Compact conglomerate reservoir identification method and system

Abstract

The embodiment of the invention provides a compact conglomerate reservoir identification method and system, and belongs to the field of unconventional oil-gas exploration and reservoir transformation. The method comprises the following steps: acquiring logging data of a reservoir section to be identified, and preprocessing the logging data; calling a corresponding lithology identification model based on the preprocessed logging data, and determining a conglomerate identification result of the corresponding reservoir section based on the lithology identification model; wherein the lithology identification model is obtained based on BP neural network training; based on the conglomerate identification result, corresponding compact conglomerate reservoir type matching is carried out in a pre-constructed compact conglomerate reservoir type library; and outputting a compact conglomerate reservoir identification result of the reservoir section to be identified based on the matched compact conglomerate reservoir type. According to the scheme, the method can effectively support the fracturing transformation decision of the reservoir, the development strategy is optimized, the oil and gas exploitation risk is reduced, and the production efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of unconventional oil and gas exploration and reservoir stimulation, specifically to a method and system for identifying tight conglomerate reservoirs. Background Technology

[0002] Existing technologies for identifying tight conglomerate reservoirs primarily rely on core observation, conventional logging data processing, and imaging logging techniques. These methods can provide information on reservoir structure and composition, such as grain size, support patterns, and lithological characteristics. However, core observation methods suffer from high sampling costs and limited sampling points, resulting in discontinuous data and hindering practical development needs. Imaging logging techniques, such as FMI (Fluorescent Imaging) logging, utilize microresistivity images of the surrounding formation to identify lithology and fracture distribution. While this technology offers advantages in conglomerate reservoir identification, it requires high-precision imaging equipment, and the analysis process is susceptible to subjective human factors, leading to low identification efficiency and accuracy.

[0003] With the development of machine learning technology, algorithms such as Support Vector Machine (SVM), Random Forest, and Backpropagation Neural Network (BPNN) have been increasingly applied to lithology identification. These algorithms rely on large amounts of well logging data and can extract reservoir features from the data through data mining and pattern recognition. However, existing machine learning methods suffer from limited model generalization ability, especially when data quality is low or feature selection is inaccurate, resulting in significantly reduced identification performance. Furthermore, for highly heterogeneous and complex geological structures such as tight conglomerate reservoirs, the prediction accuracy of existing machine learning algorithms still falls short of practical requirements.

[0004] Therefore, existing technologies have significant shortcomings in the identification and classification of tight conglomerate reservoirs, particularly in providing accurate predictions for reservoir fracturing. Traditional methods struggle to capture subtle differences in tight conglomerates, while single machine learning algorithms perform poorly in handling nonlinear and ambiguous regions, easily leading to biased lithofacies identification results. To address these challenges, a method integrating the advantages of multiple technologies is urgently needed to improve the accuracy of tight conglomerate reservoir identification and provide a reliable basis for fracturing. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for identifying tight conglomerate reservoirs, so as to at least solve the problem that the existing technology has obvious shortcomings in the identification and classification of tight conglomerate reservoirs.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for identifying tight conglomerate reservoirs. The method includes: acquiring logging data of a reservoir segment to be identified and performing preprocessing on the logging data; calling a corresponding lithology identification model based on the preprocessed logging data, and determining the conglomerate identification result of the corresponding reservoir segment based on the lithology identification model; wherein the lithology identification model is obtained by training a BP neural network; matching the corresponding tight conglomerate reservoir type in a pre-constructed tight conglomerate reservoir type library based on the conglomerate identification result; and outputting the tight conglomerate reservoir identification result of the reservoir segment to be identified based on the matched tight conglomerate reservoir type.

[0007] Optionally, the logging data of the reservoir segment to be identified includes any one or more of the following: natural gamma, density, flushed zone resistivity, neutron porosity, and sonic transit time.

[0008] Optionally, the preprocessing of the logging data includes: sequentially performing data cleaning and missing data filling on the logging data to obtain basic identification data; and performing normalization on each basic identification data to obtain preprocessed logging data.

[0009] Optionally, the method further includes training a lithology identification model; the training rules for the lithology identification model are as follows: collecting historical logging data and labeling the historical logging data with lithology samples; performing normalization processing on the historical logging data with lithology sample labeling to obtain training samples; and training the model in a pre-determined BP neural network based on the training samples to obtain the lithology identification model.

[0010] Optionally, the normalization rule for the historical well logging data with completed lithological sample annotation is as follows:

[0011] in, The i-th type of well logging data after normalization; For the i-th type of well logging data; The minimum value of the i-th logging curve; It represents the maximum value of the i-th logging curve.

