A method, system, storage medium and device for depositing microfacies type recognition

Abstract

The application provides a sedimentary microfacies type identification method, system, storage medium and device, relates to the field of geological feature identification, and comprises the following steps: acquiring facies sequence characteristics, logging curve shape and numerical characteristics and vertical probability distribution characteristics of each microfacies type in a specific sedimentary environment; performing correlation analysis on the normalized logging curve numerical characteristics, and screening out preferred logging curves; training the vertical probability distribution characteristics of the microfacies, obtaining an optimal microfacies distribution sequence and a transition probability matrix; respectively constructing a first microfacies label library and a second microfacies label library; respectively establishing a first identification model and a second identification model based on the two label libraries to form a joint model; and finally inputting non-coring well data to be predicted into the joint model, and determining the corresponding sedimentary microfacies type by weighting the highest probability output. The application can intelligently identify complex microfacies, and realizes fine interpretation of sedimentary microfacies in a research area.

Description

Technical Field

[0001] This application relates to the field of geological feature identification, and in particular to a method, system, storage medium, and device for identifying sedimentary microfacies types. Background Technology

[0002] Sedimentary microfacies are the most fundamental geological units in a sedimentary system, reflecting specific sedimentary environments and processes. The precise identification and interpretation of sedimentary microfacies are fundamental for understanding sedimentary environments, studying sedimentary distribution, and subsequent research on facies-controlled reservoirs and facies-controlled modeling. Currently, single-well sedimentary microfacies identification and interpretation mainly rely on core facies markers, well logging facies markers, and core calibration logging. Identification templates for different types of sedimentary microfacies are artificially established, and interpretation is primarily based on human experience. This leads to significant influence from expert experience, resulting in multiple interpretations and difficulties in ensuring consistency and objectivity. Summary of the Invention

[0003] The purpose of this application is to provide a method, system, storage medium, and device for identifying sedimentary microfacies types, which can improve the identification accuracy of complex sedimentary microfacies and achieve a detailed interpretation of sedimentary microfacies.

[0004] To address the aforementioned technical problems, this application provides a method for identifying sedimentary microfacies types, the specific technical solution of which is as follows:

[0005] Obtain the microfacies types of a specific sedimentary environment, and the lithofacies sequence characteristics, well logging curve morphology characteristics, well logging curve numerical characteristics, and microfacies vertical probability distribution characteristics corresponding to each microfacies type;

[0006] The numerical characteristics of the well logging curves are preprocessed to obtain standardized numerical characteristics of the well logging curves;

[0007] Correlation analysis was performed on the numerical characteristics of the standardized logging curves and the microfacies type to obtain the preferred logging curves;

[0008] The microphase vertical probability distribution features are trained using a hidden Markov model to obtain the optimal microphase distribution sequence and microphase transition probability matrix;

[0009] The preferred logging curves are normalized to obtain the normalized numerical characteristics of the preferred logging curves.

[0010] A first microfacies tag library is constructed based on the lithofacies sequence characteristics corresponding to the microfacies type of the core well, the logging curve morphology characteristics, and the optimal microfacies distribution sequence; a second microfacies tag library is constructed based on the microfacies type of the core well and the corresponding normalized optimal logging curve numerical characteristics.

[0011] A joint model comprising a first recognition model and a second recognition model is constructed, and the first recognition model and the second recognition model are trained synchronously, with the model parameters constrained during the training process using the total loss function; wherein, the first recognition model is trained based on the first micro-phase label library using the Transformer model as the base model; the second recognition model is trained based on the second micro-phase label library using the XGBoost model as the base model.

[0012] The numerical characteristics, morphological characteristics, and lithofacies sequence characteristics of the logging curves to be predicted from the non-cored wells are input into the first identification model and the second identification model in the joint model, respectively. The sedimentary microfacies type identification result corresponding to the logging curves to be predicted from the non-cored wells is determined based on the output result of the joint model. The output result is the microfacies type with the highest probability after weighting the first identification model and the second identification model.

[0013] Optionally, the methods for obtaining the numerical characteristics of the standardized logging curves include:

[0014] Extract the logging curves of the target layer, perform data cleaning to remove outliers, and standardize the curves of each well.

[0015] Optionally, the methods for obtaining the correlation results and preferred logging curves include:

[0016] Using analysis of variance and mutual information analysis, correlation analysis was performed on the numerical characteristics of the standardized logging curves of the core wells and their corresponding microfacies types. The analysis of variance was used to measure whether there was a significant difference in the mean of a logging curve under a certain microfacies. The mutual information analysis was used to measure the nonlinear dependence of each logging curve on the microfacies label.

[0017] Well logging curves with a correlation coefficient greater than a set correlation coefficient threshold are selected as curves to participate in the training of the joint model.

[0018] Optionally, the methods for obtaining the morphological characteristics of the well logging curve include:

[0019] By calling the defined complex wavelet basis function, the frequency and signal characteristics of the logging curve are analyzed, and the frequency and signal characteristics are converted into image features to obtain the morphological characteristics of the logging curve.

[0020] Optionally, the method for obtaining the microphase vertical probability distribution characteristics includes:

[0021] Referring to the Wolseau phase law criterion, analyze the microphase types above or below the current microphase type, count the number N of each specific microphase type above or below, and the total number M. Use N / M to calculate the probability distribution relationship of different microphases in the vertical direction, and obtain the vertical probability distribution characteristics of microphases.

[0022] Optionally, the optimal microphase distribution sequence and microphase transition probability matrix can be obtained in the following ways:

[0023] Define an initial state transition matrix, an observation probability matrix, and an initial state vector π to construct a hidden Markov model. The initial state transition matrix represents the transition probability between microphases in the initial state, the observation probability matrix represents the probability of observing a specific microphase type under the current microphase type, and the initial state vector π represents the most likely microphase type in the initial state.

