A seismic phase driven prestack seismic fluid detection method
Through the seismic phase-driven method, machine learning and data fusion technology are used to generate high-resolution initial models, solving the problem that existing technology is difficult to characterize complex spatial structures and improving the accuracy and stability of seismic inversion.
Patent Information
- Application Number
- CN202111627412.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-28
- Publication Date
- 2025-04-29
- Estimated Expiration
- 2041-12-28
AI Technical Summary
The prior art is difficult to effectively characterize complex spatial structures when establishing initial models, resulting in low lateral resolution of earthquake inversion results and the existence of the ‘bull eyes’ phenomenon, which affects the accurate identification of underground geological bodies.
The seismic phase-driven method is used to classify seismic phases through machine learning, combine hierarchical data and logging data, and generate a fusion initial model using a linear weighted fusion algorithm to improve the lateral and longitudinal resolution of the initial model.
The lateral resolution of earthquake inversion results is improved, the recognition ability of special underground geological bodies is enhanced, the stability and reliability of inversion results are improved, and it is suitable for fluid identification and oil and gas exploration.
Smart Images
Figure CN114265114B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of oil and gas exploration and development, and in particular to the technical field of seismic inversion. Background Art
[0002] A key step in seismic inversion is to establish an initial model, in which low-frequency information represents geological information such as the sedimentary background. Low-frequency information plays an important role in improving the stability and reliability of seismic inversion. Conventional logging constrained inversion uses geostatistical modeling methods to describe the spatial structure of sedimentary strata through variograms. However, the variogram analysis based on two points in space can only simply describe the correlation between the two points, and it is difficult to characterize complex spatial structures, which is manifested in the inconsistency between the stratigraphic occurrence of the model in the vertical direction and the occurrence of the phase axis on the seismic profile, and the interpolation result has a "bull's eye" phenomenon around the well in the horizontal direction. This greatly affects the ability of the interpretation results to describe the shape and structure of the sedimentary body.
[0003] In the process of seismic inversion, the initial model is the most important part. The richer the low-frequency information considered, the more reliable the inversion result. The existing technology usually interpolates and extrapolates the processed logging data to obtain a low-frequency model under the constraints of the stratigraphic position. This method is only applicable to relatively flat strata, it is difficult to characterize complex spatial structures, and the lateral resolution is not high, and it also has a trend effect. Therefore, the effect of identifying complex underground geological bodies is average. Summary of the invention
[0004] The main technical problem to be solved by the present invention is to use a data body that is more in line with the actual geological structure to guide the establishment of the initial model, so as to effectively improve the lateral resolution of the initial model and enhance the ability to identify special underground geological bodies.
[0005] The main technical problem to be solved by the present invention is to solve the "bull's eye effect" of conventional methods, also known as the trend effect, to improve the lateral resolution of seismic interpretation results, and to improve the stability and reliability of interpretation results based on seismic data in a more consistent with geological background.
[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a seismic phase driven pre-stack seismic fluid detection method, comprising the steps of:
[0007] S1, data preparation: prepare seismic data, stratigraphic data and well logging data;
[0008] S2, seismic phase classification based on machine learning: seismic data are identified and classified, and the classified seismic data are used as training samples for machine learning, and a three-dimensional seismic phase classification data volume is predicted through the machine learning algorithm;
[0009] S3, establishing a fusion initial model: under the constraint of the layer data, the well logging data is interpolated to obtain an initial model data volume, and the layer velocity data volume, the seismic phase classification data volume and the initial model data volume are fused by a linear weighted fusion algorithm to generate a fusion initial model;
[0010] S4, elastic impedance inversion and fluid factor extraction: Under the Bayesian framework, the likelihood function is established by superimposing partial angle gathers, and the angle seismic wavelet is extracted. In addition, the Cauchy constraint is added to ensure the stability of the inversion. The fused initial model is used as the initial value, and the elastic impedance inversion is carried out using the inversion objective function. Finally, the elastic impedance inversion results of different angles are obtained, and then combined with the rock modulus parameters, the fluid factor is further calculated.
