Carbonate rock fracture-cavity prediction method and device, storage medium and electronic equipment
Through pre-training and tuning training of deep convolutional neural networks, combined with geological modes and logging data, the problem of difficult to achieve high precision in carbonate cracked reservoir prediction is solved, and efficient reservoir prediction and efficiency is achieved promote.
Patent Information
- Application Number
- CN202310538086.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-12
- Publication Date
- 2024-11-12
AI Technical Summary
The existing technology has problems of inefficiency caused by high precision and manual intervention in the prediction of carbonate rock fracture-hole reservoirs, especially in deep oil fields, and traditional methods rely on a large number of manual interaction analysis. .
Deep convolutional neural network is used for pre-training and tuning training, geological mode is used as "pseudo-label" for pre-training, and then tuning is used for "hard label" explained by well logging, combining least squares fitting method and multi-attribute earthquake Data, build a high-precision reservoir prediction model.
High-precision prediction of carbonate rock-cut cave-type reservoirs is achieved, reducing the cost of manual interactive analysis, improving work efficiency, and ensuring that the prediction results are in line with the geological laws of underground media.
Smart Images

Figure CN118938299A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of seismic data interpretation, and in particular to a carbonate fracture-vuggy reservoir prediction method, device and electronic equipment. Background Art
[0002] The development of carbonate reservoirs in oil fields in western China is affected by multiple factors such as sedimentation, diagenesis, and tectonic movement. The reservoirs are mainly developed as fracture-cavity carbonate reservoirs, and the reservoir burial depth is generally more than 4000m. The reservoir prediction of this type of carbonate reservoir is extremely difficult and can be called a world-class problem.
[0003] The current traditional fracture-cavity carbonate oil and gas reservoir description method uses expert experience to combine logging and seismic to predict reservoirs. In 2004, Lu Xin proposed a fracture-cavity reservoir description technology that comprehensively utilizes multidisciplinary technologies such as logging geology, reservoir geophysics, geostatistics, and reservoir development engineering theory. In 2005, Zhang Songyang proposed a method for dividing the development degree of fracture-cavity reservoirs, which predicts reservoirs based on known conventional logging data and imaging data and reference to comparative core data.
[0004] However, a large amount of manual intervention will inevitably lead to a decrease in overall work efficiency. Summary of the invention
[0005] In response to the above problems, the present application proposes a carbonate fracture-cave prediction method, device, storage medium and electronic device, which uses the "pseudo-labels" of geological models to pre-train the deep convolutional neural network, and then uses the "hard labels" of well logging interpretation reservoirs to tune and train the deep convolutional neural network. The resulting mature network model can be used to predict carbonate fracture-cave reservoirs, obtain high-precision reservoir prediction results, and reduce the cost of large-scale manual interactive analysis in this field.
[0006] In a first aspect of the present application, a carbonate rock fracture-cavity prediction method is provided, the method comprising:
[0007] Determine the reservoir identification result by using the least square fitting method based on the well logging data of the preset work area;
[0008] Using the recognition result as first label data and the multi-attribute seismic data of the preset work area as input data;
[0009] Pre-training a deep convolutional neural network using the first label data and the input data to obtain a pre-trained network model;
[0010] Using a preset reservoir classification label as second label data, and performing tuning training on the pre-trained network model through the second label data and the input data to obtain a trained neural network model;
[0011] Reservoir prediction is performed on the work area to be predicted using the trained neural network model.
[0012] Furthermore, a large overdetermined equation is constructed by linear combination of effective attributes at well points in the logging data and the expectation of encountering a reservoir, and the least square solution of the large overdetermined equation is determined.
[0013] Furthermore, the multi-attribute seismic data includes:
[0014] One or more of ant tracking data, texture-energy data, absorption attenuation data, directional derivative data, and texture-entropy data.
[0015] Furthermore, the network loss function during pre-training includes a mean square error function.
[0016] Furthermore, the network loss function during tuning training includes:
[0017] A loss function including a loss criterion is obtained by adding a loss criterion to the mean square error function.
[0018] Furthermore, when there is corresponding label data for the sample to be tested in the input data, the value of the missing criterion is set to 1;
[0019] When there is no corresponding label data for the sample to be tested in the input data, the value of the missing criterion is set to 0.
[0020] Furthermore, during the tuning training process, the following steps are also included:
[0021] Dividing the second label data into data sets to obtain one or more training sets;
[0022] The deep convolutional neural network is tuned and trained according to each training set to obtain one or more deep convolutional neural network models.
[0023] The entire work area is predicted through each deep convolutional neural network model to obtain one or more groups of prediction results;
[0024] Perform Gaussian weighting processing on the prediction results corresponding to each training set according to the distance between the sample to be tested and the well logging in each training set to obtain a first result;
[0025] All the first results are fused to obtain a prediction result.
