A method and system for imaging complex geologic bodies based on adaptive mesh refinement

Abstract

This invention discloses a method and system for imaging complex geological bodies based on adaptive grid partitioning. The method includes: intelligent sensors acquiring multi-physics observation data of the target area and outputting time stamps, data quality markers, and discrete information on observation directions; constructing an initial geological skeleton and an initial grid model based on the multi-physics observation data; generating interface protection zones and identifying boundary offset corridors; establishing evidence accumulation states to filter cells to be updated; determining the grid partitioning direction and performing adaptive grid partitioning; performing topology locking review; and then performing grid updates and imaging updates to obtain the imaging results of the complex geological body. This method solves the problems of unstable grid updates near complex interfaces, misalignment of multi-physics boundaries, and easy loss of deep anomalies in multi-scale imaging.

Description

Technical Field

[0001] This invention relates to the field of geological bodies, and more specifically, to a method and system for imaging complex geological bodies based on adaptive grid subdivision. Background Technology

[0002] Imaging of complex geological bodies is typically based on geophysical exploration and iterative inversion. Seismic methods are more sensitive to abrupt changes in impedance and velocity, electromagnetic methods are better at reflecting conductivity anomalies and their extent, and gravity and magnetic methods are better suited for characterizing regional anomalies caused by density or magnetic distributions. For complex targets such as faults, thin layers, pinch-out bodies, lenses, and fracture zones, a common practice is to first establish an initial model based on multiphysics observation data, and then iteratively update it to gradually approximate the subsurface structure. When the model exhibits strong local variations, mesh refinement is usually introduced to improve local descriptive capabilities. Existing technologies have demonstrated that joint interpretation of multiphysics data, well data and seismic horizon constraints, and mesh tomography inversion can all be used for subsurface structure imaging or velocity modeling.

[0003] Among existing patents, patent document CN110058317B, entitled "Joint Inversion Method of Airborne Transient Electromagnetic Data and Airborne Magnetotelluric Data," discloses a scheme for combining airborne transient electromagnetic data and airborne magnetotelluric data. Its main contents include performing two-dimensional inversion on the airborne magnetotelluric data, gridding the results to form conductivity grid data, constructing a one-dimensional inversion model of the airborne transient electromagnetic data, and outputting the joint interpretation results. This patent illustrates the application of multi-physics data joint inversion in the interpretation of subsurface media.

[0004] Among existing patents, patent document CN111077575A, entitled "A Depth Domain Velocity Modeling Method and Device," discloses a velocity modeling scheme that combines well data, seismic stratigraphic information, and grid tomographic inversion. Its main contents include establishing a depth domain stratigraphic velocity model using three-dimensional pre-stack time migration velocity, introducing sonic logging data into the well-containing area, and optimizing the initial depth domain velocity model using grid tomographic inversion in conjunction with seismic interpretation stratigraphic layers. This patent illustrates a common technical approach to improving the accuracy of subsurface models using seismic data, well logging data, and grid tomographic inversion.

[0005] However, existing technologies still face several significant challenges in imaging complex geological bodies.

[0006] First, mesh updates near complex interfaces are easily affected by local residual fluctuations and change repeatedly. The same boundary may fall on different cells in different iterations, which can lead to unstable boundary morphology. In severe cases, problems such as false connectivity of anomalous bodies, false splitting of anomalous bodies, and inaccurate judgment of fault continuity may occur.

[0007] Second, the response locations of different physical methods to the boundary of the same geological body do not necessarily coincide. Some are closer to abrupt change surfaces, while others are closer to diffusion zones or transition zones. If these boundaries are still forcibly pressed into the same refinement area, it is easy to cause local areas to be incorrectly refined, or to misjudge the original transition zone with width as a single sharp interface.

[0008] Third, complex geological bodies commonly contain thin, elongated, steeply dipping, and pinch-out targets. If isotropic subdivision is uniformly applied to candidate anomaly areas, regions that would otherwise only require improved resolution in one direction are often divided into numerous small units. This not only increases the discrete scale but also, during multi-scale imaging, early-identified deep anomalies may be masked by strong shallow responses in later stages, resulting in poor continuity of deep targets. These problems directly affect the stability, consistency, and structural representation capabilities of imaging results for complex geological bodies. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a method and system for imaging complex geological bodies based on adaptive grid subdivision, so as to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: An imaging method for complex geological bodies based on adaptive grid partitioning is disclosed. The method involves: intelligent sensors acquiring multi-physics observation data of the target area and performing time synchronization processing on the multi-physics observation data, outputting time stamps, data quality markers, and observation direction discrete information corresponding to the multi-physics observation data; constructing an initial geological skeleton and an initial grid model based on the multi-physics observation data; generating interface protection zones around candidate interfaces in the initial geological skeleton and identifying boundary offset corridors based on candidate boundary zones corresponding to multiple physical methods; establishing evidence accumulation states for grid cells within the interface protection zones and the boundary offset corridors, and filtering cells to be updated based on the evidence accumulation states; determining the grid partitioning direction based on the observation direction discrete information of the cells to be updated, and performing adaptive grid partitioning according to the grid partitioning direction, setting ancestor cell markers for anomalous bearing cells identified in the coarse-scale stage; performing topology locking review on candidate grid operations before performing grid update and imaging update on candidate grid operations that pass the topology locking review, thereby obtaining the imaging results of the complex geological body.

[0011] Preferably, the intelligent sensor includes a seismic sensor, an electromagnetic sensor, a gravity sensor, a magnetic sensor, and a well-drilled sensor; the seismic sensor is used to detect seismic wave response data, the electromagnetic sensor is used to detect electromagnetic field response data, the gravity sensor is used to detect gravity anomaly data, the magnetic sensor is used to detect magnetic anomaly data, and the well-drilled sensor is used to detect well-drilled physical property measurement data.

[0012] Preferably, the intelligent sensor outputs a data quality marker, a time identifier, and discrete observation direction information associated with the corresponding multi-physics observation data; the data quality marker is used to characterize the data quality status of the corresponding multi-physics observation data, the time identifier is used to characterize the acquisition time sequence of the corresponding multi-physics observation data, and the discrete observation direction information is used to characterize the distribution of observation directions passing through the grid cells.

[0013] Preferably, the step of generating an interface protection zone around the candidate interface in the initial geological framework specifically includes generating an interface protection zone unit set based on the spatial distribution of grid units adjacent to the candidate interface, and limiting subsequent refinement, coarsening, and boundary relocation operations to be performed within the area covered by the interface protection zone unit set and the boundary offset corridor.

[0014] Preferably, the step of identifying the boundary offset corridor based on candidate boundary bands corresponding to multiple physical methods specifically includes extracting candidate boundary bands corresponding to multiple physical methods, comparing two spatially adjacent candidate boundary bands, and determining the area between the two candidate boundary bands as the boundary offset corridor when the distance between the two candidate boundary bands is within a preset distance range and the extension direction meets a preset consistency condition, and marking the first side sub-region, the transition sub-region, and the second side sub-region within the boundary offset corridor.

[0015] Preferably, establishing the evidence accumulation state for the grid cell specifically includes recording the local residual distribution of the grid cell to be judged, the differences in physical properties between adjacent grid cells, the location of candidate boundaries in the neighborhood of the grid cell to be judged, and the spatial orientation of the grid cell to be judged by multiple physical methods in each iteration round, and updating the evidence accumulation state of the grid cell to be judged in each iteration round; and determining the grid cell whose evidence accumulation state meets the preset accumulation conditions as the cell to be updated.

