Modeling method for fusing dynamic deposition simulation and artificial intelligence of meandering river point dam configuration

Abstract

The invention discloses a modeling method for dynamic deposition simulation and artificial intelligence fusion of meandering river point dam configuration, and belongs to the field of petroleum and natural gas development, and the modeling method comprises the following specific steps: I, collecting and preprocessing multiple types of data of a corresponding research area, simulating the distribution of a flow velocity field and a water depth field of an ancient river, and establishing a state conversion rule; according to the method, synchronous response of water flow movement and sediment migration can be achieved, the point dam side accumulation body forming process is accurately captured, the reasonability of configuration prediction is improved, and the method is suitable for point dam configuration accurate modeling in the complex sedimentary environment; meanwhile, accurate conversion from coarse resolution to high resolution can be realized, the description precision of the model is greatly improved, the geological loyalty and prediction capability of the model are remarkably improved, and a more scientific and comprehensive basis is provided for risk assessment and optimization decision of a development scheme.

Description

Technical Field

[0001] This invention relates to the field of oil and gas development, and in particular to a modeling method that integrates dynamic sedimentation simulation of meandering river point dam configurations with artificial intelligence. Background Technology

[0002] As oil and gas fields gradually enter the later stages of high water-cut development, point-bar lateral sedimentary bodies, as the core oil-controlling geological bodies for remaining oil accumulation in meandering river sand bodies, exhibit extremely complex spatial distribution, thickness variations, and connectivity characteristics of their internal lateral sedimentary interlayers. This makes accurate identification and quantitative characterization extremely difficult, becoming a key technical bottleneck restricting the efficient tapping of remaining oil potential. Existing point-bar configuration modeling methods have significant shortcomings: most are limited to qualitative descriptions of the macroscopic morphology of point-bars, lacking quantitative characterization of the fine structure of internal interlayers, failing to accurately match interlayer data interpreted from well logging, and lacking effective guidance for formulating remaining oil potential tapping schemes; moreover, the modeling process is cumbersome and lengthy, relying heavily on manual experience and judgment, with poor human-computer interaction and strong subjectivity, making it difficult to meet the needs of large-scale, refined development in terms of model repeatability and generalizability.

[0003] A search revealed that Chinese patent CN118211359A discloses a point bar configuration modeling method combining sedimentary models and geostatistics. While this invention can utilize geological models to determine the morphology and trends of abandoned channels and lateral accretion layers, existing point bar configuration modeling methods fail to achieve synchronous responses between water flow and sediment transport, cannot accurately capture the dynamic formation process of point bar lateral accretion bodies, and are ill-suited to the precise modeling requirements of complex sedimentary environments. Furthermore, existing point bar configuration modeling methods lack a precise conversion mechanism from coarse to high resolution, resulting in insufficient accuracy in depicting the fine internal structure of lateral accretion bodies, making it difficult to fully support risk assessment and optimization decisions for development schemes. Therefore, we propose a modeling method that integrates dynamic sedimentary simulation of meandering river point bar configurations with artificial intelligence. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a modeling method that integrates dynamic sedimentation simulation of meandering river point dam configurations with artificial intelligence.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A modeling method integrating dynamic sedimentary simulation and artificial intelligence for point bar configurations in meandering rivers is proposed. The specific steps of this modeling method are as follows:

[0007] Ⅰ: Collect and preprocess various types of data for the corresponding study area, and simulate the velocity and depth field distribution of ancient rivers to establish state transition rules;

[0008] II: Convert the seismic attribute volume into a probability field, use a simulator to generate the initial river channel morphology, and generate the corresponding coarse mesh model;

[0009] III: Construct and train fine-grained configuration models, generate corresponding high-resolution configuration models for each coarse-grid model, and construct the corresponding set of high-resolution models;

[0010] IV: Based on the set of high-resolution models, analyze the differences in distribution characteristics among the high-resolution models to form uncertainty characterization results.

[0011] As a further aspect of the present invention, the specific steps for establishing the state transition rule in step I are as follows:

[0012] S1.1: Collect multiple types of data for the corresponding study area, standardize the format of each type of data, calculate the standard deviation of each type of data, mark the corresponding type of data that exceeds three times the standard deviation range as outliers and remove them, and then transform the spatial coordinates of each type of data to a unified benchmark.

[0013] S1.2: Collect paleogeographic background, stratigraphic lithology distribution and paleoclimate data of the corresponding study area, set the initial topographic elevation data of the ancient river, determine the initial direction and width characteristics of the river channel, define the simulation time span and time step, clarify the initial flow rate at the inlet and the water level boundary conditions at the outlet, and set the corresponding physical parameters. Then, based on the known basic principles of fluid mechanics, set up a two-dimensional shallow water equation to describe the flow motion of the ancient river, and discretize the two-dimensional shallow water equation based on the preset computing grid, and set the convergence criterion and the upper limit of the number of iterations for solving the equation.

