Residual network-based tight oil reservoir fractured reservoir prediction method

Abstract

The invention provides a tight oil reservoir fractured reservoir prediction method based on a residual network. The prediction method comprises the steps that S1, time windows are taken from a target horizon, and multiple horizon seismic attributes are extracted; s2, using dual laterolog data to characterize the development degree of the fracture beside the well; s3, establishing a deep residual network model according to the data features, and training the deep residual network model to obtain a trained network model; and S4, predicting the tight oil reservoir fractured reservoir by adopting the training network model. And the prediction stability and efficiency of the tight oil reservoir fractured reservoir are improved.

Description

Technical Field

[0001] The invention relates to the field of geophysical exploration of tight oil reservoirs, and in particular to a method for predicting tight oil reservoir fracture reservoirs based on residual networks. Background Art

[0002] The degree of fracture development in tight oil reservoirs affects the physical properties, oil-water relationship, and fracturing effect of the reservoir. Effective prediction of fractures is of great practical significance for reservoir development, but the quantitative description of fractures in tight oil reservoirs has always been a difficult problem in reservoir prediction. Fractured reservoirs in tight oil reservoirs are a very complex reservoir type. Due to the complex reservoir space and filling properties of this type of reservoir, and the significant lateral changes in reservoir lithology, this type of reservoir has obvious heterogeneity and wide diversity, resulting in serious multi-solution problems in seismic prediction of fractured reservoirs. In order to solve this problem, a multidisciplinary, multi-method and multi-parameter combined method is used to comprehensively predict the development of fractures, thereby improving the reliability of fracture reservoir prediction and thus improving the success rate of oil and gas exploration and development.

[0003] In recent years, with the rapid development of computer science and technology, supported by the powerful computing power provided by computers, deep learning (DL) algorithms have been widely used in many fields and have achieved remarkable results. Most of the deep learning algorithm models are data-driven, that is, the algorithm obtains the laws of the data through in-depth learning and analysis of massive data. This model is consistent with the problem-solving model in many seismic exploration fields. The advantage of deep learning algorithms is that they can conduct comprehensive analysis of large amounts of data, thereby realizing multi-parameter and multi-angle analysis and learning. Introducing deep learning algorithms into solving the problem of oil and gas reservoir prediction in seismic exploration can achieve more efficient and accurate prediction of complex oil and gas reservoirs. Summary of the invention

[0004] In view of the above problems, the present invention is proposed to provide a method for predicting fractured reservoirs in tight oil reservoirs based on residual networks, which overcomes the above problems or at least partially solves the above problems.

[0005] According to one aspect of the present invention, a method for predicting fractured reservoirs in tight oil reservoirs based on residual networks is provided, the prediction method comprising:

[0006] Step S1: extracting multiple layer seismic attributes by taking a time window for the target layer;

[0007] Step S2: using dual laterolog data to characterize the degree of fracture development near the well;

[0008] Step S3: establishing a deep residual network model according to data features, and training the deep residual network model to obtain a training network model;

[0009] Step S4: using the training network model to predict tight oil reservoir fracture-type reservoirs.

[0010] Optionally, the step S1: extracting multiple layer seismic attributes from the target layer in a time window specifically includes:

[0011] The interpolated and encrypted layer data is used as the extraction index of seismic attributes, and the time window is used to extract various seismic layer attributes;

[0012] When extracting seismic attributes, each type should be extracted to ensure the comprehensiveness of the attribute information.

[0013] Optionally, the step S1: extracting multiple layer seismic attributes by taking a time window for the target layer also includes: for coherence and curvature seismic volume attributes, first extracting volume attributes of post-stack seismic data, and then extracting data by layer as layer attributes.

[0014] Optionally, the seismic attributes are divided into amplitude, frequency, phase, energy, waveform, correlation, attenuation and ratio.

[0015] Optionally, the step S2: using the dual laterolog data to characterize the degree of development of the fractures near the wellbore specifically includes:

[0016] The difference between deep and shallow lateral resistivity logging is used to characterize the fracture opening and extension of the rock formation near the well.