[0016] Preferably, the adaptive meshing based on the observation direction discrete information of the unit to be updated specifically includes determining the meshing direction according to the observation direction discrete information of the unit to be updated, and dividing the unit to be updated into multiple sub-units along the meshing direction; for the unit to be updated located in the boundary offset corridor, adaptive meshing is performed along the line connecting two candidate boundary zones or the local construction normal direction; for anomalous bearing units with ancestor unit markers, at least one descendant unit is retained to participate in the update in subsequent mesh updates.

[0017] This invention also discloses an imaging system for complex geological bodies based on adaptive grid subdivision for implementing the above-mentioned method. The system comprises an intelligent sensor collaborative acquisition module, an initial skeleton construction module, an interface sheath generation module, a boundary offset corridor identification module, an evidence accumulation and screening module, a directional subdivision and ancestor unit inheritance module, a topology locking review module, and an imaging update module. The intelligent sensor collaborative acquisition module is used to acquire multi-physics observation data of the target area and perform time synchronization processing on the multi-physics observation data, outputting time identifiers, data quality markers, and observation direction discrete information corresponding to the multi-physics observation data. The initial skeleton construction module is used to construct an initial skeleton based on the multi-physics observation data. The system comprises a geological skeleton and an initial mesh model; an interface protection generation module for generating interface protection around candidate interfaces; a boundary offset corridor identification module for identifying boundary offset corridors based on candidate boundary zones corresponding to multiple physical methods; an evidence accumulation and screening module for establishing evidence accumulation status and screening units to be updated; a directional subdivision and ancestor unit inheritance module for performing adaptive mesh subdivision based on the observation direction discrete information of the units to be updated and setting ancestor unit tags for anomalous bearing units; a topology locking review module for performing topology locking review on candidate mesh operations; and an imaging update module for performing mesh update and imaging update on candidate mesh operations that have passed the topology locking review, thereby obtaining imaging results of complex geological bodies.

[0018] Preferably, the intelligent sensors in the intelligent sensor collaborative acquisition module include seismic sensors, electromagnetic sensors, gravity sensors, magnetic sensors, and well sensors; the seismic sensors are used to detect seismic wave response data, the electromagnetic sensors are used to detect electromagnetic field response data, the gravity sensors are used to detect gravity anomaly data, the magnetic sensors are used to detect magnetic anomaly data, and the well sensors are used to detect well physical property measurement data.

[0019] Preferably, the topology locking review module is used to detect anomalous body connectivity, fault continuity, sequence adjacency, and ancestor unit inheritance before performing refinement, coarsening, cell affiliation adjustment, and boundary relocation operations, and to block the corresponding candidate mesh operation when any detection result does not meet the preset locking conditions.

[0020] The advantages of this invention over existing technologies lie in its utilization of several objective technical principles in imaging complex geological bodies. Real geological interfaces typically exhibit spatial continuity and sustained response across multiple iterations, while false boundaries induced by local noise or transient errors often appear discrete, short-lived, and abrupt. Different physical methods exhibit stable shifts in their boundary responses to the same geological body; these shifts do not necessarily indicate interpretative conflicts but may reflect natural spatial misalignments of different physical property fronts. Simultaneously, thin, elongated subsurface targets and deep anomalies exhibit directional and scale differences in observational coverage. Anomaly locations appearing at coarse-scale stages often possess a priori continuity significance, and subsequent stages should not be completely covered by strong local responses from shallow areas.

[0021] Based on the above principles, this invention collects multi-physics observation data using intelligent sensors and outputs time stamps, data quality markers, and discrete observation direction information. Based on this, an initial geological framework and initial grid model are constructed. Interface protection zones are generated around candidate interfaces, and boundary offset corridors corresponding to multiple physical methods are identified. An evidence accumulation state is established within the interface protection zones and boundary offset corridors to filter units to be updated. Then, the grid partitioning direction is determined based on the discrete observation direction information, and grid and imaging updates are performed in conjunction with ancestor unit marking and topology locking review. In this way, grid updates are no longer directly triggered by single-round local fluctuations, but are based on continuous evidence, spatial range constraints, and structural relationship review. This allows for simultaneous handling of model instability issues caused by complex interface jitter, multi-physics boundary misalignment, and competition between shallow and deep scales, making the imaging process of complex geological bodies more suitable for representing complex interfaces, transition zones, and thin, elongated targets.

[0022] Furthermore, this invention employs a smart sensor system comprised of seismic sensors, electromagnetic sensors, gravity sensors, magnetic sensors, and borehole sensors, which output data quality markers, time stamps, and discrete observation direction information corresponding to multi-physics observation data. This allows for simultaneous consideration of observation sources, acquisition sequence, data quality, and directional coverage during the imaging process, avoiding the reliance on a single physical method or single observation result to judge local grids. This further improves the reliability of selecting units to be updated and enhances the adaptability of imaging complex geological bodies to differences in multi-source observations.

[0023] Furthermore, this invention identifies candidate boundary zones corresponding to multiple physical methods and determines boundary offset corridors when the spacing and extension direction conditions are met. Within this region, a first lateral sub-region, a transitional sub-region, and a second lateral sub-region are then defined. This explicitly preserves the stable offsets between different physical boundaries, avoiding simply compressing them into a single boundary. This further alleviates the problems of over-refinement and under-refinement, and enhances the imaging representation of complex geological bodies with transitional, diffusional, or multi-frontal characteristics. Attached Figure Description

[0024] Figure 1 This is an overall flowchart of a complex geological body imaging method based on adaptive grid partitioning, demonstrating the complete core execution steps from multi-physics data acquisition to final image update.

[0025] Figure 2 This is a diagram of the imaging system architecture for complex geological bodies, showing the system composition, which includes eight core functional modules such as sensor collaborative acquisition, initial skeleton construction, and interface sheath generation.

[0026] Figure 3 It is a representative image of a real-world scenario where intelligent sensors collaboratively collect physical data, depicting a physical scenario in which earthquake, electromagnetic, gravity, and magnetic sensors deployed on the Earth's surface, as well as sensors deployed in wells, jointly transmit multi-physics observation data to a data processing center.

[0027] Figure 4 This is a schematic diagram of the initial geological framework and initial grid model, showing the basic two-dimensional grid constructed in the study area and the initial geological body boundary skeleton line running through it.

[0028] Figure 5 This is a schematic diagram of the interface sheath generation principle, highlighting the set of interface sheath units distributed in a band around the candidate geological interface, which are used to define the area for subsequent refinement and coarsening operations.

[0029] Figure 6 It is a boundary offset corridor identification structure diagram, which shows in detail the offset corridor area between two adjacent candidate boundary zones, and clearly delineates the first side sub-region, the transition sub-region, and the second side sub-region.

[0030] Figure 7 It is a logic diagram for the accumulation and screening of evidence in grid cells, which illustrates the process of updating and screening grid cells in multiple rounds by integrating various conditions such as local residuals, differences in adjacent physical properties, boundary location, and spatial orientation.

[0031] Figure 8 This is a diagram of the ancestral cell inheritance mechanism of the anomalous bearing cell. It shows that after the marked ancestral cell is refined by mesh subdivision in the coarse-scale stage, at least one marked cell is still retained in the descendant cell system to maintain the anomalous characteristics. Detailed Implementation

[0032] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. The embodiments described herein are intended to help those skilled in the art understand the technical solutions, motivations, and implementation paths of the present invention, and are not intended to limit the scope of protection of the present invention. Without departing from the concept of the present invention, those skilled in the art can make several equivalent substitutions or conventional modifications, and these equivalent substitutions or conventional modifications should all be considered to fall within the technical concept of the present invention.