[0012] Optionally, the step of training the model in a predetermined BP neural network based on training samples to obtain a lithology identification model includes initializing the predetermined BP neural network, importing training samples, and performing multiple rounds of training based on the error function of the predetermined BP neural network. In each round of training, the training results are denormalized to obtain the corresponding training identification results, and the training identification results are compared with the lithology sample annotation information of the corresponding training samples until a model with expected regression performance is obtained, which serves as the lithology identification model.

[0013] Optionally, the error function of the determined BP neural network is:

[0014] Where OK is the output of node K; N is the number of training samples; E is the error function; and TK is the lithological sample annotation information.

[0015] Optionally, the denormalization rule for performing denormalization on the training results is as follows:

[0016] in, This is the output after inverse normalization; This represents the maximum value of the lithological data; This represents the minimum value of the lithological data.

[0017] Optionally, the lithological data determination rule is as follows: each tight conglomerate reservoir type is coded, and the corresponding coded value is used as the lithological data for the corresponding tight conglomerate reservoir type; the normalization rule for normalizing the historical logging data with completed lithological sample labeling also includes:

[0018] in, These are the lithological data for the j-th type of tight conglomerate reservoir after normalization. This represents the lithological data for the j-th type of tight conglomerate reservoir.

[0019] Optionally, the tight conglomerate reservoir type includes: above-water conglomerate and / or underwater conglomerate; wherein, the above-water conglomerate includes any one or more of the following: coarse-to-giant conglomerate facies supported by gravel mainly composed of metamorphic and igneous rocks, medium-to-coarse conglomerate facies supported by matrix mainly composed of tuff and igneous rocks, medium-to-fine conglomerate facies supported by matrix mainly composed of tuff, and medium-to-fine conglomerate facies supported by grain mainly composed of tuff and igneous rocks; the underwater conglomerate includes any one or more of the following: medium-to-fine conglomerate facies supported by grain mainly composed of grain mainly composed of tuff and igneous rocks, medium-to-coarse conglomerate facies supported by gravel mainly composed of tuff and sedimentary rocks, coarse-to-giant conglomerate facies supported by gravel mainly composed of tuff, igneous rocks, and sedimentary rocks, and medium-to-coarse conglomerate facies supported by matrix mainly composed of tuff, igneous rocks, and sedimentary rocks.

[0020] Optionally, the rules for obtaining the conglomerate identification results are as follows:

[0021] in, Results of conglomerate identification; The output is the denormalized result of the lithology identification model; 1, 2, 3, 4, ... N represent the tight conglomerate reservoir type.

[0022] A second aspect of the present invention provides a tight conglomerate reservoir identification system, the system comprising: an acquisition unit for acquiring well logging data of a reservoir segment to be identified and performing preprocessing on the well logging data; a training unit for calling a corresponding lithology identification model based on the preprocessed well logging data and determining the conglomerate identification result of the corresponding reservoir segment based on the lithology identification model; wherein the lithology identification model is obtained by training a BP neural network; a matching unit for matching the corresponding tight conglomerate reservoir type in a pre-constructed tight conglomerate reservoir type library based on the conglomerate identification result; and an output unit for outputting the tight conglomerate reservoir identification result of the reservoir segment to be identified based on the matched tight conglomerate reservoir type.

[0023] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for identifying tight conglomerate reservoirs based on auxiliary variable screening.

[0024] A fourth aspect of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for identifying tight conglomerate reservoirs based on auxiliary variable screening.

[0025] The fifth aspect of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for identifying tight conglomerate reservoirs based on auxiliary variable screening.

[0026] On the other hand, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described tight conglomerate reservoir identification method.

[0027] Through the above technical solution, this invention improves the accuracy of tight conglomerate reservoir identification by collecting and preprocessing well logging data of the reservoir section to be identified. Using the preprocessed well logging data, the system calls a lithology identification model trained on a BP neural network to generate conglomerate identification results for the reservoir section. This lithology identification model can effectively handle the nonlinear relationships and complexity of well logging data, thereby improving the accuracy of lithology identification. Next, based on the identified conglomerate type, it is matched with a pre-constructed tight conglomerate reservoir type library to determine the specific tight conglomerate type of the reservoir section to be identified. The final output of the tight conglomerate reservoir identification results provides more precise guidance for reservoir development. This technical effect enables the method to effectively support reservoir fracturing and stimulation decisions, optimize development strategies, reduce oil and gas extraction risks, and improve production efficiency.

