Geological map processing method and system based on artificial intelligence

Abstract

Embodiments of the present application provide a geological map processing method and system based on artificial intelligence. The original geological map data set of the carbon dioxide geological storage target area is obtained. A pre-trained graphic element recognition model is called to perform graphic element feature recognition processing on the original geological map data set to generate an isogram graphic element set, a lithology pattern graphic element set, and a columnar pattern graphic element set. Attribute information reconstruction processing, coordinate correction and geometry repair processing are performed on the isogram graphic element set, the lithology pattern graphic element set, and the columnar pattern graphic element set to generate a standardized graphic element set with a unified target space reference system. The standardized graphic element set is transmitted to a target space database to perform regularized storage processing to obtain a carbon dioxide geological storage thematic map data set. The method can improve the accuracy and efficiency of carbon dioxide geological storage geological map processing.

Description

Technical Field

[0001] This application relates to the field of geological technology, and more specifically, to a geological map processing method and system based on artificial intelligence. Background Technology

[0002] Currently, geological carbon dioxide storage is one of the key technological pathways to address global climate change and achieve carbon neutrality. Accurate and efficient geological map processing is crucial for comprehensively assessing the geological conditions of storage sites and ensuring storage safety. At present, historical geological map data is mainly generated and stored using the MapGIS software platform. However, mainstream GIS applications, spatial analysis, and web publishing platforms are not directly compatible with or can utilize MapGIS's native format files, leading to a reliance on manual methods to convert these maps one by one into a usable format. However, existing manual processing methods suffer from low efficiency and poor accuracy, severely hindering the advancement of research related to carbon dioxide geological storage.

[0003] Therefore, improving the accuracy and efficiency of processing geological maps for carbon dioxide geological storage is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a geological map processing method and system based on artificial intelligence, so as to improve the accuracy and efficiency of processing geological maps for carbon dioxide geological sequestration.

[0005] Firstly, this application provides a geological map processing method based on artificial intelligence, comprising:

[0006] Obtain the original geological map data set of the target area for carbon dioxide geological sequestration. The original geological map data set includes contour map units, geological profile map units, and columnar map units stored in the format of the first geographic information system platform.

[0007] The pre-trained primitive recognition model is invoked to perform primitive feature recognition processing on the original geological map data set, generating a set of contour primitive features corresponding to contour map units, a set of lithological pattern primitive features corresponding to geological profile map units, and a set of columnar pattern primitive features corresponding to columnar map units.

[0008] The contour map element set, lithological pattern map element set, and columnar pattern map element set are input into the pre-trained geological feature extraction model to perform attribute information reconstruction processing, generating a map element set to be corrected carrying the original attribute information;

[0009] The set of primitive elements to be corrected is input into a pre-trained spatial reference alignment model to perform coordinate correction and geometric repair processing, generating a standardized set of primitive elements with a unified target spatial reference system;

[0010] The standardized set of map elements is transferred to the target spatial database for regularized data entry processing, and the corresponding processing metadata is generated to obtain a thematic map dataset of carbon dioxide geological storage.

[0011] Secondly, this application provides an artificial intelligence-based geological map processing system, which includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions, and when the processor executes the machine-executable instructions, the artificial intelligence-based geological map processing system implements the aforementioned artificial intelligence-based geological map processing method.

[0012] The method and system for processing geological maps based on artificial intelligence provided in this application accurately identify various map elements in the original geological map data using a pre-trained map element recognition model; then, a geological feature extraction model is used to completely reconstruct attribute information; next, a spatial reference system is unified through a spatial benchmark alignment model; finally, standardized map elements are stored in a database and metadata is generated, thereby realizing automated processing of carbon dioxide geological sequestration geological maps, reducing manual intervention, and thus improving the accuracy and efficiency of carbon dioxide geological sequestration geological map processing. Attached Figure Description

[0013] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0014] Figure 1 A flowchart illustrating an artificial intelligence-based geological map processing method provided in this application embodiment;

[0015] Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based geological map processing system provided in an embodiment of this application.

[0016] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Figure 1 This is a flowchart illustrating an artificial intelligence-based geological map processing method provided in this application embodiment. It should be understood that in other embodiments, the order of some steps in this artificial intelligence-based geological map processing method can be shared according to actual needs, or some steps can be omitted or maintained. Figure 1 As shown, the method may include the following steps:

[0019] Step S110: Obtain the original geological map data set of the target area for carbon dioxide geological sequestration. The original geological map data set includes contour map units, geological profile map units, and columnar map units stored in the format of the first geographic information system platform.

[0020] In this embodiment, for a pre-selected area for carbon dioxide geological sequestration, it is first necessary to collect basic geological data for the area. Specifically, a set of original geological map data can be obtained from historical exploration projects through a data interface. This set of original geological map data contains various types of map units, all of which are stored in a first geographic information system platform format. For example, the first geographic information system platform format can be a vector data format of the first geographic information system platform, such as a feature class or shapefile file format in a geographic database converted from an original MapGIS file.

[0021] This original geological map dataset includes contour map units reflecting stratigraphic undulations. These contour map units exist as line features in the format of the First Geographic Information System (GIS) platform, recording the elevation contours of multiple key stratigraphic interfaces (such as the top surface of reservoirs and the bottom surface of caprocks) in the target area. It also includes geological profile map units reflecting the vertical changes in subsurface geological structures. These geological profile map units also use the GIS platform format and are stored as area features or raster forms, containing lithological filling patterns of different strata and boundary information between strata. Furthermore, it may include columnar map units recording single-point vertical stratigraphic sequences, such as borehole columnar sections. These columnar map units may exist in the GIS platform format as point feature attribute tables or independent tables, describing lithological changes, grain size changes, and corresponding depth scale information from the surface to depth.

[0022] After obtaining the above data, store it in a temporary workspace for subsequent processing.

[0023] Step S120: Call the pre-trained primitive recognition model to perform primitive feature recognition processing on the original geological map data set, and generate the contour primitive feature set corresponding to the contour map unit, the lithological pattern primitive feature set corresponding to the geological profile map unit, and the columnar pattern primitive feature set corresponding to the columnar map unit.

[0024] This primitive recognition model is a deep learning-based multi-task segmentation model. Its core architecture can employ an encoder-decoder structure, with dedicated branch output heads for different types of map features. The encoder, for example, can use a residual network as the backbone to extract multi-level, multi-scale deep semantic features from the input map image. The decoder uses upsampling and skip connection operations to progressively restore the deep feature map to the resolution of the original input image, and assigns a class label to each pixel in the final output layer. During the training phase, this primitive recognition model uses a large amount of labeled geological map data for supervised learning, enabling it to accurately identify and segment various types of geological primitives.

[0025] Step S121: Input the contour map unit into the contour segmentation sub-network of the pre-trained primitive recognition model, and perform pixel-level classification on the contour map unit through the contour segmentation sub-network to identify the first pixel region corresponding to the closed contour primitive and the second pixel region corresponding to the open curve contour primitive.

[0026] First, for contour map units, this embodiment extracts them from the original dataset, performs necessary format conversion (e.g., rendering vector format into high-resolution raster images), and then inputs them into a specially designed contour segmentation sub-network in the pre-trained primitive recognition model.

[0027] The input to this contour segmentation subnetwork can be, for example, a three-channel color image, whose dimensions are uniformly adjusted to the product of the width (W_in) and height (H_in) in pixels. The image data can be standardized before entering the network; that is, the values ​​of the red, green, and blue channels of each pixel are subtracted from the mean of the training set images and then divided by the standard deviation of the training set images to obtain standardized tensor data.

[0028] The forward propagation process of this sub-network begins in the encoder section. The input tensor passes through multiple convolutional blocks sequentially. Each convolutional block contains a convolutional layer, a batch normalization layer, and a linear rectified activation function. For example, the first convolutional block uses a convolutional kernel with a stride of 2 for downsampling, outputting a feature map with 64 channels. The width and height of this feature map are halved compared to the input image, i.e., the width is W_in divided by 2 and the height is H_in divided by 2. Next, the second convolutional block continues to use a convolutional kernel with a stride of 2, outputting a feature map with 128 channels, further halving the size. After several (e.g., four to five) downsampling blocks, the deepest high-dimensional semantic feature map is obtained.

[0029] Then, the decoder starts working, gradually restoring the resolution of the feature map through upsampling layers (such as transposed convolutions), and concatenating the feature map of the current layer of the decoder with the feature map of the corresponding resolution in the encoder through skip connections, so as to fuse shallow detailed information and deep semantic information.

[0030] After a series of upsampling and convolution operations, the final output layer can use a convolutional layer with three kernels (corresponding to three categories: background, closed contour lines, and open contour lines), and apply the Softmax activation function to generate a probability map with the same size as the input image. This probability map has a width of W_in, a height of H_in, and three channels.

[0031] For each pixel location, the three values ​​in the probability map represent the probability that the pixel belongs to the background, closed contour primitive, or open contour primitive, respectively. By taking the category with the highest probability at each pixel location, the pixel-level classification result can be obtained. All pixels marked as closed contour primitives constitute the first pixel region, and all pixels marked as open contour primitives constitute the second pixel region.

[0032] Step S122: Extract the boundary coordinate sequence of closed contour primitives based on the first pixel region, and extract the path coordinate sequence of open curve contour primitives based on the second pixel region to generate a set of contour primitive elements.

[0033] In this step, for the first pixel region, i.e., the pixel region corresponding to the closed contour primitive, a boundary tracking algorithm (such as the eight-neighbor boundary tracking algorithm) can be used. Taking the eight-neighbor boundary tracking algorithm as an example, it can start from a boundary pixel point and search for its eight neighboring pixels in a clockwise or counterclockwise direction to find the next boundary point and record the image coordinates (row coordinate value, column coordinate value) of that point. This process is repeated until it returns to the starting point, forming a closed contour.

[0034] Since image coordinates are pixel-based, they need to be converted to geographic coordinates. At this point, the geographic resolution of each pixel can be calculated based on the geographic coordinates of the four corner points of the image (top left corner X_min, Y_max, bottom right corner X_max, Y_min) and the image width W_in and height H_in. Specifically, the X-direction resolution equals (X_max minus X_min) divided by W_in, and the Y-direction resolution equals (Y_max minus Y_min) divided by H_in.