[0033] This invention addresses imaging scenarios involving complex geological bodies. These complex geological bodies primarily refer to faults, thin layers, pinch-out bodies, lenses, dikes, fracture zones, and subsurface structures where multiple anomalies are adjacent and interfere with each other. Traditional methods often encounter three direct problems in such scenarios. First, the mesh near the boundary repeatedly refines and regresses due to local residual fluctuations, leading to unstable interface positions. Second, boundary positions identified by different physical methods exhibit stable offsets; forcibly merging these boundaries into a single boundary can easily misinterpret the true transition zone as a single sharp interface. Third, thin, elongated anomalies have varying resolvability in different directions; uniformly using isotropic partitioning not only increases the number of elements but also easily leads to the loss of early-identified deep anomalies in subsequent high-resolution stages. This invention addresses these problems by organizing an imaging workflow comprised of intelligent sensor collaborative acquisition, controlled boundary updates, directional partitioning, and topological constraints.

[0034] like Figure 1 As shown, the method of this invention generally includes the following steps. First, intelligent sensors collect multi-physics observation data of the target area and perform time synchronization processing on the multi-physics observation data, outputting time stamps, data quality markers, and observation direction discrete information corresponding to the multi-physics observation data. Then, an initial geological skeleton and an initial grid model are constructed based on the multi-physics observation data. Next, interface protection zones are generated around candidate interfaces in the initial geological skeleton, and boundary offset corridors are identified based on candidate boundary zones corresponding to multiple physical methods. On this basis, evidence accumulation states are established for grid cells within the interface protection zones and boundary offset corridors, and cells to be updated are selected based on the evidence accumulation states. Subsequently, the grid partitioning direction is determined based on the observation direction discrete information of the cells to be updated, and adaptive grid partitioning is performed according to this direction, setting ancestor cell markers for anomalous bearing cells identified in the coarse-scale stage. Finally, before performing grid updates, topology locking review is performed on candidate grid operations, and grid updates and imaging updates are performed on candidate grid operations that pass the topology locking review to obtain imaging results of complex geological bodies.

[0035] The process in this invention does not simply string together multiple steps, but rather uses the result of each step as an input constraint for the next. This design is because imaging complex geological bodies cannot be accomplished solely based on a single inversion result; it requires considering the spatial location of boundaries, the persistence of evidence, the rationality of the subdivision direction, and the stability of geological topological relationships. Only in this way can the final imaging result be not only a reasonable result in terms of data fitting, but also a stable result in terms of structural representation.

[0036] To facilitate a precise understanding of this invention by those skilled in the art, several terms used in this invention will be clearly explained first.

[0037] The initial geological framework refers to the set of basic structures extracted from multiphysics observation data in the early stages of imaging, which characterizes the main subsurface boundaries. In a two-dimensional scene, the initial geological framework can be represented by fault lines, stratigraphic boundaries, anomaly edges, and well control lines. In a three-dimensional scene, the initial geological framework can be expanded into a set of corresponding structural surfaces. The initial geological framework is not the final interpretation result, but rather the basic reference structure upon which subsequent interface protection zone generation, boundary offset corridor identification, and topology locking review are based.

[0038] An interface buffer zone refers to a controlled, band-shaped region surrounding a candidate interface. Its purpose is not to fix the candidate interface in a fixed position, but rather to limit subsequent boundary-related refinement, coarsening, and relocation operations to primarily occur within this region. This avoids invalid updates far from the boundary due to accidental fluctuations in local residuals. Essentially, the interface buffer zone provides a spatial buffer for the candidate interface, allowing for local migration.

[0039] A candidate boundary zone refers to the spatial response region of a subsurface boundary as determined by a specific physical method. Since many geophysical methods do not produce an infinitely thin line response to the boundary, but rather a band of a certain width, this invention uses candidate boundary zones to describe this phenomenon. In practical implementation, candidate boundary zones can originate from seismic velocity abrupt change zones, electromagnetic low-resistivity change zones, gravity gradient zones, magnetic anomaly gradient zones, etc.

[0040] A boundary offset corridor refers to the region between two spatially adjacent candidate boundary zones that extend in the same direction but are not completely coincident in location. This invention does not simply view this offset as a conflict, but rather as stable misalignment information between multi-physical boundaries. This region can be further subdivided into a first lateral sub-region, a transition sub-region, and a second lateral sub-region, to respectively bear the constraints imposed on the boundary by different physical methods.

[0041] The evidence accumulation state refers to the state description formed over multiple consecutive iterations regarding whether the grid cell to be judged continuously exhibits boundary activity characteristics. This state is not triggered by a single local anomaly, but rather is accumulated from the local residual distribution, differences in physical properties between adjacent cells, candidate boundary placement, and multi-physical spatial orientation over multiple iterations. The purpose of using this state quantity is to distinguish between continuously occurring true boundary evidence and incidentally occurring false boundary evidence.

[0042] Ancestor markers are inheritance identifiers applied to anomalous load-bearing elements during the coarse-scale stage. Anomalous load-bearing elements are mesh elements that have already demonstrated clear anomalous load-bearing capacity in the coarse-scale stage, such as elements with stable anomalous properties, high boundary activity, or significant anomalous responses across multiple consecutive rounds. After setting the ancestor marker, at least one marked element is retained in the descendant elements formed in subsequent refinement, ensuring that the anomalous information from the coarse-scale stage is not completely overwritten by shallow strong responses in the subsequent high-resolution stages.

[0043] Topology locking review refers to the process of examining whether a candidate grid operation will disrupt subsurface structural relationships before execution. These relationships include anomaly connectivity, fault continuity, sequence adjacency, and ancestor unit inheritance. The purpose of topology locking is to prevent geologically illogical model changes driven by data fitting. For example, two originally separate anomalies should not be mistakenly merged into a single entity simply due to local residual changes, and a continuous fault should not be severed without sufficient evidence.

[0044] Observation direction discretization information refers to the distribution of observation directions passing through a given grid cell. It doesn't simply record whether observations were observed, but rather which directions observations entered the cell from, whether the coverage in each direction was balanced, and whether there was a significant directional bias. This information guides subsequent orientation subdivision. Because the strength of observation support varies across different directions within a cell, ignoring these differences and uniformly subdividing it can easily lead to invalid discretization and the accumulation of directional errors.

[0045] like Figure 3 As shown, seismic sensors, electromagnetic sensors, gravity sensors, magnetic sensors, and well-drilled sensors are deployed within the target area. Multiple sensor nodes transmit data to the data processing center via wired networks, wireless networks, or a dedicated well site acquisition bus. The data processing center can be a single workstation, a server cluster, or an industrial computing platform with storage and parallel computing capabilities.

[0046] In one embodiment, seismic sensors are used to acquire seismic wave response data, specifically including reflected waves, refracted waves, direct waves, microseismic waveforms, or tomographic travel time data. Electromagnetic sensors are used to acquire electric field response data and magnetic field response data, suitable for artificial source electromagnetic, natural field electromagnetic, or transient electromagnetic scenarios. Gravity sensors are used to acquire gravity anomaly data, and magnetic sensors are used to acquire magnetic anomaly data. Downhole sensors are used to acquire downhole physical property measurement data, which may include downhole resistivity, downhole acoustic transit time, downhole density, downhole natural gamma, or downhole inclination trajectory information. Through the joint deployment of these sensors, seismic methods can more sensitively detect velocity or impedance abrupt changes, electromagnetic methods can more sensitively detect conductive anomalies, gravity and magnetic methods are better suited to reflect large-scale density and magnetic variations, and downhole sensors provide local high-precision control information. The purpose of this arrangement is to allow different physical methods to jointly constrain complex geological bodies from different perspectives.

