A lithology prediction method and system based on fusion of global and local features

Abstract

This application relates to a lithology prediction method and system based on the fusion of global and local features, belonging to the field of geological exploration technology. The method includes acquiring seismic frequency imaging data of the work area and performing feature extraction and enhancement; inputting the processed data in parallel into pre-trained global feature model and local detail model to obtain first and second lithology prediction results, respectively; dividing the predicted profile into thick stable sections, transition sections, and lithological interface transition sections based on the rate of change of the relative impedance curve in the frequency imaging data; assigning differentiated fusion weights to the prediction results of the two models for different geological sections and performing weighted fusion to obtain preliminary fused prediction results; finally, smoothing the fused results and optimizing them based on minimum layer thickness and stratigraphic sequence constraints to obtain the final lithological stratification profile. This method integrates the advantages of different models, improving the overall accuracy and geological rationality of lithology prediction, and has high engineering practical value.

Description

Technical Field

[0001] This application relates to the field of geological exploration technology, and more specifically, to a lithology prediction method and system based on the integration of global and local features. Background Technology

[0002] In the fields of oil and gas exploration and geotechnical engineering, accurate prediction of subsurface lithology distribution is a crucial foundation for reservoir evaluation, geological modeling, and engineering safety assessment. Traditional lithology prediction mainly relies on geologists' comprehensive interpretation of borehole cores, well logging curves, and seismic profiles. This method is heavily dependent on expert experience, inefficient, and exhibits significant subjectivity and inconsistency in interpretation results when faced with massive amounts of data or complex geological conditions.

[0003] With the development of machine learning technology, data-driven intelligent lithology prediction methods have gradually become a research hotspot. Existing technical solutions mainly fall into two categories: one is based on traditional statistical machine learning models. These methods typically extract global statistical features from seismic attribute or frequency imaging data for learning. They perform well in predicting thick, lithologically stable sections, but lack the ability to capture local details such as thin interbedded layers and abrupt lithological interfaces, often resulting in blurred or misjudged profiles at these interfaces. The other category is based on deep learning models. These models excel at extracting complex local patterns and nonlinear relationships from data and are sensitive to details. However, their learning process relies heavily on a large number of labeled samples, and they are prone to overfitting when training data is limited, leading to a lack of macroscopic stability in the prediction results and the potential appearance of isolated anomalies that do not conform to geological laws.

[0004] To leverage the strengths of different models, existing technologies have adopted the approach of integrating or fusing multiple prediction results. However, common strategies often involve simply averaging the outputs of multiple models or using a fixed set of weights for linear weighting. These static fusion strategies have significant drawbacks: they fail to fully consider the vertical heterogeneity of the subsurface medium—the data characteristics, prediction difficulties, and sources of reliability are fundamentally different between thick, stable sections and lithological interface transition sections. This "one-size-fits-all" fusion approach cannot fully utilize the stability advantages of the global model in thick sections, nor can it maximize the detail resolution capabilities of local models in interface sections, thus failing to fundamentally reconcile the inherent contradiction between "macroscopic stability" and "microscopic accuracy." Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a lithology prediction method and system based on the integration of global and local features.

[0006] The technical solution of this invention is as follows: This invention proposes a lithology prediction method based on the fusion of global and local features, comprising the following steps: Acquire seismic frequency imaging data of the work area, and perform feature extraction and enhancement processing on the frequency imaging data; The processed frequency imaging data is input in parallel into the pre-trained global feature model and local detail model to obtain the first lithology prediction result and the second lithology prediction result, respectively. Based on the local structural features of frequency imaging data, the predicted profile is divided into different geological sections. For different geological sections, different fusion weights are assigned to the first lithology prediction result and the second lithology prediction result, and weighted fusion is performed to obtain the preliminary fusion prediction result. The preliminary fusion prediction results were smoothed and optimized based on minimum layer thickness constraints and stratigraphic sequence constraints to obtain the final lithological stratification profile.

[0007] Preferably, the global feature model is constructed based on the gradient boosting decision tree algorithm to learn the global patterns of lithology distribution; the local detail model is constructed based on the attention mechanism neural network algorithm to capture the local detail features of lithology changes.

[0008] Preferably, the feature extraction and enhancement process includes: for each predicted point in the frequency imaging data, extracting data points in its preset depth neighborhood to form an initial feature vector; and calculating one or more derived features among its vertical gradient, horizontal gradient, and local variance based on the initial feature vector; and concatenating the initial features and the derived features to form an enhanced feature vector.

[0009] Preferably, the division of different geological sections based on local structural features of frequency imaging data includes: dividing the predicted profile into different geological sections based on the rate of change of the relative impedance curve in the frequency imaging data; dividing areas with a rate of change below a first threshold into thick stable sections, areas with a rate of change above a second threshold into lithological interface transition sections, and areas with a rate of change between the first threshold and the second threshold into transition sections; wherein the first threshold is less than the second threshold.

