Well logging analysis model construction method and system based on multistage regional feature association
By constructing a multi-level regional feature association model and utilizing graph neural networks and deep learning technology, the common geological features of the region are automatically extracted, which solves the problems of insufficient accuracy and poor generalization in traditional well logging analysis, and achieves more efficient reservoir parameter prediction and oil and gas exploration decision optimization.
Patent Information
- Application Number
- CN202511320298.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-16
- Publication Date
- 2026-01-23
AI Technical Summary
Traditional well logging analysis methods rely on data from a single well location, ignoring regional characteristics, resulting in insufficient accuracy and poor generalization, and high costs associated with manual annotation.
By constructing a multi-level regional feature association model, and utilizing graph neural networks and deep learning techniques, the commonalities of regional geology are automatically extracted, well logging data is integrated, and an oil well reservoir parameter prediction network is constructed to achieve automated analysis.
It improves the accuracy and generalization ability of well logging analysis, reduces the reliance on manual annotation, and enhances the accuracy of reservoir parameter prediction and the optimization effect of oil and gas exploration decision-making.
Smart Images
Figure CN121392166A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of oil and gas exploration artificial intelligence, and particularly relates to a well logging analysis model construction method and system based on multi-level regional feature association. BACKGROUND
[0002] Traditional oil well analysis (such as well logging interpretation) methods have many problems, such as regional feature missing problems, only relying on single well site data, not mining geological commonalities (such as sedimentary facies, tectonic units, etc.) of multiple wells in the same region, which will lead to difficulty in capturing cross-regional rules; secondly, strong artificial dependence, standard well calibration needs expert annotation of geological parameters, and the annotation cost is high; and there is a serious feature loss problem, traditional regional calibration methods (such as trend surface analysis) will cause loss of local detail features due to forced normalization of non-standard well data.
[0003] Therefore, it is urgent to solve the technical problems of insufficient precision and poor generalization caused by using standard wells relying on manual annotation and ignoring regional features in traditional methods. SUMMARY
[0004] The purpose of the present application is to provide a well logging analysis model construction method and system based on multi-level regional feature association, which automatically extracts regional geological commonalities and dynamically fuses well logging data, to solve the technical problems of insufficient precision and poor generalization caused by using standard wells relying on manual annotation and ignoring regional features in traditional methods.
[0005] The technical scheme adopted by the present application is as follows: The first aspect of the present application is a well logging analysis model construction method based on multi-level regional feature association, comprising the following steps: S1, analyzing the geographical hierarchical identifier in the well site code to construct a multi-level regional embedding vector; S2, calculating the spatial distance of adjacent regions according to the well site coordinates, generating a regional spatial relationship graph taking regions as nodes and spatial distances as edge weights; and performing feature aggregation on the relationship graph through a graph neural network to obtain neighborhood-enhanced regional features; S3, constructing an oil well reservoir parameter prediction network, the prediction network comprising a feature extraction backbone network and a task network; S4, collecting well logging data of a target oil well, inputting the well logging data into the backbone network after standardization to extract features, and then performing gated weight fusion with the multi-level regional embedding vector constructed in step S1; S5, inputting the fusion features of step S4 into the oil well reservoir parameter prediction network constructed in step S3 to output reservoir parameter prediction results; setting an MSE loss function and performing training, and after training is completed, deploying an oil and gas task to complete well logging analysis.
[0006] Further, the multi-level regional embedding vector in S1 is a three-level geographical hierarchy embedding, wherein: Hierarchy definition: First level: sedimentary facies type, including marine facies, continental facies and transitional facies; Second level: geological tectonic unit, including intra-craton depression, foreland basin, rift basin, slope belt, fault zone and depression; Third level: the number of development blocks, which is determined according to the actual development blocks of the oilfield; After determining the hierarchy and the number of each hierarchy, for each type of each hierarchy, a learnable embedding vector is generated; then the first hierarchy has three embedding vectors C1, C2, C3, the second hierarchy has six embedding vectors D1, D2, D3, D4, D5, D6, and the third hierarchy has k embedding vectors B1, B2, …, Bk according to the number of blocks k.
[0007] For a specific oil well, according to the three-level regional information of this oil well, the corresponding three embedding vectors are obtained: , wherein is the embedding vector corresponding to the sedimentary facies type of this oil well, that is, the corresponding one of C1, C2, C3; is the embedding vector corresponding to the geological tectonic unit of this oil well, is the embedding vector corresponding to the development block of this oil well.
[0008] Then the embedding vectors of different hierarchies are spliced together to form a complete regional embedding; That is: ; , wherein is the embedding vector of the corresponding three hierarchies, and concat is the vector splicing function.