[0047] The sensor in this invention is called an intelligent sensor because its output includes not only the raw observation values, but also the corresponding time stamp, data quality marker, and discrete information about the observation direction. In other words, the metadata required for subsequent imaging updates is provided during the acquisition phase, rather than being passively estimated during post-processing.

[0048] In one embodiment, time synchronization is achieved through a unified time reference. Surface sensors preferably use satellite timing or master clock pulse synchronization, while in-well sensors preferably use a combination of downhole clock synchronization and uplink time correction. For high-frequency seismic signals, the time synchronization error is preferably controlled within the range of 1 to 5 milliseconds. For low-frequency electromagnetic, gravity, and magnetic signals, the time synchronization error can be controlled within the range of 10 to 100 milliseconds. If there are significant differences in the field sampling frequencies, resampling is performed on a unified time axis. Resampling can employ linear interpolation, piecewise spline interpolation, or local smoothing interpolation. The reason for setting a unified time synchronization is that only after multi-physics observation data are aligned temporally can subsequent boundary space comparisons and evidence accumulation be comparable.

[0049] In one embodiment, data quality markers are represented by continuous values ​​from 0 to 1. The closer the value is to 1, the more suitable the observation data is as input for subsequent candidate boundary zone extraction and evidence accumulation. For seismic response data, data quality markers can be generated based on signal-to-noise ratio, first-wave clarity, bad channel ratio, and waveform continuity. For electromagnetic response data, data quality markers can be generated based on baseline drift, phase stability, repeated observation error, and field value jumps. For gravity anomaly and magnetic anomaly data, data quality markers can be generated based on instrument drift correction residuals, repeated measurement point differences, background noise levels, and the degree of external interference. For wellbore physical property measurement data, data quality markers can be generated based on well depth matching error, well section continuity, and measurement curve stability. In one embodiment, data with a data quality marker below 0.4 is not included in the candidate boundary zone extraction process; data between 0.4 and 0.7 is only used as an auxiliary reference; and data with a data quality marker of 0.7 or higher can directly participate in evidence accumulation. This grading aims to prevent low-quality observations from causing excessive interference in boundary identification.

[0050] In one embodiment, the discrete information of the observation direction is obtained through directional sector statistics. Taking a two-dimensional profile as an example, 0 degrees to 180 degrees can be divided into 8, 12, or 16 directional sectors. For seismic data, the number of sectors is counted based on the ray direction from the source to the receiver or the local tomographic sensitive direction. For electromagnetic data, the number of sectors is counted based on the main sensitive direction from the source to the receiver, the local field gradient direction, or the main direction of the sensitivity kernel. For gravity and magnetic data, the number of sectors is counted based on the local anomaly gradient direction, the main response axis direction, or the main direction of the local sensitivity kernel. For well-mounted sensors, the local directional distribution is counted based on the well trajectory crossing direction and the well-side profile direction. In one embodiment, 12 directional sectors are used for medium-scale two-dimensional imaging scenes, and 16 or 24 directional sectors are used for high-precision three-dimensional imaging scenes. The reason for using discrete directional information is that subsequent segmentation directions cannot be determined solely by boundary positions; it is also necessary to consider whether the support of the observations in each direction is balanced.

[0051] like Figure 4 As shown, after completing the acquisition and preprocessing of multi-physics observation data, the initial geological framework and initial grid model of the study area are constructed.

[0052] In one embodiment, the initial mesh model uses a two-dimensional regular rectangular mesh. Regular rectangular meshes facilitate recording cell adjacency relationships, parent-child inheritance relationships, and local refinement relationships, making them suitable for most two-dimensional profile scenarios. In other embodiments, quadtree meshes, triangular unstructured meshes, or hybrid meshes can also be used, as long as the spatial adjacency and inheritance relationships between cells can be clearly represented later. If three-dimensional imaging is used, the two-dimensional cells can be expanded into volume cells, and the mesh structure can use octree meshes, hexahedral meshes, or tetrahedral meshes.

[0053] The initial grid scale should match the exploration resolution. If the initial grid is too coarse, the candidate interface cannot be correctly represented; if the initial grid is too fine, it will introduce too many degrees of freedom in the first iteration, thus weakening the significance of subsequent adaptive grid partitioning. In one embodiment, the initial grid side length can be selected as 2 to 5 times the feature scale of the target minimum anomaly. For example, if the expected minimum effective anomaly width is 20 meters, the initial grid side length can be selected as 40 to 100 meters. For deep regions, due to the naturally lower resolution, larger cells can be appropriately used; for shallow regions or well-side high-control areas, smaller cells can be used.

[0054] The initial geological framework can be constructed through the joint interpretation of multi-physics observation data. In one embodiment, initial inversion or attribute extraction is first performed on seismic, electromagnetic, gravity, and magnetic data respectively to obtain the corresponding velocity field, resistivity field, density field, or magnetic response field. Then, boundary enhancement processing is performed on these results to extract spatially continuous gradient abrupt change regions or regions with concentrated attribute changes, forming candidate geological boundaries. Well-mounted physical property measurement data provided by well sensors can be used as a basis for local correction, making the framework location in the well-side area closer to the actual stratigraphic contact surface or fault location.

[0055] The boundary enhancement process described here can be implemented using conventional methods such as gradient magnitude extraction, local rate of change extraction, edge continuity filtering, and shortest discontinuity connections. The reason for first forming an initial geological framework, rather than allowing subsequent mesh updates to proceed freely, is that imaging complex geological bodies requires an initial structural reference. Without this reference, subsequent refinement can easily spread disorderly in high residual regions.

[0056] In one embodiment, when screening the initial geological framework, a minimum contiguous length threshold and a minimum evidence coverage threshold can be set. The minimum contiguous length threshold can be 3 to 10 basic grid side lengths, and the minimum evidence coverage threshold can require support from at least two physical methods or one physical method plus in-well sensor information. This avoids scattered local anomalies being directly identified as framework segments.

[0057] like Figure 5As shown, after forming the initial geological framework, an interface guard is generated around the candidate interfaces within the initial geological framework. The function of the interface guard is to limit subsequent boundary-related update operations from deviating too far from the actual interface, thereby reducing the back-and-forth jumps in interface position during multiple iterations.

[0058] In one embodiment, mesh cells adjacent to the candidate interface are first identified, and then several layers of cells are extended along both sides of the candidate interface's normal direction to form an interface guardrail cell set. For two-dimensional regular meshes, the guardrail width can be 2 to 6 times the side length of the base mesh. For quadtree meshes, 2 to 5 layers of adjacent cells can be used. For three-dimensional scenes, a thin-shell cell set is formed along both sides of the candidate interface's normal direction.

[0059] The width of the guard band should be determined based on the exploration resolution and interface uncertainty. If the boundary location error is expected to be small, the guard band can be appropriately narrower, for example, 2 to 3 basic grid side lengths. If the boundary location error is expected to be large, or if there is a significant diffusion of the interface response to different physical methods, the guard band can be expanded to 4 to 6 basic grid side lengths. The purpose of this setting is to allow sufficient room for the actual interface to operate, but not to allow the update range to expand into distant areas unrelated to the interface.