[0014] S1.3: Based on the preset basic parameters and boundary conditions, the two-dimensional shallow water equation is used to iteratively solve the problem and obtain the magnitude, direction and depth of the water flow in each grid cell in the simulation area. After a single iteration, the calculation results are checked to see if they meet the preset convergence threshold. If they do, the velocity field and depth field distribution data of the simulation area at each time point are output.

[0015] S1.4: According to the spatial scale consistent with the numerical simulation of the two-dimensional shallow water equation, the study area is divided into a regular cellular automata grid, and the interaction rules for state transfer between cells are formulated to ensure that each cell corresponds one-to-one with the numerical computation grid. At the same time, the core state variables of the cells are defined, and a clear value range and quantification standard are set for each state variable.

[0016] S1.5: Using the velocity field and depth field data obtained from solving the two-dimensional shallow water equation as the core driver, a trigger threshold for cell state transition is set. When the velocity at the cell location is greater than the set threshold, the erosion state is triggered, and the amount of sediment eroded by the cell per unit time is calculated. When the velocity is between the erosion threshold and the deposition threshold, the transport state is triggered, and the path and amount of sediment migration to adjacent cells along the flow direction are set and calculated. When the velocity is less than the deposition threshold, the deposition state is triggered, and the amount of sediment deposited after the cell receives sediment transported by adjacent cells is calculated.

[0017] S1.6: Establish a real-time data interaction channel between the velocity field, water depth field distribution data and the cellular automaton, and transmit the calculation results within each time step to the corresponding cell in real time. At the same time, the cell updates its state according to the received velocity, direction and water depth data, combined with its current state and the states of neighboring cells, according to the preset state transition rules. After each time step, the velocity field and water depth field response to the cell state change is updated synchronously.

[0018] As a further aspect of the present invention, the specific steps for dividing the study area into a regular cellular automata grid as described in S1.4 are as follows:

[0019] P1.1: Based on the existing topographic data of the study area and the coordinates of the two-dimensional shallow water equation numerical simulation, the planar coordinate projection method and elevation datum of the cellular automata grid are determined. Based on the geological boundary of the study area and the potential area for point dam development, the spatial coverage boundary of the cellular automata grid is delineated. The east-west and north-south extension range of the grid is determined with reference to the watershed of the ancient river basin and the upstream and downstream endpoints of the river channel.

[0020] P1.2: Based on the principle of balancing simulation accuracy requirements and computational efficiency, select the appropriate regular grid cell type, and set the grid cell size in combination with the scale of key geological phenomena in the corresponding study area and numerical computing capabilities. At the same time, use numerical modeling tools to generate regular cellular automata grids according to the set spatial reference, coverage, cell type and size.

[0021] P1.3: Using the mesh generation function of the numerical modeling tool, the cellular automata mesh is divided into uniformly distributed mesh units. The fit between the mesh boundary and the preset range is then checked. A unique identification number is assigned to each mesh unit. Based on the set spatial reference, the coordinate value of the center point of each mesh unit is calculated to determine the specific position of the mesh unit in three-dimensional space.

[0022] As a further aspect of the present invention, the specific steps for generating the initial river channel morphology using a simulator in step II are as follows:

[0023] S2.1: Based on the geological objectives of the point dam configuration modeling, the seismic attributes corresponding to the development of channel sand bodies are screened, and the noise in the seismic attributes is filtered out by mean filtering. At the same time, the distribution data of channel sand bodies calibrated by well core sampling in the study area are collected, and the seismic attributes corresponding to the known channel locations are extracted. Then, statistical analysis methods are used to compare the differences in seismic attributes between the channel area and the non-channel area, calculate the attribute threshold range indicating channel development, and construct a quantitative correlation model.

[0024] S2.2: The processed seismic attribute values ​​are mapped to the [0,1] interval, and based on the quantitative correlation model, the attribute values ​​of each grid cell are directly converted into river development probability values. Then, the weighted summation method is used to fuse the probability contribution values ​​of each attribute in the seismic attributes to generate a probability field covering the entire study area.

[0025] S2.3: Embed the constructed probability field into the built-in dynamic sedimentation process simulator, establish the correlation logic between probability values ​​and river migration path selection, and set a positive correlation between the probability values ​​of probability field grid cells and path selection weights. If the probability exceeds the preset range, assign a weight exceeding the preset threshold; otherwise, assign a weight below the preset threshold. At the same time, set the quantitative standard for constraint rules.