[0026] The second aspect of the present application provides a carbonate fracture-vuggy reservoir prediction device, the device comprising:
[0027] A first identification result determination module, used to determine a first identification result of a first reservoir by a least squares fitting method according to the well logging data;
[0028] A pre-training data determination module, used to use the first recognition result as first label data and the multi-attribute seismic data as input data;
[0029] A pre-training module, used to pre-train a deep convolutional neural network using the first label data and the input data to obtain a pre-trained network model;
[0030] A tuning and training module, used for taking a preset reservoir classification label as second label data, and performing tuning and training on the pre-trained network model through the second label data and the input data, to obtain a trained neural network model;
[0031] The prediction module is used to perform reservoir prediction on the prediction area through the trained neural network model.
[0032] According to a third aspect of the present application, a computer-readable storage medium is provided, wherein a computer program stored in the computer-readable storage medium can be executed by one or more processors to implement the method described above.
[0033] The fourth aspect of the present application provides an electronic device, including a memory and one or more processors, wherein the memory stores a computer program, the memory and the one or more processors are communicatively connected to each other, and when the computer program is executed by the one or more processors, the method described above is implemented.
[0034] Compared with the prior art, the technical solution of the present application has the following advantages or beneficial effects:
[0035] A set of deep convolutional neural network training processes with geological background constraints is disclosed, which includes three modules. The first module enables the deep convolutional neural network model to have geological background information, the second module improves the accuracy of the deep convolutional neural network model, and the third module enables the deep convolutional neural network model of this method to be applied to actual data of a large range of three-dimensional work areas. Using the deep convolutional neural network model provided by this method, high-precision prediction of carbonate fracture-cave reservoirs can be quickly achieved, laying a data foundation for fracture-cave reservoir simulation and development. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in the relevant field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0037] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions thereof in the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0038] Figure 1 A flow chart of a carbonate fracture-vuggy reservoir prediction method provided in an embodiment of the present application;
[0039] Figure 2 A schematic diagram of a deep convolutional neural network pre-training;
[0040] Figure 3 A schematic diagram of tuning and training a deep convolutional neural network;
[0041] FIG4( a ) is a schematic diagram of a pre-trained network feature provided in an embodiment of the present application;
[0042] FIG4( b ) is a schematic diagram of a terminal network feature provided in an embodiment of the present application;
[0043] Figure 5 A schematic diagram for comparing the identification results of carbonate fracture-vuggy reservoirs provided in an embodiment of the present application;
[0044] Figure 6 A schematic diagram of a verification well reservoir identification result provided in an embodiment of the present application;
[0045] Figure 7 A schematic diagram of the structure of a device provided in an embodiment of the present application;
[0046] Figure 8 A connection block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0047] The following will describe the implementation methods of the present application in detail with reference to the accompanying drawings and embodiments, so that the implementation process of how the present application applies technical means to solve technical problems and achieve corresponding technical effects can be fully understood and implemented accordingly. The embodiments of the present application and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the protection scope of the present application.
[0048] It should also be noted that, for the convenience of description, only the parts related to the present disclosure are shown in the drawings.
[0049] Embodiment 1
[0050] This embodiment provides a method for predicting carbonate fracture-cave reservoirs. Since the convolutional neural network has many parameters, a large number of training samples are often required to train the network. In the field of seismic interpretation, labeled training data is often very scarce, and manually designing labels is very time-consuming and laborious. This is a very important issue in the application of deep learning methods in the field of geophysics. As for reservoir prediction problems, since well logging data is very limited and drilling costs are extremely high, the data labels obtained by experts through well logging interpretation are also very scarce. Directly training the network with a small number of labeled samples will result in overfitting. In addition, since well logging data is one-dimensional data, it is often difficult for the network to learn geological spatial structure information by training the network using only well logging labels.
[0051] To address the above problems, this method uses a two-stage method to train a deep convolutional neural network. First, a rough reservoir identification result is obtained using traditional methods, and it is used as a "pseudo-label" to pre-train the network to learn the geological patterns contained therein, especially the underground space pattern information. Then, the pre-trained network is tuned and trained using the precise logging interpretation reservoir "hard label" to obtain the final carbonate fracture-cave reservoir identification network model, which can achieve high-precision reservoir prediction.
[0052] Figure 1 A flowchart of a carbonate fracture-vuggy reservoir prediction method provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method disclosed in this embodiment includes the following steps:
[0053] Step 110: Determine the reservoir identification result by using the least squares fitting method based on the well logging data of the preset work area.
[0054] In some embodiments, a large overdetermined equation is constructed by linear combinations of effective attributes at well points in the logging data and expectations of encountered reservoirs, and a least squares solution of the large overdetermined equation is determined.
[0055] Optionally, a least squares fitting method is used to quickly calculate a rough result of the reservoir probability based on the logging data, and the result is named as a geological model "pseudo-label".
[0056] The reservoir expectation optimization technology based on the least squares method is adopted. The linear combination of effective attributes at the well point and the expectation of the reservoir encountered are used to construct a large overdetermined equation to obtain its least squares solution. The objective assignment of the calculated weight coefficients is analyzed to conduct sensitivity analysis on the seismic attributes, establish the association between the attributes and the target prediction parameters, and achieve the purpose of attribute optimization.
[0057] For different attribute bodies in the target area, an output is obtained through linear combination, that is, the right side of the combined equation group is the mathematical expectation of carbonate fracture-vuggy reservoirs obtained based on logging or development data.
