Well-seismic joint large model driven high-resolution reservoir inversion method and system

Abstract

The application discloses a well-seismic combined large model driven high-resolution reservoir inversion method and system, and the technical scheme is that auxiliary sensing information is collected by an intelligent sensor arranged in a target work area, and target work area data is acquired; the target work area data and the auxiliary sensing information are input into a well-seismic combined large model to obtain a propagation control map and an inversion control map; according to the propagation control map, restricted propagation is performed on a well constraint detail sequence corresponding to well logging data of a target well to obtain a restricted propagation result; and according to the target work area seismic data and the restricted propagation result, reservoir inversion is performed to obtain a high-resolution reservoir inversion result, so that the problems of unconditional lateral propagation of a well constraint detail sequence, unclear local propagation boundary and insufficient spatial consistency of well-seismic combined inversion are solved.

Description

Technical Field

[0001] This invention relates to the field of reservoir inversion, and more specifically, to a high-resolution reservoir inversion method and system driven by a combined well-seismic large model. Background Technology

[0002] Well-seismic combined reservoir inversion is a crucial step in hydrocarbon geophysical interpretation. Its basic idea is to combine the high vertical resolution of well logging data at the well point, which allows for detailed characterization of lithology and physical property changes, with the wide lateral coverage of seismic data, which can continuously describe subsurface structures and stratigraphic distribution. This extends detailed local information from the well to inter-well and off-well areas, yielding wave impedance, velocity, lithology, or reservoir distribution results for the target layer. In practical applications, well data typically needs to be calibrated with adjacent seismic data before inversion modeling is performed, combining stratigraphic interpretation results, structural interpretation results, and seismic body property results. Simultaneously, with the development of downhole sensing and seismic acquisition methods, downhole pressure, temperature, vibration, and microseismic information are increasingly being used as supplementary sensing information in the reservoir evaluation process.

[0003] In the prior art, patent CN107329171A, entitled "Deep Domain Reservoir Seismic Inversion Method and Device," discloses a method for calibrating well logging curves with deep domain post-stack seismic data and combining stratigraphic interpretation and a gridded geological framework to conduct reservoir seismic inversion. The core of this patent lies in first converting time-domain seismic data to the depth domain, and then obtaining the depth-domain reservoir seismic inversion result through well logging curve models, geological framework models, and random interpolation of fractal parameter volumes. Its focus is on depth-domain modeling and fractal random interpolation.

[0004] In the prior art, patent CN112130195A, entitled "Time-Shift VSP Data Acquisition System and Method Based on Distributed Fiber Optic Acoustic Sensing," discloses a technical solution for deploying a distributed fiber optic acoustic sensing device in a well and synchronously acquiring well and surface seismic data with a surface seismic acquisition device. The core of this patent lies in acquiring vertical seismic profile data in the well and surface seismic data through the collaborative acquisition of armored fiber optic cables for distributed fiber optic acoustic sensing in the well and the surface acquisition device. This supports the evaluation of reservoir stimulation effects and dynamic monitoring of the wellbore fluid interface, with a focus on the acquisition and synchronous collection of seismic sensing information in the well.

[0005] From the perspective of current technological development, well-seismic joint inversion has already enabled the joint utilization of well data and seismic data, and schemes have emerged that involve in-well sensing devices in seismic data acquisition and interpretation. However, when it comes to high-resolution reservoir inversion, three prominent specific problems still exist. First, high-frequency details above the well cannot propagate unconditionally in space. Thin-layer rhythms, local lithological changes, or fine-grained features near the well in a particular well are often only relevant to areas belonging to the same geological semantic unit. If propagation is based solely on spatial distance, simple waveform similarity, or general stratigraphic correspondence, fine information above the well can easily cross fault boundaries, pinch-out boundaries, or phase transition boundaries, and be incorrectly carried to locations where the well's information should not be used. This results in continuous fine layers or anomalous high-frequency details in the external area that do not conform to the actual subsurface conditions. The essence of this problem is not an insufficient amount of well data, but rather that the effective propagation boundary of well data is not defined finely enough in the existing process.

[0006] Second, well-seismic mismatch does not always result in the failure of the entire well; in many cases, it occurs only locally in specific stratigraphic segments. While the well-seismic calibration relationship may be stable in some segments, others may be affected by time-depth correspondence deviations, local wavelet differences, fluctuations in well data quality, or local seismic response anomalies, leading to inaccurate correspondence between well and seismic data within those segments. If the entire well is still used as a unified constraint source, local mismatches will be amplified, misinterpreting data differences that are actually calibration issues as variations in the reservoir itself, resulting in unreasonable details in the inversion results within these segments. Although existing technologies have addressed time-depth relationship processing, well-seismic calibration, and seismic inversion separately, segmented control of localized stratigraphic mismatches remains insufficient.

[0007] Third, high-frequency information in high-resolution inversion is not suitable for injection at all locations. Whether the high-frequency details of well data can be reliably utilized in a particular analysis unit is affected not only by the well-seismic correspondence but also by the seismic resolution, seismic continuity, local noise level, and boundary complexity at that location. In existing technologies, high-resolution inversion typically emphasizes improving seismic resolution or enhancing high-frequency information. However, in real-world work areas, different locations have varying capacities for high-frequency information. Without location-related availability assessments, fine high-frequency information from the well can easily be written into locations that the seismic data itself cannot support, leading to problems such as excessive local detail, unstable spatial distribution, or insufficient interpretability in the inversion results. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a high-resolution reservoir inversion method and system driven by a combined well-seismic large model, so as to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A high-resolution reservoir inversion method driven by a combined well-seismic model includes: The system collects auxiliary sensing information and acquires target work area data by setting up intelligent sensors in the target work area. The target work area data and the auxiliary sensing information are input into the well-seismic joint large model to obtain the propagation control map and the inversion control map; Based on the propagation control map, restricted propagation is performed on the well constraint detail sequence corresponding to the target well logging data to obtain the restricted propagation result; Based on the inversion control map, segmented control is performed on the restricted propagation results, and frequency control is performed on the target well logging data to obtain segmented control results and frequency control results; Based on the seismic data of the target work area, the restricted propagation results, the segmented control results, and the frequency control results, reservoir inversion is performed to obtain high-resolution reservoir inversion results.

[0010] Preferably, the intelligent sensor is at least one of a downhole pressure sensor, a downhole temperature sensor, a distributed acoustic sensor, a distributed temperature sensor, and a microseismic sensor; the auxiliary sensing information includes at least one of pressure information, temperature information, vibration information, microseismic event information, and distributed response information along the well.

