A method and system for constructing an artificial intelligence low-frequency model based on multi-information fusion

Abstract

The present invention relates to a multi-information fusion artificial intelligence low-frequency model construction method and system, which is characterized in that the method comprises: selecting a target curve of the drilled well based on original logging data of the wells drilled in the study area; obtaining seismic layer velocity and near-, middle- and far-path partial stacked seismic data volumes based on pre-stack CMP gathers of the wells drilled in the study area; selecting an optimized seismic attribute combination based on the near-, middle- and far-path partial stacked seismic data volumes, seismic layer velocity and target curve, and obtaining an optimal nonlinear model between the target curve and the optimized seismic attribute combination; applying the optimal nonlinear model to the optimized seismic attribute combination of the entire study area to obtain a target data volume of the study area of ​​the same type as the target curve; and performing low-pass filtering on the target data volume to obtain a target low-frequency model of the study area. The present invention can be widely applied to the technical field of seismic inversion for geophysical reservoir prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic inversion for geophysical reservoir prediction, and in particular to a method and system for constructing a multi-information fusion artificial intelligence low-frequency model. Background Art

[0002] Currently, the most commonly used method for reservoir prediction is seismic inversion. Low-frequency information can control the background trend of the inversion results and is therefore of great significance for reservoir prediction. However, due to the limitations of seismic acquisition systems, direct seismic inversion results lack low-frequency components. To incorporate low-frequency trends into the inversion results, a low-frequency model must be constructed.

[0003] There are three conventional low-frequency model construction methods, namely, low-frequency modeling method based on well interpolation and extrapolation, low-frequency modeling method based on seismic velocity analysis, and low-frequency modeling method based on geological model.

[0004] However, due to the sparsity of drilling, the coarseness of seismic velocities or the subjectivity of geological models, the low-frequency models established using the above methods have the "bull's eye" phenomenon, low resolution and accuracy, and complex implementation steps. Summary of the Invention

[0005] In response to the above problems, the purpose of the present invention is to provide a multi-information fusion artificial intelligence low-frequency model construction method and system, which can effectively avoid the "bull's eye" phenomenon and improve resolution and accuracy.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions: In a first aspect, a method for constructing an artificial intelligence low-frequency model based on multi-information fusion is provided, comprising:

[0007] Based on the original logging data of the wells drilled in the study area, the target curve of the wells that have been drilled is selected;

[0008] Based on the pre-stack CMP gathers of the wells drilled in the study area, seismic layer velocities and stacked seismic data volumes of the near, middle and far tracks are obtained;

[0009] Based on the near, mid and far track partial stacked seismic data volumes, seismic layer velocities and target curves, the optimal seismic attribute combination is selected, and the optimal nonlinear model between the target curve and the optimized seismic attribute combination is obtained;

[0010] Apply the optimal nonlinear model to the optimized seismic attribute combination of the entire study area to obtain the target data volume of the study area with the same type as the target curve;

[0011] The target data volume is low-pass filtered to obtain the target low-frequency model of the study area.

[0012] Furthermore, the method further includes: fitting the obtained target low-frequency model using the result of well logging data intersection analysis to obtain other low-frequency models of the study area.

[0013] Furthermore, selecting the target curve of the drilled wells based on the original logging data of the wells drilled in the study area includes:

[0014] Obtain the original logging data of the wells drilled in the study area and optimize it;

[0015] The P-wave curve and density curve in the optimized original logging data are used for well-seismic calibration, and the target curve of the drilled well whose well-seismic correlation meets the preset requirements is selected.

[0016] Furthermore, the seismic layer velocity and the near, middle and far track partial stacked seismic data volumes are obtained based on the pre-stack CMP gathers of the wells drilled in the study area, including:

[0017] Obtain pre-stack CMP gathers of wells drilled in the study area;

[0018] Depth migration is performed on the pre-stack CMP gathers to obtain seismic layer velocity and CRP gathers;

[0019] The CRP gathers are stacked at different angles to obtain the near, middle and far track partial stacked seismic data volumes.

[0020] Furthermore, the near-, medium- and far-track partial stacked seismic data volumes refer to corresponding seismic data volumes obtained by stacking small-, medium- and large-angle pre-stack CRP gathers, respectively.

[0021] Furthermore, the method of selecting an optimized seismic attribute combination based on the near-, mid- and far-path partial stacking seismic data volumes, seismic layer velocities and target curves, and obtaining an optimal nonlinear model between the target curve and the optimized seismic attribute combination includes:

[0022] Based on the near, mid and far track partial stacking seismic data volumes, seismic layer velocities and target curves, multi-attribute linear regression analysis training is performed to select the optimal seismic attribute combination;

[0023] According to the target curve and the optimized seismic attribute combination, deep feedforward neural network analysis and training are carried out to obtain the optimal nonlinear model between the target curve and the optimized seismic attribute combination.

