A method, apparatus, equipment, medium, and product for predicting reservoir facies.

Abstract

This disclosure relates to the field of deep learning technology, and particularly to a method, apparatus, device, medium, and product for reservoir lithofacies prediction. The method includes: acquiring pre-stack seismic wave data and lithofacies annotation data; loading the pre-stack seismic wave data into the memory of a data processing device; obtaining a three-dimensional array composed of the pre-stack seismic wave data by traversing the data in the data processing device; generating a reservoir prediction model based on the three-dimensional array and lithofacies annotation data; wherein the reservoir prediction model is generated by processing the three-dimensional array and lithofacies annotation data using the HR-Net algorithm; predicting the reservoir distribution in an unknown exploration area using the reservoir prediction model, and obtaining the prediction result; wherein the prediction result is a 3D data model in Segy format. This disclosure can perform efficient and high-precision lithofacies prediction, which helps in formulating more reasonable exploration plans and risk response strategies.

Description

Technical Field

[0001] This disclosure relates to the field of deep learning technology, and in particular to a method, apparatus, equipment, medium, and product for predicting reservoir facies. Background Technology

[0002] In recent years, the field of petroleum exploration and development has effectively utilized modern science and technology, thereby promoting the rapid development of the petroleum industry. In the process of petroleum exploration and development, the identification and definition of lithofacies is of great significance. It can not only analyze the microfacies and their spatiotemporal evolution in lithofacies and establish sedimentary models, but also analyze the reservoir, caprock and other hydrocarbon accumulation elements and combinations to establish hydrocarbon accumulation models, thereby providing a basis for determining geological reserves, predicting oil and gas-bearing areas and well location deployment.

[0003] In related technologies, traditional manual lithofacies analysis methods are commonly used. However, these methods suffer from high subjectivity, are time-consuming and labor-intensive, and result in low accuracy in predicted lithofacies. Therefore, achieving efficient and high-precision lithofacies prediction has become an urgent problem to be solved. Summary of the Invention

[0004] This disclosure provides a method, apparatus, equipment, medium, and product for predicting reservoir facies, enabling efficient and high-precision prediction of facies.

[0005] Firstly, this disclosure provides a method for predicting reservoir lithology, including:

[0006] Acquire prestack seismic wave data and lithofacies annotation data, and load the prestack seismic wave data into the memory of the data processing device. By traversing the prestack seismic wave data in the data processing device, obtain a three-dimensional array composed of the prestack seismic wave data.

[0007] A reservoir prediction model is generated based on the three-dimensional array and the lithofacies annotation data; wherein, the reservoir prediction model is generated by processing the three-dimensional array and the lithofacies annotation data using the HR-Net algorithm;

[0008] The reservoir distribution in an unknown exploration area is predicted using the reservoir prediction model, and the prediction results are obtained; wherein the prediction results are 3D data models in Segy format.

[0009] In some embodiments, loading the pre-stack seismic wave data into the memory of a data processing device and obtaining a three-dimensional array composed of the pre-stack seismic wave data by traversing the pre-stack seismic wave data in the data processing device includes:

[0010] Wellside data of the pre-stack seismic wave data is extracted from the well logging curves, wherein the wellside data corresponds to the depth of the well logging lithofacies data;

[0011] Based on the wellbore data and the lithofacies annotation data, lithofacies interpretation results and conditional attributes are obtained;

[0012] The three-dimensional array is determined based on the lithofacies interpretation results and the conditional attributes.

[0013] In some embodiments, generating a reservoir prediction model based on the three-dimensional array and the lithofacies annotation data includes:

[0014] The model training set is obtained based on the three-dimensional array; wherein, the model training set includes a calibration set and a cross-validation set;

[0015] The reservoir prediction model is obtained by processing the model training set using the HR-Net algorithm; wherein the reservoir prediction model is used to predict the reservoir distribution in the unknown exploration area.

[0016] In some embodiments, processing the model training set using the HR-Net algorithm to obtain the reservoir prediction model includes:

[0017] Based on the data size and data structure of the model training set, an HR-Net algorithm model is generated; wherein, the HR-Net algorithm model includes a loss function;

[0018] The loss function is defined as a mixture of cross-entropy and BCE functions with different weights. The learning rate is optimized using the cosine annealing algorithm, and an adaptive gradient descent optimizer is used.

[0019] The model training set is processed based on the loss function to obtain the reservoir prediction model.

[0020] In some embodiments, predicting the reservoir distribution in an unknown exploration area using the reservoir prediction model and obtaining the prediction results includes:

[0021] The reservoir prediction model is evaluated based on the condition attributes required by the reservoir prediction model, and the evaluation result of the reservoir prediction model is obtained.

[0022] Based on the evaluation results, the target parameters in the HR-Net algorithm are determined;

[0023] The reservoir prediction model is processed according to the target parameters to obtain the target optimized model of the reservoir prediction model;

[0024] Based on the target optimization model, the reservoir distribution in the unknown exploration area is predicted, and the prediction results are obtained.

[0025] In some embodiments, obtaining the target optimization model based on the target parameters includes:

[0026] By comparing the lithofacies predicted for the cross-validation set in the target parameters, a confusion matrix is ​​calculated for supervised learning of the lithofacies identification method;

[0027] The lithofacies transition matrix characterizing the changes between continuous facies is calculated based on the confusion matrix;

[0028] The target optimization model of the reservoir prediction model is determined based on the lithofacies transition matrix and the confusion matrix.

[0029] Secondly, this disclosure provides a reservoir facies prediction device, comprising:

[0030] The acquisition module is used to acquire prestack seismic wave data and lithofacies annotation data, and load the prestack seismic wave data into the memory of the data processing device. By traversing the prestack seismic wave data in the data processing device, a three-dimensional array composed of the prestack seismic wave data is obtained.

[0031] A generation module is used to generate a reservoir prediction model based on the three-dimensional array and the lithofacies annotation data; wherein, the reservoir prediction model is generated by processing the three-dimensional array and the lithofacies annotation data using the HR-Net algorithm;

[0032] The prediction module is used to predict the reservoir distribution in an unknown exploration area using the reservoir prediction model and obtain prediction results; wherein the prediction results are 3D data models in Segy format.

[0033] Thirdly, this disclosure provides a computer device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the foregoing aspects.

[0034] Fourthly, this disclosure provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of the methods described in the above aspects.

[0035] Fifthly, this disclosure provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the methods described in the foregoing aspects.

[0036] This disclosure provides a reservoir lithofacies prediction method, apparatus, equipment, medium, and product. First, pre-stack seismic wave data and lithofacies annotation data are acquired, and the pre-stack seismic wave data is loaded into the memory of a data processing device. A three-dimensional array composed of the pre-stack seismic wave data is obtained by traversing the data in the data processing device. Then, a reservoir prediction model is generated based on the three-dimensional array and lithofacies annotation data. The reservoir prediction model is generated by processing the three-dimensional array and lithofacies annotation data using the HR-Net algorithm. Finally, the reservoir distribution in an unknown exploration area is predicted using the reservoir prediction model, and the prediction result is obtained. The prediction result is a 3D data model in Segy format.

[0037] As described above, the technical solution disclosed herein generates a reservoir prediction model using a three-dimensional array and lithofacies annotation data. This reservoir prediction model is generated by processing the three-dimensional array and lithofacies annotation data using the HR-Net algorithm. Compared with the traditional method of manually analyzing lithofacies, using the HR-Net algorithm to process the three-dimensional array and lithofacies annotation data can effectively capture subtle geological information, thereby more accurately classifying lithofacies and providing a more reliable basis for reservoir prediction, thus improving the efficiency and accuracy of lithofacies prediction.