[0011] Preferably, the target work area data includes target work area seismic data, target well logging data, geological interpretation data, and well-seismic calibration data; the propagation control map includes a semantic connectivity map and a blocking boundary map; the inversion control map includes a layer reliability map, a residual cause identification map, and a frequency availability map.

[0012] Preferably, before inputting the target work area data and the auxiliary sensing information into the well-seismic joint large model, the method further includes: The time reference and sampling interval for the seismic data of the target work area were unified; The depth benchmark and sampling interval were standardized for the logging data of the target well. The coordinate references for the geological interpretation data and the well-seismic calibration data shall be unified. Perform time synchronization and coordinate matching on the auxiliary sensing information; The target work area data and the auxiliary sensing information, after benchmark unification, are mapped to the unified work area coordinate system; The data within the unified work area coordinate system is divided according to a preset grid to obtain multiple analysis units; The logging data of the target well is segmented according to the preset layer interface to obtain multiple layers.

[0013] Preferably, the target work area data and the auxiliary sensing information are input into the well-seismic joint large model to obtain propagation control maps and inversion control maps, including: The seismic data of the target work area corresponding to each analysis unit, the well logging data of the target well corresponding to each layer, the geological interpretation data, the well-seismic calibration data, and the auxiliary sensing information are input into the well-seismic joint large model. The well-seismic joint large model outputs semantic connectivity identifiers for each analysis unit; The combined well-seismic model outputs boundary markers for the boundaries between adjacent analysis units. The combined well-seismic model outputs the reliability level and residual cause identifier for each segment; The combined well-seismic model outputs a frequency availability level for each analysis unit; The semantic connectivity graph is constructed based on the semantic connectivity identifiers of each of the aforementioned semantic connectivity components; The blocking boundary map is constructed based on each of the aforementioned blocking boundary identifiers; Construct the reliability map of each layer based on its reliability level; Construct the residual cause identification map based on each of the aforementioned residual cause identifications; The frequency availability map is constructed based on each of the frequency availability levels.

[0014] Preferably, based on the propagation control map, restricted propagation is performed on the well constraint detail sequence corresponding to the target well logging data to obtain the restricted propagation result, including: Extract well constraint detail sequences from each layer of each target well; For each analysis unit, target well segments with the same semantic connectivity identifier and consistent stratigraphic position as the analysis unit are selected; For each selected target well interval, an intra-stratal connection path is constructed between the analysis unit and the target well interval; When the connecting path within the layer intersects with the blocking boundary marker in the blocking boundary diagram, the propagation of the target well segment to the analysis unit is stopped; When the connecting path within the layer does not intersect with the blocking boundary marker in the blocking boundary map, a local waveform similarity comparison is performed between the seismic trace corresponding to the analysis unit and the seismic trace at the corresponding well location of the target well segment. When only one target well segment satisfies the non-intersection condition, the well constraint detail sequence corresponding to the target well segment is written into the analysis unit; When multiple target well segments meet the non-intersection condition, select the target well segment with the maximum local waveform similarity to the seismic trace corresponding to the analysis unit, and write the well constraint detail sequence corresponding to the selected target well segment into the analysis unit. The target well segment identifier corresponding to each analysis unit and the well constraint detail sequence written into each analysis unit are determined as the restricted propagation result.

[0015] Preferably, based on the inversion control map, segmented control is performed on the restricted propagation result to obtain segmented control results, including: For each target well segment in the restricted propagation results, the corresponding segment reliability level and residual cause identifier are read; When the reliability level of the segment is the first reliability level, the well constraint detail sequence corresponding to the target well segment is written into the segmented control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is a time-depth offset identifier, local time-depth correction is performed on the target well segment, and the corrected well constraint detail sequence is written into the segmented control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is the wavelet mismatch identifier, local wavelet replacement is performed on the target well segment, and the replaced well constraint detail sequence is written into the segmented control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is a well data anomaly identifier, the adjacent well segment located in the same layer as the target well segment and with the smallest distance from the corresponding target well is selected as the alternative well constraint, and the alternative well constraint is written into the segment control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is a seismic response anomaly identifier, stop writing the well constraint detail sequence corresponding to the target well segment into the segmented control results, and retain the target work area seismic data of the corresponding layer of the target well segment to participate in the reservoir inversion.

[0016] Preferably, based on the inversion control map, frequency control is performed on the target well logging data to obtain frequency control results, and reservoir inversion is performed based on the target area seismic data, the confined propagation results, the segmented control results, and the frequency control results, including: Frequency band decomposition was performed on the target well logging data corresponding to each target well segment according to the center frequency from low to high, to obtain the first frequency band data, the second frequency band data and the third frequency band data; For each analysis unit, read the target well segment identifier corresponding to that analysis unit from the restricted propagation results; For each analysis unit, read the frequency availability level corresponding to that analysis unit; When the frequency availability level is the first frequency availability level, the first frequency band data, the second frequency band data, and the third frequency band data corresponding to the target well segment identifier are written into the frequency control result corresponding to the analysis unit; When the frequency availability level is the second frequency availability level, the first frequency band data and the second frequency band data corresponding to the target well segment identifier are written into the frequency control result corresponding to the analysis unit; When the frequency availability level is the third frequency availability level, the first frequency band data corresponding to the target well segment identifier is written into the frequency control result corresponding to the analysis unit; When the frequency availability level is the fourth frequency availability level, stop writing the first frequency band data, second frequency band data and third frequency band data corresponding to the target well segment identifier into the frequency control result corresponding to the analysis unit; The restricted propagation results are determined as the well constraint basic data; The segmented control results are written into the well constraint basic data of the corresponding target well section; The frequency control results are written into the well constraint base data of the corresponding analysis unit; The seismic data of the target work area are used as seismic constraints; Solve the reservoir parameters for each analysis unit separately; The reservoir parameters of each analysis unit are stitched together to form the high-resolution reservoir inversion result.

[0017] This invention also discloses a high-resolution reservoir inversion system driven by a combined well-seismic model, comprising: Intelligent sensors are used to collect auxiliary sensing information; The data acquisition module is used to acquire data from the target work area; The map generation module is used to input the target work area data and the auxiliary sensing information into the well-seismic joint large model to obtain propagation control map and inversion control map; The restricted propagation module is used to perform restricted propagation on the well constraint detail sequence corresponding to the target well logging data based on the propagation control map, and obtain the restricted propagation result. The segmented control module is used to perform segmented control on the restricted propagation result based on the inversion control chart to obtain the segmented control result; The frequency control module is used to perform frequency control on the target well logging data based on the inversion control map to obtain the frequency control result; The inversion module is used to perform reservoir inversion based on the seismic data of the target work area, the restricted propagation results, the segmented control results, and the frequency control results to obtain high-resolution reservoir inversion results.

