Well pattern and well location optimization method based on flow field area and balanced production evaluation

Abstract

The present invention discloses a well network and well location optimization method based on flow field area and balanced utilization evaluation, which relates to the field of petroleum engineering technology. The present invention is based on streamline numerical simulation, by constructing a water drive oil reservoir streamline numerical simulation well location optimization model, clarifying the production system, optimizing variables, constraint variables and flight time balance-based objective functions, and constructing an initial population based on Latin hypercube sampling, building a water drive oil reservoir streamline numerical simulation well location optimization intelligent agent model, based on the Q-Learning enhanced marine predator algorithm, using the water drive oil reservoir streamline numerical simulation well location optimization intelligent agent model for simulation, obtaining streamline data and derivative data obtained by the water drive oil reservoir streamline numerical simulation well location optimization intelligent agent model simulation, and performing well network and well location optimization. The method of the present invention effectively improves the speed and optimization effect of well location optimization, significantly reduces the evaluation cycle of the well location layout plan, and is conducive to improving the comprehensive economic benefits of oil field development.

Description

Technical Field

[0001] The present invention relates to the technical field of petroleum engineering, and in particular to a well pattern and well location optimization method based on flow field area and balanced production evaluation. Background Art

[0002] During oil and gas field exploration and development, the spatial heterogeneity of reservoirs and fluids has a crucial impact on development efficiency. Constrained by multiple factors, including reservoir permeability distribution, fluid viscosity differences, and fault structure, well production rates vary significantly across different spatial locations within a reservoir, with single-well production fluctuations in local areas reaching orders of magnitude. This strong heterogeneity makes the scientific formulation of well pattern deployment under complex geological conditions and the maximization of recovery through spatially optimized well placement a core challenge in improving the economic efficiency of oilfield development.

[0003] At present, there are two main methods for well location design: one is the traditional design method based on reservoir engineering experience. This method relies on geological modeling and numerical simulation, and by screening reservoir physical parameters, it prioritizes well placement in areas with high oil saturation and structural highs. Although this method can avoid the risk of edge and bottom water coning, its decision-making mechanism has significant flaws: First, the generation of well placement plans relies on engineers' subjective interpretation of static geological models, with a high weight of experience, and randomness in the exploration of the plan space. Second, when analyzing multi-source heterogeneous data coupling, the efficiency of manual data processing is limited, and the final well location often deviates from the theoretical optimal position because the cumulative error of the ambiguity of key parameters exceeds the allowable threshold.

[0004] Another approach is an automated well location optimization method based on optimization theory. Its technical process includes three core steps: establishing a multi-constraint optimization model with economic net present value (NPV) or stage recovery as the objective function; using intelligent optimization algorithms such as differential evolution to drive iterative updates of well location coordinate parameters; and quantitatively evaluating the production dynamic response of each candidate solution through a reservoir numerical simulator. Although this method reduces human intervention by introducing an automated optimization mechanism, it faces severe computational efficiency bottlenecks, especially when applied to reservoir models with a million-grid scale. A single full-cycle numerical simulation is time-consuming, and the optimization algorithm typically requires evaluating a large number of candidate solutions before convergence, resulting in a long overall optimization cycle. More seriously, some iterative solutions become invalid solutions due to violations of inter-well interference suppression criteria or deviations from the laws of seepage dynamics, resulting in a significant waste of computing resources and severely limiting the engineering applicability of this method.

[0005] Therefore, it is urgent to propose a well network and well location optimization method based on flow field area and balanced production evaluation to improve the well network and well location optimization efficiency, shorten the well network and well location optimization cycle, reduce the time and money spent on oil field well location deployment, and improve the comprehensive economic benefits of oil field development. Summary of the Invention

[0006] The present invention aims to overcome the defects of the existing well location design method. carry A well pattern and well location optimization method based on flow field area and balanced production evaluation is provided. Streamline numerical simulation is used to obtain streamline-derived information for the streamline numerical simulation and well location optimization model of water-flooding reservoirs. A streamline characteristic objective function based on flight time balance is constructed. The meta-heuristic QQLMPA algorithm is used for well location optimization. During the well pattern and well location optimization process, a better well location distribution scheme can be obtained with fewer numerical simulations, significantly shortening the evaluation cycle of the well location scheme.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The well pattern and well location optimization method based on flow field area and balanced production evaluation includes the following steps:

[0009] Step 1: Build a well location optimization model for numerical simulation of water-flooding reservoir streamlines, clarify the production system, determine the optimization variables, constraint variables, and objective function, and construct the initial population based on Latin hypercube sampling;

[0010] Step 2: Based on the Q-Learning enhanced marine predator algorithm, an intelligent agent model for well location optimization in numerical simulation of waterflooding reservoir streamlines is constructed;

[0011] Step 3: Use the intelligent agent-based model for optimizing well locations and numerical simulation of streamlines in water-flooding reservoirs to perform simulations, obtain and process streamline data and derived data obtained by the intelligent agent-based model for optimizing well locations and numerical simulation of streamlines in water-flooding reservoirs, and establish a streamline data set;

[0012] Step 4: Combine historical data from the elite data retention area, update the Q-table through Q-Learning to optimize the phase selection strategy, and use QOL quasi-adversarial learning to generate candidate solutions, expand the search range, and use the updated strategy to guide MPA to perform next-generation well location optimization;

[0013] Step 5: Based on the reservoir streamline numerical simulator, the water drive reservoir streamline numerical simulation well location optimization intelligent agent model continuously interacts with the reservoir numerical simulation to evaluate the fitness of the new population, select elite samples and store them in the elite data retention area;

[0014] Step 6: Repeat steps 4 and 5 until the preset maximum number of iterations is reached to determine the optimal well location arrangement.