[0038] 1. Acquire pre-stack seismic wave data and lithofacies annotation data. Load the pre-stack seismic wave data into the memory of the data processing device. By traversing the pre-stack seismic wave data in the data processing device, obtain a three-dimensional array composed of the pre-stack seismic wave data. This provides the most original and crucial information source for the entire reservoir lithology interpretation system. The pre-stack seismic wave data contains various information about the propagation process of seismic waves underground, which can reflect rich details of the subsurface geological structure and lithological characteristics. The lithofacies annotation data provides an accurate reference standard for subsequent model training, enabling the model to learn the performance characteristics of different lithofacies in the data, thereby improving the accuracy of lithofacies prediction.

[0039] 2. Reservoir prediction models generated based on three-dimensional arrays and lithofacies annotation data can effectively capture subtle geological information, thereby more accurately classifying lithofacies and providing a more reliable basis for reservoir prediction, improving the efficiency and accuracy of lithofacies prediction.

[0040] 3. By using reservoir prediction models to predict the reservoir distribution in unknown exploration areas, the uncertainty in the exploration process can be reduced. Reservoir prediction models can provide quantitative and qualitative information about reservoir distribution, such as reservoir thickness and lithological distribution, giving explorers a clearer understanding of the subsurface geological conditions. In a complex, unknown area, traditional methods provide almost no information about the reservoir situation, while model predictions can indicate the probability and general characteristics of reservoirs in different locations, reducing uncertainty by 40%-60%. This helps in developing more reasonable exploration plans and risk management strategies. Attached Figure Description

[0041] The present disclosure will be described in more detail below based on embodiments and with reference to the accompanying drawings:

[0042] Figure 1 A schematic flowchart illustrating a reservoir facies prediction method provided in this embodiment of the disclosure;

[0043] Figure 2 A flowchart of the HR-Net algorithm model provided in this embodiment of the disclosure;

[0044] Figure 3 A schematic diagram of training results provided in an embodiment of this disclosure;

[0045] Figure 4 This is a schematic diagram of the prediction results provided in the embodiments of this disclosure;

[0046] Figure 5 This is a schematic diagram of a data entry interface provided in an embodiment of the present disclosure;

[0047] Figure 6 This is a schematic diagram of the model training interface provided in an embodiment of the present disclosure;

[0048] Figure 7 A schematic block diagram of the functional modules of a reservoir facies prediction device provided for an exemplary embodiment of the present disclosure;

[0049] Figure 8 A structural block diagram of an electronic device provided as an exemplary embodiment of this disclosure;

[0050] Figure 9 A structural block diagram of a computer system provided as an exemplary embodiment of this disclosure;

[0051] Figure 10 A structural block diagram of a computer program product provided for an exemplary embodiment of this disclosure.

[0052] In the accompanying drawings, the same parts are referred to by the same reference numerals, and the drawings are not drawn to scale. Detailed Implementation

[0053] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0054] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0055] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0056] In the process of oil exploration and development, the identification and definition of lithofacies is of great significance. It can not only analyze the microfacies and their spatiotemporal evolution in lithofacies and establish sedimentary models, but also analyze the reservoir, caprock and other hydrocarbon accumulation elements and combinations to establish hydrocarbon accumulation models. This allows for further exploration of the relationship between hydrocarbon accumulation and sedimentary microfacies, interpretation of the distribution morphology of known sand bodies, guidance on sand body prediction in uncontrolled areas, and provides a basis for determining geological reserves, predicting oil and gas-bearing areas and well location deployment.

[0057] Lithofacies analysis is a crucial step in interpreting seismic data for reservoir characterization. It plays a vital role in initial exploration prospect assessment, reservoir characterization, and ultimately, oilfield development. A lithofacies is a stratigraphic unit or region characterized by distinctive reflection patterns that differentiate it from other areas. Regions of different lithofacies are typically described using descriptive terms reflecting large-scale seismic patterns, such as reflection amplitude, continuity, and the internal configuration of reflectors defined by stratigraphic horizons. The applications and scales of lithofacies analysis vary widely, from basin-wide applications to detailed reservoir characterization. Within basins, lithofacies analysis has been applied to hydrocarbon system studies for the broad identification of source, reservoir, and sealing-prone areas. These areas are typically identified based on their reflection geometry, amplitude intensity, and continuity. High-amplitude, semi-continuous reflectors are often used to identify potential hydrocarbon-bearing reservoirs, such as deepwater channels, while low-amplitude continuous to semi-continuous areas can be used to identify sealing-prone units.

[0058] Lithofacies analysis can also be applied to single reservoirs to help constrain detailed physical property descriptions. In these localized applications, the definitions of continuity and amplitude are often not rigorous and are based on rock property calibration or sedimentary interpretation environments. The relationship between seismic characteristics and physical properties can be demonstrated, and lithofacies volumes can then be used to predict rock property distributions and conditional geological models.

[0059] The standard technique for lithofacies analysis and mapping is a manual process in which a seismic interpreter makes visual decisions about the characteristics of seismic reflection data within the region of interest and maps this data. The lithofacies are then used for various purposes, but primarily to interpret the distribution of lithofacies and rock properties. Intuition and experience have contributed significantly to the success of lithofacies studies; however, this approach can also lead to lithofacies analysis becoming a subjective, time-consuming, and often laborious inefficient task. Several related techniques have been used in the petroleum industry to improve automation and enhance the interpretation of lithofacies from seismic data.

[0060] Lithofacies refers to rocks or rock assemblages formed in a specific sedimentary environment. It is a major component of sedimentary facies and refers to sedimentary bodies with distinctly different rock physical characteristics and seismic properties. The lithofacies of this invention are classified according to lithology and pore fluid aggregation. Thus, for reservoir types, water-bearing sandstone and oil-bearing sandstone will be classified as two different lithofacies. Lithofacies plays a decisive role in the elastic properties of rocks; therefore, lithofacies identification is crucial for predicting reservoir porosity, permeability, and other physical parameters. Incorrect lithofacies identification may reduce the accuracy of reservoir property predictions and increase the uncertainty of the prediction results. Therefore, accurate lithofacies identification and analysis play a key role in predicting reservoir physical parameters. In stratigraphic correlation, sedimentary facies analysis, and other studies, lithofacies identification is a very important pioneering task.

[0061] Currently, commonly used reservoir facies prediction methods in the oil and gas exploration and development industry include reservoir facies inversion based on seismic information and prediction methods constrained by sedimentary facies models. This invention is a reservoir facies prediction method based on production data, well logging data, and seismic data, representing a comprehensive application of measurement methods from these various fields.

[0062] Because seismic response is influenced by numerous factors, seismic responses of different lithofacies are often very similar, posing significant challenges to lithofacies identification using seismic data. This invention incorporates multivariate statistical and machine learning methods as optimization modules for lithofacies identification and reservoir property parameter prediction. Commonly used machine learning algorithms include K-nearest neighbor discrimination, decision trees, K-means, and Bayesian discrimination. The basic idea behind these algorithms is to statistically analyze the classification features of training data, establish discrimination criteria, and compare and discriminate new data samples. Different discrimination methods have different discrimination criteria, and each method has different assumptions, discrimination bases, and processing techniques, thus each has its own specific application conditions.

[0063] Seismic-based reservoir facies prediction is a commonly used and effective method in the industry, but it also has problems. The main problem is that seismic data is generally limited by resolution, making it difficult to predict the facies of thin reservoirs. This invention addresses these issues by proposing a high-resolution (HR-Net) algorithm model and inventing a new kernel function, which effectively solves the above problems.

