Dry layer and effective reservoir deep learning identification method and device based on four parameters

Abstract

The invention provides a deep learning identification method and device for a dry layer and an effective reservoir based on four parameters, and the method comprises the steps: finding out one or more representative wells from a research region, and carrying out the interpretation of a layer section of a research target layer; digitalizing the lithology interpretation result of the target stratum; analyzing the logging data of the well in the research area, and normalizing the logging data of the representative well; constructing a deep learning network structure with an input layer with input neurons and bias neurons, a hidden layer and a neuron output layer by using a TensorFlow framework; setting a learning rate of a deep learning network gradient descent algorithm; setting a cost function of the deep learning network as a square error cost function; setting training times, and training the deep learning network on the basis of the training sample to obtain deep learning network parameters; and testing the training sample and the test sample by using the obtained deep learning network parameters. According to the method, the dry reservoir and the effective reservoir can be effectively distinguished from the complex and diversified reservoirs.

Description

Technical Field

[0001] The present invention belongs to the field of lithological interpretation in exploration, and in particular, relates to a deep learning identification method and device for dry layers and effective reservoirs based on four parameters. Background Art

[0002] With the continuous deepening of oil field exploration and development in my country, the exploration depth of thin interbedded continental sandstone and mudstone in eastern oil fields has been continuously deepened. At present, most of them have entered deep exploration. With the increase of exploration depth, the porosity of reservoirs has been greatly reduced relative to shallow layers. Most reservoirs have a porosity of less than 10%. Many of these reservoirs are not effective reservoirs, but dry layers. In order to improve the success rate of exploration and development, it is necessary to conduct in-depth research on dry layers and effective reservoirs, especially to correctly identify dry layers and effective reservoirs from the well. Regarding the identification of dry layers and effective reservoirs from the well, some scholars have conducted certain research, such as:

[0003] The invention patent "Method and device for determining effective dry layer" with application number 201210183588.X provides a method for determining effective dry layer. The method is based on multiple discrimination parameters and combines theoretical knowledge such as principal component analysis and fuzzy proximity to distinguish dry layer and effective reservoir.

[0004] The invention patent with application number 202211152464.5, "A logging identification method for dry layers in medium and high porosity", provides a method for identifying dry layers in medium and high porosity. This method is based on the dual parameters of porosity and resistivity, combined with the regional division method to distinguish dry layers and effective reservoirs.

[0005] The invention patent with application number 202010918769.7, "A method and system for distinguishing oil, water and dry layers based on weighted average value of resistivity", provides a method for distinguishing different reservoirs. This method is based on the dual parameters of resistivity and permeability, and uses weighted average and other means to distinguish oil, water layers and dry layers.

[0006] The invention patent with application number 202210943261.1 "Method for identifying sandstone gas layers and dry layers using flushing zones and formation water porosity" provides a method for distinguishing sandstone gas layers and dry layers. This method is based on calculating the resistivity of mud filtrate, formation resistivity, flushing zone resistivity, etc. to obtain the formation water porosity and flushing zone water porosity parameters, and then obtains the ratio of the flushing zone water porosity to the formation water porosity, and distinguishes gas layers and dry layers based on the meter plate constructed based on the ratio.

[0007] The document “Lower limits and main controlling factors of effective reservoir properties of Box 7-Box 8 segments in Linxing Central District, Ordos Basin” determines the cumulative reservoir properties under different porosity and permeability conditions based on core permeability and porosity data, and then determines the lower limits of permeability and porosity data of effective reservoirs in different layers.

[0008] From the existing literature and patents, the existing effective reservoir and dry layer discrimination methods have the following characteristics: (1) at least two parameters are required; (2) the original data needs to be properly processed and transformed; (3) some methods have unclear parameters, and the principal component analysis method is used to determine the final main parameters, and the parameters used are difficult to adapt to diverse reservoirs; some methods have clear parameters, but the final discrimination parameters used to determine the effective reservoir and dry layer are relatively few, which is difficult to adapt to complex reservoir conditions.

[0009] In conventional logging interpretation, the distinction between dry layers and effective reservoirs is mainly based on some logging data (such as acoustic data, gamma data, density data, resistivity data, etc.), and some new parameters are obtained by means of principal component analysis, ratios, etc., and the dry layers and effective reservoirs are distinguished by means of intersection analysis. This method generally requires the establishment of interpretation standards for stratified sections when applied, and is more suitable for simple reservoirs when applied. The existing methods have two shortcomings: one is that they are not suitable for complex reservoirs (reservoirs with complex pore structures and lithology characteristics); the other is that the interpretation efficiency is low. Therefore, in-depth research on the discrimination research methods of dry layers and effective reservoirs that are suitable for diverse reservoirs is of great significance for the exploration and development of deep sandstone and mudstone interbeds in eastern my country.

