A method for predicting formation pore pressure

Abstract

The present invention belongs to the technical field of oil and gas exploration, and more specifically, relates to a method for predicting formation pore pressure, including using a machine learning method to establish a statistical regression model of logging lithology parameters and seismic attributes, and predicting the longitudinal wave, density, resistivity and gamma lithology parameters of each channel of cross-well seismic according to the model; then using a machine learning method to establish a statistical regression model of well logging lithology parameters and logging pore pressure, and then predicting the formation pore pressure body according to the model. The present invention avoids complex mathematical derivation and calculation processes, and improves prediction efficiency and prediction accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of oil and gas exploration, and more specifically, relates to a method for predicting formation pore pressure. Background Art

[0002] Formation pressure refers to the pressure of the fluid contained in rock pores, also known as formation pore fluid pressure. Under normal formation pressure conditions, the formation pore fluid pressure is equal to the corresponding hydrostatic pressure. When the formation pore fluid pressure trend line deviates from the hydrostatic pressure trend line, it is called pore pressure. During oil and gas exploration and development, inaccurate predictions of formation pore pressure often lead to wellbore collapse and fracture, which not only hinders project progress but also results in significant economic losses. Therefore, accurate prediction of formation pore pressure is crucial for drilling design.

[0003] The theories of formation pore pressure prediction methods are compaction theory, equilibrium theory and effective pressure theory. Commonly used pore pressure prediction methods include equivalent depth method, Eaton method, effective stress method, etc. The basic principle of the equivalent depth method is that, without considering formation temperature, for pure rock formations without gas, the magnitude of the acoustic transit time depends on the lithology, degree of compaction, porosity, and the fluid content in the pores. Therefore, under normal compaction, the porosity of shale decreases with increasing depth H, exhibiting an exponential decay pattern. Pore pressure prediction is based on this pattern. The Eaton method is based on the principle that the relationship between the ratio of the actual value of the compaction observation parameter to the normal trend and the formation pore pressure is determined by the change in the overburden pressure gradient. The Eaton method is a commonly used empirical method for predicting formation pore pressure. It comprehensively considers the role of other high-pressure formation mechanisms in addition to compaction conditions and summarizes and references the relationship between measured drilling pressure and various logging information. It is a relatively practical method. The effective stress method is based on the principle that the vertical effective stress of the sediment continues to increase during formation compaction and remains at its maximum value during compaction after compaction and diagenesis. The vertical effective stress should be determined according to the loading conditions, comprehensively considering porosity, shale content, and effective stress. The pore pressure of the formation is determined based on the relationship between pore pressure and effective stress properties.

[0004] However, there is an error between the formation pore pressure calculated by any of the above porosity prediction methods based on well logging data and the measured pressure. This is mainly because the formulas of different methods are different and their applicability to different formations is also different. Due to the different compaction levels, porosities and mud contents of different formations, the different formulas themselves are imperfect, which increases the error of the prediction results.

[0005] To address the above technical issues, a machine learning-based formation pore pressure prediction method disclosed in the prior art includes the following prediction steps: a. Data processing and preparation: collecting relevant well logging data and related petrophysical parameters; b. Sensitivity curve determination: preparing reference and comparison sequences for the grey correlation method to determine sensitive well logging curves; c. Model training and testing: dividing the original dataset into training and test sets, inputting the training set into a gradient boosting regression tree model to obtain the optimal model; and d. Formation pore pressure prediction: using the sensitive well logging curves as input feature vectors for the optimal model to predict reservoir formation pressure. This method significantly improves prediction accuracy, prediction range, and reliability. However, the final prediction results remain inaccurate based solely on well logging data and petrophysical parameters. Summary of the Invention

[0006] In order to overcome at least one of the above-mentioned defects in the prior art, the present invention provides a formation pore pressure prediction method with high accuracy and high speed.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for predicting formation pore pressure is provided, comprising the following steps:

[0009] S1: extracting the wellside instantaneous phase based on the wellside seismic data, and decomposing the wellside seismic data to obtain the wellside transient decomposition waveform;

[0010] S2: Using the time-domain well P-wave, density, resistivity, and gamma-ray lithologic parameters as training label data, and the wellside instantaneous phase and wellside transient decomposition waveforms as training sample data, a machine learning method is used to establish a statistical regression model between logging lithologic parameters and seismic attributes, which is recorded as the first statistical regression model.