[0064] In conventional classification algorithms, the distance-based lithofacies discrimination method assumes a normal distribution for the entire dataset. This method simply considers the mean and variance of the data, the prior probability of different samples, and the covariance matrix between lithofacies categories without systematic analysis. Therefore, it cannot provide a reasonable evaluation of the accuracy of the discrimination results. This invention finds that when the correlation between input attributes is strong, the accuracy of lithofacies discrimination results is low. Furthermore, during the discrimination process, the attributes are added with unequal weights, and the weight coefficients are not linked to the discrimination efficiency. In the improved discrimination method proposed in this invention, addressing the above problems and the difficulty in estimating the conditional probability density function, while retaining the assumption of a Gaussian distribution, an improved Bayesian discrimination method is introduced. This method can directly discriminate lithofacies in their original state without the need for selection and combination transformation of independent variables. The method of this invention can organically combine weight coefficients and discrimination efficiency, comprehensively consider the data characteristics of each sample, enhance the statistical analysis capability of data characteristics, and estimate the posterior probability and misclassification rate of the discrimination results.

[0065] Reservoir facies prediction based on production data is highly dependent on the abundance of such data, which is difficult to obtain systematically and comprehensively in most oilfields. Relying solely on sedimentary facies model constraints for reservoir facies prediction suffers from the problem of facies model diversity. To reduce the ambiguity of inter-well reservoir facies prediction, this invention applies well-side attribute constraints, and the organic integration and mutual verification of seismic data with the HR-Net algorithm model, successfully mitigating the ambiguity problem in seismic image-based reservoir facies prediction.

[0066] Besides well logging data, generated data, and seismic data, the experience of geophysical interpretation experts often plays a decisive role. Therefore, this invention specifically designed expert parameters during model training to record and store the experts' interpretation experience, making the trained model more robust. This effectively achieves the goal of efficient and high-precision reservoir facies prediction.

[0067] This invention addresses the shortcomings of existing technologies by providing an intelligent lithofacies prediction method based on well logging information and seismic data, which can effectively solve the problems existing in the current technology.

[0068] Example 1

[0069] In one embodiment, such as Figure 1 As shown, a reservoir lithofacies prediction method is provided, including the following steps:

[0070] Step 101: Obtain prestack seismic wave data and lithofacies annotation data, and load the prestack seismic wave data into the memory of the data processing device. By traversing the prestack seismic wave data in the data processing device, obtain a three-dimensional array composed of the prestack seismic wave data.

[0071] Here, the executing entity can acquire pre-stack seismic wave data and lithofacies annotation data, and load the pre-stack seismic wave data into the memory of the data processing equipment. By traversing the pre-stack seismic wave data in the data processing equipment, a three-dimensional array composed of pre-stack seismic wave data is obtained. Here, lithofacies refers to a three-dimensional seismic reflection unit within a certain distribution range. The seismic characteristic parameters (such as reflection structure, geometry, amplitude, frequency, continuity, etc.) within this unit differ from adjacent units, representing the lithological assemblage, bedding, and sedimentary characteristics of the sediments that produced its reflection. Lithofacies prediction involves identifying and mapping lithofacies units within sedimentary stratigraphic units based on seismic characteristic parameters, combined with other downhole and surface data, according to a certain procedure. This lays the necessary foundation for the comprehensive interpretation of sedimentary environments and systems. Seismic data, essential for lithofacies prediction, is indispensable basic data in petroleum exploration. It can be obtained in the early stages of exploration and generally covers the entire work area, containing extremely rich stratigraphic, structural, and sedimentary facies information. The purpose of this invention to identify and predict lithofacies is to interpret regional stratigraphy, determine sedimentary systems, lithofacies characteristics, and interpret sedimentary development history. Finally, lithofacies are converted into sedimentary facies, which serves as the basis for studying the generation, reservoir, and caprock assemblages and their distribution patterns in petroleum geology, thereby predicting favorable oil-generating areas and reservoir facies zones.

[0072] In one possible embodiment, pre-stack seismic wave data is loaded into the memory of a data processing device, and a three-dimensional array composed of the pre-stack seismic wave data is obtained by traversing the pre-stack seismic wave data in the data processing device, including the following steps:

[0073] Extract well-side data from pre-stack seismic wave data from well logging curves;

[0074] Lithofacies interpretation results and conditional attributes are obtained based on wellbore data and lithofacies annotation data;

[0075] The three-dimensional array was determined based on the lithofacies interpretation results and conditional properties.

[0076] Specifically, in a relatively simple geological structure area of ​​an oilfield, firstly, high-precision seismic exploration instruments are used to extract pre-stack seismic wave data from well logging curves. The well-side data corresponds in depth to the well logging lithofacies data. The strata in this area are relatively stable, with relatively uniform lithological variations. Then, the system analyzes drilled core samples and interprets well logging data to obtain accurate lithofacies annotation data. This annotation data covers the main lithological types in the area, such as sandstone and mudstone. Next, the system loads the acquired pre-stack seismic wave data into the memory of the data processing equipment. By traversing the pre-stack seismic wave data within the data processing equipment and obtaining lithofacies interpretation results and conditional attributes based on the well-side data and lithofacies annotation data, a three-dimensional array is formed for subsequent processing. The system can select a portion of the data with known lithofacies classifications as training data, dividing it into a calibration set at 70% (adjustable according to actual conditions) and a cross-validation set at the remaining 30%.

[0077] Step 102: Generate a reservoir prediction model based on the three-dimensional array and lithofacies annotation data.

[0078] Here, the executing entity traverses the prestack seismic wave data in the data processing device to obtain a three-dimensional array composed of the prestack seismic wave data. Based on the three-dimensional array and lithofacies annotation data, it can generate a reservoir prediction model. The reservoir prediction model is generated by processing the three-dimensional array and lithofacies annotation data using the HR-Net algorithm.

[0079] In one possible embodiment, generating a reservoir prediction model based on a three-dimensional array and lithofacies annotation data includes the following steps:

[0080] Obtain the model training set based on a three-dimensional array;

[0081] The HR-Net algorithm is used to process the model training set to obtain a reservoir prediction model.

[0082] Specifically, first, the executing entity can obtain the model training set from the three-dimensional array, where the model training set includes a calibration set and a cross-validation set; then, the executing entity constructs the HR-Net algorithm model based on the size and structural characteristics of the regional data in the three-dimensional array, such as... Figure 2 As shown, Figure 2An exemplary flowchart of the HR-Net algorithm model is shown. The HR-Net (High-Resolution Network) algorithm model is a deep learning neural network model that includes multiple convolutional units. In this invention, it is used to interpret reservoir lithology in seismic images. It aims to solve the problem of feature fusion at different scales in tasks such as image semantic segmentation, so as to achieve high-precision identification and interpretation of reservoir lithology in seismic images. The model training set is processed by the HR-Net algorithm model to obtain a reservoir prediction model, which is used to predict the reservoir distribution in unknown exploration areas.

[0083] In one possible embodiment, the reservoir prediction model is obtained by processing the model training set using the HR-Net algorithm, including the following steps:

[0084] Generate the HR-Net algorithm model based on the size and structure of the model training set;

[0085] The loss function is defined as a mixture of cross-entropy and BCE with different weights. The learning rate is optimized using the cosine annealing algorithm, and an adaptive gradient descent optimizer is used.

[0086] The reservoir prediction model is obtained by processing the model training set based on the loss function.

[0087] Specifically, after obtaining the model training set based on a three-dimensional array, the execution entity first generates an HR-Net algorithm model according to the data size and structure of the training set. The HR-Net model includes a loss function, and the execution entity conducts random tensor experiments to determine the initial structure and parameter settings of the model through multiple trials, ensuring that the model can adapt to the data characteristics in the early stages of construction. Next, the execution entity defines a hybrid loss function of cross-entropy and BCE with different weights, setting appropriate weight values ​​based on data characteristics and model requirements. For example, the execution entity can use a cosine annealing algorithm to define a learning rate optimization scheme, with an initial learning rate set to 0.001 (which can be adjusted based on experiments), and define an adaptive gradient descent optimizer to ensure that the model can converge quickly and stably during training. Cosine annealing is a dynamically adjusting learning rate optimization algorithm mainly used for training neural network models. Its core idea is to gradually reduce the learning rate through a cosine function, thereby avoiding oscillations during training and improving the model's training stability and generalization ability. Finally, the execution entity substitutes the loss function into the model training set to obtain the reservoir prediction model.