[0011] Preferably, step S1 further includes the step of screening the seismic data:
[0012] Choose to calculate the correlation coefficient of each attribute data of the seismic data under the same interlayer slice, take the absolute value of the correlation coefficient of each attribute data, and then sum them up, select the three attributes with the smallest sum of correlation coefficients as the screening conditions, and screen the seismic data corresponding to the three attributes.
[0013] Preferably, step S1 further includes the step of screening the seismic data: selecting seismic data with three attributes: arc length, coherence and instantaneous main frequency.
[0014] Preferably, in step S2, the machine learning and prediction process is completed through the machine learning module of MATLAB2021b.
[0015] Preferably, in step S2, the machine learning algorithm adopts a kernel naive Bayes classification algorithm.
[0016] Preferably, in step S3, the frequency bands of the layer velocity data volume, the seismic phase classification data volume and the initial model data volume are 0-8 Hz, 0-23 Hz and 0-55 Hz respectively, and the fused initial model is:
[0017] Model fusion =a*Model 1 +b*Model 2 +c*Model 3
[0018] Among them, a, b, c are weight coefficients, Model 1 、Model 2 、Model 3 They are layer velocity data volume, seismic phase classification data volume and initial model data volume, Model fusion The initial fusion model.
[0019] Compared with the prior art, the present invention has the following beneficial effects:
[0020] The present invention integrates the seismic inversion technology of the seismic phase data body and the well logging data interpolation model data body, and can perform pre-stack seismic inversion and oil and gas identification.
[0021] The present invention incorporates a new seismic phase classification data body when establishing the initial model, and couples the seismic phase classification data body and the logging data interpolation data body through attribute fusion to obtain an initial model with relatively high vertical and horizontal resolutions, and the model can well reflect special geological bodies such as river channels.
[0022] The seismic inversion performed by the present invention can make the inversion result more consistent with the actual situation, enhance the accuracy and stability of the interpretation of special geological bodies such as river channels, can be used for fluid identification, and is conducive to improving the accuracy of oil and gas exploration.
[0023] The seismic phase driven seismic inversion constructed by the present invention takes into account more low-frequency information and enhances the recognition capability of special geological bodies. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 A flow chart of a seismic phase driven pre-stack seismic fluid detection method provided by the present invention.
[0025] Figure 2 Confusion matrix plot of machine learning classification results.
[0026] Figure 3(a) is a schematic diagram of the phase classification profile results of machine learning.
[0027] Figure 3(b) is a schematic cross-sectional view of the initial model established using the traditional method.
[0028] Figure 3(c) is a schematic diagram of the initial model cross section after data fusion algorithm processing.
[0029] Figure 4(a) is a schematic diagram of the phase classification plane of machine learning.
[0030] Figure 4(b) is a schematic diagram of the initial model plane established by the traditional method.
[0031] Figure 4(c) is a schematic diagram of the initial model plane after data fusion algorithm processing.
[0032] Figure 5(a) is a schematic cross-sectional view of the initial model of small-angle elastic impedance.
[0033] Figure 5(b) is a schematic cross-sectional view of the initial model of mid-angle elastic impedance.
[0034] Figure 5(c) is a schematic cross-sectional view of the initial model of large-angle elastic impedance.
[0035] Figure 6(a) is a schematic cross-sectional view of the small-angle elastic impedance inversion results.
[0036] Figure 6(b) is a cross-sectional view of the mid-angle elastic impedance inversion results.
[0037] Figure 6(c) is a schematic cross-sectional view of the large-angle elastic impedance inversion results.
[0038] Figure 7(a) is a schematic diagram of the three-dimensional effect of river channel depiction of GT32-GT34.
[0039] Figure 7(b) is a schematic diagram of the three-dimensional effect of river channel depiction of GT31-GT32.
[0040] Figure 8 Schematic diagram of fluid factor inversion results.
[0041] FIG9( a ) is a schematic diagram of the interval velocity data volume and its frequency band range.