[0028] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0029] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the steps of a tight conglomerate reservoir identification method provided in one embodiment of the present invention; Figure 2 This is an identification diagram of the above-water and underwater conglomerate logging curves of the Baikouquan Formation on the northern slope of the Mahu Depression, provided by one embodiment of the present invention. Figure 3 This is an underwater conglomerate logging curve identification diagram of the Baikouquan Formation on the northern slope of the Mahu Depression provided by one embodiment of the present invention; Figure 4 This is a logging curve identification diagram of the water conglomerate of the Baikouquan Formation on the northern slope of the Mahu Depression provided by one embodiment of the present invention; Figure 5 This is a diagram illustrating the iterative process of a BP neural network according to one embodiment of the present invention; Figure 6 This is a training trend graph of a BP neural network provided in one embodiment of the present invention; Figure 7 This is a BP neural network linear regression analysis provided by one embodiment of the present invention; Figure 8 This is a system structure diagram of a tight conglomerate reservoir identification system provided in one embodiment of the present invention. Detailed Implementation

[0030] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0031] Figure 1 This is a flowchart of a method for identifying tight conglomerate reservoirs according to one embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a method for identifying tight conglomerate reservoirs, the method comprising: Step S10: Collect logging data of the reservoir section to be identified and perform preprocessing on the logging data.

[0032] Specifically, the logging data for the reservoir section to be identified includes any one or more of the following: natural gamma ray, density, resistivity of the flushed zone, neutron porosity, and sonic transit time. The preprocessing of the logging data includes: sequentially performing data cleaning and missing data filling on the logging data to obtain basic identification data; and performing normalization on each basic identification data to obtain preprocessed logging data.

[0033] In this embodiment of the invention, the method analyzes various logging curve data and selects logging curves that are sensitive to changes in different types of conglomerate. Logging data that provide good indications for different types of conglomerate include natural gamma (GR), density (DEN), flushed zone resistivity (Rxo), neutron porosity (CNL), and acoustic transit time (AC). Impedance per minute (IMP) is calculated using density and acoustic transit time logging data. The original logging curve ranges corresponding to tight conglomerate reservoirs are shown in Table 1.

[0034] Table 1. Identification parameters for different types of sandstone and conglomerate on the northern slope of the Mahu Depression

[0035] Therefore, when performing speech and behavior recognition, the present invention needs to collect these well logging data.

[0036] Furthermore, the present invention first acquires well logging data of the reservoir section to be identified. This data includes any one or more of the following: natural gamma (GR), density (DEN), flushed zone resistivity (Rxo), neutron porosity (CNL), and acoustic transit time (AC). This well logging data provides basic physical properties and lithological characteristics of the reservoir, directly reflecting its composition, structure, and porosity, and is of great significance for reservoir identification. However, well logging data is often affected by logging tools, wellbore environment, and human factors, and may contain noise, outliers, and missing values, which can affect the accuracy of subsequent identification models. Therefore, preprocessing the well logging data is a crucial step in identifying reservoir types.

[0037] Furthermore, in the preprocessing stage, the acquired logging data is first cleaned. Data cleaning includes removing noise and outliers generated during the logging process to ensure the authenticity and consistency of the data. Specific methods can employ filtering algorithms or statistical methods for removing outliers to clean up outliers and noisy data in the logging curves, thereby improving data reliability. The cleaned data better reflects the actual situation of the reservoir, providing a more accurate foundation for subsequent analysis.

[0038] Furthermore, missing data imputation is performed on the cleaned logging data. During the logging process, due to equipment or wellbore limitations, some logging curves may be incomplete or missing, resulting in gaps in the logging data. Common imputation methods include linear interpolation, mean imputation, and machine learning-based predictive imputation. Imputation effectively reduces the impact of missing data on the identification results, ensuring the integrity of the logging data and thus providing sufficient data support for subsequent model input.

[0039] Furthermore, after data cleaning and missing data imputation, normalization processing needs to be performed on each identification baseline data. The purpose of normalization is to transform data with different dimensions to a uniform scale range (e.g., 0 to 1) to facilitate subsequent model calculations and parameter optimization. Different well logging data often have significantly different dimensions and numerical ranges; for example, natural gamma has a wide range, while density varies less. Normalization can eliminate these differences in numerical range, ensuring that each data point has equal influence weight in the model and preventing any feature from being overemphasized or ignored. Common normalization methods include max-min normalization and Z-score standardization. These methods provide preprocessed well logging data, offering standardized values ​​for the input of the lithology identification model.

[0040] Furthermore, the preprocessed logging data was used to train a lithology identification model based on a backpropagation (BP) neural network. The BP neural network is a deep learning model with powerful nonlinear mapping capabilities, suitable for processing complex and nonlinear geological data. This model, trained on a large number of historical samples, learns the logging curve characteristics corresponding to different lithologies, forming a classification model for reservoir identification. In practical applications, the preprocessed logging data is input into the BP neural network, and through the model's nonlinear calculations and predictions, the conglomerate identification results for the reservoir section are obtained. These identification results can be further used for matching with a tight conglomerate reservoir type library, thereby accurately determining the reservoir type and providing a basis for subsequent reservoir stimulation and fracturing decisions.