[0010] The current identification methods of dry layers and effective reservoirs are difficult to adapt to complex and diverse reservoirs. The existence of this problem affects researchers' correct understanding of dry layers and effective reservoirs, thereby affecting the effective interpretation of seismic data and increasing exploration risks. Summary of the invention

[0011] In order to improve the recognition accuracy and efficiency of dry layers and effective reservoirs, the embodiment of the present application provides a deep learning recognition method and device for dry layers and effective reservoirs based on four parameters, which can effectively distinguish dry layers and effective reservoirs in complex and diverse reservoirs, greatly improving the researchers' recognition ability and efficiency, and reducing the risks of subsequent exploration and development.

[0012] In the first aspect, an embodiment of the present application provides a deep learning identification method for dry layers and effective reservoirs based on four parameters, including: S1 finding one or more representative wells from the study area and interpreting the stratified sections of the target layer; S2 digitizing the lithological interpretation results of the target layer; S3 analyzing the logging data of the wells in the study area and normalizing the logging data of the representative wells; S4 using the TensorFlow framework to construct a deep learning network structure with an input layer, a hidden layer, and a neuron output layer having input neurons and bias neurons; S5, setting the learning rate of the deep learning network gradient descent algorithm; setting the cost function of the deep learning network to the square error cost function; setting the number of training times, training the deep learning network based on the training samples, and obtaining the deep learning network parameters; S6 using the obtained deep learning network parameters to test the training samples and test samples.

[0013] Wherein, step S1 includes interpreting the dry layer, effective reservoir and mudstone of the target layer segment.

[0014] In step S3, the logging data includes acoustic wave data, gamma data, density data, and resistance data. The AC range of the acoustic wave data is determined to be [AC min ,AC max ]、Gamma data GR range is [GR min ,GR max ]、Den range of density data Den is [Den min ,Den max ]、Resistance data R 25 The range is [R min ,R max ].

[0015] Among them, in step S3, the normalized data is AC T , GR T 、Den T , R T .

[0016] Among them, in step S5, AC T , GR T 、Den T , R T As training input samples, digital lithology interpretations of representative wells are taken as training output samples. The deep learning network is trained using the training samples to obtain the deep learning network parameters.

[0017] Among them, in step S6, if the test is consistent with the prediction, the trained network parameters are used for application and promotion in other wells; if it does not meet expectations, it is improved in terms of sample representativeness, network structure, and number of training times.

[0018] In the second aspect, an embodiment of the present application provides a deep learning identification device for dry layers and effective reservoirs based on four parameters, including: an interpretation unit, which finds one or more representative wells from the study area and interprets the stratified segments of the study target layer; a digitization unit, which is used to digitize the lithology interpretation results of the target layer; an analysis unit, which analyzes the logging data of the wells in the study area and normalizes the logging data of the representative wells; a construction unit, which is used to use the TensorFlow framework to construct a deep learning network structure with an input layer, a hidden layer, and a neuron output layer having input neurons and bias neurons; a setting unit, which is used to set the learning rate of the deep learning network gradient descent algorithm; set the cost function of the deep learning network to be a square error cost function; set the number of training times, train the deep learning network based on the training samples, and obtain the deep learning network parameters; a testing unit, which is used to test the training samples and test samples using the obtained deep learning network parameters.

[0019] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when executed by a processor.

[0020] In a fourth aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the program.

[0021] The deep learning identification method and device of dry layers and effective reservoirs based on four parameters in the embodiment of the present application have the following beneficial effects: the method can automatically discover the inherent laws of distinguishing dry layers and effective reservoirs from the data, and can adapt to complex and diverse reservoirs; the interpretation efficiency is high. The method provides a more convenient and effective measure for distinguishing dry layers and effective reservoirs from well logging, which can better guide practical applications, reduce actual exploration risks, and improve the success rate of exploration. It is conducive to promoting the development of interpretation technology in sandstone and mudstone interbedded areas in my country, and has a great promoting effect on the exploration of effective reservoirs in my country. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a schematic flow chart of a deep learning identification method for dry layers and effective reservoirs based on four parameters in an embodiment of the present application;

[0023] Figure 2 This is a schematic diagram of a deep learning network according to an embodiment of the present application;