[0058] This can be simplified to a more general mathematical problem. Given a function y = f(x) at m points x i The value y at i (i=1,2,…,m), requiring approximate function:
[0059]
[0060] where c 0 ,c 1 ,…,c n is a constant to be determined, are known functions. Obviously, these functions are linearly independent. If we want to take c 0 ,c 1 ,…,c n Make The minimum, then the obtained c 0 ,c 1 ,…,c n That is the least squares solution of each weight coefficient.
[0061] Obviously, at this time, because the number of equations in the system is greater than the number of unknowns, this system is an overdetermined system of equations. At this time, it is necessary to find a "most approximate" solution to the system of equations, that is, the least squares solution.
[0062] Specifically, the methods include:
[0063] If there is an n-dimensional vector x * , so that A T Ax=A T b. Take any n-dimensional vector make Then y≠0, and
[0064]
[0065] At this time x * is the least squares solution of Ax=b.
[0066] The method steps for solving the attribute coefficient weight include:
[0067] ①First, find out the attribute values of the target area;
[0068] ② Extract the central traces of each attribute of the target area within an effective radius area centered on the well point, then extract the values of the seismic trace within the boundary of the longitudinal layered segment carved by the cave or fracture and calculate their average (multi-channel average);
[0069] ③ Normalize the attribute values of each boundary belonging to the central channel;
[0070] ④ When b on the right side of the equation is the expected value of the fracture-cavity body prediction, the equation is an overdetermined system of equations. Solve the obtained normalized system of equations to find the least squares solution x of the equation;
[0071] ⑤The least squares solution x is the coefficient weight corresponding to each attribute value.
[0072] Although the result is relatively rough and lacks accuracy, it contains geological spatial structure information and can therefore be regarded as a "pseudo-label".
[0073] Step 120: Use the recognition result as first label data, and use the multi-attribute seismic data of the preset work area as input data.
[0074] In some embodiments, the multi-attribute seismic data includes:
[0075] One or more of ant tracking data, texture-energy data, absorption attenuation data, directional derivative data, and texture-entropy data.
[0076] Optionally, the input data for network training is multi-attribute seismic data. In this embodiment, five types of seismic attribute data, namely ant tracking data, texture-energy data, absorption attenuation data, directional derivative data, and texture-entropy data, are used, and the training label data is the geological model "pseudo-label". The actual result is a pre-trained network model containing geological background information.
[0077] Step 130: pre-train a deep convolutional neural network using the first label data and the input data to obtain a pre-trained network model.
[0078] In some embodiments, the network loss function during pre-training includes a mean square error function.
[0079] Optionally, the geological model "pseudo-label" and seismic attribute data are used to form training sample pairs, which can be used to pre-train the deep convolutional neural network so that the network can learn the geological model of the corresponding work area. At the same time, since this method can obtain the prediction results of the entire work area, it can generate very rich "pseudo-label" training data.
[0080] The deep convolutional neural network is trained using "pseudo labels", and the network loss function is the mean square error (MSE) function:
[0081]
[0082] Where z is the actual output of the network, y is the sample label, and n is the number of wells.
[0083] Step 140: Use the preset reservoir classification label as the second label data, and use the second label data and the input data to tune and train the pre-trained network model to obtain a trained neural network model.
[0084] Optionally, a small amount of well logging interpretation reservoir “hard labels” based on manual interpretation is used to fine-tune the pre-trained network model to improve the accuracy of the network model prediction.
[0085] In some embodiments, the network loss function during fine-tuning training includes:
[0086] A loss function including a loss criterion is obtained by adding a loss criterion to the mean square error function.
[0087] In some embodiments, when there is corresponding label data for the sample to be tested in the input data, the value of the loss criterion is set to 1;
[0088] When there is no corresponding label data for the sample to be tested in the input data, the value of the missing criterion is set to 0.
[0089] Since the well logging data is one-dimensional data, and the network input is two-dimensional seismic data, that is, not all points in the input training samples have their corresponding labels, it is impossible to directly train the network based on the MSE loss function. In order to achieve end-to-end network training and effectively utilize the spatial texture features of seismic data, we add a mask to the loss function so that only the loss calculated by the well logging samples is back-propagated and used to update the network parameters. The loss function with a mask is:
[0090]
[0091] Here, let m be a mask. If the sample is located on the well log (i.e., it has an accurate label), then m = 1, otherwise m = 0:
[0092]
[0093] Optionally, in this step, the input data for network training is multi-attribute seismic data, and the training label data is the logging interpretation reservoir "hard label". The actual result is a deep convolutional neural network model that can be used to identify carbonate fracture-cavity reservoirs.
[0094] In some embodiments, the process of performing tuning training further includes:
[0095] Dividing the second label data into data sets to obtain one or more training sets;
[0096] The deep convolutional neural network is tuned and trained according to each training set to obtain one or more deep convolutional neural network models.
[0097] The entire work area is predicted through each deep convolutional neural network model to obtain one or more groups of prediction results;
[0098] Perform Gaussian weighting processing on the prediction results corresponding to each training set according to the distance between the sample to be tested and the well logging in each training set to obtain a first result;
[0099] All the first results are fused to obtain a prediction result.