[0041] Step S20: Based on the preprocessed logging data, call the corresponding lithology identification model, and determine the conglomerate identification result of the corresponding reservoir section based on the lithology identification model.

[0042] Specifically, the method further includes training a lithology identification model; the training rules for the lithology identification model are as follows: collecting historical logging data and labeling the historical logging data with lithology samples; performing normalization processing on the historical logging data with lithology sample labeling to obtain training samples; and training the model in a pre-determined BP neural network based on the training samples to obtain the lithology identification model.

[0043] In this embodiment of the invention, acquiring historical logging data is fundamental to the entire training process. Historical logging data includes logging curves for natural gamma ray, density, flushed zone resistivity, neutron porosity, and sonic transit time, which reflect the geological characteristics of the reservoir. Through the accumulation of a large amount of historical logging data, not only can different lithological types be covered, but formation variation characteristics can also be effectively reflected, providing rich data support for the model.

[0044] Furthermore, lithological sample labeling of historical well logging data is a crucial step in model training. Lithological sample labeling refers to classifying and labeling each well logging data point according to the results of geological experts or experimental analysis, thus determining its corresponding reservoir lithology category. This process can be accomplished through methods such as core observation and geological report analysis. The labeled lithological samples serve as "labels" for model training, enabling the neural network to learn the differences between different lithologies. Since the accuracy of lithology identification largely depends on the precision of sample labeling, the scientific validity and reliability of the labels must be ensured during the labeling process.

[0045] After annotation, the annotated logging data undergoes normalization to eliminate dimensional differences between different logging data, allowing for comparison and learning on the same scale. Normalization standardizes the logging data to a specific range (e.g., 0 to 1), thereby improving the stability and efficiency of the training process and preventing certain features from being ignored or amplified due to excessively large or small values. The normalized data is then used as training samples input into the neural network.

[0046] Based on training samples, a lithology identification model is constructed by training a predefined backpropagation (BP) neural network. BP neural networks possess powerful nonlinear mapping capabilities; through the backpropagation algorithm, the network weights and biases are continuously adjusted to make the network output as close as possible to the target value. The training process mainly includes two stages: forward propagation and backpropagation. In forward propagation, the training sample data is passed through the neural network layer by layer, outputting preliminary predicted values. Then, the difference between the predicted and actual values ​​is calculated using an error function, and the error is continuously corrected and the network parameters are updated through backpropagation until the error reaches the set convergence condition. After training, a BP neural network model capable of accurately identifying lithology is obtained.

[0047] Specifically, the normalization rules for historical well logging data with completed lithological sample annotation are as follows:

[0048] in, The i-th type of well logging data after normalization; For the i-th type of well logging data; The minimum value of the i-th logging curve; It represents the maximum value of the i-th logging curve.

[0049] Specifically, the step of training a model in a pre-determined BP neural network based on training samples to obtain a lithology identification model includes initializing the pre-determined BP neural network, importing training samples, and performing multiple rounds of training based on the error function of the determined BP neural network. In each round of training, the training results are denormalized to obtain the corresponding training identification results, and the training identification results are compared with the lithology sample annotation information of the corresponding training samples until a model with expected regression performance is obtained, which serves as the lithology identification model.

[0050] In this embodiment of the invention, the training of the lithology identification model is based on a pre-defined backpropagation (BP) neural network and includes the processes of initializing the neural network, importing training samples, setting an error function, and performing multiple rounds of training. First, the BP neural network is initialized to determine its structure and parameters (such as the number of layers, nodes, and learning rate), laying the foundation for subsequent training. Then, normalized training samples are imported to keep the network input features on a uniform scale, ensuring training stability and computational efficiency.

[0051] During training, the BP neural network continuously adjusts its parameters using an error function. After each training round, the training recognition results are denormalized to restore them to the original data scale, facilitating comparison with the actual lithological annotation data of the training samples. This comparison allows for the evaluation of the current model's recognition accuracy. If the recognition results deviate significantly from the annotation information, the error function adjusts the network weights and biases through a backpropagation algorithm to reduce the error. After multiple iterations, as training progresses, the model output gradually approaches the target value. Finally, when the regression performance of the training recognition results reaches the expected accuracy, the model is considered to have converged, forming a lithological identification model that meets the requirements.

[0052] Specifically, the error function of the determined BP neural network is:

[0053] Where OK is the output of node K; N is the number of training samples; E is the error function; and TK is the lithological sample annotation information.

[0054] Furthermore, the denormalization rule for performing denormalization on the training results is as follows:

[0055] in, This is the output after inverse normalization; This represents the maximum value of the lithological data; This represents the minimum value of the lithological data.

