Gas saturation prediction method and device based on multi-source information fusion and medium

Abstract

The invention provides a gas saturation prediction method and device based on multi-source information fusion and a medium. The method comprises the following steps: acquiring multi-source geological data of a reservoir to be predicted; performing time-frequency analysis on the pre-stack seismic data in the multi-source geological data, and determining the frequency division attribute of the pre-stack seismic data; determining a frequency gradient corresponding to the pre-stack seismic data according to the frequency division attribute; generating a region mask and a lithology mask according to the multi-source geological data; according to correlation analysis between the logging gas saturation data in the multi-source geological data and the frequency gradient and between the logging gas saturation data and other data in the multi-source geological data, determining prediction related characteristics; and obtaining a gas saturation prediction result of the reservoir to be predicted according to the region mask, the lithologic mask and the initial prediction result, the initial prediction result being a result obtained by the gas saturation prediction model through prediction based on the prediction related characteristics. According to the method, through analysis of multi-source data, the accuracy and reliability of gas saturation prediction are improved.

Description

Technical Field

[0001] The present application relates to the technical field of oil and gas geophysical exploration, and in particular to a method, device and medium for predicting gas saturation based on multi-source information fusion. Background Art

[0002] Seismic reservoir prediction technology plays a vital role in oil and gas exploration. With the deepening of exploration and development, tight sandstone reservoirs have become a new challenge for oil and gas exploration due to their deep burial, low porosity and permeability, complex pore structure, and strong heterogeneity. In high-water-content tight sandstone reservoirs, the difference between gas layers and water layers in seismic response and porosity and permeability is weak, which further increases the difficulty of predicting reservoir gas saturation.

[0003] Existing technologies usually use seismic inversion technology and machine learning technology to predict reservoir gas content. By learning the nonlinear relationship between seismic data and well logging data, a mapping function from input (seismic information, well logging information and geological information) to output (gas content prediction result) is established. For example, CNN (Convolutional Neural Network) can extract the spatial features of seismic data, and RNN (Recurrent Neural Network) can process the time series features of well logging data. Through this fusion of multiple models and multiple data sources, gas content prediction can be achieved.

[0004] However, the existing methods have the problem of poor prediction effect when predicting gas saturation. Summary of the invention

[0005] The embodiments of the present application provide a gas saturation prediction method, device and medium based on multi-source information fusion, so as to solve the problem that the existing methods have poor prediction effect when predicting gas saturation, and improve the prediction accuracy.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting gas saturation based on multi-source information fusion, comprising:

[0007] Obtain multi-source geological data of the reservoir to be predicted, including pre-stack seismic data, seismic attribute data, gamma inversion results, seismic lithology results, fault information, and well logging gas saturation data;

[0008] Conduct time-frequency analysis on pre-stack seismic data in multi-source geological data to determine the frequency division properties of pre-stack seismic data;

[0009] According to the frequency division attribute, the frequency gradient corresponding to the pre-stack seismic data is determined;

[0010] Generate regional masks and lithology masks based on multi-source geological data;

[0011] Determine prediction-related features based on correlation analysis between well logging gas saturation data and frequency gradient and other data in multi-source geological data;

[0012] According to the regional mask, lithology mask and initial prediction result, the gas saturation prediction result of the reservoir to be predicted is obtained. The initial prediction result is the result obtained by the gas saturation prediction model based on prediction-related features. The gas saturation prediction model is a machine learning model constructed based on the K nearest neighbor algorithm.

[0013] In a possible implementation, time-frequency analysis is performed on pre-stack seismic data in multi-source geological data to determine frequency-division attributes of the pre-stack seismic data, including:

[0014] Determine angle data of pre-stack seismic data, the angle data including near angle data, middle angle data and far angle data;

[0015] For each angle data in the pre-stack seismic data, time-frequency analysis is performed on the angle data to determine the target frequency of the angle data and the frequency division attribute corresponding to the target frequency, where the target frequency includes low frequency, medium frequency and high frequency;

[0016] Accordingly, according to the frequency division attribute, the frequency gradient corresponding to the pre-stack seismic data is determined, including:

[0017] For each angle data in the pre-stack seismic data, the frequency gradient of the angle data is determined according to each frequency division attribute corresponding to the angle data.

[0018] In a possible implementation, generating a regional mask and a lithology mask based on multi-source geological data includes:

[0019] Generate a regional mask of the reservoir to be predicted based on the fault information in the multi-source geological data, where the regional mask represents the positional relationship between the reservoir to be predicted and the fault;

[0020] Determine the sandstone and mudstone positions of the reservoir to be predicted based on the seismic lithology results in multi-source geological data;

[0021] According to the sandstone position and the mudstone position, a lithology mask of the reservoir to be predicted is generated, and the lithology mask includes a sandstone mask and a mudstone mask.

[0022] In a possible implementation, prediction-related features are determined based on correlation analysis between well logging gas saturation data and frequency gradient in multi-source geological data and other data in multi-source geological data, including:

[0023] Extract features of well logging gas saturation data, other data and frequency gradient in multi-source geological data to obtain features of well logging gas saturation data, other data features and frequency gradient features;

[0024] Performing correlation analysis on the well logging gas saturation data characteristics and the frequency gradient characteristics to generate a first analysis result, the first analysis result characterizing the influence of the frequency gradient characteristics on the gas saturation;

[0025] Performing correlation analysis on the well logging gas saturation data features and other data features to generate a second analysis result, the second analysis result characterizing the influence of other data features on the gas saturation;

[0026] According to the first analysis result and the second analysis result, prediction-related features are obtained.

[0027] In a possible implementation, obtaining prediction-related features according to the first analysis result and the second analysis result includes:

[0028] According to the first analysis result and the second analysis result, other data features and frequency gradient features are sorted to obtain a feature sorting result;

[0029] According to the preset prediction-related feature requirements and feature sorting results, prediction-related features are obtained from other data features and frequency gradient features.

[0030] In a possible implementation, before obtaining the gas saturation prediction result of the reservoir to be predicted according to the regional mask, the lithology mask and the initial prediction result, the method further includes:

[0031] According to the fault information of the reservoir to be predicted, a first target layer and a second target layer are determined, wherein the first target layer is a reservoir located above the fault in the reservoir to be predicted, and the second target layer is a reservoir located below the fault in the reservoir to be predicted;

[0032] Constructing a first gas saturation prediction model and a second gas saturation prediction model for the first target layer; and a third gas saturation prediction model and a fourth gas saturation prediction model for the second target layer, wherein the first gas saturation prediction model and the third gas saturation prediction model are used to predict the gas saturation of the reservoir with sandstone attributes in the reservoir to be predicted, and the second gas saturation prediction model and the fourth gas saturation prediction model are used to predict the gas saturation of the reservoir with mudstone attributes in the reservoir to be predicted;

[0033] According to the prediction-related characteristics of each target layer in the reservoir to be predicted, the initial prediction results corresponding to each gas saturation prediction model in each target layer are determined.

[0034] In a possible implementation, the gas saturation prediction result satisfies:

[0035] Sg'=Mf(Ms*KNN1(X)+Md*KNN2(X))+(1-Mf)(Ms*KNN3(X)+Md*KNN4(X));

[0036] Among them, Sg' is the gas saturation prediction result; Mf is the regional mask; Ms is the sandstone mask; Md is the mudstone mask; KNN1 is the first gas saturation prediction model; KNN2 is the second gas saturation prediction model; KNN3 is the third gas saturation prediction model; KNN4 is the fourth gas saturation prediction model; X is the prediction related feature.