[0024] The state transition matrix, observation probability matrix, and initial state vector π between states in the model are defined using the characteristics of the microphase vertical probability distribution.

[0025] Each microfacies type is defined as a hidden state, and the numerical features of the logging curves selected after correlation analysis are used as the observation sequence. The three core parameters of the hidden Markov model are estimated based on the observation sequence, and the model is optimized through repeated iterations until convergence or the maximum number of iterations is reached.

[0026] Based on the numerical characteristics of the well logging curves and the trained Hidden Markov Model, the characteristic parameters of the microphase probability distribution are obtained, including the optimal microphase distribution sequence and the microphase transition probability matrix.

[0027] Optionally, the method for obtaining the normalized preferred logging curve numerical characteristics includes:

[0028] The maximum and minimum values ​​of different curves are obtained, and the min-max normalization is used to normalize them so that the response values ​​contained in each logging curve can be mapped to the normalized interval.

[0029] Optionally, the training process of the first recognition model includes:

[0030] The optimal micro-phase transition probability matrix is ​​introduced into the extended Transformer model. In the extended Transformer model, a first weighted layer is added sequentially after the multi-head attention mechanism layer at the encoder end, and the transition probability matrix is ​​passed into the first weighted layer. The data after the attention mechanism is processed is weighted by a linear layer with the micro-phase transition probability matrix to add probability constraints to the extended Transformer model.

[0031] A second weighted layer is added sequentially after the masked self-attention mechanism layer at the decoder end, and the optimal transition probability matrix is ​​passed into the second weighted layer. By utilizing the information masking and extraction capabilities of the masked self-attention mechanism, the extended Transformer model's attention to important geological probability information is enhanced.

[0032] Optionally, the training process of constraining the model parameters during training using the total loss function includes:

[0033] In the joint training framework, the weighted sum of the first loss and the second loss output by the two models in parallel is used as the total loss, and the convergence condition of the total loss is used as the standard for training completion; wherein, the convergence condition is that the validation set does not significantly decrease or fall below a set threshold for k consecutive rounds, and k is a natural number.

[0034] Optionally, the process of obtaining the micro-phase type with the highest probability after weighting the first identification model and the second identification model includes:

[0035] A linear ensemble is performed on the probability distribution of the first micro-phase category output by the first recognition model and the probability distribution of the second micro-phase category output by the second recognition model. The micro-phase with the highest probability is output through the argmax function as the final result of the joint model.

[0036] Optionally, after determining the sedimentary microfacies type identification result corresponding to the well logging curve data to be predicted based on the output result of the joint model, the method further includes:

[0037] The lithofacies assemblage sequence data within the microfacies identified in the sedimentary microfacies type identification results are statistically analyzed to determine whether they are consistent with known lithofacies assemblage sequences;

[0038] If so, statistically analyze the vertical probability characteristics of the microfacies in the sedimentary microfacies type identification results, and determine whether the probability characteristics are consistent with the known vertical probability distribution of microfacies;

[0039] If the probability characteristics are consistent with the known vertical probability distribution of microfacies, the identification result of the sedimentary microfacies type is confirmed to be valid.

[0040] This application also provides a sedimentary microfacies type identification system, including:

[0041] The feature acquisition module is used to acquire the microfacies type of a specific sedimentary environment, the lithofacies sequence characteristics, well logging curve morphology characteristics, well logging curve numerical characteristics, and microfacies vertical probability distribution characteristics corresponding to each microfacies type;

[0042] The feature preprocessing module is used to preprocess the numerical features of the logging curves to obtain standardized numerical features of the logging curves.

[0043] The correlation analysis module is used to perform correlation analysis between the numerical characteristics of the standardized logging curve and the microfacies type to obtain the preferred logging curve;

[0044] The model training module is used to train a hidden Markov model on the vertical probability distribution features of the microphase to obtain the optimal microphase distribution sequence and microphase transition probability matrix.

[0045] The normalization processing module is used to normalize the preferred logging curves to obtain the normalized numerical characteristics of the preferred logging curves.

[0046] The tag library construction module is used to construct a first microfacies tag library based on the lithofacies sequence characteristics corresponding to the microfacies type of the core well, the logging curve morphology characteristics, and the optimal microfacies distribution sequence, and to construct a second microfacies tag library based on the microfacies type of the core well and the corresponding normalized optimal logging curve numerical characteristics.

[0047] The joint model construction module is used to construct a joint model containing a first recognition model and a second recognition model, train the first recognition model and the second recognition model simultaneously, and constrain the model parameters during the training process using the total loss function; wherein, the first recognition model is trained based on the first micro-phase label library using a Transformer model as the base model; the second recognition model is trained based on the second micro-phase label library using an XGBoost model as the base model.

[0048] The identification module is used to input the numerical characteristics, morphological characteristics, and lithofacies sequence characteristics of the logging curve to be predicted from the non-cored well into the first identification model and the second identification model in the joint model, respectively, and determine the sedimentary microfacies type identification result corresponding to the logging curve to be predicted from the non-cored well based on the output result of the joint model; the output result is the microfacies type with the highest probability after weighting the first identification model and the second identification model.

[0049] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0050] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described above when it invokes the computer program in the memory.