[0026] S2.4: Based on the paleogeographic background of the study area, the initial parameters of the dynamic sedimentation process simulator are set, and under the action of the prior probability constraint rules, the dynamic sedimentation process simulator is started. The river migration path is expanded in the high probability area first. At the same time, the river direction and width are adjusted in combination with the initial topographic elevation changes of the study area to simulate the process of sediment erosion, transportation and deposition, forming the initial river morphology. Then, based on the initial river morphology, an initial coarse grid model is generated.

[0027] As a further aspect of the present invention, the specific steps for generating the corresponding coarse mesh model in step II are as follows:

[0028] S3.1: Extract the spatial distribution information of the initial channel morphology, calculate the spatial matching degree between the initial channel morphology and the probability field, measure the coverage of the initial channel morphology in the high probability area, and then convert the calculated spatial matching degree into the first likelihood function. After that, collect historical production dynamic data in the study area and use time series interpolation to supplement the missing data in the historical production dynamic data.

[0029] S3.2: Import the initial coarse mesh model into the built-in streamline simulator, set the fluid parameters of the streamline simulator, and then input the processed historical production data into the streamline simulator. The streamline simulator simulates the reservoir fluid flow process and outputs the simulation dynamic results. Compare the fit between the simulation dynamic results and the actual production data of the study area, and convert the fit measure into the second likelihood function.

[0030] S3.3: The particle swarm optimization algorithm is adopted to maximize the joint probability of the first likelihood function and the second likelihood function. The parameters of the dynamic deposition process simulator are adjusted, the optimization parameters of the particle swarm optimization algorithm are set, and the parameters of the dynamic deposition process simulator are used as optimization variables. In each iteration, the joint likelihood function value corresponding to each particle is calculated, the optimal position of the particle and the global optimal position are updated, and the parameters of the dynamic deposition process simulator are continuously adjusted until the preset number of iterations is reached or the joint likelihood function value converges to obtain the optimal parameter combination.

[0031] S3.4: Based on the optimized parameter combination, the dynamic deposition process simulator is started multiple times. Random perturbation is introduced during each simulation to generate multiple sets of different coarse grid models. The joint likelihood function value corresponding to each set of coarse grid models is calculated. Coarse grid models whose joint likelihood function value exceeds the preset range are selected to finally form the coarse grid model.

[0032] As a further aspect of the present invention, the specific steps for generating a corresponding high-resolution configuration model for each coarse mesh model in step III are as follows:

[0033] S4.1: Construct and train the fine-configuration model, organize all coarse-mesh models to form a complete model set, and then number and sort each coarse-mesh model in the model set according to a preset order. At the same time, based on the computing resource carrying capacity of the fine-configuration model, set the number of input models and the scale of parallel processing, and determine the computing channels of each coarse-mesh model.

[0034] S4.2: The coarse grid model is input into the generator of the fine configuration model one by one in the sorting order. The generator extracts the macroscopic geological features of the input coarse grid model and converts the extracted macroscopic geological features into feature vectors that the generator can recognize. At the same time, the generator performs intelligent downscaling operation on the macroscopic geological features of the coarse grid model based on the geological mapping relationship. With the macroscopic geological features of the coarse grid model as constraints, microscopic geological details are added inside each coarse grid cell. Then, through iterative calculation of multi-layer neural network, the coarse grid cell is split into multiple high-resolution grid cells.