[0088] Step 103: Predict the reservoir distribution in the unknown exploration area using a reservoir prediction model.

[0089] Here, after generating a reservoir prediction model based on a 3D array and lithofacies annotation data, the executing entity can use the reservoir prediction model to predict the reservoir distribution in unknown exploration areas and obtain prediction results. These prediction results are 3D data models in SEGY format. SEGY is a standard magnetic tape data format proposed by SEG (Society of Exploration Geophysicists), and it is one of the most common formats for seismic data in the petroleum exploration industry.

[0090] In one possible embodiment, the reservoir distribution in an unknown exploration area is predicted using a reservoir prediction model, and the prediction results are obtained, including the following steps:

[0091] The reservoir prediction model is evaluated based on the condition attributes required by the reservoir prediction model, and the evaluation results of the reservoir prediction model are obtained.

[0092] Determine the target parameters in the HR-Net algorithm based on the evaluation results;

[0093] The reservoir prediction model is processed according to the target parameters to obtain the target optimized model of the reservoir prediction model;

[0094] The reservoir distribution in unknown exploration areas is predicted based on the target optimization model, and the prediction results are obtained.

[0095] Specifically, firstly, the implementing entity evaluates the reservoir prediction model based on the required condition attributes, obtaining the evaluation results. That is, after model training is complete, the implementing entity imports new block data for lithofacies prediction to obtain the evaluation results. For example, the implementing entity inputs seismic data from a new exploration area into the trained model, and the model predicts the lithofacies of that area based on learned features and patterns. Then, the implementing entity evaluates and analyzes the prediction results, comparing and verifying them with lithofacies data obtained from actual drilling to obtain the evaluation results. Figure 3 As shown, Figure 3 An illustrative diagram of the training results is shown. In a specific slope area, the lithofacies distribution predicted by the model highly matches the actual lithofacies situation. Subsequently, the executing entity determines the target parameters in the HR-Net algorithm based on the evaluation results, and processes the reservoir prediction model according to the target parameters to train and optimize the reservoir prediction model to obtain the target optimized model. Finally, the executing entity predicts the reservoir distribution in the unknown exploration area based on the target optimized model, obtaining the prediction results, such as... Figure 4 As shown, Figure 4An exemplary schematic diagram of the prediction results is shown, in which the predicted lithology and the actual lithology are in high agreement, demonstrating the effectiveness and accuracy of the method in reservoir facies prediction and providing an important basis for decision-making in oil and gas exploration and development.

[0096] In one possible embodiment, obtaining the target optimization model based on the target parameters includes the following steps:

[0097] The confusion matrix for supervised learning of lithofacies identification methods is calculated by comparing the lithofacies predicted for the cross-validation set in the target parameters.

[0098] The lithofacies transition matrix, which characterizes the changes between continuous facies, is calculated based on the confusion matrix.

[0099] The target optimization model for reservoir prediction is determined based on the lithofacies transition matrix and confusion matrix.

[0100] Specifically, after determining the target parameters in the HR-Net algorithm based on the evaluation results, the executing entity first calculates the confusion matrix used for supervised learning of the lithofacies recognition method by comparing the lithofacies predicted for the cross-validation set using the target parameters. Then, based on the confusion matrix, the executing entity calculates the lithofacies transition matrix, representing the changes between continuous facies. Finally, based on the lithofacies transition matrix and the confusion matrix, the executing entity determines the target optimization model for the reservoir prediction model, i.e., continuously adjusting the model parameters during training, such as the weights and biases of the convolution kernels. Simultaneously, the executing entity monitors the model's training performance using the validation set, and adjusts the training strategy in a timely manner based on metrics such as loss value and accuracy on the validation set, such as adjusting the learning rate and increasing the number of training epochs, to optimize the model and improve its performance.

[0101] This disclosure provides a reservoir lithofacies prediction method, apparatus, equipment, medium, and product. First, pre-stack seismic wave data and lithofacies annotation data are acquired, and the pre-stack seismic wave data is loaded into the memory of a data processing device. A three-dimensional array composed of the pre-stack seismic wave data is obtained by traversing the data in the data processing device. Then, a reservoir prediction model is generated based on the three-dimensional array and lithofacies annotation data. The reservoir prediction model is generated by processing the three-dimensional array and lithofacies annotation data using the HR-Net algorithm. Finally, the reservoir distribution in an unknown exploration area is predicted using the reservoir prediction model, and the prediction result is obtained. The prediction result is a 3D data model in Segy format.

[0102] As described above, the technical solution disclosed herein generates a reservoir prediction model using a three-dimensional array and lithofacies annotation data. This reservoir prediction model is generated by processing the three-dimensional array and lithofacies annotation data using the HR-Net algorithm. Compared with the traditional method of manually analyzing lithofacies, using the HR-Net algorithm to process the three-dimensional array and lithofacies annotation data can effectively capture subtle geological information, thereby more accurately classifying lithofacies and providing a more reliable basis for reservoir prediction, thus improving the efficiency and accuracy of lithofacies prediction.

[0103] 1. Acquire pre-stack seismic wave data and lithofacies annotation data. Load the pre-stack seismic wave data into the memory of the data processing device. By traversing the pre-stack seismic wave data in the data processing device, obtain a three-dimensional array composed of the pre-stack seismic wave data. This provides the most original and crucial information source for the entire reservoir lithology interpretation system. The pre-stack seismic wave data contains various information about the propagation process of seismic waves underground, which can reflect rich details of the subsurface geological structure and lithological characteristics. The lithofacies annotation data provides an accurate reference standard for subsequent model training, enabling the model to learn the performance characteristics of different lithofacies in the data, thereby improving the accuracy of lithofacies prediction.

[0104] 2. Reservoir prediction models generated based on three-dimensional arrays and lithofacies annotation data can effectively capture subtle geological information, thereby more accurately classifying lithofacies and providing a more reliable basis for reservoir prediction, improving the efficiency and accuracy of lithofacies prediction.

[0105] 3. By using reservoir prediction models to predict the reservoir distribution in unknown exploration areas, the uncertainty in the exploration process can be reduced. Reservoir prediction models can provide quantitative and qualitative information about reservoir distribution, such as reservoir thickness and lithological distribution, giving explorers a clearer understanding of the subsurface geological conditions. In a complex, unknown area, traditional methods provide almost no information about the reservoir situation, while model predictions can indicate the probability and general characteristics of reservoirs in different locations, reducing uncertainty by 40%-60%. This helps in developing more reasonable exploration plans and risk management strategies.

[0106] Example 2

[0107] Based on the above embodiments, in one embodiment, data processing and model training of the intelligent lithofacies identification system can be implemented through software. This system consists of an artificial intelligence module comprising machine learning and deep learning modules, as well as a Python web framework. The client and server are connected via a network. The user opens the webpage of the intelligent lithofacies identification system and enters the data entry interface, such as... Figure 5 As shown, Figure 5An exemplary data entry interface diagram is shown. In this interface, well logging data for an oilfield is entered via file upload. This includes resistivity curves (used to reflect formation conductivity and distinguish different lithologies and fluid properties), sonic transit time curves (used to estimate formation porosity and lithology), and natural gamma curves (used to identify clay content and lithology). The data covers 30 wells in the oilfield, with depths ranging from 1000 meters to 3000 meters. Simultaneously, 3D seismic data for the area is uploaded, with a sampling interval of 2 milliseconds and a coverage area of ​​50 square kilometers. Calculation parameters are set according to actual needs, such as a data sampling interval of 1 meter (for subsequent data processing), a processing window size of 100 meters × 100 meters (for calculating seismic texture attributes), and module parameters, such as an initial learning rate of 0.001 for the deep learning model and 3 hidden layers.