[0009] Further, the S2 aggregates features of the relationship graph through a graph neural network, specifically: regarding the oil well as a node in the graph, calculating the distance between each well according to the coordinate information of the well, and constructing the adjacent node relationship of the graph; when the distance meets the set aggregation radius condition, the edge connection is established, and the regional space relationship graph is formed; on the basis of constructing the regional space relationship graph, the graph neural network architecture is built for aggregating the adjacent regional features; and the graph neural network algorithm is used, and the aggregation formula followed by the node feature update in the network is: ; , wherein is the feature of the lth layer regional node i, and the neighbor node set N(i) satisfies the spatial distance <5km; is the ReLU activation function; the mean function is the mean value function; and the concat function is the vector splicing function.
[0010] Further, S3 specifically involves: constructing an oil well reservoir parameter prediction network, the prediction network comprising: (a) The feature extraction backbone network is used to extract deep features from the input preprocessed logging data, and may include a multi-level network architecture based on convolutional neural networks or self-attention mechanisms. (b) The task network is used to calculate and predict the values of reservoir parameters based on logging characteristics and regional embedded fusion characteristics. The prediction layer may include at least one fully connected layer followed by a ReLU activation function, and finally output the predicted values of porosity, permeability and oil saturation.
[0011] Further, S4 specifically involves: inputting the logging data obtained from a comprehensive well inspection using logging equipment into the backbone network constructed in S3; then performing gated weight fusion between the extracted logging data features and the multi-level region embedding vector constructed in S2. The gated weight fusion is achieved by calculating the similarity weight between the logging features and the region embedding, with the specific formula as follows: ; ; in For the output characteristics of the well logging data backbone network, The current well's multi-level region embedding fusion vector is Trainable parameter matrix.
[0012] Furthermore, the logging data in step S4 includes natural gamma, resistivity, and sonic transit time.
[0013] Specifically, the MSE loss function expression in step S5 is as follows: ; Where L is the loss value and N is the number of training samples. For the sample target true value, Predict the target value for the model.
[0014] The second aspect of the present invention: a well logging analysis model construction system based on multi-level regional feature correlation, comprising the following modules: Multi-level region nested embedding vector construction module: Parses the geographic hierarchy identifier in well location codes and constructs multi-level region nested embedding vectors; Feature aggregation module: Calculates the spatial distance between adjacent areas based on well location coordinates, generates a regional spatial relationship graph with areas as nodes and spatial distance as edge weights; performs feature aggregation on this relationship graph through a graph neural network to obtain neighborhood-enhanced regional features; A prediction network construction module: a reservoir parameter prediction network of an oil well is constructed, and the prediction network comprises a feature extraction backbone network and a task network; A gated weight fusion module: logging data of a target oil well is collected, and after the logging data is standardized, the logging data is input into the backbone network to extract features, and then the features are fused with the multi-level regional embedding vectors constructed by the multi-level regional embedding vector construction module through gated weight fusion; A logging analysis module: the fused features of the gated weight fusion module are input into the reservoir parameter prediction network of the oil well constructed by the prediction network construction module, and a reservoir parameter prediction result is output; an MSE loss function is set and trained, and after the training is completed, oil and gas task deployment is performed, and logging analysis is completed.
[0015] The third aspect of the present application: an electronic device comprising a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to realize the logging analysis model construction method based on multi-level regional feature association.
[0016] The fourth aspect of the present application: a computer readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to realize the logging analysis model construction method based on multi-level regional feature association.
[0017] The beneficial effects of the present application are as follows: The present application solves the technical problems of insufficient precision and poor generalization caused by the use of standard wells relying on manual annotation and the neglect of regional features in the conventional method. It is suitable for reservoir parameter prediction, oil reservoir characterization and exploration decision optimization scenarios. BRIEF DESCRIPTION OF DRAWINGS
[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0019] Figure 1 The flowchart of the present application is shown in the figure; Figure 2 The model building architecture diagram of the present application is shown in the figure. DETAILED DESCRIPTION
[0020] With reference to the drawings and embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.
[0021] As shown in Figure 1 and Figure 2 , the present application provides a well analysis model construction method based on multi-level regional embedding and spatial neighborhood modeling, including the following processes: Step S1, analyzing the geographic level identifier in the well site code, and constructing a multi-level regional nested embedding vector Firstly, define the sedimentary type as three types of marine, terrestrial and transitional facies, define the geological structure as six types of intra-craton depression, foreland basin, rift basin, slope belt, fault zone and depression, define the number of development blocks, and determine the number according to the actual oilfield development block arrangement, which is defined as 128 blocks in this example. The dimension of each level embedding is defined as 512, and the dimensions of the three-level regional embedding are 3*512, 6*512 and 128*512 respectively.
