A method for constructing a braided river sand body structure training image based on deposition numerical simulation

Abstract

The present application belongs to the technical field of oil and gas field development geological reservoir identification, and specifically provides a braided river sand body structure training image construction method based on sedimentary numerical simulation, which comprises the following steps: obtaining the parameters of the braided river under similar conditions in the research area, and using a numerical simulation software to perform numerical simulation of the braided river deposition; then, the data of the numerical simulation result of the deposition is screened, and the hydrodynamic data for constructing the training image in the simulation result is screened out; the thickness surface data of the numerical simulation evolution of the deposition is first used to construct a sedimentary body framework model through a rule-based method; the braided river configuration unit is identified by selecting a plurality of hydrodynamic parameter data of the simulation result; and finally, the braided river configuration unit is filled into the sedimentary body framework model established in the early stage through the assignment filling method. The present application provides statistically robust and geologically reasonable prior information for multipoint geostatistical geological modeling, artificial intelligence geological modeling and the like, and helps to improve the prediction accuracy of the reservoir geological model.

Description

Technical Field

[0001] This invention belongs to the field of geological reservoir identification technology for oil and gas field development, and particularly relates to a method for constructing training images of braided river sand body structures based on sedimentary numerical simulation. Background Technology

[0002] Reservoir geological modeling is one of the core technologies in oil and gas exploration and development, and its accuracy directly affects resource reserve assessment and development plan optimization. The complex heterogeneity of fluvial reservoirs not only determines the pore structure and seepage characteristics of the reservoir, but also has a decisive impact on the accurate prediction of remaining oil distribution and underground fluid migration. Constructing a geological model that can realistically reflect its internal structure has always been a major challenge. In particular, braided river sedimentary systems are characterized by frequent migration of multiple channels, repeated superposition and alteration of mid-channel bars, and strong erosion and infilling, resulting in complex internal structures and variable connectivity of sand bodies, and significant reservoir heterogeneity. Mainstream geological modeling methods for strongly heterogeneous reservoirs, such as multi-point geostatistics and deep learning, all require training images to capture the geometric morphology and spatial features of geological bodies. Therefore, constructing training images that contain reasonable geological features is crucial.

[0003] Currently, academia and industry have developed various methods for acquiring training images. Manual drawing transforms the experience and knowledge of geologists into two-dimensional training images. While convenient and expanded in recent years through 2D-to-3D algorithms, it suffers from high workload, strong subjectivity, and difficulty in quality control. Target model-based methods, though easy to parameterize, struggle to accurately reproduce the geometry and spatial superposition of facies at multiple scales, and their commonly used geometric forms often do not match actual geological conditions. Geophysical data extraction methods establish correspondences with sedimentary facies through 3D seismic attribute analysis. This method is highly dependent on the quality, resolution, and correlation with reservoir properties of the seismic data; failure to meet any of these conditions will lead to method failure. Existing geological model methods directly utilize detailed geological models of mature oil and gas fields as training images. While this allows for data synchronization and model updates, it struggles to overcome the inherent limitations of existing models. The outcrop analogy method uses surface outcrops as analogies to the geometry of underground lithofacies, effectively filling the scale gap between well logging and seismic data. However, due to the limited outcrop exposure, even the most complete outcrop can only provide fragmented "windows" and is difficult to fully reveal the internal structure of large three-dimensional geological bodies.

[0004] In summary, the construction of training images is crucial for reservoir geological modeling, and a new method for constructing training images is urgently needed. Numerical simulation methods based on sedimentary physics processes not only consider the sediment accumulation process, but more importantly, fully simulate the control effect of hydrodynamic conditions on sedimentary configurations. This allows for the generation of simulation results that are more consistent with sedimentary laws, retains more refined reservoir structural features that modeling methods can capture, and can generate full 3D training images in batches. Summary of the Invention

[0005] This invention aims to address the technical problems of existing training image construction methods, such as reliance on human experience, difficulty in reflecting the dynamic characteristics of sedimentary processes, low generation efficiency, and insufficient diversity. It provides a method for constructing training images of braided river sand bodies based on sedimentary numerical simulation. This method can automatically generate high-quality training images that are both geologically plausible and diverse, providing reliable data support for multi-point geostatistics and deep learning geological modeling. This invention includes the following technical solutions:

[0006] S1. Set the simulation parameters for the sedimentation numerical simulation, run the sedimentation numerical simulation, and obtain the time series hydrodynamic and sedimentation outputs.