[0037] In a second aspect, an embodiment of the present application provides a gas saturation prediction device based on multi-source information fusion, comprising:

[0038] An acquisition module is used to acquire multi-source geological data of the reservoir to be predicted, wherein the multi-source geological data includes pre-stack seismic data, seismic attribute data, gamma-ray inversion results, seismic lithology results, fault information, and well logging gas saturation data;

[0039] The first analysis module is used to perform time-frequency analysis on pre-stack seismic data in multi-source geological data to determine the frequency division properties of the pre-stack seismic data;

[0040] A determination module, for determining a frequency gradient corresponding to pre-stack seismic data according to a frequency division attribute;

[0041] A generation module, used to generate regional masks and lithology masks based on multi-source geological data;

[0042] The second analysis module is used to determine the prediction related features based on the correlation analysis between the well logging gas saturation data and the frequency gradient and other data in the multi-source geological data;

[0043] The prediction module is used to obtain the gas saturation prediction result of the reservoir to be predicted based on the regional mask, lithology mask and initial prediction result. The initial prediction result is the result obtained by the gas saturation prediction model based on prediction-related features. The gas saturation prediction model is a machine learning model constructed based on the K nearest neighbor algorithm.

[0044] In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0045] Memory stores computer-executable instructions;

[0046] The processor executes the computer-executable instructions stored in the memory to implement the above first aspect and / or various possible implementations of the first aspect.

[0047] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed, they are used to implement the first aspect above and / or various possible implementation methods of the first aspect.

[0048] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0049] The gas saturation prediction method, device and medium based on multi-source information fusion provided in the embodiment of the present application obtain multi-source geological data of the reservoir to be predicted, and the multi-source geological data include pre-stack seismic data, seismic attribute data, gamma inversion results, seismic lithology results, fault information, and well logging gas saturation data; perform time-frequency analysis on the pre-stack seismic data in the multi-source geological data to determine the frequency division attributes of the pre-stack seismic data; determine the frequency gradient corresponding to the pre-stack seismic data according to the frequency division attributes; generate a regional mask and a lithology mask according to the multi-source geological data; determine the prediction according to the correlation analysis between the well logging gas saturation data and the frequency gradient in the multi-source geological data and other data in the multi-source geological data. According to the regional mask, lithology mask and initial prediction result, the gas saturation prediction result of the reservoir to be predicted is obtained. The initial prediction result is the result obtained by the gas saturation prediction model based on the prediction-related characteristics. The gas saturation prediction model is a machine learning model constructed based on the K nearest neighbor algorithm. By integrating a variety of geological data sources, it takes advantage of the advantages of different data sources to provide more comprehensive and accurate geological information. Then, the position masks are established respectively using the fault and lithology information, and combined with the machine learning model of the K nearest neighbor algorithm, the uncertainty caused by the complexity of geological conditions is reduced, and the accurate prediction of gas saturation under different lithology hard constraints in different fault division areas is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0051] Figure 1 A flow chart of a gas saturation prediction method based on multi-source information fusion provided in this application;

[0052] Figure 2 A schematic flow chart of another gas saturation prediction method based on multi-source information fusion provided in this application;

[0053] Figure 3 A specific flow chart of a gas saturation prediction method based on multi-source information fusion provided in this application;

[0054] Figure 4 A schematic diagram of actual pre-stack seismic data provided for this application;

[0055] Figure 5 A schematic diagram of a frequency gradient of pre-stack seismic data provided in this application;

[0056] Figure 6 The well logging gas saturation curve and the curve diagram of each characteristic data provided in this application;

[0057] Figure 7a A schematic diagram of the correlation analysis between the gas saturation curve and various characteristics provided in this application;

[0058] Figure 7b A schematic diagram of the feature correlation ranking results provided for this application;

[0059] Figure 8 A schematic diagram of the results of a gas saturation prediction profile provided in this application;

[0060] Fig. 9 A schematic diagram showing the comparison between the predicted gas saturation curve and the actual logging curve of a test well A5 provided in this application;

[0061] Fig.10 A schematic diagram of a cross-validation model test result provided in this application;

[0062] Fig.11 A schematic diagram of a target layer slice of a gas saturation prediction result provided in this application;

[0063] Fig.12 A schematic diagram of the structure of a gas saturation prediction device based on multi-source information fusion provided in this application;

[0064] Fig.13 A schematic diagram of the structure of the electronic device provided in this application.

[0065] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0066] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0067] In the prior art, seismic inversion technology and machine learning technology are usually used to predict the gas content of reservoirs. By learning the nonlinear relationship between seismic data and well logging data, a mapping function from input (seismic information, well logging information and geological information) to output (gas content prediction result) is established. For example, CNN can extract the spatial features of seismic data, and RNN can process the time series features of well logging data. Through the fusion of multiple models and multiple data sources, gas content prediction is achieved. However, in practical applications, the fusion of multi-source data by existing machine learning models is usually completed at the input feature level, while the integration of geological information and prior knowledge is relatively limited, resulting in low prediction accuracy of gas saturation. In addition, as a black box model, the machine learning model has poor interpretability of the prediction results. Multi-source data is integrated into the prediction process in the form of input features, which is a soft constraint. In the process of model parameter learning, the model is limited by geological information and prior knowledge, and is more inclined to learn input features with higher correlation. The amount of data required for model training is very large and the training time is long, resulting in low prediction efficiency.

[0068] In order to solve the above problems, the embodiment of the present application provides a method, device and medium for predicting gas saturation based on multi-source information fusion, using a variety of geological data sources, including pre-stack seismic data, seismic attribute data, gamma inversion results, seismic lithology results, fault information and well logging gas saturation data. By performing time-frequency analysis on pre-stack seismic data, extracting frequency division attributes and frequency gradients, and combining other geological data to generate regional masks and lithology masks. Then, by analyzing the correlation between well logging gas saturation data and frequency gradients and other geological data, the prediction-related features are determined. Finally, the machine learning model constructed using the K nearest neighbor algorithm is used for initial prediction, and a more accurate gas saturation prediction result is obtained by combining regional masks and lithology masks. By fusing multi-source geological data and comprehensively utilizing information from different data sources, the accuracy of gas saturation prediction is improved: by generating regional and lithology masks, the prediction uncertainty caused by the complexity of geological conditions can be effectively reduced; using machine learning models (such as the K nearest neighbor algorithm) for prediction can better capture the relationship between complex data and improve the reliability of prediction.

[0069] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0070] The execution subject of the gas saturation prediction method based on multi-source information fusion provided in the embodiment of the present application may be a server, wherein the server may be a mobile phone, a computer, a tablet or other device. The embodiments of the present application do not impose any particular restrictions on the implementation method of the execution subject, as long as the execution subject can obtain multi-source geological data of the reservoir to be predicted, the multi-source geological data includes pre-stack seismic data, seismic attribute data, gamma inversion results, seismic lithology results, fault information, and well logging gas saturation data; perform time-frequency analysis on the pre-stack seismic data in the multi-source geological data to determine the frequency division attributes of the pre-stack seismic data; determine the frequency gradient corresponding to the pre-stack seismic data based on the frequency division attributes; generate regional masks and lithology masks based on the multi-source geological data; determine prediction-related features based on the correlation analysis between the well logging gas saturation data and the frequency gradient in the multi-source geological data and other data in the multi-source geological data; obtain the gas saturation prediction result of the reservoir to be predicted based on the regional mask, lithology mask and initial prediction result, the initial prediction result is the result obtained by the gas saturation prediction model based on the prediction-related features, and the gas saturation prediction model is a machine learning model constructed based on the K nearest neighbor algorithm.