[0100] Optionally, carbonate reservoirs are complex in structure and have poor consistency in underground structures, which leads to some contradictory logging labels. In actual data, there are often two wells with similar waveforms, but the actual reservoir distribution and reservoir category are completely different. Conflicting labels can easily cause the network to fail to converge. In order to solve this problem, this method adopts a method of first predicting a single well and then performing spatial weighting. First, the labeled well logs are divided, and a well or several similar and adjacent wells are divided into a training set. For each training set, the network is trained according to the method of the second module to obtain multiple deep convolutional neural network models for the work area. Different deep convolutional neural network models are used to predict the entire work area to obtain multiple sets of prediction results. Then, the prediction results are Gaussian weighted according to the distance of the predicted sample from the well log in the training set.
[0101] An optional Gaussian weighting function includes:
[0102]
[0103] Among them, a, b, and c are control parameters, and d is the distance between the sample and the well logging in the training set.
[0104] The fusion function is:
[0105]
[0106] Among them, g(d i ) is the weighting coefficient, ri is the prediction result of group i, r 0 is the current prediction result, g 0 is the current weighting coefficient, and mix is the fusion result.
[0107] In some possible embodiments, the weighting coefficient may also be obtained by taking some common methods according to actual conditions, which are not specifically limited here.
[0108] Figure 4(a) and Figure 4(b) show the features of the pre-trained network model and the tuned network model on the ninth layer of the network. 2 features, the receptive domain size is 23×23. As can be seen from Figure 4(a) and Figure 4(b), the tuned network has learned more features of different scales of seismic data.
[0109] Step 150: Perform reservoir prediction on the work area to be predicted using the trained neural network model.
[0110] Optionally, after performing reservoir prediction on the work area to be predicted, the actual result obtained is a carbonate fracture-vuggy reservoir identification data volume for the entire work area.
[0111] In this embodiment, the current mainstream deep learning technology is introduced into the field of fracture-cavity reservoir description of fracture-cavity oil reservoirs, and a two-step method is used for network training. The pre-trained network uses the geological model as a label, and the tuned network uses the fracture-cavity reservoir classification as a label. Its advantage is that it can not only ensure the accuracy of the prediction, but also ensure that the predicted reservoir conforms to the geological laws of the underground medium.
[0112] The carbonate fracture-cave reservoir prediction method provided in this embodiment adopts a two-stage method to train a deep convolutional neural network. First, a rough reservoir identification result is obtained by using a traditional method, and it is used as a "pseudo-label" to pre-train the network to learn the geological pattern contained therein, especially the underground space pattern information. Then, the pre-trained network is tuned and trained using the precise logging interpretation reservoir "hard label" to obtain the final carbonate fracture-cave reservoir identification network model, which can achieve high-precision reservoir prediction. Specifically: the identification result of the reservoir is determined by the least squares fitting method according to the logging data of the preset work area; the identification result is used as the first label data, and the multi-attribute seismic data of the preset work area is used as the input data; the deep convolutional neural network is pre-trained by the first label data and the input data to obtain a pre-trained network model; the preset reservoir classification label is used as the second label data, and the pre-trained network model is tuned and trained by the second label data and the input data to obtain a trained neural network model; the trained neural network model is used to predict the reservoir of the predicted work area. A set of deep convolutional neural network training processes with geological background constraints is disclosed, which includes three modules. The first module enables the deep convolutional neural network model to have geological background information, the second module improves the accuracy of the deep convolutional neural network model, and the third module enables the deep convolutional neural network model of this method to be applied to actual data of a large-scale three-dimensional work area. Using the deep convolutional neural network model provided by this method, high-precision prediction of carbonate fracture-cave reservoirs can be quickly achieved, laying a data foundation for fracture-cave reservoir simulation and development.
[0113] Embodiment 2
[0114] This embodiment provides a specific example based on the first embodiment.
[0115] In this embodiment, the carbonate rock fracture-cavity data of a certain work area in western China is used as an example.
[0116] Step 1: Determine the reservoir identification result by using the least squares fitting method based on the logging data of the preset work area.
[0117] In some embodiments, a large overdetermined equation is constructed by linear combinations of effective attributes at well points in the logging data and expectations of encountered reservoirs, and a least squares solution of the large overdetermined equation is determined.
[0118] Optionally, a least squares fitting method is used to quickly calculate a rough result of the reservoir probability based on the logging data, and the result is named as a geological model "pseudo-label".
[0119] The reservoir expectation optimization technology based on the least squares method is adopted. The linear combination of effective attributes at the well point and the expectation of the reservoir encountered are used to construct a large overdetermined equation to obtain its least squares solution. The objective assignment of the calculated weight coefficients is analyzed to conduct sensitivity analysis on the seismic attributes, establish the association between the attributes and the target prediction parameters, and achieve the purpose of attribute optimization.
[0120] For different attribute bodies in the target area, an output is obtained through linear combination, that is, the right side of the combined equation group is the mathematical expectation of carbonate fracture-vuggy reservoirs obtained based on logging or development data.