[0018] Preferably, the map generation module is used to output a semantic connectivity map, a blocking boundary map, a segment reliability map, a residual cause identification map, and a frequency availability map; the restricted propagation module is used to filter target well segments with the same semantic connectivity identifier and consistent layer as the analysis unit, construct intra-layer connection paths, stop propagation when the intra-layer connection path intersects with the blocking boundary identifier, and select the target well segment with the maximum local waveform similarity to the seismic trace corresponding to the analysis unit when multiple target well segments meet the propagation conditions and write it into the corresponding analysis unit; the segment control module is used to perform well constraint detail sequence writing, local time depth correction, local wavelet replacement, substitute well constraint writing, and well constraint detail sequence stop writing on the target well segments according to the segment reliability level and residual cause identifier; the frequency control module is used to write the corresponding frequency band data from the first frequency band data, second frequency band data, and third frequency band data corresponding to the target well segment identifier to the analysis unit according to the frequency availability level.

[0019] The advantages of this invention over existing technologies lie in its utilization of three technical principles in well-seismic joint inversion. First, while the fine sequence stratigraphy and physical property details in well data have high resolution, their effective lateral propagation range is not solely determined by spatial distance, but is constrained by geological semantic connectivity and blocking boundaries. Second, well-seismic differences exhibit significant segmental characteristics; residuals in different segments often correspond to different causes, therefore, all residuals cannot be directly fitted as reservoir variations. Third, the effect of high-frequency information from well data on seismic inversion varies spatially; different analysis units have inconsistent capacities for carrying high-frequency information. Existing technologies have already demonstrated the importance of well-seismic calibration, frequency-division inversion, and in-well sensing acquisition, which also indirectly illustrates the objective existence of the aforementioned principles.

[0020] This invention inputs auxiliary sensing information collected by intelligent sensors along with target area data into a combined well-seismic model. The combined model generates propagation control maps and inversion control maps. Based on the propagation control maps, restricted propagation is performed on the well-constrained detail sequence. Subsequently, reservoir inversion is performed by combining the segmented control results and frequency control results. Because the propagation range of the well-constrained detail sequence is limited before subsequent inversion, this invention ensures that well details only participate in modeling within the permitted propagation area. This prevents high-resolution well data from crossing boundaries that should not propagate into areas outside the well, allowing the local accuracy of well data and the spatial continuity of seismic data to be used synergistically in the same process, thereby improving the spatial and geological consistency of high-resolution reservoir inversion results.

[0021] In a further technical solution, segmented control of the restricted propagation results is implemented using segment reliability maps and residual cause identification maps. This allows different segments to be differentiated and processed according to their reliability and residual causes before entering the inversion process. This separates local time-depth migrations, local wavelet mismatches, well data anomalies, and seismic response anomalies from the unified inversion residuals, preventing the misinterpretation of local data relationship problems as actual reservoir changes in the results. It also ensures that the use of well constraints within a segment aligns with the actual matching state of that segment, further improving the stability of the local segment inversion results.

[0022] In a further technical solution, a frequency availability map is used to perform frequency control on the target well logging data, allowing different analysis units to select data from different frequency bands for inversion according to their corresponding frequency availability levels. This ensures that the introduction of high-frequency information matches the seismic resolution capability of the specific analysis unit, avoiding the forced writing of surface high-frequency details in locations unsuitable for high-frequency injection. This makes the scope and intensity of high-frequency information application more consistent with the actual spatial conditions of the subsurface target layer, further improving the interpretability of the inversion results in thin layers, boundaries, and complex areas. Attached Figure Description

[0023] Figure 1 This is the overall flowchart of the high-resolution reservoir inversion method driven by the well-seismic joint large model of the present invention. It intuitively shows the top-down implementation steps from data acquisition and input in the work area, through map generation and multi-dimensional fine control execution, to finally obtain high-resolution inversion results. Figure 2 This is a schematic diagram of the deployment of the physical layer hardware structure and the spatial interaction relationship between the geological layer and the present invention. It shows the layout of the physical carrier for assisting in the acquisition of sensing information by connecting various types of intelligent sensor devices deep in the target wellbore through the surface processing module. Figure 3 This is an abstract diagram of the core input-output logical relationship of the well-seismic joint large model of this invention. The left side integrates various basic logging and sensing inputs, while the right side outputs five key control charts in two major categories for propagation control and inversion control, respectively. Figure 4 This is a partition diagram of the benchmark unification customization and subsequent networked mapping and segmentation process of multi-source data in this invention, showing the transformation steps of the original data from time depth and other coordinate matching preprocessing to subsequent completion of analysis units and fine segmentation; Figure 5 This invention presents a frequency band progressive decomposition and multi-tier availability classification control model diagram, demonstrating the control principle of dynamically filtering and writing into the entire frequency band, eliminating high frequencies, or even directly truncating all frequency band features based on the four determined availability levels. Detailed Implementation

[0024] The invention will now be further described with reference to the accompanying drawings.

[0025] The core idea of ​​this invention is not to have the model directly output the final inversion parameters, but rather to first have a large-scale well-seismic joint model interpret the correspondence between well, seismic, and auxiliary sensing information. This involves generating propagation control maps and inversion control maps, and then using these maps to constrain the propagation range, layer usage, and frequency writing method of the well-constrained detail sequences, ultimately completing the high-resolution reservoir inversion. The reason for this design is that well data has the advantage of local realism and high vertical resolution, seismic data has the advantage of horizontal continuity and large coverage, and the large-scale model has the advantage of simultaneously processing multiple inputs and making unified judgments on various relationships. Therefore, performing control first and then inversion is more conducive to integrating the advantages of all three into a single technological chain.

[0026] like Figure 1 As shown, the overall implementation process of this invention includes nine stages: auxiliary sensing information acquisition, target work area data acquisition, benchmark unification, spatial mapping and segmentation, well-seismic joint large model mapping, well constraint detail sequence restricted propagation, segmented control, frequency control, and reservoir inversion. These stages are both sequential and data transfer-related; the input used in a later stage is not the original data, but rather the processed result of the previous stage.