[0071] In seismic exploration, pre-stack usually refers to the state before the seismic data is stacked.

[0072] Figure 1 The flowchart of a method for predicting gas saturation based on multi-source information fusion provided in this application is shown in FIG. The execution subject of the method may be a server or other server storing the method for predicting gas saturation based on multi-source information fusion. This embodiment is not particularly limited here. Figure 1 As shown, the method may include:

[0073] S101. Acquire multi-source geological data of a reservoir to be predicted, wherein the multi-source geological data includes pre-stack seismic data, seismic attribute data, gamma-ray inversion results, seismic lithology results, fault information, and well logging gas saturation data.

[0074] Among them, multi-source geological data can refer to geological related data obtained from multiple sources. Multi-source geological data can also include other data related to gas saturation prediction, such as impedance inversion results and low-frequency trends of gas saturation. Adding relevant data of impedance inversion results can provide additional information about the elastic properties of the formation, improve the ability to identify reservoir fluid and lithological characteristics, and thus improve the accuracy of gas saturation prediction. Adding low-frequency trends of gas saturation can provide information about the macroscopic change trend of gas saturation. Low-frequency trends can help capture large-scale changes in gas saturation and provide a basic background trend, so that the prediction model can better correct local anomalies and improve the stability of the overall prediction.

[0075] It should be noted that the multi-source geological data or fault information can be data obtained based on a two-dimensional well profile or data obtained based on a three-dimensional space.

[0076] Multi-source geological data of the reservoir to be predicted is obtained by collecting and integrating data from different geological surveys and measurement methods. For example, pre-stack seismic data is obtained using seismic exploration technology, gas saturation data is obtained using well logging technology, and seismic attributes and lithology results are extracted from geological modeling software.

[0077] In one example, the pre-stack seismic data may include near-angle data S n , medium angle data S m , Far angle data S f , that is, S = [S n ,S m ,S f ]; Seismic attribute data may include amplitude envelope, 90-degree phase shift, amplitude second derivative, instantaneous amplitude, instantaneous frequency, instantaneous phase and other attribute data, namely

[0078] S102: Perform time-frequency analysis on pre-stack seismic data in the multi-source geological data to determine frequency division properties of the pre-stack seismic data.

[0079] In this step, time-frequency analysis may refer to a method of analyzing the distribution of a signal in time and frequency, and frequency division properties may refer to the characteristics of a seismic signal in different frequency ranges.

[0080] The method for performing time-frequency analysis on pre-stack seismic data may be to use methods such as short-time Fourier transform and wavelet transform to extract seismic signal characteristics at different frequencies.

[0081] For example, wavelet transform is used to analyze seismic data to extract energy distribution in different frequency bands; short-time Fourier transform is applied to obtain the frequency spectrum characteristics of seismic signals.

[0082] In one example, the frequency division properties of pre-stack seismic data are obtained based on the optimal order S transform. The S transform is a method that combines the characteristics of short-time Fourier transform (STFT) and wavelet transform. The S transform satisfies:

[0083]

[0084] in, represents the S transform at frequency ω and time t, d(t) represents the amplitude value of the prestack seismic signal d, and τ as the integral variable represents the time domain range of the entire signal. g(t-τ,ω) is a Gaussian window function centered at t, and the window width is controlled by the frequency ω.

[0085] Therefore, based on the above S transformation, the frequency division property of prestack seismic data can be obtained as follows: d = S (w, t; S). When there are different frequencies w, such as low frequency w 1 、Intermediate frequency w 2 , high frequency w 3 When , the obtained frequency division property can be

[0086] S103. Determine the frequency gradient corresponding to the pre-stack seismic data according to the frequency division attribute.

[0087] Among them, the frequency gradient can characterize the change of frequency attributes in space or time. By calculating the change rate of different frequency attributes in prestack seismic data, the frequency gradient is formed, and the change of geological structure can be identified.

[0088] Furthermore, the frequency gradient can be obtained by calculating the spatial derivative of the frequency attribute; the frequency gradient corresponding to the prestack seismic data can be determined using the three-point difference method.

[0089] For example, when there are different frequencies w, the frequency gradient obtained by the three-point difference method is

[0090] S104. Generate a regional mask and a lithology mask based on multi-source geological data.

[0091] In this step, the region mask may be a mask used to identify a specific geographical region in the reservoir, and the lithology mask may be a mask used to identify specific lithology features.

[0092] Furthermore, masks are generated based on geological data so as to take into account the spatial variation of geological conditions and lithological characteristics in the prediction process. For example, lithological masks are generated using seismic lithological results to identify the distribution of different lithologies; regional masks are generated based on fault information to identify fault-affected areas.

[0093] S105. Determine prediction-related features based on correlation analysis between well logging gas saturation data and frequency gradient in the multi-source geological data and other data in the multi-source geological data.

[0094] Among them, correlation analysis can be a statistical method for determining the relationship between different variables. Prediction-related features can refer to features that have an important influence in the prediction model.

[0095] In this step, the relationship between the well logging data and other geological data is evaluated by statistical analysis methods (such as Pearson correlation coefficient, regression analysis) to identify the features that have a significant impact on the prediction of gas saturation.

[0096] For example, the Pearson correlation coefficient is used to analyze the relationship between well logging gas saturation and frequency gradient. The impact of seismic attribute data on gas saturation is determined through regression analysis.

[0097] S106. Obtain a gas saturation prediction result of the reservoir to be predicted based on the regional mask, lithology mask and initial prediction result. The initial prediction result is a result obtained by a gas saturation prediction model based on prediction-related features. The gas saturation prediction model is a machine learning model constructed based on a K nearest neighbor algorithm.

[0098] The initial prediction result may be a preliminary prediction result obtained based on prediction-related features and a machine learning model. The K nearest neighbor algorithm is an instance-based non-parametric learning algorithm for classification and regression.

[0099] Furthermore, the machine learning model constructed using the K nearest neighbor algorithm was used for initial prediction, and then adjusted in combination with the regional and lithology masks to obtain the final gas saturation prediction results.

[0100] In one example, the K nearest neighbor algorithm is used to model prediction-related features to obtain initial prediction results. The initial prediction results are spatially adjusted using regional and lithology masks to obtain more accurate prediction results.

[0101] The gas saturation prediction method based on multi-source information fusion provided in the embodiment of the present application significantly improves the accuracy and reliability of gas saturation prediction through the fusion and analysis of multi-source geological data; extracts key features using time-frequency analysis and correlation analysis, and combines the machine learning model of the K nearest neighbor algorithm to achieve data-driven accurate prediction; and further reduces the uncertainty caused by the complexity of geological conditions by generating regional and lithology masks, enhances the spatial resolution and adaptability of the prediction, and improves the accuracy and reliability of gas saturation prediction.