[0024] Furthermore, the target data volume is subjected to low-pass filtering to obtain a target low-frequency model of the study area, including:

[0025] Set the cutoff frequency value according to the actual frequency of the target data body;

[0026] Only the frequency components of the target data volume that are not greater than the cutoff frequency value are retained to obtain the target low-frequency model of the study area that is consistent with the target curve and target data volume type.

[0027] Secondly, a multi-information fusion artificial intelligence low-frequency model construction system is provided, including:

[0028] The target curve selection module is used to select the target curve of the drilled wells based on the original logging data of the wells drilled in the study area;

[0029] Pre-stack CMP gather processing module is used to obtain seismic layer velocity and near-, mid- and far-track partial stack seismic data volumes based on pre-stack CMP gathers of wells drilled in the study area;

[0030] The optimal nonlinear model determination module is used to select the optimal seismic attribute combination based on the near-, mid- and far-path partial stacking seismic data volumes, seismic layer velocities and target curves, and obtain the optimal nonlinear model between the target curve and the optimized seismic attribute combination;

[0031] The target data volume determination module is used to apply the optimal nonlinear model to the optimized seismic attribute combination of the entire study area to obtain the target data volume of the study area with the same type as the target curve;

[0032] The low-frequency model determination module is used to perform low-pass filtering on the target data volume to obtain the target low-frequency model of the study area.

[0033] According to a third aspect, a processing device is provided, comprising computer program instructions, wherein the computer program instructions, when executed by the processing device, are used to implement the steps corresponding to the above-mentioned method for constructing an artificial intelligence low-frequency model of multi-information fusion.

[0034] In a fourth aspect, a computer-readable storage medium is provided, on which computer program instructions are stored, wherein when the computer program instructions are executed by a processor, they are used to implement the steps corresponding to the above-mentioned multi-information fusion artificial intelligence low-frequency model construction method.

[0035] The present invention has the following advantages due to the adoption of the above technical solution:

[0036] 1. The present invention simultaneously inputs the near-, mid- and far-channel partial stacked seismic data volumes for training, thereby increasing the information of different offset distances / incident angles reflecting different reservoir parameters and improving the low-frequency model's ability to describe actual geological conditions.

[0037] 2. This method incorporates depth-migrated seismic interval velocities into the training process, incorporating low-frequency components from 0 to 5 Hz into the target low-frequency model. Compared to initial models derived using conventional well interpolation and extrapolation, these interval velocities better reflect geological background trends, making them particularly effective in areas with a small number of wells and large depth differences between wells. Furthermore, depth-migrated interval velocities offer greater accuracy than interval velocities derived using time-migrated RMS velocities.

[0038] 3. The present invention connects multi-attribute linear regression analysis and deep feedforward neural network in series. The principle of multi-attribute linear regression analysis is simple and the calculation speed is fast. It can preliminarily fit a good relationship between various seismic attributes and the target curve. Compared with traditional machine learning, deep feedforward neural network can automatically learn from simple features and extract useful information to obtain deeper and more complex features, and obtain a more reasonable nonlinear model. Compared with simply using multi-attribute linear regression analysis or deep feedforward neural network, connecting multi-attribute linear regression analysis and deep feedforward neural network in series can further obtain the optimal nonlinear model through deep learning on the basis of selecting the optimal seismic attribute combination, and then obtain the target data body and target low-frequency model.

[0039] 4. When conducting multi-attribute linear regression analysis training and establishing the relationship between the target curve and multiple seismic attributes, the present invention eliminates attributes without geological meaning and frequency-related attributes, which can effectively improve the resolution, accuracy and planar predictability of the target low-frequency model.

[0040] 5. The present invention can effectively avoid the "bull's eye" phenomenon of conventional low-frequency models and provide a low-frequency model containing richer geological background and sedimentary characteristic information for seismic inversion, which is of great significance for subsequent reservoir fine prediction work.

[0041] In summary, the present invention can be widely applied in the field of seismic inversion technology for geophysical reservoir prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. Throughout the drawings, the same reference numerals are used to denote the same components. In the drawings:

[0043] Figure 1 This is a flow chart of a method provided by one embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the depth migration seismic interval velocity spectrum of the target layer segment in the study area provided by one embodiment of the present invention;

[0045] Figure 3 1 is a schematic diagram of a seismic longitudinal wave spectrum provided by an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram of a seismic longitudinal wave well cross section provided by an embodiment of the present invention;

[0047] Figure 5 It is a schematic plan view of a target layer section in a study area using a low-frequency model of lithologic indicator factors constructed using a well interpolation and extrapolation method provided by an embodiment of the present invention;

[0048] Figure 6 It is a schematic plan view of a target layer section in a study area using a low-frequency model of lithologic indicator factors constructed using the method of the present invention, provided by one embodiment of the present invention;