[0007] Simulation parameters include grid size, dimensions, simulation domain extent, slope, simulation time, sedimentary composition, sediment grain size, initial water level, flow conditions, constraint type, transport conditions, and geomorphic acceleration factor.

[0008] The sedimentary numerical simulation adopts a constant flow control strategy to ensure the physical consistency of sediment transport during the simulation process; the geomorphological acceleration factor is determined based on the stability of sedimentary geomorphological evolution and the simulation calculation efficiency, and its value ranges from 1 to 10.

[0009] S2. The data from the numerical simulation results of braided river sediments are filtered to select the hydrodynamic data for constructing the training image framework model and dividing the braided river configuration units.

[0010] Specifically, based on the theory of fluvial reservoir configuration, the braided river sedimentary system is divided into four configuration units: braided channel, mid-channel bar, floodplain, and mid-channel bar edge. Seven parameters are selected as data for constructing the training image framework model and as discrimination parameters for the configuration unit division.

[0011] S3. Identify braided river configuration units by inputting hydrodynamic parameter data.

[0012] At least 300 time steps of data on six hydrodynamic parameters were manually labeled to create a labeled training set. The six hydrodynamic parameters from the training set were used as input features to train a random forest classification model. The trained model was then applied to unlabeled data from other time steps to identify braided river morphological units. The model automatically identified the category labels for three types of morphological units: braided channel (code 1), mid-channel bar (code 2), and floodplain (code 3). Mid-channel bar edges (code 4) were defined by identifying locations with a sedimentary thickness of less than 1 m as mid-channel bar edges after determining the distribution range of the mid-channel bars.

[0013] S4. A rule-based method is used to construct a sedimentary body framework model from the thickness surface data of the sedimentary numerical simulation evolution.

[0014] Specifically, all sediment thickness data were converted to GSLIB format and imported sequentially into Petrel software to establish sediment interfaces for each time step. Using a rule-based method, the sediment interfaces were superimposed sequentially in chronological order using the Make Horizon function to construct a three-dimensional sedimentary framework model. This framework model fully preserves information on the dynamic evolution of the bed morphology, channel migration, mid-channel bar composite modification, and erosion-deposition alternation.

[0015] S5. Based on the configuration unit results identified in S3, fill the constructed sedimentary body framework model according to the assignment method, establish the configuration unit model, then orthogonalize it, and output the training image three-dimensional model that conforms to geological laws for use in multi-point geostatistics or deep learning geological modeling.

[0016] Specifically, all configurable unit partitioning data output from the random forest model are imported into Petrel in GSLIB format, and assignment surfaces are constructed using the partitioning data at each time step. An empty configurable unit model is created within the established framework model, and the Assign Values ​​function is used to sequentially assign the configurable unit data from all time steps to the corresponding vertical stratification zones, completing the initial construction of the braided river sand body structure. The Scale Up Property function is used to orthogonalize the model, and the resulting 3D mesh model is the final 3D training image of the braided river sand body structure. By changing the simulation parameters in S1, 3D training images of different sedimentary modes are generated in batches, forming a training image dataset that can be directly imported into multi-point geostatistical modeling software or deep learning modeling networks.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] (1) The present invention generates training images based on numerical simulation of sedimentary hydrodynamics, follows the physical laws of sediment transport, deposition and erosion, and constructs sedimentary bodies layer by layer according to the sedimentary evolution interface under the constraint of a three-dimensional sedimentary framework, which can reproduce the self-organized evolution process of braided river landforms. Compared with traditional methods based on random simulation of geometric parameters or artificial construction, the generated training images can reflect the dynamic processes such as channel migration, mid-channel bar formation and composite transformation, and alternation of erosion and deposition, while preserving the spatial distribution and geomorphological statistical characteristics of the braided river system, and have high geological rationality.