[0022] According to the information obtained in the early exploration, the three-level embedding attributes corresponding to the well are determined. For example, a well corresponds to a marine sedimentary facies of an oilfield, so this field is 0, and the corresponding embedding is obtained. The geological structure is a slope belt, so this field is 3, and the corresponding embedding is obtained. The development block is located in the 90th block, so the corresponding field is 90, and the corresponding embedding is obtained. Then according to the formula: ; the complete three-level regional embedding is obtained.
[0023] S2, generating a regional space relationship diagram according to the well site coordinates, and aggregating the features of adjacent regions through a graph neural network; The oil well is regarded as a node in the graph, the distance between each well is calculated according to the coordinate information of the well, and the adjacent node relationship of the graph is constructed. The distance <5km is defined as the adjacent node. The three-level regional embedding obtained in step S1 is the learning parameter of each well node. In the subsequent training process, the node parameter (three-level embedding) is updated according to the following formula: ; wherein is the feature of the lth layer regional node i, and the neighbor node set N(i) satisfies the spatial distance <5km. ReLU is an activation function. The mean function is a mean value function. The concat function is a vector concatenation function.
[0024] S3, input the well logging data into the backbone network to extract features, and perform gate weight fusion with the multi-level regional embedding vector; A general deep learning technology is used to build a backbone network for well logging data feature extraction. If the well logging data includes 9 well logging curves, a 9*1536 fully connected layer is established for example, so that a 1536-dimensional feature vector is obtained, which matches the 512*3-dimensional three-level embedding vector in step S1.
[0025] The well logging feature vector obtained in step S3 is denoted as The three-level embedding obtained in S2 is denoted as The weight fusion is performed through the following formula: ; ; Wherein is a trainable parameter matrix.
[0026] S4, output the reservoir parameter prediction result and perform end-to-end training.
[0027] Build a network corresponding to a specific task, input the fused vector obtained in step S3 into the task network, and perform training.
[0028] Build a network corresponding to a specific task, such as building a task network for porosity prediction, establish a 1536*1 fully connected layer as the task network, input the fused vector obtained in step S3 into the task network, and output the value of porosity. Using MSE as the loss function, the whole process of training can be performed. The expression of the MSE loss function is as follows: ; Wherein L is the loss value, N is the number of training samples, is the sample target true value, is the model predicted target value.
[0029] After training, it can be deployed in oil and gas business, inputting the regional information and well logging data of the well to complete the prediction of porosity.
[0030] The application also provides a well logging analysis model construction system based on multi-level regional feature association, comprising the following modules: A multi-level regional nested embedding vector construction module: analyzing the geographic hierarchical identifier in the well site code to construct a multi-level regional nested embedding vector; The feature aggregation module: calculate the spatial distance of the adjacent area according to the well site coordinates, generate a regional spatial relationship graph taking the area as a node and the spatial distance as an edge weight; and aggregate the features of the relationship graph through a graph neural network to obtain the enhanced regional features of the neighborhood; The prediction network construction module: construct an oil well reservoir parameter prediction network, which includes a feature extraction backbone network and a task network; The gating weight fusion module: collect the logging data of the target oil well, input the logging data into the backbone network after standardization to extract features, and then perform gating weight fusion with the multi-level regional embedding vectors constructed by the multi-level regional embedding vector construction module; The logging analysis module: input the fusion features of the gating weight fusion module into the oil well reservoir parameter prediction network constructed by the prediction network construction module, and output the reservoir parameter prediction results; set the MSE loss function and perform training, and after the training is completed, deploy the oil and gas task to complete the logging analysis.
[0031] In addition, an electronic device is provided, including a memory and a processor, wherein the memory is coupled with the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to realize the logging analysis model construction method based on multi-level regional feature association; and a computer readable storage medium is provided, which stores a computer program, wherein the program is executed by the processor to realize the logging analysis model construction method based on multi-level regional feature association.
[0032] The above only describes the embodiments of the present application and is not used to limit the present application. The present application can have various changes and variations for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the scope of claims of the present application.
Claims
1. A method for constructing a well logging analysis model based on multi-level regional feature correlation, characterized in that, Includes the following steps: S1. Parse the geographical hierarchy identifier in the well location code and construct a multi-level regional nested embedding vector; S2. Calculate the spatial distance between adjacent areas based on the well location coordinates, and generate a regional spatial relationship graph with the area as the node and the spatial distance as the edge weight; perform feature aggregation on the relationship graph through a graph neural network to obtain the neighborhood-enhanced regional features; S3. Construct an oil well reservoir parameter prediction network, which includes a feature extraction backbone network and a task network; S4. Collect logging data from the target oil well, standardize the logging data, input it into the backbone network to extract features, and then perform gated weight fusion with the multi-level regional embedding vector constructed in step S1. S5. Input the fusion features from step S4 into the oil well reservoir parameter prediction network constructed in step S3, and output the reservoir parameter prediction results; set the MSE loss function and train it; after training, deploy oil and gas tasks and complete the well logging analysis.