[0024] Figure 3 This is a comprehensive cross-sectional view of Well S1 in the embodiment of the present application;

[0025] Figure 4 This is a comprehensive cross-sectional view of Well S2 of the present application embodiment;

[0026] Figure 5 A comprehensive cross-section of the S1 well with superimposed lithological interpretation in the embodiment of the present application;

[0027] Figure 6 This is the S1 comprehensive cross-section with digital lithological interpretation of the embodiment of this application;

[0028] Figure 7 The S2 comprehensive cross-section diagram with digital lithological interpretation of the embodiment of the present application;

[0029] Figure 8 A comprehensive cross-sectional diagram comparing the dry layer and effective reservoir predicted by deep learning of Well S1 in the present embodiment with the original interpretation;

[0030] Fig. 9 A comprehensive cross-sectional diagram comparing the dry layer and effective reservoir predicted by deep learning of Well S2 in the present application embodiment with the original interpretation;

[0031] Fig.10 A comprehensive cross-sectional diagram of the dry layer and effective reservoir predicted by deep learning of Well S3 in the present embodiment;

[0032] Fig.11 This is a schematic diagram of the structure of a deep learning identification device for dry layers and effective reservoirs based on four parameters in an embodiment of the present application. DETAILED DESCRIPTION

[0033] The present application is further described below in conjunction with the accompanying drawings and embodiments.

[0034] In the following introduction, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance. The following introduction provides multiple embodiments of the present invention, and different embodiments can be replaced or combined, so this application can also be considered to include all possible combinations of the same and / or different embodiments recorded. Therefore, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then this application should also be considered to include embodiments containing one or more of all other possible combinations of features A, B, C, and D, even though the embodiment may not be clearly recorded in the following text.

[0035] Embodiment 1

[0036] This application proposes a deep learning identification method for dry layers and effective reservoirs based on four parameters. 25 , AC, and Den parameters, and utilizing the better feature learning ability of deep learning methods, a method for automatically distinguishing diverse dry layers and effective reservoirs was formed based on the TensorFlow framework.

[0037] like Figure 1As shown, the present application provides a deep learning identification method for dry layers and effective reservoirs based on four parameters: S1, find one or more representative wells from the study area, and interpret the stratified sections of the study target layer; S2, digitize the lithological interpretation results of the target layer; S3, analyze the logging data of the wells in the study area, and normalize the logging data of the representative wells; S4, use the TensorFlow framework to construct a deep learning network structure with an input layer, a hidden layer, and a neuron output layer having input neurons and bias neurons; S5, set the learning rate of the deep learning network gradient descent algorithm; set the cost function of the deep learning network to the square error cost function; set the number of training times, train the deep learning network based on the training samples, and obtain the deep learning network parameters; S6, use the obtained deep learning network parameters to test the training samples and test samples.

[0038] The deep learning identification method for dry layers and effective reservoirs based on four parameters provided in this application can relatively effectively distinguish dry layers and effective reservoirs in complex and diverse reservoirs, greatly improving the researchers' discrimination ability and efficiency and reducing the risks of subsequent exploration and development.

[0039] Embodiment 2

[0040] The current logging methods used to distinguish dry layers and effective reservoirs have two shortcomings: one is that they are difficult to adapt to complex reservoirs (reservoirs with complex pore structures and lithology, etc.); the other is that the interpretation efficiency is low. In view of the shortcomings of the existing logging methods in interpreting dry layers and effective reservoirs, this application provides a deep learning identification method for dry layers and effective reservoirs based on four parameters. It is based on the conventional logging data in the study area, including acoustic data AC, gamma data GR, density data Den, and resistance data R 25 Based on this, a deep learning network constructed using the TensorFlow framework is used to form an interpretation method for automatically distinguishing dry layers and effective reservoirs, which can be applied to complex reservoirs.

[0041] The specific implementation steps are as follows:

[0042] (1) Find one or more representative wells in the study area;

[0043] (2) Based on conventional methods, the dry layer, effective reservoir and mudstone of the target layer are interpreted;

[0044] (3) Digitize the lithology interpretation results of the target layer, where 1 represents an effective reservoir, 0.5 represents a dry layer, and 0 represents a mudstone layer;

[0045] (4) Acoustic wave data AC, gamma data GR, density data Den, and resistance data R of the wells in the study area 25Analyze and determine the sound wave data range as [AC min ,AC max ], the gamma data range is [GR min ,GR max ], density data range is [Den min ,Den max ], the resistance data range is [R min ,R max ];