[0035] S4.3: After the generator completes the downscaling operation, it outputs the high-resolution model corresponding to each coarse grid model. Then, it performs quality verification on the high-resolution model to verify whether the extension direction and thickness variation of the lateral accretion layer in the high-resolution model conform to the law of sedimentary dynamics. Then, it checks the consistency of the macroscopic geological features between the high-resolution model and the corresponding coarse grid model. Finally, it assigns a unique identifier to the high-resolution model that passes the verification.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] 1. This invention collects various types of data from the corresponding study area, unifies the format of each data type, and removes outliers beyond three standard deviations through standard deviation analysis. Then, it converts each data type to a unified benchmark. Simultaneously, it collects paleogeographic background, stratigraphic lithology, and paleoclimate data for the study area, sets the initial topographic elevation, channel direction, and width of the ancient river, defines the simulation time span and step size, and clarifies the inlet and outlet flow rates, water level boundary conditions, and corresponding physical parameters. Then, based on fluid mechanics principles, a two-dimensional shallow water equation is established. The equation is then discretized according to a preset computational grid, and convergence criteria and iteration limits are set. The parameters and boundary conditions are substituted to iteratively solve the problem, obtaining velocity, direction, and depth data for each grid. After meeting the convergence threshold, the velocity field and depth field at each time point are output. Following a spatial scale consistent with the two-dimensional shallow water equation numerical simulation, the study area is divided into regular cellular automata grids, ensuring a one-to-one correspondence between each cell and the numerical computation grid. Simultaneously, the core state variables of the cells are defined. The method establishes quantitative and value standards, using velocity and depth fields as drivers to set state transition thresholds. It calculates sediment migration and deposition under erosion, transport, and deposition states, establishes data interaction channels, filters corresponding seismic attributes of channel sand bodies and performs mean filtering for noise reduction, and combines well calibration data of the study area to establish a quantitative correlation model through statistical analysis. After normalizing attribute values, it converts them into channel development probability values, integrates the probability contributions of multiple attributes to generate a regional probability field, embeds the probability field into a dynamic sedimentary process simulator, sets path selection weight association rules, initializes simulator parameters based on paleogeographic background, prioritizes channel expansion in areas where the probability exceeds the preset threshold, and then adjusts the channel morphology in combination with topography to simulate the sediment transport and deposition process. Finally, it generates an initial coarse-grid configuration model, which can achieve synchronous response of water flow and sediment transport, accurately capture the point bar lateral accumulation formation process, improve the rationality of configuration prediction, and is suitable for accurate modeling of point bar configurations in complex sedimentary environments.

[0038] 2. This invention extracts the spatial distribution information of the initial channel morphology, calculates the spatial matching degree between the initial channel morphology and the probability field, and transforms the spatial matching degree into a first likelihood function. Simultaneously, it collects historical production dynamic data of the study area, supplements missing data using time series interpolation, imports the initial coarse-grid model into a streamline simulator, sets the corresponding fluid parameters, inputs the processed historical production data, simulates reservoir fluid flow, and outputs the results. The fit between the simulation and actual production data is compared to quantify the second likelihood function. Then, a particle swarm optimization algorithm is used to maximize the joint probability of the first and second likelihood functions. The optimization parameters of the algorithm are set, and the parameters of the dynamic sedimentation process simulator are used as optimization variables. The joint likelihood function value is iteratively calculated, and the optimal particle position is updated until the iteration reaches the target or converges, obtaining the optimal parameter combination. Based on the optimized parameter combination, the simulator is started multiple times. By introducing random perturbations, multiple sets of coarse-grid models are generated. Models whose joint likelihood function values ​​conform to a preset range are then selected, organized, numbered, and sorted. Resource carrying capacity is calculated based on the refined configuration models. Batch processing rules are then set, and the models are input one by one into the generator of the refined configuration models. The generator extracts the macroscopic geological features of the coarse-grid models and converts them into corresponding feature vectors. Simultaneously, intelligent downscaling operations are used to supplement details, and coarse-grid units are split to generate high-resolution models. Quality verification is performed on the high-resolution models to verify whether the lateral sedimentary features conform to sedimentary dynamics and to check their macroscopic consistency with the corresponding coarse-grid models. A unique identifier is assigned to models that pass the verification. This process enables accurate conversion from coarse to high resolution, significantly improving the model's characterization accuracy, geological fidelity, and predictive ability, providing a more scientific and comprehensive basis for risk assessment and optimization decisions in development schemes. Attached Figure Description

[0039] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0040] Figure 1 This is a flowchart of the modeling method that integrates dynamic sedimentation simulation and artificial intelligence for point dam configurations in meandering rivers, as proposed in this invention. Detailed Implementation

[0041] Example 1

[0042] Reference Figure 1 A modeling method integrating dynamic sedimentary simulation of point dam configurations in meandering rivers with artificial intelligence is proposed. The specific steps of this modeling method are as follows:

[0043] We collected and preprocessed various types of data for the corresponding study area, and simulated the velocity and depth fields of ancient rivers to establish state transition rules.

