A method for predicting fracture parameters based on waveform classification

Abstract

The application discloses a method for predicting fracture parameters based on waveform classification, which comprises the following steps: S1, calculating and data reconstructing the "central angle-attribute" of each data point of each central angle; S2, performing curve fitting on the "central angle-attribute" data pair in a two-dimensional coordinate axis, and taking the fitting curve as the fracture response waveform of the data point; S3, designing the classification number of the fracture response waveform, the model trace, and performing waveform classification on the plane, and determining the fracture strength and azimuth angle value of each type of model trace. The application is a technology for classifying and predicting the development conditions of different types of fractures by using the fracture response waveform of the "central angle-attribute" data pair of the data point, and can accurately and effectively predict the distribution conditions of different types of fractures.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of seismic exploration, belongs to the technical field of seismic data interpretation, and is a method for predicting fracture development parameters by classifying waveforms of fracture response waveforms of a "central angle-attribute" data pair of each data point (CDP point). BACKGROUND

[0002] Fractures are important channels for underground oil and gas accumulation and migration, and in most cases, fractures are mainly tectonic fractures: fractures due to or associated with local tectonic events, including fault-related fracture systems, uplift-related fracture systems, and fold-related fracture systems. The tight sandstone gas reservoirs of the continental facies Xujiahe Formation in the Sichuan Basin are closely related to fracture development. Large fractures cause damage to sandstone reservoirs, while medium and small fractures play a positive role in post-fracturing fracture formation of sandstone reservoirs and communication of gas-bearing reservoirs. Therefore, different types of fractures have different effects on tight sandstone reservoirs.

[0003] At present, there are only two methods for predicting underground fractures using seismic data, both of which are related to prestack and poststack seismic data. In addition, some geologic experience analysis techniques such as finite element analysis and tectonic stress field analysis are also used to predict fractures. One of them uses azimuthal seismic stacking data and is guided by azimuthal anisotropy theory. By determining an ellipse with a center coordinate (0, 0) and an arbitrary angle θ with the X axis, three parameters need to be determined: the long axis radius a, the short axis radius b, and θ. Therefore, at least three known point coordinates are needed to determine an ellipse. By calculating the ratio of the long axis and the short axis of the fitted ellipse, the fracture strength and direction are calculated. Some geophysical workers have also achieved relatively more results in predicting fracture reservoirs using seismic reflection waveform classification. The core algorithm "self-organizing neural network" in waveform classification is a neural network with learning function. It actually reflects the topological relationship of each mode in the signal space on the network output almost unchanged, mainly composed of input and output. The input is one-dimensional, and the output can be multi-dimensional, and the output nodes are widely connected to other nodes in the neighborhood. Each input neuron and each output neuron has a feedforward connection. Through adaptive and self-organizing learning and continuous adjustment of weights, the neural network has the same output for a certain input when it is stable.

[0004] There are quite a few patent technical literatures for the prediction of different types of cracks, which shows that the detection of cracks has been a hot spot of research and exploration. Different types of cracks may have different seismic reflection waveforms, but some such as small cracks have similar seismic reflection waveforms, which are difficult to distinguish and identify from the seismic reflection waveform. Some patents and technical literatures such as the establishment of crack distribution model and its seismic response proposed by Fan Guozhang and Mou Yongguang (Fan Guozhang and Mou Yongguang published "Changes of P-wave velocity in anisotropic medium with cracks and its influence on common-midpoint gather stacking" in Petroleum Exploration in 2002; the patent of "A method and device for seismic waveform analysis and reservoir prediction" (Patent No. 201210135544.X) proposes to establish multiple model traces using the seismic reflection waveform of the target layer, classify different waveforms according to the self-organizing neural network, generate a seismic facies classification map, and further form a sedimentary facies map for reservoir prediction; the patent of "A crack prediction method and device" (Patent No. 201010205983.4) proposes to obtain the reflection amplitude of each seismic trace by picking up the target layer time window, and to perform elliptical fitting by obtaining the azimuth and reflection amplitude to predict the direction and density of cracks. However, these methods are quite difficult to implement for crack prediction, and are easily affected by various factors during operation, mainly having the following problems:

[0005] (1) The prediction results of conventional crack prediction technology for different types of cracks are often not consistent with well data, and the prediction difficulty is relatively large.