[0017] If the well bed is developed with high-angle fractures, the deep lateral resistivity logging response value R LLD Greater than the shallow lateral resistivity logging response value R LLS , the difference between deep and shallow lateral resistivity logging shows a positive difference;

[0018] Rock formations with developed low-angle fractures will show the opposite behavior, with the difference between deep and shallow lateral resistivity logging showing a negative difference.

[0019] Optionally, the step S3: establishing a deep residual network model according to data features, and training the deep residual network model, and obtaining the training network model specifically includes:

[0020] The deep residual network model is adjusted according to the number of data channels of the seismic attribute combination, the convolution kernel is reduced during the convolution operation, and the deep residual network model is established;

[0021] The deep residual network model is established by training to obtain a training network model.

[0022] Optionally, the input layer of the deep residual network is set to a multi-channel convolution layer with a number of channels equal to the number of input seismic attribute combinations, and a total of 8 convolution kernels of size 3×3 are used to convolve the input;

[0023] After the maximum pooling process, the main network structure part of the deep residual network is entered;

[0024] The main part of the network structure consists of three stacked residual modules;

[0025] Finally, a linear fully connected layer is used as the output layer of the network.

[0026] Optionally, the main part of the network structure includes three residual modules stacked together, specifically including: each of the residual modules includes three convolutional layers and an identity mapping, and there is a maximum pooling layer after the three residual modules.

[0027] Optionally, the training and establishing the deep residual network model to obtain the training network model specifically includes:

[0028] The Adam optimization algorithm is used to train the deep residual neural network. The activation function uses the ReLU function as the activation function of the hidden layer, and the loss function of the network is the mean square error function.

[0029] The training data set of the network is a combination of various well-side seismic attributes. For each seismic attribute, 9 seismic attribute values ​​next to the corresponding position of the well coordinates are extracted as data;

[0030] The training label set of the network is the characterization value of the degree of fracture development near the well and the difference between deep and shallow lateral resistivity logging;

[0031] According to the number of seismic attribute combinations, a corresponding deep residual network model is established;

[0032] During training, the initial learning rate and number of iterations are adjusted according to the actual data volume and the training status of the network to ensure that the network converges stably while avoiding overfitting.

[0033] Optionally, the step S4: using the training network model to predict tight oil reservoir fractured reservoirs specifically includes:

[0034] After the deep residual network model is trained, multiple seismic attributes of the target layer are used as inputs of the network model to obtain the prediction results of the fracture development degree in the entire target layer.

[0035] The invention provides a method for predicting tight oil reservoir fractured reservoir based on residual network, the method comprising: step S1: extracting multiple layer seismic attributes from the target layer by taking a time window; step S2: using dual laterolog data to characterize the development degree of wellside fractures; step S3: establishing a deep residual network model according to data characteristics, and training the deep residual network model to obtain a training network model; step S4: using the training network model to predict tight oil reservoir fractured reservoirs. The prediction stability and efficiency of tight oil reservoir fractured reservoirs are improved.

[0036] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for predicting fractured reservoirs in tight oil reservoirs based on a deep residual network provided in an embodiment of the present invention;

[0039] Figure 2 Part of the seismic attribute data extracted by the embodiment of the present invention;

[0040] Figure 3 This is a structural diagram of a deep residual network model according to an embodiment of the present invention;

[0041] Figure 4 This is a training error curve diagram of an embodiment of the present invention;

[0042] Figure 5 This is a diagram showing the prediction result of the degree of crack development according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0044] The terms "comprises" and "having" and any variations thereof in the description embodiments, claims and drawings of the present invention are intended to cover non-exclusive inclusions, for example, including a series of steps or units.

[0045] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0046] A method for predicting fractured reservoirs in tight oil reservoirs based on a deep residual network comprises the following steps:

[0047] S1. Extract multiple layer seismic attributes from the target layer time window;

[0048] S2. Use dual laterolog data to characterize the degree of fracture development near the wellbore;

[0049] S3. Establish and train a deep residual network model based on data characteristics;

[0050] S4. Use the trained network model to predict fractured reservoirs in tight oil reservoirs.