[0042] Figure 9(b) is a schematic diagram of the seismic phase classification data volume and its frequency band range.
[0043] Figure 9(c) is a schematic diagram of the well interpolation initial model data volume and its frequency band range.
[0044] Fig.10 Flowchart of weighted fusion method. DETAILED DESCRIPTION
[0045] In order to facilitate a further understanding of the method of the present invention and the effects achieved, the present invention is described in detail below with reference to the accompanying drawings. However, the accompanying drawings are only provided for reference and illustration and are not intended to limit the present invention.
[0046] The present invention starts from seismic data and obtains a seismic phase classification data volume through a machine learning method, and then obtains an initial model through layer constraints and logging data interpolation. Through a data fusion algorithm, the normalized seismic phase classification data volume, layer velocity data volume and logging data interpolation data volume are fused to generate an initial model data volume with high lateral and vertical resolutions. Then, a seismic interpretation result can be obtained through an inversion algorithm, thereby achieving the purpose of fluid identification.
[0047] The method of the present invention is based on the initial model of traditional logging data interpolation. The layer velocity data body, the seismic phase classification data body and the logging data interpolation data body are fused through a data fusion algorithm to generate an initial model data body with high lateral and vertical resolutions. The seismic interpretation result is obtained through an inversion algorithm, thereby achieving the purpose of reservoir fluid identification.
[0048] like Figure 1 As shown, it is a flow chart of a seismic phase driven pre-stack seismic fluid detection method provided by the present invention, and the specific steps of the method are as follows.
[0049] S1, Data Preparation
[0050] The method of the present invention requires three types of data, namely seismic data (also called pre-stack seismic data), stratum data and well logging data (also called well data). Pre-stack seismic data needs to be processed by dynamic correction, static correction, denoising, etc. to maintain a high signal-to-noise ratio and fidelity. After interpretation and interpolation, the stratum data needs to be smooth and accurate. Well logging data includes data such as P-wave velocity, S-wave velocity, density, and acoustic time difference.
[0051] Seismic data is currently the most abundant and accurate data body containing underground information. Therefore, various attribute bodies can be calculated to reflect the underground sedimentary characteristics. The attribute types calculated based on seismic data are rich and varied, and need to be optimized in combination with the characteristics of the work area. The method of the present invention selects three attributes (arc length, coherence, and instantaneous main frequency) from many attributes as the data basis for machine learning seismic phase classification.
[0052] In some embodiments, the attributes are preferably based on two principles: first, they must conform to the paleo-geomorphic characteristics. The attribute analysis data is required to clearly depict the shape of the river channel and to display the internal characteristics of the river channel. Second, the correlation between the attributes is low. If the correlation between different attributes is high, the geological background information of the work area cannot be objectively reflected. Different attributes have different characteristics. On the basis of conforming to the paleo-geomorphic characteristics, the combination of attributes with lower correlation can more completely represent the geological background.
[0053] In some embodiments, the correlation coefficient of the attribute data under the same inter-layer slice is calculated, and the absolute value of the correlation coefficient is taken and then summed up, and the three attributes with the smallest sum of the correlation coefficients (arc length, coherence, and instantaneous main frequency are selected in the present invention) are selected as the data source for machine learning.
[0054] S2, Seismic facies classification based on machine learning
[0055] Establish seismic phase category labels: select a part of the filtered seismic data, manually mark them, and classify them;
[0056] Machine learning prediction: The classified seismic data are used as training samples, and the three-dimensional seismic phase classification data body can be predicted through the machine learning algorithm. The machine learning and prediction process in the method of the present invention is completed by the machine learning module of MATLAB2021b. The method of the present invention selects the training model generated by the kernel naive Bayes classification algorithm and applies it to the three-dimensional seismic phase classification. In the prediction process, a cross-validation ratio of 50% is used to avoid overfitting. Figure 2The confusion matrix of the kernel naive Bayes classification algorithm is shown. It can be seen from the figure that the classification accuracy of the three types of seismic phases is above 98%.