[0102] On the basis of the above-mentioned embodiment, the method of performing time-frequency analysis on pre-stack seismic data in multi-source geological data in S102 to determine the frequency division properties of the pre-stack seismic data may include: determining angle data of the pre-stack seismic data, the angle data including near-angle data, medium-angle data and far-angle data; for each angle data in the pre-stack seismic data, performing time-frequency analysis on the angle data to determine the target frequency of the angle data and the frequency division properties corresponding to the target frequency, the target frequency including low frequency, medium frequency and high frequency;

[0103] Accordingly, the method of determining the frequency gradient corresponding to the prestack seismic data according to the frequency division attribute in S103 may include: for each angle data in the prestack seismic data, determining the frequency gradient of the angle data according to each frequency division attribute corresponding to the angle data.

[0104] Among them, angle data can refer to the incident angle of seismic waves in seismic exploration, which will affect the characteristics of the received signal. Angle data is usually divided into near angle, medium angle and far angle, corresponding to different incident angles. Target frequency can refer to a specific frequency range of interest in time-frequency analysis, including low frequency, medium frequency and high frequency. For each angle data, it can refer to analyzing the near angle, medium angle and far angle data separately.

[0105] In one example, when the prestack seismic data includes near-angle data S n , medium angle data S m , Far angle data S f , and the target frequency of each angle data includes the low frequency w 1 、Intermediate frequency w 2 , high frequency w 3 When the near angle data S n Perform time-frequency analysis and obtain the frequency division properties: Medium angle data S m Perform time-frequency analysis and obtain the frequency division properties: Far angle data S f Perform time-frequency analysis and obtain the frequency division properties:

[0106] Correspondingly, the near angle data S n The corresponding frequency gradient is Medium angle data S m The corresponding frequency gradient is Far angle data S f The corresponding frequency gradient is

[0107] By segmenting and analyzing the angle data of pre-stack seismic data, geological features at different depths and angles can be more accurately identified. Time-frequency analysis combined with frequency gradient calculation can provide more detailed geological information and improve the accuracy of gas saturation prediction. The specific implementation method can be adjusted according to specific geological conditions and exploration goals to optimize the prediction results.

[0108] On the basis of the above-mentioned embodiment, the method for generating a regional mask and a lithology mask according to multi-source geological data in S104 may include: generating a regional mask of the reservoir to be predicted according to the fault information in the multi-source geological data, the regional mask characterizing the positional relationship between the reservoir to be predicted and the fault; determining the sandstone position and mudstone position of the reservoir to be predicted according to the seismic lithology results in the multi-source geological data; generating a lithology mask of the reservoir to be predicted according to the sandstone position and the mudstone position, the lithology mask including a sandstone mask and a mudstone mask.

[0109] Among them, the regional mask can be used to represent the spatial relationship between the reservoir and geological features (such as faults), helping to identify the location of the reservoir in the geological structure. The lithology mask can be used to represent the distribution of different lithologies (such as sandstone and mudstone) in the reservoir, helping to identify different lithology areas for more accurate geological analysis. Sandstone location and mudstone location can refer to the specific spatial location of sandstone and mudstone identified in the seismic lithology results.

[0110] Furthermore, the spatial relationship between the reservoir and the fault is determined using fault information. The fault information can be obtained through seismic exploration data or geological models, and a regional mask is created to identify the relative positions of the reservoir and the fault. The seismic lithology results are used to identify the spatial distribution of sandstone and mudstone in the reservoir, and sandstone masks and mudstone masks are created to respectively identify the positions of sandstone and mudstone in the reservoir.

[0111] Among them, each value M in the region mask Mf f (x,y) is represented as:

[0112]

[0113] Wherein, Fy represents the vertical coordinate position of the fault.

[0114] When the sandstone mask Ms is 1, it indicates the sandstone position, and when the sandstone mask Ms is 0, it indicates other positions except sandstone; when the mudstone mask Md is 1, it indicates the mudstone position, and when the mudstone mask Md is 0, it indicates other positions except mudstone.

[0115] For example, fault data can be analyzed using seismic interpretation software to generate a binary mask identifying areas in the reservoir close to faults, or geological modeling tools can be applied to generate a three-dimensional regional mask in combination with fault dip and strike information. Seismic attribute analysis can be used to identify high-amplitude areas as sandstone locations and low-amplitude areas as mudstone locations, and machine learning algorithms (such as support vector machines or random forests) can be used to classify seismic lithology results to generate masks for sandstone and mudstone.

[0116] By generating regional masks and lithology masks, the relationship between reservoirs and geological features (such as faults) and lithology distribution can be more accurately described, and the complexity of geological conditions can be better considered in gas saturation prediction, thereby improving the accuracy and reliability of prediction. Different implementation methods can be adjusted according to specific geological data and analysis tools to optimize the generation and application of masks.

[0117] On the basis of the above-mentioned embodiment, in S105, according to the correlation analysis between the well logging gas saturation data and the frequency gradient and other data in the multi-source geological data, the method for determining the prediction related features may include: extracting features of the well logging gas saturation data, other data in the multi-source geological data, and the frequency gradient in the multi-source geological data to obtain well logging gas saturation data features, other data features, and frequency gradient features; performing correlation analysis on the well logging gas saturation data features and the frequency gradient features to generate a first analysis result, the first analysis result characterizing the influence of the frequency gradient features on the gas saturation; performing correlation analysis on the well logging gas saturation data features and other data features to generate a second analysis result, the second analysis result characterizing the influence of other data features on the gas saturation; and obtaining the prediction related features according to the first analysis result and the second analysis result.

[0118] In this step, other data features may be features obtained by extracting features from other data in the multi-source geological data except the well logging gas saturation data.

[0119] For example, the well logging gas saturation data feature can be a feature describing the gas saturation extracted from the well logging data, such as the mean value, variance, maximum value, minimum value, etc. Other data features can be related features extracted from other geological data, such as the amplitude, frequency, phase, etc. of seismic attributes. Frequency gradient features can be features extracted from frequency gradient data, such as the mean value and rate of change of the gradient, etc. Optionally, these features need to be normalized or standardized before correlation analysis is performed to solve the problem of inaccurate correlation analysis caused by different dimensions and ranges of different data features.

[0120] Further, based on the first analysis result and the second analysis result, features that have a significant impact on the prediction of gas saturation are selected as inputs of the prediction model. For example, a correlation coefficient is set, and based on the correlation analysis result, features with a higher correlation coefficient are selected as model inputs; or a machine learning algorithm (such as recursive feature elimination) is used to automatically select the features with the most predictive ability.

[0121] Through feature extraction and correlation analysis, the most influential features for gas saturation prediction can be identified. This method can help optimize the input of the prediction model and improve the accuracy and reliability of the model. The implementation method can be adjusted according to the specific data characteristics and analysis requirements to optimize the feature selection process.

[0122] On the basis of the above-mentioned embodiment, according to the first analysis result and the second analysis result, the method for obtaining prediction-related features may include: according to the first analysis result and the second analysis result, sorting other data features and frequency gradient features to obtain feature sorting results; according to preset prediction-related feature requirements and feature sorting results, obtaining prediction-related features from other data features and frequency gradient features.

[0123] Using the results of the correlation analysis, other data features and frequency gradient features are sorted, where the sorting basis can be the absolute value of the correlation coefficient, the contribution of the feature to the prediction accuracy, etc. The feature sorting result can be a sorted list or a sorted graph to show all features and their corresponding importance indicators.

[0124] The preset prediction-related feature requirements can be pre-set standards or conditions for selecting which features should be included in the prediction model, for example, selecting the top N (such as the top 4 in the ranking) most important features, or selecting features whose correlation coefficient exceeds a certain threshold (such as a correlation coefficient greater than 0.2).