[0044] Specifically, multiple types of data for the corresponding study area were collected, and the format of each type of data was standardized. The standard deviation of each type of data was then calculated, and data exceeding three times the standard deviation were marked as outliers and removed. The spatial coordinates of each type of data were then transformed to a unified benchmark. Paleogeographical background, stratigraphic lithology distribution, and paleoclimate data for the corresponding study area were collected. Initial topographic elevation data for the ancient river were established, and the initial direction and width characteristics of the river channel were determined. The simulation time span and time step were defined, and the initial flow rate at the inlet and the water level boundary conditions at the outlet were clarified. Simultaneously, corresponding physical parameters were set. Then, based on known fundamental principles of fluid mechanics, a two-dimensional shallow-water equation describing the ancient river's flow motion was established. This equation was discretized using a pre-defined computational grid. Convergence criteria and an upper limit for the number of iterations were set. Based on pre-defined basic parameters and boundary conditions, the two-dimensional shallow-water equation was iteratively solved to obtain the flow velocity, direction, and depth values ​​for each grid cell within the simulation area. After each iteration, the calculation results were checked to ensure they met the pre-defined convergence threshold. If they did, the velocity and depth field distribution data for the simulation area at each time point were output, following the same numerical simulation method as the two-dimensional shallow-water equation. To achieve a consistent spatial scale, the study area was divided into a regular cellular automata grid, and interaction rules for state transfer between cells were established to ensure a one-to-one correspondence between each cell and the numerical computation grid. Core state variables for each cell were defined, with specific value ranges and quantification standards set for each variable. Using velocity and depth field data obtained from solving the two-dimensional shallow water equations as the core driving force, trigger thresholds for cell state transitions were set. When the velocity at the cell's location exceeds the set threshold, an erosion state is triggered, and the amount of sediment eroded per unit time is calculated. When the velocity is between the erosion and deposition thresholds, a transport state is triggered. The system calculates the path and amount of sediment migration along the flow direction to adjacent cells. When the flow velocity is less than the deposition threshold, a deposition state is triggered, and the amount of sediment deposited after the cell receives sediment transported by adjacent cells is calculated. A real-time data interaction channel is established between the velocity field, water depth field distribution data and the cellular automaton, and the calculation results within each time step are transmitted to the corresponding cell in real time. At the same time, the cell updates its state according to the received velocity, flow direction and water depth data, combined with its current state and the state of adjacent cells, according to the preset state transition rules. After each time step, the velocity field and water depth field responses to the cell state changes are updated synchronously.

[0045] It should be further explained that the specific steps for dividing the study area into a regular cellular automata grid are as follows:

[0046] Based on existing topographic data and coordinates from two-dimensional shallow water equation numerical simulations of the study area, the planar coordinate projection method and elevation datum of the cellular automata grid were determined. The spatial coverage boundary of the cellular automata grid was delineated based on the geological boundaries and potential areas for point dam development in the study area. Using the watershed of the ancient river basin and the upstream and downstream endpoints of the river channel as references, the east-west and north-south extension ranges of the grid were determined. According to the principle of balancing simulation accuracy requirements and computational efficiency, appropriate regular grid cell types were selected. The grid cell size was set based on the scale of key geological phenomena in the corresponding study area and numerical computation capabilities. Simultaneously, using numerical modeling tools, regular cellular automata grids were generated according to the set spatial datum, coverage range, cell type, and size. The grid generation function of the numerical modeling tools was used to divide the cellular automata grid into uniformly distributed grid cells. The fit between the grid boundaries and the preset range was checked, and a unique identification number was assigned to each grid cell. Based on the set spatial datum, the coordinate values ​​of the center point of each grid cell were calculated to determine the specific location of the grid cell in three-dimensional space.

[0047] The seismic attribute volume is converted into a probability field, and the initial river channel morphology is generated using a simulator to generate the corresponding coarse mesh model.

[0048] Specifically, based on the geological objectives of point dam configuration modeling, seismic attributes corresponding to channel sand body development are selected, and mean filtering is used to filter noise in the seismic attributes. Simultaneously, channel sand body distribution data calibrated by well core sampling within the study area are collected, and seismic attributes corresponding to known channel locations are extracted. Then, statistical analysis methods are used to compare the differences in seismic attributes between channel and non-channel areas, calculate the attribute threshold interval indicating channel development, and construct a quantitative correlation model. The processed seismic attribute values ​​are mapped to the [0,1] interval. Based on the quantitative correlation model, the attribute value of each grid cell is directly converted into a channel development probability value. Then, a weighted summation method is used to fuse the probability contribution values ​​of each attribute in the seismic attributes, generating a probability field covering the entire study area. The constructed probability field is then embedded into the built-in... A dynamic sedimentation process simulator establishes a logical correlation between probability values ​​and river migration path selection, and sets a positive correlation between the probability values ​​of probability field grid cells and path selection weights. If the probability exceeds a preset range, a weight exceeding a preset threshold is assigned; conversely, a weight below a preset threshold is assigned. Quantitative standards for constraint rules are also set. Based on the paleogeographic background of the study area, initial parameters for the dynamic sedimentation process simulator are set. Under the action of prior probability constraint rules, the dynamic sedimentation process simulator is started, prioritizing the expansion of river migration paths in high-probability areas. Simultaneously, the river direction and width are adjusted in conjunction with the initial topographic elevation changes of the study area to simulate sediment erosion, transportation, and deposition processes, forming an initial river morphology. Then, based on the initial river morphology, an initial coarse grid model is generated.