[0015] Preferably, the step 1 comprises the following steps:

[0016] Step 1.1, construct a well location optimization model for numerical simulation of water drive reservoir flow lines;

[0017] Step 1.2: The two-dimensional coordinates of the injection well and production well are used as optimization variables; the constraint variables are the minimum well spacing constraint and the well location non-repetition constraint; the objective function is the streamline numerical simulation well location optimization objective function , the goal is to maximize the cumulative oil production under the constraints of well spacing and well location non-repetition;

[0018] Step 1.3, construct the initial population based on Latin hypercube sampling;

[0019] Latin hypercube sampling is used to sample from a multivariate parameter distribution, and the number of injection well location plans to be generated is Determine the number of samples to be drawn and divide the (0,1) interval into Segments are randomly sampled in each segment, the sampled samples are mapped to standard normal distribution samples through the inverse function of the standard normal distribution and randomly sorted to obtain the initial population.

[0020] Preferably, the water drive reservoir streamline numerical simulation well location optimization model is:

[0021] (1)

[0022] The constraints are:

[0023] ; ;

[0024] ; ;

[0025] Where, is the well location coordinate set; Optimize the objective function for well location in streamline numerical simulation; is the minimum function; is the streamline displacement potential coefficient; is the coefficient of variation of streamline flight time; is the penalty term between injection-production well pairs; It is the internal penalty item for injection and production wells; and are the upper and lower bounds of the boundary constraints of the well location coordinate set respectively; is the well spacing constraint; 、 All are numbered; 、 All are single well location coordinates; is the number of injection wells, is the number of producing wells.

[0026] Preferably, the step 2 comprises the following steps:

[0027] Step 2.1: Determine the well location coordinates of the expected production cycle based on the particle coordinate values ​​in the initial population and obtain the well location coordinate set :

[0028] (2);

[0029] Step 2.2, set the simulation time step according to the time step of the expected production cycle and perform streamline numerical simulation;

[0030] Step 2.3: Determine the streamline displacement potential coefficient at the current time step based on the proportion of the number of grids that the streamline passes through to the number of grids in the water flooding reservoir streamline numerical simulation well location optimization model. ;

[0031] Step 2.4, based on the coefficient of variation of streamline flight time at the current time step Set the balance degree of each injection and production well on the average flight time;

[0032] Step 2.5: Penalty term between injection and production wells at the current time step Set the positive difference between the average flight time of each injection-production well pair and the expected production cycle;

[0033] Step 2.6, based on the internal penalty item of the injection and production well at the current time step Set the average value of the positive difference between the flight time of the streamlines contained in each injection-production well pair and the expected production period;

[0034] Step 2.7, determine the initial parameters of the learning environment of the intelligent agent model for well location optimization in the numerical simulation of water flooding reservoir streamlines.

[0035] Preferably, the streamline displacement potential coefficient at the current time step is for:

[0036] (3)

[0037] Where, is the total number of grids in the well location optimization model for the numerical simulation of the water drive reservoir streamline that the streamline passes through; The total number of grids in the well location optimization model for numerical simulation of waterflood reservoir flow lines;

[0038] The coefficient of variation of streamline flight time at the current time step for:

[0039] (4)

[0040] Where, is the sample standard deviation of the flight time of the injection and production wells; is the average flight time of the injection and production wells; is the number of injection-production well pairs; For the Flight time of each injection-production well pair;

[0041] The penalty term between injection and production wells at the current time step for:

[0042] (5)

[0043] Where, is the number of injection-production well pairs whose flight time is less than the expected production cycle, ; is the penalty intensity power series, ;

[0044] The internal penalty item of the injection and production well at the current time step for:

[0045] (6)

[0046] Where, is the sequence number of the streamline, For the the number of streamlines within the group; is the streamline water inrush index, ,in, For the No. The flight time at the end of the root streamline, Estimated production cycle.

[0047] Preferably, the initial parameters of the learning environment include population size, maximum number of iterations, particle dimension, number of Q-Learning states, number of Q-Learning actions, Q-Learning training rounds, step size coefficient, discount factor, Levy random factor, Brownian random factor, and adaptive adjustment factor.

[0048] Preferably, the step 3 comprises the following steps:

[0049] Step 3.1, obtaining a streamline data file obtained by simulating a water drive reservoir streamline numerical simulation well location optimization intelligent agent model, decoding the streamline data file, and obtaining streamline data;

[0050] Obtain streamline data files for each time step during the simulation of the intelligent agent model for well location optimization in numerical simulation of waterflooding reservoir streamlines. Divide the storage content of the streamline data files into string type data and numeric type data. Obtain string type data through ASCII decoding. Directly obtain numeric type data through binary decoding according to IEEE rules. Obtain streamline attributes for each time step from the streamline data files, including the position, saturation, and flight time of each streamline in the flow field.

[0051] Step 3.2, streamline data preprocessing;

[0052] The streamline data preprocessing includes deleting characteristic indicating coordinates in the streamline data and averaging streamline attribute data of repeated streamline points;

[0053] Step 3.3, obtain other properties of the streamline field;

[0054] Calculate derived data based on streamline data, obtain the grid pressure and grid number of the grid through which the streamline passes, establish a pressure mapping relationship, obtain the pressure at the current streamline point, and then use the interpolation algorithm to determine the pressure at any position in the streamline field; determine the streamline injection and production well pair number based on the streamline starting and ending well numbers, and calculate the streamline water inrush index based on the streamline flight time;

[0055] According to Darcy's law and grid pressure, the fluid flow velocity within any grid interface is determined as:

[0056] (7)

[0057] Where, For Grid The average velocity of the internal fluid; is the grid permeability; is the fluid viscosity; For the The grid pressure of each grid; For the The grid pressure of each grid; is the grid length;

[0058] In step 3.4, streamlines are grouped according to the injection-production well pair numbers, abnormal boundary streamlines are eliminated, and a streamline dataset is constructed.