[0060] In one embodiment, three types of operations are allowed within the interface sheath. The first type is refinement, used to improve local expressiveness. The second type is coarsening, used to reclaim local cells that have proven unnecessary to maintain high resolution. The third type is boundary relocation, used to fine-tune the actual placement of boundaries within the sheath. Boundary relocation is generally not performed outside the sheath; local adjustments are only permitted when there is a genuine need for global imaging and topology locking is approved. This allows subsequent computational resources to be concentrated on locations where boundary migration is truly likely to occur.

[0061] like Figure 6 As shown, after the initial geological framework and interface protection zone are determined, candidate boundary zones are extracted using multiple physical methods, and boundary offset corridors are identified.

[0062] In one embodiment, candidate boundary zone extraction is performed separately for each physical method. For seismic data, concentrated bands of property variation can be extracted from the velocity field or wave impedance field as seismic candidate boundary zones. For electromagnetic data, electrical variation stripes can be extracted from resistivity inversion results as electromagnetic candidate boundary zones. For gravity and magnetic data, concentrated gradient bands can be extracted from local gradient enhancement results as gravity or magnetic candidate boundary zones. When well data does not form a separate zonal boundary, it can be used as a basis for position correction to correct the position of the centerline of adjacent boundary zones or adjust the width of the boundary zones.

[0063] Candidate boundary band extraction is preferably performed in areas where data quality markers meet requirements, and the minimum and maximum widths of the boundary bands can be set. For seismic scenarios, the boundary band width can be 1 to 3 basic grid side lengths; for electromagnetic diffusion scenarios, the boundary band width can be 2 to 6 basic grid side lengths; for gravity and magnetic responses, the boundary band width can be 2 to 5 basic grid side lengths depending on the gradient expansion range. A band is used instead of a line because different methods inherently possess a certain thickness in their spatial representation of the boundary.

[0064] In one embodiment, two spatially adjacent candidate boundary bands are compared. When the minimum distance between the two candidate boundary bands is within a preset distance range and their extension directions meet a preset consistency condition, the area between them is determined as a boundary offset corridor.

[0065] The preset spacing range can be determined based on the study scale and grid scale. For medium-scale 2D profile scenarios, the minimum spacing between two candidate boundary zones can be 1 to 5 basic grid side lengths. If the spacing is too small, it means that the two can be considered almost the same boundary, and there is no need to establish a corridor. If the spacing is too large, it means that the two may not correspond to the same geological interface, and it is not appropriate to forcibly classify them as offset corridors.

[0066] The consistency condition of extension direction can be achieved by comparing the directional angles between the local centerlines of two boundary zones. In one embodiment, the consistency condition of extension direction is considered satisfied when the directional angle is no greater than 15 to 25 degrees. If the angle is too large, it indicates that the two are not different physical responses of the same geological interface in terms of morphology, but may correspond to different structural units.

[0067] After the boundary offset corridor is established, it can be further divided into a first lateral sub-region, a transitional sub-region, and a second lateral sub-region. The first lateral sub-region is close to the first candidate boundary zone, the second lateral sub-region is close to the second candidate boundary zone, and the transitional sub-region is located in between. This division allows different physical methods to apply different update intensities to different locations within the corridor. For example, the seismic candidate boundary zone is more likely to constrain the first lateral sub-region, the electromagnetic candidate boundary zone is more likely to constrain the second lateral sub-region, while the transitional sub-region is reserved as a region to be determined. This design does not absolutely bind any one method to a particular sub-region, but rather allows multi-physical boundary offset relationships to be explicitly expressed, rather than being crudely compressed.

[0068] like Figure 7 As shown, within the area covered by the interface guardrail and the boundary offset corridor, an evidence accumulation state is established for the grid cells to determine whether a cell exhibits boundary activity characteristics in multiple consecutive iterations.

[0069] In one embodiment, the grid cell to be judged receives four types of evidence input in each iteration. The first type of evidence is the local residual distribution. Here, local residual refers to the distribution of observation fitting errors near the grid cell after the current imaging update. If a local residual concentration phenomenon consistently exists near a cell in multiple iterations, it indicates that there is likely still an incorrectly described boundary or transition structure at the current location. The second type of evidence is the difference in physical properties between adjacent grid cells. Here, physical property differences can refer to differences in velocity, resistivity, density, magnetism, or the combined differences of multiple physical properties in a local area. If the difference in physical properties between adjacent cells persists in multiple iterations, it indicates that the location is more likely to be an actual geological boundary rather than noise-induced. The third type of evidence is the location of candidate boundaries in the neighborhood of the grid cell to be judged. If candidate boundaries extracted in different iterations repeatedly fall near the cell, it indicates that the cell has a boundary bearing tendency. The fourth type of evidence is the spatial orientation of multiple physical methods for the grid cell to be judged. Here, spatial orientation refers to whether different physical methods all exhibit boundary orientation near the cell, and whether these orientations are concentrated in close proximity.

[0070] In one embodiment, the evidence accumulation state is represented by a hierarchical state quantity, such as level 0, level 1, level 2, level 3, and level 4. Level 0 represents no obvious boundary activity evidence, and level 4 represents highly credible boundary activity units. After each iteration, the state quantity is updated based on the occurrence of the four types of evidence, data quality markers, and temporal continuity.

[0071] One feasible update method is to use a rolling time window. The length of the rolling time window can be 3 to 8 iterations. If a cell simultaneously satisfies local residual concentration, significant physical property differences, and repeated candidate boundary placements for at least 3 iterations within the rolling time window, its evidence accumulation state is improved by 1 level. If, in addition, at least two physical methods point to this cell, and the quality labels of the participating data are all at least 0.7, it is improved by an additional 1 level. Conversely, if the above evidence significantly weakens in 2 to 4 consecutive iterations, the state is reduced by 1 level.

[0072] The reason for this design is that evidence of true boundaries is usually persistent and spatially coherent, while false boundaries often only appear in a few rounds. If we do not accumulate round by round, but instead immediately refine the mesh once a residual peak appears in a certain round, it is easy to cause boundary jitter and mesh back and forth.

[0073] In one embodiment, when the evidence accumulation status reaches level 3 or 4, the grid cell is identified as a cell to be updated. If the evidence accumulation status is only level 1 or 2, observation can be retained, and subdivision can be delayed. If the evidence accumulation status remains at level 0 for an extended period, high-resolution cells in that region can be preferentially recycled during subsequent coarsening.

[0074] After selecting the units to be updated, it is necessary to determine the grid partitioning direction. Instead of simply dividing the units into smaller, uniform subdivisions, directional partitioning is performed based on the discrete information of the observation direction. The fundamental reason for this design is that many targets in complex geological bodies have a clear directionality. For example, thin layers require higher resolution in the thickness direction, and steeply dipping faults require improved boundary positioning capabilities along their normal direction. Ignoring this and performing uniform partitioning would result in a large number of unnecessary small units.

[0075] In one embodiment, the distribution of discrete information about the observation direction is statistically analyzed for each cell to be updated. If most observations are concentrated in a few adjacent sectors, it indicates that the cell already has strong support in these directions, while the direction approximately perpendicular to these directions is still relatively lacking. In this case, the subdivision direction can be set as the normal direction of the current main observation direction. The intuitive understanding of this is that if most observations pass through the cell along a certain direction, the resolution in that direction is already good, and what really needs to be improved is the detail resolution perpendicular to that direction.

[0076] In one embodiment, when the coverage ratio of the largest sector is not less than 0.45 to 0.6 of the total coverage, and the coverage ratio in the direction approximately perpendicular to it is not higher than 0.2 to 0.35, the cell can be considered to have a significant directional bias, and directional partitioning should be performed. If the coverage in each direction is approximately balanced, conventional half partitioning or cross partitioning can be used.