[0035] Then, for each image coordinate point (row coordinate value R_pixel, column coordinate value C_pixel) obtained by the boundary tracking algorithm, it can be converted into geographic coordinates using the affine transformation formula: geographic coordinate X equals X_min plus C_pixel multiplied by the X-direction resolution, and geographic coordinate Y equals Y_max minus R_pixel multiplied by the Y-direction resolution.

[0036] Arrange all the converted geographic coordinate points in the tracking order to obtain the boundary coordinate sequence of the closed contour map element.

[0037] For the second pixel region, i.e., the pixel region corresponding to the open curve contour primitive, vectorization is also required. For example, a skeleton extraction algorithm, such as a morphological thinning algorithm, can be used to thin the pixel region of a certain width into a center line of single pixel width before tracking. The tracking algorithm used in this process is similar to that used for the second pixel region, except that it does not require a closure condition. The tracking of a curve is completed when no next neighboring point is found or an image boundary is encountered.

[0038] The same affine transformation is applied to each tracked pixel to convert it into geographic coordinates, thus obtaining the path coordinate sequence of the open curve contour primitive.

[0039] By traversing all identified connected regions, a corresponding coordinate sequence is generated for each closed or open contour feature. This geometric data, along with metadata extracted from the original map (such as map name, stratigraphy, etc.), is then organized to form a contour feature set. Each feature in this set contains at least a unique identifier and an ordered set of coordinate pairs.

[0040] Step S123: Input the geological profile map unit into the profile segmentation sub-network of the pre-trained primitive recognition model, and perform pixel-level classification on the geological profile map unit through the profile segmentation sub-network to identify the third pixel region corresponding to the lithological filling pattern primitive and the fourth pixel region corresponding to the stratigraphic boundary primitive.

[0041] In this step, geological profile map elements from the original geological map dataset can be input into the profile segmentation subnetwork of the pre-trained primitive recognition model. The input to the profile segmentation subnetwork is a geological profile raster image that has undergone size unification and standardization. Its network structure is similar to that of the contour line segmentation subnetwork, but the output layer category settings are different, targeting specific components of the profile map.

[0042] The encoder-decoder structure of the profile segmentation subnetwork can be the same as described in step S121, extracting deep features through multi-layer convolution and pooling downsampling, and then restoring resolution through upsampling. The final output layer has four convolutional kernels, corresponding to the four categories: background, lithological filling pattern primitives, stratigraphic boundary primitives, and other elements (such as text labels). After the Softmax activation function, a four-channel probability map of the same size as the input image is obtained. By selecting the category with the highest probability at each pixel location, pixel-level fine segmentation of the geological profile is achieved.

[0043] In the segmentation results, all pixels classified as lithological filling pattern primitives constitute the third pixel region. These regions appear as planar areas with certain textures and colors, representing strata of different lithologies. All pixels classified as stratigraphic boundary primitives constitute the fourth pixel region. These regions are thin lines or narrow bands, representing the boundaries between different stratigraphic units.

[0044] Step S124: Extract the color distribution features and texture distribution features of the lithological filling pattern primitives based on the third pixel region, and extract the extension direction features of the stratigraphic boundary primitives based on the fourth pixel region to generate a set of lithological pattern primitive elements and a set of stratigraphic boundary primitive elements.

[0045] In this step, for the third pixel region, i.e. the lithology-filled pattern primitive region, it is necessary to extract visual features that can characterize its lithology type.

[0046] First, the color distribution characteristics can be calculated for each independent connected region. For example, the values ​​of all pixels in the red-green-blue color space within the region can be obtained, and the mean and variance of the red, green, and blue channels can be calculated to form a six-dimensional color statistical feature vector. Furthermore, to capture its texture information, the gray-level co-occurrence matrix (GLCM) method can be used to convert the region into a grayscale image. Then, its GLCM in multiple directions can be calculated, and four texture feature parameters—contrast, correlation, energy, and homogeneity—can be extracted from each matrix to obtain texture feature parameters. These texture parameters are then concatenated with the aforementioned color statistical parameters to form a feature vector representing the color and texture distribution characteristics of the lithological filling pattern primitives.

[0047] For the fourth pixel region, i.e., the stratigraphic boundary primitive region, a skeleton extraction can be performed on each independent boundary connected region to obtain a center line with a single pixel width. Then, straight lines or curves are fitted to all pixels on this center line to determine its overall orientation. For example, the least squares method can be used to fit the points on the line to a straight line, and the slope of this line can be calculated using the arctangent function to obtain its azimuth relative to the horizontal direction. In addition, the local orientation of each point on the center line can be calculated, i.e., by calculating the direction of the vector formed by several adjacent points, to reflect the curvature of the boundary within a local range. Statistical processing of the above orientation information yields a orientation histogram, which serves as the extended orientation feature vector of this stratigraphic boundary primitive.

[0048] Through the above steps, a feature vector containing color and texture features is generated for each lithological filling pattern primitive, and a feature vector containing extension direction features is generated for each stratigraphic boundary primitive. These feature vectors are then associated with their respective geometric coordinate sequences to form a set of lithological pattern primitive elements and a set of stratigraphic boundary primitive elements.

[0049] Step S125: Input the columnar image unit into the columnar segmentation sub-network of the pre-trained primitive recognition model, and perform pixel-level classification on the columnar image unit through the columnar segmentation sub-network to identify the fifth pixel region corresponding to the lithological columnar pattern primitive and the sixth pixel region corresponding to the depth scale line primitive.

[0050] In this step, columnar plot units from the original geological map dataset are input into the columnar segmentation subnetwork of a pre-trained primitive recognition model. The input to the columnar segmentation subnetwork is a standardized and uniformly sized columnar raster image. The output layer of the network structure has four convolutional kernels, corresponding to the background, lithological columnar pattern primitives, depth scale line primitives, and other auxiliary primitives (such as lithological symbols and text). Through forward propagation of the model, a four-channel probability map of the same size as the input image is finally output. Pixel-level classification based on this probability map yields the fifth pixel region corresponding to the lithological columnar pattern primitive and the sixth pixel region corresponding to the depth scale line primitive.

[0051] The fifth pixel region appears as a vertical band composed of color bands with different gray levels or textures, reflecting lithological information that varies with depth. The sixth pixel region, however, is along a columnar... Figure 1 Short, thin lines perpendicular to the depth direction, distributed on the sides or both sides, and possibly accompanied by scale value areas.

[0052] Step S126: Extract the gray value sequence of the lithological columnar pattern primitives based on the fifth pixel region, and extract the scale value sequence of the depth scale line primitives based on the sixth pixel region to generate a set of columnar pattern primitive elements and a set of depth scale primitive elements.

[0053] In this step, for the fifth pixel region, i.e., the lithological columnar pattern primitive, the core information is the variation along the depth direction. Specifically, the main axis direction of the columnar pattern can be determined first, for example, vertically. Then, along the main axis direction, the pixels of each row (or a strip of fixed height) are analyzed from top to bottom (or from bottom to top, corresponding to depths from shallow to deep). The average grayscale value of all pixels belonging to the fifth pixel region within each row is calculated. Next, the average brightness value is calculated for each row, resulting in a grayscale value sequence that varies with the row index. Each value in this sequence represents the average grayscale at that depth position, indirectly reflecting the grain size or compositional variation of the lithology. This grayscale value sequence is a one-dimensional vector whose length is equal to the number of pixel rows in the columnar pattern along the depth direction.

[0054] For the sixth pixel region, i.e., the depth scale primitive, it is necessary to identify the position of each scale line and the depth value it represents. First, each scale line can be treated as an independent primitive through connected component analysis. For each scale line primitive, its vertical position (row coordinate) in the image is determined by the centroid coordinates of its pixel region. Then, optical character recognition (OCR) technology is used to identify the numerical regions adjacent to the scale line primitive and convert the numbers in the image into their corresponding values.

[0055] This method yields a series of point pairs corresponding to depth values ​​and image row coordinates. These point pairs constitute a mapping relationship between depth scales and image coordinates, allowing the fitting of a transformation function between depth and row coordinates. For example, the depth value equals the slope multiplied by the row coordinate plus the intercept. This slope represents the actual depth difference represented by each pixel row. Using this transformation function, the row coordinates corresponding to each grayscale value in the grayscale value sequence obtained in step S126 can be converted into the actual formation depth, thus obtaining the depth value corresponding to each grayscale value.

[0056] By integrating this information, a sequence of data pairs containing depth and corresponding grayscale values ​​can be generated for each lithological columnar pattern primitive, thus forming a set of columnar pattern primitive elements. Simultaneously, the identified scale line positions and their numerical information can be organized into a set of depth scale primitive elements.

[0057] Step S130: Input the contour map element set, lithological pattern map element set, and columnar pattern map element set into the pre-trained geological feature extraction model to perform attribute information reconstruction processing, and generate a map element set to be corrected carrying the original attribute information.

[0058] The goal of the geological feature extraction model is to reconstruct the geological attribute information represented by these raw primitives that only contain geometric and visual features, such as elevation values, lithological types, and stratigraphic thicknesses, so as to lay the foundation for subsequent spatial analysis and correction.

[0059] Step S131: Input the boundary coordinate sequence and path coordinate sequence into the first attribute reconstruction sub-network of the pre-trained geological feature extraction model. The first attribute reconstruction sub-network calculates the first elevation attribute value corresponding to each closed contour map element based on the boundary coordinate sequence, and generates the second elevation attribute value corresponding to each open curve contour map element based on the path coordinate sequence and the preset legend elevation reference table.

[0060] First, the boundary coordinate sequence of closed contour lines and the path coordinate sequence of open contour lines from the contour feature set can be input into the first attribute reconstruction sub-network. This first attribute reconstruction sub-network can employ different elevation attribute reconstruction strategies for different types of contour lines. For closed contour lines, they typically represent a closed loop, and the elevation values ​​within the loop are either the same (e.g., contour lines of islands in lakes) or need to be inferred through other methods. In this embodiment, the first attribute reconstruction sub-network first calculates the geometric center point of the region enclosed by the boundary coordinate sequence of the closed contour lines. Specifically, the center point coordinates are obtained by calculating the average of the coordinates of all vertices on the boundary. Then, the sub-network does not directly calculate the elevation of the closed contour line itself, but rather uses it as an important reference. Its elevation value is often related to its adjacent open contour lines, so the determination of its attribute value will be combined with contextual information and processed uniformly in subsequent steps.