[0059] Preferably, the step 4 comprises the following steps:

[0060] Step 4.1: Set the state space based on the initialized Q-Learning parameters and action space ; Initialize the Q table to a zero matrix ; Initialize reward table for:

[0061] (8)

[0062] Where, is the state space, ,in, 、 、 Corresponding to the three search phases of the marine predator algorithm MPA, is the Brownian phase, used for local development, is the Levy-Brownian phase, used for mixed exploration; is the Levy phase, used for global exploration; is the action space, , is the switching strategy of the Brownian phase, is the switching strategy of the Levy-Brownian phase, is the switching strategy of Levy phase;

[0063] Step 4.2, based on the current status , using a greedy strategy to select actions , after executing the action, switch the search phase and update the individual position:

[0064] (9)

[0065] Where, For the Time step The location of an individual, For the Time step the location of each individual; is the step length coefficient; is a random number factor between (0,1); For the Individual The elite solution used in the current iteration; is the Brownian random factor; is the adaptive adjustment factor;

[0066] Step 4.3: Determine the immediate reward based on the fitness difference before and after optimization , and update the Q table using the Bellman equation:

[0067] (10)

[0068] in,

[0069] (11)

[0070] Where, is the learning decay rate function; is the discount factor; For the The immediate reward of time steps; For the The state of the time step; For the corresponding status the action chosen; is the time step, is the maximum number of iterations;

[0071] Step 4.4, based on the quasi-adversarial learning mechanism, expand the search range and Construct its quasi-opposite solution :

[0072] (12)

[0073] Where, is a random number function;

[0074] Comparison individuals Its opposite solution The fitness of , retains the optimal solution;

[0075] Step 4.5: Determine whether the disturbance mechanism has been entered based on the fish aggregation mechanism :

[0076] (13)

[0077] Where, and are two groups of random individual positions; is the upper boundary position vector of the well coordinates, is the lower boundary position vector of the well coordinates;

[0078] Step 4.6: Store the individuals with the best fitness in the current iteration into the elite reserve area and use them as the next generation update. Vector, completing the search iteration of the current iteration round.

[0079] Preferably, the step 5 comprises the following steps:

[0080] Step 5.1, a single simulation of the individuals in the current population is performed based on the reservoir streamline numerical simulator. The water flooding reservoir streamline numerical simulation well location optimization intelligent agent model processes the streamline information and calculates the fitness of each individual in the current population in combination with the objective function.

[0081] Step 5.2: For each individual in the current population, the well location coordinates are checked for legitimacy to ensure that they meet the distance approximation and repeatability filtering requirements. If not, the nearest neighbor feasible coordinate search function is called to make corrections.

[0082] Step 5.3, compare the current optimal fitness with the previous generation fitness, retain the better individual, update the position coordinates of the best individual, and record the current optimal fitness in the convergence curve.

[0083] The beneficial technical effects brought about by the present invention are:

[0084] Among the well network and well location optimization methods, the traditional well network and well location optimization method uses cumulative oil production (COP) or net present value (NPV) as the objective function. When using a reservoir numerical simulator to complete the full-time simulation of the expected production cycle, it takes a long time and the objective function has poor convergence, which cannot meet the time cost requirements in the actual optimization process.

[0085] The present invention proposes a well pattern and well location optimization method based on flow field area and balanced production evaluation, which overcomes the shortcomings of traditional well pattern and well location optimization methods. First, by constructing a streamline numerical simulation well location optimization model for water-flooded reservoirs with the goals of flight time balance, streamline coverage maximization, and water inrush penalty minimization, the well location optimization control is closer to the dynamic response of the reservoir, thereby improving the physical consistency and engineering operability of the well pattern well location optimization model. Second, by utilizing streamline simulation, the method of the present invention only needs to complete the simulation of the first time step of the expected production cycle to quickly obtain the injection-production connectivity and its changing law, effectively reducing the high computational cost brought by traditional fully implicit numerical simulation in the optimization process and significantly improving the optimization efficiency of the well pattern well location. Third, by introducing the Q-Learning reinforcement learning mechanism, the method of the present invention dynamically learns the advantages and disadvantages of each phase in the search process and guides strategy switching with the Q value, thereby improving the algorithm's adaptability and global search performance in multi-peak optimization problems. At the same time, the method combines the quasi-adversarial learning strategy to expand the search space and avoid local convergence, and integrates the FADs aggregation disturbance mechanism to simulate ecological group intelligence, thereby enhancing population diversity. Finally, by constructing an elite individual retention mechanism, the method achieves continuous tracking and inheritance of high-quality solutions, making the optimization process more robust and reliable.

[0086] It can be seen that the method of the present invention significantly reduces the optimization cycle of the well location layout plan, and the optimized well location distribution plan is better, providing a basis for improving the comprehensive economic benefits of oil field development. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 The present invention is a flow chart of a well pattern and well location optimization method based on flow field area and balanced production evaluation.

[0088] Figure 2This is the permeability field map of the reservoir model in the embodiment of the present invention.

[0089] Figure 3 This is the porosity field map of the reservoir model in the embodiment of the present invention.