[0011] S3: Extract the instantaneous phase and transient decomposition waveform of the cross-well seismic data one by one as prediction sample data, and use the first statistical regression model to predict the longitudinal wave, density, resistivity and gamma lithologic parameters of each cross-well seismic data;

[0012] S4: Using the pore pressure of the time domain well as the label data, a machine learning method is used to establish a statistical regression model of well logging lithology parameters and logging pore pressure, which is recorded as the second statistical regression model;

[0013] S5: Using the P-wave, density, resistivity and gamma lithologic parameters of each cross-well seismic trace obtained in step S3 as prediction sample data, a second statistical regression model is used to predict the formation pore pressure body.

[0014] In this scheme, the P-wave, density, resistivity, and gamma-ray lithologic parameters of each crosswell seismic trace are first predicted using near-well seismic data and time-domain well data. The P-wave, density, resistivity, and gamma-ray lithologic parameters of each crosswell seismic trace are then used to predict the formation pore pressure body. This avoids complex mathematical derivation and calculation processes and improves the prediction rate.

[0015] Preferably, in the above step S1, the empirical mode decomposition method is used to decompose the wellbore seismic data to obtain the wellbore transient decomposition waveform.

[0016] Preferably, in the above step S2, a support vector machine statistical learning method is used to establish a statistical regression model of well logging lithologic parameters and seismic attributes.

[0017] Preferably, the above-mentioned use of the support vector machine statistical learning method to establish a statistical regression model of well logging lithologic parameters and seismic attributes specifically includes the following steps:

[0018] S21: Create sample training set x = {SWave ij ,Sphase j}(i=1,2,…,num,j=1,2,…,n), the sample label is y, where SWave is the wellside transient decomposition waveform, its waveform dimension is num, and Sphase is the wellside instantaneous phase;

[0019] S22: Establish the support vector machine regression mapping objective function based on the sample training set and sample labels:

[0020] S23: introducing a relaxation factor to improve the support vector machine regression mapping objective function in step S22;

[0021] S24: using Lagrangian optimization to solve the constrained optimization problem for the formula optimized in step S23 to obtain a regression estimation function;

[0022] S25: A heuristic sequential minimum optimization algorithm is used to solve the regression estimation function to obtain a statistical regression model of the well logging lithologic parameters and seismic attributes, which is recorded as the first statistical regression model.

[0023] Preferably, the sample labels are respectively the longitudinal wave, density, resistivity and gamma of the time domain well in step S1, and the number of sample labels is n.

[0024] Preferably, in the process of establishing the first statistical regression model in the above step S2, high-pass filtering is performed on the P-wave, density, resistivity and gamma lithologic parameters.

[0025] Preferably, the support vector machine regression mapping objective function in the above step S22 is specifically:

[0026]

[0027] Among them, l ε is the loss function, C is the soft constraint coefficient, ε is the allowable deviation between f(φ(x)) and y, w and b are parameters to be determined, and m is the length of the sample training set and sample labels.

[0028] Preferably, in the above step S23, a relaxation factor is introduced to improve the support vector machine regression mapping objective function in step S22, specifically:

[0029]

[0030] Among them, ξ i and is the relaxation factor, and m is the length of the sample training set and sample labels.

[0031] Preferably, in the above step S24, the constrained optimization problem is solved for the improved formula in step S23 to obtain the regression estimation function. Specifically, the original problem is transformed into a dual problem by using the Lagrangian optimization idea with inequality constraints, and the solution of the original problem is obtained by solving the dual problem. The specific formula is:

[0032]

[0033] The estimated function obtained is:

[0034]

[0035] Among them, α and is the Lagrange multiplier, K(x i ,x j )=φ(x i )φ(x j ) is the kernel function.

[0036] Preferably, in the above step S4, the same support vector machine statistical learning method as in step S2 is used to establish a statistical regression model of well logging lithologic parameters and well logging pore pressure.

[0037] Compared with the prior art, the beneficial effects are:

[0038] On the one hand, the present invention adopts the time-frequency analysis method of empirical mode decomposition to obtain the earthquake transient decomposition waveform to overcome the problem that the basis function cannot adapt to non-stationary earthquake signals. There is no need to conduct preliminary analysis and research on the earthquake signal, and the transient decomposition waveform is directly obtained by hierarchical decomposition to improve the final prediction effect and rate; on the other hand, the model is established through the support vector machine statistical learning method, which can rely on a small number of key samples for training, and the obtained model has good robustness and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic block diagram of the process of establishing a first statistical regression model of the formation pore pressure prediction method according to Example 1 of the present invention;