[0061] For open-curve contour map elements, their path coordinate sequence directly reflects the strike and location of the strata. The first attribute reconstruction sub-network is used to assign specific elevation attribute values ​​to the aforementioned open curves. Specifically, the model can maintain a pre-set legend elevation reference table. This legend elevation reference table can be constructed by learning and parsing the legends of a large number of historical geological maps, storing the mapping relationship between different line types, colors, label styles, and specific elevation values. For example, the legend elevation reference table records that "brown solid line, line width 0.5 mm" corresponds to an elevation value of -1000 meters, and "blue dashed line, line width 0.25 mm" corresponds to an elevation value of -500 meters, etc.

[0062] The first attribute reconstruction subnetwork first extracts the appearance features of each input open curve contour map element, including its color (extracted from the original map), line type (solid, dashed, dotted, etc.), and line width. Then, these appearance features are matched against entries in a legend elevation lookup table. The matching process can be accomplished, for example, by calculating the similarity between feature vectors. For instance, the line type can be encoded as a category vector, the color as a three-channel (red, green, blue) value, and the line width as a scalar value. These are then compared with each entry in the lookup table, and the most similar entry is found. The elevation value associated with this entry is then determined as the second elevation attribute value of the open curve contour map element.

[0063] Optionally, if a contour line is directly labeled with a number, the number can be identified as its elevation value using optical character recognition (OCR).

[0064] In the above manner, an elevation value is assigned to each open curve contour map element.

[0065] Step S132: Add the first elevation attribute value and the second elevation attribute value as attribute fields to the contour map element set to generate a contour map element set carrying elevation attribute information.

[0066] In this step, a field named "Elevation Value" can be added to the attribute table of each contour feature. For open contour features, the second elevation attribute value obtained in step S131 can be written into this field. For closed contour features, the first elevation attribute value can be temporarily marked as null or a special identifier, waiting to be interpolated or assigned later based on neighborhood information.

[0067] Optionally, to maintain information integrity, other visual features of contour primitives, such as line type codes and color values, can also be stored as auxiliary attribute fields. After the above processing, the original set of contour primitive features containing only geometric coordinates can be transformed into a set of contour primitive features carrying key elevation attribute information, providing the necessary data foundation for subsequent 3D stratigraphic reconstruction.

[0068] Step S133: Input the color distribution features, texture distribution features, and extension direction features into the second attribute reconstruction subnetwork of the pre-trained geological feature extraction model. The second attribute reconstruction subnetwork matches the color distribution features and texture distribution features with the pre-set lithological standard feature library to determine the lithological type identifier corresponding to each lithological filling pattern element, and determines the stratigraphic sequence boundary position corresponding to the stratigraphic boundary element based on the extension direction features.

[0069] In this step, the color distribution feature vector and texture distribution feature vector of each lithological filling pattern element extracted in step S124, as well as the extension direction feature vector of each stratigraphic boundary element, can be input into the second attribute reconstruction sub-network of the geological feature extraction model. The core of this second attribute reconstruction sub-network is a classifier and a regressor, and it has a built-in lithological standard feature library. This lithological standard feature library is established by extracting features from a large number of standard legends of known lithologies. Each record in the library contains a standard lithology type (such as "medium-grained sandstone", "silty mudstone", "limestone", etc.) and its corresponding typical color distribution feature vector and texture distribution feature vector.

[0070] For each input lithology-filled pattern primitive's feature vector, the second attribute reconstruction subnetwork compares it with each record in the standard feature library, calculating the distance between feature vectors (such as Euclidean distance, Mahalanobis distance, etc.). The lithology type identifier corresponding to the standard lithology record with the smallest distance, such as "SS-M" (representing medium-grained sandstone) or "MS" (representing silty mudstone), is the lithology type identifier for that lithology-filled pattern primitive.

[0071] Simultaneously, for stratigraphic boundary primitives, the second attribute reconstruction subnetwork can analyze their extension direction feature vectors and, combined with their spatial location, infer their boundary position within the stratigraphic sequence. For example, if the extension direction of a stratigraphic boundary primitive is primarily near-horizontal, and there is a significant difference in lithological type identifiers on its upper and lower sides (e.g., one side is sandstone, and the other is mudstone), then the second attribute reconstruction subnetwork can identify it as an important stratigraphic sequence boundary. The model can generate an attribute for this stratigraphic boundary primitive describing its type as a sequence boundary, such as a "lithological abrupt change interface" or a "sedimentary discontinuity interface." This inference process relies on the knowledge of stratigraphic stacking patterns learned by the model during the training phase.

[0072] Step S134: Add the lithology type identifier and stratigraphic sequence boundary location as attribute fields to the lithology pattern primitive element set to generate a lithology pattern primitive element set carrying lithology attribute information and sequence information.

[0073] After determining the lithology type identifier and stratigraphic boundary attribute of each lithology filling pattern element in step S133, the above information is added as a new attribute field to the corresponding feature set.

[0074] For a set of lithological pattern primitives, for example, a "lithological code" field could be added to each element. This field would be of character type and would be used to store standardized codes such as "SS-F" (fine sandstone) or "SH" (shale). Additionally, a "lithological description" field could potentially be added to store more detailed information, such as "gravelly coarse sandstone".

[0075] For a set of stratigraphic boundary elements, an "interface type" field can be added to each element to identify whether it is a sequence boundary, lithological boundary, or other type of boundary. In addition, to establish the spatial relationship between lithological elements and stratigraphic boundaries, "overlying boundary identifier" and "underlying boundary identifier" fields can be added to the lithological elements to point to the stratigraphic boundary elements above and below them.

[0076] In this way, the lithological filling areas that were originally just visual patterns are now given clear geological meanings and established with stratigraphic boundaries, forming a set of lithological pattern primitive elements carrying lithological attribute information and sequence information.

[0077] Step S135: Input the gray value sequence and scale value sequence into the third attribute reconstruction sub-network of the pre-trained geological feature extraction model. The third attribute reconstruction sub-network identifies the grain size variation cycle based on the gray value sequence and determines the depth range corresponding to each grain size variation cycle based on the scale value sequence, generating the grain size type identifier and thickness value for each grain size variation cycle.

[0078] In this step, the grayscale value sequence and depth scale value sequence of the lithological columnar pattern obtained in step S126 can be input into the third attribute reconstruction subnetwork of the geological feature extraction model. This third attribute reconstruction subnetwork first uses the depth scale value sequence to fit a precise conversion relationship between depth and row coordinates, such as the linear function mentioned earlier. Then, using this conversion relationship, the row coordinates corresponding to each grayscale value in the grayscale value sequence are converted into depth values, thus obtaining a grayscale curve that varies with depth, i.e., the depth-grayscale value curve.

[0079] The core task of the third attribute reconstruction subnetwork is to automatically stratify and interpret the lithology of the depth-grayscale curve. This subnetwork can integrate, for example, a sequence analysis model, such as a segmented model based on a recurrent neural network or a one-dimensional convolutional neural network. This model can analyze the depth-grayscale curve and identify continuous segments with similar statistical characteristics. For example, the model can detect changes in the mean, variance, and other statistical quantities of grayscale values ​​along the depth direction. These changes correspond to interfaces of lithological changes. The model divides the curve into several continuous segments, each representing a grain size variation cycle. For each segment, the model can calculate the average grayscale value within that segment and, based on this average, refer to a pre-defined grayscale-grain size comparison table (which can be established, for example, by statistically analyzing well logging curves from known boreholes, where low grayscale values ​​correspond to coarse grain and high grayscale values ​​correspond to fine grain) to determine the grain size type identifier for that segment, such as "coarse grain," "medium grain," "fine grain," or "muddy." At the same time, the thickness of the lithological segment can be calculated by subtracting the depth value from the starting and ending depths of the segment (e.g., subtracting the starting depth from the ending depth).

[0080] Thus, for each identified granularity change cycle, a corresponding granularity type identifier and thickness value are generated.

[0081] Step S136: Add the granularity type identifier and thickness value as attribute fields to the column pattern element set to generate a column pattern element set carrying granularity attribute information and thickness attribute information.

[0082] After generating granularity type identifiers and thickness values ​​for each depth segment of the bar chart in step S135, this embodiment can structure this information into attribute fields and add them to the original set of bar pattern element features.

[0083] Specifically, the original columnar pattern primitives may contain only a single continuous sequence of grayscale values. This sequence needs to be segmented into multiple records. That is, each identified granularity variation period will serve as an independent sub-feature or an independent attribute record within the columnar pattern feature set. For example, a borehole columnar pattern can be segmented into several lithological segments. Each segment record may include, for example, the following: starting depth (Depth_start_i), ending depth (Depth_end_i), thickness value (Thickness_i, equal to Depth_end_i minus Depth_start_i), and granularity type identifier (Litho_i). This information can be stored in an association table or embedded as a multi-valued attribute within the columnar pattern features. In this way, the original grayscale image information is transformed into a structured set of geologically significant columnar pattern primitives, where each feature (or sub-feature) carries granularity and thickness attribute information.

[0084] Step S137: Merge the set of contour map elements carrying elevation attribute information, the set of lithological pattern map elements carrying lithological attribute information and sequence information, and the set of columnar pattern map elements carrying grain size attribute information and thickness attribute information to generate a set of map elements to be corrected carrying the original attribute information.

[0085] Finally, after generating the feature sets carrying their respective geological attribute information in steps S132, S134 and S136, these features from different map units, but which may describe the same geological body or the same spatial location, can be logically integrated.

[0086] Specifically, all elements can be centrally managed within a pre-defined, unified geospatial framework. For example, a closed contour map element might represent the boundary of a reservoir structure, while its interior might contain multiple lithological pattern elements extracted from geological profiles, describing the lithological distribution within the reservoir. Similarly, the location of a borehole columnar plot element might fall within the area covered by contour map elements. These elements are then merged to generate a new, comprehensive dataset—the set of map elements to be corrected.