[0125] Through feature sorting and selection, the features that have the greatest impact on gas saturation prediction can be effectively identified, which can help optimize the model input and improve the accuracy and efficiency of prediction.

[0126] On the basis of the above embodiment, before S106, the method also includes: determining the first target layer and the second target layer according to the fault information of the reservoir to be predicted, the first target layer being the reservoir located above the fault in the reservoir to be predicted, and the second target layer being the reservoir located below the fault in the reservoir to be predicted; constructing a first gas saturation prediction model and a second gas saturation prediction model for the first target layer; and a third gas saturation prediction model and a fourth gas saturation prediction model for the second target layer, the first gas saturation prediction model and the third gas saturation prediction model are used to predict the gas saturation of the reservoir under sandstone attributes in the reservoir to be predicted, and the second gas saturation prediction model and the fourth gas saturation prediction model are used to predict the gas saturation of the reservoir under mudstone attributes in the reservoir to be predicted; determining the initial prediction results corresponding to each gas saturation prediction model in each target layer according to the prediction-related characteristics of each target layer in the reservoir to be predicted.

[0127] Among them, the fault position in the reservoir is identified by using seismic data or geological models; according to the spatial position of the fault, the reservoir is divided into a first target layer above and a second target layer below. It should be noted that the above and below here are used to distinguish the position space on both sides of the fault position, not limited to the upper and lower positions, and the actual positions on both sides of the fault position need to be determined according to the specific setting requirements of the regional mask.

[0128] The K-Nearest Neighbors (KNN) algorithm was used to build the model, and training was performed on the characteristics of different lithologies in different regions.

[0129] Based on the above method, the gas saturation prediction results in S106 can meet the following requirements:

[0130] Sg'=Mf(Ms*KNN1(X)+Md*KNN2(X))+(1-Mf)(Ms*KNN3(X)+Md*KNN4(X));

[0131] Among them, Sg' is the gas saturation prediction result; Mf is the regional mask; Ms is the sandstone mask; Md is the mudstone mask; KNN1 is the first gas saturation prediction model; KNN2 is the second gas saturation prediction model; KNN3 is the third gas saturation prediction model; KNN4 is the fourth gas saturation prediction model; X is the prediction related feature.

[0132] By dividing the target layer according to the fault information and establishing different KNN gas saturation prediction models for different lithologies in different regions, the initial prediction results of the gas saturation are then integrated through masks to obtain the final gas saturation prediction results. The multivariate information (seismic, well logging, geology) is integrated with prior knowledge (fault separation areas), and the influence of geological structure and lithology changes on gas saturation can be considered, thereby improving the prediction accuracy.

[0133] Figure 2 A flow chart of another method for predicting gas saturation based on multi-source information fusion provided in an embodiment of the present application. Figure 2 As shown, the method may include:

[0134] S201. Data preparation.

[0135] In this step, the acquired data may include: pre-stack seismic data S = [S n ,S m ,S f ], earthquake attribute data (representing amplitude envelope, 90-degree phase shift, second-order derivative of amplitude, instantaneous amplitude, instantaneous frequency and instantaneous phase respectively), impedance inversion result Ip, gamma inversion result Gr, gas saturation low-frequency trend Sgl, seismic lithology prediction result L, fault F(x,y) and well logging gas saturation data Sg.

[0136] S202. Calculate frequency gradient data of pre-stack seismic data based on optimal order S-transform.

[0137] Based on the S transform, the frequency division properties of the angle-divided seismic data are obtained:

[0138]

[0139] in, to Respectively represent the frequency division properties at different angles. 1 、w 2 、w 3 The target frequencies correspond to low frequency, medium frequency and high frequency (such as 10Hz, 20Hz and 30Hz) respectively.

[0140] The frequency gradient of the prestack gather is calculated by three-point difference:

[0141]

[0142] Among them, G n is the frequency gradient corresponding to the near-angle data; G m is the frequency gradient corresponding to the mid-angle data; G f is the frequency gradient corresponding to the far-angle data.

[0143] S203. Masks are established respectively based on the fault position and the seismic lithology prediction results.

[0144] Since faults can seal, divert or destroy reservoirs, their impact on gas saturation should be considered when predicting reservoirs. The regional mask Mf can be obtained according to the fault location, and the reservoir gas saturation model can be learned for different regions. f (x,y) is represented as:

[0145]

[0146] Wherein, Fy represents the vertical coordinate position of the fault.

[0147] Seismic lithology prediction results can also constrain reservoir gas saturation prediction. There is a large difference in gas saturation between sandstone and mudstone. The accuracy of gas saturation prediction can be further improved by lithology constraints. The sandstone mask Ms (i.e., sandstone position is 1, other positions are 0) and the mudstone mask Md (i.e., mudstone position is 1, other positions are 0) can be obtained through the lithology result L.

[0148] S204. Perform sensitivity analysis on existing features and select gas saturation sensitive attributes as model input.

[0149] By performing correlation analysis on the existing features and sorting the correlations from large to small, the first feature is selected as the model input X.

[0150] S205, using the KNN model and the mask to predict the gas saturation of different lithologies in different regions, and combining them based on the mask to obtain the final gas saturation prediction result.

[0151] Among them, the gas saturation prediction results can meet the following requirements:

[0152] Sg'=Mf(Ms*KNN1(X)+Md*KNN2(X))+(1-Mf)(Ms*KNN3(X)+Md*KNN4(X)).

[0153] The gas saturation prediction method based on multi-source information fusion provided in the embodiment of the present application establishes fault area masks and lithology masks for the areas separated by faults and different lithologies (sandstone and mudstone). Different KNN gas saturation prediction models are established for different lithologies in different areas, and the gas saturation prediction results are integrated through masks to obtain the final gas saturation prediction results. Through this process, multivariate information (seismic, well logging, geology) can be fused with prior knowledge (fault-separated areas), and interpretable gas saturation prediction can be achieved through the KNN model, thereby improving the prediction accuracy.

[0154] Figure 3 A specific flow chart of a method for predicting gas saturation based on multi-source information fusion provided in an embodiment of the present application is as follows: Figure 3 As shown in Figure 1, sub-figure (a) shows the principle of the KNN model. The KNN model is a non-parametric method based on similarity. When predicting the output of a sample, the KNN model first calculates the correlation coefficient between the sample and each sample in the training set, and then selects the K samples closest to it as neighbors. The output values ​​of these K neighbors are combined by weighted average of spatial distance to obtain the prediction result. Sub-figure (b) shows the process of using the KNN model to fuse multivariate information to predict gas saturation. Seismic and well logging information are integrated into the process in the form of samples. The input features include the frequency division properties of pre-stack seismic data, the results of longitudinal impedance inversion, the results of gamma curve inversion, and the low-frequency trend of the gas saturation curve. The label is the well logging gas saturation curve. At the same time, geological information (faults, lithology) is integrated into the process by modifying the KNN model. In sub-figure (b), KNN1 and KNN2 respectively represent the establishment of different gas saturation prediction models on both sides of the fault to learn different gas saturation characteristics, so that accurate prediction of gas saturation can be achieved in high water-content tight sandstone reservoirs. It has good interpretability and can achieve efficient incremental gas saturation prediction when there is incremental data (such as new drilling and logging data).