[0051] This application provides a method for identifying sedimentary microfacies types, comprising: acquiring microfacies types of a specific sedimentary environment, and corresponding lithofacies sequence characteristics, well logging curve morphology characteristics, well logging curve numerical characteristics, and microfacies vertical probability distribution characteristics for each microfacies type; preprocessing the well logging curve numerical characteristics to obtain standardized well logging curve numerical characteristics; performing correlation analysis between the standardized well logging curve numerical characteristics and the microfacies type to obtain preferred well logging curves; training a hidden Markov model on the microfacies vertical probability distribution characteristics to obtain an optimal microfacies distribution sequence and microfacies transition probability matrix; normalizing the preferred well logging curves to obtain normalized preferred well logging curve numerical characteristics; constructing a first microfacies tag library based on the lithofacies sequence characteristics, well logging curve morphology characteristics, and optimal microfacies distribution sequence corresponding to the microfacies type of the core well; and using the microfacies type of the core well and its corresponding normalized preferred well logging curve... A second microfacies label library is constructed based on the numerical features of the logging curves. A joint model containing a first identification model and a second identification model is constructed, and the first identification model and the second identification model are trained simultaneously. The model parameters during the training process are constrained by the total loss function. The first identification model is trained based on the first microfacies label library using a Transformer model as the base model. The second identification model is trained based on the second microfacies label library using an XGBoost model as the base model. The numerical features, morphological features, and lithofacies sequence features of the logging curves to be predicted from the non-cored wells are input into the first identification model and the second identification model in the joint model, respectively. The sedimentary microfacies type identification result corresponding to the logging curves to be predicted from the non-cored wells is determined based on the output result of the joint model. The output result is the microfacies type with the highest probability after weighting the first identification model and the second identification model.

[0052] This application significantly improves the accuracy and reliability of sedimentary microfacies type identification by introducing multi-source information coupling and a dual-model collaborative mechanism. By simultaneously incorporating lithofacies sequences, curve morphology, numerical characteristics, and vertical probability distributions into the training, the model can capture subtle differences in microfacies in the vertical direction, avoiding the information gaps caused by single curves in previous methods. During the training of the Hidden Markov Model, the inheritance and transfer patterns of microfacies in the vertical direction are fully explored. The generated optimal distribution sequence and transition probability matrix provide the model with vertical constraints consistent with sedimentological logic, significantly reducing the "facies jump" artifact. A first microfacies label library is constructed using lithofacies, curve morphology, and vertical distribution from core wells, while a second microfacies label library is constructed using normalized optimized curve values. This gives the labels both genetic interpretation and numerical discrimination advantages, providing high-quality training samples for the joint model. The joint model uses Transformer to capture long-order dependencies and XGBoost to enhance local nonlinearities. Both are trained simultaneously and share parameter update directions under the constraint of the total loss function, achieving complementarity between global and local information and avoiding the local extrema that single models are prone to. Ultimately, by inputting the characteristics of the logging curves to be predicted from non-cored wells, the microfacies type with the highest confidence level can be output through weighted probability, providing rapid and reliable single-well microfacies interpretation results for subsequent studies on sedimentary facies distribution. It is evident that this application, by constructing an intelligent interpretation model constrained by geological laws, achieves a refined interpretation of sedimentary microfacies, effectively improving the interpretation accuracy of complex microfacies and providing a solid and reliable foundation for subsequent geological research. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0054] Figure 1 A flowchart illustrating a method for identifying sedimentary microphase types provided in this application embodiment;

[0055] Figure 2 This is an example diagram of the complex Morlet wavelet transform of the logging curve provided in the embodiments of this application;

[0056] Figure 3 This is a statistical diagram of the probability characteristics of microphase distribution provided in the embodiments of this application;

[0057] Figure 4 This is a correlation analysis diagram of the variance of logging curves provided in the embodiments of this application;

[0058] Figure 5This is a correlation analysis diagram of the mutual information of logging curves provided in the embodiments of this application;

[0059] Figure 6 A flowchart for establishing the microphase probability distribution characteristic parameters provided in the embodiments of this application;

[0060] Figure 7 Flowchart for establishing the microphase sample label library provided in the embodiments of this application;

[0061] Figure 8 This is a diagram of the extended Transformer model architecture provided in the embodiments of this application;

[0062] Figure 9 This is a diagram of the XGBoost model architecture provided in the embodiments of this application;

[0063] Figure 10 This is a flowchart of the XGBoost+Extended Transformer joint model training provided in the embodiments of this application;

[0064] Figure 11 This is a diagram showing the results of single-well sedimentary microfacies identification provided in an embodiment of this application. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] See Figure 1 , Figure 1 A flowchart illustrating a method for identifying sedimentary microfacies types provided in this application embodiment, the method comprising:

[0067] S101: Obtain the microfacies types of a specific sedimentary environment, and the lithofacies sequence characteristics, well logging curve morphology characteristics, well logging curve numerical characteristics, and microfacies vertical probability distribution characteristics corresponding to each microfacies type;

[0068] S102: Preprocess the numerical characteristics of the well logging curves to obtain standardized numerical characteristics of the well logging curves;

[0069] S103: Perform correlation analysis between the numerical characteristics of the standardized logging curve and the microfacies type to obtain the preferred logging curve;

[0070] S104: Train a hidden Markov model on the vertical probability distribution features of the microphase to obtain the optimal microphase distribution sequence and microphase transition probability matrix;

[0071] S105: Normalize the preferred logging curves to obtain the normalized preferred logging curve numerical characteristics;

[0072] S106: Construct a first microfacies tag library based on the lithofacies sequence characteristics corresponding to the microfacies type of the core well, the logging curve morphology characteristics, and the optimal microfacies distribution sequence; construct a second microfacies tag library based on the microfacies type of the core well and the corresponding normalized optimal logging curve numerical characteristics.

[0073] S107: Construct a joint model including a first recognition model and a second recognition model, train the first recognition model and the second recognition model simultaneously, and use the total loss function to constrain the model parameters during the training process; wherein, the first recognition model is a Transformer model as the base model, trained based on the first micro-phase label library; the second recognition model is an XGBoost model as the base model, trained based on the second micro-phase label library;

[0074] S108: Input the numerical characteristics, morphological characteristics, and lithofacies sequence characteristics of the logging curve to be predicted from the non-cored well into the first identification model and the second identification model in the joint model, respectively. Determine the sedimentary microfacies type identification result corresponding to the logging curve to be predicted from the non-cored well based on the output result of the joint model. The output result is the microfacies type with the highest probability after weighting the first identification model and the second identification model.