[0040] Figure 2 This is a schematic block diagram of the process of establishing a second statistical regression model of the formation pore pressure prediction method according to Example 1 of the present invention;

[0041] Figure 3 1 is a schematic diagram of a wellside transient decomposition waveform of a formation pore pressure prediction method according to Example 1 of the present invention;

[0042] Figure 4 is with Figure 3 The corresponding original seismic waveform;

[0043] Figure 5 is the original seismic profile;

[0044] Figure 6 The present invention is the embodiment 1 of the formation pore pressure prediction method and Figure 5 The corresponding formation pore pressure profile;

[0045] Figure 7 is the original well logging curve of the formation pore pressure prediction method of Example 2 of the present invention;

[0046] Figure 8 The method for predicting formation pore pressure in Example 2 of the present invention is Figure 7 The original logging curve and the logging curve after high-pass filtering. DETAILED DESCRIPTION

[0047] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting the present invention. To better illustrate the present embodiment, some components in the accompanying drawings may be omitted, enlarged, or reduced in size, and do not represent actual product dimensions. Those skilled in the art will appreciate that some well-known structures and their descriptions may be omitted from the accompanying drawings. The positional relationships depicted in the accompanying drawings are for illustrative purposes only and are not to be construed as limiting the present invention.

[0048] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "long", "short", etc. indicating the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting this patent. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0049] The technical solution of the present invention is further described in detail below through specific embodiments and in conjunction with the accompanying drawings:

[0050] Example 1:

[0051] like Figures 1 to 6 A first embodiment of a formation pore pressure prediction method is shown, comprising the following steps:

[0052] S1: extracting the wellside instantaneous phase based on the wellside seismic data, and decomposing the wellside seismic data to obtain the wellside transient decomposition waveform;

[0053] S2: Using the time-domain well P-wave, density, resistivity, and gamma-ray lithologic parameters as training label data, and the wellside instantaneous phase and wellside transient decomposition waveforms as training sample data, a machine learning method is used to establish a statistical regression model between logging lithologic parameters and seismic attributes, which is recorded as the first statistical regression model.

[0054] S3: Extract the instantaneous phase and transient decomposition waveform of the cross-well seismic data one by one as prediction sample data, and use the first statistical regression model to predict the longitudinal wave, density, resistivity and gamma lithologic parameters of each cross-well seismic data;

[0055] S4: Using the pore pressure of the time-domain well as the label data and the P-wave, density, resistivity and gamma-ray lithology curves of the time-domain well as the training sample data, a machine learning method is used to establish a statistical regression model of well logging lithology parameters and logging pore pressure, which is recorded as the second statistical regression model;

[0056] S5: Using the P-wave, density, resistivity and gamma lithologic parameters of each cross-well seismic trace obtained in step S3 as prediction sample data, a second statistical regression model is used to predict the formation pore pressure body.

[0057] It should be noted that the sequence of steps in this embodiment is only a reference implementation method and cannot be understood as a limitation of this solution. During the specific implementation process, step S4 can be performed at any time before step S5.

[0058] In step S1 of this embodiment, empirical mode decomposition (EMD) is used to decompose the wellbore seismic data to obtain a wellbore transient decomposition waveform. EMD overcomes the problem of basis functions being unable to adapt to non-stationary seismic signals. During the entire signal analysis process, there is no need for prior analysis or research of the signal. Instead, hierarchical decomposition is performed directly to obtain layered basis functions, i.e., transient decomposition waveforms. This method offers superior performance compared to traditional Fourier transforms and continuous wavelet transforms.

[0059] In step S2 of this embodiment, a support vector machine statistical learning method is used to establish a statistical regression model between well logging lithologic parameters and seismic attributes. This method avoids the inductive and deductive deduction process of training data, relying on a small number of key samples for model training. The results exhibit good robustness and generalization. Furthermore, well logging lithologic parameters are low-dimensional, so the use of kernel functions in the support vector machine modeling process enables nonlinear mapping and the mining of latent features in the data.

[0060] In this embodiment, the use of the support vector machine statistical learning method to establish a statistical regression model of well logging lithology parameters and seismic attributes specifically includes the following steps:

[0061] S21: Create sample training set x = {SWave ij ,Sphase j}(i=1,2,…,num,j=1,2,…,n), the sample label is y, where SWave is the wellside transient decomposition waveform, its waveform dimension is num, and Sphase is the wellside instantaneous phase;

[0062] S22: Establish the support vector machine regression mapping objective function based on the sample training set and sample labels:

[0063] S23: introducing a relaxation factor to improve the support vector machine regression mapping objective function in step S22;

[0064] S24: using Lagrangian optimization to solve the constrained optimization problem for the improved formula in step S23 to obtain a regression estimation function;

[0065] S25: A heuristic sequential minimum optimization algorithm is used to solve the regression estimation function to obtain a statistical regression model of the well logging lithologic parameters and seismic attributes, which is recorded as the first statistical regression model.