[0108] After entering the data and setting the parameters, you will enter the model training interface, such as... Figure 6 As shown, Figure 6 An exemplary schematic diagram of the model training interface is shown, which can be used to calibrate the target layer (the main reservoir section of this oilfield is located in the depth range of 2000-2500 meters) using well logging, well logging, and synthetic seismic records. By comparing the characteristic changes of the well logging curves and the seismic reflection interface, the top and bottom interfaces and internal subdivisions of the target layer are accurately determined.

[0109] After determining the top and bottom interfaces and internal subdivisions of the target layer, the well logging data were specified and classified. Based on core descriptions and geological expert experience, the data were labeled as lithofacies categories such as sandstone, mudstone, and carbonate rocks. Data with known lithofacies classifications were divided into training and testing subsets in a 7:3 ratio. A backpropagation neural network was used as the supervised learning algorithm for model training. During training, a confusion matrix was generated, where the diagonal elements represent the number of times the model correctly classified each lithofacies category, and the off-diagonal elements represent the number of misclassifications. For example, after one training iteration, the confusion matrix showed that 80% of sandstone was correctly classified, but 10% was misclassified as mudstone, and 10% was misclassified as carbonate rock. A facies transition matrix was generated, and analysis revealed a transition probability of 0.3 from sandstone to mudstone and 0.2 from mudstone to sandstone. Based on these matrices, the target probability matrix was calculated, and the weights for model training were adjusted.

[0110] Seismic data was extracted along the target stratigraphic level within a specified time window width (e.g., 50 milliseconds) as training data for the deep learning pre-trained model, with a time window shift distance of 25 milliseconds. Through 100 experiments, the optimal model depth was selected as 5 layers, with the number of neurons per layer set to [128, 64, 32, 16, 8], the neuron activation function to ReLU, and a sparsity constraint of 0.4. Then, well-side seismic data was extracted along the target stratigraphic level as training data for the deep learning fine-tuning model, including oil, gas, and water samples. The deep learning model parameters were fine-tuned using batch stochastic gradient descent with a batch size of 32. After 50 rounds of fine-tuning, the model's accuracy on the test subset improved by 10%.

[0111] During model training, the basis function of each layer of the deep learning model is calculated, and the seismic response value of the target layer is extracted. Analysis reveals that the frequency characteristics of seismic waves are crucial in distinguishing between reservoirs and non-reservoirs, and these frequency characteristics are used as target features for deep learning. For example, energy variations within the frequency range of 20Hz-50Hz are closely related to the hydrocarbon content of reservoirs; utilizing this feature can effectively enhance the seismic response characteristics of hydrocarbons and improve the ability to distinguish between reservoirs and non-reservoirs.

[0112] Example 3

[0113] Based on the above embodiments, in one embodiment, the geological conditions of a certain oil and gas exploration area are extremely complex. In terms of subsurface structure, there are multiple large folds and fault zones, with fold amplitudes ranging from tens to hundreds of meters, and fault zones extending for thousands of meters. Numerous small fault blocks are irregularly distributed, resulting in a large lateral variation in subsurface velocity ranging from 1500 m / s to 4500 m / s. The reservoir lithology is complex, encompassing various types from coarse sandstone to mudstone and shale, with frequent transition zones between different lithologies. Potential reservoir types are diverse, including both porous and fracture-porous reservoirs, and gas reservoir types include conventional structural gas reservoirs, lithological gas reservoirs, and tight gas reservoirs. This complex geological situation makes it difficult for traditional exploration methods to accurately characterize reservoir features and predict oil and gas distribution; detailed description is extremely challenging, and the exploration risk is extremely high.

[0114] Using the software described in the above embodiments, detailed seismic data interpretation was first performed. Regarding stratigraphic correlation, multiple stratigraphic interfaces were accurately identified by comparing seismic reflection characteristics and well logging curve characteristics, with stratigraphic thickness errors controlled within 5%. In structural interpretation, the software accurately depicted the morphology, orientation, and amplitude of folds, as well as the location, extension direction, and displacement of fault zones. The generated structural maps clearly demonstrated the complexity of the subsurface structures. Sedimentary facies studies were conducted. Based on information such as the reflection structure and amplitude variations in the seismic data, combined with lithology and sedimentary structural characteristics from well logging data, various sedimentary facies types, including alluvial fans, rivers, and deltas, were identified. Sedimentary facies planar maps were drawn, accurately reflecting the distribution range and interrelationships of different sedimentary facies.

[0115] Then, pre-stack and post-stack seismic inversion and reservoir prediction were performed. Utilizing the rich information from pre-stack seismic data and combining it with well data, an initial seismic inversion model was established. A co-kriging interpolation algorithm was employed for modeling, fully considering the influence of geological structures and lithological variations on seismic properties. High-resolution acoustic impedance / velocity data were obtained through iterative inversion. Reservoir prediction results showed that the reservoir thickness prediction error was within 10%, and the prediction accuracy for favorable reservoir distribution ranges reached 85%. The distribution ranges of several favorable reservoirs were determined, providing clear target areas for exploration and development. For example, within an area of ​​30 square kilometers, two favorable reservoir areas were delineated, with predicted high hydrocarbon saturation.

[0116] Finally, exploration and development objectives were proposed, and a detailed exploration and development plan was formulated based on the software analysis results. Regarding well location, the locations of five exploratory wells were determined based on reservoir prediction results and structural characteristics. For the selection of extraction methods, corresponding extraction strategies were formulated for different types of reservoirs (e.g., conventional water injection for porous reservoirs, and fracturing for fractured-porous reservoirs). In practical application, the success rate of exploratory wells drilled under software guidance reached 60%, effectively improving the efficiency and success rate of oil and gas exploration and development, and reducing exploration risks. Simultaneously, the application of the software fully exploited the geological information in seismic data, transforming previously underutilized high-precision seismic data (previously only used for simple structural interpretation) into comprehensive reservoir analysis and prediction data, avoiding data waste and improving the economic benefits and technical level of the entire exploration and development project. Furthermore, the integrated data management platform provided by the software facilitated data query and management, the seismic data optimization processing tools improved the quality of seismic profiles, and the multi-well integrated geological interpretation function assisted geological engineers in conducting more in-depth geological analysis, providing comprehensive support for the smooth progress of the project.

[0117] Example 4

[0118] Based on the above embodiments, in one embodiment, for a seismic dataset and well logging dataset of an oil field, multiple initial texture attributes representing the amount of seismic data are first calculated. For example, for a study area of ​​50 square kilometers with a depth ranging from 1000 meters to 3000 meters, initial texture attributes such as amplitude and phase are calculated using seismic data.

[0119] Then, a probabilistic neural network is constructed from the calculated initial texture attributes. Representative seismic trace data within the region are selected as training data, and the texture attributes of these data are used to construct the probabilistic neural network. The texture attributes of the training images become the weight vectors of the first layer of the network. For example, 10 seismic traces traversing different geological structures are selected. These data contain the reflection characteristics of different lithologies such as sandstone and shale, and their texture attributes are used to construct the network weights.

[0120] Subsequently, the final texture attributes are calculated from the entire seismic data set. During the calculation, 2D texture attributes are preferentially selected and filtered in time slices to simulate 3D operations. Simultaneously, the window size is adjusted to account for changes in data frequency with depth. For example, at depths of 1000-1500 meters, where the data frequency is higher, a smaller window size (e.g., 50m x 50m) is used to calculate texture attributes; at depths of 2500-3000 meters, where the data frequency decreases, the window size increases to 100m x 100m. Furthermore, a dynamic window size adjustment strategy based on user-defined confidence levels is employed to ensure the accuracy of the calculation results.