[0049] Figure 7 It is a schematic plan view of the target layer section of the study area of ​​the absolute inversion result of the lithologic indicator factor obtained by fusing the low-frequency model constructed by the well interpolation and extrapolation method and the direct inversion result provided by one embodiment of the present invention;

[0050] Figure 8 It is a planar schematic diagram of the absolute inversion result of the lithologic indicator factor in the target layer section of the study area obtained by fusing the low-frequency model constructed by the method of the present invention with the direct inversion result provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0051] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0052] It should be understood that the terms used herein are for the purpose of describing specific example embodiments only and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "comprise", "include", "contain" and "have" are inclusive and therefore specify the presence of stated features, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not to be construed as necessarily requiring them to be performed in the specific order described or illustrated, unless the order of execution is clearly indicated. It should also be understood that additional or alternative steps may be used.

[0053] Although the terms first, second, third, etc. can be used in the text to describe multiple elements, components, regions, layers and / or sections, these elements, components, regions, layers and / or sections should not be limited by these terms. These terms can only be used to distinguish an element, component, region, layer or section from another region, layer or section. Unless the context clearly indicates otherwise, terms such as "first", "second" and other numerical terms do not imply order or sequence when used in the text. Therefore, the first element, component, region, layer or section discussed below can be referred to as the second element, component, region, layer or section without departing from the teaching of the example embodiments.

[0054] The multi-information fusion artificial intelligence low-frequency model construction method and system provided by the embodiment of the present invention inputs the superimposed seismic data volume of the near, middle, and long paths and the deep offset seismic layer velocity for training. By using multi-attribute linear regression analysis and deep feedforward neural network in series, the optimal nonlinear model is further obtained through deep learning based on the selection of the optimal seismic attribute combination, thereby obtaining the target data volume and target low-frequency model. The low-frequency model constructed by the present invention has higher resolution and accuracy, stronger planar predictability and description of actual geological conditions, contains richer geological background and sedimentary characteristics information, can effectively avoid the "bull's eye" phenomenon, and is of great significance for subsequent reservoir fine prediction and quantitative analysis.

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment provides a method for constructing an artificial intelligence low-frequency model based on multi-information fusion, comprising the following steps:

[0057] 1) Obtain the original logging data and prestack CMP (Common Middle Point) gathers of the wells drilled in the study area.

[0058] Specifically, the original logging data of the drilled wells include well curves, P-wave (acoustic) curves, P-wave impedance curves, and density curves in the target interval that are consistent with the data types of the low-frequency model to be constructed.

[0059] For example: if it is intended to construct a low-frequency P-wave model of the target layer H1, the original logging data should at least include the P-wave curve within the H1 layer; if it is intended to construct a low-frequency density model of the target layer H2, the original logging data should at least include the density curve and P-wave curve within the H2 layer.

[0060] 2) Optimize the acquired original logging data and use the optimized original logging data for fine well-seismic calibration. Select the target curve of the drilled well whose well-seismic correlation meets the preset requirements, specifically:

[0061] 2.1) Optimize the acquired raw logging data.

[0062] Specifically, optimization processing usually includes environmental correction and standardization, the purpose of which is to correct the measurement deviation of logging curves caused by the logging and drilling environment, and eliminate the systematic differences in measurement results between multiple wells caused by different instruments, different measurement environments, different acquisition times and different wellbore conditions.

[0063] 2.2) The P-wave curve and density curve in the optimized raw logging data are used for well-seismic calibration, and the target curve of the drilled well whose well-seismic correlation meets the preset requirements is selected.

[0064] Specifically, the detailed process of fine well-seismic calibration is as follows:

[0065] ① Obtain the P-wave impedance curve based on the P-wave curve and density curve in the optimized original logging data. If there is no density curve, the density curve can be obtained using the Gardner empirical formula. If the original logging data includes the P-wave impedance curve, skip this step.

[0066] ②Convert the longitudinal wave impedance curve into a reflection coefficient sequence.

[0067] ③ Perform full stacking of the pre-stack CMP gathers to obtain the original seismic records.

[0068] ④ Create an initial zero-phase amplitude spectrum wavelet based on the amplitude and frequency of the original seismic record near the target layer.

[0069] ⑤ Convolve the zero-phase amplitude spectrum wavelet extracted in step ④ and the reflection coefficient sequence obtained in step ② to obtain a synthetic seismic record.

[0070] ⑥ Adjust the synthetic seismic record obtained in step ⑤ by means of overall time shift, local stretching or compression, so as to improve the correlation and wave group characteristic consistency between the synthetic seismic record obtained in step ③ and the original seismic record.

[0071] ⑦ Re-extract the zero-phase amplitude spectrum wavelet based on the adjusted synthetic seismic record and enter step ⑤ until the correlation between the synthetic seismic record and the original seismic record and the morphology, amplitude spectrum and phase spectrum of the extracted zero-phase amplitude spectrum wavelet meet the preset requirements.