[0090] Figure 4 This is an optimization convergence diagram with cumulative oil production COP as the objective function in an embodiment of the present invention.

[0091] Figure 5 The objective function for optimizing well location using streamline numerical simulation in the embodiment of the present invention is Optimization convergence diagram.

[0092] Figure 6 3 is a comparison chart of the method of the present invention and the traditional well pattern and well location optimization method in this embodiment.

[0093] Figure 7 This is a coordinate diagram of optimized well locations with cumulative oil production (COP) as the objective function in an embodiment of the present invention.

[0094] Figure 8 This is an oil saturation diagram with cumulative oil production (COP) as the objective function in an embodiment of the present invention.

[0095] Figure 9 This is a streamline distribution diagram with cumulative oil production COP as the objective function in an embodiment of the present invention.

[0096] Figure 10 The objective function for optimizing well location by numerical simulation of streamlines in the embodiment of the present invention is Optimized well location coordinate diagram for the objective function.

[0097] Figure 11 The objective function for optimizing well location by numerical simulation of streamlines in the embodiment of the present invention is Oil saturation diagram of the objective function.

[0098] Figure 12 The objective function for optimizing well location by numerical simulation of streamlines in the embodiment of the present invention is Oil saturation diagram of the objective function.

[0099] In the figure, P1, P2, P3, P4, P5, P6, P7, P8, and P9 are production wells, and INJ1, INJ2, INJ3, and INJ4 are water injection wells. DETAILED DESCRIPTION

[0100] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0101] Example 1

[0102] This embodiment proposes a well pattern and well location optimization method based on flow field area and balanced production evaluation, such as Figure 1 As shown, it specifically includes the following sub-steps:

[0103] Step 1: Build a well location optimization model for numerical simulation of waterflooding reservoir streamlines, clarify the production system, determine the optimization variables, constraint variables, and objective function, and construct the initial population based on Latin hypercube sampling, which includes the following sub-steps:

[0104] Step 1.1: Construct a well location optimization model for numerical simulation of water drive reservoir streamlines.

[0105] The water drive reservoir streamline numerical simulation well location optimization model is:

[0106] (1)

[0107] The constraints are:

[0108] ; ;

[0109] ; ;

[0110] Where, is the well location coordinate set; Optimize the objective function for well location in streamline numerical simulation; is the minimum function; is the streamline displacement potential coefficient; is the coefficient of variation of streamline flight time; is the penalty term between injection-production well pairs; It is the internal penalty item for injection and production wells; and are the upper and lower bounds of the boundary constraints of the well location coordinate set respectively; is the well spacing constraint; 、 All are numbered; 、 All are single well location coordinates; is the number of injection wells, is the number of producing wells.

[0111] Step 1.2: The two-dimensional coordinates of the injection well and production well are used as optimization variables; the constraint variables are the minimum well spacing constraint and the well location non-repetition constraint; the objective function is the streamline numerical simulation well location optimization objective function The goal is to maximize the cumulative oil production under the constraints of well spacing and well non-repetition, ensure the balance of flight time between each injection and production well pair, maximize the streamline coverage, and minimize the occurrence of water inrush to maximize the cumulative oil production.

[0112] Step 1.3: Construct the initial population based on Latin hypercube sampling.

[0113] Latin hypercube sampling is used to sample from a multivariate parameter distribution, and the number of injection well location plans to be generated is Determine the number of injection well location plans that need to be generated , according to the number of water injection well location plans to be generated Determine the number of samples to be drawn and divide the (0,1) interval into Segments are randomly sampled in each segment, the sampled samples are mapped to standard normal distribution samples through the inverse function of the standard normal distribution and randomly sorted to obtain the initial population.

[0114] Step 2: Based on the Q-Learning enhanced marine predator algorithm (QQLMPA), an intelligent agent model for well location optimization in water-flooding reservoir streamline numerical simulation is constructed. The Q-Learning enhanced marine predator algorithm aims to improve the optimization performance of the traditional marine predator algorithm (MPA). The introduction of Q-Learning enables MPA to utilize the information generated in historical iterations, adaptively select the optimal strategy (Lévy motion, Brownian motion, or a combination of the two), balance exploration and development, and accelerate convergence. The introduction of quasi-opponent learning (QOL) can generate quasi-opponent solutions to the current solution, expand the search space, increase population diversity, and thus reduce the risk of falling into local optimality. QQLMPA significantly improves the convergence speed and global search capability of MPA through the dual improvements of quantum behavior improvement QL and quasi-opponent learning QOL. It specifically includes the following sub-steps:

[0115] Step 2.1: Determine the well location coordinates of the expected production cycle based on the particle coordinate values ​​in the initial population and obtain the well location coordinate set :

[0116] (2).

[0117] Step 2.2: Based on the characteristics of the streamline numerical simulation method, the attribute information of the streamline can be obtained after the first time step of the numerical simulation. Therefore, the simulation time step is set according to the time step of the expected production cycle to perform the streamline numerical simulation.

[0118] Step 2.3: Determine the streamline displacement potential coefficient at the current time step based on the proportion of the number of grids that the streamline passes through to the number of grids in the water flooding reservoir streamline numerical simulation well location optimization model. for:

[0119] (3)

[0120] Where, The total number of grids in the numerical simulation well location optimization model of the water drive reservoir streamline that the streamline passes through is determined by extracting and removing duplicate grid numbers of all the streamlines; is the total number of grids in the well location optimization model for numerical simulation of water drive reservoir streamlines, and is determined according to the model parameters of the well location optimization model for numerical simulation of water drive reservoir streamlines.