[0049] Specifically, the spatial distribution information of the initial channel morphology is extracted, and the spatial matching degree between the initial channel morphology and the probability field is calculated to measure the coverage of the initial channel morphology in high-probability areas. The calculated spatial matching degree is then converted into a first likelihood function. Historical production dynamic data of the study area are collected, and time series interpolation is used to supplement missing data in the historical production dynamic data. The initial coarse-grid model is imported into the built-in streamline simulator, and the fluid parameters of the streamline simulator are set. The processed historical production data is then input into the streamline simulator, which simulates the reservoir fluid flow process and outputs the simulated dynamic results. The fit between the simulated dynamic results and the actual production data of the study area is compared, and the fit measure is converted into a second likelihood function. A particle swarm optimization algorithm is used to maximize the first likelihood. The parameters of the dynamic deposition process simulator are adjusted to achieve the joint probability of the first and second likelihood functions. The optimization parameters of the particle swarm optimization algorithm are set, and the parameters of the dynamic deposition process simulator are used as optimization variables. In each iteration, the joint likelihood function value corresponding to each particle is calculated, and the optimal position of the particle and the global optimal position are updated. The parameters of the dynamic deposition process simulator are continuously adjusted until the preset number of iterations is reached or the joint likelihood function value converges, obtaining the optimal parameter combination. Based on the optimized parameter combination, the dynamic deposition process simulator is started multiple times. Random perturbations are introduced in each simulation to generate multiple sets of different coarse-grid models. The joint likelihood function value corresponding to each set of coarse-grid models is calculated, and coarse-grid models whose joint likelihood function values ​​exceed the preset range are selected, ultimately forming the coarse-grid model.

[0050] Example 2

[0051] Reference Figure 1 A modeling method integrating dynamic sedimentary simulation of point dam configurations in meandering rivers with artificial intelligence is proposed. The specific steps of this modeling method are as follows:

[0052] Construct and train fine-grained configuration models, generate corresponding high-resolution configuration models for each coarse-grid model, and construct a set of corresponding high-resolution models.

[0053] Specifically, a fine-grained model is constructed and trained, and all coarse-grid models are organized to form a complete model set. Each coarse-grid model in the set is then numbered and sorted according to a preset order. Simultaneously, based on the computational resource capacity of the fine-grained model, the number of models input per run and the scale of parallel processing are set, and the computational channels for each coarse-grid model are determined. The coarse-grid models are then input one by one into the generator of the fine-grained model in the sorted order. The generator extracts the macroscopic geological features of the input coarse-grid models and converts these features into feature vectors that the generator can recognize. Simultaneously, based on geological mapping relationships, the generator performs macroscopic mapping on the coarse-grid models. The geological features are intelligently downscaled. The macroscopic geological features of the coarse grid model are used as constraints. Microscopic geological details are added to each coarse grid cell. Then, through iterative calculation of a multi-layer neural network, the coarse grid cell is split into multiple high-resolution grid cells. After the generator completes the downscaling operation, it outputs the high-resolution model corresponding to each coarse grid model. The high-resolution model is then quality-verified to verify whether the extension direction and thickness variation of the lateral accretion layer in the high-resolution model conform to the laws of sedimentary dynamics. The consistency of the macroscopic geological features between the high-resolution model and the corresponding coarse grid model is then checked. Finally, a unique identifier is assigned to the high-resolution model that passes the verification.

[0054] Based on the set of high-resolution models, the differences in distribution characteristics among the high-resolution models are analyzed to form uncertainty characterization results.

Claims

1. A modeling method integrating dynamic sedimentary simulation of point dam configurations in meandering rivers with artificial intelligence, characterized in that... The specific steps of this modeling method are as follows: Ⅰ: Collect and preprocess various types of data for the corresponding study area, and simulate the velocity and depth field distribution of ancient rivers to establish state transition rules; II: Convert the seismic attribute volume into a probability field, use a simulator to generate the initial river channel morphology, and generate the corresponding coarse mesh model; III: Construct and train fine-grained configuration models, generate corresponding high-resolution configuration models for each coarse-grid model, and construct the corresponding set of high-resolution models; IV: Based on the set of high-resolution models, analyze the differences in distribution characteristics among the high-resolution models to form uncertainty characterization results.