[0121] This can be simplified to a more general mathematical problem. Given a function y = f(x) at m points x i The value y at i (i=1,2,…,m), requiring approximate function:
[0122]
[0123] where c 0 ,c 1 ,…,c n is a constant to be determined, are known functions. Obviously, these functions are linearly independent. If we want to take c 0 ,c 1 ,…,c n Make The minimum, then the obtained c 0 ,c 1 ,…,c n That is the least squares solution of each weight coefficient.
[0124] Obviously, at this time, because the number of equations in the system is greater than the number of unknowns, this system is an overdetermined system of equations. At this time, it is necessary to find a "most approximate" solution to the system of equations, that is, the least squares solution.
[0125] Specifically, the methods include:
[0126] If there is an n-dimensional vector x * , so that A T Ax=A T b. Take any n-dimensional vector make Then y≠0, and
[0127]
[0128] At this time x * is the least squares solution of Ax=b.
[0129] The method steps for solving the attribute coefficient weight include:
[0130] ①First, find out the attribute values of the target area;
[0131] ② Extract the central traces of each attribute of the target area within an effective radius area centered on the well point, then extract the values of the seismic trace within the boundary of the longitudinal layered segment carved by the cave or fracture and calculate their average (multi-channel average);
[0132] ③ Normalize the attribute values of each boundary belonging to the central channel;
[0133] ④ When b on the right side of the equation is the expected value of the fracture-cavity body prediction, the equation is an overdetermined system of equations. Solve the obtained normalized system of equations to find the least squares solution x of the equation;
[0134] ⑤The least squares solution x is the coefficient weight corresponding to each attribute value.
[0135] Although the result is relatively rough and lacks accuracy, it contains geological spatial structure information and can therefore be regarded as a "pseudo-label".
[0136] Optionally, in this embodiment, five common seismic attributes are used as network input, the region-specific geological pattern labels are used as network training labels, and the U-net network is used to train the reservoir prediction model. The transmission relationship between the data is as follows Figure 2 shown.
[0137] Step 2: Use the recognition result as the first label data and use the multi-attribute seismic data of the preset work area as input data.
[0138] In some embodiments, the multi-attribute seismic data includes:
[0139] One or more of ant tracking data, texture-energy data, absorption attenuation data, directional derivative data, and texture-entropy data.
[0140] Optionally, the input data for network training is multi-attribute seismic data. In this embodiment, five types of seismic attribute data, namely ant tracking data, texture-energy data, absorption attenuation data, directional derivative data, and texture-entropy data, are used, and the training label data is the geological model "pseudo-label". The actual result is a pre-trained network model containing geological background information.
[0141] Step three: pre-train a deep convolutional neural network using the first label data and the input data to obtain a pre-trained network model.
[0142] In some embodiments, the network loss function during pre-training includes a mean square error function.
[0143] Optionally, the geological model "pseudo-label" and seismic attribute data are used to form training sample pairs, which can be used to pre-train the deep convolutional neural network so that the network can learn the geological model of the corresponding work area. At the same time, since this method can obtain the prediction results of the entire work area, it can generate very rich "pseudo-label" training data.
[0144] The deep convolutional neural network is trained using "pseudo labels", and the network loss function is the mean square error (MSE) function:
[0145]
[0146] Where z is the actual output of the network, y is the sample label, and n is the number of wells.
[0147] Step 4: Use the preset reservoir classification label as the second label data, and use the second label data and the input data to tune and train the pre-trained network model to obtain a trained neural network model.
[0148] Optionally, a small amount of well logging interpretation reservoir “hard labels” based on manual interpretation is used to fine-tune the pre-trained network model to improve the accuracy of the network model prediction.
[0149] In some embodiments, the network loss function during fine-tuning training includes:
[0150] A loss function including a loss criterion is obtained by adding a loss criterion to the mean square error function.
[0151] In some embodiments, when there is corresponding label data for the sample to be tested in the input data, the value of the loss criterion is set to 1;
[0152] When there is no corresponding label data for the sample to be tested in the input data, the value of the missing criterion is set to 0.
[0153] Since the well logging data is one-dimensional data, and the network input is two-dimensional seismic data, that is, not all points in the input training samples have their corresponding labels, it is impossible to directly train the network based on the MSE loss function. In order to achieve end-to-end network training and effectively utilize the spatial texture features of seismic data, we add a mask to the loss function so that only the loss calculated by the well logging samples is back-propagated and used to update the network parameters. The loss function with a mask is:
[0154]
[0155] Here, let m be a mask. If the sample is located on the well log (i.e., it has an accurate label), then m = 1, otherwise m = 0:
[0156]
[0157] Optionally, in this step, the input data for network training is multi-attribute seismic data, and the training label data is the logging interpretation reservoir "hard label". The actual result is a deep convolutional neural network model that can be used to identify carbonate fracture-cavity reservoirs.
[0158] In some embodiments, the process of performing tuning training further includes:
[0159] Dividing the second label data into data sets to obtain one or more training sets;
[0160] The deep convolutional neural network is tuned and trained according to each training set to obtain one or more deep convolutional neural network models.