[0072] For example, when the correlation between the synthetic seismic record and the original seismic record reaches 70% near the target layer, the preset requirement is met.

[0073] Specifically, the target curve refers to a well curve that is consistent with the data type of the low-frequency model to be constructed. For example, if a longitudinal wave low-frequency model is to be constructed, the target curve is a longitudinal wave curve.

[0074] 3) Perform depth migration on the pre-stack CMP gathers to obtain seismic layer velocities and pre-stack CRP gathers.

[0075] 4) The pre-stack CRP gathers are stacked at different angles to obtain the near, middle and far stacked seismic data volumes.

[0076] Specifically, the near-, medium- and far-channel partial stacked seismic data volumes refer to the corresponding seismic data volumes obtained by stacking small-, medium- and large-angle pre-stack CRP gathers, respectively. The specific ranges of small, medium and large angles can be determined based on actual tests and will not be elaborated here.

[0077] 5) Using the target curve selected in step 2) as the learning target, multi-attribute linear regression analysis training is performed based on the seismic layer velocity obtained in step 3) and the near-, mid-, and far-path partial stacking seismic data volumes obtained in step 4), to establish the relationship between the target curve and multiple seismic attributes, and select the optimal seismic attribute combination that minimizes the prediction error.

[0078] Specifically, seismic attributes include the input data (i.e., the near-, mid-, and far-path stacked seismic data volumes and seismic layer velocities) and their derived attributes. Derived attributes are attributes derived from the input data through mathematical or attribute transformations, such as the inverse of the input seismic layer velocity and the amplitude envelope of the input far-path stacked seismic data volume. Furthermore, derived attributes should not include attributes without geological meaning or frequency-based attributes.

[0079] Specifically, each sample point on the target curve is regarded as a linear combination of several seismic attributes at the same time. At each sample point, the target curve can be expressed by the following linear regression equation:

[0080] L(t)=ω0+ω1A1(t)+ω2A2(t)+…+ω i A i (t)+…+ω n A n (t) (1)

[0081] Where L(t) represents the value of the target curve at time t; A i (t) represents the value of the i-th seismic attribute of the well bypass at time t; ω i represents the weight coefficient of the i-th seismic attribute in the linear regression model, and n represents the total number of seismic attributes.

[0082] When the mean square prediction error in the following formula (2) reaches the minimum value, the optimal weight coefficient of each seismic attribute in the linear regression model can be obtained:

[0083]

[0084] Where, E2 represents the mean squared prediction error; t = 1, 2, ..., N represents the data time series; N represents the total number of data time series; A nt Represents the value of the nth earthquake attribute at time t.

[0085] Since the frequency components of the target curve and the seismic attributes are quite different, the cross-correlation based on a single sample point may not be optimal. To solve this problem, it is assumed that each sample point of the target curve is correlated with a group of adjacent sample points of the seismic attributes. Therefore, the weight coefficient in formula (1) can be replaced by a convolution operator with a certain length, that is, formula (1) can be rewritten as:

[0086] L(t)=w0+w1*A1(t)+w2*A2(t)+…+w i *A i (t)+…+w n *A n (t) (3)

[0087] In the formula, * represents the convolution operation, w i Represents a convolution operator of a specified length.

[0088] Similarly, when the mean square prediction error in the following formula (4) reaches the minimum value, the optimal convolution operator length can be obtained:

[0089]

[0090] It can be seen from formula (4) above that the mean square prediction error of the multi-attribute linear regression analysis is not only related to the length of the convolution operator, but also to the seismic attributes and their number. Therefore, this step requires testing the seismic attributes and the length of the convolution operator to obtain the optimal results of both.

[0091] Specifically, the testing of seismic attributes and convolution operator length includes: fixing other parameters, testing the seismic attribute category, the number of seismic attributes and the convolution operator length respectively, and the results of the seismic attribute category, the number of seismic attributes and the convolution operator length when the prediction error reaches the minimum value are the optimal results.

[0092] Specifically, for example: the optimized seismic attribute combination that minimizes the verification error includes the inverse of the seismic layer velocity, the instantaneous amplitude derivative of the near-track partially stacked seismic data volume, the instantaneous amplitude derivative of the far-track partially stacked seismic data volume, the absolute amplitude integral of the far-track partially stacked seismic data volume, the square of the middle-track partially stacked seismic data volume, the square of the far-track partially stacked seismic data volume, the amplitude envelope of the far-track partially stacked seismic data volume, the orthogonal tracks of the near-track partially stacked seismic data volume, and the derivative of the middle-track partially stacked seismic data volume.

[0093] 6) Using the target curve selected in step 2) as the learning target, and performing deep feedforward neural network analysis training based on the optimized seismic attribute combination selected in step 5), an optimal nonlinear model between the target curve and the optimized seismic attribute combination is obtained.

