Geological-seismic depth fusion intelligent reservoir prediction method

Abstract

The present application relates to a kind of geological-seismic depth fusion intelligent reservoir prediction method, method as follows: in the area with earthquake, construct the reservoir prediction model based on geological data and the reservoir prediction model based on seismic data, first the coordinate matching of two prediction results is carried out, fuses expert experience formula reservoir mode chart, then through A 3 Res:All2Res integrated learning mode, finally obtain the effective reservoir prediction result of the area with earthquake fusion.The present application first carries out the research of the sedimentary characteristics of typical block, reservoir sand body distribution characteristics, summarizes the sedimentary geological model on this basis, forms reliable geological knowledge, obtains the reservoir prediction model based on geological data;On the other hand, carry out seismic data processing and interpretation and available information mining, obtain the reservoir prediction model based on seismic data: after coordinate matching, then fusing expert experience formula reservoir mode chart converted from expert experience, form reliable intelligent reservoir prediction method by domain knowledge+deep learning.

Description

Technical Field

[0001] This invention relates to the field of oil and gas field development technology, and more specifically to an intelligent reservoir prediction method that deeply integrates geology and seismic data. Background Technology

[0002] The reservoirs in the Yanchang block have complex stacked characteristics in the vertical direction, which makes it difficult to identify the dominant reservoirs. The main reasons for the difficulty and inaccuracy in identifying effective reservoirs in the Yanchang exploration area are as follows: (1) Strong heterogeneity of reservoirs in both horizontal and vertical directions; (2) Thin reservoirs are developed, making them difficult to identify accurately; (3) The sedimentary environment of the target layer changes rapidly; (4) The main controlling factors of the reservoirs in the study area are complex; (5) There is a lack of high-quality three-dimensional seismic data; (6) Traditional reservoir research methods are difficult to be effective.

[0003] To address the aforementioned challenges, it is necessary to leverage new intelligent technologies to conduct research on effective reservoir prediction methods and develop new reservoir prediction techniques. Summary of the Invention

[0004] This invention aims to address the aforementioned problems by proposing an intelligent reservoir prediction method that deeply integrates geology and seismic data.

[0005] The technical solution of this invention is as follows: A smart reservoir prediction method that deeply integrates geology and seismic data is presented below.

[0006] (a) Areas prone to earthquakes; In seismic areas, reservoir prediction models based on geological data and seismic data are constructed. First, coordinate matching of the prediction results from both models is performed. Then, reservoir model diagrams based on expert empirical formulas are integrated, and finally, A... 3 The Res:All2Res ensemble learning model ultimately yields effective reservoir prediction results for seismic zones. The specific construction process of the reservoir prediction model based on geological data is as follows: Parameters related to reservoir prediction are selected, lithological interpretation results obtained from well logging curves and effective reservoir interpretation results are fused to obtain a sparse data volume, and the sparse data spatial distribution of the data volume containing well logging information is expanded by Dirichlet mosaicking to obtain a spatial data volume; then, the relationship between the spatial data volume and the effective reservoir geological modeling results interpreted by experts is learned through deep learning to train the reservoir prediction model based on geological data. The specific construction process of the reservoir prediction model based on seismic data is as follows: after quality control of the seismic data, the coordinates of the seismic data coverage area and the well logging data coverage area are matched. After resampling the seismic data, the relationship between the seismic amplitude data and the effective reservoir geological modeling results interpreted by experts is learned through deep learning, and the reservoir prediction model based on seismic data is trained. In the construction of the reservoir prediction model based on geological data, deep learning is employed using U-Net A, a residual network and encoder-decoder architecture. 3 Res: Geo2Res model; In the construction of reservoir prediction models for seismic data, deep learning is employed using U-Net A, a model based on residual networks and an encoder-decoder architecture. 3 Res:Seis2Res model.

[0007] (ii) Seismic-free areas.

[0008] The area where effective reservoir prediction results of seismic area fusion are obtained is taken as the verification area. The effective reservoir prediction results of seismic area fusion in the verification area are obtained, and the distribution characteristics of logging data in the verification area are first matched with the coordinates of the effective reservoir prediction results of seismic area fusion. The distribution characteristics of well logging data expanded by sparse data spatial distribution are combined with the effective reservoir prediction results of seismic fusion in the validation area, and then the effective reservoir prediction results of seismic fusion are obtained through deep learning. Specifically, the U-Net A3Res:Geo2Res model is trained using incremental learning technology to learn the correlation between the distribution characteristics of well logging data in the validation area and the effective reservoir, thereby obtaining the effective reservoir prediction results of seismic fusion.