[0056] Furthermore, the rules for determining the lithological data are as follows: each tight conglomerate reservoir type is coded, and the corresponding coded value is used as the lithological data for the corresponding tight conglomerate reservoir type; the normalization rules for normalizing the historical logging data after lithological sample labeling also include:

[0057] in, These are the normalized lithological data for the j-th type of tight conglomerate reservoir. This represents the lithological data for the j-th type of tight conglomerate reservoir.

[0058] Step S30: Based on the conglomerate identification results, perform corresponding tight conglomerate reservoir type matching in the pre-constructed tight conglomerate reservoir type library.

[0059] Specifically, the tight conglomerate reservoir types include: above-water conglomerate and / or underwater conglomerate; wherein, the above-water conglomerate includes any one or more of the following: coarse-to-giant conglomerate facies supported by gravel mainly composed of metamorphic and igneous rocks, medium-to-coarse conglomerate facies supported by matrix mainly composed of tuff and igneous rocks, medium-to-fine conglomerate facies supported by matrix mainly composed of tuff, and medium-to-fine conglomerate facies supported by grain mainly composed of tuff and igneous rocks; the underwater conglomerate includes any one or more of the following: medium-to-fine conglomerate facies supported by grain mainly composed of grain mainly composed of tuff and igneous rocks, medium-to-coarse conglomerate facies supported by gravel mainly composed of tuff and sedimentary rocks, coarse-to-giant conglomerate facies supported by gravel mainly composed of tuff, igneous rocks, and sedimentary rocks, and medium-to-coarse conglomerate facies supported by matrix mainly composed of tuff, igneous rocks, and sedimentary rocks.

[0060] Furthermore, the rules for obtaining the conglomerate identification results are as follows:

[0061] in, Results of conglomerate identification; The output is the denormalized result of the lithology identification model; 1, 2, 3, 4, ... N represent the tight conglomerate reservoir types.

[0062] In this invention, the tight conglomerate reservoir is classified into two main categories: above-water conglomerate and underwater conglomerate. Each category is further subdivided into different lithofacies types, which are further classified according to their main rock composition and support method.

[0063] For subsea conglomerates, the classification in the type library includes four main conglomerate facies: coarse-grained to mega-conglomerate facies supported by gravel, dominated by metamorphic and igneous rocks; medium- to coarse-grained conglomerate facies supported by matrix, dominated by tuff and igneous rocks; medium- to fine-grained conglomerate facies supported by matrix, dominated by tuff; and medium- to fine-grained conglomerate facies supported by grain, dominated by tuff. These facies formed in relatively shallow sedimentary environments and are typically subject to more weathering, resulting in relatively complex structural characteristics and pore distribution. These characteristics directly affect reservoir permeability and porosity, and are of great significance for the assessment and development of oil and gas reservoirs.

[0064] In the classification of underwater conglomerates, the tight conglomerate reservoir type pool also includes four main lithofacies types: grain-supported medium- to fine-grained conglomerate facies dominated by tuff and igneous rocks; gravel-supported medium- to coarse-grained conglomerate facies dominated by tuff and sedimentary rocks; gravel-supported coarse- to mega-conglomerate facies dominated by tuff; and matrix-supported medium- to coarse-grained conglomerate facies dominated by tuff, igneous rocks, and sedimentary rocks. These lithofacies typically form in deeper-water sedimentary environments, with lower porosity and permeability due to compaction, resulting in higher reservoir heterogeneity and posing greater challenges to oil and gas accumulation and extraction. These unique characteristics of underwater conglomerates make accurate identification and matching particularly important to provide a scientific basis for the rational development of reservoirs.

[0065] In practical applications, the conglomerate identification results obtained from the model can be compared with a tight conglomerate reservoir type library to quickly and accurately classify the identified conglomerate types into the most suitable reservoir types. This matching process not only improves the accuracy of the identification results but also further refines the reservoir classification, making the reservoir model more targeted. For oil and gas development, refined reservoir classification can better guide the formulation of fracturing and stimulation strategies, accurately identify ideal areas for efficient development, optimize well location layout, and reduce the risk of ineffective exploitation.

[0066] Based on the present invention, by matching with a type library, the specific type of tight conglomerate can be accurately identified according to the subtle characteristics of different conglomerate facies, avoiding identification errors caused by relying solely on well logging data. This type library-based matching process significantly improves the accuracy of identification. Superaquatic and subaquatic conglomerates differ significantly in geological origin and reservoir characteristics; further subdivision into different lithofacies types clarifies the reservoir's structure and physical properties. This refined classification not only facilitates the accurate construction of reservoir models but also provides clear guidance for subsequent reservoir stimulation and development.