[0121] Step 2.4, based on the coefficient of variation of streamline flight time at the current time step Set the balance degree of each injection and production well on the average flight time; the coefficient of variation of the streamline flight time at the current time step for:

[0122] (4)

[0123] Where, is the sample standard deviation of the flight time of the injection and production wells; is the average flight time of the injection and production wells; is the number of injection-production well pairs; For the The flight time of an injection-production well pair is used to represent the average flight time of all streamlines contained in the injection-production well pair.

[0124] Step 2.5: Penalty term between injection and production wells at the current time step Set the positive difference between the average flight time of each injection-production well pair and the expected production cycle; the penalty term between the injection-production well pairs at the current time step for:

[0125] (5)

[0126] Where, is the number of injection-production well pairs whose flight time is less than the expected production cycle, ; is the penalty intensity power series, .

[0127] Step 2.6, based on the internal penalty item of the injection and production well at the current time step Set the average value of the positive difference between the flight time of the streamlines contained in each injection-production well pair and the expected production cycle; the penalty item within the injection-production well pair at the current time step is for:

[0128] (6)

[0129] Where, is the sequence number of the streamline, For the the number of streamlines within the group; is the streamline water inrush index, ,in, For the No. The flight time at the end of the root streamline, Estimated production cycle.

[0130] Step 2.7, determining the initial parameters of the learning environment of the intelligent agent model for well location optimization in numerical simulation of waterflooding reservoir streamlines;

[0131] The initial parameters of the learning environment include the population size , maximum number of iterations , particle dimension , the number of states of Q-Learning , the number of Q-Learning actions , Q-Learning training rounds , step coefficient , discount factor , Levy random factor , Brownian random factor , adaptive adjustment factor .

[0132] Step 3, using the intelligent agent-based model for optimizing well locations for numerical simulation of streamlines in water-flooding reservoirs to perform simulations, obtains and processes streamline data and derived data obtained by the intelligent agent-based model for optimizing well locations for numerical simulation of streamlines in water-flooding reservoirs, and establishes a streamline dataset, including the following sub-steps:

[0133] Step 3.1, obtaining a streamline data file obtained by simulating a water drive reservoir streamline numerical simulation well location optimization intelligent agent model, decoding the streamline data file, and obtaining streamline data;

[0134] During each simulation process, the streamline data files of each time step of the water drive reservoir streamline numerical simulation well location optimization intelligent agent model are obtained, and the storage content of the streamline data files is divided into string type and value type. Among them, the string type data is obtained through ASCII decoding, and the value type data is directly decoded from the binary according to the specific value type according to the IEEE rules. The streamline properties of the time step are obtained from the streamline data file, including the position, saturation, flight time and other related streamline properties of each streamline in the flow field.

[0135] Step 3.2, streamline data preprocessing;

[0136] The streamline data preprocessing includes deleting the characteristic indicating coordinates in the streamline data and averaging the streamline attribute data of repeated streamline points. The characteristic indicating coordinates in the streamline data are used to achieve alignment in the relevant data dimensions. By averaging the streamline attribute data of repeated streamline points, repeated calculations and deletion of redundant repeated streamline points can be effectively avoided, thereby ensuring that each streamline point data is unique.

[0137] Step 3.3, obtain other properties of the streamline field;

[0138] Derivative data is calculated based on streamline data. The grid pressure and grid number of the grid through which the streamline passes are obtained. A pressure mapping relationship is established to obtain the pressure at the current streamline point. The pressure at any position in the streamline field is then determined using an interpolation algorithm. The streamline injection-production well pair number is determined based on the streamline start and end well numbers, and the streamline water inrush index is calculated based on the streamline flight time.

[0139] According to Darcy's law and grid pressure, the fluid flow velocity within any grid interface is determined as:

[0140] (7)

[0141] Where, For Grid The average velocity of the internal fluid; is the grid permeability, in units of ; is the fluid viscosity, in units of ; For the The grid pressure of the grid, in units of ; For the The grid pressure of the grid, in units of ; is the grid length in units of .

[0142] In step 3.4, all streamlines are divided into groups corresponding to their injection-production well pair numbers and abnormal boundary streamlines are removed to ensure that the streamline data is not contaminated, and a streamline dataset is constructed.

[0143] Step 4: Combine historical data from the elite data reserve, update the Q-table through Q-Learning to optimize the phase selection strategy, and use quasi-adversarial learning (QOL) to generate candidate solutions, expand the search scope, and use the updated strategy to guide MPA to perform next-generation well location optimization. This includes the following sub-steps:

[0144] Step 4.1: Set the state space based on the initialized Q-Learning parameters and action space ; Initialize the Q table to a zero matrix ; Initialize reward table for:

[0145] (8)

[0146] Where, is the state space, ,in, 、 、 Corresponding to the three search phases of the marine predator algorithm MPA, is the Brownian phase, used for local development, is the Levy-Brownian phase, used for mixed exploration; is the Levy phase, used for global exploration; is the action space, , is the switching strategy of the Brownian phase, is the switching strategy of the Levy-Brownian phase, is the switching strategy of the Levy phase.

[0147] Step 4.2, based on the current status , using a greedy strategy to select actions , after executing the action, switch the search phase and update the individual position:

[0148] (9)

[0149] Where, For the Time step The location of an individual, For the Time step the location of each individual; is the step length coefficient; is a random number factor between (0,1); For the Individual The elite solution used in the current iteration; is the Brownian random factor; is the adaptive adjustment factor.

[0150] Step 4.3: Determine the immediate reward based on the fitness difference before and after optimization , and update the Q table using the Bellman equation:

[0151] (10)

[0152] in,

[0153] (11)

[0154] Where, is the learning decay rate function; is the discount factor; For the The immediate reward of time steps; For the The state of the time step; For the corresponding status the action chosen; is the time step, is the maximum number of iterations.