[0057] S3, build the initial fusion model
[0058] Seismic records can reflect the lithology, contact relationship and other characteristics of underground media. Different sedimentary bodies have different amplitudes, frequencies and continuity on seismic profiles. According to the different reflection characteristics, the boundaries between different seismic phases are represented by the seismic phase classification data obtained by machine learning. The sedimentary boundary can effectively characterize special underground geological bodies and enhance the lateral resolution of seismic inversion modeling.
[0059] The traditional modeling method is to interpolate the well logging data under the constraint of the stratigraphic data to obtain an initial model data body (interpolation of well data). At the same time, the variogram is used to simply describe the correlation between two underground points to characterize the spatial structure of the sedimentary strata. Based on this feature, it is difficult for this method to accurately characterize the complex structural characteristics of the reservoir, and it will produce a "bull's eye" effect (as shown in Figure 4). Therefore, the present invention provides a method for establishing a seismic phase-driven fusion initial model. The information fusion algorithm derives more useful information by combining information at multiple levels and in multiple aspects. This not only utilizes the advantages of the coordinated operation of multiple data bodies, but also integrates the advantages of other information source data. The data bodies selected by the method of the present invention are layer velocity data body, seismic phase classification data body, and initial model data body. The frequency bands are 0-8Hz, 0-23Hz, and 0-55Hz, respectively. Then, a linear weighted fusion algorithm is used to fuse the three data bodies of different frequency bands to generate a fused initial model.
[0060] The fusion initial model can be expressed as: Model fusion =a*Model 1 +b*Model 2 +c*Model 3 .
[0061] Among them, a, b, c are weight coefficients, Model 1 、Model 2 、Model 3 They are layer velocity data volume, seismic phase classification data volume and initial model data volume, Model fusion The initial fusion model.
[0062] The cross-section diagrams and frequency band ranges of different data volumes are shown in Figure 9. The specific fusion process is as follows Fig.10The weighted fusion method flow chart is shown in the figure. According to the results of the fusion of each two models, the weight coefficients a, b, c of the output results are continuously adjusted until the model fusion results are consistent with the sedimentary background and have good lateral and vertical resolutions.
[0063] S4, Elastic Impedance Inversion and Fluid Factor Extraction
[0064] In the Bayesian framework, the likelihood function is established by stacking partial angle gathers, and the angle seismic wavelet is extracted. The Cauchy constraint is added to ensure the stability of the inversion. The fusion initial model is used as the initial value, and the elastic impedance inversion is carried out using the inversion objective function. Finally, the elastic impedance inversion results of each angle are obtained. Combined with seismic rock physics analysis, rock modulus parameters such as P-wave velocity and S-wave velocity and density can be obtained, and then the fluid factor can be further calculated. The interpretation of reservoir oil and gas content is realized, and the quality control of the inversion results is achieved by comparing the logging interpretation results with the fluid factor data body. Finally, the fluid factor profile with the highest inversion quality is selected as the basis for oil and gas identification, and pre-stack seismic fluid detection driven by seismic phases is realized.
[0065] Figure 3(a), Figure 3(b), and Figure 3(c) are schematic diagrams of the machine learning phase classification profile, the initial model profile established by the traditional method, and the profile after the data fusion algorithm is processed. As can be seen from the figure, Figure 3(c) overcomes the problem of low lateral resolution of the event axis on the profile in Figure 3(b). Combined with the plane diagram shown in Figure 4, it can be seen that the initial model diagram after the fusion algorithm has basically overcome the "bull's eye" effect produced by the traditional modeling method, and can effectively highlight special geological bodies such as rivers. Figure 4(a), Figure 4(b), and Figure 4(c) are schematic diagrams of the machine learning phase classification plane, the initial model plane established by the traditional method, and the initial model plane after the data fusion algorithm is processed.