[0121] After calculating the final texture attributes from the entire seismic data set, a constructed probabilistic neural network is used to classify these attributes. During the classification process, the neural network's ability to generate nonlinear boundaries is utilized to distinguish between different lithofacies. For example, to differentiate between sandstone and shale, the neural network learns the differences in texture attributes between sandstone and shale in the training data, establishes a decision surface, and accurately classifies new data.

[0122] After classification, scalar statistical measurements of texture attributes are calculated, including first-order and second-order descriptors. First-order statistics, such as the mean absolute amplitude and standard deviation of amplitude values ​​within texels, can be used to depict amplitude anomalies and reflection intensity. Analysis revealed that the sandstone reservoirs in the study area had higher mean absolute amplitudes at certain locations, reflecting stronger reflection intensity. Second-order statistics, such as texture correlation, texture inertia, texture entropy, and texture energy, are used to quantify the spatial relationships between pixels within an image. For example, in a sandstone reservoir area with good continuity, the texture correlation value is high, the texture inertia is low, the texture entropy is low, and the texture energy is high, indicating that the pixel values ​​in this area have high similarity, low contrast, spatially ordered organization, and a relatively concentrated gray-level distribution.

[0123] After calculating the scalar statistical measurements of texture attributes, lithofacies identification is performed based on the characteristics of the gray-level co-occurrence matrix. In homogeneous regions, elements closer to the diagonal of the gray-level co-occurrence matrix have higher values; for example, in shale regions, the diagonal elements have relatively high values, reflecting low pixel value differences. In heterogeneous regions, such as the transition zone between sandstone and shale, elements farther from the diagonal have larger values. In low-amplitude regions, gray-level co-occurrence matrix values ​​cluster near the center, while in high-amplitude regions, they are more widely distributed. These characteristics allow for accurate identification of different lithofacies and their transitional zones.

[0124] It should be noted that, in order to improve model performance, when dealing with degraded seismic data quality, the data is first filtered using convolution or median filters to smooth it. For example, in areas with high data noise, a median filter (with a filter window size of 3×3) is used to preprocess the seismic data to reduce the impact of noise on texture attribute calculation and lithofacies identification.

[0125] Subsequently, the facies were recalculated and classified based on the stratigraphic dip angle. In an area with a large stratigraphic dip angle, the depressions within the image were estimated using gradient-based techniques, and the direction of texture analysis was adjusted to make the calculation results more consistent with the actual geological conditions. For example, the texture attributes that were originally calculated horizontally were recalculated according to the stratigraphic dip direction after taking the stratigraphic dip angle into account, which improved the accuracy of lithofacies identification in this area.

[0126] Finally, the model was continuously optimized by training and adjusting it with new data to improve its generalization ability. As exploration progressed, more seismic data and lithofacies annotation information were obtained. This new data was added to the training set, and the probabilistic neural network was retrained, enabling the model to better adapt to lithofacies identification and prediction tasks under different geological conditions within the study area.

[0127] Example 5

[0128] Based on the above embodiments, this embodiment provides an application example. In an oil field, dense seismic acquisition equipment was deployed to obtain high-resolution pre-stack seismic data, covering an area of ​​100 square kilometers, with a sampling interval of 2 milliseconds. Simultaneously, detailed core analysis and well logging measurements were performed on 25 existing wells in the oil field, obtaining accurate lithofacies annotation data covering various lithofacies types such as sandstone, shale, and carbonate rocks.

[0129] First, the pre-stack seismic wave data was loaded into the memory of a high-performance computing server and combined according to three-dimensional spatial coordinates to form a three-dimensional array of size (1000×1000×500) (where 1000×1000 is the planar spatial resolution and 500 is the number of time sampling points). Preprocessing of the data included noise removal (using a median filtering algorithm with a filter window size of 5×5×3), amplitude compensation (exponential compensation based on formation absorption characteristics), and normalization (mapping amplitude values ​​to the [-1,1] interval), which improved data quality.

[0130] Then, the preprocessed data was divided into training, validation, and test sets in a 7:2:1 ratio to ensure a balanced proportion of different lithofacies in each subset. Spatial random sampling was used during the partitioning process to avoid the impact of spatial correlation on model training.

[0131] Subsequently, based on the characteristics of the data, the HR-Net algorithm model was constructed. The model contains four stages, with resolution branches of [1 / 4, 1 / 8, 1 / 16, 1 / 32] for each stage. During the construction process, ten random tensor experiments were conducted, adjusting the convolutional kernel size (from 3×3 to 7×7) and the number of channels (from 32 to 128) of the network layers in each experiment. Finally, a network structure with a convolutional kernel size of 5×5 and 64 channels was determined, resulting in a lower loss function value and faster convergence speed in the initial experiments.

[0132] After constructing the HR-Net algorithm model, a hybrid loss function of cross-entropy and BCE is defined, with a weight of 0.7 for the cross-entropy loss function and a weight of 0.3 for the BCE loss function, to balance classification accuracy and performance for binary classification tasks (such as reservoir vs. non-reservoir). The learning rate is optimized using cosine annealing, with an initial learning rate of 0.005, which decays to 0.9 times every 10 training epochs. An adaptive gradient descent optimizer is defined with a momentum parameter set to 0.9, and L2 regularization (weight decay coefficient of 0.001) is used to prevent overfitting.

[0133] The model was then trained using the training set, with performance evaluated on the validation set every 5 epochs. Initially, the model's loss function decreased rapidly, but overfitting became apparent in the 30th epoch (the validation set loss function began to rise). By adding random transformations to the training data (such as random rotations and flips, with angles ranging from -10° to 10°) and employing early stopping (stopping training when the validation set loss function did not decrease for 5 consecutive epochs), the model achieved an accuracy of 88% on the validation set after 100 epochs of training.

[0134] After training, seismic data from a new block (20 square kilometers) was input into the model for lithofacies prediction. The prediction results showed that the eastern part of the block was more likely to contain sandstone reservoirs, the central part was mainly composed of shale, and the western part was more likely to contain carbonate reservoirs.

[0135] To evaluate the predictive performance, drilling verification was conducted at eight locations within the new block. Comparative results showed that the model achieved an accuracy rate of 85% in predicting lithofacies, with 90% accuracy for sandstone reservoirs, 80% for shale, and 75% for carbonate rocks. In a typical slope area (as shown in the attached figure)... Figure 2 As shown in the attached figure, the model accurately predicted the location of the transition zone between sandstone and shale, which is in high agreement with the actual core analysis results. In the prediction of the entire new block (as shown in the attached figure), the model accurately predicted the location of the transition zone between sandstone and shale, which is in high agreement with the actual core analysis results. Figure 3 As shown in the figure, the lithofacies distribution trend is consistent with the geological background and sedimentary environment analysis results, providing an important basis for decision-making for the further exploration and development of the oilfield, and effectively guiding well location deployment and reserve assessment.

[0136] In this embodiment, detailed data acquisition (such as 100 square kilometers of high-resolution pre-stack seismic data and accurate lithofacies annotation data from 25 wells) and meticulous data preprocessing (denoising, amplitude compensation, normalization, etc.) provided high-quality input to the model, enabling the HR-Net model to learn lithofacies characteristics more accurately. For example, after data preprocessing, the noise level of the seismic data was reduced by 30%, and the amplitude stability was improved by 25%, laying the foundation for accurate subsequent predictions.