[0087] In this system, the set of map elements to be corrected no longer distinguishes the source of the original map, but stores all map elements and their original attribute information (elevation, lithology, grain size, sequence stratigraphy, etc.) after reconstruction in a unified manner. This set of map elements to be corrected is the basis for all subsequent spatial correction and geometric restoration work, and it completely preserves all geological information extracted and inferred from the original map.

[0088] Step S140: Input the set of primitive elements to be corrected into the pre-trained spatial reference alignment model to perform coordinate correction and geometric repair processing, and generate a standardized set of primitive elements with a unified target spatial reference system.

[0089] This spatial reference alignment model is used to solve the problems of inconsistent spatial coordinates and geometric distortion caused by different sources, different projection methods, and scanning deformation of the original maps. Ultimately, it unifies all map elements under a preset target spatial reference system, such as a national geodetic coordinate system or an internationally used geographic coordinate system.

[0090] Step S141: Input the spatial coordinate sequence of each graphic element in the set of graphic elements to be corrected into the reference system identification sub-network of the pre-trained spatial reference alignment model. The reference system identification sub-network matches the spatial coordinate sequence with multiple preset coordinate reference system feature templates based on the distribution characteristics of the spatial coordinate sequence and the original attribute information carried by the set of graphic elements to be corrected, and determines the original coordinate reference system type corresponding to each graphic element.

[0091] In this step, the spatial coordinate sequences of all graphic elements in the set of graphic elements to be corrected, along with their original attribute information (such as map source descriptions, map sheet numbers, etc.), can be input into the reference system identification sub-network of the spatial reference alignment model. This reference system identification sub-network contains a pre-built coordinate reference system feature template library, which includes feature descriptions of various geographic coordinate systems and projected coordinate systems. Each template in this library not only includes the mathematical definition of the coordinate system (such as ellipsoid parameters, central meridian, standard latitude, etc.), but also the typical coordinate value distribution patterns and geometric deformation characteristics exhibited by geographic elements (such as contour lines, coastlines, and national boundaries) under that coordinate system.

[0092] The reference frame identification subnetwork first analyzes the statistical characteristics of the input coordinate sequence. For example, it can calculate the minimum bounding rectangle of all coordinate points to analyze whether their latitude and longitude ranges fall within a specific region; analyze the magnitude of the coordinate values, such as whether they are in metric or degree units; and analyze the curvature direction of contour lines to determine if they produce specific deformations under a particular projection. Then, these feature vectors extracted from the coordinate data are matched with the feature vectors of each template in a template library. This matching process can employ machine learning-based classification algorithms, such as training a random forest classifier that takes the features of the coordinate sequence as input and outputs the most probable coordinate system category.

[0093] In this way, the original coordinate reference system type corresponding to each graphic element can be identified.

[0094] Step S142: According to the original coordinate reference system type, call the corresponding coordinate transformation parameters from the preset coordinate transformation parameter library, and input the spatial coordinate sequence of all primitive elements in the set of primitive elements to be corrected into the coordinate projection transformation sub-network of the pre-trained spatial reference alignment model to perform coordinate projection transformation and generate a preliminary projection coordinate sequence set.

[0095] In this step, the corresponding coordinate transformation parameters can be retrieved from a preset coordinate transformation parameter library based on the original coordinate reference system type corresponding to each graphic element. This coordinate transformation parameter library stores the mathematical parameters required to transform from a common coordinate reference system to a preset target spatial reference system. For example, to transform coordinates in a first-type projected coordinate system to a target coordinate system, the required parameters may include translation parameters (such as DeltaX, DeltaY, DeltaZ), rotation parameters (such as Rx, Ry, Rz), and scale parameters.

[0096] Then, the original spatial coordinate sequences of all elements in the set of elements to be corrected can be input into the coordinate projection transformation sub-network of the spatial datum alignment model. This coordinate projection transformation sub-network is used to perform the corresponding mathematical transformation according to the called parameters. For transformations between geographic coordinate systems, geodetic datum conversion may be involved, such as the conversion from one geodetic coordinate system to another, which can be implemented using, for example, the Bursa model or the Mologinski model. For projection transformations, such as the conversion from geographic coordinates to projected coordinates, or from one projection to another, it can be implemented using the Gauss-Kruger projection forward or inverse calculation formulas. This coordinate projection transformation sub-network calculates the coordinates of each element point by point. Assuming the original coordinates are (longitude value Lon, latitude value Lat), after geodetic datum conversion and projection transformation, new coordinate values ​​(X_new, Y_new) are obtained. This operation is performed for all coordinate points of all elements, ultimately generating a preliminary projected coordinate sequence set. In this preliminary projected coordinate sequence set, all elements have been transformed to the target spatial reference system.

[0097] Step S143: Input the preliminary projected coordinate sequence set into the topology-preserving subnetwork of the pre-trained spatial reference alignment model, use the topology-preserving subnetwork to calculate the spatial distance change and azimuth change between adjacent primitives in the preliminary projected coordinate sequence set, and identify primitives whose spatial distance change exceeds a preset distance change threshold and whose azimuth change exceeds a preset angle change threshold as distorted primitives.

[0098] Although the preliminary projected coordinate sequence set obtained in step S142 has correct coordinate values, it may contain local geometric distortions due to the quality of the original map. This can cause unwanted gaps or overlaps between elements that should be adjacent, or lines that should be parallel to each other to become non-parallel. To identify these distortions, this embodiment can input the preliminary projected coordinate sequence set into the topology-preserving sub-network of the spatial reference alignment model. This topology-preserving sub-network is used to evaluate and identify geometric distortions. It can first construct a topological adjacency graph between elements based on their type and spatial relationships.

[0099] For example, for contour primitives, the spatial distance between adjacent contour lines and the normal direction at key points can be calculated. Specifically, the topology-preserving subnetwork can select a set of key point pairs, for example, finding a series of corresponding point pairs (Point_A_i and Point_B_i) on two adjacent contour lines through nearest-neighbor search. Then, the original distance D_original_i and the original azimuth (e.g., the direction angle of vector A->B) Az_original_i between these point pairs are calculated in the original set of primitive features to be corrected. Next, in the initial set of projected coordinate sequences, the distance D_proj_i and the azimuth Az_proj_i between identical point pairs are calculated again. Then, the change in spatial distance DeltaD_i for each point pair is calculated, which can be, for example, equal to the absolute value of D_proj_i minus D_original_i divided by D_original_i, expressed as a relative change; and the change in azimuth DeltaAz_i, which can be, for example, equal to the absolute value of Az_proj_i minus Az_original_i.

[0100] For each primitive feature, the DeltaD_i and DeltaAz_i of all keypoint pairs on it can be statistically analyzed, for example, by calculating their average or maximum value. If the statistical DeltaD_i of a primitive feature exceeds a preset distance change threshold, such as 5%, or its statistical DeltaAz_i exceeds a preset angle change threshold, such as 3 degrees, then the topology-preserving subnetwork can mark this primitive feature as a distorted primitive feature that may have undergone geometric distortion.

[0101] Step S144: Perform local adjustment processing on the preliminary projected coordinate sequence of distorted primitive features through the geometric correction sub-network of the pre-trained spatial reference alignment model, adjust the coordinate point position of the distorted primitive features to restore the original topological relationship between the distorted primitive features and the adjacent primitive features, and generate a preliminary corrected primitive feature set with correct topology.

[0102] In this step, the preliminary projected coordinate sequence of the distorted primitives and the topological relationship information between the distorted primitives and their surrounding undistorted features can be input into the geometric correction sub-network of the spatial reference alignment model. This geometric correction sub-network is used to fine-tune the locally distorted regions while maintaining the overall coordinate frame, in order to restore the correct geometry and topological relationships. This geometric correction sub-network can, for example, employ a deformation algorithm based on a physical model for fine-tuning, such as thin-plate spline interpolation or a deformation algorithm based on a linear elastic model.

[0103] Specifically, points on undistorted primitives can be used as control points, and a smooth deformation field can be constructed to "pull" points on distorted primitives from their current incorrect positions to their correct positions, while minimizing the energy of the entire region to ensure the continuity and smoothness of the deformation.

[0104] For example, for a contour line that has undergone local distortion, the geometric correction subnetwork can use points on the undistorted contour lines surrounding the distorted contour line as a reference, and calculate the correction amount (DeltaX_corr, DeltaY_corr) for each point on the distorted contour line using an interpolation function. Then, the correction amount can be added to the initial projected coordinates (X_proj, Y_proj) to obtain the corrected coordinates (X_corr, Y_corr).

[0105] After the aforementioned local adjustments, the distorted primitives can be pulled back to their proper spatial positions, thus restoring the correct distance and orientation relationships with adjacent primitives. This process is repeated for all identified distorted primitives, and then merged with undistorted primitives to obtain a preliminary set of topologically corrected primitives.

[0106] Step S145: Identify complex graphic elements in the preliminary correction set of graphic elements whose graphic symbols or annotation information have been deformed, and output the complex graphic elements as an intermediate exchange format file.

[0107] In this step, after the geometric correction in step S144, the geometric shape of most graphic elements has been restored. However, for some complex graphic elements, such as lithological filling patterns with specific graphic symbols (e.g., triangle symbols representing conglomerate) or graphic elements with complex annotation information (e.g., text containing superscripts, subscripts, or special fonts), simple adjustment of coordinate points may cause these symbols to be squeezed or stretched, resulting in distorted symbols and blurred text that cannot be identified.

[0108] Therefore, in this step, these complex graphic elements that are sensitive to deformation can be identified based on the attribute information of the graphic elements. For example, if a lithological pattern graphic element has the attribute of "pattern fill" and its internal texture is complex, or if a label graphic element has the attribute of "text" and its font is non-standard, it can be classified as a complex graphic element.

[0109] For all identified complex graphic elements, this embodiment can extract them from the initial corrected graphic element set and output them as intermediate exchange format files, such as Scalable Vector Graphics Format or Computer-Aided Design Exchange Format (such as DXF and other intermediate exchange formats). The intermediate exchange format file can losslessly preserve the vector definition of graphic symbols, the font, font size, style of text, and their precise relationship with geometric positions, providing a foundation for subsequent fine-tuning and conversion.