[0161] The entire work area is predicted through each deep convolutional neural network model to obtain one or more groups of prediction results;
[0162] Perform Gaussian weighting processing on the prediction results corresponding to each training set according to the distance between the sample to be tested and the well logging in each training set to obtain a first result;
[0163] All the first results are fused to obtain a prediction result.
[0164] Optionally, carbonate reservoirs are complex in structure and have poor consistency in underground structures, which leads to some contradictory logging labels. In actual data, there are often two wells with similar waveforms, but the actual reservoir distribution and reservoir category are completely different. Conflicting labels can easily cause the network to fail to converge. In order to solve this problem, this method adopts a method of first predicting a single well and then performing spatial weighting. First, the labeled well logs are divided, and a well or several similar and adjacent wells are divided into a training set. For each training set, the network is trained according to the method of the second module to obtain multiple deep convolutional neural network models for the work area. Different deep convolutional neural network models are used to predict the entire work area to obtain multiple sets of prediction results. Then, the prediction results are Gaussian weighted according to the distance of the predicted sample from the well log in the training set.
[0165] An optional Gaussian weighting function includes:
[0166]
[0167] Among them, a, b, and c are control parameters, and d is the distance between the sample and the well logging in the training set.
[0168] The fusion function is:
[0169]
[0170] Where g(d i ) is the weighting coefficient, r i is the prediction result for group i.
[0171] Optionally, in this embodiment, five common seismic attributes are used as network inputs, and reservoir classification labels after logging interpretation are used as network training labels. The transmission relationship between data is as follows: Figure 3 shown.
[0172] The prediction results obtained are as follows Figure 5 (right figure) as shown, where Figure 5 The left picture is a schematic diagram of the seismic profile, the middle picture is the result of the traditional carbonate fracture-vuggy reservoir prediction method, and the right picture is a schematic diagram of the carbonate fracture-vuggy reservoir prediction results based on geological background constraints. Figure 5 (Left) shows the seismic data of the carbonate reservoir. Figure 5 (Middle figure) shows the carbonate reservoir prediction results obtained by traditional interpretation methods. Well A in the figure represents an actual exploration well in the work area. The red, blue and yellow blocks in the figure represent reservoirs encountered in actual drilling exploration. It can be seen that the carbonate reservoir prediction results obtained by the present invention are Figure 5 (Right) Prediction results obtained with traditional interpretation methods Figure 5 (Middle Figure) Compared with the well logging, the prediction results are highly consistent with the well logging, and can accurately detect missed reservoirs and accurately identify non-reservoir parts with higher recognition resolution.
[0173] Step 5: Perform reservoir prediction on the work area to be predicted using the trained neural network model.
[0174] Optionally, after performing reservoir prediction on the work area to be predicted, the actual result obtained is a carbonate fracture-vuggy reservoir identification data volume for the entire work area.
[0175] Optionally, the seismic attributes of the area where the sample well is located are used as network input, and the prediction results are as follows: Figure 6 (right figure) as shown, among which, Figure 6 The left picture is a schematic diagram of the seismic profile, and the right picture is a schematic diagram of the carbonate fracture-vuggy reservoir type prediction results based on geological background constraints. Figure 6 The left picture in the figure shows the seismic data of carbonate reservoir in this work area. Figure 6Well B in the figure represents an actual exploration well in the work area, and the information of this well has not been used as a sample for network training for the training of the network model. Well B in the figure shows the actual detected P-wave velocity curve. In this work area, the low value of the velocity curve is often used as an indication of the reservoir. In the area designated by the circle in the right figure of Figure 6, the reservoir prediction result obtained using the present invention shows an obvious red area, indicating that the probability of the development of the reservoir at this location is very high, which is completely consistent with the low value area of the Well B velocity curve actually detected, indicating that the prediction result has a higher consistency on the non-sample well, which can illustrate that the network model has good generalization.
[0176] In this embodiment, the current mainstream deep learning technology is introduced into the field of fracture-cavity reservoir description of fracture-cavity oil reservoirs, and a two-step method is used for network training. The pre-trained network uses the geological model as a label, and the tuned network uses the fracture-cavity reservoir classification as a label. Its advantage is that it can not only ensure the accuracy of the prediction, but also ensure that the predicted reservoir conforms to the geological laws of the underground medium.
[0177] Embodiment 3
[0178] This embodiment provides a device. This device embodiment can be used to execute the method embodiment of the present application. For details not disclosed in the device embodiment, please refer to the method embodiment of the present application. Figure 7 A schematic diagram of a structure of a device provided in an embodiment of the present application is shown in FIG. Figure 7 As shown, the device 700 disclosed in this embodiment includes:
[0179] A first identification result determination module 701 is used to determine a first identification result of a first reservoir by using a least squares fitting method according to the well logging data;
[0180] A pre-training data determination module 702 is used to use the first recognition result as first label data and the multi-attribute seismic data as input data;
[0181] A pre-training module 703, configured to pre-train a deep convolutional neural network using the first label data and the input data to obtain a pre-trained network model;
[0182] A tuning and training module 704 is used to use the preset reservoir classification label as the second label data, and perform tuning and training on the pre-trained network model through the second label data and the input data to obtain a trained neural network model;
[0183] The prediction module 705 is used to perform reservoir prediction on the work area to be predicted by using the trained neural network model.