[0019] (2) This invention proposes a quantitative classification rule based on multi-parameter hydrodynamic characteristics, which automatically converts continuous simulation output into discrete configuration units, avoiding the subjective bias of traditional outcrop interpretation or manual annotation, and ensuring that the classification results are repeatable and physically consistent. On this basis, the edge configuration unit of the core bar is introduced to characterize the contact, erosion and superposition interface between core bar bodies of different phases, thereby depicting the internal structure and heterogeneity of the composite core bar, distinguishing the erosion contact relationship between different dam body envelopes, and making up for the deficiency that traditional training images are difficult to characterize the internal heterogeneity of the composite core bar.

[0020] (3) By adjusting sedimentary numerical simulation parameters such as source supply rate, flow rate, and sediment grain size distribution, this invention can generate three-dimensional training images with different sedimentary patterns and configuration combinations in batches, forming a large-scale, label-controllable training image dataset. The training images can provide statistically robust and geologically reasonable prior information for multi-point geostatistical simulations, and can also serve as training samples for deep learning geological modeling, thereby improving the richness and coverage of modeling data and the prediction accuracy of reservoir geological models. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method for constructing training images of braided river sand body structures based on sedimentary numerical simulation according to the present invention.

[0022] Figure 2 This is a planar representation of the deposition numerical simulation output data in an embodiment of the present invention;

[0023] Figure 3 Examples of qualitative and quantitative identification standards for six hydrodynamic parameter identification configuration units in this invention;

[0024] Figure 4 This is an example of the procedure for verifying the angle consistency of braided river configuration units according to an embodiment of the present invention;

[0025] Figure 5 This is an example of using thickness data volume to identify the edge of a braided river mid-slope dam according to an embodiment of the present invention;

[0026] Figure 6This is an embodiment of the invention's intelligent identification process for configuration units based on the random forest algorithm;

[0027] Figure 7 This is a modeling example based on a thickness evolution interface, as described in an embodiment of the present invention.

[0028] Figure 8 The training image construction result is shown in the embodiment of the present invention. Detailed Implementation

[0029] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. Taking the braided river of the Guantao Formation in the XX Oilfield of the Bohai Sea as an example, the solution of this application will be described in conjunction with the accompanying drawings. Figure 1 As shown, the present invention includes the following steps:

[0030] S1. Based on the principle of sedimentary similarity, and according to the geological information of the target braided river reservoir in the study area and the sedimentary information of modern braided rivers with similar sedimentary conditions, numerical simulation parameters for braided river sedimentation are set. These parameters include the simulation domain extent, grid size and resolution, riverbed slope, simulation time, sediment composition, sediment grain size, initial water level, flow conditions, boundary constraint type, sediment transport conditions, and geomorphic acceleration factor. Taking the Guantao Formation braided river reservoir in the XX oilfield of the Bohai Sea as an example, based on previous geological research results, core test data, well logging interpretation data, and modern sedimentary analogy data, various numerical simulation parameters for sedimentation are set, specifically including:

[0031] (1) Setting the initial riverbed morphology and slope. The sedimentary numerical simulation model adopts a planar riverbed sloping along the river direction. The riverbed slope is determined based on the paleoslope data obtained from the paleogeographic reconstruction of the study area, and is placed within the typical slope range of a sandy braided river. The width of the simulation area is kept constant, and no specific initial riverbed bottom shape is preset, so as to reduce the influence of artificially set initial landforms on the simulation results, allowing sedimentary landform units such as braided river channels, mid-channel bars, and sandbars to form and evolve naturally during water flow and sediment transport. In this embodiment, the initial riverbed is set as an sloping plane, the riverbed slope is set to 0.1%, and no initial riverbed bottom shape is preset.