[0066] Figures 5 and 6 are respectively the initial model and inversion results of the elastic impedance inversion process at three angles obtained by the method of the present invention. From the profile of the initial model, it can be seen that the fusion initial model driven by the seismic phase has a good lateral resolution, and there is no excessive continuity on the profile of the initial model established by the traditional method. From the profile of the inversion result, the elastic impedance inversion result has a high degree of consistency with the logging data and has a good resolution, which meets the requirements for further seismic interpretation.
[0067] In order to determine whether the elastic impedance obtained by inversion conforms to the geological background, the inversion results of the method of the present invention are displayed in three dimensions. As can be seen from FIG7 , the elastic impedance inversion results can clearly depict the river channel in stages and show the change process of the river channel in different geological periods.
[0068] After obtaining the elastic impedance inversion results at different angles, the method of the present invention further carried out research on fluid prediction methods. Figure 8 This is a fluid factor profile calculated based on elastic impedance at three angles. Compared with the well data, it can be seen that the inversion result is highly consistent with the well data. The predicted fluid factor is consistent with the thickness of the oil-bearing section on the well, and the resolution is high. It has a good prospect for predicting the oil and gas content of the reservoir.
[0069] The above embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Any equivalent substitution or change made by a person skilled in the art based on the present invention is within the protection scope of the present invention.
Claims
1. A seismic phase driven pre-stack seismic fluid detection method, characterized in that: Includes steps: S1, data preparation: prepare seismic data, stratigraphic data and well logging data; S2, seismic phase classification based on machine learning: the seismic data is marked and classified, and the classified seismic data is used as training samples for machine learning, and a three-dimensional seismic phase classification data volume is predicted by a machine learning algorithm; S3, establishing a fusion initial model: under the constraint of the layer data, interpolating the well logging data to obtain an initial model data volume, and using a linear weighted fusion algorithm to fuse the layer velocity data volume, the seismic phase classification data volume and the initial model data volume to generate a fusion initial model; S4, elastic impedance inversion and fluid factor extraction: Under the Bayesian framework, the likelihood function is established by superimposing partial angle gathers, and the angle seismic wavelet is extracted. In addition, the Cauchy constraint is added to ensure the stability of the inversion. The fused initial model is used as the initial value, and the elastic impedance inversion is carried out using the inversion objective function. Finally, the elastic impedance inversion results of different angles are obtained, and then combined with the rock modulus parameters, the fluid factor is further calculated.
2. A seismic phase driven pre-stack seismic fluid detection method as claimed in claim 1, characterized in that: The step S1 also includes the step of screening the seismic data: Select and calculate the correlation coefficients of the various attribute data of the seismic data under the same interlayer slice, take the absolute value of the correlation coefficient of each attribute data, and then sum them up, select the three attributes with the smallest sum of correlation coefficients as screening conditions, and screen the seismic data corresponding to the three attributes.
3. A seismic phase driven pre-stack seismic fluid detection method as claimed in claim 1, characterized in that: The step S1 also includes the step of screening the seismic data: selecting seismic data with three attributes: arc length, coherence and instantaneous main frequency.
4. A seismic phase driven pre-stack seismic fluid detection method as claimed in claim 1, 2 or 3, characterized in that: In step S2, the machine learning and prediction process is completed through the machine learning module of MATLAB2021b.
5. A seismic phase driven pre-stack seismic fluid detection method as claimed in claim 1, 2 or 3, characterized in that: In step S2, the machine learning algorithm adopts a kernel naive Bayes classification algorithm.
6. A seismic phase driven pre-stack seismic fluid detection method as claimed in claim 1, 2 or 3, characterized in that: In step S3, the frequency bands of the layer velocity data volume, the seismic phase classification data volume and the initial model data volume are 0-8 Hz, 0-23 Hz and 0-55 Hz respectively, and the fused initial model is: Model fusion =a*Model1+b*Model2+c*Model3 Among them, a, b, c are weight coefficients, Model1, Model2, and Model3 are the layer velocity data volume, seismic phase classification data volume, and initial model data volume, respectively. fusion is the fusion initial model.