2. The modeling method for dynamic sedimentary simulation and artificial intelligence integration of meandering river point dam configuration according to claim 1, characterized in that, The specific steps for establishing the state transition rules described in step I are as follows: S1.1: Collect multiple types of data for the corresponding study area, standardize the format of each type of data, calculate the standard deviation of each type of data, mark the corresponding type of data that exceeds three times the standard deviation range as outliers and remove them, and then transform the spatial coordinates of each type of data to a unified benchmark. S1.2: Collect paleogeographic background, stratigraphic lithology distribution and paleoclimate data of the corresponding study area, set the initial topographic elevation data of the ancient river, determine the initial direction and width characteristics of the river channel, define the simulation time span and time step, clarify the initial flow rate at the inlet and the water level boundary conditions at the outlet, and set the corresponding physical parameters. Then, based on the known basic principles of fluid mechanics, set up a two-dimensional shallow water equation to describe the flow motion of the ancient river, and discretize the two-dimensional shallow water equation based on the preset computing grid, and set the convergence criterion and the upper limit of the number of iterations for solving the equation. S1.3: Based on the preset basic parameters and boundary conditions, the two-dimensional shallow water equation is used to iteratively solve the problem and obtain the magnitude, direction and depth of the water flow in each grid cell in the simulation area. After a single iteration, the calculation results are checked to see if they meet the preset convergence threshold. If they do, the velocity field and depth field distribution data of the simulation area at each time point are output. S1.4: According to the spatial scale consistent with the numerical simulation of the two-dimensional shallow water equation, the study area is divided into a regular cellular automata grid, and the interaction rules for state transfer between cells are formulated to ensure that each cell corresponds one-to-one with the numerical computation grid. At the same time, the core state variables of the cells are defined, and a clear value range and quantification standard are set for each state variable. S1.5: Using the velocity field and depth field data obtained from solving the two-dimensional shallow water equation as the core driver, a trigger threshold for cell state transition is set. When the velocity at the cell location is greater than the set threshold, the erosion state is triggered, and the amount of sediment erosion per unit time is calculated. When the velocity is between the erosion threshold and the deposition threshold, the transport state is triggered, and the path and amount of sediment migration to adjacent cells along the flow direction are set and calculated. When the flow velocity is less than the deposition threshold, the deposition state is triggered, and the amount of sediment deposited after the cell receives sediment transported by the adjacent cells is calculated. S1.6: Establish a real-time data interaction channel between the velocity field, water depth field distribution data and the cellular automaton, and transmit the calculation results within each time step to the corresponding cell in real time. At the same time, the cell updates its state according to the received velocity, direction and water depth data, combined with its current state and the states of neighboring cells, according to the preset state transition rules. After each time step, the velocity field and water depth field response to the cell state change is updated synchronously.

3. The modeling method for dynamic sedimentary simulation and artificial intelligence integration of meandering river point dam configuration according to claim 2, characterized in that, The specific steps for dividing the study area into a regular cellular automata grid, as described in S1.4, are as follows: P1.1: Based on the existing topographic data of the study area and the coordinates of the two-dimensional shallow water equation numerical simulation, the planar coordinate projection method and elevation datum of the cellular automata grid are determined. Based on the geological boundary of the study area and the potential area for point dam development, the spatial coverage boundary of the cellular automata grid is delineated. The east-west and north-south extension range of the grid is determined with reference to the watershed of the ancient river basin and the upstream and downstream endpoints of the river channel. P1.2: Based on the principle of balancing simulation accuracy requirements and computational efficiency, select the appropriate regular grid cell type, and set the grid cell size in combination with the scale of key geological phenomena in the corresponding study area and numerical computing capabilities. At the same time, use numerical modeling tools to generate regular cellular automata grids according to the set spatial reference, coverage, cell type and size. P1.3: Using the mesh generation function of the numerical modeling tool, the cellular automata mesh is divided into uniformly distributed mesh units. The fit between the mesh boundary and the preset range is then checked. A unique identification number is assigned to each mesh unit. Based on the set spatial reference, the coordinate value of the center point of each mesh unit is calculated to determine the specific position of the mesh unit in three-dimensional space.