[0032] (2) Setting the simulation area and boundary conditions. Based on the actual distribution scale of braided river sediments in the study area, and combined with the channel width, extension length, and planar distribution characteristics of similar modern braided rivers, the computational domain of the sedimentary numerical simulation is determined. In this embodiment, the length of the computational domain along the river direction is set to 20 km, and the width perpendicular to the river direction is set to 3 km. The upstream boundary along the river direction is set as an open boundary for water flow and sediment input, and the downstream boundary is set as an open boundary for water flow and sediment output; the two sides perpendicular to the river direction are set as fixed inerosible boundaries to restrict the lateral diffusion of water flow and sediment to the simulation area, so that sediments mainly undergo erosion, transportation, accumulation, and deposition within the simulation area. Corresponding to the model plane coordinates, the left and right ends along the X direction are set as open boundaries, and the upper and lower sides along the Y direction are set as fixed inerosible boundaries.

[0033] (3) Set the simulation grid, sediment transport parameters, hydrodynamic conditions, simulation time, and geomorphological acceleration factor. The simulation grid resolution is set to balance computational efficiency and the fineness of the simulation results, so that the simulation results can effectively characterize the morphological features of braided river sandbars extending, migrating, and overlapping in the downstream direction. In this embodiment, a non-uniform grid is used, wherein the grid resolution along the river direction (X direction) is set to 50m, and the grid resolution in the transverse direction (Y direction) is set to 25m.

[0034] The riverbed transported material consisted of non-cohesive sandy sediments and cohesive muddy sediments. The grain size parameters of the sandy sediments were determined based on core grain size analysis data from the study area, and their grain size composition was characterized using a log-normal distribution. In this embodiment, the grain size range of the non-cohesive sandy sediments was set to 0.15–0.9 mm, with a median grain size of 0.3 mm. The upstream sandy sediment input concentration was determined based on modern sedimentary analogy principles, referencing the average annual sediment transport concentration of modern braided sandy rivers with similar sediment grain size, riverbed slope, and hydrodynamic conditions to the study area. In this embodiment, the sediment transport concentration was set to 3 kg / m³. The input concentration of cohesive muddy sediments was determined based on the sand-to-mud ratio obtained from well logging interpretation data in the study area, to ensure that the input ratio of sandy to muddy sediments during the simulation matched the actual sediment composition of the study area.

[0035] The hydrodynamic parameters were calculated and determined by comprehensively considering the sand body thickness information of structural units such as mid-channels, channels, and floodplains obtained from well point statistics in the study area, and in conjunction with the sediment transport initiation rate formula. In this embodiment, the simulated flow rate range was set to 1800–5000 m³ / s, and a constant flow rate input constraint type was adopted to ensure the physical consistency of water flow conditions and sediment transport processes during the simulation. The simulation period was set to 5 years, and the geomorphological acceleration factor was set to 5 to improve the simulation efficiency while maintaining the evolution law of braided river sedimentary geomorphology. The complete settings of various sedimentary numerical simulation parameters are shown in Table 1.

[0036] Table 1. Parameter settings for braided river sedimentation simulation

[0037]

[0038] S2. The numerical simulation results of braided river sediments obtained in step S1 are filtered, and time-series sedimentary thickness data for constructing the training image framework model are extracted, along with data on average flow velocity, bed shear stress, cumulative erosion sedimentation, water depth, bed shear stress direction, and horizontal flow velocity direction for dividing braided river morphological units. In this example, the total simulation duration is 5 years, and the output time step is set to 0.5 days. Each type of hydrodynamic data outputs approximately 3650 time steps. Data cleaning is performed on these 3650 time steps to remove redundant data. Based on the accuracy requirements of morphological unit identification and the computational efficiency of the subsequent random forest algorithm, the data volume is thinned by a coefficient of 5, resulting in 730 time steps for each type of hydrodynamic data. Specifically, this includes:

[0039] (1) Analyze the hydrodynamic characteristics of different braided river configuration units and extract the corresponding sedimentary thickness and hydrodynamic parameter data. Based on the theory of fluvial reservoir configuration, the braided river configuration units are divided into braided channels, mid-channel bars, floodplains, and mid-channel bar edges. Braided channels are characterized by greater water depth, higher flow velocity, stronger bed shear stress, and more consistent flow direction; mid-channel bars are characterized by shallower water depth, moderate flow velocity and bed shear stress, and are dominated by sedimentation; floodplains are characterized by lower water depth, lower flow velocity, and lower bed shear stress, and are dominated by low-energy sedimentation; mid-channel bar edges are located between mid-channel bars of different phases and are used to characterize the erosion superposition interface between bars, characterized by thin thickness and distribution controlled by the erosion relationship of mid-channel bars.