[0027] In one embodiment, the target work area is a continental sandstone-mudstone reservoir area, or a marine clastic rock reservoir area or a carbonate rock reservoir area. The target well can be any of a vertical well, a directional well, or a horizontal well. The target work area data includes at least the target work area seismic data, target well logging data, geological interpretation data, and well-seismic calibration data. The target work area seismic data can be a three-dimensional post-stack seismic body, or in another embodiment, it can be extended to a multi-body input including seismic attribute bodies. The target well logging data can include sonic transit time curves, density curves, natural gamma curves, resistivity curves, neutron curves, and P-wave impedance curves converted from these curves. The geological interpretation data can include stratigraphic interpretation results, fault interpretation results, structural interpretation results, and sedimentary facies interpretation results. The well-seismic calibration data can include synthetic records, well-side seismic traces, time-depth relation tables, and well-seismic matching results.

[0028] like Figure 2As shown, this invention deploys smart sensors at the physical layer. These smart sensors can be installed inside the wellbore, near the wellhead, or at surface microseismic observation points. A commonly used combination is to deploy downhole pressure sensors, downhole temperature sensors, distributed acoustic sensors, and distributed temperature sensors inside the wellbore, and microseismic sensors at the surface or near the well. The purpose of this design is not to replace seismic or logging data with smart sensors, but to provide dynamic and auxiliary information for joint well-seismic analysis. Pressure information reflects reservoir connectivity and local fluid activity; temperature information reflects the state of the wellbore and near-wellbore zone; vibration and microseismic event information reflects fracture activation, micro-disturbances, and local energy activity; and distributed response information along the wellbore can help determine whether there are anomalous disturbance zones near a certain well section. While these information may not directly determine reservoir parameters when interpreted individually, they are very helpful in determining whether well-constrained detail sequences are propagable and whether the well-seismic relationship in a certain section is reliable.

[0029] In one embodiment, the sampling interval of the downhole pressure sensor can be from 1 second to 60 seconds, the sampling interval of the downhole temperature sensor can be from 1 second to 60 seconds, the channel spacing of the distributed acoustic sensor can be from 1 meter to 10 meters, the sampling spacing of the distributed temperature sensor can be from 1 meter to 10 meters, and the event localization accuracy of the microseismic sensor can be controlled within the range of 5 meters to 30 meters. These parameters are not necessarily fixed to a single value, but can be adjusted according to the well depth, work area scale, and sensor deployment density. The principle for parameter selection is to ensure that the auxiliary sensing information corresponds spatially with the seismic data and well logging data, is synchronized temporally, and can participate in unified analysis in terms of resolution.

[0030] Before target work area data and auxiliary sensing information are incorporated into the large model, baseline unification and spatial mapping need to be completed. For example... Figure 4 As shown, seismic data for the target work area are first unified to the same time reference and sampling interval. In actual processing, 3D seismic volumes often use a 2ms or 1ms sampling interval; if the sources are different, they can be uniformly resampled to 2ms. Well logging data for the target well is first unified to the same depth reference, then converted into time-domain curves via well-seismic calibration data, and further unified to a sampling interval consistent with the seismic volume. Geological interpretation data and well-seismic calibration data need to be unified to the same coordinate reference, typically using a unified plane coordinate system and depth reference surface for the work area. Auxiliary sensing information requires time synchronization and coordinate matching; for example, mapping microseismic event information to event points in the unified coordinate system of the work area, and mapping well-side data from distributed acoustic sensors and distributed temperature sensors to discrete sampling points on the well trajectory. After benchmark unification, all data is mapped to the unified work area coordinate system.

[0031] After mapping, network segmentation is required. In one embodiment, the work area is divided into multiple analysis units according to a preset grid. The analysis units can be voxel units or gather units. If voxel units are used, the horizontal grid size can be selected from 10m to 25m, and the vertical grid size can be selected from 1ms to 4ms. If the thin layers in the work area are significantly developed, the vertical size is preferably 1ms to 2ms. Using smaller analysis units is to make propagation control and frequency control truly spatially targeted, rather than making it a uniform control across the entire area. At the same time, the logging data of the target well is segmented according to a preset stratigraphic interface to obtain multiple segments. A segment can be a single small layer or a micro-segment composed of several fine layers. If the segment division is too coarse, local mismatches will be difficult to handle individually; if the segment division is too fine, it will cause noise sensitivity. In practice, 0.5m to 5m can be preferred as the initial segment thickness range, and local areas can be merged or subdivided according to the stratigraphic interpretation results.

[0032] like Figure 3 As shown, the combined well-seismic model outputs two types of control maps after receiving multi-source inputs. The propagation control maps include semantic connectivity maps and blocking boundary maps. The inversion control maps include segment reliability maps, residual cause identification maps, and frequency availability maps. These five maps are not ordinary static images, but data maps that can be processed by software and directly called by the inversion module. In actual implementation, the semantic connectivity map can be represented as a three-dimensional raster, with each analysis unit corresponding to an integer label; different labels represent different semantic connectivity domains. The blocking boundary map can be represented as a three-dimensional boundary volume or as a binary mask composed of multiple boundary voxels, where voxels with a value of 1 represent the location of the blocking boundary. The segment reliability map can be represented as a well segment index table or a well-side columnar classification map, with each segment corresponding to a reliability level. The residual cause identification map can be represented as a segment category table or a well-side classification strip map, with each segment corresponding to a residual cause category. The frequency availability map can be represented as a three-dimensional hierarchical raster map of the work area, with each analysis unit corresponding to a frequency availability level. To facilitate manual verification, the above-mentioned images can be further visualized as color cross-sections, layer slices, well-side columnar sections, and 3D volume renderings at the display level. However, within the system, they are essentially computable label images and mask images.

[0033] In one embodiment, the combined well-seismic model employs a multimodal coding and fusion decoding architecture. The model includes a seismic coding branch, a well logging coding branch, an auxiliary sensing coding branch, a geological interpretation coding branch, and a fusion decoding branch. The seismic coding branch can use a hybrid structure of 3D convolutional networks and Transformers, where the 3D convolutional network is used to extract local spatial textures, and the Transformer is used to extract spatial dependencies over a larger area. The well logging coding branch can use a combination of 1D convolutional networks and bidirectional gated recurrent units to extract sequence patterns along the well's longitudinal direction. The auxiliary sensing coding branch can use a time-series Transformer to process the time-series features of pressure, temperature, vibration, and microseismic events. The geological interpretation coding branch can use an embedded layer plus attention module to process discrete category features formed by horizon, fault, structure, and facies interpretation. The fusion decoding branch can use a multi-head cross-attention structure to fuse the seismic features at the location of the analysis unit, the features of adjacent well segments, auxiliary sensing features, and interpretation features, and then output semantic connectivity identifiers, blocking boundary identifiers, segment reliability levels, residual cause identifiers, and frequency availability levels, respectively.