[0009] Preferably, the parameters related to reservoir prediction are: lithology interpretation, effective reservoir interpretation (Pay), acoustic AC, natural gamma ray (GR), density (DEN), neutron density (CNL), deep lateral resistivity (LLD), shallow lateral resistivity (LLS), wellbore diameter (CAL), porosity (POR), permeability (PERM), and saturation (SG).

[0010] Preferably, the specific process of the Dirichlet mosaicking is as follows: the sparse logging grid point data of each layer is expanded into a 2D screen polygon partition map, such that all points in each polygon are closest to the control unit inside it, and the value in each polygon is determined by the control unit.

[0011] Preferably, the modules involved in the resampling process include horizontal sampling rate matching, vertical sampling rate matching, main direction finding sampling, and communication direction finding sampling.

[0012] The technical advantages of this invention are as follows: This invention first studies the sedimentary characteristics and reservoir sand body distribution characteristics of typical blocks, and on this basis, summarizes sedimentary geological models, forms reliable geological knowledge, and obtains a reservoir prediction model based on geological data. On the other hand, it conducts seismic data processing and interpretation and mining of available information to obtain a reservoir prediction model based on seismic data. After coordinate matching, it integrates the reservoir model diagram of expert experience formula transformed from expert experience, and forms a reliable intelligent reservoir prediction method through domain knowledge + deep learning. Attached Figure Description

[0013] Figure 1 The flowchart of this invention is for use in seismic areas.

[0014] Figure 2 The flowchart of the present invention is shown for use in a seismic-free region. Detailed Implementation

[0015] Example 1 - Seismic area.

[0016] A smart reservoir prediction method that deeply integrates geology and seismic data is as follows: In seismic areas, reservoir prediction models based on geological data and seismic data are constructed. First, coordinate matching of the prediction results from both models is performed. Then, reservoir model diagrams based on expert empirical formulas are integrated, and finally, A... 3 The Res:All2Res ensemble learning model ultimately yields effective reservoir prediction results for seismic zones. The specific construction process of the reservoir prediction model based on geological data is as follows: Parameters related to reservoir prediction are selected, lithological interpretation results obtained from well logging curves and effective reservoir interpretation results are fused to obtain a sparse data volume, and the sparse data spatial distribution of the data volume containing well logging information is expanded by Dirichlet mosaicking to obtain a spatial data volume; then, the relationship between the spatial data volume and the effective reservoir geological modeling results interpreted by experts is learned through deep learning to train the reservoir prediction model based on geological data. The specific construction process of the reservoir prediction model based on seismic data is as follows: after quality control of the seismic data, the coordinates of the seismic data coverage area and the well logging data coverage area are matched. After resampling the seismic data, the relationship between the seismic amplitude data and the effective reservoir geological modeling results interpreted by experts is learned through deep learning, and the reservoir prediction model based on seismic data is trained. In the construction of the reservoir prediction model based on geological data, deep learning is employed using U-Net A, a residual network and encoder-decoder architecture. 3 Res: Geo2Res model; In the construction of reservoir prediction models for seismic data, deep learning is employed using U-Net A, a model based on residual networks and an encoder-decoder architecture.3 Res:Seis2Res model.

[0017] Example 2—Seismic-free zone The area where effective reservoir prediction results of seismic area fusion are obtained is taken as the verification area. The effective reservoir prediction results of seismic area fusion in the verification area are obtained, and the distribution characteristics of logging data in the verification area are first matched with the coordinates of the effective reservoir prediction results of seismic area fusion. The logging data distribution characteristics, which are augmented by expanding the sparse data volume spatial distribution, are combined with the effective reservoir prediction results from the seismic fusion of the validation area. Then, deep learning is used to obtain the effective reservoir prediction results from the seismic fusion of the non-seismic areas. Specifically, U-Net A is trained using incremental learning techniques. 3 The Res:Geo2Res model learns the correlation between the distribution characteristics of logging data in the validation area and the effective reservoir, thereby obtaining the effective reservoir prediction results of seismic-free area fusion.

[0018] Example 3 Based on Example 2, it also includes: The parameters related to reservoir prediction are specifically: lithology interpretation, effective reservoir interpretation (Pay), acoustic AC, natural gamma ray (GR), density (DEN), neutron density (CNL), deep lateral resistivity (LLD), shallow lateral resistivity (LLS), wellbore diameter (CAL), porosity (POR), permeability (PERM), and saturation (SG).