[0010] Preferably, in the thick stable section, the fusion weight assigned to the first lithology prediction result is greater than the fusion weight assigned to the second lithology prediction result; in the lithology interface transition section, the fusion weight assigned to the second lithology prediction result is greater than the fusion weight assigned to the first lithology prediction result.

[0011] Preferably, the optimization based on the minimum layer thickness constraint involves merging predicted thin layers with a thickness less than a preset threshold.

[0012] Preferably, the stratigraphic sequence constraint is to verify and correct the optimized lithological stratification profile according to a preset stratigraphic sedimentary sequence rule.

[0013] On the other hand, the present invention also provides a lithological prediction system based on the fusion of global and local features, comprising: The data preprocessing module acquires seismic frequency imaging data of the work area and performs feature extraction and enhancement processing on the frequency imaging data. The model prediction module inputs the processed frequency imaging data in parallel into the pre-trained global feature model and local detail model to obtain the first lithology prediction result and the second lithology prediction result, respectively. The dynamic fusion module divides the predicted profile into different geological sections based on the local structural features of the frequency imaging data. For different geological sections, different fusion weights are assigned to the first lithology prediction result and the second lithology prediction result, and weighted fusion is performed to obtain the preliminary fusion prediction result. The post-processing optimization module smooths the preliminary fusion prediction results and optimizes them based on minimum layer thickness constraints and stratigraphic sequence constraints to obtain the final lithological stratification profile.

[0014] In another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a lithology prediction method based on the fusion of global and local features as described in any embodiment of the present invention.

[0015] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a lithology prediction method based on the fusion of global and local features as described in any embodiment of the present invention.

[0016] The present invention has the following beneficial effects: 1. By constructing a dual predictor framework of "gradient boosting tree (global model) + attention network (local model)" with complementary mechanisms, and combining it with the local structural features of the data itself for dynamic weighted fusion, the goal of relying on global laws to maintain stability in thick layers and relying on local details to improve accuracy in interface layers is achieved. This overcomes the limitations of a single model from a mechanistic perspective, making the final prediction profile better in both macroscopic and detailed aspects.

[0017] 2. Feature enhancement through neighborhood feature extraction and derived feature calculation effectively taps into the information potential of limited data. Simultaneously, the dual-model independent parallel training architecture reduces model complexity and dependence on massive amounts of labeled data, making it more suitable for real-world engineering scenarios where borehole data is scarce.

[0018] 3. Two post-processing steps were introduced: merging based on minimum layer thickness and verification and correction based on stratigraphic sequence rules. This combined the "data-driven" results predicted by the model with the "knowledge-driven" experience of geologists and the requirements of engineering interpretability, automatically outputting continuous, smooth, and sequence-reasonable lithological profiles, which greatly reduced the workload of manual interpretation. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a diagram showing the frequency imaging results; Figure 3 This is a lithological prediction stratigraphic profile. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0021] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0022] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0023] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0024] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0025] Example 1: To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present application and with reference to the accompanying drawings.

[0026] To address the problems of existing technologies, this invention provides a lithology prediction method based on the fusion of global and local features, comprising the following steps: Acquire seismic frequency imaging data of the work area, and perform feature extraction and enhancement processing on the frequency imaging data; In this embodiment, the steps for feature extraction and enhancement processing of frequency imaging data, i.e., relative impedance data, include: For each depth point to be predicted in the imaging data, M sampling points (M=5 in this embodiment) are selected above and below it along the depth direction, forming a local window. The relative impedance values ​​of all data points within this window are used as the initial feature vector for that point. To further enhance the representational ability of the features, the first-order vertical gradient, horizontal gradient, and local variance within the window are calculated based on this initial feature vector. These calculated derived features are then concatenated with the initial feature vector to form the final enhanced feature vector.

[0027] The processed frequency imaging data is input in parallel into the pre-trained global feature model and local detail model to obtain the first lithology prediction result and the second lithology prediction result, respectively. In this embodiment, the obtained enhanced feature vectors are input in parallel into two pre-trained machine learning models: Global Feature Model: This embodiment uses the LightGBM model built based on the gradient boosting decision tree algorithm. This model aims to learn the macroscopic statistical regularities and global trends of lithology distribution from the data of the entire work area, and output the first lithology prediction result, specifically the probability vector of each depth point belonging to each type of lithology.

[0028] Local detail model: This embodiment uses TabNet, a neural network based on an attention mechanism. Through its built-in attention mechanism, this model can focus on key local information of the input features, excelling at capturing subtle changes in lithology such as abrupt changes and thin interbedded layers in the vertical direction, and outputting a second lithology prediction result.