[0040] Based on the aforementioned hydrodynamic characteristics, data on sedimentary thickness, average flow velocity, bed shear stress, cumulative erosion sedimentation, water depth, horizontal flow velocity direction, and bed shear stress direction were extracted for each time step. Sedimentary thickness data was used to construct the training image framework model, while the remaining parameters were used to identify braided river morphological units. Taking time step 1330 as an example... Figure 2 As shown, Figure 2 The AG uses this as data for sediment thickness, average flow velocity, bed shear stress, cumulative erosion sedimentation, average water depth, horizontal flow velocity direction, and bed shear stress direction.

[0041] (2) Establish quantitative classification rules for braided river configuration units. First, the simulation results of 300 time steps are manually interpreted to obtain pre-classification labels for braided channels, mid-channel bars, floodplains, and mid-channel bar edges. Then, the hydrodynamic parameter characteristics corresponding to each configuration unit are statistically analyzed to determine the following classification criteria:

[0042] Average flow velocity was used as a threshold of 1 m / s. Areas with an average flow velocity greater than or equal to 1 m / s were preferentially classified as braided channels, while areas with an average flow velocity less than 1 m / s were classified as mid-channel bars or floodplains based on other parameters. Figure 3 As shown in Figure A. The bed shear stress was set at 1.5 N / m² as the threshold for distinguishing braided channels from mid-channel bars, and 0.5 N / m² as the threshold for identifying floodplains. Areas with high bed shear stress were preferentially classified as braided channels, and the abrupt changes or convergence points between high and low stress areas were used to identify the edges of mid-channel bars, such as... Figure 3 As shown in Figure B. The cumulative erosion and deposition amount is used as the threshold for distinguishing between erosion and deposition zones, with areas less than 0m primarily classified as braided channels, and areas greater than 0m classified as mid-channel bars or floodplains based on other parameters; 3m is used as an auxiliary threshold for floodplains, as shown in Figure B. Figure 3 As shown in Figure C. A water depth of 1.5m is used as the threshold for distinguishing braided channels from mid-channel bars or floodplains. Areas with a depth greater than or equal to 1.5m are preferentially classified as braided channels, while areas less than 1.5m are further subdivided based on other parameters, such as... Figure 3 As shown in D.

[0043] Furthermore, the direction of bed shear stress and the direction of horizontal flow velocity are superimposed for analysis. Let the angle difference between the two be θ, and the angle consistency index I is calculated according to the following formula: I = 1 - sinθ. When I is 0.7–1.0, it is preferentially classified as a braided channel; when I is 0–0.3, it is preferentially classified as a mid-channel bar or mid-channel bar edge; when I is greater than 0.3 and less than 0.7, a comprehensive classification is made by combining average flow velocity, bed shear stress, cumulative erosion and deposition, and water depth, such as... Figure 3 China E and Figure 4 As shown. Based on the above multi-parameter quantitative rules, the classification results of braided river morphological units at each time step are obtained.

[0044] S3. Based on the hydrodynamic parameter classification criteria determined in step S2, the simulation results of some time steps are manually interpreted and training labels are created. A random forest algorithm is then used to classify the grid data for all time steps to obtain the time series identification results of braided river morphological units, specifically including:

[0045] (1) Constructing Random Forest Training Samples. Fifty time steps were selected from each hydrodynamic parameter data, for a total of 300 time steps, for manual interpretation. Three types of structural units—channel, mid-channel bar, and floodplain—were labeled to form a training label set Y. The average flow velocity, average bed shear stress, erosion, sedimentation, average water depth, and the consistency between the flow velocity direction and the bed shear stress direction corresponding to each grid unit were used as feature parameters to construct random forest training samples, such as... Figure 5 As shown.