[0066] In this embodiment, the sample labels are the P-wave Vp, density Rhob, resistivity Rt, and gamma GR of the time-domain wells in step S1. The number of sample labels is n. That is, the sample label length matches the length of the sample training set. Four models are established using the P-wave Vp, density Rhob, resistivity Rt, and gamma GR of the time-domain wells as labels, respectively, to facilitate the subsequent prediction of P-wave Vp, density Rhob, resistivity Rt, and gamma GR lithologic parameters from cross-well seismic data.

[0067] The support vector machine regression mapping objective function in step S22 of this embodiment is specifically:

[0068]

[0069]

[0070] Among them, l ε is the loss function, C is the soft constraint coefficient, ε is the allowable deviation between f(φ(x)) and y (soft interval parameter constraint), w and b are parameters to be determined, m is the length of the sample training set and sample label, otherwise represents other cases (i.e. countercondition )

[0071] In addition, the goal of support vector machine modeling is to learn a mapping f(φ(x))=wφ(x)+b so that it is as close as possible to the value of y, where f is the regression mapping relationship and φ(x) is a nonlinear function that maps the training sample x to a high-dimensional feature space.

[0072] In step S23 of this embodiment, a relaxation factor is introduced to improve the support vector machine regression mapping objective function in step S22, specifically:

[0073]

[0074] Among them, ξ i and is the relaxation factor, and m is the length of the sample training set and sample labels.

[0075] In step S24 of this embodiment, the constrained optimization problem is solved for the improved formula in step S23 to obtain the regression estimation function. Specifically, the original problem is transformed into a dual problem by using the Lagrangian optimization idea with inequality constraints. The solution of the original problem is obtained by solving the dual problem. The specific formula is:

[0076]

[0077] The estimated function obtained is:

[0078]

[0079] Among them, α and is the Lagrange multiplier, K(x i ,x j )=φ(x i )φ(x j ) is the kernel function.

[0080] In addition, in step S25 of this embodiment, since the Lagrangian dual problem needs to satisfy the constraints of the KKT condition, a heuristic sequential minimum optimization algorithm is used to solve the regression estimation function.

[0081] Example 2:

[0082] like Figure 7 、 Figure 8The second embodiment of a formation pore pressure prediction method is shown. The only difference between this embodiment and embodiment 1 is that in the process of establishing the first statistical regression model in step S2 of this embodiment, the compressional wave, density, resistivity and gamma lithologic parameters are high-pass filtered. This is because the low-frequency trend of the logging data is highly correlated with the formation hydrostatic pressure, while the medium and high frequency parts of the logging data are related to the formation pore pressure. In this way, the low-frequency trend of the logging data can be filtered out by high-pass filtering, so as to enhance the statistical regression mapping relationship between the logging lithologic parameters and the seismic attributes in the process of formation pore pressure prediction, simplify the dependence on the empirical value of the hydrostatic pressure, and thus improve the overall prediction accuracy.

[0083] Example 3:

[0084] The only difference between this embodiment and embodiment 1 or embodiment 2 is that in step S4 of this embodiment, the same support vector machine statistical learning method as in step S2 is used to establish a statistical regression model of well logging lithologic parameters and well logging pore pressure. That is, the following steps are performed:

[0085] Create a sample training set x = {Vp j ,Rhob j ,Rt j ,GR j}(j=1,2,…n), the sample label is the pore pressure of the time domain well, where Vp is the compressional wave, Rhob is the density, Rt is the resistivity Rt, and GR is the gamma;

[0086] Establish a support vector machine regression mapping objective function based on the sample training set and sample labels;

[0087] Introducing a relaxation factor to improve the support vector machine regression mapping objective function in step S22;

[0088] Using Lagrangian optimization to solve the constrained optimization problem for the improved formula in step S23, a regression estimation function is obtained;

[0089] The regression estimation function is solved by using a heuristic sequential minimum optimization algorithm, and a statistical regression model of logging lithologic parameters and logging pore pressure is obtained, which is recorded as the second statistical regression model.