[0110] Step S146: After importing the intermediate exchange format file through the standard interface and converting it into a target platform compatible format, replace the corresponding complex graphic elements in the preliminary calibration graphic element set to generate a standardized graphic element set.

[0111] In this step, the intermediate exchange format file generated in step S145 can be imported into professional graphic editing software or conversion tools capable of handling complex graphic symbols via a standard interface. The conversion tool can then utilize its symbol processing and text rearrangement functions to repair deformed symbols and text. For example, a flattened triangle symbol can be readjusted into an equilateral triangle, and distorted text can be rearranged neatly. After repair, the files are exported from the tool's format to a format compatible with the target spatial database, such as the annotation feature class or cartographic symbol library supported by the First Geographic Information System platform.

[0112] Then, the repaired and transformed complex primitives can be replaced back in their original positions within the initially corrected primitive set. Specifically, based on the unique identifier of each complex primitive, the original, possibly deformed geometric shapes or annotations in the set can be replaced with correctly shaped versions exported from the tool. After this replacement, a standardized primitive set is generated. All primitives in this standardized primitive set have a unified target space reference frame, correct geometric topological relationships, and clearly readable graphic and textual information.

[0113] Step S150: Transmit the standardized map element set to the target spatial database to perform regularized data entry processing, and generate the processing metadata corresponding to the standardized map element set to obtain the thematic map dataset of carbon dioxide geological storage.

[0114] The target spatial database is a system specifically designed for storing and managing spatial data related to carbon dioxide geological sequestration; for example, it could be a geographic database using a first-level geographic information system platform. Data import processing must adhere to predefined rules to ensure that the data can be effectively organized, retrieved, and shared.

[0115] Step S151: Based on the original attribute information carried by each element in the standardized map element set, extract the geological map theme type identifier, basin name identifier, and reservoir type identifier corresponding to the standardized map element set.

[0116] Before writing the standardized set of map elements into the database, it is necessary to analyze its content and determine its affiliation within the database. This embodiment can extract key classification identifiers by parsing the attribute information of each map element. For example, the entire set can be traversed, and information about the "data source map" can be read from the attribute fields of each map element. If the attribute field "Original Map Name" of a contour map element contains the phrase "Reservoir Top Surface Structure Map", the geological map theme type identifier "Reservoir Top Surface Structure" can be extracted from it. If the attribute field "Breakhole Location Description" of a columnar map element contains the phrase "a basin", the basin name identifier can be extracted as "a basin". Similarly, if the "Lithotype Code" field of a lithological map element indicates that it is "sandstone", and its spatial location falls within a known reservoir area, its reservoir type identifier can be inferred to be "sandstone reservoir".

[0117] By aggregating and reasoning about the metadata of all elements, we can obtain the thematic type identifier of the geological map, the name identifier of the basin to which it belongs, and the reservoir type identifier corresponding to the entire dataset. For example, the dataset to be added to the database can be identified as "structural map of the top surface of a deep saline reservoir in a certain basin".

[0118] Step S152: Based on the geological map topic type identifier, the basin name identifier, the reservoir type identifier, and the predefined data entry rules, determine the target dataset, target feature class, and data hierarchical storage path in the target spatial database.

[0119] The data entry rules define which dataset (equivalent to a folder), feature class (equivalent to a table), and layering should be used for different types of data. For example, the rules could stipulate that all "reservoir top surface structure" thematic data belonging to "a certain basin" should be stored in a dataset named "AAA," and further subdivided into the reservoir top surface structure line feature class and fault line feature class according to feature type (point, line, polygon). The data layering storage path can be set according to actual needs to store the above data in different layers under different paths.

[0120] Step S153: Store the first element in the standardized map element set that conforms to the geological map theme type identifier into the target element class, and store the second element in the standardized map element set that conforms to the reservoir type identifier into the reservoir theme element class corresponding to the reservoir type identifier.

[0121] Step S154: Obtain the processing metadata of the standardized primitive feature set. The processing metadata includes source data file name information, processing timestamp information, operator identification information, coordinate transformation parameter path information used, and final data status information.

[0122] While data is being stored, key metadata for the entire processing process can be recorded for subsequent data traceability, quality assessment, and future updates. In this step, the processing program can automatically record logs at each key node of the entire process (e.g., pre-installed logging points at each key node to capture key metadata generated during processing), thus obtaining the processing process metadata.

[0123] For example, the metadata of the processing process may include source data file name information (i.e., the list of original map file names initially obtained), processing timestamp information (such as the start and end times of each major processing step), operator identification information (such as the system user or account identifier that executes the processing flow), coordinate transformation parameter path information used, and final data status information (such as a brief summary of the data entered into the database, such as the total number of features, spatial range, etc.).

[0124] Step S155: Associate and store the processing metadata in the metadata table of the target spatial database to generate a thematic map dataset of carbon dioxide geological sequestration.

[0125] In this step, the processing metadata obtained in step S154 can be associated with the geometric and attribute data actually entered into the database in step S153, and stored in a dedicated metadata table of the target spatial database. Specifically, for example, a new record can be created in the metadata table, which contains a unique dataset identifier associated with the feature class into which the data was written in step S153. Then, the various contents of the processing metadata are filled into the corresponding fields of this record. For example, the "Source Data Filename" field is filled with a list of files, the "Processing Time" field is filled with a timestamp, and the "Coordinate Transformation Parameters" field is filled with the path to the parameter file.

[0126] This approach enables integrated storage of both data and metadata. The entire process, from raw maps to a standardized, traceable database, is now complete, resulting in a high-quality, structured dataset of thematic maps on carbon dioxide geological storage. This dataset can be directly used for subsequent specialized applications such as storage potential assessment, reservoir modeling, and risk analysis.

[0127] Step S210: Input the reservoir top surface contour map elements from the standardized map element set into the pre-trained reservoir structure analysis model, and use the reservoir structure analysis model to perform spatial interpolation processing on the reservoir top surface contour map elements to generate a reservoir top surface elevation raster.

[0128] After generating a thematic map dataset of carbon dioxide geological storage, this embodiment further utilizes this dataset for reservoir feature analysis. Specifically, a standardized set of primitive elements can be extracted from the dataset, and the reservoir top surface contour map elements can be selected. These reservoir top surface contour map elements are then input into a pre-trained reservoir structure analysis model for processing. This reservoir structure analysis model is a geostatistical deep neural network model used to generate a continuous elevation surface from discrete contour data.

[0129] The reservoir structure analysis model can first read the coordinates and corresponding elevation attribute values ​​of each contour map element on the top surface of the reservoir. Since the points on the contour map elements are discretely distributed, the model can first perform point densification processing on each contour line, that is, interpolating new points between adjacent points according to a linear relationship, ensuring that the maximum distance between all points does not exceed a preset threshold, thus obtaining a dense discrete point cloud with elevation attribute values. Next, the model can first analyze the spatial autocorrelation of these discrete points, calculate the experimental semivariogram (i.e., analyze the relationship between the distance between point pairs and the elevation difference), and fit a theoretical semivariogram model. Then, for the center point of each raster cell to be interpolated within the study area, using the theoretical semivariogram model and known points within a certain range around it, the predicted elevation value of that point is calculated by solving the Kriging equations. The predicted elevation value is a weighted average of the elevation values ​​of the surrounding known points, with the weights determined by the semivariogram model to minimize the prediction variance.

[0130] This process is repeated for all grid cells to eventually generate a continuous reservoir top elevation grid covering the entire study area, where the value of each cell in the elevation prediction represents the reservoir top elevation at that location.

[0131] Step S220: Based on the local maxima and local minima of the reservoir top surface elevation grid, obtain the structural features of the reservoir top surface. The structural features of the reservoir top surface include the location of the structural high point corresponding to the local maxima and the location of the structural low point corresponding to the local minima.

[0132] After generating the reservoir top surface elevation grid, the terrain analysis module in the reservoir structural analysis model is used to extract key structural features.

[0133] Specifically, a sliding window analysis algorithm can be used to traverse the entire reservoir top surface elevation grid by defining a fixed-size neighborhood window, such as a three-by-three pixel neighborhood. For each central pixel in the grid, its elevation value is compared with the elevation values ​​of the other eight pixels in its neighborhood.

[0134] If the elevation value of the central pixel is strictly greater than the elevation values ​​of all eight neighboring pixels, then the central pixel is marked as a local maximum point, and its geographical location is recorded as a construction high point. Conversely, if the elevation value of the central pixel is strictly less than the elevation values ​​of all eight neighboring pixels, then it is marked as a local minimum point, and its geographical location is recorded as a construction low point.

[0135] Optionally, to prevent boundary effects, cells located on grid boundaries can also be ignored.

[0136] By traversing the entire grid, the model generates a set of discrete point features containing the locations of all structural highs and lows. The attribute tables for these point features also record their elevation values. These structural highs and lows reveal the undulating morphology of the reservoir top surface; structural highs typically correspond to the top of anticlines or the high parts of nose-like structures, indicating areas of carbon dioxide migration. The model outputs these point features as vector data, as part of the structural features of the reservoir top surface.

[0137] Step S230: Input the cover layer distribution surface primitives in the standardized primitive set into the pre-trained cover layer continuity evaluation model, and use the cover layer continuity evaluation model to calculate the ratio of the overlapping area to the non-overlapping area between adjacent cover layer distribution surface primitives to obtain the cover layer continuity quantification index.

[0138] In this step, all caprock distribution surface primitives are extracted from the standardized primitive set. These primitives exist in polygonal form, representing the distribution range of the caprock. The caprock distribution surface primitives are then input into a pre-trained caprock continuity evaluation model for processing. This model, based on the principle of spatial overlay analysis, is used to assess the lateral continuity of the caprock.

[0139] The model first performs a topological check on all input polygon features to ensure there are no unwanted overlaps or gaps. Then, the model iterates through each cap layer distribution polygon feature, using spatial indexing to find all other adjacent cap layer polygon features. For each pair of adjacent cap layer polygon features, the model calculates their spatial relationships.