[0051] Step S1 specifically includes:

[0052] The interpolated and encrypted layer data is used as the extraction index of seismic attributes, and the appropriate time window size is used to extract various seismic layer attributes. Seismic attributes can be roughly divided into several categories, such as amplitude, frequency, phase, energy, waveform, correlation, attenuation and ratio. When extracting seismic attributes, some of each type should be extracted to ensure the comprehensiveness of attribute information. For seismic volume attributes such as coherence and curvature, it is necessary to first extract the volume attributes of the post-stack seismic data, and then extract its data by layer as layer attributes.

[0053] Step S2 specifically includes:

[0054] The difference between deep and shallow lateral resistivity logging can be used to characterize the fracture opening and extension of the rock formation near the well. In general, the greater the difference between the deep and shallow lateral resistivity logging response values, the better the fracture opening and extension of the rock formation near the well.

[0055] If the rock formation near the well is developed with high-angle fractures, then the deep lateral resistivity logging response value (R LLD ) will be greater than the shallow lateral resistivity logging response value (R LLS ), that is, the difference between deep and shallow lateral resistivity logging shows a positive difference; the rock formation with developed low-angle fractures will show the opposite, that is, the difference between deep and shallow lateral resistivity logging will show a negative difference.

[0056] Step S3 specifically includes:

[0057] Since multiple seismic attributes are used as network input, their data features are quite different from the image data processed by the traditional residual network. First, image data is mostly single-channel or three-channel (RGB) data, while multiple seismic attributes are data with a higher number of channels, and the number of channels of different numbers of seismic attribute combinations is also different, so corresponding adjustments need to be made to the network structure. Secondly, since the seismic attributes are near-well seismic attributes and are not as large as image data, the size of the convolution kernel is appropriately reduced during the convolution operation. Compared with the traditional convolution operation, more attention is paid to the retention of data, allowing the network to obtain more useful information.

[0058] According to the characteristics of the above-mentioned seismic attribute data, the input layer of the deep residual network in the present invention is set to a multi-channel convolution layer with the same number of channels as the number of input seismic attribute combinations, and a total of 8 3×3 convolution kernels are used to convolve the input, and then enter the main network structure of the residual network after the maximum pooling process. The main structure of the network is composed of three residual modules stacked together, each of which contains three convolution layers and an identity mapping, and there is a maximum pooling layer after the three residual modules, and finally a linear fully connected layer is used as the output layer of the network.

[0059] The Adam optimization algorithm is used in the training of the deep residual neural network. The activation function uses the ReLU function as the activation function of the hidden layer. The loss function of the network is the mean square error (MSE) function commonly used in regression problems. The training data set of the network is a combination of various well-side seismic attributes. For each seismic attribute, 9 seismic attribute values ​​corresponding to the well coordinates are extracted as data. The training label set of the network is the characterization value of the degree of fracture development near the well and the difference between deep and shallow lateral resistivity logging. Since the input channel of the deep residual network will be different when using different numbers of seismic attribute combinations as training data sets, it is necessary to establish a corresponding deep residual network model according to the number of seismic attribute combinations. During training, the initial learning rate and the number of iterations also need to be adjusted according to the actual data volume and the training status of the network to ensure that the network converges stably while avoiding overfitting.

[0060] Step S4 specifically includes:

[0061] After the deep residual network model is trained, the multiple seismic attributes of the target layer are used as the input of the network model to obtain the prediction results of the degree of fracture development in the entire target layer.

[0062] The beneficial effects of the technical solution provided by the present invention are: based on the deep residual network, a prediction method for tight oil reservoir fracture type reservoir is provided. By using a variety of seismic attributes as data to train the deep residual network model, and using the trained network model to obtain the degree of fracture development of the target layer, a reliable and efficient solution is provided for the prediction of tight oil reservoir fracture type reservoir.

[0063] like Figure 1 As shown, Figure 1 The present invention is a flowchart of a method for predicting fractured reservoirs in tight oil reservoirs based on a deep residual network according to an embodiment of the present invention. The method comprises the following steps:

[0064] S1. Extract seismic attributes of various layers in the target layer time window.

[0065] In this embodiment, S1 specifically includes:

[0066] First, the target layer is subjected to seismic attribute extraction, and the layer attributes are extracted by taking a 10ms time window up and down along the layer. A total of four seismic attributes are extracted: average instantaneous frequency, average amplitude, reflection intensity slope, and coherence volume.