[0046] (2) Identifying channels, bar centers, and floodplains. The random forest model was trained using the training samples, and the trained random forest model was applied to all time steps without manual annotation. For each grid cell in each time step, its six feature parameters were input into the random forest model, and the corresponding configuration cell category was output. Among them, channels have the characteristics of high flow velocity, high bed shear stress, net erosion or weak deposition, and high flow direction consistency, and are labeled with code 1; bar centers have the characteristics of medium to low flow velocity, low bed shear stress, strong deposition, weak erosion, and low flow direction consistency, and are labeled with code 2; floodplains have the characteristics of low flow velocity, low bed shear stress, net deposition, and large flow direction disturbance, and are labeled with code 3.

[0047] (3) Identifying the edges of core bars. Within the core bar area identified by random forest, the superimposed interfaces between core bars of different phases are identified using sedimentary thickness data. For example... Figure 6 As shown, the sedimentation thickness threshold was set to 1m, and grid cells with a sedimentation thickness of less than or equal to 1m inside the midstrip bar were classified as midstrip bar edges and labeled with code 4. Finally, the classification results of braided river configuration units for all time steps were obtained.

[0048] S4. Import the time-series sedimentary thickness data obtained in step S2 into Petrel software. Using a method of layering sedimentary interfaces sequentially over time, construct a three-dimensional sedimentary framework model of the braided river training images. Specifically, this includes:

[0049] (1) Determine the construction data for the sedimentary body framework model. The algebraic sum of the sedimentary increment and erosion within each time step is used as the sedimentary thickness for that time step, which is used to characterize the net change in the bed morphology. The sedimentary thickness surface at each time step is used as the sedimentary interface at the corresponding time, recording changes in channel migration, mid-channel bar formation, and erosion deposition.

[0050] (2) Establish time series deposition interfaces. Extract deposition thickness data for 730 time steps, convert them to GSLIB format and import them into Petrel software in sequence. Establish corresponding Horizon surfaces based on the deposition thickness data for each time step.

[0051] (3) Constructing a three-dimensional sedimentary body framework model. Following the chronological order of the time steps, the MakeHorizons function in Petrel software is used to load each Horizon plane sequentially, and these planes are stacked layer by layer along the vertical direction to form a complete three-dimensional sedimentary body framework model, such as... Figure 7 As shown, the framework model preserves spatial information on substrate morphology evolution, channel migration, mid-channel bar reconstruction, and alternating erosion and deposition.

[0052] S5. Import the time-series morphological unit identification results obtained in step S3 into Petrel software and assign them to the sedimentary framework model established in step S4 to construct a three-dimensional braided river morphological unit model; then perform mesh orthogonalization processing on the model to obtain the final braided river training image, specifically including:

[0053] (1) Establish configuration unit assignment surfaces. Convert the 730 time-step configuration unit classification data output by the random forest model into GSLIB format and import them into Petrel software in chronological order to establish configuration unit assignment surfaces corresponding to each time step.

[0054] (2) Construct a three-dimensional configuration unit model. In the sedimentary body framework model established in step S4, a blank configuration unit attribute model is established through the Geometrical module, and the Assign Values ​​function is used to assign the configuration unit assignment surfaces of each time step to the corresponding sedimentary segments in sequence, forming a three-dimensional braided river configuration unit model.

[0055] (3) Generate braided river training images. The Scale UpProperty function in the Property Modeling module of Petrel software is used to orthogonalize the mesh of the morphological unit model. The orthogonalized mesh size is set to 1000×128×64, and the mesh parameters are determined according to the study area size and the morphological unit representation accuracy. The generated three-dimensional mesh model is the braided river training image, such as... Figure 8 As shown.

[0056] By adjusting simulation parameters such as the source supply rate, flow rate, and sediment grain size distribution in step S1, three-dimensional training images with different sedimentary characteristics can be generated in batches, forming a training image dataset. These training images can be used for multi-point geostatistical modeling or deep learning reservoir modeling.