[0077] In one embodiment, directional subdivision can be performed as a 1-to-2 or 1-to-3 subdivision. For rectangular elements, if the subdivision direction is horizontal, the element is divided into two sub-elements along the vertical direction; if the subdivision direction is vertical, it is divided into two sub-elements along the horizontal direction. For elements that need to represent transition regions, a 1-to-3 subdivision can also be used, where the middle sub-element retains the transition attributes, and the two side sub-elements correspond to the two sides of the boundary, respectively. For three-dimensional volume elements, 1-to-2 or 1-to-4 refinement can be performed along the normal plane corresponding to the selected subdivision direction.

[0078] For cells to be updated located within a boundary offset corridor, this invention preferentially uses the direction of the line connecting two candidate boundary zones or the local construction normal direction as the partitioning direction. The reason for this is that the boundary offset corridor itself represents the misalignment relationship between different physical boundaries. Partitioning along the corridor's connecting line direction is more conducive to expressing the first side sub-region, the transition sub-region, and the second side sub-region separately. Furthermore, if the local construction normal direction is more clearly defined, partitioning along the normal direction is more beneficial for precise interface positioning.

[0079] In one embodiment, the internal units of the corridor can be preferentially divided into three sub-regions. Sub-units closer to the first candidate boundary zone are assigned to the first side sub-region, sub-units closer to the second candidate boundary zone are assigned to the second side sub-region, and intermediate sub-units are assigned to the transition sub-region. During subsequent imaging updates, the first side sub-region can preferentially accept physical method constraints corresponding to the first candidate boundary zone, the second side sub-region can preferentially accept physical method constraints corresponding to the second candidate boundary zone, and the transition sub-region is handled by balanced update or pending decision update. In this way, boundary misalignment is no longer compressed into a line, but is specifically expressed at the mesh level.

[0080] like Figure 8 As shown, the anomalous bearing elements identified in the coarse-scale stage need to be marked with ancestral elements to ensure that deep or weak response anomalies are not completely diluted in the subsequent high-resolution stage.

[0081] In one embodiment, the determination of anomaly-bearing cells is based on multi-round imaging results at the coarse-scale stage. A coarse-grid cell can be identified as an anomalous-bearing cell if it simultaneously meets at least two of the following conditions in two to five consecutive iterations: Condition 1: A stable anomalous physical property value consistently exists in the vicinity of the cell. Condition 2: The cell has an evidence accumulation state of at least level 3. Condition 3: The cell is continuously pointed to by at least two physical methods. Condition 4: In-well sensors or nearby high-confidence control points support the presence of an anomalous body at the cell's location.

[0082] Once a unit is identified as an abnormal carrier unit, an ancestor unit tag is set for it. The ancestor unit tag can be stored as a unit attribute field, such as adding an inheritance flag, ancestor unit number, and ancestor unit level number to the unit data structure.

[0083] In one embodiment, when the ancestral unit is divided into multiple descendant units in subsequent iterations, at least one descendant unit is retained to continue carrying the corresponding ancestral unit label. The selection of this descendant unit can be achieved as follows: The descendant unit whose central position is closest to the center of the original ancestral unit's anomalous response is preferentially selected; if multiple descendant units are all close to this central position, their evidence accumulation state and anomalous property stability are further compared, and the unit with the strongest overall performance is selected to inherit the ancestral unit label.

[0084] If multiple descendant units all possess significant carrying capacity, multiple descendant units may be allowed to inherit the ancestor unit number simultaneously, but at least one primary inheriting unit should be retained. The purpose of this design is to maintain the continuity of coarse-scale anomaly information in the fine-scale stage without sacrificing local refinement degrees of freedom.

[0085] In one embodiment, if a descendant unit with an ancestor unit tag experiences an evidence accumulation state of 0 over 3 to 6 consecutive iterations, and there are no physical methods in the vicinity that continue to point to that unit, an ancestor unit tag transfer or revocation review can be triggered. Revocation is not executed directly, but rather undergoes a topology locking review first to prevent deep anomalous chains from being prematurely severed.

[0086] Topology locking review is a crucial part of this invention. Its function is not to prevent all model changes, but to prevent candidate grid operations that, while locally improving data fitting, would significantly disrupt the underground structural relationships.

[0087] In one embodiment, candidate mesh operations include refinement, coarsening, cell assignment adjustment, and boundary relocation. Refinement involves dividing the current cell into multiple smaller cells. Coarsening involves merging multiple adjacent high-resolution cells into a larger cell. Cell assignment adjustment involves changing the cell's association with a specific geological region, boundary region, or anomaly during the update. Boundary relocation involves adjusting the boundary center position within the interface sheath.

[0088] After each candidate operation is formed, the topology locking review module first performs a structural consistency check on the operation before deciding whether to allow its execution.

[0089] In one embodiment, the connectivity of anomalous entities is represented by an anomalous entity adjacency graph. Nodes in the adjacency graph correspond to anomalous entities or anomalous unit clusters, and edges represent spatial connectivity. During topology locking review, the number of connected components and the main connected paths before and after the candidate operation are compared. If an operation causes two originally separate anomalous entities to merge without new, continuous evidence, or causes a originally continuous anomalous entity to break into multiple parts without reasonable basis, then the connectivity of the anomalous entities is determined not to meet the preset locking conditions.

[0090] The preset locking condition here can be set to ensure that the number of connected components of the main anomaly does not change in a single round of candidate operations, unless there is evidence of boundary breakage or merger of level 3 or higher in multiple consecutive iterations.

[0091] In one embodiment, fault continuity is detected by the continuous length of fault skeleton segments, the spacing between segments, and the consistency of their orientation. If a candidate operation would cause an interruption of a confirmed fault, and the interruption spacing exceeds one to three basic mesh side lengths, then the fault continuity is determined not to meet the preset locking conditions. This constraint is set because once a fault is artificially cut off in the model, subsequent stratigraphic relationships and anomalous distribution relationships will be affected.

[0092] In one embodiment, sequence adjacency is used to describe the order of adjacent strata and local contact relationships. It can be stored through a sequence adjacency table or a sequence directed graph. If, after a candidate operation is executed, a stratum cell that should be located in the upper part is incorrectly assigned to a lower stratum, or the adjacency order is reversed, it is determined that the sequence adjacency does not meet the preset locking condition.

[0093] This examination is particularly applicable to thin layers, pinch-out bodies, and lenticular bodies near faults, because these structures are most prone to sequence disorder during local refinement.

[0094] In one embodiment, the ancestor unit inheritance relationship is used to detect whether the ancestor unit tag chain still exists. If a coarsening operation or attribution adjustment operation causes all descendant units of an ancestor unit to lose their inheritance tags and there is no legal transfer path, then the ancestor unit inheritance relationship is determined not to meet the preset locking condition.

[0095] The significance of this check is to prevent anomalies identified at the coarse-scale from being completely forgotten by the system at the fine-scale stage.

[0096] In one embodiment, candidate mesh operations are submitted to the imaging update module for execution only after the topology locking review is passed. If any detection result does not meet the preset locking conditions, the corresponding candidate mesh operation is blocked, and the operation is recorded as a rejected operation for reference in the next round of parameter adjustment. Rejection does not mean that the region will never be updated, but rather that the currently proposed operation method is unreasonable and needs to be re-evaluated in subsequent rounds with new evidence and new subdivision directions.

[0097] After the topology lock review is approved, mesh updates and imaging updates are performed. Mesh updates refer to the actual refinement, coarsening, boundary relocation, or attribution adjustment of the cells to be updated that have passed the review. Imaging updates refer to resolving subsurface properties under multiphysics constraints on the updated mesh.