[0075] First, several characteristics need to be obtained. Based on the sedimentary background of the study area, the sedimentary facies zones and microfacies types developed in the study area can be determined. Taking the deltaic sedimentary environment as an example, meandering river deltas and braided river deltas are developed, which can be further subdivided into microfacies types such as braided channels, mid-channel bars, interdistributary bays, crevasse fans, underwater distributary channels, mouth bars, and overflow sands.

[0076] Based on the core description data and lithofacies logging interpretation data of the core well, the microfacies types of a specific sedimentary environment and the lithofacies sequence characteristics corresponding to each microfacies type can be obtained. These are used to determine the microfacies type and its internal lithofacies sequence characteristics corresponding to the complete core well section. The specific lithofacies types, microfacies types and their corresponding lithofacies sequences are shown in Tables 1 and 2. Table 1 is a lithofacies code explanation table, and Table 2 is a microfacies type and its lithofacies sequence number table.

[0077] Table 1. Explanation of Lithofacies Codes

[0078]

[0079] Table 2. Microfacies types and their lithofacies sequence numbering table

[0080]

[0081] Furthermore, it is necessary to analyze and obtain the morphological characteristics of logging curves for different microfacies. Specifically, logging curves corresponding to different microfacies in the core section can be obtained, and then the frequency and signal characteristics of the logging curves can be analyzed using a set complex wavelet basis function. These frequency and signal characteristics are then converted into image features to obtain the logging curve morphological characteristics. In one feasible implementation, based on the microfacies type determined by the core wells in the study area, the morphological characteristics of the logging curves corresponding to each microfacies type can be analyzed. The frequency and signal characteristics of the logging curves can be analyzed using complex Morlet wavelet technology, and the curve morphological characteristics can be converted into data features and presented in image form to obtain the logging curve morphological characteristics.

[0082] Common logging curve morphologies for sedimentary microfacies include box-shaped, bell-shaped, finger-shaped, and funnel-shaped. For example, the logging curve morphology of the braided channel microfacies in the study area is bell-shaped, while the logging curve morphology of the mid-shoal microfacies is box-shaped. The microfacies logging images converted using complex Morlet wavelet technology are shown below. Figure 2 As shown, Figure 2 This is an example diagram of the complex Morlet wavelet transform of the logging curve provided in the embodiments of this application. Figure 2 Each channel represents a logging curve, which serves as the image data input to the model. The channels are then superimposed to form a curve-shaped data volume.

[0083] Numerical characteristics of well logging curves can be directly measured using well logging instruments.

[0084] For the vertical probability distribution characteristics of microfacies, the Wolseau phase law criterion can be referenced. Stacked bar charts can be used to characterize the probability distribution characteristics of specific microfacies developing above and below the current microfacies in the vertical direction, and then the probability distribution relationship of different microfacies in the vertical direction can be calculated.

[0085] See Figure 3 , Figure 3 This is a statistical diagram of the probability characteristics of microphase distribution provided in an embodiment of this application. Figure 3 The left figure shows the probability distribution of the previous microphase of the current microphase, and the right figure shows the probability distribution of the next microphase of the current microphase. Based on this, the probability distribution relationship of different microphases in the vertical direction can be analyzed. The specific probability calculation method is as follows: Analyze the microphase types above or below the current microphase type, that is, the microphase types immediately above or below the current microphase type. Count the number N of each specific microphase type above or below the current microphase type, that is, the number of the microphase types immediately above or below the current microphase type, and the total number M. The calculation formula is as follows:

[0086] ;

[0087] In step S102, data cleaning is performed on the logging curves of the target layer. Specifically, the logging curves of the target layer from the cored well are extracted, and data cleaning is performed to remove outliers. The curves from each well are first standardized to unify the scale. This involves counting the number N of each specific microfacies type in the upper and lower layers.

[0088] In step S103, a correlation analysis is performed between the numerical characteristics of standardized logging curves and microfacies types. Specifically, analysis of variance and mutual information analysis can be used to perform correlation analysis between the numerical characteristics of standardized logging curves from cored wells and their corresponding microfacies types. Analysis of variance is used to measure whether there is a significant difference in the mean of a logging curve under a certain microfacies. Mutual information analysis is used to measure the nonlinear dependence of each logging curve on the microfacies label. Finally, logging curves with a correlation coefficient greater than a set correlation coefficient threshold are selected as curves to participate in the joint model training.

[0089] See Figure 4 and Figure 5 , Figure 4 This is a correlation analysis diagram of the variance of the logging curves provided in the embodiments of this application. Figure 5 The well logging curve mutual information correlation analysis diagram provided in this application embodiment preferably uses well logging curves with higher correlation coefficients with microfacies as curves participating in model training. Using this method, this embodiment selects natural gamma (GR), spontaneous potential (ΔSP), neutron (CN), and photoelectric absorption interface index (PE) as well logging correlation parameters (e.g., ...). Figure 4 (as shown), which will serve as the training set data for later use.

[0090] In step S104, a hidden Markov model is trained on the vertical probability distribution features of the microphases to obtain the optimal microphase distribution sequence and microphase transition probability matrix. This can be implemented using the following steps:

[0091] The first step is to define the initial state transition matrix, the observation probability matrix, and the initial state vector to construct the Hidden Markov Model. The initial state transition matrix represents the transition probability between microphases in the initial state, the observation probability matrix represents the probability of observing a specific microphase type under the current microphase type, and the initial state vector represents the most likely microphase type in the initial state.