[0094] Specifically, the structure of a deep feedforward neural network includes an input layer, a hidden layer, and an output layer, and the full connection between each layer. The connection between each layer represents the weight of the feature, and the input and output have the following mapping relationship:

[0095] y=f(x,θ0) (5)

[0096] Where x and y represent input and output respectively; θ0 represents the optimal parameter solution for the mapping between input and output.

[0097] For this step, the output y is the target curve selected in step 2), the input x is the optimized seismic attribute combination selected in step 5), and the optimal parameter solution θ0 mapped between the input and output is the optimal nonlinear model between the target curve and the optimized seismic attribute combination to be obtained through this step (the reason why it is "nonlinear" is because y = f(x, θ0) is a relatively complex nonlinear function relationship).

[0098] Specifically, the optimal parameter solution θ0 can be obtained using the conjugate gradient method (CG) or the steepest descent method (SD). Taking the conjugate gradient method as an example, the steps include: vector initialization, calculation of the residual vector, calculation of the direction vector, calculation of the step size, updating the solution vector, and repeating the above steps until the residual vector is sufficiently small. The solution vector at this point is the desired result. Since the conjugate gradient method and the steepest descent method are both algorithms disclosed in the prior art, the specific process is not detailed here.

[0099] 7) Applying the optimal nonlinear model obtained in step 6) to the optimized seismic attribute combination of the entire study area to obtain a target data volume of the study area with the same type as the target curve.

[0100] Specifically, the optimal nonlinear model is obtained by training known samples at the well locations through a multi-attribute linear regression analysis coupled with a deep feedforward neural network. It represents the optimal nonlinear relationship between the target curve and the optimized seismic attribute combination. Because the target curve is of the same data type as the proposed low-frequency model (e.g., if a P-wave low-frequency model is proposed, the target curve is a P-wave curve), the optimal nonlinear model can be extended from the well location to the entire study area (generally, the overall tectonic setting and sedimentary environment of the study area should be consistent with those of the drilled wells). This means that the optimal nonlinear model is applied to the optimized seismic attribute combination for the entire study area, resulting in the target data volume for the study area. The target data volume is the data volume before the target low-frequency model is low-pass filtered (e.g., if a P-wave low-frequency model is proposed, the target data volume is the P-wave data volume before low-pass filtering), and its type is consistent with the target curve.

[0101] In addition, the "curve" is the logging data at the well location (or the drilled well), while the "data volume" and "low-frequency model" are both data volumes of the entire study area; when training the optimal nonlinear model, the wellside waveform curve extracted from the seismic attributes in the optimized seismic attribute combination is used; when obtaining the target data volume of the study area, the seismic attribute volume of each seismic attribute in the optimized seismic attribute combination in the entire study area is used.

[0102] 8) Perform low-pass filtering on the target data volume obtained in step 7) to obtain the target low-frequency model of the study area, specifically:

[0103] 8.1) Set the cutoff frequency value according to the actual frequency of the target data volume.

[0104] 8.2) Assign all frequency components of the target data volume that are greater than the cutoff frequency value to 0, and only retain the frequency components of the target data volume that are not greater than the cutoff frequency value, to obtain the target low-frequency model of the study area that is consistent with the target curve and target data volume type.

[0105] 9) Based on the target low-frequency model obtained in step 8), the optimized well logging data intersection analysis results in step 2) are used for fitting to obtain other low-frequency models of the study area.

[0106] Specifically, there is a Gardner empirical relationship between the longitudinal wave velocity and density: Among them, V p represents the longitudinal wave velocity, ρ represents the density, and C1 and C2 are both constants. There is a Castagna empirical relationship between the longitudinal wave velocity and the shear wave velocity: V s =C3V p +C4, where V p represents the longitudinal wave velocity, V srepresents shear wave velocity, and C3 and C4 are constants (the low-frequency models used in seismic inversion can be calculated from the low-frequency models of compressional wave velocity, shear wave velocity, and density). The values ​​of the constants C1, C2, C3, and C4 in the above two empirical relationships must be obtained by fitting the results of the intersection analysis of actual well logging data for the target layer in the study area.

[0107] Specifically, the fitting is performed using the results of well logging data intersection analysis, including:

[0108] 9.1) Intersect the density logging data and the P-wave velocity logging data of the target layer in the study area and fit them to form the relationship Then we can get the values ​​of constants C1 and C2.

[0109] 9.2) Intersect the shear wave velocity logging data and the compressional wave velocity logging data of the target layer in the study area and fit them to form the relationship V s =C3V p +C4, and then get the values ​​of constants C3 and C4.

[0110] 9.3) Based on the values ​​of constants C1, C2, C3, and C4, the Gardner and Castagna empirical relationship for the target layer in the study area is obtained.

[0111] 9.4) Using the obtained Gardner and Castagna empirical relationship, according to the target low-frequency model (any one of the P-wave velocity, S-wave velocity, or density low-frequency model) of the target layer segment in the study area obtained in step 8), obtain other low-frequency models of the study area.