[0140] Specifically, the model can calculate the area of ​​the overlapping region between the two polygons, as well as their individual areas. The cap layer continuity quantification index is defined as the area of ​​the overlapping region divided by the sum of the areas of the two polygons minus the area of ​​the overlapping region. The larger this ratio, the higher the degree of overlap between the two cap layer blocks, and the better the vertical continuity of the stacking. If the two polygons are adjacent (i.e., share a common edge), the model can also calculate the length of the common edge as an auxiliary evaluation index; the longer the common edge, the better the lateral continuity.

[0141] For each pair of adjacent surface features, the model calculates its corresponding continuity index, and finally outputs a quantification index of cap continuity containing multiple index values.

[0142] Step S240: Obtain the distribution characteristics of cap layer continuity based on the spatial distribution characteristics of the cap layer continuity quantification index.

[0143] The cap layer continuity quantification index is attached to each pair of adjacent cap layer blocks. To obtain the cap layer continuity distribution characteristics of the entire region, this step requires spatializing these paired indices to cover the entire study area.

[0144] Specifically, a regular grid of the study area can be constructed first. For each grid cell, all cover layer distribution surface primitives falling within that grid cell are identified. If multiple cover layer distribution surface primitives exist within a grid cell, the continuity quantification index between each pair of these primitives is calculated. Then, these indices are statistically analyzed, for example, by taking the average value, as the representative continuity value for that grid cell.

[0145] If a grid cell contains only one cap layer element, it indicates that the cell is located inside the cap layer block, and the continuity can be considered good. Therefore, the continuity index of this grid cell is assigned a maximum value. If a grid cell contains no cap layer elements, it indicates that the cell is located in a gap area of ​​the cap layer distribution, and the continuity index is assigned a minimum value.

[0146] By traversing all grid cells, a two-dimensional array of the same size as the grid is obtained, where each value represents the caprock continuity quantization index at that location. Then, interpolation methods can be used to resample this regular grid data into a continuous raster surface with the same resolution and range as the reservoir top elevation raster, i.e., the caprock continuity distribution characteristic raster. Each pixel value on this caprock continuity distribution characteristic raster represents the caprock continuity quantization index at that location, clearly demonstrating the spatial integrity variations of the caprock.

[0147] Step S250: Input the fault line element elements from the standardized set of element elements into the pre-trained fault closure analysis model. Use the fault closure analysis model to extract the strike azimuth and dip azimuth of the fault line element elements. Combine the lithological combination information of the strata on both sides of the fault line element elements to obtain the fault closure performance characteristics. The fault closure performance characteristics are related to the mudstone smearing coefficient of the fault surface.

[0148] In this step, all fault line primitives are extracted from the standardized primitive set and then input into a pre-trained fault closure analysis model for processing. This fault closure analysis model is an expert system integrating geometric and geomechanical analysis.

[0149] The fault closure analysis model first performs a geometrical analysis of each fault line to calculate its overall strike. By performing linear regression on the coordinates of each point on the fault line, the best-fit line for the fault is obtained, and the direction angle of this line is the fault strike azimuth. For the dip azimuth, since the fault line is two-dimensional, it cannot be directly obtained from two-dimensional line features; the model needs to combine regional geological stress field information or infer it from adjacent three-dimensional data. In this embodiment, the model can combine regional tectonic stress field direction data of the area, or infer the possible dip of the fault by spatially correlating it with adjacent profile features.

[0150] Based on this, the fault closure analysis model can perform spatial overlay analysis between fault line primitives and lithological pattern primitives carrying lithological attribute information generated in previous steps. Specifically, a buffer zone of a specified width can be established on both sides of the fault line first, then all stratigraphic primitives falling within the buffer zone can be identified, and the lithological type identifiers carried by these stratigraphic primitives can be read. Through this operation, the model can accurately grasp the lithological assemblage characteristics of the strata on both sides of the fault, especially the specific distribution location, number of layers, and spatial distribution morphology of mudstone layers.

[0151] Based on the aforementioned lithological assemblage information, the model further calculates the mudstone smearing coefficient of the fault surface. Specifically, the layer thickness data of each stratum can be obtained from the columnar pattern primitive element set; secondly, by comparing the spatial displacement of the same marker layer (e.g., a specific lithological interface or standard layer) on both sides of the fault, the vertical or horizontal displacement of the fault is estimated. Based on this, the model can simulate the process of mudstone layers being smeared and dragged across the fault surface during fault slip along the fault line. The specific value of the mudstone smearing coefficient can be determined by calculating the ratio of the cumulative thickness of the mudstone layers continuously distributed along the fault slip direction on the fault surface to the total fault displacement. The larger this ratio, the more continuous and thicker the mudstone smearing layer formed on the fault surface, and the better the lateral sealing performance of the fault.

[0152] Ultimately, the fault sealing performance analysis model generates corresponding fault sealing performance characteristic data for each fault line, or by dividing each fault into multiple independent evaluation segments based on differences in fault properties. These characteristic data are directly related to the calculated mudstone smearing coefficient of the fault surface and are attached to the fault line primitive elements in the form of attribute fields.

[0153] In this way, the model can quantitatively reflect the fault's ability to laterally impede the migration of carbon dioxide fluids, providing key fault sealing parameters for subsequent comprehensive evaluation of favorable storage areas.

[0154] Step S260: Perform spatial overlay analysis on the structural features of the reservoir top surface, the continuity distribution features of the caprock, and the fault sealing performance features to identify spatially connected regions that simultaneously meet the reservoir structural amplitude conditions, caprock continuity conditions, and fault sealing conditions.

[0155] After obtaining the key features of the reservoir, caprock, and fault, this embodiment performs a comprehensive spatial overlay analysis in a geographic information system environment to identify areas that simultaneously possess good storage conditions, good capping conditions, and good preservation conditions, i.e., potential favorable areas for carbon dioxide sequestration.

[0156] Specifically, the process begins by identifying all areas delineated by closed structural contour lines using a reservoir top elevation grid. These closed contour lines constitute potential structural traps. Next, these trap areas are overlaid with a caprock continuity distribution map to filter out traps located in areas with high caprock continuity indices, excluding traps located in areas with low caprock continuity or fracture zones. Finally, the remaining trap areas are overlaid with fault sealing performance characteristics, and each fault within each trap area is evaluated individually. If a trap's boundary is assessed as being cut by an open fault, or if a continuous open fault exists within the trap, the trap's lateral sealing conditions and overall integrity are compromised, resulting in poor preservation. Therefore, traps whose boundaries and all internal faults exhibit good sealing performance characteristics need to be selected.

[0157] After the above-mentioned step-by-step screening, the finally identified region is a spatially connected region that simultaneously meets the reservoir structural amplitude condition, caprock continuity condition, and fault sealing condition.

[0158] Step S270: Extract the boundary point coordinate sequence of the spatially connected region to obtain the storage space boundary coordinate sequence, and add the storage space boundary coordinate sequence to the carbon dioxide geological storage thematic map dataset.

[0159] For each identified spatially connected region, its geometry can be extracted. Taking a polygon as an example, all vertices on the boundary of the polygon can be extracted and arranged in order to obtain the boundary coordinate sequence of the storage space. Then, these boundary coordinate sequences are added as new surface features to the previously generated thematic map dataset of carbon dioxide geological storage.

[0160] Simultaneously, the attribute table of this new element can record various evaluation indicators of the storage space, such as its internal average caprock continuity index and comprehensive evaluation of fault sealing. Through this step, the results of the analysis and evaluation are also incorporated as part of the thematic information into a unified data management framework, making the dataset richer and more in-depth.

[0161] Step S310: Calculate the three-dimensional volume of the reservoir space based on the boundary coordinate sequence of the reservoir space, and multiply the three-dimensional volume of the reservoir space with the preset reservoir porosity parameters and carbon dioxide density parameters to obtain the carbon dioxide geological storage potential value.

[0162] After determining the boundary of the reservoir space, this embodiment can further quantify its storage potential. First, the three-dimensional volume of the reservoir space needs to be calculated. The top of the reservoir space is defined by the reservoir top surface elevation grid, and the bottom is usually a structural isosurface. The volume calculation uses a grid-based discrete integral method for approximation. Specifically, the reservoir top surface elevation grid within the area defined by the boundary polygon of the reservoir space can be traversed. For each grid cell falling inside the boundary polygon, it is considered as an independent micro-cylinder. The base area of ​​the cylinder is the planar area of ​​the grid cell, and the height of the cylinder is the difference between the reservoir top surface elevation value and the bottom reference surface elevation value at the cell location. By summing the volumes of all cylinders located within the boundary polygon, the three-dimensional volume of the entire reservoir space can be obtained. This process is essentially a discretized summation of an irregular geological body.

[0163] After obtaining the three-dimensional volume of the reservoir space, two key parameters corresponding to the reservoir are retrieved from the target space database: the average effective porosity parameter, representing the proportion of pore space in the rock that can be used for fluid storage; and the carbon dioxide density parameter under the formation temperature and pressure conditions corresponding to the reservoir depth. The three-dimensional volume of the reservoir space is then multiplied sequentially by the average effective porosity parameter and the carbon dioxide density parameter. The product is the theoretical geological carbon dioxide sequestration potential of the reservoir space. This potential value, expressed in mass, represents the total mass of carbon dioxide that the structural trap can theoretically sequester under the current reservoir parameters.

[0164] Step S320: Compare the carbon dioxide geological storage potential value with the preset potential level classification threshold to determine the storage potential level identifier corresponding to the storage space.

[0165] To more intuitively represent the magnitude of the potential for landfilling, this step involves classifying the calculated potential values ​​into different levels. Specifically, multiple threshold values ​​for potential levels can be pre-defined based on industry standards or project requirements, such as setting threshold ranges for low, low-medium, medium-high, and high levels. The calculated potential values ​​are then compared to these thresholds, and the corresponding landfill potential level identifier is determined based on the range in which they fall. This quantitative classification transforms complex numerical values ​​into easily understood and decision-making level labels.

[0166] Step S330: Associate the storage potential level identifier as an attribute field with the reservoir space map element in the carbon dioxide geological storage thematic map dataset to generate a carbon dioxide storage potential classification map dataset containing potential level labels.