[0092] The second step is to use the micro-phase vertical probability distribution characteristics to define the state transition matrix, observation probability matrix and initial state vector between states in the model;

[0093] The third step is to define each microfacies type as a hidden state, and use the numerical features of the well logging curves selected after correlation analysis as the observation sequence; predict the three core parameters of the hidden Markov model based on the observation sequence, and optimize the model through repeated iterations until convergence or the maximum number of iterations is reached.

[0094] The fourth step involves obtaining the characteristic parameters of the microphase probability distribution based on the numerical characteristics of the well logging curves and the trained hidden Markov model, including the optimal microphase distribution sequence and the microphase transition probability matrix.

[0095] First, construct the initial Hidden Markov Model (HMM): define the initial state transition matrix (A), the observation probability matrix (B), and the initial state vector. See also Figure 6 , Figure 6 A flowchart is provided to establish the microphase probability distribution characteristic parameters provided in the embodiments of this application. The state transition matrix (A) between states in the model is defined using the vertical probability distribution characteristics of the microphase obtained from the analysis, which is used to illustrate the initial transition probability between microphases. The observation probability matrix (B) is defined to represent the probability of observing a specific microphase type under the current microphase type. The initial state vector π is used to indicate which type of microphase is most likely to be initially.

[0096] The model was then trained, with each microfacies type defined as a hidden state. The selected logging curves were used as the observation sequence. The Baum-Welch unsupervised learning algorithm was used to automatically estimate the three core parameters of the initial hidden Markov model based on the observation sequence. The model was then iteratively optimized until convergence or the maximum number of iterations was reached.

[0097] Finally, as Figure 6 As shown, the characteristic parameters of the microphase probability distribution can be extracted using the Viterbi algorithm. Based on the numerical characteristics of the logging curves and the trained Hidden Markov Model, the characteristic parameters of the microphase probability distribution, including the optimal microphase distribution sequence and the microphase transition probability matrix, can be obtained.

[0098] In step S105, the preferred logging curves are normalized to obtain the normalized numerical characteristics of the preferred logging curves. The maximum and minimum values ​​of different curves can be obtained, and min-max normalization is used to normalize them, so that the response values ​​contained in each logging curve are mapped to a normalized interval. The range of this one-to-one normalization interval is not specifically limited and can be set by those skilled in the art. The maximum and minimum values ​​of different curves are obtained respectively, and min-max normalization is used to normalize the different curves, thereby mapping the response values ​​contained in each logging curve to the range [0-1].

[0099] See Figure 7 , Figure 7 The flowchart for establishing the microfacies sample tag library provided in the embodiments of this application is as follows: In step S106, the first microfacies tag library is constructed based on the strata and lithology as boundaries, the lithofacies sequence characteristics corresponding to the microfacies type of the core well, the logging curve morphology characteristics, and the optimal microfacies distribution sequence. The second microfacies tag library is constructed based on the microfacies type of the core well and the corresponding normalized optimal logging curve numerical characteristics.

[0100] Subsequently, a first recognition model and a second recognition model were constructed. For the first microfacies label library of probability + sequence, the Transformer model was first selected as the base model. Due to its core self-attention mechanism, the model can capture the dependencies between any positions in the sequence when processing sequence data, and the multi-head attention mechanism (MHA) also makes it easier for the model to obtain the features of the training data. Therefore, by inputting lithofacies sequence data and microfacies labels, the dependencies of the lithofacies sequences corresponding to the microfacies can be obtained by training the Transformer model. The self-attention mechanism can also better focus on important lithofacies information and reduce the interference of noisy lithofacies. In addition, the self-attention mechanism enables the Transformer to better capture long-term dependencies when processing long sequences, avoiding the gradient vanishing or exploding problems that may occur in RNN or LSTM in long sequence modeling, and solving problems such as model training non-convergence and overfitting. At the same time, the Transformer can handle sequences of different lengths by adjusting the position encoding, which has good flexibility, thus making it more efficient and effective in processing long sequences.

[0101] However, there is still the issue of vertical probability in geology; therefore, in one feasible implementation approach, the Transformer model can be extended. See [link / reference]. Figure 8 , Figure 8 The diagram below shows the architecture of the extended Transformer model provided in this application. The optimal micro-phase transition probability matrix is ​​introduced into the extended Transformer model. After the multi-head attention mechanism layer at the encoder end in the extended Transformer model, a first weighted layer is added sequentially, and the transition probability matrix is ​​passed into the first weighted layer. The data after the attention mechanism is processed is weighted by the linear layer and the micro-phase transition probability matrix to add probability constraints to the extended Transformer model.

[0102] Figure 8In this context, the multi-head attention mechanism divides the input query, key, and value into multiple subspaces, calculates attention weights independently for each subspace, and then concatenates the results and generates the final output through a linear transformation. This allows the model to focus on different aspects of information in the input sequence, thereby capturing richer features.

[0103] The masked multi-head attention mechanism combines multi-head attention and masking mechanisms. The masking mechanism is used to shield parts that do not need to participate in the calculation, ensuring that the model does not see future information during training, thereby maintaining autoregressive properties, avoiding information leakage, and improving the quality of model generation.

[0104] Adding normalization involves introducing residual connections between the input and output of each sublayer and applying layer normalization after the output. This helps mitigate the vanishing and exploding gradient problems while accelerating the training process. Bit-wise feedforward networks are used to perform non-linear transformations on the representation at each location, extracting more complex features.

[0105] A second weighted layer is added sequentially after the masked self-attention mechanism layer at the decoder end, and the optimal transition probability matrix is ​​passed into the second weighted layer. By utilizing the information masking and extraction capabilities of the masked self-attention mechanism, the attention of the extended Transformer model to important geological probability information is strengthened.

[0106] See Figure 9 , Figure 9 The diagram shows the XGBoost model architecture provided in this application embodiment. For the second microfacies label library based on well logging curves, the XGBoost model is used for small sample model prediction. Previous experiments have shown that XGBoost has a high accuracy rate when training small sample models. Therefore, under a single-layer scale framework, well logging curve data is used to intelligently interpret microfacies.