[0046] (5) The acoustic wave data AC, gamma data GR, density data Den, and resistance data R of the representative well are 25 Normalize, and assume that the normalized data is AC T , GR T 、Den T , R T ;

[0047] (6) Use the TensorFlow framework to build Figure 2 The deep learning network structure shown has an input layer with 4 input neurons + 1 bias neuron, N (no less than 3) hidden layers, and a neuron output layer;

[0048] (7) Set the learning rate LR of the deep learning network gradient descent algorithm. The recommended value is between [0.01, 0.001];

[0049] (8) Setting the cost function of the deep learning network to the square error cost function;

[0050] (9) Set the number of training times TrainN, with AC T , GR T 、Den T , R T As training input samples, digital lithology interpretations of representative wells are used as training output samples, and the deep learning network is trained using the training samples to obtain the deep learning network parameters;

[0051] (10) The obtained deep learning network parameters are used to test the training samples and test samples. If they meet the prediction, the trained network parameters can be used for application and promotion in other wells. If they do not meet the expectations, they are improved from three aspects: sample representativeness, network structure, and number of trainings. In this way, the generalizable deep learning network parameters can be obtained.

[0052] By using the method of this application to carry out research on dry layers and effective reservoirs, the inherent laws for distinguishing dry layers and effective reservoirs can be automatically discovered from the data, the interpretation accuracy and efficiency of dry layers and effective reservoirs in complex reservoir areas can be improved, and practical applications can be better guided, the actual exploration risks can be reduced, and the exploration success rate can be improved.

[0053] The earthquake forward modeling cases based on this application method are as follows:

[0054] (1) Figure 3 and Figure 4 This is a comprehensive cross-section of representative wells S1 and S2 in a certain study area of ​​S Oilfield;

[0055] (2) According to conventional methods, the lithology of dry layers and effective reservoirs of Wells S1 and S2 is interpreted, such as Figure 5 Shown is a comprehensive cross-section of Well S1 with superimposed lithological interpretations of dry layers and effective reservoirs (including oil layers, water layers, oil-water layers, etc.) (at the right layer stratification);

[0056] (3) Based on the lithology interpretation results, different lithologies of representative wells are assigned values, where the effective reservoir is assigned a value of 1, the dry layer is assigned a value of 0.5, and the mudstone is assigned a value of 0. Figure 6 Shown and Figure 7 Composite cross-section with digital lithological interpretation shown;

[0057] (4) Acoustic wave data AC, gamma data GR, density data Den, and resistance data R of the wells in the study area 25 After analysis, it is determined that the sound wave data range is [150,450] (unit: us / m), the gamma data range is [20,150] (unit: Api), and the density data range is [2.0,2.82] (unit: g / cm 3 ), the resistance data range is [0,60] (unit: Ω);

[0058] (5) According to the data value range in the fourth step above, the acoustic data AC, gamma data GR, density data Den, and resistance data R of the representative well are converted into 25 Normalize, and set the normalized data to be AC T , GR T 、Den T , R T ;

[0059] (6) Use the TensorFlow framework to build a deep learning network structure with an input layer of 4 input neurons, 1 bias neuron, 3 hidden layers, and an output layer of 1 neuron;

[0060] (7) Set the learning rate LR of the deep learning network gradient descent algorithm to 0.01;

[0061] (8) Setting the cost function of the deep learning network to the square error cost function;

[0062] (9) Set the number of training times to 10,000 to train sample AC T , GR T、Den T , R T Based on this, the deep learning network is trained to obtain the deep learning network parameters;

[0063] (10) Use the obtained deep learning network parameters to test the training samples and test samples, such as Figure 8 and Fig. 9 As shown in the figure, the gray solid line on the far right is the dry layer and effective reservoir of the original interpretation, and the black dotted line is the dry layer and effective reservoir predicted by deep learning. By comparing the predicted values ​​with the original interpretation, it can be found that most of the deep learning results are consistent with the original results, indicating that the deep learning network parameters are effective. It is extended to the S3 well and the following is obtained: Fig.10 The results shown are from Fig.10 It is easy to see which depth sections are dry layers and which depth sections are effective reservoirs.

[0064] The deep learning identification method for dry layers and effective reservoirs based on four parameters in this application provides a more convenient and effective measure to distinguish dry layers and effective reservoirs from well logging. Compared with the existing methods with poor adaptability to complex reservoirs and inefficient interpretation, it has great advantages, which is conducive to enriching the interpretation methods of dry layers and effective reservoirs in my country's sandstone and mudstone interbedded areas, promoting the development of interpretation technology in my country's sandstone and mudstone interbedded areas, reducing actual exploration risks, and improving the success rate of exploration.