[0067] Furthermore, precise reservoir classification can effectively support fracturing and stimulation decisions. Different types of conglomerate exhibit varying fracture propagation and support effects during fracturing; therefore, identifying reservoir types helps select the most suitable fracturing technology and well placement strategy, thereby improving oil and gas extraction efficiency. High-precision reservoir type matching can help identify "sweet spots" (i.e., areas that are easy to extract and have high production capacity), reducing the risks of blind drilling and ineffective fracturing. This has a significant effect on reducing extraction costs and improving the economics of oil and gas field development.

[0068] Step S40: Output the tight conglomerate reservoir identification results of the reservoir segment to be identified based on the matched tight conglomerate reservoir type.

[0069] Specifically, after matching the tight conglomerate reservoir types, the system outputs the final tight conglomerate reservoir identification result based on the matching results. This identification result includes detailed information such as the specific lithological category, structural characteristics, and reservoir properties of the reservoir segment to be identified. This process associates the matched reservoir type with the reservoir segment to be identified, thus providing accurate reservoir classification information. In this way, geological engineers can intuitively understand the lithological characteristics of each reservoir segment, providing a reliable basis for further fracturing, development planning, and well placement. Furthermore, the identification results can also serve as important basic data in reservoir simulation and reservoir modeling to accurately predict oil and gas production capacity and optimize extraction strategies.

[0070] Example: Using the tight conglomerate reservoir identification and prediction technology that integrates neural network model and well logging data processing mentioned above, the tight conglomerate reservoir of the Triassic Baikouquan Formation on the northern slope of the Mahu Depression in the Junggar Basin of western China was identified and predicted. The specific process and results are shown below.

[0071] (1) Well logging data processing technology The well logging data is processed and analyzed using cross-plots. When the sample size is sufficiently large, large conglomerate types are identified in DEN-GR (…). Figure 2 ) and IMP-Rxo ( Figure 2The data is well-identified on the chart. The logging data for Type A conglomerate has a narrower range: GR values ​​of 50 API-90 API, DEN values ​​of 2.56 g / cm³-2.64 g / cm³, RXO of 10 Ω·m-30 Ω·m, and IMP of 11.2×10⁹ g / m³·s-12.2×10⁹ g / m³·s. The logging data for Type B conglomerate has a wider range: GR values ​​of 20 API-90 API, DEN values ​​of 2.44 g / cm³-2.58 g / cm³, RXO of 30 Ω·m-200 Ω·m, and IMP of 10×10⁹ g / m³·s-12.8×10⁹ g / m³·s. The blurred areas on the chart are small, indicating good identification.

[0072] This method can identify four subclasses within Class A conglomerate. Although there is some overlap in area between different lithofacies, the overlap area is small, indicating good identification results. Figure 3 However, the four subclasses of conglomerate B have a large overlap area in the well logging chart, resulting in poor identification performance. Figure 4 In such cases, a method capable of handling nonlinearity and fuzzy judgments is needed to identify lithology in data intersection areas of cross-plots. Backpropagation (BP) neural networks, a deep learning method, can effectively handle this type of problem. BP neural networks, through their unique sample learning capabilities, can perform lithology discrimination, overcoming the shortcomings of complex equation algorithms and multivariate geological analysis. Their strong self-learning and fault-tolerant capabilities can solve the problem of highly nonlinear mapping between well logging data and formation lithology. Furthermore, another advantage of the BP neural network method is that it can achieve a nonlinear mapping relationship between system input and output without establishing complex mathematical equations.

[0073] Well logging curve processing technology performs well in identifying large-scale conglomerate reservoirs, but its effectiveness is poor in identifying sub-type tight conglomerate reservoirs, resulting in large blurred areas on the chart. To address this issue, 700 m core data from 26 wells in the study area were selected, corresponding to five well logging curves (DEN, GR, Rxo, CNL, and AC), totaling over 2000 data points, as training samples. Using Matlab software as the platform, a BP neural network algorithm was developed to predict sandstone and conglomerate bodies in sections without coring.

[0074] 1) Backpropagation (BP) neural network structure: The network structure consists of three layers: one input layer with 5 nodes, one hidden layer with 20 nodes, and two output layers with 2 and 8 nodes respectively. After setting the parameters of the BP neural network model, the network is trained.

[0075] 2) Iterative process diagram display ( Figure 5 The goal is to achieve convergence in step 29.

[0076] 3) Network training trends: Network training trend chart ( Figure 6 The data shows that as the number of training steps increases, the mean square error in the vertical axis becomes smaller and smaller, and the training trend gets closer and closer to the ideal error. When the training step size is 23, the iteration does not produce a difference, and the training curve intersects with the target error curve, indicating that the training has reached convergence and the sample training ends.