[0006] (2) Waveform classification technology is usually applied to seismic facies identification, and it is relatively difficult to further distinguish different types of crack seismic reflection waveforms using this technology. SUMMARY

[0007] In order to overcome the shortcomings of the above conventional technical methods and solve the problem of crack strength prediction and evaluation, the present application provides a corresponding technical process to solve the prediction, which can conveniently, simply and accurately predict the crack strength of the target layer using existing P-wave 3D seismic acquisition data. The present application aims to overcome the above and other shortcomings of the existing conventional crack prediction technology, and fully utilize the azimuth information of the pre-stack gather which is relatively sensitive to crack response. Therefore, the present application provides a method for classifying and predicting the development parameters of different types of cracks based on the "central angle-attribute" data pair of crack response waveform of pre-stack data processing. SUMMARY:

[0009] S1, reconstruct the data of each data point in each central angle seismic attribute data body and extract the attribute data of the target layer to form a "central angle-attribute" data pair of each data point.

[0010] S1-1, after the CDP gather data is corrected for moveout, the corresponding azimuth range is divided using appropriate azimuth division parameters, and relevant central angle gather data is obtained. The median of the azimuth range in actual seismic data is the central angle, and the central angle direction represents the response direction of the seismic data in the azimuth range, i.e., the central angle gather data is obtained, and the maximum offset distance in the central angle gather data should not be greater than 30° for the incident angle of the target layer. The azimuth of the seismic data is set to 0° in the north direction, and is rotated clockwise to 360°. In general, the number of divided central angles should be greater than or equal to five in principle, and the division of the azimuth range requires that the stacking times of each central angle data after division should be substantially uniform, so that the extracted attributes and attribute fitting curves have relatively high precision and are beneficial to the calculation of the next step.

[0011] S1-2, "well-seismic" calibration is performed using post-stack 3D seismic data to determine the top and bottom interfaces of the target layer. The top and bottom interfaces are artificially interpreted, interpolated and smoothed to obtain the interpreted horizon of the target layer and facilitate subsequent attribute extraction and calculation of the target layer.

[0012] S1-3, each central angle gather data is stacked and migrated to obtain corresponding central angle seismic data volumes, and after reconstruction and attribute extraction, a "central angle-attribute" data pair for each data point is formed. Specifically:

[0013] S1-3-1, according to the same data point (CDP point), each central angle seismic data trace is reconstructed to form the gather data of the data point (CDP point) after reconstruction.

[0014] S1-3-2, according to the horizon data of the interpreted target layer and the gather data of the reconstructed data point, seismic attribute extraction of the target layer is performed to obtain a "central angle-attribute" data pair for each data point in the target layer. The types of seismic attributes can include frequency, amplitude, double-layer travel time, velocity and other attribute data related to anisotropy calculation caused by fractures, and one of the attribute data is preferably selected for attribute extraction and calculation, which requires that the attribute can effectively highlight the response of the fracture, and the difference between the fracture and the non-fracture in the attribute data value is more obvious.

[0015] S2, curve fitting is performed on the "central angle-attribute" data pair of each data point to obtain the fracture response waveform of each data point.

[0016] S2-1, a two-dimensional coordinate system (θ, f) is established, where the horizontal coordinate is the angle value θ, and the vertical coordinate is the attribute value f.

[0017] S2-2, the "central angle (θ i ) - seismic attribute (fi ) data pairs into the coordinate system, and performs a least square binary curve fitting on the data pairs of the point to obtain a relevant fitting curve. The fitting curve is taken as the fracture response waveform (0i, fi, 0i e (0°, 180°)) of the target layer section of the data point, and enters the next step.

[0018] S2-3, and so on, to complete the fracture response waveform calculation of each data point.

[0019] S3, classifying the fracture response waveforms of each data point, and predicting the distribution range of different types of fractures.

[0020] S3-1, setting the number of fracture classifications and creating a plurality of fracture response waveform model traces, the number of which is the same as the number of fracture classifications. Generally, the number of fracture response waveform model traces is 3-5 times the number of sedimentary facies in the study area. In actual operation, the number of fracture classifications and model traces can be determined according to relevant waveform classification test conditions, expert experience, fracture prediction accuracy, etc.