[0067] S2. Use dual laterolog data to characterize the degree of wellbore fracture development.

[0068] In this embodiment, S2 specifically includes:

[0069] The difference between deep and shallow lateral resistivity logging can be used to characterize the fracture opening and extension of the rock formation near the well. In general, the greater the difference between the deep and shallow lateral resistivity logging response values, the better the fracture opening and extension of the rock formation near the well.

[0070] If the rock formation near the well is developed with high-angle fractures, then the deep lateral resistivity logging response value (R LLD ) will be greater than the shallow lateral resistivity logging response value (R LLS ), that is, the difference between deep and shallow lateral resistivity logging shows a positive difference; the rock formation with low-angle fracture development will show the opposite, that is, the difference between deep and shallow lateral resistivity logging will show a negative difference. The following table lists the difference between deep and shallow lateral resistivity logging of the target layer in the work area.

[0071] hashtag Dual lateral resistivity difference (Ω) A -2.038 B 0.778 C 2.58 D -0.567 E -0.745 F 1.918 G -0.353 H 1.023 I 1.275 J 0.434

[0072] S3. Establish and train the deep residual network model based on data characteristics.

[0073] In this embodiment, step S3 specifically includes:

[0074] According to the above extracted earthquake attribute combination, the input layer of the deep residual network is set to a convolution layer with 4 channels, and 8 3×3 convolution kernels are used to convolve the input to extract features, and then enter the main network structure of the residual network after maximum pooling. The main structure of the network is composed of three residual modules stacked together, each of which contains three convolution layers and an identity mapping. There is a maximum pooling layer after the three residual modules, and finally a linear fully connected layer is used as the output layer of the network. The network structure used is as follows: Figure 3 shown.

[0075] The Adam optimization algorithm is used in the training of the deep residual neural network. The activation function uses the ReLU function as the activation function of the hidden layer. The loss function of the network is the mean square error (MSE) function commonly used in regression problems. The training data set of the network is a combination of various well-side seismic attributes. For each seismic attribute, 9 seismic attribute values ​​corresponding to the well coordinates are extracted as data. The training label set of the network is the characterization value of the degree of fracture development near the well and the difference between deep and shallow lateral resistivity logging. During training, the initial learning rate was set to 0.001, and the training was performed for a total of 200 iterations. The error curve of the training process is shown in the figure below. Figure 4 shown.

[0076] S4. Use the trained network model to predict fractured reservoirs in tight oil reservoirs.

[0077] In this embodiment, step S4 specifically includes:

[0078] After the deep residual network model is trained, the four seismic attributes of the target layer are used as the input of the network model to obtain the prediction results of the fracture development degree in the entire target layer. The prediction results obtained in this embodiment are as follows: Figure 5 shown.

[0079] In one embodiment of the present invention, Figure 2 is the preferred seismic attribute data result of the embodiment of the present invention, wherein Figure 2 (a) is the average instantaneous frequency, Figure 2 (b) is the average amplitude, Figure 2 (c) is the reflection intensity slope, Figure 2 (d) is a coherent body. Figure 3The deep residual network model structure used in the embodiment of the present invention has an input layer that is set to a multi-channel convolution layer with the same number of channels as the number of input seismic attribute combinations. A total of 8 3×3 convolution kernels are used to convolve the input. After the maximum pooling process, it enters the main network structure of the residual network. The main network structure is composed of three residual modules stacked together, each of which contains three convolution layers and an identity mapping. The size of the convolution layer is the number marked in the figure, and finally it is processed by the maximum pooling process again. The output of the network is obtained. Figure 4 This is a training error curve diagram of an embodiment of the present invention. It can be seen that after 200 iterations, the network model converges smoothly. Figure 5 This is the fracture development degree prediction result of the embodiment of the present invention. By comparing it with the wellside fracture development degree data of other wells in the work area, the effectiveness of the present invention is verified.