[0184] It can be understood by those skilled in the art that Figure 7The structure shown in the figure does not constitute a limitation on the device of the embodiment of the present application, and may include more or fewer modules / units than shown in the figure, or a combination of certain modules / units, or different module / unit arrangements.
[0185] It should be understood by those skilled in the art that the above modules or steps of the present application can be implemented by a general-purpose computing device, and they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation.
[0186] The device provided in this embodiment includes: a first recognition result determination module 701, which is used to determine the first recognition result of the first reservoir by the least squares fitting method according to the logging data; a pre-training data determination module 702, which is used to use the first recognition result as the first label data and the multi-attribute seismic data as the input data; a pre-training module 703, which is used to pre-train the deep convolutional neural network through the first label data and the input data to obtain a pre-trained network model; a tuning training module 704, which is used to use the preset reservoir classification label as the second label data, and to tune and train the pre-trained network model through the second label data and the input data to obtain a trained neural network model; a prediction module 705, which is used to perform reservoir prediction on the predicted area through the trained neural network model. The current mainstream deep learning technology is introduced into the field of fracture-cavity reservoir description of fracture-cavity oil reservoirs, and the two-step method is used for network training. The pre-training network uses the geological model as the label, and the tuning network uses the fracture-cavity reservoir classification as the label. Its advantage is that it can not only ensure the accuracy of the prediction, but also ensure that the predicted reservoir conforms to the geological laws of the underground medium.
[0187] Embodiment 4
[0188] This embodiment further provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method steps in the aforementioned method embodiment can be implemented, and this embodiment will not be repeated here.
[0189] Among them, the computer-readable storage medium may also include computer programs, data files, data structures, etc. alone, or include a combination thereof. The computer-readable storage medium or computer program may be specifically designed and understood by a technician in the field of computer software, or the computer-readable storage medium may be known and available to a technician in the field of computer software. Examples of computer-readable storage media include: magnetic media, such as hard disks, floppy disks, and tapes; optical media, such as CDROM disks and DVDs; magneto-optical media, such as optical disks; and hardware devices, specifically configured to store and execute computer programs, such as read-only memory (ROM), random access memory (RAM), flash memory; or servers, app application stores, etc. Examples of computer programs include machine code (e.g., code generated by a compiler) and files containing high-level code, which can be executed by a computer using an interpreter. The described hardware device may be configured to be used as one or more software modules to perform the operations and methods described above, and vice versa. In addition, the computer-readable storage medium may be distributed in a networked computer system, and program code or computer programs may be stored and executed in a decentralized manner.
[0190] Embodiment 5
[0191] Figure 8 A connection block diagram of an electronic device provided in an embodiment of the present application, such as Figure 8 As shown, the electronic device 800 may include: one or more processors 801 , a memory 802 , a multimedia component 803 , an input / output (I / O) interface 804 , and a communication component 805 .
[0192] The one or more processors 801 are used to execute all or part of the steps in the above method embodiment. The memory 802 is used to store various types of data, which may include instructions of any application or method in the electronic device, and data related to the application.
[0193] One or more processors 801 can be an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a controller, a microcontroller, a microprocessor or other electronic components, and are used to execute the method in the aforementioned method embodiment.
[0194] The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.
[0195] The multimedia component 803 may include a screen and an audio component, the screen may be a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, and the microphone is used to receive external audio signals. The received audio signal may be further stored in a memory or sent via a communication component. The audio component also includes at least one speaker for outputting audio signals.
[0196] The I / O interface 804 provides an interface between one or more processors 801 and other interface modules, which may be a keyboard, a mouse, buttons, etc. These buttons may be virtual buttons or physical buttons.
[0197] The communication component 805 is used for wired or wireless communication between the electronic device 800 and other devices. Wired communication includes communication through a network port, a serial port, etc.; wireless communication includes: Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, 5G, or one or a combination of them. Therefore, the corresponding communication component 805 may include: a Wi-Fi module, a Bluetooth module, and an NFC module.
[0198] In summary, the present application provides a carbonate fracture prediction method, device, storage medium and electronic device. The method trains a deep convolutional neural network by adopting a two-stage method. First, a rough reservoir identification result is obtained by using a traditional method, and the result is used as a "pseudo-label" to pre-train the network to learn the geological pattern contained therein, especially the underground space pattern information. Then, the pre-trained network is tuned and trained using the precise logging interpretation reservoir "hard label" to obtain the final carbonate fracture reservoir identification network model, which can achieve high-precision reservoir prediction. Specifically: the identification result of the reservoir is determined by the least squares fitting method according to the logging data of the preset work area; the identification result is used as the first label data, and the multi-attribute seismic data of the preset work area is used as the input data; the deep convolutional neural network is pre-trained by the first label data and the input data to obtain a pre-trained network model; the preset reservoir classification label is used as the second label data, and the pre-trained network model is tuned and trained by the second label data and the input data to obtain a trained neural network model; the trained neural network model is used to predict the reservoir of the predicted work area. A set of deep convolutional neural network training processes with geological background constraints is disclosed, which includes three modules. The first module enables the deep convolutional neural network model to have geological background information, the second module improves the accuracy of the deep convolutional neural network model, and the third module enables the deep convolutional neural network model of this method to be applied to actual data of a large-scale three-dimensional work area. Using the deep convolutional neural network model provided by this method, high-precision prediction of carbonate fracture-cave reservoirs can be quickly achieved, laying a data foundation for fracture-cave reservoir simulation and development.