[0029] The two models use the same sample set during the training phase but are trained independently, and operate independently during the prediction phase without interfering with each other; the sample set consists of borehole data and its corresponding seismic frequency imaging data.

[0030] Based on the local structural features of frequency imaging data, the predicted profile is divided into different geological sections. For different geological sections, different fusion weights are assigned to the first lithology prediction result and the second lithology prediction result, and weighted fusion is performed to obtain the preliminary fusion prediction result. The local structural characteristics of the frequency imaging data and relative impedance curves are calculated. Specifically, a window is slid along the depth direction (in this embodiment, the length is 2 meters), and the standard deviation of the relative impedance value within each window is calculated as the rate of change at the center point of the window. This value quantifies the degree of local fluctuation in the formation properties.

[0031] Regions with a rate of change below a first threshold are classified as thick, stable sections; regions with a rate of change above a second threshold are classified as lithological interface transition sections; and regions with a rate of change between the first and second thresholds are classified as transition zones. The first threshold is less than the second threshold. In this embodiment, the first threshold is set to 0.2, and the second threshold is set to 0.6. For transition zones with a rate of change between the first and second thresholds, interpolation is performed using a linear ratio.

[0032] Differential weighting is performed based on the divided geological sections, specifically as follows: In the thick, stable section, global patterns are considered to dominate. A higher fusion weight is assigned to the first lithology prediction result output by the global feature model LightGBM (0.6-0.8 in this embodiment), while a lower fusion weight is assigned to the second lithology prediction result output by the local detail model TabNet (0.2-0.4 in this embodiment).

[0033] In the lithological interface transition section, local details are considered more important. A higher fusion weight is assigned to the second lithological prediction result output by the local detail model TabNet (0.6-0.8 in this embodiment), while a lower fusion weight is assigned to the first lithological prediction result output by the global feature model LightGBM (0.2-0.4 in this embodiment).

[0034] For each depth point, the probability vectors output by the two models are weighted and summed according to the weight of the segment to which it belongs, to obtain the preliminary fusion prediction result.

[0035] The preliminary fusion prediction results were smoothed and optimized based on minimum layer thickness constraints and stratigraphic sequence constraints to obtain the final lithological stratification profile.

[0036] In this embodiment, the preliminary fusion prediction results are processed into rules to make them more consistent with geological understanding and engineering needs: Smoothing: Median filtering or mean filtering is used to smooth the fused lithological probability profile in order to suppress random noise.

[0037] Minimum layer thickness constraint: A minimum layer thickness is set (0.5 meters in this embodiment). The predicted lithological profile is scanned, and thin layers with a thickness less than this threshold are merged with their adjacent layers predicted to have the same lithology. This step eliminates ultrathin layers that have no geological or engineering significance.

[0038] Stratigraphic sequence constraints: A rule base for stratigraphic sedimentary sequences is established based on fundamental principles of stratigraphy and sedimentology (such as a priori rules like "sandstone does not directly overlie mudstone"). The profiles after thickness constraints are validated, and if sequences violating the rules are found, they are corrected according to the rules. For example, isolated thin layers that violate the rules are adjusted to reasonable lithology based on the lithological probability of their context.

[0039] After the above steps, a geologically sound, smooth and continuous final lithological stratification profile is finally obtained.

[0040] like Figure 2 As shown, to facilitate the demonstration of prediction results, a typical survey line is used as an example. This survey line has 30 measuring points, with a point spacing of 2m and a line length of approximately 60m. One borehole is located at each end of the survey line. The data of the survey line to be predicted is simultaneously input into two models, resulting in two sets of lithological classification probability results. The thick stable section and the lithological interface transition section are identified based on the rate of change of the relative impedance curve: areas with a rate of change below 0.2 are identified as thick stable sections, and areas with a rate of change above 0.6 are identified as lithological interface transition sections. In the thick stable section, the global model weight is set to 0.75 and the local model weight is set to 0.25; in the lithological interface transition section, the global model weight is set to 0.25 and the local model weight is set to 0.75; the transition zone is interpolated using a linear ratio. After fusion, the overall classification accuracy of the survey line improved to 93.5%, an improvement of 6.3 percentage points compared to LightGBM alone (87.2%), and an improvement of 8.9 percentage points compared to TabNet alone (84.6%). The accuracy within thick layers reached 95.2%, and the interface positioning error decreased to 0.35m. Median filtering (window of 5 depth points) was applied to the fused results to eliminate isolated noise points. A minimum layer thickness of 0.5m was set, and thin layers with insufficient thickness were merged into adjacent dominant strata. Inverse sequence anomalies were corrected based on stratigraphic sequence relationships, ultimately forming a continuous and regular lithological stratigraphic profile, such as... Figure 3 As shown.