[0167] Finally, the determined storage potential level identifier is added as a new attribute field to the attribute table of the storage space primitives. In this way, the isometric features representing favorable traps are no longer just geometric shapes, but carry key information on storage potential evaluation. After assigning this level identifier to all identified storage space primitives, the entire dataset evolves into a carbon dioxide storage potential grading map dataset containing potential level labels. This dataset can be directly used to generate thematic maps, with different levels rendered using different colors, thus visually displaying the spatial distribution of carbon dioxide storage potential in different areas within the study area.

[0168] Step S410: Input the standardized set of map elements into the pre-trained geological map quality assessment model, and use the geological map quality assessment model to extract the geometric integrity features and attribute integrity features of the map elements in the standardized set of map elements.

[0169] After generating a standardized set of primitive features with a unified target spatial reference frame, this step can further assess the quality of the set to ensure the reliability and usability of the data. Specifically, the entire standardized set of primitive features can be input into a pre-trained geological map quality assessment model. This geological map quality assessment model is a deep learning model built on graph neural networks and attention mechanisms, used to comprehensively evaluate the geometric and attribute quality of the primitive features.

[0170] For geometric integrity features, geological map quality assessment models can analyze whether the geometry of each map element conforms to geological laws. For example, for contour lines, the model checks whether they are smooth and whether there are any unwanted sharp corners or self-intersections; for areal features, the model checks whether their boundaries are closed and whether there are any narrow "gaps" or elongated "peninsulas." The model can extract topological relationship features between map elements through graph neural networks, such as whether adjacent features should be connected but have gaps, or whether features that should be separated have unwanted overlaps. These geometric features are encoded into a multidimensional geometric integrity feature vector. For attribute integrity features, the model can analyze whether the attribute fields of each map element are complete and whether the attribute values ​​are within a reasonable range. For example, for contour lines with elevation attributes, the model checks whether the elevation values ​​are within a reasonable regional elevation range and whether the elevation differences between adjacent contour lines are consistent; for areal features with lithology codes, the model can check whether the lithology codes are valid codes in the standard library and whether the lithological combinations of adjacent features conform to geological sedimentary laws. The above attribute features can be encoded into a multidimensional attribute integrity feature vector.

[0171] Step S420: Calculate the geometric error rate of the primitive elements based on the geometric integrity characteristics.

[0172] Based on the extracted geometric integrity feature vector, this step quantifies the geometric error rate of each map element. Specifically, the geological map quality assessment model can include a geometric error rate regression subnetwork. This subnetwork takes the geometric integrity feature vector as input, processes it through multiple fully connected layers and nonlinear activation functions, and finally outputs a scalar value as the geometric error rate. This geometric error rate is a value between zero and one, representing the probability or severity of geometric defects in the map element. For example, a self-intersecting contour line will have a geometric error rate close to one; while a smooth, reasonable contour line will have a geometric error rate close to zero. For different types of map elements, the model can use different weights and evaluation criteria to ensure that the calculation of the geometric error rate is targeted and accurate.

[0173] Step S430: Calculate the attribute missing rate of the graphic element based on the attribute integrity feature.

[0174] Simultaneously, based on the extracted attribute integrity feature vector, this step quantifies the attribute missing rate for each map element. The geological map quality assessment model may include an attribute missing rate regression subnetwork. This subnetwork takes the attribute integrity feature vector as input, processes it through multiple fully connected layers and nonlinear activation functions, and finally outputs a scalar value as the attribute missing rate. This attribute missing rate is a value between zero and one, representing the degree to which the attribute information of the map element is incomplete or the attribute values ​​are unreasonable. For example, a contour map element with an empty key attribute field (such as elevation value) will have a high attribute missing rate; while a map element with all attribute fields filled with reasonable values ​​will have a low attribute missing rate.

[0175] Step S440: Generate a comprehensive data quality score based on the geometric error rate and the attribute missing rate.

[0176] In this step, the geological map quality assessment model can use geometric error rate and attribute missing rate as inputs, which are then combined through a weighted fusion layer. This weighted fusion layer assigns preset weight coefficients to both the geometric error rate and attribute missing rate before weighting and fusing them. These preset weight coefficients include geometric quality weights and attribute quality weights. The overall data quality score is the sum of the geometric error rate multiplied by its geometric quality weight and the attribute missing rate multiplied by its attribute quality weight.

[0177] Optionally, to more intuitively represent the quality level, the comprehensive score can be defined as the sum of the weighted averages mentioned above, where a higher score indicates better quality. For example, the comprehensive data quality score can be equal to one minus the geometric error rate multiplied by the geometric quality weight, and then minus the attribute missing rate multiplied by the attribute quality weight. In this way, the comprehensive score also falls between zero and one, with the score closer to one indicating higher quality of the graphic elements.

[0178] Step S450: Identify the graphic elements whose comprehensive data quality score is less than or equal to the preset quality threshold as graphic elements to be corrected, and obtain a quality assessment report containing the graphic element identifier to be corrected and the error type code.

[0179] In this step, the overall score of each graphic element can be compared with a preset quality threshold. If the overall score of a graphic element is less than or equal to the quality threshold, it can be identified as a graphic element that needs to be corrected, meaning that its quality does not meet the standard and requires further inspection and correction.

[0180] Furthermore, based on the intermediate analysis results of geometric integrity features and attribute integrity features, the main error type of the primitive feature to be corrected can be determined. For example, error types can be coded as "geometric self-intersection," "missing attribute value," and "topological inconsistency." Finally, a structured quality assessment report can be generated, which includes a list of primitive features to be corrected. Each record in the list corresponds to a primitive feature to be corrected, and records the unique identifier of that primitive feature and its corresponding error type code.

[0181] Step S460: Associate the quality assessment report with the metadata information of the standardized graphic element set to generate a standardized graphic element set with quality traceability information.

[0182] In this step, the generated quality assessment report can be stored in association with the original standardized primitive feature set. Specifically, the quality assessment report can be added as a new metadata item to the metadata information of the standardized primitive feature set. This allows every user using the dataset or subsequent processing program to quickly understand the quality issues present in the dataset by consulting the metadata. This standardized primitive feature set, with accompanying quality traceability information, not only contains high-quality spatial geometry and attribute data but also a detailed description of the data's quality, providing crucial reference for data use, correction, and updates, significantly improving the data's reliability and usability.

[0183] The method provided in this application accurately identifies various primitive elements in the original geological map data using a pre-trained primitive recognition model; then, it completely reconstructs attribute information using a geological feature extraction model; next, it unifies the spatial reference system using a spatial benchmark alignment model; and finally, it stores the standardized primitive elements in the database and generates metadata, thereby achieving automated processing of carbon dioxide geological sequestration geological maps, reducing manual intervention, and thus improving the accuracy and efficiency of carbon dioxide geological sequestration geological map processing.

[0184] Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based geological map processing system 100 provided in an embodiment of this application. Figure 2 As shown, the processor 120 can be used in the artificial intelligence-based geological map processing system 100 and to perform the functions in this invention.

[0185] The AI-based geological map processing system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the AI-based geological map processing method of this invention. Although only one server is shown in this invention, for convenience, the functions described in this invention can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0186] For example, an AI-based geological map processing system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI-based geological map processing system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The AI-based geological map processing system 100 also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0187] For ease of explanation, only one processor is described in the AI-based geological map processing system 100. However, it should be noted that the AI-based geological map processing system 100 of this invention may also include multiple processors, and therefore the steps performed by one processor described in this invention may also be performed jointly or individually by multiple processors. For example, if the processor of the AI-based geological map processing system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0188] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0189] Finally, it should be noted that the above-disclosed embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A geological map processing method based on artificial intelligence, characterized in that, The method includes: Obtain a set of original geological map data for the target area of ​​carbon dioxide geological sequestration. The set of original geological map data includes contour map units, geological profile map units, and columnar map units stored in the format of a first geographic information system platform. The pre-trained primitive recognition model is invoked to perform primitive feature recognition processing on the original geological map data set, generating the contour primitive feature set corresponding to the contour map unit, the lithological pattern primitive feature set corresponding to the geological profile map unit, and the columnar pattern primitive feature set corresponding to the columnar map unit. The contour map element set, the lithological pattern map element set, and the columnar pattern map element set are input into a pre-trained geological feature extraction model to perform attribute information reconstruction processing, generating a map element set to be corrected carrying the original attribute information; The set of primitive elements to be corrected is input into a pre-trained spatial reference alignment model to perform coordinate correction and geometric repair processing, generating a standardized set of primitive elements with a unified target spatial reference system; The standardized set of map elements is transmitted to the target spatial database for regularized data entry processing, and the corresponding processing metadata is generated to obtain a thematic map dataset of carbon dioxide geological storage.

2. The geological map processing method based on artificial intelligence according to claim 1, characterized in that, The step of calling the pre-trained primitive recognition model to perform primitive feature recognition processing on the original geological map data set, generating a set of contour primitive features corresponding to the contour map unit, a set of lithological pattern primitive features corresponding to the geological profile map unit, and a set of columnar pattern primitive features corresponding to the columnar map unit, including: The contour map unit is input into the contour segmentation subnetwork of the pre-trained primitive recognition model. The contour segmentation subnetwork performs pixel-level classification on the contour map unit, and identifies the first pixel region corresponding to the closed contour primitive and the second pixel region corresponding to the open curve contour primitive. The boundary coordinate sequence of closed contour primitives is extracted from the first pixel region, and the path coordinate sequence of open contour primitives is extracted from the second pixel region to generate the contour primitive feature set. The geological profile map unit is input into the profile segmentation sub-network of the pre-trained primitive recognition model. The geological profile map unit is classified at the pixel level through the profile segmentation sub-network, and the third pixel region corresponding to the lithological filling pattern primitive and the fourth pixel region corresponding to the stratigraphic boundary primitive are identified. The color distribution features and texture distribution features of the lithological filling pattern primitives are extracted based on the third pixel region, and the extension direction features of the stratigraphic boundary primitives are extracted based on the fourth pixel region, thereby generating the lithological pattern primitive element set and the stratigraphic boundary primitive element set. The columnar graphic unit is input into the columnar segmentation sub-network of the pre-trained primitive recognition model. The columnar segmentation sub-network performs pixel-level classification on the columnar graphic unit, and identifies the fifth pixel region corresponding to the lithological columnar pattern primitive and the sixth pixel region corresponding to the depth scale line primitive. The grayscale value sequence of the lithological columnar pattern primitives is extracted based on the fifth pixel region, and the scale value sequence of the depth scale line primitives is extracted based on the sixth pixel region, thereby generating the set of columnar pattern primitive elements and the set of depth scale primitive elements.