[0065] The present application also provides a deep learning identification device for dry layers and effective reservoirs based on four parameters, such as Fig.11 As shown, it includes: an interpretation unit 10, which finds one or more representative wells from the study area and interprets the stratified sections of the study target layer; a digitization unit 11, which is used to digitize the lithology interpretation results of the target layer; an analysis unit 12, which analyzes the logging data of the wells in the study area and normalizes the logging data of the representative wells; a construction unit 13, which is used to use the TensorFlow framework to construct a deep learning network structure with an input layer, a hidden layer, and a neuron output layer having input neurons and bias neurons; a setting unit 14, which is used to set the learning rate of the deep learning network gradient descent algorithm; set the cost function of the deep learning network to be a square error cost function; set the number of training times, train the deep learning network based on the training samples, and obtain the deep learning network parameters; a testing unit 15, which is used to test the training samples and the test samples using the obtained deep learning network parameters.

[0066] In the present application, an embodiment of a deep learning automatic identification device for dry layers and effective reservoirs based on four parameters is basically similar to an embodiment of a deep learning identification method for dry layers and effective reservoirs based on four parameters. For relevant matters, please refer to the introduction of the embodiment of a deep learning identification method for dry layers and effective reservoirs based on four parameters.

[0067] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned deep learning identification method steps of dry layers and effective reservoirs based on four parameters are implemented. The computer-readable storage medium may include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A deep learning identification method for dry layers and effective reservoirs based on four parameters, characterized in that: include: S1, find one or more representative wells from the study area and interpret the target layer segment; S2, digitize the lithology interpretation results of the target layer; S3, analyzing the logging data of the wells in the study area and normalizing the logging data of representative wells; S4, using the TensorFlow framework to build a deep learning network structure with input neurons, bias neurons, input layer, hidden layer, and neuron output layer; S5, setting the learning rate of the deep learning network gradient descent algorithm; setting the cost function of the deep learning network to be the square error cost function; setting the number of training times, training the deep learning network based on the training samples, and obtaining the deep learning network parameters; S6, using the obtained deep learning network parameters to test the training samples and test samples.

2. The deep learning identification method of dry layer and effective reservoir based on four parameters according to claim 1 is characterized in that: Step S1 includes interpreting the dry layer, effective reservoir and mudstone of the target layer segment.

3. The deep learning identification method of dry layer and effective reservoir based on four parameters according to claim 1 is characterized in that: In step S3, the logging data includes acoustic wave data, gamma data, density data, and resistance data. The AC range of the acoustic wave data is determined to be [AC min ,AC max ], gamma data GR range is [GR min ,GR max ]、Den range of density data Den is [Den min ,Den max ]、Resistance data R 25 The range is [R min ,R max ].

4. The deep learning identification method for dry layers and effective reservoirs based on four parameters according to any one of claims 1 to 3, characterized in that: In step S3, the normalized data is AC T , GR T 、Den T , R T .

5. The deep learning identification method for dry layers and effective reservoirs based on four parameters according to claim 4 is characterized in that: In step S5, AC T , GR T 、Den T , R T As training input samples, digital lithology interpretations of representative wells are taken as training output samples. The deep learning network is trained using the training samples to obtain the deep learning network parameters.

6. The deep learning identification method for dry layers and effective reservoirs based on four parameters according to any one of claims 1 to 3, characterized in that: In step S6, if the test is consistent with the prediction, the trained network parameters are used for application and promotion in other wells; if it does not meet expectations, it is improved in terms of sample representativeness, network structure, and number of training times.

7. A deep learning identification device for dry layers and effective reservoirs based on four parameters, characterized in that: include: Interpretation unit: find one or more representative wells from the study area and interpret the target layer segment; A digitization unit is used to digitize the lithology interpretation results of the target layer; The analysis unit analyzes the logging data of the wells in the study area and normalizes the logging data of the representative wells; A construction unit is used to construct a deep learning network structure with an input layer of input neurons and bias neurons, a hidden layer, and a neuron output layer using the TensorFlow framework; Setting unit, used to set the learning rate of the deep learning network gradient descent algorithm; Set the cost function of the deep learning network to the square error cost function; set the number of training times, train the deep learning network based on the training samples, and obtain the deep learning network parameters; The testing unit is used to test the training samples and the testing samples by using the obtained deep learning network parameters.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method described in any one of claims 1 to 6 are implemented.

Images