[0057] Through the above specific implementation scheme, the present invention realizes an intelligent construction method for braided river training images based on sedimentary numerical simulation, which reduces the influence of human subjective factors and provides important technical support for the exploration and development of braided river reservoirs.

[0058] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the protection scope of the claims and specification of the present invention.

Claims

1. A method for constructing training images of braided river sand body structures based on sedimentary numerical simulation, characterized in that, Includes the following steps: S1. Set the simulation parameters for the sedimentation numerical simulation, run the sedimentation numerical simulation, and obtain the time-series hydrodynamic and sedimentation outputs; S2. The data from the numerical simulation results of braided river sedimentation are filtered to select the hydrodynamic data for constructing the training image framework model and dividing the braided river configuration units; Based on the theory of river configuration hierarchy, the braided river sedimentary system is divided into four configuration units: braided channel, mid-channel bar, mid-channel bar edge, and floodplain. Seven parameters are selected as data for constructing the training image framework model and as discrimination parameters for configuration unit division: sedimentary thickness, average flow velocity, maximum bed shear stress, cumulative net sedimentation, water depth, flow velocity direction, and shear stress direction. S3. Identify braided river morphological units by inputting hydrodynamic parameter data; manually label six types of hydrodynamic parameter data from at least 300 time steps to establish a label training set; use the six types of hydrodynamic parameter data in the training set as input features to train a random forest classification model; apply the trained model to unlabeled data from other time steps to identify braided river morphological units; automatically identify the category labels of three types of morphological units: braided channel (code 1), mid-channel bar (code 2), and floodplain (code 3); and define the mid-channel bar edge (code 4) as the edge of a mid-channel bar after identifying the distribution range of the mid-channel bar. S4. A rule-based method is used to first construct a sedimentary body framework model from the thickness surface data of the sedimentary numerical simulation evolution. S5. Based on the configuration unit results identified in S3, fill the constructed sedimentary body framework model according to the assignment method, establish the configuration unit model, and then orthogonally output the training image three-dimensional model that conforms to geological laws.

2. The method for constructing training images of braided river sand body structures based on sedimentary numerical simulation according to claim 1, characterized in that, The simulation parameters described in S1 include grid size, dimensions, simulation domain range, slope, simulation time, sedimentary composition, sediment grain size, initial water level, flow conditions, constraint type, transport conditions, and geomorphic acceleration factor.

3. The method for constructing training images of braided river sand body structures based on sedimentary numerical simulation according to claim 1, characterized in that, The sedimentation numerical simulation described in S1 employs a constant flow control strategy to ensure the physical consistency of sediment transport during the simulation process.

4. The method for constructing training images of braided river sand body structures based on sedimentary numerical simulation according to claim 2, characterized in that, The geomorphic acceleration factor is determined based on the stability of sedimentary geomorphic evolution and the efficiency of simulation calculation, and its value ranges from 1 to 10.

5. The method for constructing training images of braided river sand body structures based on sedimentary numerical simulation according to claim 1, characterized in that, The specific steps of S4 are as follows: convert the deposition thickness data into GSLIB format, import it into Petrel software in sequence to create the deposition interface for each time step, and use the Make Horizon function to overlay the deposition interfaces in chronological order to construct a three-dimensional depositional body framework model.

6. The method for constructing training images of braided river sand body structures based on sedimentary numerical simulation according to claim 1, characterized in that, The specific steps for S5 are as follows: import all configuration unit partitioning data output by the random forest model into Petrel in GSLIB format, construct assignment surfaces based on the partitioning data at each time step, create empty configuration unit models within the established framework model, and use the Assign Values ​​function to sequentially assign configuration unit data from all time steps to the corresponding vertical layer zone to complete the initial construction of the three-dimensional braided river configuration unit. Use the Scale Up Property function to orthogonalize the model, and the generated three-dimensional mesh model is the final braided river training image. By changing the simulation parameters in S1, three-dimensional training images of different sedimentation modes are generated in batches.

Images