[0137] In this embodiment, the HR-Net model's unique architecture (4 stages and multi-resolution branches) and optimized training strategies (such as random tensor experiments to determine the optimal network structure, hybrid loss functions to balance classification tasks, and cosine annealing to optimize the learning rate) enable it to perform exceptionally well in lithofacies prediction. In this oilfield application, the model achieved a prediction accuracy of over 85% for various lithofacies, including sandstone, shale, and carbonate rocks, representing a 20%-30% improvement compared to traditional methods. In a typical slope area, the model accurately predicted the location of the lithofacies transition zone, closely matching the actual core analysis results, effectively solving the problem of traditional methods' difficulty in accurately identifying complex lithofacies transition zones.

[0138] In this embodiment, accurate lithofacies prediction results provided crucial information for well placement. Based on the lithofacies distribution predicted by the model, the locations of 10 exploratory wells were rationally planned within a 20-square-kilometer area of ​​the new block, improving the drilling success rate. For example, in areas predicted to have a high probability of sandstone reservoirs, the drilling success rate reached 70%, while the success rate of well placement using traditional methods was only around 40%.

[0139] Furthermore, this embodiment facilitates reserve assessment and exploitation scheme design. Through accurate lithofacies prediction, oil and gas reserves in reservoirs of different lithofacies can be estimated more precisely, and personalized exploitation schemes can be developed for different lithofacies. For example, acid fracturing and production enhancement measures can be adopted for carbonate reservoirs, and water injection schemes can be optimized for sandstone reservoirs, thereby improving oil and gas recovery rates, which are expected to increase recovery rates by 10%-15%.

[0140] In this embodiment, unnecessary drilling work is reduced. Due to the high accuracy of model predictions, ineffective drilling in unfavorable lithofacies areas is avoided. In this oilfield application, it is expected to reduce the number of drilling wells by 20%-30%, significantly reducing drilling costs.

[0141] In this embodiment, the exploration and development cycle is shortened. The entire process from data acquisition to the application of lithofacies prediction results is more efficient. For example, the traditional method requires 12 months to complete a round of exploration and development decisions, while the method shortens this to 8 months, enabling oil fields to start production faster and obtain economic benefits sooner.

[0142] In this embodiment, accurate prediction and analysis of lithofacies through the model can provide a deeper understanding of the sedimentary environment and geological evolution of the oilfield. For example, based on the distribution patterns of lithofacies, it can be inferred that the oilfield experienced multiple marine transgressions and regressions during geological history, with the sedimentary environment gradually transitioning from shallow sea to terrestrial facies, providing important clues for further research on the geological characteristics of the oilfield.

[0143] Furthermore, this embodiment helps to discover potential oil and gas reservoirs. Based on a deep understanding of lithofacies and geological characteristics, it can identify some hidden oil and gas reservoirs that are difficult to discover using traditional methods. For example, small lenticular oil and gas reservoirs were discovered near lithofacies transition zones, providing a new direction for increasing oilfield reserves and production.

[0144] It should be noted that this embodiment can be applied to a variety of scenarios:

[0145] (1) The geological conditions are complex and the structural interpretation is difficult. Most of the exploration areas have relatively complex underground structures, with fault zones and small fault blocks developed, and the underground velocity changes rapidly in the lateral direction. The reservoir lithology is complex, the potential reservoir types are diverse, and the gas reservoir types are complex and diverse, making it difficult to describe in detail.

[0146] (2) The utilization rate of existing seismic data is low. Due to the limitations of hardware and software conditions, the high-precision seismic data collected by customers at great cost can only be used for structural interpretation and simple post-stack reservoir prediction. In particular, the high-precision all-round three-dimensional seismic data collected cannot be used to independently and effectively carry out pre-stack reservoir and fracture prediction, which seriously restricts the exploration process of low-permeability blocks, fails to tap the potential of data information to reduce exploration risks, and causes a certain degree of data waste.

[0147] (3) Due to limitations such as professional seismic software and professional technology, seismic exploration projects make it difficult for exploration deployment personnel and decision-makers to obtain relatively real, reliable and scientific evaluation data in a timely manner, which to some extent increases exploration risks. At the same time, the client's scientific researchers also lose the opportunity to further study and master the core exploration technologies, which restricts the establishment of the core technology system of the client's oil and gas exploration assets.

[0148] As described above, this embodiment can construct an integrated geophysical platform for exploration and development, which is of great significance for clients to improve their technical system. This integrated platform is needed for detailed seismic data interpretation (stratigraphic correlation, structural interpretation), sedimentary facies research, pre-stack and post-stack seismic inversion, and reservoir prediction to determine the distribution range of favorable reservoirs and propose exploration and development targets. It can maximize the exploitation of the rich geological information in existing massive seismic data, effectively reduce the risks of oil and gas exploration and development, and contribute to clients' efforts in discovering geological reserves.

[0149] Example 6

[0150] By dividing each function into corresponding functional modules, this disclosure provides a reservoir facies prediction device, which can be a server or a chip applied to a server. Figure 7 A schematic block diagram of the functional modules of a reservoir facies prediction device provided for an exemplary embodiment of this disclosure.

[0151] like Figure 7 As shown, the reservoir facies prediction device includes:

[0152] The acquisition module 701 is used to acquire prestack seismic wave data and lithofacies annotation data, and load the prestack seismic wave data into the memory of the data processing device. By traversing the prestack seismic wave data in the data processing device, a three-dimensional array composed of the prestack seismic wave data is obtained.

[0153] The generation module 702 is used to generate a reservoir prediction model based on the three-dimensional array and the lithofacies annotation data; wherein, the reservoir prediction model is generated by processing the three-dimensional array and the lithofacies annotation data using the HR-Net algorithm;

[0154] The prediction module 703 is used to predict the reservoir distribution in an unknown exploration area using the reservoir prediction model and obtain prediction results; wherein the prediction results are 3D data models in Segy format.

[0155] In one embodiment, the acquisition module 701 includes:

[0156] An extraction unit is used to extract well-side channel data from the pre-stack seismic wave data from the logging curves, wherein the well-side channel data corresponds to the depth of the logging lithofacies data;

[0157] The first acquisition unit is used to acquire lithofacies interpretation results and conditional attributes based on the wellbore data and the lithofacies annotation data;

[0158] The first determining unit is used to determine the three-dimensional array based on the lithofacies interpretation results and the conditional attributes.

[0159] In one embodiment, the generation module 702 includes:

[0160] The second acquisition unit is used to acquire a model training set based on the three-dimensional array; wherein the model training set includes a calibration set and a cross-validation set;

[0161] The third acquisition unit is used to process the model training set using the HR-Net algorithm to acquire the reservoir prediction model; wherein the reservoir prediction model is used to predict the reservoir distribution in the unknown exploration area.

[0162] In one embodiment, the generation module 702 includes:

[0163] A generation unit is used to generate an HR-Net algorithm model based on the data size and data structure of the model training set; wherein the HR-Net algorithm model includes a loss function;

[0164] The first computational unit is used to define the loss function as a mixture of cross-entropy and BCE functions with different weights, optimize the learning rate using the cosine annealing algorithm, and use an adaptive gradient descent optimizer.

[0165] The fourth acquisition unit is used to process the model training set based on the loss function to obtain the reservoir prediction model.

[0166] In one embodiment, the prediction module 703 includes:

[0167] The fifth acquisition unit is used to evaluate the reservoir prediction model according to the condition attributes required by the reservoir prediction model, and to obtain the evaluation result of the reservoir prediction model.

[0168] The second determining unit is used to determine the target parameters in the HR-Net algorithm based on the evaluation results;

[0169] The sixth acquisition unit is used to process the reservoir prediction model according to the target parameters to obtain the target optimized model of the reservoir prediction model;

[0170] The seventh acquisition unit is used to predict the reservoir distribution in the unknown exploration area based on the target optimization model and obtain the prediction results.