[0077] 4) Linear Regression Analysis: Linear regression analysis was performed on the training objective, validation objective, and test dataset objective output by the BP neural network. The results are as follows: Figure 7 As shown, all data output values ​​fit the expected values ​​well, and the data all decrease along a 45° angle. The corresponding R values ​​all reach above 0.9, indicating that the BP neural network has a good response effect and the model has high accuracy. This model can be used to predict sandstone and conglomerate bodies in the study area.

[0078] 5) Verification of recognition results The prediction results were further verified using wells M154 and X82, which have relatively long coring sections. Well M154 has a large coring thickness in the Baikouquan Formation, with continuous coring reaching 70 m. The prediction results show that the coring section develops A-1 type conglomerate. A-1 type conglomerate is characterized by reduced color, coarse grain size, gravel support, strong cementation, and the gravel composition is mainly volcanic clastic rocks, metamorphic rocks, and igneous rocks. The conglomerate composition identification results and the detailed description of the gravel structure in the core of well M154 show that this type of conglomerate belongs to A-1 type, which is consistent with the prediction results of the BP neural network algorithm.

[0079] The prediction results from well M136 indicated that the cored section contained A-2 type conglomerate. A-2 type conglomerate is characterized by its reduced color, coarse grain size, matrix support, and weak cementation. The gravel composition is mainly volcanic clastic and igneous rocks. However, gravel composition analysis and core observations show that the M136 gravel is primarily composed of tuff, felite, and andesite, with high matrix content and low cement content, consistent with the prediction results. Furthermore, the actual observations and analysis results from wells such as X82 are consistent with the prediction results.

[0080] Figure 8 This is a system structure diagram of a tight conglomerate reservoir identification system provided in one embodiment of the present invention. Figure 8As shown, this invention provides a tight conglomerate reservoir identification system. The system includes: an acquisition unit for acquiring well logging data of the reservoir segment to be identified and performing preprocessing on the well logging data; a training unit for calling the corresponding lithology identification model based on the preprocessed well logging data and determining the conglomerate identification result of the corresponding reservoir segment based on the lithology identification model; wherein the lithology identification model is obtained by training a BP neural network; a matching unit for matching the corresponding tight conglomerate reservoir type in a pre-constructed tight conglomerate reservoir type library based on the conglomerate identification result; and an output unit for outputting the tight conglomerate reservoir identification result of the reservoir segment to be identified based on the matched tight conglomerate reservoir type.

[0081] Based on the present invention, the extremely low porosity and ultra-low permeability of tight conglomerate reservoirs result in very low natural production, necessitating hydraulic fracturing to maintain stable production. However, the inherent characteristics of tight conglomerate reservoirs, such as their strong structural and compositional heterogeneity (geological factors), lead to poor hydraulic fracturing effects and difficulty in controlling fracture morphology after fracturing. Although current methods for classifying and evaluating tight conglomerate reservoirs considering reservoir stimulation effects have been developed, their effectiveness in identifying fracturing sweet spots remains limited due to a lack of effective identification and prediction technologies. Focusing on this problem, after trying various identification methods, a multidisciplinary integrated study was conducted, and two identification technologies—a BP neural network model and well logging data processing—were selected to identify tight conglomerate reservoirs. The results show that different identification technologies respond differently to the classification and evaluation results of different levels of tight conglomerate reservoirs. The sensitive logging curve intersection chart obtained through well logging data processing responds well to the primary classification and evaluation results, with a small overlapping area in the chart; however, it responds poorly to the secondary classification and evaluation results, with a large fuzzy area in the chart. Leveraging the ability of the BP neural network algorithm to handle fuzzy regions, the secondary classification evaluation results were processed. The model results show that all data output values ​​fit the expected values ​​well, with corresponding R values ​​exceeding 0.9, indicating that the BP neural network algorithm is effective in identifying tight conglomerate reservoirs and has high model accuracy. The prediction results were validated using well sections with longer core samples. The model predictions were consistent with the actual observations and descriptions, demonstrating the good identification effect of combining the two techniques.

[0082] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for identifying tight conglomerate reservoirs based on auxiliary variable screening.

[0083] A fourth aspect of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for identifying tight conglomerate reservoirs based on auxiliary variable screening.

[0084] The fifth aspect of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for identifying tight conglomerate reservoirs based on auxiliary variable screening.

[0085] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0086] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0087] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for identifying tight conglomerate reservoirs, characterized in that, The method includes: Collect logging data of the reservoir section to be identified, and perform preprocessing on the logging data; Based on the preprocessed well logging data, the corresponding lithology identification model is invoked, and the conglomerate identification result of the corresponding reservoir section is determined based on the lithology identification model; wherein, The lithology identification model is obtained based on BP neural network training; Based on the conglomerate identification results, the corresponding tight conglomerate reservoir type is matched in the pre-constructed tight conglomerate reservoir type library; Based on the matched tight conglomerate reservoir type, the output shows the tight conglomerate reservoir identification results for the reservoir segment to be identified.