[0155] Step 4.4, based on the quasi-adversarial learning mechanism, expand the search range and Construct its quasi-opposite solution :

[0156] (12)

[0157] Where, is a random number function.

[0158] Comparison individuals Its opposite solution fitness, and retain the optimal solution.

[0159] Step 4.5: Determine whether the disturbance mechanism has been entered based on the fish aggregating device. :

[0160] (13)

[0161] Where, and are two groups of random individual positions; is the upper boundary position vector of the well coordinates, is the lower boundary position vector of the well coordinates.

[0162] Step 4.6: Store the individuals with the best fitness in the current iteration into the elite reserve area and use them as the next generation update. Vector, completing the search iteration of the current iteration round.

[0163] Step 5: Based on the reservoir streamline numerical simulator, the water drive reservoir streamline numerical simulation well location optimization intelligent agent model continuously interacts with the reservoir numerical simulation to evaluate the fitness of the new population, select elite samples and store them in the elite data retention area. This includes the following sub-steps:

[0164] Step 5.1, perform a single simulation of the individuals in the current population based on the reservoir streamline numerical simulator, process the streamline information through the water drive reservoir streamline numerical simulation well location optimization intelligent agent model and calculate the fitness of each individual in the current population in combination with the objective function.

[0165] In step 5.2, for each individual in the current population, the well location coordinates will be judged for legitimacy to ensure that they meet the distance approximation and repeatability filtering requirements. If they do not meet the requirements, the nearest neighbor feasible coordinate search function will be called for correction.

[0166] In step 5.3, compare the current optimal fitness with the fitness of the previous generation, retain the better individuals, update the position coordinates of the top predator, that is, update the position coordinates of the best individual, and record the current optimal fitness in the convergence curve.

[0167] Step 6: Repeat steps 4 and 5 until the preset maximum number of iterations is reached to determine the optimal well location arrangement.

[0168] Example 2

[0169] Taking the oil reservoir in a certain study area as an example, the well pattern and well location optimization method based on flow field area and balanced production evaluation described in Example 1 is used to optimize the well location of the water drive oil reservoir in the study area. In this example, the reservoir model adopts a two-dimensional reservoir model, and the grid system of the reservoir model is 60 220 1. The grid size is 15m 5m 5m, the permeability field of the reservoir model is as follows Figure 2 As shown in Figure 2, the porosity field of the reservoir model is as follows: Figure 3 As shown in FIG4 , the well location coordinates that need to be optimized for this oil reservoir include 9 production wells and 4 water injection wells, wherein the production wells are numbered P1 to P9 and the water injection wells are numbered INJ1 to INJ4. The production time of the oil reservoir is 1000 days, which is divided into 20 time steps, each of which is 50 days. However, when the well pattern and well location optimization method based on flow field area and balanced production evaluation proposed in Example 1 is used to optimize the well location of a water drive oil reservoir, only one time step needs to be selected. The well location spacing constraint is based on the grid distance, and the spacing between adjacent well locations is not less than 10 m.

[0170] The well pattern and well location optimization method based on flow field area and balanced production evaluation proposed in Example 1 is used to optimize the well pattern and well location of the oil reservoir, including the following steps:

[0171] Step 1: Build a well location optimization model for numerical simulation of water-flooding reservoir streamlines, clarify the production system, determine the optimization variables, constraint variables and objective function, and construct the initial population based on Latin hypercube sampling.

[0172] Step 2: Based on the Q-Learning enhanced marine predator algorithm, an intelligent agent model for well location optimization in numerical simulation of waterflooding reservoir streamlines is constructed.

[0173] Step 3: Use the intelligent agent model for numerical simulation of water-flooding reservoir streamlines and well location optimization to perform simulation, obtain and process the streamline data and derivative data obtained by the intelligent agent model for numerical simulation of water-flooding reservoir streamlines and well location optimization, and establish a streamline data set grouped by injection-production well pairs.

[0174] In step 4, the phase selection strategy is optimized by updating the Q-table through Q-Learning based on the historical data in the elite data retention area. Quasi-adversarial learning is used to generate candidate solutions, expand the search scope, and use the updated strategy to guide MPA to perform the next generation of well location optimization.

[0175] Step 5: Based on the reservoir streamline numerical simulator, the water drive reservoir streamline numerical simulation well location optimization intelligent agent model continuously interacts with the reservoir numerical simulation to evaluate the fitness of the new population, screen the elite samples and store them in the elite data retention area.

[0176] Step 6: Repeat steps 4 and 5 until the preset maximum number of iterations is reached to determine the optimal well location arrangement.

[0177] In order to verify the superiority of the streamline numerical simulation well location optimization objective function in the method of the present invention compared with the traditional well pattern well location optimization method, the well pattern well location optimization was carried out using the same well location optimization model as the present invention with the cumulative oil production COP as the objective function, and the convergence status within a total of 20 time steps in the entire expected production cycle was obtained, as shown in the figure. Figure 4 shown. Figure 5 The figure is the convergence diagram of the optimization method of the present invention. The cumulative oil production and optimization time of the well position obtained by optimizing with the cumulative oil production COP as the objective function are compared with the cumulative oil production and optimization time of the well position obtained by optimizing with the streamline numerical simulation well position optimization objective function proposed by the present invention as the objective function. Figure 6 As shown, it is found that the method of the present invention is better in terms of time consumption and oil production than the case where the cumulative oil production COP is used as the objective function;