2. The method according to claim 1, characterized in that, The multi-level region nested embedding vector in S1 contains a three-level geographic hierarchy embedding, wherein: Hierarchical definition: Level 1: Sedimentary facies types, including marine, terrestrial, and transitional facies; The second level consists of geological structural units, including intracratonic depressions, foreland basins, rift basins, slope zones, fault zones, and depressions. Level 3: The number of development blocks, which is determined based on the actual development blocks of the oilfield; After determining the levels and the number of elements in each level, learnable embedding vectors are generated for each type in each level. For example, level 1 has 3 embedding vectors C1, C2, and C3; level 2 has 6 embedding vectors D1, D2, D3, D4, D5, and D6; and level 3 has k embedding vectors B1, B2, ..., Bk based on the number of blocks k. For a specific oil well, based on the well's three-level regional information, three corresponding embedding vectors are obtained: ,in For this oil well, the embedding vector corresponding to the sedimentary facies type is which one of C1, C2, and C3; This is the embedding vector corresponding to the geological structural unit of the oil well. Therefore, the embedding vector corresponding to the oil well development block; Then, the different levels of embedding are spliced together to form a complete region embedding; Right now: ； in, For the three levels of embedding vectors mentioned above, concat is the vector concatenation function.
3. The method according to claim 1, characterized in that, In step S2, a graph neural network is used to aggregate features of the relationship graph. Specifically, oil wells are treated as nodes in the graph. Based on the coordinate information of the wells, the distance between each well is calculated to construct the relationship between adjacent nodes in the graph. When the distance meets the set aggregation radius condition, an edge connection is established to form a regional spatial relationship graph. Based on the constructed regional spatial relationship graph, a graph neural network architecture is built to aggregate features of adjacent regions. A graph neural network algorithm is used, and the aggregation formula followed by the node feature update in the network is: ； in For a node i in the l-th layer region, the set of neighboring nodes N(i) satisfies that the spatial distance is <5km; is the ReLU activation function; mean is the mean function; concat is the vector concatenation function.
4. The method according to claim 1, characterized in that, Specifically, S3 involves constructing an oil well reservoir parameter prediction network, the prediction network comprising: (a) The feature extraction backbone network is used to extract deep features from the input preprocessed logging data, and may include a multi-level network architecture based on convolutional neural networks or self-attention mechanisms. (b) The task network is used to calculate and predict the values of reservoir parameters based on logging characteristics and regional embedded fusion characteristics. The prediction layer may include at least one fully connected layer followed by a ReLU activation function, and finally output the predicted values of porosity, permeability and oil saturation.
5. The method according to claim 1, characterized in that, Specifically, S4 involves inputting the logging data obtained from a comprehensive well inspection using logging equipment into the backbone network constructed in S3. Then, the extracted logging data features are combined with the multi-level region embedding vectors constructed in S2 using gated weight fusion. This gated weight fusion is achieved by calculating the similarity weights between the logging features and the region embeddings. The specific formula is as follows: ； ； in For the output characteristics of the well logging data backbone network, The current well's multi-level region embedding fusion vector is Trainable parameter matrix.
6. The method according to claim 1, characterized in that, The logging data in step S4 includes natural gamma, resistivity, and sonic transit time.
7. The method according to claim 1, characterized in that, The MSE loss function expression in step S5 is as follows: ； Where L is the loss value and N is the number of training samples. For the sample target true value, Predict the target value for the model.
8. The system for constructing a well logging analysis model based on multi-level regional feature correlation according to any one of claims 1-7, characterized in that, Includes the following modules: Multi-level region nested embedding vector construction module: Parses the geographic hierarchy identifier in well location codes and constructs multi-level region nested embedding vectors; Feature aggregation module: Calculates the spatial distance between adjacent areas based on well location coordinates, generates a regional spatial relationship graph with areas as nodes and spatial distance as edge weights; performs feature aggregation on this relationship graph through a graph neural network to obtain neighborhood-enhanced regional features; Prediction network construction module: Constructs an oil well reservoir parameter prediction network, which includes a feature extraction backbone network and a task network; Gated weight fusion module: Collects logging data from the target oil well, standardizes the logging data, inputs it into the backbone network to extract features, and then performs gated weight fusion with the multi-level region embedding vector constructed by the multi-level region nesting embedding vector construction module; Well logging analysis module: Input the fusion features of the gated weight fusion module into the oil well reservoir parameter prediction network constructed by the prediction network construction module, and output the reservoir parameter prediction results; set the MSE loss function and train it; after training, deploy oil and gas tasks and complete well logging analysis.
9. An electronic device comprising a memory and a processor, characterized in that, The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the well logging analysis model construction method based on multi-level regional feature association as described in any one of claims 1-7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the well logging analysis model construction method based on multi-level regional feature association as described in any one of claims 1-7.