[0199] In addition, it should be understood that the method or system disclosed in the embodiments provided in the present application can also be implemented in other ways. The method or system embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the methods and devices according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a computer program segment or a part of a computer program, and the module, computer program segment or a part of a computer program contains one or more computer programs for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings, and can actually be executed in parallel. They can also be executed in the opposite order sometimes, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer programs.
[0200] In this application, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements that are inherent to such process, method, article, or apparatus. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, device or equipment including the elements; if there is a description of "first", "second", etc., it is only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features; in the description of this application, unless otherwise specified, the terms "multiple" and "multiple" mean at least two; if there is a description of a server, it should be noted that the server can be an independent physical server or terminal, or a server cluster composed of multiple physical servers, or a cloud server that can provide basic cloud computing services such as cloud servers, cloud databases, cloud storage and CDN; if there is a description of a smart terminal or mobile device in this application, it should be noted that the smart terminal or mobile device can be a mobile phone, a tablet computer, a smart watch, a netbook, a wearable electronic device, a personal digital assistant (PDA), an augmented reality technology device (AR), a virtual reality device (VR), a smart TV, a smart speaker, a personal computer (PC), a tablet computer, a smart phone, a smart TV, a smart speaker, a smart phone ... Computer, PC), etc., but not limited to this, and the present application does not specifically limit the specific form of the smart terminal or mobile device.
[0201] Finally, it should be noted that, in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "an example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0202] Although the embodiments of the present application have been shown and described above, it is to be understood that the above embodiments are exemplary, and the contents described are only embodiments adopted for the convenience of understanding the present application, and are not intended to limit the present application. Any technician in the technical field to which the present application belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present application, but the scope of protection of the present application shall still be subject to the scope defined in the attached claims.
Claims
1. A method for predicting carbonate fracture-vuggy reservoirs. It is characterized in that The method comprises: Determine the reservoir identification result by using the least square fitting method based on the well logging data of the preset work area; Using the recognition result as first label data and the multi-attribute seismic data of the preset work area as input data; Pre-training a deep convolutional neural network using the first label data and the input data to obtain a pre-trained network model; Using a preset reservoir classification label as second label data, and performing tuning training on the pre-trained network model through the second label data and the input data to obtain a trained neural network model; Reservoir prediction is performed on the work area to be predicted using the trained neural network model.
2. The method according to claim 1, It is characterized in that A large overdetermined equation is constructed by linear combination of effective attributes at well points in the logging data and the expectation of drilling into a reservoir, and a least square solution of the large overdetermined equation is determined.
3. The method according to claim 1, It is characterized in that The multi-attribute seismic data includes: One or more of ant tracking data, texture-energy data, absorption attenuation data, directional derivative data, and texture-entropy data.
4. The method according to claim 1, It is characterized in that The network loss function during pre-training includes the mean square error function.
5. The method according to claim 4, It is characterized in that The network loss functions during tuning training include: A loss function including a loss criterion is obtained by adding a loss criterion to the mean square error function.
6. The method according to claim 5, It is characterized in that When there is corresponding label data for the sample to be tested in the input data, the value of the missing criterion is set to 1; When there is no corresponding label data for the sample to be tested in the input data, the value of the missing criterion is set to 0.
7. The method according to claim 1, It is characterized in that The process of tuning training also includes: Dividing the second label data into data sets to obtain one or more training sets; The deep convolutional neural network is tuned and trained according to each training set to obtain one or more deep convolutional neural network models. The entire work area is predicted through each deep convolutional neural network model to obtain one or more groups of prediction results; Perform Gaussian weighting processing on the prediction results corresponding to each training set according to the distance between the sample to be tested and the well logging in each training set to obtain a first result; All the first results are fused to obtain a prediction result.
8. A carbonate fracture-cavity reservoir prediction device, It is characterized in that include: A first identification result determination module, used to determine a first identification result of a first reservoir by a least squares fitting method according to the well logging data; A pre-training data determination module, used to use the first recognition result as first label data and the multi-attribute seismic data as input data; A pre-training module, used to pre-train a deep convolutional neural network using the first label data and the input data to obtain a pre-trained network model; A tuning and training module, used for taking a preset reservoir classification label as second label data, and performing tuning and training on the pre-trained network model through the second label data and the input data, to obtain a trained neural network model; The prediction module is used to perform reservoir prediction on the prediction area through the trained neural network model.
9. A computer-readable storage medium, It is characterized in that The computer program stored in the computer-readable storage medium, when executed by one or more processors, implements the method according to any one of claims 1 to 7.
10. An electronic device, It is characterized in that The method comprises a memory and one or more processors, wherein a computer program is stored in the memory, and the memory and the one or more processors are communicatively connected with each other, and when the computer program is executed by the one or more processors, the method according to any one of claims 1 to 7 is executed.
