings of traditional "black box" optimization that is detached from geological mechanisms, making the final well network scheme not only optimal in NPV, but also more in line with underground geological laws. The decision-making basis is scientific, reliable, and easily accepted by geological engineers. Significantly improving optimization efficiency and saving huge computing resources, this embodiment pre-constrains the search space by geological rules, which can automatically avoid a large number of invalid or high-risk well placement schemes, greatly reducing the number of time-consuming numerical simulation calculations. Examples show that this method can reduce invalid simulation calculations by about 30%-50%, and significantly improve optimization efficiency. This embodiment provides a general technical framework that deeply integrates domain knowledge, such as sedimentology, with artificial intelligence optimization algorithms to form an advanced new paradigm of "knowledge-guided optimization". It is not only applicable to reservoir well network optimization, but can also provide a reference for other complex system optimization problems that require the integration of professional knowledge.

[0061] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for optimizing reservoir well patterns, characterized in that, Includes the following steps: S1. Preprocess the geological model and establish a microfacies lookup table; S2. Define decision variables and hyperparameters; S3. Automated screening and iterative evolution are performed based on the fitness function and the hyperparameters. S4. Output the well placement plan and conduct geological verification.

2. The reservoir well pattern optimization method according to claim 1, characterized in that, In step S1, the geological model is set to a three-dimensional geological model.

3. The reservoir well pattern optimization method according to claim 2, characterized in that, Step S1 includes: S1.

1. Import the three-dimensional geological model containing the distribution of sedimentary microfacies into numerical simulation software for preprocessing; S1.

2. The model mesh of each partition is attribute-encoded using the partitioning function of the numerical simulation software to form microfacies type codes, and a corresponding mapping relationship is established between the three-dimensional spatial coordinates (x, y, z) of the three-dimensional geological model and the microfacies type codes. S1.3 Extract the center coordinates of each model grid and the corresponding microfacies type code from the three-dimensional geological model as the microfacies partition data of that partition; S1.4 Summarize the microphase partition data of each partition to establish a microphase query table that can be queried in real time at the second level by the optimization algorithm.

4. The reservoir well pattern optimization method according to claim 1, characterized in that, Step S2 includes: S2.1 Using an intelligent optimization algorithm, the planar coordinates of the planned well are encoded as the decision variables; S2.2 Define at least two sets of hyperparameters to be optimized, including the global geological penalty coefficient α and the microfacies well type weight vector W.

5. The reservoir well pattern optimization method according to claim 4, characterized in that, The geological penalty coefficient α is used to adjust the intensity of geological laws' intervention in the optimization calculation process; Each element in the microfacies well type weight vector W represents the degree of preference for well placement of a specific sedimentary microfacies relative to a specific well type.

6. The reservoir well pattern optimization method according to claim 1, characterized in that, The fitness function in step S3 is: F = NPV[1 – α(Penalty)] facies / Penalty base )]; Where F is the fitness function value; NPV stands for Net Present Value in the numerical simulation scheme. α is the geological penalty coefficient; Penalty facies This is the geological penalty value; Penalty base This is the baseline penalty value.

7. The reservoir well pattern optimization method according to claim 6, characterized in that, The method for calculating the geological penalty value is as follows: The microfacies lookup table is queried based on the coordinates of all planned wells to obtain the microfacies type of each planned well location; Combining the micro-phase well type weight vector W in the currently mentioned hyperparameters, calculate the (1-W) of all the planned wells. i The sum of these values ​​is the geological penalty value. Among them, W i The value of the micro-phase well type weight vector W corresponding to the i-th planned well.

8. The reservoir well pattern optimization method according to claim 6, characterized in that, The baseline penalty value is a constant, and / or the baseline penalty value is the average of the geological penalty values ​​of the initial population generated by the plane coordinate encoding initialization of the planned well.

9. The reservoir well pattern optimization method according to claim 1, characterized in that, The automated screening in step S3 involves comparing the fitness function values ​​calculated based on the fitness function, retaining excellent individuals, eliminating inferior individuals, and generating a new population through crossover and mutation, thereby iterating and evolving repeatedly. And / or, the condition for iterative evolution is that the convergence condition is met or the number of iterations reaches the target value.

10. The reservoir well pattern optimization method according to claim 1, characterized in that, In step S4, the geological verification is completed by overlaying the well locations of each planned well in the well layout scheme with the sedimentary microfacies plane distribution map.

Images