[0155] Next, taking the high-water-content tight sandstone reservoir data of a certain actual work area as an example, it is explained how to use the gas saturation prediction method based on multi-source information fusion provided in the embodiment of the present application to realize the gas saturation prediction of the reservoir. Figure 4 A schematic diagram of actual pre-stack seismic data provided for this application, such as Figure 4As shown in the figure, horizontal and vertical lines are terms used to describe the direction of seismic data acquisition. The vertical line usually refers to the main direction of seismic exploration (usually the direction of the exploration line), while the horizontal line refers to the direction perpendicular to the vertical line. Combining sub-figure (a) - a well-connected profile in the work area, sub-figure (b) - near-angle prestack seismic data, sub-figure (c) - medium-angle prestack seismic data, and sub-figure (d) - far-angle prestack seismic data, it can be seen that the seismic data has a total of 1200 channels, each of which has 400 time sampling points. The thickness of the target layer is about 50ms. According to well logging analysis, thin reservoirs are relatively developed in the target layer. At the same time, there are 8 wells (A1, A2, A3, A4, A5, A6, A7, A8) on the seismic profile that vertically pass through the target layer. From the profile, we can see that there is an obvious fault in the part from CDP=400 to CDP=600. The fault may lead to the redistribution of underground fluids. Therefore, the fault information can be incorporated into the reservoir prediction model as a priori information. Among them, CDP (CDP gather) can be represented as the X coordinate of the profile in the seismic data. When the seismic data has a total of 1200 channels, there are 1200 corresponding CDP gather values.

[0156] The frequency gradients of near-angle, medium-angle and far-angle seismic data can be calculated from pre-stack seismic data by using the optimal order S transform, such as Figure 5 As shown, sub-figure (a) is the frequency gradient of near-angle seismic data, sub-figure (b) is the frequency gradient of medium-angle seismic data, and sub-figure (c) is the frequency gradient of far-angle seismic data.

[0157] Take Well A6 among the above 8 wells as an example. Figure 6 As shown in the figure, the lithology distribution, gas saturation curve, gamma curve and various seismic attributes of Well A6 are shown. The target reservoir is located between 1000 and 1050 ms, and the main lithologies are sandstone and mudstone. Different lithologies also have a significant impact on the preservation and flow of fluids. Therefore, incorporating lithology information into the model can enhance the spatial constraint ability of the model and help predict the location of gas-bearing reservoirs more accurately. It should be noted that Figure 6 The lithology attributes in represent the seismic lithology prediction results in the above data, Lithology; the well logging gas saturation data can be expressed as Sg; the gamma curve represents the gamma inversion result, Gr; the frequency gradient represents the frequency gradient attributes at near, middle and far angles, FreqGrad; the intercept attribute is the P attribute; the gradient attribute is the G attribute; the 90-degree phase shift attribute is the QT attribute; the second-order derivative attribute is the SD attribute; the instantaneous frequency attribute is InstFreq; the instantaneous amplitude attribute is the InstAmp attribute; the instantaneous phase is the InstPhase attribute; the amplitude envelope attribute is the AmpEnv attribute.

[0158] The gas saturation curve is analyzed with multiple features to determine the sensitive features with strong correlation with the gas saturation curve, such as Figure 7a As shown, sub-figure (a) is the intersection diagram of P attribute and gas saturation; sub-figure (b) is the intersection diagram of G attribute and gas saturation; sub-figure (c) is the intersection diagram of 90-degree phase shift attribute and gas saturation; sub-figure (d) is the intersection diagram of second-order derivative attribute and gas saturation; sub-figure (e) is the intersection diagram of gamma curve and gas saturation; sub-figure (f) is the intersection diagram of near-angle frequency gradient attribute and gas saturation; sub-figure (g) is the intersection diagram of medium-angle frequency gradient attribute and gas saturation; sub-figure (h) is the intersection diagram of far-angle frequency gradient attribute and gas saturation; sub-figure (i) is the intersection diagram of instantaneous frequency attribute and gas saturation; sub-figure (j) is the intersection diagram of instantaneous amplitude attribute and gas saturation; sub-figure (k) is the intersection diagram of instantaneous phase attribute and gas saturation; sub-figure (l) is the intersection diagram of amplitude envelope attribute and gas saturation. Figure 7b A schematic diagram of the feature correlation ranking results provided for this application.

[0159] from Figure 7a From the results of the various intersection diagrams shown, AVO (Amplitude Variation with Offset) attributes such as P and G attributes are less sensitive to the gas saturation curve, while the 90-degree phase shift attribute, prestack frequency gradient attribute (near, middle, and far angles), and gamma curve are more correlated with the gas saturation curve. Figure 7b Based on the correlation ranking results, the 90-degree phase shift attribute, the near-angle frequency gradient, the medium-angle frequency gradient, the far-angle frequency gradient, and gamma are selected as the input features of the model. Optionally, the A5 well can be used as a test set to test the accuracy of the model, and the remaining wells can be used as training sets to train the model.

[0160] The hyperparameter k in the KNN model is determined through multiple experiments. When k=6, the gas saturation prediction profile of the well section is obtained as follows: Figure 8 As shown, it can be observed that the gas saturation prediction results have a good correspondence with the well logging gas saturation curve near the well location and have a high vertical resolution.

[0161] The trained model is used for testing in Well A5. Fig. 9As shown, sub-figure (a) shows the comparison between the prediction result (solid line) of the gas saturation prediction method based on multi-source information fusion provided in the embodiment of the present application and the well logging gas saturation curve (dashed line); sub-figure (b) shows the comparison between the prediction result (solid line) of the conventional KNN model and the well logging gas saturation curve (dashed line). The correlation coefficient of the prediction result of the gas saturation prediction method based on multi-source information fusion provided in the embodiment of the present application is 0.73, while the correlation coefficient of the prediction result of the conventional KNN model is 0.57.

[0162] In order to verify the prediction stability of the model, the cross-validation method was used for testing. For example, each well was used as a test well, and the other wells were used as training wells to establish multiple test models. The prediction results were compared with the test wells. At the same time, the conventional KNN model was used as a comparison model for testing. The test results are shown in Figure 2. Fig.10 As shown, among them, when sub-figure (a) A1 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model; when sub-figure (b) A2 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model; when sub-figure (c) A3 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model; when sub-figure (d) A4 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model; when sub-figure (e) A5 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model; when sub-figure (f) A6 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model; when sub-figure (g) A7 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model; when sub-figure (h) A8 well is a test well, the prediction results of the present invention are compared with those of the conventional KNN model.

[0163] It can be seen that in most of the test wells, the prediction effect of the prediction model of the gas saturation prediction method based on multi-source information fusion provided in the embodiment of the present application is better than that of the conventional KNN model.

[0164] Based on the good results of predicting gas saturation with two-dimensional well-connected profiles, further research on predicting gas saturation in three-dimensional space was carried out. The goal of three-dimensional gas saturation prediction is to predict the spatial distribution of gas-bearing reservoirs by jointly analyzing seismic data and well location information in the entire work area. In order to achieve this goal, the prediction method of two-dimensional profiles is extended to three dimensions, and the spatial position of each well location in the work area and its relationship with the gas saturation characteristics of the target layer are considered.

[0165] First, based on the pre-stack seismic data in the work area, angle stacking was performed to obtain angle seismic data volumes. This data volume covers the entire work area, with 795 main survey lines and 1011 contact survey lines. Each seismic data has 400 time sampling points. At the same time, a total of 19 wells in the work area vertically pass through the target layer. Through the analysis of the well data, important features such as the gas saturation curve, gamma curve and lithology information at the well point were obtained. These well data not only provide important target parameters for three-dimensional gas saturation prediction, but also provide high-quality label data for model training.