[0090] The present invention is described with reference to the flowcharts or block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each process or block in the flowchart or block diagram, as well as the combination of processes or blocks in the flowchart 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 processes in the flowchart or block diagram. 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.

[0091] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

[0092] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the claims of the present invention.

Claims

1. A formation pore pressure prediction method, characterized in that: The following steps are involved: Extracting the wellside instantaneous phase according to the wellside seismic data, and decomposing the wellside seismic data to obtain the wellside transient decomposition waveform; The longitudinal wave, density, resistivity and gamma lithology parameters of the time domain well are respectively used as training label data, and the wellside instantaneous phase and wellside transient decomposition waveform are used as training sample data, and a machine learning method is used to establish a statistical regression model of well logging lithology parameters and seismic attributes, which is recorded as the first statistical regression model; Extracting the interwell seismic instantaneous phase and the interwell seismic transient decomposition waveform one by one as prediction sample data, and using the first statistical regression model to predict the longitudinal wave, density, resistivity and gamma lithology parameters of each interwell seismic channel; The pore pressure of the time domain well is used as the label data, and the P-wave, density, resistivity and gamma lithology curve of the time domain well is used as the training sample data. The machine learning method is used to establish a statistical regression model of well logging lithology parameters and logging pore pressure, which is recorded as the second statistical regression model. The predicted longitudinal wave, density, resistivity and gamma lithology parameters of each well-to-well seismic trace are used as prediction sample data, and the second statistical regression model is used to predict the formation pore pressure body; Decomposing the wellside seismic data to obtain the wellside transient decomposition waveform specifically comprises using an empirical mode decomposition method to decompose the wellside seismic data to obtain the wellside transient decomposition waveform; The use of a machine learning method to establish a statistical regression model of well logging lithology parameters and seismic attributes specifically includes using a support vector machine statistical learning method to establish a statistical regression model of well logging lithology parameters and seismic attributes; The method of establishing a statistical regression model of well logging lithology parameters and seismic attributes by using a support vector machine statistical learning method specifically includes the following steps: Create a sample training set x = {SWave ij ,Sphase j }(i=1,2,…,num,j=1,2,…,n), the sample label is y, where SWave is the wellside transient decomposition waveform, its waveform dimension is num, and Sphase is the wellside transient phase; Establish the support vector machine regression mapping objective function based on the sample training set and sample labels: Introducing a relaxation factor to improve the support vector machine regression mapping objective function; The Lagrangian optimization is used to solve the constrained optimization problem of the improved support vector machine regression mapping objective function and obtain the regression estimation function; The regression estimation function is solved by using a heuristic sequential minimum optimization algorithm, and a statistical regression model of logging lithology parameters and seismic attributes is obtained, which is recorded as the first statistical regression model.

2. A formation pore pressure prediction method according to claim 1, characterized in that: The sample labels y are the P-wave, density, resistivity and gamma of the well in the time domain, and the number of sample labels is n.

3. A formation pore pressure prediction method according to claim 2, characterized in that: In the process of establishing the first statistical regression model, high-pass filtering was performed on the P-wave, density, resistivity and gamma lithology parameters.

4. A formation pore pressure prediction method according to claim 3, characterized in that: The support vector machine regression mapping objective function is specifically: Among them, l ε is the loss function, C is the soft constraint coefficient, ε is the allowable deviation between f(φ(x)) and y, w and b are parameters to be determined, and m is the length of the sample training set and sample labels.

5. A formation pore pressure prediction method according to claim 4, characterized in that: The relaxation factor is introduced to improve the support vector machine regression mapping objective function, specifically: Among them, ξ i and is the relaxation factor, and m is the length of the sample training set and sample labels.

6. A formation pore pressure prediction method according to claim 5, characterized in that: The improved support vector machine regression mapping objective function solves the constrained optimization problem to obtain the regression estimation function. Specifically, the original problem is transformed into a dual problem by using the Lagrangian optimization idea with inequality constraints, and the solution of the original problem is obtained by solving the dual problem. The specific formula is: 0≤α i ≤C The estimated function obtained is: Among them, α and is the Lagrange multiplier, K(x i ,x j )=φ(x i )φ(x j ) is the kernel function.

7. A formation pore pressure prediction method according to claim 6, characterized in that: The use of machine learning methods to establish a statistical regression model of well logging lithology parameters and well logging pore pressure specifically adopts the same support vector machine statistical learning method as that used to establish the first statistical regression model to establish a statistical regression model of well logging lithology parameters and well logging pore pressure.

Images