[0034] During large-scale model training, labeled training samples need to be constructed. These samples can come from historical work areas, known blocks that have already been interpreted, or manually labeled samples from the current work area. Semantic connectivity labels can be manually delineated by interpreters based on fault, facies zone, stratigraphic level, and well-seismic correspondence, or they can be automatically inherited from validated reservoir units in historical work areas and then manually corrected. Blocking boundary labels can be generated comprehensively based on fault interpretation, pinch-out boundary interpretation, phase transition boundary interpretation, and well-side microseismic anomaly zones. Stratum reliability level labels can be determined based on historical well-seismic matching quality, repeat calibration consistency, and known well-side validation results. Residual cause identification labels can be supervised and labeled from clearly identified time-depth migration segments, wavelet mismatch segments, well data anomaly segments, and seismic response anomaly segments in historical work areas. Frequency availability level labels can be formed based on the seismic dominant frequency, signal-to-noise ratio, continuity, well-seismic consistency, and interpretation stability of the target location. The training process can employ a multi-task joint loss function, where semantic connectivity identifiers, blocking boundary identifiers, layer reliability levels, residual cause identifiers, and frequency availability levels each correspond to a classification loss term, and the total loss is the weighted sum of all classification loss terms. To ensure the model is not dominated by a single task, the loss weights for semantic connectivity and blocking boundaries can be set to 0.25 to 0.35, the loss weights for layer reliability and residual cause can be set to 0.15 to 0.25, and the loss weight for frequency availability level can be set to 0.10 to 0.20. The training batch size can be 4 to 16, the learning rate can be 1×10^-5 to 5×10^-4, and the number of training epochs can be 50 to 300. During inference, the large model outputs corresponding labels and their confidence levels for each analysis unit and each layer. Regions with excessively low confidence levels can undergo manual review, and the confidence threshold can be set to 0.60 to 0.85.

[0035] The purpose of designing the semantic connectivity graph in this invention is to clearly define the propagable region of well constraint detail sequences. Traditional approaches often only consider spatial distance or the similarity of single waveforms, while this invention first determines whether two locations are within the same semantic connectivity region. The same semantic connectivity region can be understood as a geological unit belonging to the same set of shared well constraint information. If a high-frequency detail corresponding to a well segment belongs to the interior of a channel sand body, it should not propagate across the channel boundary to the adjacent mudstone zone; similarly, if a well segment is located within a fault block, it should not propagate across faults to another fault block. Therefore, the semantic connectivity graph and the blocking boundary graph are used in conjunction: one tells the system where propagation is allowed, and the other tells the system where propagation must stop.

[0036] The constrained propagation process first extracts well constraint detail sequences from each segment of each target well. These well constraint detail sequences can be the P-wave impedance detail sequences corresponding to the target well segment, or the intra-layer property variation sequences obtained by converting acoustic, density, and other logging curves. Extraction is preferably performed in the time domain for direct comparison with seismic traces. For each analysis unit, the system first filters target well segments with the same semantic connectivity identifier and consistent layer position. Layer consistency can be determined by layer number or the overlap ratio of the top and bottom time windows of the layer. The time window overlap ratio threshold can be set to 0.70 to 0.95. Subsequently, for each selected target well segment, an intra-layer connection path is constructed between the analysis unit and the target well segment. The intra-layer connection path can be obtained by shortest path search along the target layer plane in a unified work area coordinate system, or recursively along adjacent cells of the layer grid. The reason for constructing paths within the layer, rather than using straight lines, is that straight lines may extend beyond the target layer, introducing cross-layer propagation errors.

[0037] When the connecting path within a layer intersects with the blocking boundary marker on the blocking boundary map, the propagation of the target well segment to the analysis unit is immediately stopped. This action reflects a hard constraint relationship, aiming to prevent analysis units with seemingly similar local waveforms from crossing faults, pinch-out boundaries, or phase transition boundaries to borrow well details that should not be borrowed. If the connecting path does not intersect with the blocking boundary marker, the local waveform similarity comparison continues. Local waveform similarity can be achieved using normalized cross-correlation coefficients. Specifically, a time window centered on the analysis unit is extracted from the seismic trace corresponding to the analysis unit, and a time window of the same length is extracted from the seismic trace at the well location corresponding to the target well segment. Then, the normalized cross-correlation value of the waveform sequences within the two time windows is calculated. The time window length can be selected based on the dominant frequency and the target layer thickness, preferably 8ms to 40ms. If the dominant frequency is low, it can be appropriately increased. The local waveform similarity threshold can be set to 0.65 to 0.95. When only one target well segment meets the non-intersection condition, its well constraint detail sequence can be directly written into the analysis unit. When multiple target well segments simultaneously meet the non-intersecting condition, the target well segment with the highest local waveform similarity is selected and written into the analysis unit. If the difference in local waveform similarity among multiple target well segments is less than 0.03, a distance factor can be added as a secondary discrimination condition in another embodiment to avoid multiple candidate well segments competing for the same analysis unit for a long time. Through this process, the target well segment identifier and the corresponding well constraint detail sequence can be determined for each analysis unit, and these two together constitute the constrained propagation result.

[0038] After obtaining the restricted propagation results, they are not directly fed into the inversion; segmented control is still required. The reason for segmented control is that well-seismic mismatch is often a problem in a local segment, rather than an overall well failure. Without segmented control, local errors will be carried over to the entire well. Segmented control first reads the segment reliability level and residual cause identifier corresponding to each target well segment. The segment reliability level can be divided into three levels: Level 1, Level 2, and Level 3. Level 1 indicates that the well-seismic relationship of this segment is stable and can be directly used as a well constraint. Level 2 indicates that this segment can be used, but requires additional correction. Level 3 indicates that this segment is not suitable for direct use as the original well constraint and needs to be replaced or restricted. The segment reliability level can be determined by the output of the well-seismic joint large model, or by adding a rule verification step after training. Rule verification can be performed using local waveform similarity, well-seismic time difference, intelligent sensor anomaly degree, and neighboring well consistency. For example, when the local waveform similarity is higher than 0.85, the well-vibration time difference is less than 2ms, and there are no obvious abnormalities in the auxiliary sensing information, it can be preferentially classified as the first reliability level; when the local waveform similarity is between 0.70 and 0.85, and the well-vibration time difference is between 2ms and 6ms, it can be classified as the second reliability level; those below the above range can be classified as the third reliability level.