[0166] Similar to the 2D profile prediction, the 3D prediction process still uses frequency gradient, 90-degree phase shift attribute, gamma curve, etc. as input features. Furthermore, the 90-degree phase shift attribute is obtained by calculating the seismic attributes of the 3D seismic data, and the gradient is obtained along the frequency dimension through the optimal order S transform to calculate the frequency gradients of the near angle, medium angle and far angle respectively. In addition, considering the existence of faults inside the reservoir, the spatial prior information of the fault position is introduced, and the fault information is used as part of the spatial constraint to assist the model in inferring the gas saturation distribution.

[0167] Based on the processed data, the KNN model is used to predict the three-dimensional gas saturation. In three-dimensional space, the KNN model not only considers the relationship between the seismic characteristics at each well point and the gas saturation curve of the target reservoir, but also comprehensively considers the spatial distribution of the well point and the impact of the fault on the fluid distribution. To this end, the KNN model is optimized, combining seismic attributes, lithology information and fault prior information to enhance the model's prediction ability.

[0168] In the modeling process, based on the analysis of the two-dimensional profile prediction results, the hyperparameter k value was set to 6, and 10 wells (A1 to A10) were selected for three-dimensional gas saturation prediction. The input features of the model still include 90-degree phase shift attributes, near-angle frequency gradients, mid-angle frequency gradients, far-angle frequency gradients, and gamma inversion results. The neighboring sample selection mechanism of the KNN model in three-dimensional space has been improved, using a more complex spatial proximity judgment method to ensure that the model can accurately capture gas saturation changes in three-dimensional space.

[0169] The target layer slice of the three-dimensional gas saturation prediction result is as follows Fig.11As shown in the figure, the model captures the changes in gas saturation inside the reservoir well. Near the fault, the prediction results of the model show obvious changes in gas saturation, which is consistent with the redistribution of underground fluids by the fault. In addition, the reservoir interpretation results of the remaining 9 wells (A11 to A19) were selected as tests. The gas saturation prediction results are consistent with the logging reservoir interpretation conclusions. This shows that the model can accurately predict the location of gas-bearing reservoirs and the strength of gas saturation in three-dimensional space. Overall, the three-dimensional gas saturation prediction results show that within the work area, gas-bearing reservoirs are mainly distributed on the south side of the fault, and the gas saturation is higher near the fault. This result is consistent with the two-dimensional profile prediction results, which further verifies the applicability and stability of the model in three-dimensional space.

[0170] Fig.12 The schematic diagram of the structure of the gas saturation prediction device based on multi-source information fusion provided in this application is as follows: Fig.12 As shown, the gas saturation prediction device 30 based on multi-source information fusion provided in this embodiment includes:

[0171] An acquisition module 301 is used to acquire multi-source geological data of the reservoir to be predicted, wherein the multi-source geological data includes pre-stack seismic data, seismic attribute data, gamma-ray inversion results, seismic lithology results, fault information, and well logging gas saturation data;

[0172] The first analysis module 302 is used to perform time-frequency analysis on pre-stack seismic data in multi-source geological data to determine the frequency division properties of the pre-stack seismic data;

[0173] A determination module 303 is used to determine the frequency gradient corresponding to the pre-stack seismic data according to the frequency division attribute;

[0174] A generating module 304, for generating a regional mask and a lithology mask according to multi-source geological data;

[0175] The second analysis module 305 is used to determine prediction related features based on correlation analysis between well logging gas saturation data and frequency gradient in multi-source geological data and other data in multi-source geological data;

[0176] The prediction module 306 is used to obtain the gas saturation prediction result of the reservoir to be predicted based on the regional mask, the lithology mask and the initial prediction result. The initial prediction result is the result obtained by the gas saturation prediction model based on the prediction-related features. The gas saturation prediction model is a machine learning model constructed based on the K nearest neighbor algorithm.

[0177] In a possible implementation, the first analysis module 302 may also be used to:

[0178] Determine angle data of pre-stack seismic data, the angle data including near angle data, middle angle data and far angle data;

[0179] For each angle data in the pre-stack seismic data, time-frequency analysis is performed on the angle data to determine the target frequency of the angle data and the frequency division attribute corresponding to the target frequency, where the target frequency includes low frequency, medium frequency and high frequency;

[0180] Accordingly, according to the frequency division attribute, the frequency gradient corresponding to the pre-stack seismic data is determined, including:

[0181] For each angle data in the pre-stack seismic data, the frequency gradient of the angle data is determined according to each frequency division attribute corresponding to the angle data.

[0182] In a possible implementation, the generating module 304 may also be used to:

[0183] Generate a regional mask of the reservoir to be predicted based on the fault information in the multi-source geological data, where the regional mask represents the positional relationship between the reservoir to be predicted and the fault;

[0184] Determine the sandstone and mudstone positions of the reservoir to be predicted based on the seismic lithology results in multi-source geological data;

[0185] According to the sandstone position and the mudstone position, a lithology mask of the reservoir to be predicted is generated, and the lithology mask includes a sandstone mask and a mudstone mask.

[0186] In a possible implementation, the second analysis module 305 may also be used to:

[0187] Extract features of well logging gas saturation data, other data and frequency gradient in multi-source geological data to obtain features of well logging gas saturation data, other data features and frequency gradient features;

[0188] Performing correlation analysis on the well logging gas saturation data characteristics and the frequency gradient characteristics to generate a first analysis result, the first analysis result characterizing the influence of the frequency gradient characteristics on the gas saturation;

[0189] Performing correlation analysis on the well logging gas saturation data features and other data features to generate a second analysis result, the second analysis result characterizing the influence of other data features on the gas saturation;

[0190] According to the first analysis result and the second analysis result, prediction-related features are obtained.

[0191] In a possible implementation, the second analysis module 305 may also be used to:

[0192] According to the first analysis result and the second analysis result, other data features and frequency gradient features are sorted to obtain a feature sorting result;

[0193] According to the preset prediction-related feature requirements and feature sorting results, prediction-related features are obtained from other data features and frequency gradient features.

[0194] In a possible implementation, the prediction module 306 may also be used to:

[0195] According to the fault information of the reservoir to be predicted, a first target layer and a second target layer are determined, wherein the first target layer is a reservoir located above the fault in the reservoir to be predicted, and the second target layer is a reservoir located below the fault in the reservoir to be predicted;

[0196] Constructing a first gas saturation prediction model and a second gas saturation prediction model for the first target layer; and a third gas saturation prediction model and a fourth gas saturation prediction model for the second target layer, wherein the first gas saturation prediction model and the third gas saturation prediction model are used to predict the gas saturation of the reservoir with sandstone attributes in the reservoir to be predicted, and the second gas saturation prediction model and the fourth gas saturation prediction model are used to predict the gas saturation of the reservoir with mudstone attributes in the reservoir to be predicted;

[0197] According to the prediction-related characteristics of each target layer in the reservoir to be predicted, the initial prediction results corresponding to each gas saturation prediction model in each target layer are determined.