2. The method according to claim 1, characterized in that, The logging data for the reservoir section to be identified includes: Any one or more of the following: natural gamma, density, wash band resistivity, neutron porosity, and acoustic transit time.

3. The method according to claim 1, characterized in that, The preprocessing of the well logging data includes: The well logging data is sequentially cleaned and missing data is filled to obtain basic identification data. Normalization is performed on each identification base data to obtain preprocessed well logging data.

4. The method according to claim 1, characterized in that, The method also includes training a lithology identification model; The training rules for the lithology identification model are as follows: Collect historical logging data and perform lithological sample annotation on the historical logging data; Normalize the historical well logging data with completed lithological sample annotation to obtain training samples; Based on the training samples, the model is trained in a pre-determined BP neural network to obtain a lithology identification model.

5. The method according to claim 4, characterized in that, The processing rule for normalizing historical well logging data with completed lithological sample annotation is as follows: ； in, The i-th type of well logging data after normalization; For the i-th type of well logging data; The minimum value of the i-th logging curve; It represents the maximum value of the i-th logging curve.

6. The method according to claim 4, characterized in that, The process of training a model in a pre-determined BP neural network based on training samples to obtain a lithology identification model includes: Initialize a predetermined BP neural network, import training samples, and perform multiple rounds of training based on the predetermined error function of the BP neural network; In each round of training, the training results are denormalized to obtain the corresponding training identification results. The training identification results are then compared with the lithological sample annotation information of the corresponding training samples until a model with expected regression performance is obtained, which serves as the lithological identification model.

7. The method according to claim 6, characterized in that, The error function of the determined BP neural network is: ； Where OK is the output of node K; N is the number of training samples. E is the error function; TK represents the lithological sample annotation information.

8. The method according to claim 6, characterized in that, The denormalization rule for performing denormalization on the training results is as follows: ； in, This is the output after inverse normalization; This represents the maximum value of the lithological data; This represents the minimum value of the lithological data.

9. The method according to claim 8, characterized in that, The rules for determining the lithological data are as follows: Each type of tight conglomerate reservoir is coded, and the corresponding coded value is used as the lithological data for the corresponding tight conglomerate reservoir type. The normalization rules for normalizing historical logging data with completed lithological sample annotation also include: ； in, These are the normalized lithological data for the j-th type of tight conglomerate reservoir. This represents the lithological data for the j-th type of tight conglomerate reservoir.

10. The method according to claim 1, characterized in that, The tight conglomerate reservoir types include: Submarine conglomerate and / or underwater conglomerate; wherein, The marine conglomerate includes: Any one or more of the following: coarse-to-giant conglomerate facies supported by gravel, medium-to-coarse conglomerate facies supported by matrix, medium-to-fine conglomerate facies supported by matrix, and medium-to-fine conglomerate facies supported by grain, mainly composed of metamorphic and igneous rocks; The underwater conglomerate includes: The facies can be any one or more of the following: medium-fine conglomerate facies supported by tuff and igneous rocks as the main grains; medium-coarse conglomerate facies supported by tuff and sedimentary rocks as the main gravels; coarse-macro conglomerate facies supported by tuff and igneous rocks as the main gravels; and medium-coarse conglomerate facies supported by matrix composed of tuff, igneous rocks, and sedimentary rocks.

11. The method according to claim 1, characterized in that, The rules for obtaining the conglomerate identification results are as follows: ； in, Results of conglomerate identification; This is the output of the lithology identification model after inverse normalization; 1, 2, 3, 4, ... N represent tight conglomerate reservoir types.

12. A tight conglomerate reservoir identification system, characterized in that, The system includes: The acquisition unit is used to acquire logging data of the reservoir section to be identified and to perform preprocessing on the logging data; The training unit is used to call the corresponding lithology identification model based on the preprocessed well logging data, and determine the conglomerate identification result of the corresponding reservoir section based on the lithology identification model; wherein, The lithology identification model is obtained based on BP neural network training; The matching unit is used to match the corresponding tight conglomerate reservoir type in a pre-built tight conglomerate reservoir type library based on the conglomerate identification results. The output unit is used to output the tight conglomerate reservoir identification results of the reservoir segment to be identified based on the matched tight conglomerate reservoir type.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the tight conglomerate reservoir identification method based on auxiliary variable screening as described in any one of claims 1-11.

14. 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 computer program, it implements the tight conglomerate reservoir identification method based on auxiliary variable screening as described in any one of claims 1-11.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the tight conglomerate reservoir identification method based on auxiliary variable screening as described in any one of claims 1-11.

Images