[0019] The specific process of the Dirichlet mosaic is as follows: the sparse logging grid point data of each layer is expanded into a 2D screen polygon partition map, so that all points in each polygon are closest to the control unit inside it, and the value in each polygon is determined by the control unit.

[0020] The modules involved in the resampling process include horizontal sampling rate matching, vertical sampling rate matching, main direction finding sampling, and communication direction finding sampling.

Claims

1. A smart reservoir prediction method that deeply integrates geology and seismic data, characterized in that, This includes reservoir prediction in seismically active areas and reservoir prediction in seismically resistant areas. The specific prediction methods are as follows: Reservoir prediction in earthquake-prone areas: In seismic areas, on the one hand, parameters related to reservoir prediction are selected, and lithological interpretation results and effective reservoir interpretation results obtained from well logging curves are integrated to obtain a sparse data volume, and then a reservoir prediction model based on geological data is constructed. On the other hand, construct reservoir prediction models based on seismic data; First, coordinate matching of the prediction results from both methods is performed, then the reservoir model diagram based on expert empirical formulas is integrated, and finally, A is used... 3 The Res:All2Res ensemble learning model ultimately yields effective reservoir prediction results for seismic zones. Reservoir prediction in seismic-free areas: The area where effective reservoir prediction results of seismic area fusion are obtained is taken as the verification area. The effective reservoir prediction results of seismic area fusion in the verification area are obtained, and the distribution characteristics of logging data in the verification area are first matched with the coordinates of the effective reservoir prediction results of seismic area fusion in the verification area. The distribution characteristics of well logging data, which are expanded by sparse data spatial distribution, are combined with the effective reservoir prediction results of seismic areas in the validation zone, and then the effective reservoir prediction results of seismic-free areas are obtained through deep learning.

2. The intelligent reservoir prediction method based on deep geology-seismic fusion according to claim 1, characterized in that, The specific construction process of the reservoir prediction model based on geological data is as follows: the sparse data volume is expanded by Dirichlet mosaicking to expand the sparse data spatial distribution of the data volume containing well logging information to obtain a spatial data volume; then, the relationship between the spatial data volume and the effective reservoir geological modeling results interpreted by experts is learned through deep learning to train the reservoir prediction model based on geological data.

3. The intelligent reservoir prediction method based on deep geology-seismic fusion according to claim 2, characterized in that, The specific construction process of the reservoir prediction model based on seismic data is as follows: after quality control of the seismic data, coordinate matching is performed between the seismic data coverage area and the well logging data coverage area. After resampling the seismic data, the relationship between the seismic amplitude data and the effective reservoir geological modeling results interpreted by experts is learned through deep learning, and the reservoir prediction model based on seismic data is trained.

4. The intelligent reservoir prediction method based on deep geology-seismic fusion according to claim 3, characterized in that, In the construction of the reservoir prediction model based on geological data, deep learning is employed using U-Net A, a residual network and encoder-decoder architecture. 3 Res: Geo2Res model; In the construction of reservoir prediction models for seismic data, deep learning is employed using U-Net A, a model based on residual networks and an encoder-decoder architecture. 3 Res:Seis2Res model.

5. The intelligent reservoir prediction method based on deep geology-seismic fusion according to claim 4, characterized in that, The specific process of obtaining the effective reservoir prediction results through seismic-free region fusion via deep learning is as follows: U-Net A is trained using incremental learning techniques. 3 The Res:Geo2Res model learns the correlation between the distribution characteristics of logging data in the validation area and the effective reservoir, thereby obtaining the effective reservoir prediction results of seismic-free area fusion.

6. The intelligent reservoir prediction method based on deep geology-seismic fusion according to claim 1, characterized in that, The parameters related to reservoir prediction are specifically: lithology interpretation, effective reservoir interpretation (Pay), acoustic AC, natural gamma ray (GR), density (DEN), neutron density (CNL), deep lateral resistivity (LLD), shallow lateral resistivity (LLS), wellbore diameter (CAL), porosity (POR), permeability (PERM), and saturation (SG).

7. The intelligent reservoir prediction method based on deep geology-seismic fusion according to claim 2, characterized in that, The specific process of the Dirichlet mosaic is as follows: the sparse logging grid point data of each layer is expanded into a 2D screen polygon partition map, so that all points in each polygon are closest to the control unit inside it, and the value in each polygon is determined by the control unit.

8. The intelligent reservoir prediction method based on deep geology-seismic fusion according to claim 3, characterized in that, The modules involved in the resampling process include horizontal sampling rate matching, vertical sampling rate matching, main direction finding sampling, and communication direction finding sampling.

Images