[0039] When the reliability level of a segment is the first reliability level, the well constraint detail sequence corresponding to the target well segment is written into the segmented control results. When the reliability level of a segment is the second or third reliability level, it needs to be processed according to the residual cause identifier. The residual cause identifier includes at least the time-depth migration identifier, wavelet mismatch identifier, well data anomaly identifier, and seismic response anomaly identifier. The time-depth migration identifier indicates that the problem of this segment mainly comes from the local migration of the time-depth relationship. At this time, local time-depth correction can be performed on the target well segment. The local time-depth correction can be implemented by locally stretching or compressing the time correspondence between the synthetic record and the well-side earthquake under the constraints of the top and bottom interfaces of this segment, so that the key reflection corresponding points within this segment are realigned. The local time adjustment amount is preferably controlled within the range of 1ms to 8ms. The wavelet mismatch identifier indicates that the seismic wavelet of this segment does not match the wavelet used to construct the synthetic record. At this time, local wavelet replacement can be performed on this segment. Local wavelet replacement can be achieved by re-estimating the dominant frequency, phase, and waveform morphology of the wavelet within a window surrounding the layer. The dominant frequency adjustment range can be 5Hz to 20Hz, and the phase adjustment range can be -45° to 45°. Well data anomaly indicators suggest that the logging data for this layer may be affected by enlargement, invasion, instrument malfunction, etc. In this case, the adjacent well layer located at the same level as the target well and with the smallest distance from the target well can be selected as a substitute well constraint. The adjacent well search radius can be set from 100m to 1500m. Seismic response anomaly indicators suggest that the seismic response around this layer may have local noise, acquisition defects, or unstable interpretation. In this case, the well constraint detail sequence corresponding to this layer should not be written into the segmented control results, while the seismic data of the target work area at the corresponding layer should be retained for subsequent inversion. The motivation for this processing is to avoid forcibly imposing unreliable well constraints on seismic data, but rather to allow seismic constraints to dominate in this area.

[0040] After completing segmented control, frequency control is applied to the target well logging data. For example... Figure 5As shown, the goal of frequency control is not to uniformly enhance high frequencies, but rather to determine which frequency bands can be written based on the analysis unit. First, frequency band decomposition is performed on the target well logging data corresponding to each target well segment, from low to high center frequency, to obtain first, second, and third frequency band data. In practice, bandpass filter decomposition can be used based on the time-domain curves after well-seismic calibration. If the dominant seismic frequency in the work area is between 20Hz and 35Hz, it is preferable to set the first frequency band data to 0Hz to 20Hz, the second frequency band data to 20Hz to 40Hz, and the third frequency band data to 40Hz to 80Hz. If the dominant seismic frequency in the work area is lower, such as 15Hz to 25Hz, then the first frequency band data can be set to 0Hz to 15Hz, the second frequency band data to 15Hz to 30Hz, and the third frequency band data to 30Hz to 60Hz. The frequency band boundaries are not fixed values; the selection principle is to match the seismic resolution and well curve sampling accuracy of the target work area. The first frequency band data mainly carries background trends, the second frequency band data mainly carries medium-scale stratigraphic changes, and the third frequency band data mainly carries fine high-frequency changes.

[0041] Subsequently, for each analysis unit, the target well segment identifier corresponding to the restricted propagation results is read, and the frequency availability level corresponding to that analysis unit is read. The frequency availability level is divided into four levels. The first frequency availability level indicates that the analysis unit can carry all frequency band data; in this case, all first, second, and third frequency band data for the corresponding target well segment are written into the analysis unit. The second frequency availability level indicates that the high-frequency carrying capacity is somewhat reduced; in this case, only first and second frequency band data are written. The third frequency availability level indicates that only background and lower-frequency trends are allowed to participate; in this case, only first frequency band data is written. The fourth frequency availability level indicates that the analysis unit is not suitable for writing any well frequency band features; in this case, writing first, second, and third frequency band data is stopped. The frequency availability level division can be output from a large model or superimposed with rule constraints. Rule constraints can comprehensively consider local signal-to-noise ratio, seismic continuity, well-seismic corresponding stability, and the degree of anomaly sensing. For example, when the local signal-to-noise ratio is greater than 6dB, the continuity index is higher than 0.75, the waveform consistency between adjacent analysis units is high, and there are no obvious auxiliary sensing anomalies, it can be judged as the first frequency usable level; when the local signal-to-noise ratio is low or the boundary is complex, it can be downgraded to the second or third frequency usable level; when there are obvious fault fracture zones, microseismic anomaly clusters, or well-seismic instability, it can be judged as the fourth frequency usable level.

[0042] After the frequency control results are generated, the restricted propagation results are determined as the well constraint basic data. Then, the segmented control results are written into the well constraint basic data of the corresponding target well segment, and the frequency control results are written into the well constraint basic data of the corresponding analysis unit. The well constraint basic data generated in this way is no longer a simple extrapolation of the original single-well details, but a refined well constraint body after triple constraints of propagation range, segment reliability, and frequency level. Subsequently, the seismic data of the target work area is used as the seismic constraint, and the well constraint basic data is used as the well constraint input to the inversion module to solve for reservoir parameters for each analysis unit. The inversion module can employ any of the following: model constraint inversion, sparse pulse inversion, or geostatistical inversion. In one embodiment, model constraint inversion is used, with the goal of solving for the P-wave impedance body that satisfies both seismic and well constraints. If model constraint inversion is used, iterative least squares can be used. The number of iterations can be from 10 to 100, and the termination condition can be that the change in the objective function between two adjacent iterations is less than 1×10^-4 to 1×10^-6. Finally, the reservoir parameters obtained from each analysis unit are stitched together to obtain high-resolution reservoir inversion results. This result can be a three-dimensional P-wave impedance volume, or in other embodiments, it can be further converted into a porosity volume, lithology volume, or reservoir probability volume.