[0080] Beneficial effects: The present invention first uses post-stack seismic data to extract a variety of seismic attributes, and then uses dual lateral logging data to characterize the degree of wellbore fracture development; then a deep residual network model matching the extracted seismic attributes is established and trained; finally, the trained deep residual network model is used to predict the degree of fracture development in the target layer, thereby improving the prediction stability and efficiency of fractured reservoirs in tight oil reservoirs, and the effectiveness of the method is verified through examples.

[0081] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting fractured reservoirs in tight oil reservoirs based on residual networks, characterized in that: The prediction method comprises: Step S1: extracting multiple layer seismic attributes by taking a time window for the target layer; Step S2: using dual laterolog data to characterize the degree of fracture development near the well; Step S3: establishing a deep residual network model according to data features, and training the deep residual network model to obtain a training network model; Step S4: using the training network model to predict tight oil reservoir fracture-type reservoirs.

2. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 1, characterized in that: The step S1: extracting multiple layer seismic attributes from the target layer in a time window specifically includes: The interpolated and encrypted layer data is used as the extraction index of seismic attributes, and the time window is used to extract various seismic layer attributes; When extracting seismic attributes, each type should be extracted to ensure the comprehensiveness of the attribute information.

3. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 1, characterized in that: The step S1: extracting multiple layer seismic attributes from the target layer in a time window also includes: for coherence and curvature seismic volume attributes, first extracting volume attributes of post-stack seismic data, and then extracting data according to the layer as layer attributes.

4. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 1, characterized in that: The seismic attributes are classified into amplitude, frequency, phase, energy, waveform, correlation, attenuation and ratio.

5. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 1, characterized in that: The step S2: using the dual laterolog data to characterize the development degree of the fractures near the wellbore specifically includes: The difference between deep and shallow lateral resistivity logging is used to characterize the fracture opening and extension of the rock formation near the well. If the well bed is developed with high-angle fractures, the deep lateral resistivity logging response value R LLD Greater than the shallow lateral resistivity logging response value R LLS , the difference between deep and shallow lateral resistivity logging shows a positive difference; Rock formations with developed low-angle fractures will show the opposite behavior, with the difference between deep and shallow lateral resistivity logging showing a negative difference.

6. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 1, characterized in that: The step S3: establishing a deep residual network model according to data features, and training the deep residual network model, and obtaining the training network model specifically includes: The deep residual network model is adjusted according to the number of data channels of the seismic attribute combination, the convolution kernel is reduced during the convolution operation, and the deep residual network model is established; The deep residual network model is established by training to obtain a training network model.

7. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 6, characterized in that: The input layer of the deep residual network is set to a multi-channel convolution layer with the same number of channels as the number of input seismic attribute combinations, and a total of 8 convolution kernels of size 3×3 are used to convolve the input; After the maximum pooling process, the main network structure part of the deep residual network is entered; The main part of the network structure consists of three stacked residual modules; Finally, a linear fully connected layer is used as the output layer of the network.

8. The method for predicting tight oil reservoir fractures based on residual network according to claim 7, characterized in that: The main part of the network structure includes three residual modules stacked together, specifically including: each residual module includes three convolutional layers and an identity mapping, and there is a maximum pooling layer after the three residual modules.

9. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 6, characterized in that: The training and establishing of the deep residual network model to obtain the training network model specifically includes: The Adam optimization algorithm is used to train the deep residual neural network. The activation function uses the ReLU function as the activation function of the hidden layer, and the loss function of the network is the mean square error function. The training data set of the network is a combination of various well-side seismic attributes. For each seismic attribute, 9 seismic attribute values ​​next to the corresponding position of the well coordinates are extracted as data; The training label set of the network is the characterization value of the degree of fracture development near the well and the difference between deep and shallow lateral resistivity logging; According to the number of seismic attribute combinations, a corresponding deep residual network model is established; During training, the initial learning rate and number of iterations are adjusted according to the actual data volume and the training status of the network to ensure that the network converges stably while avoiding overfitting.

10. The method for predicting fractured reservoirs in tight oil reservoirs based on residual networks according to claim 1, characterized in that: The step S4: using the training network model to predict the tight oil reservoir fracture type reservoir specifically includes: After the deep residual network model is trained, multiple seismic attributes of the target layer are used as inputs of the network model to obtain the prediction results of the fracture development degree in the entire target layer.

Images