[0178] Then compare the optimal well location map, oil saturation map and streamline distribution map obtained by using the cumulative oil production COP as the objective function and the streamline numerical simulation well location optimization objective function as the objective function. Figures 7 to 12 As shown, Figure 7This is a coordinate diagram of optimized well locations with cumulative oil production (COP) as the objective function in an embodiment of the present invention. Figure 8 This is an oil saturation diagram with cumulative oil production COP as the objective function in the embodiment of the present invention. Figure 9 is a streamline distribution diagram with cumulative oil production COP as the objective function in an embodiment of the present invention, Figure 10 The objective function for optimizing well location by numerical simulation of streamlines in the embodiment of the present invention is is the optimized well location coordinate diagram of the objective function, Figure 11 The objective function for optimizing well location by numerical simulation of streamlines in the embodiment of the present invention is is the oil saturation diagram of the objective function, Figure 12 The objective function for optimizing well location by numerical simulation of streamlines in the embodiment of the present invention is Oil saturation diagram of the objective function.

[0179] This further verifies that the method of the present invention can better formulate well location distribution plans compared to the traditional well location optimization method with cumulative oil production COP as the objective function, significantly reduces the evaluation cycle of the well location layout plan, and the well location distribution plan has a better effect.

[0180] In summary, the method of the present invention can obtain a better well location distribution plan in a shorter optimization time, significantly reducing the optimization cycle of the well location layout plan, making the optimization process more robust and reliable, and helping to improve the comprehensive economic benefits of oilfield development.

[0181] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the scope of protection of the present invention.

Claims

1. A well pattern and well location optimization method based on flow field area and balanced production evaluation, characterized in that: The following steps are involved: Step 1: Build a well location optimization model for numerical simulation of water-flooding reservoir streamlines, clarify the production system, determine the optimization variables, constraint variables, and objective function, and construct the initial population based on Latin hypercube sampling; Step 2: Based on the Q-Learning enhanced marine predator algorithm, an intelligent agent model for well location optimization in numerical simulation of waterflooding reservoir streamlines is constructed; Step 3: Use the intelligent agent-based model for optimizing well locations and numerical simulation of streamlines in water-flooding reservoirs to perform simulations, obtain and process streamline data and derived data obtained by the intelligent agent-based model for optimizing well locations and numerical simulation of streamlines in water-flooding reservoirs, and establish a streamline data set; Step 4: Combine historical data from the elite data retention area, update the Q-table through Q-Learning to optimize the phase selection strategy, and use QOL quasi-adversarial learning to generate candidate solutions, expand the search range, and use the updated strategy to guide MPA to perform next-generation well location optimization; Step 5: Based on the reservoir streamline numerical simulator, the water drive reservoir streamline numerical simulation well location optimization intelligent agent model continuously interacts with the reservoir numerical simulation to evaluate the fitness of the new population, select elite samples and store them in the elite data retention area; Step 6: Repeat steps 4 and 5 until the preset maximum number of iterations is reached to determine the optimal well location arrangement; The step 1 comprises the following steps: Step 1.1, construct a well location optimization model for numerical simulation of water drive reservoir flow lines; Step 1.2: The two-dimensional coordinates of the injection well and production well are used as optimization variables; the constraint variables are the minimum well spacing constraint and the well location non-repetition constraint; the objective function is the streamline numerical simulation well location optimization objective function , the goal is to maximize the cumulative oil production under the constraints of well spacing and well location non-repetition; Step 1.3, construct the initial population based on Latin hypercube sampling; Latin hypercube sampling is used to sample from a multivariate parameter distribution, and the number of injection well location plans to be generated is Determine the number of samples to be drawn and divide the (0,1) interval into Segments are randomly sampled in each segment, the sampled samples are mapped to standard normal distribution samples through the inverse function of standard normal distribution and randomly sorted to obtain the initial population; The water drive reservoir streamline numerical simulation well location optimization model is: （1） The constraints are: ； ； ； ； Where, is the well location coordinate set; Optimize the objective function for well location in streamline numerical simulation; is the minimum function; is the streamline displacement potential coefficient; is the coefficient of variation of streamline flight time; is the penalty term between injection-production well pairs; It is the internal penalty item for injection and production wells; and are the upper and lower bounds of the boundary constraints of the well location coordinate set respectively; is the well spacing constraint; 、 All are numbered; 、 All are single well location coordinates; is the number of injection wells, is the number of producing wells; The step 2 comprises the following steps: Step 2.1: Determine the well location coordinates of the expected production cycle based on the particle coordinate values ​​in the initial population and obtain the well location coordinate set : （2）； Step 2.2, set the simulation time step according to the time step of the expected production cycle and perform streamline numerical simulation; Step 2.3: Determine the streamline displacement potential coefficient at the current time step based on the proportion of the number of grids that the streamline passes through to the number of grids in the water flooding reservoir streamline numerical simulation well location optimization model. ; Step 2.4, based on the coefficient of variation of streamline flight time at the current time step Set the balance degree of each injection and production well on the average flight time; Step 2.5: Penalty term between injection and production wells at the current time step Set the positive difference between the average flight time of each injection-production well pair and the expected production cycle; Step 2.6, based on the internal penalty item of the injection and production well at the current time step Set the average value of the positive difference between the flight time of the streamlines contained in each injection-production well pair and the expected production period; Step 2.7, determining the initial parameters of the learning environment of the intelligent agent model for well location optimization in numerical simulation of waterflooding reservoir streamlines; The streamline displacement potential coefficient at the current time step for: （3） Where, is the total number of grids in the well location optimization model for the numerical simulation of the water drive reservoir streamline that the streamline passes through; The total number of grids in the well location optimization model for numerical simulation of waterflood reservoir flow lines; The coefficient of variation of streamline flight time at the current time step for: （4） Where, is the sample standard deviation of the flight time of the injection and production wells; is the average flight time of the injection and production wells; is the number of injection-production well pairs; For the Flight time of each injection-production well pair; The penalty term between injection and production wells at the current time step for: （5） Where, is the number of injection-production well pairs whose flight time is less than the expected production cycle, ; is the penalty intensity power series, ; The internal penalty item of the injection and production well at the current time step for: （6） Where, is the sequence number of the streamline, For the the number of streamlines within the group; is the streamline water inrush index, ,in, For the No. The flight time at the end of the root streamline, Estimated production cycle.