[0112] Specifically, other low-frequency models refer to low-frequency models that may be used in seismic inversion other than the target low-frequency model, including low-frequency models of elastic parameters such as P-wave velocity, S-wave velocity, density, P-wave impedance, S-wave impedance, P-wave velocity ratio, Poisson's ratio and Lame coefficient. Among them, the low-frequency models of P-wave velocity, S-wave velocity and density can be directly obtained by the above two empirical formulas, and the low-frequency models of elastic parameters such as P-wave impedance, S-wave impedance, P-wave velocity ratio, Poisson's ratio and Lame coefficient can be calculated based on the target low-frequency model and the low-frequency models of P-wave velocity, S-wave velocity and density obtained according to the above two empirical formulas through the relationship between elastic parameters.

[0113] It should be noted that this step is not mandatory. If the target low-frequency model is the desired low-frequency model, this step is not necessary. If a P-wave impedance low-frequency model is required in post-stack seismic inversion and the target low-frequency model is the P-wave impedance low-frequency model, this step is not necessary.

[0114] The following describes the multi-information fusion artificial intelligence low-frequency model construction method of the present invention in detail, taking the low-frequency modeling and reservoir prediction of a gravity flow channel-lobed complex on a continental slope in a deepwater area as a specific example. The geological and seismic conditions in this study area are complex, and it is difficult to determine the boundaries of sand bodies and characterize high-quality reservoirs. In addition, the number of wells drilled is small, the logging data is limited, the well control range is large, and the seismic data quality is low. The application of conventional low-frequency modeling methods in this study area is severely restricted:

[0115] 1) Obtain the original logging data and pre-stack CMP gathers of Well01 and Well02 drilled on the deepwater fan lithology on the continental slope of the study area, and Well03, Well04, Well05 and Well06 drilled on the shelf margin delta.

[0116] The six drilled wells selected in this embodiment all contain high-quality P-wave curves in the target layer, and the well-seismic calibration results are good, which meets the requirements for establishing a P-wave low-frequency model. In addition, the target layer has a large depth range, which is more conducive to training and learning.

[0117] 2) The original well logging data obtained is optimized and processed, and fine well-seismic calibration is performed using the optimized original well logging data. The target curve of the drilled well whose well-seismic correlation meets the preset requirements is selected.

[0118] 3) Perform depth migration on the pre-stack CMP gathers to obtain seismic layer velocities and pre-stack CRP gathers.

[0119] 4) The pre-stack CRP gathers are stacked at different angles to obtain the near, middle and far stacked seismic data volumes.

[0120] 5) Using the target curve selected in step 2) as the learning target, multi-attribute linear regression analysis training is performed based on the seismic layer velocity obtained in step 3) and the near-, mid-, and far-path partial stacking seismic data volumes obtained in step 4), to establish the relationship between the target curve and multiple seismic attributes, and select the optimal seismic attribute combination that minimizes the prediction error.

[0121] Specifically, the optimized seismic attribute combination that minimizes the verification error includes the inverse of the seismic layer velocity, the instantaneous amplitude derivative of the near-track partially stacked seismic data volume, the instantaneous amplitude derivative of the far-track partially stacked seismic data volume, the absolute amplitude integral of the far-track partially stacked seismic data volume, the square of the middle-track partially stacked seismic data volume, the square of the far-track partially stacked seismic data volume, the amplitude envelope of the far-track partially stacked seismic data volume, the orthogonal tracks of the near-track partially stacked seismic data volume, and the derivative of the middle-track partially stacked seismic data volume.

[0122] 6) Using the target curve selected in step 2) as the learning target, and performing deep feedforward neural network analysis training based on the optimized seismic attribute combination selected in step 5), an optimal nonlinear model between the target curve and the optimized seismic attribute combination is obtained.

[0123] 7) Applying the optimal nonlinear model obtained in step 6) to the optimized seismic attribute combination of the entire study area to obtain a target data volume of the study area with the same type as the target curve.

[0124] like Figure 2 and Figure 3 As shown in the figure, compared with the seismic interval velocity, the seismic P-wave results obtained by the method of the present invention have a significant frequency compensation. In particular, the compensation of low-frequency information in the 5-10 Hz range is of great significance for reservoir prediction, as this low-frequency information cannot be obtained by conventional low-frequency modeling methods.

[0125] like Figure 4 As shown in FIG, the P-wave data volume obtained by the method of the present invention is highly consistent with the P-wave measured results of the drilled well, which indicates that a relatively accurate target data volume can be obtained by using the method.

[0126] 8) Perform low-pass filtering on the target data volume obtained in step 7) to obtain a target low-frequency model of the study area.

[0127] 9) Since the logging data intersection analysis results show that the lithologic indicator factor "P-wave impedance × (Vp / Vs)" is most sensitive to the sandstone in the target layer of the study area, this embodiment uses the above intersection analysis results for fitting, and obtains the low-frequency model of the lithologic indicator factor of the study area based on the target low-frequency model obtained in step 8).