[0107] After obtaining the first and second recognition models, simultaneous joint training of the first and second recognition models can be performed, i.e., XGBoost + Extended Transformer joint model training. During training, the model parameters are constrained using the total loss function. Specifically, in the joint training framework, the weighted sum of the first and second losses output by the two models in parallel is used as the total loss, and the convergence condition of the total loss is used as the standard for training completion. The convergence condition is that the validation set does not significantly decrease or falls below a set threshold for k consecutive rounds, where k is a natural number that can be defined by those skilled in the art.

[0108] One feasible model parameter setting is as follows. Those skilled in the art can also use different model parameters for different study areas based on the exemplary model settings described below, and these should also be within the scope of protection of this application. See also Figure 10 As shown, Figure 10 This is a flowchart illustrating the training process of the XGBoost+ Extended Transformer joint model provided in an embodiment of this application.

[0109] Figure 10 In this example, the Transformer model is extended: the number of Transformer layers is 12; the number of attention heads is 12; the hidden layer dimension is 768; the activation function is SwishGLU; dropout is 0.1; the initial learning rate and weight decay of the SGD optimizer are set to 0.1 and 0.0005 respectively; the learning rate of the AdamW optimizer is set to 0.0003, and the weight decay is set to 0.02; the maximum positional encoding length is 512.

[0110] XGBoost model: base learner type is bgtree; objective function is multi:softprob; number of classes is 7; number of weak learners is 300; learning rate is 0.001; evaluation metric is rmse.

[0111] Training and validation data preparation: The first microphase label library and the second microphase label library are used as training data and validation data, respectively. According to the microphase type, 70% of the data in the label library is randomly selected as the training set and 30% of the data is selected as the validation set.

[0112] As can be seen, in this embodiment, the first and second microfacies label libraries are input into the extended Transformer model and XGBoost, respectively. The XGBoost model utilizes the second microfacies label library, leveraging its advantage in few-shot learning, to interpret microfacies. The extended Transformer model, using the first microfacies label library, identifies the corresponding curve morphology of microfacies and interprets the corresponding lithofacies sequences. Combining this with microfacies probability parameters, it ultimately outputs the most likely next microfacies type under the current microfacies, thus constraining the interpretation results of the XGBoost model. Through continuous iterative updates of model parameters, the number of model layers, neurons, learning rate, and other parameters are optimized. The model is then validated using a validation set, ultimately achieving joint model learning of data features.

[0113] Finally, the model prediction is executed. The numerical characteristics, morphological characteristics, and lithofacies sequence characteristics of the logging curves to be predicted from the non-cored wells are input into the first and second identification models in the joint model, respectively. Based on the output of the joint model, the sedimentary microfacies type corresponding to the logging curves to be predicted from the non-cored wells is determined. It should be emphasized that the output result is the microfacies type with the highest probability after weighting the first and second identification models.

[0114] To ensure the accuracy of the model identification results, model validation can be performed. Specifically, the lithofacies assemblage sequence data within the sedimentary microfacies type identification results can be statistically analyzed to determine whether they are consistent with known lithofacies assemblage sequences.

[0115] If so, analyze the probability characteristics of the vertical microfacies in the sedimentary microfacies type identification results, and determine whether the probability characteristics are consistent with the known vertical probability distribution of microfacies.

[0116] If the probability characteristics are consistent with the known vertical probability distribution of microfacies, the identification results of sedimentary microfacies types can be confirmed as valid.

[0117] In practical applications, geological experts can also evaluate the prediction results to determine whether there are any unreasonable aspects from a geological perspective; ultimately, the effectiveness of the intelligent joint model's interpretation can be determined. (See also...) Figure 11 , Figure 11 This is a diagram showing the results of single-well sedimentary microfacies identification provided in an embodiment of this application. Figure 11 The results of intelligent interpretation of sedimentary microfacies in a single well are shown. The microfacies interpretation column represents manually interpreted sedimentary microfacies, the model prediction column represents the results of intelligent joint model interpretation, and the right side is a legend. There are a total of 7 microfacies types. The first and second columns on the left are depth and sub-layer, respectively. The sub-layer is manually calibrated. The last three columns are the original logging curve columns, including (GR, SP, CAL, etc.). The lithology column is the lithology of the core section, and the single-layer column is a further refinement based on the sub-layer.

[0118] This application employs a Hidden Markov Model to extract key microfacies probability distribution transition matrices and microfacies probability parameters from the vertical probability sequence of microfacies. The extracted geological probability constraints (microfacies transition matrices) are integrated into the Transformer model architecture to construct an extended Transformer model guided by geological laws, thus embedding geological laws into the intelligent interpretation process. The XGBoost model's efficient learning advantage under small sample conditions is fully utilized, effectively merging it with the aforementioned extended Transformer model to construct a joint intelligent interpretation model, thereby forming an intelligent sedimentary microfacies interpretation model system that integrates the advantages of geological probability constraints, deep learning, and small sample learning. In this process, by fusing multi-source geological information, the reliability and spatial resolution of sedimentary microfacies identification are significantly improved. The Transformer model, leveraging lithofacies sequences, well logging morphologies, and probabilistic features from the first microfacies tag library, can capture long-distance vertical correlations, providing a more refined characterization of complex superimposed microfacies. The XGBoost model, guided by the second microfacies tag library, utilizes the high correlation between well logging curves and microfacies to quickly provide robust probability estimates. In the two-level model coupling process, the prediction of XGBoost is set as a constraint condition, which not only suppresses the false details that the Transformer may produce, but also retains its high-resolution advantage. By comprehensively analyzing the lithofacies assemblage sequence within microfacies, the probabilistic development characteristics between microfacies, and the well logging response characteristics of different lithofacies and microfacies themselves, an intelligent interpretation model constrained by geological laws is constructed to achieve a fine interpretation of sedimentary microfacies.