2. The well pattern and well location optimization method based on flow field area and balanced production evaluation according to claim 1 is characterized in that: The initial parameters of the learning environment include population size, maximum number of iterations, particle dimension, number of Q-Learning states, number of Q-Learning actions, Q-Learning training rounds, step size coefficient, discount factor, Levy random factor, Brownian random factor, and adaptive adjustment factor.

3. The well pattern and well location optimization method based on flow field area and balanced production evaluation according to claim 2, characterized in that: The step 3 comprises the following steps: Step 3.1, obtaining a streamline data file obtained by simulating a water drive reservoir streamline numerical simulation well location optimization intelligent agent model, decoding the streamline data file, and obtaining streamline data; Obtain streamline data files for each time step during the simulation of the intelligent agent model for well location optimization in numerical simulation of waterflooding reservoir streamlines. Divide the storage content of the streamline data files into string type data and numeric type data. Obtain string type data through ASCII decoding. Directly obtain numeric type data through binary decoding according to IEEE rules. Obtain streamline attributes for each time step from the streamline data files, including the position, saturation, and flight time of each streamline in the flow field. Step 3.2, streamline data preprocessing; The streamline data preprocessing includes deleting characteristic indicating coordinates in the streamline data and averaging streamline attribute data of repeated streamline points; Step 3.3, obtain other properties of the streamline field; Calculate derived data based on streamline data, obtain the grid pressure and grid number of the grid through which the streamline passes, establish a pressure mapping relationship, obtain the pressure at the current streamline point, and then use the interpolation algorithm to determine the pressure at any position in the streamline field; determine the streamline injection and production well pair number based on the streamline starting and ending well numbers, and calculate the streamline water inrush index based on the streamline flight time; According to Darcy's law and grid pressure, the fluid flow velocity within any grid interface is determined as: （7） Where, For Grid The average velocity of the internal fluid; is the grid permeability; is the fluid viscosity; For the The grid pressure of each grid; For the The grid pressure of each grid; is the grid length; In step 3.4, streamlines are grouped according to the injection-production well pair numbers, abnormal boundary streamlines are eliminated, and a streamline dataset is constructed.

4. The well pattern and well location optimization method based on flow field area and balanced production evaluation according to claim 3 is characterized in that: The step 4 comprises the following steps: Step 4.1: Set the state space based on the initialized Q-Learning parameters and action space ; Initialize the Q table to a zero matrix ; Initialize reward table for: （8） Where, is the state space, ,in, 、 、 Corresponding to the three search phases of the marine predator algorithm MPA, is the Brownian phase, used for local development, is the Levy-Brownian phase, used for mixed exploration; is the Levy phase, used for global exploration; is the action space, , is the switching strategy of the Brownian phase, is the switching strategy of the Levy-Brownian phase, is the switching strategy of Levy phase; Step 4.2, based on the current status , using a greedy strategy to select actions , after executing the action, switch the search phase and update the individual position: （9） Where, For the Time step The location of an individual, For the Time step the location of each individual; is the step size coefficient; is a random number factor between (0,1); For the Individual The elite solution used in the current iteration; is the Brownian random factor; is the adaptive adjustment factor; Step 4.3: Determine the immediate reward based on the fitness difference before and after optimization , and update the Q table using the Bellman equation: （10） in, （11） Where, is the learning decay rate function; is the discount factor; For the The immediate reward of time steps; For the The state of the time step; For the corresponding status the action chosen; is the time step, is the maximum number of iterations; Step 4.4, based on the quasi-adversarial learning mechanism, expand the search range and Construct its quasi-opposite solution : （12） Where, is a random number function; Comparison individuals Its opposite solution The fitness of , retains the optimal solution; Step 4.5: Determine whether the disturbance mechanism has been entered based on the fish aggregation mechanism : （13） Where, and are two groups of random individual positions; is the upper boundary position vector of the well coordinates, is the lower boundary position vector of the well coordinates; Step 4.6: Store the individuals with the best fitness in the current iteration into the elite reserve area and use them as the next generation update. Vector, completing the search iteration of the current iteration round.

5. The well pattern and well location optimization method based on flow field area and balanced production evaluation according to claim 4 is characterized in that: The step 5 comprises the following steps: Step 5.1, a single simulation of the individuals in the current population is performed based on the reservoir streamline numerical simulator. The water flooding reservoir streamline numerical simulation well location optimization intelligent agent model processes the streamline information and calculates the fitness of each individual in the current population in combination with the objective function. Step 5.2: For each individual in the current population, the well location coordinates are checked for legitimacy to ensure that they meet the distance approximation and repeatability filtering requirements. If not, the nearest neighbor feasible coordinate search function is called to make corrections. Step 5.3, compare the current optimal fitness with the previous generation fitness, retain the better individual, update the position coordinates of the best individual, and record the current optimal fitness in the convergence curve.

Images