[0128] like Figure 5 and Figure 6 As shown, the low-frequency model of lithologic indicator factors obtained using conventional well interpolation and extrapolation methods has low resolution and accuracy, exhibits a significant "bull's eye" phenomenon at the well points, and fails to demonstrate the reasonable geological patterns of the target strata in the study area of ​​this embodiment. However, the low-frequency model of lithologic indicator factors constructed using the method of the present invention has improved resolution and accuracy, not only eliminating the "bull's eye" phenomenon but also more clearly characterizing the geological background and sedimentary characteristics of the target strata in the study area of ​​this embodiment. This allows for the clear identification of the shelf-edge delta, the continental slope deepwater fan, and the shelf slope break between them. Therefore, the method of the present invention has a better implementation effect.

[0129] The low-frequency model is combined with the relative inversion results of the lithologic indicator factors to obtain the absolute inversion results of the lithologic indicator factors. Figure 7 and Figure 8As shown, compared with the results of low-frequency model fusion using conventional methods, the results of low-frequency model fusion obtained using the method of the present invention can more clearly characterize the continental slope deep-water fan lithologic body of the target layer section in the study area in this embodiment, which is of great significance for reservoir prediction in the study area.

[0130] Example 2

[0131] This embodiment provides a multi-information fusion artificial intelligence low-frequency model construction system, including:

[0132] The target curve selection module is used to select the target curve of the drilled wells based on the original logging data of the drilled wells in the study area.

[0133] The pre-stack CMP gather processing module is used to obtain seismic layer velocity and near-, middle- and far-track partial stack seismic data volumes based on the pre-stack CMP gathers of the wells drilled in the study area.

[0134] The optimal nonlinear model determination module is used to select the optimal seismic attribute combination based on the near, middle and far track partial stacking seismic data volumes, seismic layer velocity and target curve, and obtain the optimal nonlinear model between the target curve and the optimized seismic attribute combination.

[0135] The target data volume determination module is used to apply the optimal nonlinear model to the optimized seismic attribute combination of the entire study area to obtain the target data volume of the study area with the same type as the target curve.

[0136] The low-frequency model determination module is used to perform low-pass filtering on the target data volume to obtain the target low-frequency model of the study area.

[0137] In a preferred embodiment, it further comprises:

[0138] The other low-frequency model determination module is used to fit the obtained target low-frequency model using the logging data intersection analysis results to obtain other low-frequency models of the study area.

[0139] Example 3

[0140] This embodiment provides a processing device corresponding to the multi-information fusion artificial intelligence low-frequency model construction method provided in this embodiment 1. The processing device can be suitable for client processing devices, such as mobile phones, laptops, tablet computers, desktop computers, etc., to execute the method of embodiment 1.

[0141] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to facilitate communication between them. The memory stores a computer program executable on the processing device. When the processing device executes the computer program, it executes the method for constructing a multi-information fusion artificial intelligence low-frequency model provided in Example 1.

[0142] In some implementations, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage.

[0143] In other implementations, the processor may be a central processing unit (CPU), a digital signal processor (DSP), or other general-purpose processors, which are not limited herein.

[0144] In addition, the logical instructions in the above-mentioned memory can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. 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 and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0145] Those skilled in the art will understand that the structure of the above-mentioned computing device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computing device to which the solution of the present application is applied. The specific computing device may include more or fewer components, or combine certain components, or have a different component arrangement.

[0146] Example 4

[0147] This embodiment provides a computer program product corresponding to the method for constructing an artificial intelligence low-frequency model of multi-information fusion provided in this embodiment 1. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the method for constructing an artificial intelligence low-frequency model of multi-information fusion described in this embodiment 1 are loaded.

[0148] Computer readable storage media can be tangible devices that hold and store instructions used by instruction execution devices. Computer readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any combination thereof.

[0149] The above embodiment provides a computer-readable storage medium, whose implementation principle and technical effects are similar to those of the above method embodiment, and will not be repeated here.

[0150] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0151] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0152] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0153] The above embodiments are only used to illustrate the present invention, wherein the structure, connection mode and manufacturing process of each component can be changed. Any equivalent transformations and improvements based on the technical solution of the present invention should not be excluded from the scope of protection of the present invention.