[0098] In one embodiment, imaging updates can be performed using a conventional iterative inversion framework, without relying on machine learning or reinforcement learning models. In other words, the key to this invention is not to invent a new inversion solver, but to reorganize the mesh update logic before and after inversion through interface guardrails, boundary offset corridors, evidence accumulation states, directional partitioning, ancestor cell inheritance, and topology locking review.

[0099] Specifically, forward response models for seismic, electromagnetic, gravity, and magnetic forces can be established separately, and the corresponding physical property fields can be updated on a unified or mapped grid. In-well physical property measurement data can be used for constraint correction of adjacent cells. Updates can be performed using alternating single methods or joint updates. In the joint update scenario, different data weighting strategies can be set for the first, transition, and second sub-regions within the boundary offset corridor. For example, the first sub-region can preferentially receive seismic boundary constraints, the second sub-region can preferentially receive electromagnetic boundary constraints, and the transition sub-region can use balanced weights. This allows the influence of different physical methods on the same complex boundary to be spatially differentiated.

[0100] In one embodiment, the imaging update termination condition can be set to stop when any of the following conditions are met: First, the global residual improvement is no higher than 1% to 5% in 2 to 4 consecutive iterations. Second, the number of units to be updated decreases to 5% to 15% of the initial number of units to be updated. Third, a preset maximum number of iterations is reached, for example, 10 to 30 iterations.

[0101] The purpose of setting these termination conditions is to avoid performing too much refinement when the boundaries are already basically stable, thereby controlling the computational scale.

[0102] The following is a more complete embodiment to further illustrate the implementation of the present invention.

[0103] In a two-dimensional profile region characterized by well-developed faults and the coexistence of thin layers and lenticular bodies, seismic sensors, electromagnetic sensors, gravity sensors, and magnetic sensors were deployed along the surface, and well-drilled sensors were also installed in two control wells. The acquired seismic wave response data, electromagnetic field response data, gravity anomaly data, magnetic anomaly data, and well-drilled physical property measurement data were first uniformly sent to the data processing center. After time synchronization processing, the system added time stamps, data quality markers, and observation direction discretization information to each data point.

[0104] Subsequently, an initial grid model was established using a two-dimensional regular rectangular grid with a basic grid side length of 50 meters. The initial geological framework was then extracted by combining seismic velocity variation zones, resistivity variation zones, and well-level constraints. If a candidate interface has a continuous length of not less than 5 basic grid side lengths and is supported by at least two physical methods, it is included in the initial geological framework.

[0105] Three layers of basic grid cells are extended to each side of each candidate interface in the initial geological framework to generate interface protection zones. Next, seismic candidate boundary zones are extracted from seismic velocity variation results, and electromagnetic candidate boundary zones are extracted from resistivity variation results. When the minimum distance between the two is between 2 and 4 basic grid side lengths, and the angle between the local extension directions is no greater than 20 degrees, the region between them is defined as a boundary offset corridor and divided into a first lateral sub-region, a transition sub-region, and a second lateral sub-region.

[0106] In the subsequent five rolling time windows, the local residual distribution, differences in physical properties between adjacent grid cells, candidate boundary placement, and multi-physical spatial orientation of the grid cells within the interface guardrail and boundary offset corridor are recorded round by round. If a cell simultaneously satisfies the following conditions in at least three out of the five rounds: local residual concentration, significant differences in adjacent physical properties, and repeated candidate boundary placement, and at least two physical methods point to the cell, its evidence accumulation status is upgraded to level 3 or above, and it is identified as a cell to be updated.

[0107] For cells to be updated, if the observation direction is mainly concentrated in three adjacent sectors out of the 12 directional sectors, the directional offset is considered significant, and directional meshing is performed along the normal of the main observation direction. For cells to be updated located in the boundary offset corridor, a one-to-three meshing is preferentially performed along the line connecting two candidate boundary zones. For mesh cells that have shown anomalous bearing capacity for three consecutive rounds in the coarse-scale stage, ancestor cell labels are added.

[0108] Before each refinement, coarsening, cell assignment adjustment, or boundary relocation, the system examines the connectivity of anomalous bodies, fault continuity, sequence adjacency, and ancestor cell inheritance. If an operation would cause a previously continuous fault to be interrupted by more than two base grid side lengths, or would completely eliminate an ancestor cell inheritance chain, the operation would be rejected.

[0109] For operations that pass the review, perform grid updates and imaging updates, repeat the above process until the global residual improvement is no higher than 2% for three consecutive rounds, and then output the imaging results of complex geological bodies.

[0110] As can be seen from the above process, this invention does not simply add a new boundary extraction algorithm, nor does it simply add a new inversion solver. Instead, it introduces a controlled boundary activity space, a multi-round evidence accumulation mechanism, a directional subdivision mechanism oriented towards directional differences, a ancestor unit inheritance mechanism oriented towards preserving deep anomalies, and a topology locking review mechanism oriented towards structural stability throughout the entire imaging update process. This makes the imaging of complex geological bodies more suitable for handling scenarios with multiple adjacent interfaces, multiple physical boundary misalignments, and thin and elongated targets.

[0111] like Figure 2As shown, the present invention also provides an imaging system for complex geological bodies to implement the above-described method. This system includes an intelligent sensor collaborative acquisition module, an initial skeleton construction module, an interface sheath generation module, a boundary offset corridor identification module, an evidence accumulation and screening module, a directional subdivision and ancestral unit inheritance module, a topology locking review module, and an imaging update module.

[0112] The intelligent sensor collaborative acquisition module is used to collect multi-physics observation data of the target area, perform time synchronization processing on the multi-physics observation data, and output time identifiers, data quality markers and observation direction discrete information corresponding to the multi-physics observation data.

[0113] The initial skeleton construction module is used to construct the initial geological skeleton and initial grid model based on multi-physics observation data.

[0114] The interface guard generation module is used to generate interface guards around candidate interfaces.

[0115] The boundary offset corridor identification module is used to identify boundary offset corridors based on candidate boundary bands corresponding to multiple physical methods.

[0116] The evidence accumulation and filtering module is used to establish the evidence accumulation status and filter the units to be updated.

[0117] The directional meshing and ancestor element inheritance module is used to perform adaptive meshing based on the observation direction discretization information of the element to be updated and to set ancestor element tags for anomalous bearing elements.

[0118] The Topology Lock Review module is used to perform topology lock reviews on candidate mesh operations.

[0119] The imaging update module is used to perform mesh updates and imaging updates on candidate mesh operations that have passed the topology locking review, so as to obtain imaging results of complex geological bodies.

[0120] In one embodiment, the above modules can be deployed in software on the same server, or distributed in a hybrid software and hardware manner. The intelligent sensor collaborative acquisition module can be deployed between the edge acquisition terminal and the central server. The initial skeleton construction module, interface guardrail generation module, boundary offset corridor recognition module, evidence accumulation and screening module, directional subdivision and ancestor unit inheritance module, topology locking review module, and imaging update module can be deployed on the central computing platform. For large-scale 3D scenes, the imaging update module can run on a multi-core processor and graphics processor collaborative platform to accelerate forward simulation and inversion updates.

[0121] like Figure 3 As shown, the intelligent sensors in the system include earthquake sensors, electromagnetic sensors, gravity sensors, magnetic sensors, and well sensors. The detection objects of the above sensors are the same as those described in the aforementioned method embodiments.