3. The geological map processing method based on artificial intelligence according to claim 2, characterized in that, The step of inputting the contour map element set, the lithological pattern map element set, and the columnar pattern map element set into a pre-trained geological feature extraction model to perform attribute information reconstruction processing, generating a map element set to be corrected carrying the original attribute information, includes: The boundary coordinate sequence and the path coordinate sequence are input into the first attribute reconstruction sub-network of the pre-trained geological feature extraction model. The first attribute reconstruction sub-network calculates the first elevation attribute value corresponding to each closed contour map element based on the boundary coordinate sequence, and generates the second elevation attribute value corresponding to each open curve contour map element based on the path coordinate sequence and the preset legend elevation reference table. The first elevation attribute value and the second elevation attribute value are added as attribute fields to the contour map element set to generate a contour map element set carrying elevation attribute information; The color distribution features, texture distribution features, and extension direction features are input into the second attribute reconstruction subnetwork of the pre-trained geological feature extraction model. The second attribute reconstruction subnetwork matches the color distribution features and texture features with a pre-set lithology standard feature library to determine the lithology type identifier corresponding to each lithology filling pattern element, and determines the stratigraphic sequence boundary position corresponding to the stratigraphic boundary element based on the extension direction features. The lithology type identifier and the stratigraphic sequence boundary location are added as attribute fields to the lithology pattern primitive element set to generate a lithology pattern primitive element set carrying lithology attribute information and sequence information. The grayscale value sequence and the scale value sequence are input into the third attribute reconstruction subnetwork of the pre-trained geological feature extraction model. The third attribute reconstruction subnetwork identifies the grain size variation cycle based on the grayscale value sequence and determines the depth range corresponding to each grain size variation cycle based on the scale value sequence, generating a grain size type identifier and thickness value for each grain size variation cycle. The granularity type identifier and the thickness value are added as attribute fields to the columnar pattern element set to generate a columnar pattern element set carrying granularity attribute information and thickness attribute information. The set of contour map elements carrying elevation attribute information, the set of lithological pattern map elements carrying lithological attribute information and sequence information, and the set of columnar pattern map elements carrying grain size attribute information and thickness attribute information are merged to generate the set of map elements to be corrected carrying the original attribute information.

4. The geological map processing method based on artificial intelligence according to claim 3, characterized in that, The step of inputting the set of primitive features to be corrected into a pre-trained spatial reference alignment model to perform coordinate correction and geometric repair processing, generating a standardized set of primitive features with a unified target spatial reference system, includes: The spatial coordinate sequence of each graphic element in the set of graphic elements to be corrected is input into the reference system identification sub-network of the pre-trained spatial reference alignment model. The reference system identification sub-network matches the spatial coordinate sequence with multiple preset coordinate reference system feature templates based on the distribution characteristics of the spatial coordinate sequence and the original attribute information carried by the set of graphic elements to be corrected, thereby determining the original coordinate reference system type corresponding to each graphic element. According to the original coordinate reference system type, the corresponding coordinate transformation parameters are called from the preset coordinate transformation parameter library, and the spatial coordinate sequence of all primitive elements in the set of primitive elements to be corrected is input into the coordinate projection transformation sub-network of the pre-trained spatial reference alignment model to perform coordinate projection transformation and generate a preliminary set of projected coordinate sequences. The initial projection coordinate sequence set is input into the topology-preserving subnetwork of the pre-trained spatial reference alignment model. The topology-preserving subnetwork is used to calculate the spatial distance change and azimuth change between adjacent primitives in the initial projection coordinate sequence set. Primitives whose spatial distance change exceeds a preset distance change threshold and whose azimuth change exceeds a preset angle change threshold are identified as distorted primitives. The geometric correction subnetwork of the pre-trained spatial reference alignment model performs local adjustment processing on the preliminary projected coordinate sequence of the distorted primitive features, adjusts the coordinate point position of the distorted primitive features to restore the original topological relationship between the distorted primitive features and the adjacent primitive features, and generates a preliminary corrected primitive feature set with correct topology. Identify complex graphic elements in the preliminary correction set of graphic elements whose graphic symbols or annotation information have been deformed, and output the complex graphic elements as an intermediate exchange format file; After importing the intermediate exchange format file through the standard interface and converting it into a target platform-compatible format, the corresponding complex graphic elements in the preliminary calibration graphic element set are replaced to generate the standardized graphic element set.

5. The geological map processing method based on artificial intelligence according to claim 4, characterized in that, The process involves transmitting the standardized map element set to the target spatial database for rule-based data entry and generating corresponding processing metadata for the standardized map element set, resulting in a carbon dioxide geological storage thematic map dataset, including: Based on the original attribute information carried by each element in the standardized map element set, the geological map theme type identifier, the basin name identifier, and the reservoir type identifier corresponding to the standardized map element set are extracted. Based on the geological map topic type identifier, the basin name identifier, the reservoir type identifier, and the predefined data entry rules, the target dataset, target feature class, and data hierarchical storage path are determined in the target spatial database. Store the first primitive element in the standardized primitive element set that conforms to the geological map theme type identifier into the target element class, and store the second primitive element in the standardized primitive element set that conforms to the reservoir type identifier into the reservoir theme element class corresponding to the reservoir type identifier; Obtain the processing metadata of the standardized primitive element set. The processing metadata includes source data file name information, processing timestamp information, operator identification information, coordinate transformation parameter path information used, and final data status information. The metadata of the processing procedure is associated and stored in the metadata table of the target spatial database to generate the thematic map dataset of carbon dioxide geological sequestration.

6. The geological map processing method based on artificial intelligence according to claim 5, characterized in that, After transmitting the standardized set of graphic elements to the target spatial database for rule-based data entry processing, the process further includes: Read the standardized map element set that has been stored in the carbon dioxide geological storage thematic map dataset, and extract the spatial range information and attribute field information of the standardized map element set; Configure the spatial reference system and display range parameters of the map service according to the spatial range information, and configure the list of queryable fields and symbolic rendering rules of the map service according to the attribute field information; The standardized set of graphic elements is published as a network map service layer and a network element service layer by calling a service interface that conforms to a preset standard. Standardized map service resources are generated based on the access addresses of the network map service layer and the network element service layer.

7. The geological map processing method based on artificial intelligence according to claim 1, characterized in that, After inputting the set of primitive features to be corrected into the pre-trained spatial reference alignment model to perform coordinate correction and geometric repair processing, the process further includes: The reservoir top surface contour map elements in the standardized map element set are input into the pre-trained reservoir structure analysis model. The reservoir structure analysis model is then used to perform spatial interpolation on the reservoir top surface contour map elements to generate a reservoir top surface elevation raster. Based on the local maxima and local minima of the reservoir top surface elevation grid, the structural features of the reservoir top surface are obtained. The structural features of the reservoir top surface include the location of the structural high point corresponding to the local maxima and the location of the structural low point corresponding to the local minima. The cover layer distribution surface primitives in the standardized primitive set are input into a pre-trained cover layer continuity evaluation model. The cover layer continuity evaluation model is used to calculate the ratio of the overlapping area to the non-overlapping area between adjacent cover layer distribution surface primitives to obtain the cover layer continuity quantification index. The distribution characteristics of cap layer continuity are obtained based on the spatial distribution characteristics of the cap layer continuity quantification index; The standardized set of map elements is used to input the fault line map elements into a pre-trained fault closure analysis model. The strike azimuth and dip azimuth of the fault line map elements are extracted using the fault closure analysis model. The fault closure performance characteristics are obtained by combining the lithological combination information of the strata on both sides of the fault line map elements. The fault closure performance characteristics are related to the mudstone smearing coefficient of the fault surface. Spatial overlay analysis is performed on the structural features of the reservoir top surface, the continuity distribution features of the caprock, and the fault sealing performance features to identify spatially connected regions that simultaneously satisfy the reservoir structural amplitude condition, the caprock continuity condition, and the fault sealing condition. The boundary point coordinate sequence of the spatially connected region is extracted to obtain the storage space boundary coordinate sequence, and the storage space boundary coordinate sequence is added to the carbon dioxide geological storage thematic map dataset.

8. The geological map processing method based on artificial intelligence according to claim 1, characterized in that, After transmitting the standardized set of graphic elements to the target spatial database for rule-based data entry processing, the process further includes: The three-dimensional volume of the reservoir space is calculated based on the boundary coordinate sequence of the reservoir space, and the three-dimensional volume of the reservoir space is multiplied with the preset reservoir porosity parameters and carbon dioxide density parameters to obtain the carbon dioxide geological storage potential value. The carbon dioxide geological storage potential value is compared with a preset potential level classification threshold to determine the storage potential level identifier corresponding to the storage space; The storage potential level identifier is associated as an attribute field with the reservoir space map element in the carbon dioxide geological storage thematic map dataset to generate a carbon dioxide storage potential classification map dataset containing potential level labels.

9. The geological map processing method based on artificial intelligence according to claim 1, characterized in that, After generating the standardized primitive feature set with a unified target space reference system, the process further includes: The standardized set of map elements is input into a pre-trained geological map quality assessment model, and the geological map quality assessment model is used to extract the geometric integrity features and attribute integrity features of the map elements in the standardized set of map elements. The geometric error rate of the primitive element is calculated based on the geometric integrity characteristics. The attribute missing rate of the graphic element is calculated based on the attribute integrity feature. A comprehensive data quality score is generated based on the geometric error rate and the attribute missing rate. The graphic elements whose overall data quality score is less than or equal to the preset quality threshold are identified as graphic elements to be corrected, and a quality assessment report containing the graphic element identifier to be corrected and the error type code is obtained. The quality assessment report is associated and stored in the metadata information of the standardized graphic element set to generate a standardized graphic element set with quality traceability information.

10. A geological map processing system based on artificial intelligence, characterized in that, The method includes a processor and a computer-readable storage medium storing machine-executable instructions, which, when executed by a computer, implement the artificial intelligence-based geological map processing method of any one of claims 1-9.

Images