Claims

1. A method for constructing an artificial intelligence low-frequency model based on multi-information fusion, characterized in that: include: Based on the original logging data of the wells drilled in the study area, the target curve of the wells that have been drilled is selected; Based on the pre-stack CMP gathers of the wells drilled in the study area, seismic layer velocities and stacked seismic data volumes of the near, middle and far tracks are obtained; Based on the near, mid and far track partial stacked seismic data volumes, seismic layer velocities and target curves, the optimal seismic attribute combination is selected, and the optimal nonlinear model between the target curve and the optimized seismic attribute combination is obtained; Apply the optimal nonlinear model to the optimized seismic attribute combination of the entire study area to obtain the target data volume of the study area with the same type as the target curve; Perform low-pass filtering on the target data volume to obtain the target low-frequency model of the study area; Based on the pre-stack CMP gathers of the wells drilled in the study area, seismic layer velocities and stacked seismic data volumes of near, middle and far tracks are obtained, including: Obtain pre-stack CMP gathers of wells drilled in the study area; Perform depth migration on the pre-stack CMP gathers to obtain seismic layer velocity and CRP gathers; The CRP gathers are stacked at different angles to obtain the near, middle and far track partial stacked seismic data volumes; The method selects an optimized seismic attribute combination based on the near-, mid- and far-path partial stacking seismic data volumes, seismic layer velocities and target curves, and obtains an optimal nonlinear model between the target curve and the optimized seismic attribute combination, including: Based on the near, mid and far track partial stacking seismic data volumes, seismic layer velocities and target curves, multi-attribute linear regression analysis training is performed to select the optimal seismic attribute combination; According to the target curve and the optimized seismic attribute combination, deep feedforward neural network analysis and training are carried out to obtain the optimal nonlinear model between the target curve and the optimized seismic attribute combination.

2. The method for constructing a multi-information fusion artificial intelligence low-frequency model according to claim 1, characterized in that: Also includes: According to the obtained target low-frequency model, the logging data intersection analysis results were used for fitting to obtain other low-frequency models of the study area.

3. The method for constructing a multi-information fusion artificial intelligence low-frequency model according to claim 1, characterized in that: The target curve of the drilled wells is selected based on the original logging data of the wells drilled in the study area, including: Obtain the original logging data of the wells drilled in the study area and optimize it; The P-wave curve and density curve in the optimized original logging data are used for well-seismic calibration, and the target curve of the drilled well whose well-seismic correlation meets the preset requirements is selected.

4. The method for constructing a multi-information fusion artificial intelligence low-frequency model according to claim 1, wherein: The near, medium and far track partial stacked seismic data volumes refer to corresponding seismic data volumes obtained by stacking small, medium and large angle pre-stack CRP gathers, respectively.

5. The method for constructing a multi-information fusion artificial intelligence low-frequency model according to claim 1, wherein: The target data volume is subjected to low-pass filtering to obtain a target low-frequency model of the study area, including: Set the cutoff frequency value according to the actual frequency of the target data body; Only the frequency components of the target data volume that are not greater than the cutoff frequency value are retained to obtain the target low-frequency model of the study area that is consistent with the target curve and target data volume type.

6. A multi-information fusion artificial intelligence low-frequency model construction system, characterized by: include: The target curve selection module is used to select the target curve of the drilled wells based on the original logging data of the wells drilled in the study area; Pre-stack CMP gather processing module is used to obtain seismic layer velocity and near-, mid- and far-track partial stack seismic data volumes based on pre-stack CMP gathers of wells drilled in the study area; The optimal nonlinear model determination module is used to select the optimal seismic attribute combination based on the near-, mid- and far-path partial stacking seismic data volumes, seismic layer velocities and target curves, and obtain the optimal nonlinear model between the target curve and the optimized seismic attribute combination; The target data volume determination module is used to apply the optimal nonlinear model to the optimized seismic attribute combination of the entire study area to obtain the target data volume of the study area with the same type as the target curve; The low-frequency model determination module is used to perform low-pass filtering on the target data volume to obtain the target low-frequency model of the study area; Based on the pre-stack CMP gathers of the wells drilled in the study area, seismic layer velocities and stacked seismic data volumes of near, middle and far tracks are obtained, including: Obtain pre-stack CMP gathers of wells drilled in the study area; Perform depth migration on the pre-stack CMP gathers to obtain seismic layer velocity and CRP gathers; The CRP gathers are stacked at different angles to obtain the near, middle and far track partial stacked seismic data volumes; The method selects an optimized seismic attribute combination based on the near-, mid- and far-path partial stacking seismic data volumes, seismic layer velocities and target curves, and obtains an optimal nonlinear model between the target curve and the optimized seismic attribute combination, including: Based on the near, mid and far track partial stacking seismic data volumes, seismic layer velocities and target curves, multi-attribute linear regression analysis training is performed to select the optimal seismic attribute combination; According to the target curve and the optimized seismic attribute combination, deep feedforward neural network analysis and training are carried out to obtain the optimal nonlinear model between the target curve and the optimized seismic attribute combination.

7. A processing device, characterized in that The method comprises computer program instructions, wherein when the computer program instructions are executed by a processing device, they are used to implement the steps corresponding to the method for constructing an artificial intelligence low-frequency model of multi-information fusion according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, wherein the computer program instructions, when executed by the processor, are used to implement the steps corresponding to the method for constructing an artificial intelligence low-frequency model of multi-information fusion according to any one of claims 1 to 5.

Images