[0171] In one embodiment, the prediction module 703 includes:

[0172] The second computing unit is used to calculate a confusion matrix for supervised learning of the lithofacies identification method by comparing the lithofacies predicted for the cross-validation set in the target parameters.

[0173] The third calculation unit is used to calculate the lithofacies transition matrix representing the changes between continuous facies based on the confusion matrix;

[0174] The third determining unit is used to determine the target optimization model of the reservoir prediction model based on the lithofacies transition matrix and the confusion matrix.

[0175] Example 7

[0176] Based on the above embodiments, this disclosure also provides an electronic device, including: at least one processor; a memory for storing executable instructions of the at least one processor; wherein the at least one processor is configured to execute the instructions to implement the above-described method disclosed in this disclosure.

[0177] Figure 8 This is a schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this disclosure. For example... Figure 8 As shown, the electronic device 800 includes at least one processor 801 and a memory 802 coupled to the processor 801. The processor 801 can perform the corresponding steps in the methods disclosed in the embodiments of this disclosure.

[0178] The processor 801 described above can also be called a central processing unit (CPU), which can be an integrated circuit chip with signal processing capabilities. Each step in the method disclosed in this embodiment can be implemented by the integrated logic circuitry in the processor 801 or by software instructions. The processor 801 can be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this embodiment can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can be located in the memory 802, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor 801 reads information from the memory 802 and, in conjunction with its hardware, completes the steps of the method described above.

[0179] Furthermore, various operations / processes according to this disclosure, implemented via software and / or firmware, can be transmitted from a storage medium or network to a computer system with a dedicated hardware architecture, such as... Figure 9 The computer system 900 shown is equipped with the programs that constitute the software. When various programs are installed, the computer system is able to perform various functions, including functions such as those described above. Figure 9 A block diagram of a computer system provided for an exemplary embodiment of this disclosure.

[0180] Computer system 900 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0181] like Figure 9As shown, the computer system 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. The RAM 903 may also store various programs and data required for the operation of the computer system 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0182] Multiple components in the computer system 900 are connected to the I / O interface 905, including: an input unit 906, an output unit 907, a storage unit 908, and a communication unit 909. The input unit 906 can be any type of device capable of inputting information into the computer system 900. The input unit 906 can receive input numerical or character information and generate key signal inputs related to user settings and / or function control of the electronic device. The output unit 907 can be any type of device capable of presenting information and may include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. The storage unit 908 may include, but is not limited to, a hard disk and an optical disk. The communication unit 909 allows the computer system 900 to exchange information / data with other devices via a network such as the Internet, and may include, but is not limited to, a modem, network card, infrared communication device, wireless communication transceiver, and / or chipset, such as Bluetooth™ device, WiFi device, WiMax device, cellular communication device, and / or the like.

[0183] The computing unit 901 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above. For example, in some embodiments, the methods disclosed in this disclosure can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via ROM 902 and / or communication unit 909. In some embodiments, the computing unit 901 can be configured to perform the methods disclosed in this disclosure by any other suitable means (e.g., by means of firmware).

[0184] This disclosure also provides a computer-readable storage medium, wherein when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform the methods disclosed in this disclosure.

[0185] The computer-readable storage medium in this disclosure can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. The aforementioned computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specifically, the aforementioned computer-readable storage medium may include electrical connections based on one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0186] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0187] Figure 10 A computer program product 1000 is provided as an exemplary embodiment of the present disclosure. The computer program product 1000 includes a computer program 1001, wherein the computer program 1001, when executed by a processor, implements the methods disclosed in the embodiments of the present disclosure.

[0188] In embodiments of this disclosure, computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network (including a local area network (LAN) or a wide area network (WAN)), or it can be connected to an external computer.

[0189] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0190] The modules, components, or units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the modules, components, or units do not necessarily constitute a limitation on the module, component, or unit itself.

[0191] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary hardware logic components that can be used include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0192] The above description is merely an embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0193] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A method for predicting reservoir lithofacies, characterized in that, include: Acquire prestack seismic wave data and lithofacies annotation data, and load the prestack seismic wave data into the memory of the data processing device. By traversing the prestack seismic wave data in the data processing device, obtain a three-dimensional array composed of the prestack seismic wave data. A reservoir prediction model is generated based on the three-dimensional array and the lithofacies annotation data; wherein, the reservoir prediction model is generated by processing the three-dimensional array and the lithofacies annotation data using the HR-Net algorithm; The reservoir distribution in an unknown exploration area is predicted using the reservoir prediction model, and the prediction results are obtained; wherein the prediction results are 3D data models in Segy format.

2. The method according to claim 1, characterized in that, The step of loading the pre-stack seismic wave data into the memory of the data processing device, and obtaining a three-dimensional array composed of the pre-stack seismic wave data by traversing the pre-stack seismic wave data in the data processing device, includes: Wellside data of the pre-stack seismic wave data is extracted from the well logging curves, wherein the wellside data corresponds to the depth of the well logging lithofacies data; Based on the wellbore data and the lithofacies annotation data, lithofacies interpretation results and conditional attributes are obtained; The three-dimensional array is determined based on the lithofacies interpretation results and the conditional attributes.

3. The method according to claim 1, characterized in that, The generation of the reservoir prediction model based on the three-dimensional array and the lithofacies annotation data includes: The model training set is obtained based on the three-dimensional array; wherein, the model training set includes a calibration set and a cross-validation set; The reservoir prediction model is obtained by processing the model training set using the HR-Net algorithm; wherein the reservoir prediction model is used to predict the reservoir distribution in the unknown exploration area.

4. The method according to claim 3, characterized in that, The step of processing the model training set using the HR-Net algorithm to obtain the reservoir prediction model includes: Based on the data size and data structure of the model training set, an HR-Net algorithm model is generated; wherein, the HR-Net algorithm model includes a loss function; The loss function is defined as a mixture of cross-entropy and BCE functions with different weights. The learning rate is optimized using the cosine annealing algorithm, and an adaptive gradient descent optimizer is used. The model training set is processed based on the loss function to obtain the reservoir prediction model.

5. The method according to claim 1, characterized in that, The process of predicting the reservoir distribution in an unknown exploration area using the reservoir prediction model and obtaining the prediction results includes: The reservoir prediction model is evaluated based on the condition attributes required by the reservoir prediction model, and the evaluation result of the reservoir prediction model is obtained. Based on the evaluation results, the target parameters in the HR-Net algorithm are determined; The reservoir prediction model is processed according to the target parameters to obtain the target optimized model of the reservoir prediction model; Based on the target optimization model, the reservoir distribution in the unknown exploration area is predicted, and the prediction results are obtained.

6. The method according to claim 5, characterized in that, The step of obtaining the target optimization model based on the target parameters includes: By comparing the lithofacies predicted for the cross-validation set in the target parameters, a confusion matrix is ​​calculated for supervised learning of the lithofacies identification method; The lithofacies transition matrix characterizing the changes between continuous facies is calculated based on the confusion matrix; The target optimization model of the reservoir prediction model is determined based on the lithofacies transition matrix and the confusion matrix.

7. A reservoir facies prediction device, characterized in that, include: The acquisition module is used to acquire prestack seismic wave data and lithofacies annotation data, and load the prestack seismic wave data into the memory of the data processing device. By traversing the prestack seismic wave data in the data processing device, a three-dimensional array composed of the prestack seismic wave data is obtained. A generation module is used to generate a reservoir prediction model based on the three-dimensional array and the lithofacies annotation data; wherein, the reservoir prediction model is generated by processing the three-dimensional array and the lithofacies annotation data using the HR-Net algorithm; The prediction module is used to predict the reservoir distribution in an unknown exploration area using the reservoir prediction model and obtain prediction results; wherein the prediction results are 3D data models in Segy format.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 6.

Images