[0119] This application also provides a sedimentary microfacies type identification system, including:

[0120] The feature acquisition module is used to acquire the microfacies type of a specific sedimentary environment, the lithofacies sequence characteristics, well logging curve morphology characteristics, well logging curve numerical characteristics, and microfacies vertical probability distribution characteristics corresponding to each microfacies type;

[0121] The feature preprocessing module is used to preprocess the numerical features of the logging curves to obtain standardized numerical features of the logging curves.

[0122] The correlation analysis module is used to perform correlation analysis between the numerical characteristics of the standardized logging curve and the microfacies type to obtain the preferred logging curve;

[0123] The model training module is used to train a hidden Markov model on the vertical probability distribution features of the microphase to obtain the optimal microphase distribution sequence and microphase transition probability matrix.

[0124] The normalization processing module is used to normalize the preferred logging curves to obtain the normalized numerical characteristics of the preferred logging curves.

[0125] The tag library construction module is used to construct a first microfacies tag library based on the lithofacies sequence characteristics corresponding to the microfacies type of the core well, the logging curve morphology characteristics, and the optimal microfacies distribution sequence, and to construct a second microfacies tag library based on the microfacies type of the core well and the corresponding normalized optimal logging curve numerical characteristics.

[0126] The joint model construction module is used to construct a joint model containing a first recognition model and a second recognition model, train the first recognition model and the second recognition model simultaneously, and constrain the model parameters during the training process using the total loss function; wherein, the first recognition model is trained based on the first micro-phase label library using a Transformer model as the base model; the second recognition model is trained based on the second micro-phase label library using an XGBoost model as the base model.

[0127] The identification module is used to input the numerical characteristics, morphological characteristics, and lithofacies sequence characteristics of the logging curve to be predicted from the non-cored well into the first identification model and the second identification model in the joint model, respectively, and determine the sedimentary microfacies type identification result corresponding to the logging curve to be predicted from the non-cored well based on the output result of the joint model; the output result is the microfacies type with the highest probability after weighting the first identification model and the second identification model.

[0128] This application also provides an embodiment corresponding to a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in the above method embodiments.

[0129] It is understood that if the methods in the above embodiments are implemented as software functional units and sold or used as independent products, they 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 all or 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 executes 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.

[0130] The computer-readable storage medium provided in this embodiment includes the method mentioned above, and has the same effect.

[0131] This application also provides an electronic device that may include a processor and a memory, wherein the computer-readable storage medium stored in the memory includes the methods mentioned above, with the same effect.

[0132] The processor may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor can be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor may also include a main processor and coprocessors. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0133] The memory may include one or more computer-readable storage media, which may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory is used to store at least the following computer program, which, after being loaded and executed by a processor, is capable of implementing the relevant steps in the method executed by the electronic device side as disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory may also include an operating system and data, and the storage method may be temporary or permanent storage. The operating system may include Windows, Linux, Android, etc.

[0134] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. As the system provided in the embodiments corresponds to the method provided in the embodiments, the description is relatively simple; relevant parts can be found in the method section.

[0135] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

[0136] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for identifying sedimentary microfacies types, characterized in that, include: Obtain the microfacies types of a specific sedimentary environment, and the lithofacies sequence characteristics, well logging curve morphology characteristics, well logging curve numerical characteristics, and microfacies vertical probability distribution characteristics corresponding to each microfacies type; The numerical characteristics of the well logging curves are preprocessed to obtain standardized numerical characteristics of the well logging curves; Correlation analysis was performed on the numerical characteristics of the standardized logging curves and the microfacies type to obtain the preferred logging curves; The microphase vertical probability distribution characteristics are trained using a hidden Markov model to obtain the optimal microphase distribution sequence and microphase transition probability matrix; The preferred logging curves are normalized to obtain the normalized numerical characteristics of the preferred logging curves. A first microfacies tag library is constructed based on the lithofacies sequence characteristics corresponding to the microfacies type of the core well, the logging curve morphology characteristics, and the optimal microfacies distribution sequence; a second microfacies tag library is constructed based on the microfacies type of the core well and the corresponding normalized optimal logging curve numerical characteristics. A joint model comprising a first recognition model and a second recognition model is constructed, and the first recognition model and the second recognition model are trained synchronously, with the model parameters constrained during the training process using the total loss function; wherein, the first recognition model is trained based on the first micro-phase label library using the Transformer model as the base model; the second recognition model is trained based on the second micro-phase label library using the XGBoost model as the base model. The numerical characteristics, morphological characteristics, and lithofacies sequence characteristics of the logging curves to be predicted from the non-cored wells are input into the first identification model and the second identification model in the joint model, respectively. The sedimentary microfacies type identification result corresponding to the logging curves to be predicted from the non-cored wells is determined based on the output result of the joint model. The output result is the microfacies type with the highest probability after weighting the first identification model and the second identification model.

2. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The methods for obtaining the numerical characteristics of the standardized logging curves include: Extract the logging curves of the target layer, perform data cleaning to remove outliers, and standardize the curves of each well.

3. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The methods for obtaining the correlation results and preferred logging curves include: Using analysis of variance and mutual information analysis, correlation analysis was performed on the numerical characteristics of the standardized logging curves of the core wells and their corresponding microfacies types. The analysis of variance was used to measure whether there was a significant difference in the mean of a logging curve under a certain microfacies. The mutual information analysis was used to measure the nonlinear dependence of each logging curve on the microfacies label. Well logging curves with a correlation coefficient greater than a set correlation coefficient threshold are selected as curves to participate in the training of the joint model.

4. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The methods for obtaining the morphological characteristics of the well logging curves include: By calling the defined complex wavelet basis function, the frequency and signal characteristics of the logging curve are analyzed, and the frequency and signal characteristics are converted into image features to obtain the morphological characteristics of the logging curve.

5. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The methods for obtaining the vertical probability distribution characteristics of the microphase include: Referring to the Wolseau phase law criterion, analyze the microphase types above or below the current microphase type, count the number N of each specific microphase type above or below, and the total number M. Use N / M to calculate the probability distribution relationship of different microphases in the vertical direction, and obtain the vertical probability distribution characteristics of microphases.

6. The method for identifying sedimentary microfacies types according to claim 5, characterized in that, The methods for obtaining the optimal microphase distribution sequence and microphase transition probability matrix include: Define an initial state transition matrix, an observation probability matrix, and an initial state vector π to construct a hidden Markov model. The initial state transition matrix represents the transition probability between microphases in the initial state, the observation probability matrix represents the probability of observing a specific microphase type under the current microphase type, and the initial state vector π represents the most likely microphase type in the initial state. The state transition matrix, observation probability matrix, and initial state vector π between states in the model are defined using the characteristics of the microphase vertical probability distribution. Each microfacies type is defined as a hidden state, and the numerical features of the logging curves selected after correlation analysis are used as the observation sequence. The three core parameters of the hidden Markov model are estimated based on the observation sequence, and the model is optimized through repeated iterations until convergence or the maximum number of iterations is reached. Based on the numerical characteristics of the well logging curves and the trained Hidden Markov Model, the characteristic parameters of the microphase probability distribution are obtained, including the optimal microphase distribution sequence and the microphase transition probability matrix.

7. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The methods for obtaining the normalized preferred logging curve numerical characteristics include: The maximum and minimum values ​​of different curves are obtained, and the min-max normalization is used to normalize them so that the response values ​​contained in each logging curve can be mapped to the normalized interval.

8. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The training process of the first recognition model includes: The optimal micro-phase transition probability matrix is ​​introduced into the extended Transformer model. In the extended Transformer model, a first weighted layer is added sequentially after the multi-head attention mechanism layer at the encoder end, and the transition probability matrix is ​​passed into the first weighted layer. The data after the attention mechanism is processed is weighted by a linear layer with the micro-phase transition probability matrix to add probability constraints to the extended Transformer model. A second weighting layer is sequentially added after the masking self-attention mechanism layer at the decoder end, and the optimal transition probability matrix is ​​passed into the second weighting layer.

9. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The training process that uses the total loss function to constrain the model parameters during training includes: In the joint training framework, the weighted sum of the first loss and the second loss output by the two models in parallel is used as the total loss, and the convergence condition of the total loss is used as the standard for training completion; wherein, the convergence condition is that the validation set does not significantly decrease or fall below a set threshold for k consecutive rounds, and k is a natural number.

10. The method for identifying sedimentary microfacies types according to claim 1, characterized in that, The process of obtaining the micro-phase type with the highest probability after weighting the first and second identification models includes: A linear ensemble is performed on the probability distribution of the first micro-phase category output by the first recognition model and the probability distribution of the second micro-phase category output by the second recognition model. The micro-phase with the highest probability is output through the argmax function as the final result of the joint model.

11. The method for identifying sedimentary microfacies types according to claim 1 or 10, characterized in that, After determining the sedimentary microfacies type identification result corresponding to the well logging curve data to be predicted based on the output of the joint model, the following is also included: The lithofacies assemblage sequence data within the microfacies identified in the sedimentary microfacies type identification results are statistically analyzed to determine whether they are consistent with known lithofacies assemblage sequences; If so, statistically analyze the vertical probability characteristics of the microfacies in the sedimentary microfacies type identification results, and determine whether the probability characteristics are consistent with the known vertical probability distribution of microfacies; If the probability characteristics are consistent with the known vertical probability distribution of microfacies, the identification result of the sedimentary microfacies type is confirmed to be valid.

12. A sedimentary microfacies type identification system, characterized in that, include: The feature acquisition module is used to acquire the microfacies type of a specific sedimentary environment, the lithofacies sequence characteristics, well logging curve morphology characteristics, well logging curve numerical characteristics, and microfacies vertical probability distribution characteristics corresponding to each microfacies type; The feature preprocessing module is used to preprocess the numerical features of the logging curves to obtain standardized numerical features of the logging curves. The correlation analysis module is used to perform correlation analysis between the numerical characteristics of the standardized logging curve and the microfacies type to obtain the preferred logging curve; The model training module is used to train a hidden Markov model on the vertical probability distribution features of the microphase to obtain the optimal microphase distribution sequence and microphase transition probability matrix. The normalization processing module is used to normalize the preferred logging curves to obtain the normalized numerical characteristics of the preferred logging curves. The tag library construction module is used to construct a first microfacies tag library based on the lithofacies sequence characteristics corresponding to the microfacies type of the core well, the logging curve morphology characteristics, and the optimal microfacies distribution sequence, and to construct a second microfacies tag library based on the microfacies type of the core well and the corresponding normalized optimal logging curve numerical characteristics. The joint model construction module is used to construct a joint model containing a first recognition model and a second recognition model, train the first recognition model and the second recognition model simultaneously, and constrain the model parameters during the training process using the total loss function; wherein, the first recognition model is trained based on the first micro-phase label library using a Transformer model as the base model; the second recognition model is trained based on the second micro-phase label library using an XGBoost model as the base model. The identification module is used to input the numerical characteristics, morphological characteristics, and lithofacies sequence characteristics of the logging curve to be predicted from the non-cored well into the first identification model and the second identification model in the joint model, respectively, and determine the sedimentary microfacies type identification result corresponding to the logging curve to be predicted from the non-cored well based on the output result of the joint model; the output result is the microfacies type with the highest probability after weighting the first identification model and the second identification model.

13. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the method as described in any one of claims 1 to 11.

Images