4. The modeling method for dynamic sedimentary simulation and artificial intelligence fusion of meandering river point dam configuration according to claim 3, characterized in that, The specific steps for generating the initial river channel morphology using the simulator in step II are as follows: S2.1: Based on the geological objectives of the point dam configuration modeling, the seismic attributes corresponding to the development of channel sand bodies are screened, and the noise in the seismic attributes is filtered out by mean filtering. At the same time, the distribution data of channel sand bodies calibrated by well core sampling in the study area are collected, and the seismic attributes corresponding to the known channel locations are extracted. Then, statistical analysis methods are used to compare the differences in seismic attributes between the channel area and the non-channel area, calculate the attribute threshold range indicating channel development, and construct a quantitative correlation model. S2.2: The processed seismic attribute values ​​are mapped to the [0,1] interval, and based on the quantitative correlation model, the attribute values ​​of each grid cell are directly converted into river development probability values. Then, the weighted summation method is used to fuse the probability contribution values ​​of each attribute in the seismic attributes to generate a probability field covering the entire study area. S2.3: Embed the constructed probability field into the built-in dynamic sedimentation process simulator, establish the correlation logic between probability values ​​and river migration path selection, and set a positive correlation between the probability values ​​of probability field grid cells and path selection weights. If the probability exceeds the preset range, assign a weight exceeding the preset threshold; otherwise, assign a weight below the preset threshold. At the same time, set the quantitative standard for constraint rules. S2.4: Based on the paleogeographic background of the study area, the initial parameters of the dynamic sedimentation process simulator are set, and under the action of the prior probability constraint rules, the dynamic sedimentation process simulator is started. The river migration path is expanded in the high probability area first. At the same time, the river direction and width are adjusted in combination with the initial topographic elevation changes of the study area to simulate the process of sediment erosion, transportation and deposition, forming the initial river morphology. Then, based on the initial river morphology, an initial coarse grid model is generated.

5. The modeling method for dynamic sedimentary simulation and artificial intelligence integration of meandering river point dam configuration according to claim 4, characterized in that, The specific steps for generating the corresponding coarse mesh model in step II are as follows: S3.1: Extract the spatial distribution information of the initial channel morphology, calculate the spatial matching degree between the initial channel morphology and the probability field, measure the coverage of the initial channel morphology in the high probability area, and then convert the calculated spatial matching degree into the first likelihood function. After that, collect historical production dynamic data in the study area and use time series interpolation to supplement the missing data in the historical production dynamic data. S3.2: Import the initial coarse mesh model into the built-in streamline simulator, set the fluid parameters of the streamline simulator, and then input the processed historical production data into the streamline simulator. The streamline simulator simulates the reservoir fluid flow process and outputs the simulation dynamic results. Compare the fit between the simulation dynamic results and the actual production data of the study area, and convert the fit measure into the second likelihood function. S3.3: The particle swarm optimization algorithm is adopted to maximize the joint probability of the first likelihood function and the second likelihood function. The parameters of the dynamic deposition process simulator are adjusted, the optimization parameters of the particle swarm optimization algorithm are set, and the parameters of the dynamic deposition process simulator are used as optimization variables. In each iteration, the joint likelihood function value corresponding to each particle is calculated, the optimal position of the particle and the global optimal position are updated, and the parameters of the dynamic deposition process simulator are continuously adjusted until the preset number of iterations is reached or the joint likelihood function value converges to obtain the optimal parameter combination. S3.4: Based on the optimized parameter combination, the dynamic deposition process simulator is started multiple times. Random perturbation is introduced during each simulation to generate multiple sets of different coarse grid models. The joint likelihood function value corresponding to each set of coarse grid models is calculated. Coarse grid models whose joint likelihood function value exceeds the preset range are selected to finally form the coarse grid model.

6. The modeling method for dynamic sedimentary simulation and artificial intelligence integration of meandering river point dam configuration according to claim 5, characterized in that, The specific steps for generating the corresponding high-resolution configuration model for each coarse mesh model described in step III are as follows: S4.1: Construct and train the fine-configuration model, organize all coarse-mesh models to form a complete model set, and then number and sort each coarse-mesh model in the model set according to a preset order. At the same time, based on the computing resource carrying capacity of the fine-configuration model, set the number of input models and the scale of parallel processing, and determine the computing channels of each coarse-mesh model. S4.2: The coarse grid model is input into the generator of the fine configuration model one by one in the sorting order. The generator extracts the macroscopic geological features of the input coarse grid model and converts the extracted macroscopic geological features into feature vectors that the generator can recognize. At the same time, the generator performs intelligent downscaling operation on the macroscopic geological features of the coarse grid model based on the geological mapping relationship. With the macroscopic geological features of the coarse grid model as constraints, microscopic geological details are added inside each coarse grid cell. Then, through iterative calculation of multi-layer neural network, the coarse grid cell is split into multiple high-resolution grid cells. S4.3: After the generator completes the downscaling operation, it outputs the high-resolution model corresponding to each coarse grid model. Then, it performs quality verification on the high-resolution model to verify whether the extension direction and thickness variation of the lateral accretion layer in the high-resolution model conform to the law of sedimentary dynamics. Then, it checks the consistency of the macroscopic geological features between the high-resolution model and the corresponding coarse grid model. Finally, it assigns a unique identifier to the high-resolution model that passes the verification.

Images