[0043] The five types of maps in this invention can be invoked by the algorithm and also visualized to help technicians understand the control process. The semantic connectivity map can be displayed as different colored blocks on seismic profiles or slices, with different colors representing different semantic connectivity domains. The blocking boundary map can be displayed in a 3D volume as boundary surfaces, boundary lines, or highlighted masks. The segment reliability map can be overlaid next to the wellbore column chart, with different color bands for different levels. The residual cause identification map can be displayed as classification label strips along the well segments. The frequency availability map can be displayed as a four-level color band map on horizontal slices or profiles. The reason for providing both visualizations is to facilitate manual review by interpreters; however, these maps are primarily algorithmic control charts in this invention, not simply display charts.

[0044] In one embodiment, the system comprises intelligent sensors, a data acquisition module, a map generation module, a restricted propagation module, a segmented control module, a frequency control module, and an inversion module. The intelligent sensors continuously collect auxiliary sensing information. The data acquisition module receives and processes data from the target work area. The map generation module incorporates a large-scale well-seismic joint model to output semantic connectivity maps, blocking boundary maps, layer reliability maps, residual cause identification maps, and frequency availability maps. The restricted propagation module performs intra-layer connectivity path search, boundary intersection discrimination, and local waveform similarity comparison. The segmented control module performs layer-level time-depth correction, wavelet replacement, substitution well constraint writing, and well constraint stop-write control. The frequency control module performs frequency band decomposition, frequency level reading, and frequency band writing. The inversion module solves for reservoir parameters and stitches the results. Figure 2 As shown, the surface processing module can serve as the data aggregation center for the entire system, communicating with downhole intelligent sensors via the network and connecting with the seismic interpretation workstation to form an integrated implementation platform from acquisition to inversion.

[0045] In another embodiment, the well-seismic joint large model can output all five types of maps in parallel instead of all at once, using a shared encoder and multiple task heads. Alternatively, it can output the propagation control map first, followed by the inversion control map. If the training samples in the work area are insufficient, a transfer training method can be used, i.e., pre-training on historical work areas and then fine-tuning with a small number of labeled samples from the target work area. If the number of wells in the work area is small, the window range of adjacent wells can be increased to improve the candidate coverage of the target well segment. If the dominant frequency of the seismic data is low, the upper limit of the third frequency band data can be appropriately narrowed to avoid writing false high frequencies that are not carried by seismic events. If there are short-term missing data from smart sensors, time proximity interpolation or adjacent depth interpolation within the same well can be used to complete the data, but the completion length should preferably not exceed 30 minutes to avoid over-reconstruction of auxiliary sensing information.

[0046] In another embodiment, local waveform similarity comparison is not limited to normalized cross-correlation; phase consistency index, envelope similarity index, or principal component projection similarity index can also be used. As long as the local waveform similarity between the seismic trace corresponding to the analysis unit and the seismic trace corresponding to the target well section can be quantified, it can be used as an alternative. When selecting the threshold, it is advisable to first perform a sensitivity analysis on historical work areas to observe the impact of threshold changes on the restricted propagation range, and then select a threshold range in the target work area that maintains connectivity while preventing mis-borrowing. Generally, a waveform similarity threshold that is too low will lead to cross-boundary mis-borrowing, while one that is too high will make it difficult to propagate well constraints. Therefore, a threshold of approximately 0.75 is preferred as the initial value, and then adjusted according to the complexity of the work area.

[0047] The above embodiments provide a relatively complete description of the present invention. Those skilled in the art can adjust the size of the analysis unit, the layer division method, the specific network structure of the large model, the training parameters, the frequency band boundaries, and the threshold settings without departing from the concept of the present invention; all such adjustments fall within the technical scope of the present invention.

[0048] 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 high-resolution reservoir inversion method driven by a combined well-seismic model, characterized in that, include: The system collects auxiliary sensing information and acquires target work area data by setting up intelligent sensors in the target work area. The target work area data and the auxiliary sensing information are input into the well-seismic joint large model to obtain the propagation control map and the inversion control map; Based on the propagation control map, restricted propagation is performed on the well constraint detail sequence corresponding to the target well logging data to obtain the restricted propagation result; Based on the inversion control map, segmented control is performed on the restricted propagation results, and frequency control is performed on the target well logging data to obtain segmented control results and frequency control results; Based on the seismic data of the target work area, the restricted propagation results, the segmented control results, and the frequency control results, reservoir inversion is performed to obtain high-resolution reservoir inversion results.

2. The high-resolution reservoir inversion method driven by a combined well-seismic model according to claim 1, characterized in that, The intelligent sensor is at least one of a downhole pressure sensor, a downhole temperature sensor, a distributed acoustic sensor, a distributed temperature sensor, and a microseismic sensor; the auxiliary sensing information includes at least one of pressure information, temperature information, vibration information, microseismic event information, and distributed response information along the well.

3. The high-resolution reservoir inversion method driven by a combined well-seismic model according to claim 1, characterized in that, The target work area data includes target work area seismic data, target well logging data, geological interpretation data, and well-seismic calibration data; the propagation control map includes a semantic connectivity map and a blocking boundary map; the inversion control map includes a layer reliability map, a residual cause identification map, and a frequency availability map.

4. The high-resolution reservoir inversion method driven by a combined well-seismic model according to claim 3, characterized in that, Before inputting the target work area data and the auxiliary sensing information into the well-seismic joint large model, the following steps are also included: The time reference and sampling interval for the seismic data of the target work area were unified; The depth benchmark and sampling interval were standardized for the logging data of the target well. The coordinate references for the geological interpretation data and the well-seismic calibration data shall be unified. Perform time synchronization and coordinate matching on the auxiliary sensing information; The target work area data and the auxiliary sensing information, after benchmark unification, are mapped to the unified work area coordinate system; The data within the unified work area coordinate system is divided according to a preset grid to obtain multiple analysis units; The logging data of the target well is segmented according to the preset layer interface to obtain multiple layers.

5. The high-resolution reservoir inversion method driven by a combined well-seismic model according to claim 4, characterized in that, The target work area data and the auxiliary sensing information are input into the well-seismic joint large model to obtain propagation control maps and inversion control maps, including: The seismic data of the target work area corresponding to each analysis unit, the well logging data of the target well corresponding to each layer, the geological interpretation data, the well-seismic calibration data, and the auxiliary sensing information are input into the well-seismic joint large model. The well-seismic joint large model outputs semantic connectivity identifiers for each analysis unit; The combined well-seismic model outputs boundary markers for the boundaries between adjacent analysis units. The combined well-seismic model outputs the reliability level and residual cause identifier for each segment; The combined well-seismic model outputs a frequency availability level for each analysis unit; The semantic connectivity graph is constructed based on each of the semantic connectivity identifiers; The blocking boundary map is constructed based on each of the aforementioned blocking boundary identifiers; Construct the reliability map of each layer based on its reliability level; Construct the residual cause identification map based on each of the aforementioned residual cause identifications; The frequency availability map is constructed based on each of the frequency availability levels.