[0041] Example 2: This embodiment provides a lithology prediction system based on the fusion of global and local features, including: The data preprocessing module acquires seismic frequency imaging data of the work area and performs feature extraction and enhancement processing on the frequency imaging data. The model prediction module inputs the processed frequency imaging data in parallel into the pre-trained global feature model and local detail model to obtain the first lithology prediction result and the second lithology prediction result, respectively. The dynamic fusion module divides the predicted profile into different geological sections based on the local structural features of the frequency imaging data. For different geological sections, different fusion weights are assigned to the first lithology prediction result and the second lithology prediction result, and weighted fusion is performed to obtain the preliminary fusion prediction result. The post-processing optimization module smooths the preliminary fusion prediction results and optimizes them based on minimum layer thickness constraints and stratigraphic sequence constraints to obtain the final lithological stratification profile.

[0042] Example 3: This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a lithology prediction method based on the fusion of global and local features as described in any embodiment of the present invention.

[0043] Example 4: This embodiment proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements a lithology prediction method based on the fusion of global and local features as described in any embodiment of the present invention.

[0044] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0045] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0046] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0047] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0048] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A lithological prediction method based on the fusion of global and local features, characterized in that, Includes the following steps: Acquire seismic frequency imaging data of the work area, and perform feature extraction and enhancement processing on the frequency imaging data; The processed frequency imaging data is input in parallel into the pre-trained global feature model and local detail model to obtain the first lithology prediction result and the second lithology prediction result, respectively. Based on the local structural features of frequency imaging data, the predicted profile is divided into different geological sections. For different geological sections, different fusion weights are assigned to the first lithology prediction result and the second lithology prediction result, and weighted fusion is performed to obtain the preliminary fusion prediction result. The preliminary fusion prediction results were smoothed and optimized based on minimum layer thickness constraints and stratigraphic sequence constraints to obtain the final lithological stratification profile.

2. The lithological prediction method based on the fusion of global and local features according to claim 1, characterized in that: The global feature model is constructed based on the gradient boosting decision tree algorithm and is used to learn the global patterns of lithology distribution; the local detail model is constructed based on the attention mechanism neural network algorithm and is used to capture the local detail features of lithology changes.

3. The lithological prediction method based on the fusion of global and local features according to claim 1, characterized in that: The feature extraction and enhancement process includes: for each predicted point in the frequency imaging data, extracting data points in its preset depth neighborhood to form an initial feature vector; and calculating one or more derived features among its vertical gradient, horizontal gradient, and local variance based on the initial feature vector; and concatenating the initial features and the derived features to form an enhanced feature vector.

4. The lithological prediction method based on the fusion of global and local features according to claim 1, characterized in that: The method of dividing different geological sections based on local structural features of frequency imaging data includes: dividing the predicted profile into different geological sections based on the rate of change of the relative impedance curve in the frequency imaging data; dividing the area with a rate of change below a first threshold into a thick stable section, dividing the area with a rate of change above a second threshold into a lithological interface transition section, and dividing the area with a rate of change between the first threshold and the second threshold into a transition section; wherein the first threshold is less than the second threshold.

5. The lithological prediction method based on the fusion of global and local features according to claim 4, characterized in that: In the thick, stable section, the fusion weight assigned to the first lithological prediction result is greater than the fusion weight assigned to the second lithological prediction result; in the lithological interface transition section, the fusion weight assigned to the second lithological prediction result is greater than the fusion weight assigned to the first lithological prediction result.

6. The lithological prediction method based on the fusion of global and local features according to claim 1, characterized in that: The optimization based on the minimum layer thickness constraint is to merge predicted thin layers with a thickness less than a preset threshold.

7. The lithological prediction method based on the fusion of global and local features according to claim 1, characterized in that: The stratigraphic sequence constraint is to verify and correct the optimized lithological stratification profile according to the preset stratigraphic sedimentary sequence rules.

8. A lithological prediction system based on the integration of global and local features, characterized in that, include: The data preprocessing module acquires seismic frequency imaging data of the work area and performs feature extraction and enhancement processing on the frequency imaging data. The model prediction module inputs the processed frequency imaging data in parallel into the pre-trained global feature model and local detail model to obtain the first lithology prediction result and the second lithology prediction result, respectively. The dynamic fusion module divides the predicted profile into different geological sections based on the local structural features of the frequency imaging data. For different geological sections, different fusion weights are assigned to the first lithology prediction result and the second lithology prediction result, and weighted fusion is performed to obtain the preliminary fusion prediction result. The post-processing optimization module smooths the preliminary fusion prediction results and optimizes them based on minimum layer thickness constraints and stratigraphic sequence constraints to obtain the final lithological stratification profile.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a lithology prediction method based on the fusion of global and local features as described in claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a lithology prediction method based on the fusion of global and local features as described in claims 1-7.

Images