[0122] like Figure 1 and Figure 2 As shown, the data flow relationships between each module correspond one-to-one with the method flow, thus enabling closed-loop processing from multi-physical data acquisition to final imaging output.

[0123] This invention does not incorporate machine learning or reinforcement learning models as essential components; therefore, its core implementation does not depend on the model training process. This invention can be directly implemented on a conventional geophysical inversion platform by adding logic for interface protection zone generation, boundary offset corridor identification, evidence accumulation state update, directional subdivision, ancestor unit inheritance, and topology locking review to the existing grid management, boundary extraction, and structure review modules. This allows those skilled in the art to implement this invention without replacing the underlying forward solver.

[0124] Furthermore, the parameter ranges in this invention are not fixed but are related to the exploration scale, data quality, subsurface medium complexity, and imaging target depth. Those skilled in the art can routinely adjust the basic grid side length, guardrail width, candidate boundary zone width, boundary offset corridor spacing range, directional consistency conditions, evidence accumulation time window length, cell selection threshold for updates, and topology locking conditions according to field conditions. As long as the overall technical logic described in this invention is still followed—that is, providing multi-physics observations and metadata through intelligent sensors, defining the structural update range through the initial geological framework and interface guardrail, expressing multi-physics boundary misalignment through boundary offset corridors, selecting real boundary active cells through evidence accumulation status, guiding directional subdivision through discrete observation direction information, maintaining the continuation of deep anomalies through ancestor cell marking, and preventing unreasonable grid operations through topology locking review—the technical solution of this invention can be achieved.

[0125] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for imaging complex geological bodies based on adaptive grid partitioning, characterized in that, include: The intelligent sensor collects multi-physics observation data of the target area, performs time synchronization processing on the multi-physics observation data, and outputs time identifiers, data quality markers and observation direction discrete information corresponding to the multi-physics observation data; An initial geological framework and an initial grid model were constructed based on the aforementioned multiphysics observation data; An interface protection zone is generated around the candidate interfaces in the initial geological framework, and a boundary offset corridor is identified based on the candidate boundary zones corresponding to multiple physical methods. Within the interface guardrail and the boundary offset corridor, an evidence accumulation state is established for the grid cells, and cells to be updated are filtered based on the evidence accumulation state. The grid partitioning direction is determined based on the discrete information of the observation direction of the unit to be updated, and adaptive grid partitioning is performed according to the grid partitioning direction. Ancestor unit tags are set for the abnormal bearing units identified in the coarse-scale stage. Before performing a grid update, a topology lock check is performed on the candidate grid operations. For the candidate grid operations that pass the topology lock check, grid updates and imaging updates are performed to obtain imaging results of complex geological bodies.

2. The method according to claim 1, characterized in that, The intelligent sensors include a seismic sensor, an electromagnetic sensor, a gravity sensor, a magnetic sensor, and a well-drilled sensor; the seismic sensor is used to detect seismic wave response data, the electromagnetic sensor is used to detect electromagnetic field response data, the gravity sensor is used to detect gravity anomaly data, the magnetic sensor is used to detect magnetic anomaly data, and the well-drilled sensor is used to detect well-drilled physical property measurement data.

3. The method according to claim 2, characterized in that, The intelligent sensor outputs data quality markers, time markers, and discrete observation direction information associated with the corresponding multi-physics observation data. The data quality markers are used to characterize the data quality status of the corresponding multi-physics observation data, the time markers are used to characterize the acquisition time sequence of the corresponding multi-physics observation data, and the discrete observation direction information is used to characterize the distribution of observation directions passing through the grid cells.

4. The method according to claim 1, characterized in that, The process of generating an interface protection zone around the candidate interfaces in the initial geological framework specifically includes generating an interface protection zone unit set based on the spatial distribution of grid units adjacent to the candidate interfaces, and limiting subsequent refinement, coarsening, and boundary relocation operations to be performed within the area covered by the interface protection zone unit set and the boundary offset corridor.

5. The method according to claim 1, characterized in that, The method of identifying the boundary offset corridor based on candidate boundary bands corresponding to multiple physical methods specifically includes extracting candidate boundary bands corresponding to multiple physical methods, comparing two spatially adjacent candidate boundary bands, and determining the area between the two candidate boundary bands as the boundary offset corridor when the distance between the two candidate boundary bands is within a preset distance range and the extension direction meets a preset consistency condition. The method also includes marking the first side sub-region, the transition sub-region, and the second side sub-region within the boundary offset corridor.

6. The method according to claim 1, characterized in that, The establishment of evidence accumulation status for grid cells specifically includes recording the local residual distribution of the grid cell to be judged, the differences in physical properties between adjacent grid cells, the location of candidate boundaries in the neighborhood of the grid cell to be judged, and the spatial orientation of the grid cell to be judged by multiple physical methods in each iteration round, and updating the evidence accumulation status of the grid cell to be judged in each iteration round. Grid cells whose evidence accumulation status meets the preset accumulation conditions are identified as cells to be updated.

7. The method according to claim 1, characterized in that, The adaptive meshing based on the observation direction discrete information of the unit to be updated specifically includes determining the meshing direction according to the observation direction discrete information of the unit to be updated, and dividing the unit to be updated into multiple sub-units along the meshing direction; for the unit to be updated located in the boundary offset corridor, adaptive meshing is performed along the line connecting two candidate boundary zones or the local construction normal direction; for anomalous bearing units with ancestor unit tags, at least one descendant unit is retained to participate in the update in subsequent mesh updates.

8. A complex geological body imaging system based on adaptive mesh partitioning for implementing the method of claim 1, characterized in that, The system comprises an intelligent sensor collaborative acquisition module, an initial skeleton construction module, an interface protection zone generation module, a boundary offset corridor identification module, an evidence accumulation and screening module, a directional subdivision and ancestor unit inheritance module, a topology lock review module, and an imaging update module. The intelligent sensor collaborative acquisition module collects multi-physics observation data of the target area and performs time synchronization processing on the multi-physics observation data, outputting time stamps, data quality markers, and observation direction discrete information corresponding to the multi-physics observation data. The initial skeleton construction module constructs an initial geological skeleton and an initial grid model based on the multi-physics observation data. The interface protection zone generation module generates interface protection zones around candidate interfaces. The boundary offset corridor identification module identifies boundary offset corridors based on candidate boundary zones corresponding to multiple physical methods. The evidence accumulation and screening module establishes evidence accumulation status and filters units to be updated. The directional subdivision and ancestor unit inheritance module performs adaptive grid subdivision based on the observation direction discrete information of the units to be updated and sets ancestor unit markers for anomalous bearing units. The topology lock review module performs topology lock review on candidate grid operations. The imaging update module performs grid update and imaging update on candidate grid operations that pass the topology lock review, obtaining imaging results for complex geological bodies.

9. The system according to claim 8, characterized in that, The intelligent sensors in the intelligent sensor collaborative acquisition module include seismic sensors, electromagnetic sensors, gravity sensors, magnetic sensors, and well sensors. The seismic sensors are used to detect seismic wave response data, the electromagnetic sensors are used to detect electromagnetic field response data, the gravity sensors are used to detect gravity anomaly data, the magnetic sensors are used to detect magnetic anomaly data, and the well sensors are used to detect well physical property measurement data.

10. The system according to claim 8, characterized in that, The topology locking review module is used to detect anomalous body connectivity, fault continuity, sequence adjacency, and ancestor cell inheritance before performing refinement, coarsening, cell affiliation adjustment, and boundary relocation operations, and to block the corresponding candidate mesh operation if any detection result does not meet the preset locking conditions.

Images