6. The high-resolution reservoir inversion method driven by a combined well-seismic model according to claim 5, characterized in that, Based on the propagation control map, constrained propagation is performed on the well constraint detail sequence corresponding to the target well logging data to obtain the constrained propagation results, including: Extract well constraint detail sequences from each layer of each target well; For each analysis unit, target well segments with the same semantic connectivity identifier and consistent stratigraphic position as the analysis unit are selected; For each selected target well interval, an intra-stratal connection path is constructed between the analysis unit and the target well interval; When the connecting path within the layer intersects with the blocking boundary marker in the blocking boundary diagram, the propagation of the target well segment to the analysis unit is stopped; When the connecting path within the layer does not intersect with the blocking boundary marker in the blocking boundary map, a local waveform similarity comparison is performed between the seismic trace corresponding to the analysis unit and the seismic trace at the corresponding well location of the target well segment. When only one target well segment satisfies the non-intersection condition, the well constraint detail sequence corresponding to the target well segment is written into the analysis unit; When multiple target well segments meet the non-intersection condition, select the target well segment with the maximum local waveform similarity to the seismic trace corresponding to the analysis unit, and write the well constraint detail sequence corresponding to the selected target well segment into the analysis unit. The target well segment identifier corresponding to each analysis unit and the well constraint detail sequence written into each analysis unit are determined as the restricted propagation result.

7. The high-resolution reservoir inversion method driven by a combined well-seismic model according to claim 6, characterized in that, Based on the aforementioned inversion control chart, segmented control is performed on the restricted propagation result to obtain segmented control results, including: For each target well segment in the restricted propagation results, the corresponding segment reliability level and residual cause identifier are read; When the reliability level of the segment is the first reliability level, the well constraint detail sequence corresponding to the target well segment is written into the segmented control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is a time-depth offset identifier, local time-depth correction is performed on the target well segment, and the corrected well constraint detail sequence is written into the segmented control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is the wavelet mismatch identifier, local wavelet replacement is performed on the target well segment, and the replaced well constraint detail sequence is written into the segmented control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is a well data anomaly identifier, the adjacent well segment located in the same layer as the target well segment and with the smallest distance from the corresponding target well is selected as the alternative well constraint, and the alternative well constraint is written into the segment control result; When the reliability level of the segment is the second or third reliability level, and the residual cause identifier is a seismic response anomaly identifier, stop writing the well constraint detail sequence corresponding to the target well segment into the segmented control results, and retain the target work area seismic data of the corresponding layer of the target well segment to participate in the reservoir inversion.

8. The high-resolution reservoir inversion method driven by a combined well-seismic model according to claim 7, characterized in that, Based on the inversion control map, frequency control is performed on the target well logging data to obtain frequency control results. Then, based on the target area seismic data, the confined propagation results, the segmented control results, and the frequency control results, reservoir inversion is performed, including: Frequency band decomposition was performed on the target well logging data corresponding to each target well segment according to the center frequency from low to high, to obtain the first frequency band data, the second frequency band data and the third frequency band data; For each analysis unit, read the target well segment identifier corresponding to that analysis unit from the restricted propagation results; For each analysis unit, read the frequency availability level corresponding to that analysis unit; When the frequency availability level is the first frequency availability level, the first frequency band data, the second frequency band data, and the third frequency band data corresponding to the target well segment identifier are written into the frequency control result corresponding to the analysis unit; When the frequency availability level is the second frequency availability level, the first frequency band data and the second frequency band data corresponding to the target well segment identifier are written into the frequency control result corresponding to the analysis unit; When the frequency availability level is the third frequency availability level, the first frequency band data corresponding to the target well segment identifier is written into the frequency control result corresponding to the analysis unit; When the frequency availability level is the fourth frequency availability level, stop writing the first frequency band data, second frequency band data and third frequency band data corresponding to the target well segment identifier into the frequency control result corresponding to the analysis unit; The restricted propagation results are determined as the well constraint basic data; The segmented control results are written into the well constraint basic data of the corresponding target well section; The frequency control results are written into the well constraint base data of the corresponding analysis unit; The seismic data of the target work area are used as seismic constraints; Solve the reservoir parameters for each analysis unit separately; The reservoir parameters of each analysis unit are stitched together to form the high-resolution reservoir inversion result.

9. A high-resolution reservoir inversion system driven by a combined well-seismic large model as described in claim 1, characterized in that, include: Intelligent sensors are used to collect auxiliary sensing information; The data acquisition module is used to acquire data from the target work area; The map generation module is used to input the target work area data and the auxiliary sensing information into the well-seismic joint large model to obtain propagation control map and inversion control map; The restricted propagation module is used to perform restricted propagation on the well constraint detail sequence corresponding to the target well logging data based on the propagation control map, and obtain the restricted propagation result. The segmented control module is used to perform segmented control on the restricted propagation result based on the inversion control chart to obtain the segmented control result; The frequency control module is used to perform frequency control on the target well logging data based on the inversion control map to obtain the frequency control result; The inversion module is used to perform reservoir inversion based on the seismic data of the target work area, the restricted propagation results, the segmented control results, and the frequency control results to obtain high-resolution reservoir inversion results.

10. A high-resolution reservoir inversion system driven by a combined well-seismic model according to claim 9, characterized in that, The map generation module is used to output semantic connectivity maps, blocking boundary maps, segment reliability maps, residual cause identification maps, and frequency availability maps. The restricted propagation module is used to filter target well segments with the same semantic connectivity identifier and consistent layer as the analysis unit, construct intra-layer connection paths, stop propagation when the intra-layer connection path intersects with the blocking boundary identifier, and select the target well segment with the maximum local waveform similarity to the seismic trace corresponding to the analysis unit when multiple target well segments meet the propagation conditions and write it into the corresponding analysis unit. The segment control module is used to perform well constraint detail sequence writing, local time depth correction, local wavelet replacement, substitute well constraint writing, and well constraint detail sequence stop writing on the target well segments according to the segment reliability level and residual cause identifier. The frequency control module is used to write the corresponding frequency band data from the first frequency band data, second frequency band data, and third frequency band data corresponding to the target well segment identifier to the analysis unit according to the frequency availability level.

Images