[0198] In a possible implementation, the prediction module 306 may also be used to:

[0199] The gas saturation prediction results meet the following requirements:

[0200] Sg'=Mf(Ms*KNN1(X)+Md*KNN2(X))+(1-Mf)(Ms*KNN3(X)+Md*KNN4(X));

[0201] Among them, Sg' is the gas saturation prediction result; Mf is the regional mask; Ms is the sandstone mask; Md is the mudstone mask; KNN1 is the first gas saturation prediction model; KNN2 is the second gas saturation prediction model; KNN3 is the third gas saturation prediction model; KNN4 is the fourth gas saturation prediction model; X is the prediction related feature.

[0202] The gas saturation prediction device based on multi-source information fusion provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be described in detail here.

[0203] Fig.13 This is a schematic diagram of the structure of the electronic device provided in this application. Fig.13 As shown, the electronic device 40 provided in this embodiment includes: at least one processor 401 and a memory 402. Optionally, the device 40 also includes a communication component 403. The processor 401, the memory 402 and the communication component 403 are connected via a bus 404.

[0204] In a specific implementation process, at least one processor 401 executes the computer-executable instructions stored in the memory 402, so that at least one processor 401 executes the above method.

[0205] The specific implementation process of the processor 401 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0206] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0207] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk memory.

[0208] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0209] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0210] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0211] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0212] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0213] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0214] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0215] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0216] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0217] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0218] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A gas saturation prediction method based on multi-source information fusion, characterized in that: include: Acquire multi-source geological data of the reservoir to be predicted, wherein the multi-source geological data includes pre-stack seismic data, seismic attribute data, gamma-ray inversion results, seismic lithology results, fault information, and well logging gas saturation data; Performing time-frequency analysis on the pre-stack seismic data in the multi-source geological data to determine the frequency division properties of the pre-stack seismic data; Determining a frequency gradient corresponding to the pre-stack seismic data according to the frequency division attribute; generating a regional mask and a lithology mask according to the multi-source geological data; Determine prediction-related features based on correlation analysis between well logging gas saturation data in the multi-source geological data and the frequency gradient and other data in the multi-source geological data; According to the regional mask, the lithology mask and the initial prediction result, the gas saturation prediction result of the reservoir to be predicted is obtained. The initial prediction result is the result obtained by the gas saturation prediction model based on the prediction-related features. The gas saturation prediction model is a machine learning model constructed based on the K nearest neighbor algorithm.

2. The method according to claim 1, characterized in that The performing time-frequency analysis on the pre-stack seismic data in the multi-source geological data to determine the frequency division properties of the pre-stack seismic data includes: Determining angle data of the pre-stack seismic data, the angle data comprising near angle data, medium angle data and far angle data; For each angle data in the pre-stack seismic data, time-frequency analysis is performed on the angle data to determine a target frequency of the angle data and a frequency division attribute corresponding to the target frequency, wherein the target frequency includes a low frequency, a medium frequency and a high frequency; Accordingly, determining the frequency gradient corresponding to the pre-stack seismic data according to the frequency division attribute includes: For each angle data in the pre-stack seismic data, the frequency gradient of the angle data is determined according to each frequency division attribute corresponding to the angle data.

3. The method according to claim 1, characterized in that The step of generating a regional mask and a lithology mask according to the multi-source geological data comprises: Generate a regional mask of the reservoir to be predicted according to the fault information in the multi-source geological data, wherein the regional mask represents the positional relationship between the reservoir to be predicted and the fault; Determining the sandstone position and mudstone position of the reservoir to be predicted according to the seismic lithology results in the multi-source geological data; A lithology mask of the reservoir to be predicted is generated according to the sandstone position and the mudstone position, and the lithology mask includes a sandstone mask and a mudstone mask.

4. The method according to claim 1, characterized in that: The step of determining prediction-related features based on correlation analysis between the well logging gas saturation data in the multi-source geological data and the frequency gradient and other data in the multi-source geological data includes: Extracting features of the well logging gas saturation data in the multi-source geological data, other data in the multi-source geological data, and the frequency gradient to obtain features of the well logging gas saturation data, features of other data, and features of the frequency gradient; Performing a correlation analysis on the well logging gas saturation data feature and the frequency gradient feature to generate a first analysis result, wherein the first analysis result represents the influence of the frequency gradient feature on the gas saturation; Performing a correlation analysis on the well logging gas saturation data feature and the other data features to generate a second analysis result, wherein the second analysis result represents the influence of the other data features on the gas saturation; According to the first analysis result and the second analysis result, prediction-related features are obtained.

5. The method according to claim 4, characterized in that The obtaining prediction-related features according to the first analysis result and the second analysis result includes: According to the first analysis result and the second analysis result, the other data features and the frequency gradient features are sorted to obtain a feature sorting result; According to preset prediction-related feature requirements and feature sorting results, prediction-related features are obtained from the other data features and the frequency gradient features.

6. The method according to claim 3, characterized in that Before obtaining the gas saturation prediction result of the reservoir to be predicted according to the regional mask, the lithology mask and the initial prediction result, the method further includes: Determine a first target layer and a second target layer according to the fault information of the reservoir to be predicted, wherein the first target layer is a reservoir located above the fault in the reservoir to be predicted, and the second target layer is a reservoir located below the fault in the reservoir to be predicted; Constructing a first gas saturation prediction model and a second gas saturation prediction model for the first target layer; and a third gas saturation prediction model and a fourth gas saturation prediction model for the second target layer, wherein the first gas saturation prediction model and the third gas saturation prediction model are used to predict the gas saturation of the reservoir with sandstone attributes in the reservoir to be predicted, and the second gas saturation prediction model and the fourth gas saturation prediction model are used to predict the gas saturation of the reservoir with mudstone attributes in the reservoir to be predicted; According to the prediction-related characteristics of each target layer in the reservoir to be predicted, the initial prediction results corresponding to each gas saturation prediction model in each target layer are determined.

7. The method according to claim 6, characterized in that The gas saturation prediction result satisfies: Sg'=Mf(Ms*KNN1(X)+Md*KNN2(X))+(1-Mf)(Ms*KNN3(X)+Md*KNN4(X)); Among them, Sg' is the gas saturation prediction result; Mf is the regional mask; Ms is the sandstone mask; Md is the mudstone mask; KNN1 is the first gas saturation prediction model; KNN2 is the second gas saturation prediction model; KNN3 is the third gas saturation prediction model; KNN4 is the fourth gas saturation prediction model; X is the prediction-related feature.

8. A gas saturation prediction device based on multi-source information fusion, characterized in that: include: An acquisition module is used to acquire multi-source geological data of the reservoir to be predicted, wherein the multi-source geological data includes pre-stack seismic data, seismic attribute data, gamma inversion results, seismic lithology results, fault information, and well logging gas saturation data; A first analysis module is used to perform time-frequency analysis on the pre-stack seismic data in the multi-source geological data to determine the frequency division properties of the pre-stack seismic data; A determination module, used for determining a frequency gradient corresponding to the pre-stack seismic data according to the frequency division attribute; A generation module, used for generating a regional mask and a lithology mask according to the multi-source geological data; A second analysis module is used to determine prediction related features based on correlation analysis between the well logging gas saturation data in the multi-source geological data and the frequency gradient and other data in the multi-source geological data; A prediction module is used to obtain a gas saturation prediction result of the reservoir to be predicted based on the regional mask, the lithology mask and the initial prediction result, wherein the initial prediction result is a result obtained by a gas saturation prediction model based on the prediction-related features, and the gas saturation prediction model is a machine learning model constructed based on a K nearest neighbor algorithm.

9. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed.

Images