[0021] S3-2, classifying the fracture response waveforms of different data points according to the self-organizing neural network to distinguish different types of fractures. The self-organizing neural network is trained on actual fracture response waveforms. After several iterations, the self-organizing neural network constructs a synthetic fracture response waveform, which is then compared with the actual fracture response waveform. Through adaptive experiments and error processing, the synthetic waveform is changed after each iteration to find a better correlation between the model trace and the actual fracture response waveform. Generally, the more classifications, the more model traces are generated. Generally, the larger the number of classifications, the higher the prediction accuracy of different types of fractures, and the longer the operation time. The smaller the number of classifications, the lower the prediction accuracy, and the shorter the operation time. The calculation steps of the self-organizing neural network are as follows:

[0022] Let the number of neurons be N, and the neuron full vector be:

[0023] The input mode is:

[0024] The implementation steps are:

[0025] 1. Initialize the weight vector Select a random value, select the initial learning rate a (0) and the neighborhood function ω m (0);

[0026] 2. Input the sample vector to the input layer;

[0027] 3. Select the weight vector that best matches neurons as the winning neuron, the winning neuron label m, satisfying

[0028]

[0029] IV. Training the weight vector so that the neurons in the active bubble range move toward the input vector direction:

[0030]

[0031] V. Updating the learning rate a(t);

[0032] VI. Reducing the neighborhood function: ω m (t); finally checking the end condition, and exiting when the weights no longer change significantly, otherwise continuing to step ②.

[0033] Through competitive learning, the weight vectors of the neighboring neurons approach the weight vector of the winning neighborhood neuron, and the weight vector of the winning neuron is closest to a certain learning sample vector. Therefore, if there is a significant classification clustering relationship in the learning sample vector, the distribution of the neuron weight vectors in the neural network through competitive learning presents a zoning phenomenon, the neuron weight vectors in the same zone are close to each other and close to a certain class of sample vectors, thereby achieving the purpose of classification.

[0034] S3-3, according to the fracture response waveform model trace, the fracture intensity and azimuth value are calculated, so as to obtain the fracture intensity and direction of each fracture classification. Specifically, the maximum and minimum attribute values in the fracture response waveform model trace are determined, and after calculation by the correlation fracture intensity calculation formula, the data value obtained is taken as the fracture intensity value of the fracture response waveform model trace. In addition, the azimuth angle corresponding to the maximum attribute value or plus or minus 90° can be taken as the fracture azimuth of the model. Whether to add or subtract 90° depends on the nature of the attribute to the fracture response and expert experience. In principle, the fracture azimuth can be understood as the horizontal angle between the north direction line from a point on the fracture and the fracture strike line in the clockwise direction. The azimuth and relative intensity of the fracture strike predicted on the design of the seismic rose are symmetrically distributed at 180°, and the azimuth value is in the range of 0°-180°. Among them, the fracture intensity calculation formula is as follows:

[0035] P i = |P k | / |P l | (1)

[0036] In formula (1), P i is the fracture intensity value of the model trace, P k is the maximum attribute value of the model trace, and P l is the minimum attribute value of the model trace.

[0037] If there are well data in some model traces, the fracture intensity and azimuth of the relevant model trace can be determined, and the fracture intensity and azimuth of the well can be used as the fracture intensity and direction of each data point in the relevant model trace; the fracture intensity and azimuth of the relevant model trace can be calculated by using formula (1) for other model traces without well data. In addition, the fracture intensity of each model trace can be calculated by fitting the fracture intensity of the well and the fracture intensity value of the corresponding data point calculated by using the fracture intensity calculation method of the present application (this method requires two or more well data for fitting calculation). Another method for calculating the fracture intensity and azimuth of the model trace is to calculate the fracture intensity and azimuth of a data point in the model trace by using elliptical fitting, and the obtained fracture parameters are used as the fracture intensity and direction of the model trace, and the fracture intensity and azimuth of each model trace are calculated in this way, so that the fracture intensity and direction of the waveform classified plane can be interpreted.

[0038] S4, the display method of the fracture waveform classification result is to obtain a waveform classification plane, three display markers are arranged in the right blank part of the graph, one display marker is the color of the classified model trace, the second display marker is the fracture intensity corresponding to the classified color, and the third display marker is the fracture azimuth value corresponding to the classified color.

[0039] Preferably, the present application also provides an electronic device, comprising: a memory storing executable instructions; a processor running the executable instructions in the memory to implement the method for predicting fractures based on curve waveform classification.

[0040] Preferably, the present application also provides a computer readable storage medium storing a computer program, which is executed by a processor to implement the method for predicting fractures based on curve waveform classification.

[0041] Advantages of the present application

[0042] The present application is a kind of technology for predicting different types of fracture development by using the "central angle-attribute" data pair of the fracture response waveform of data points, which can accurately and effectively predict the distribution of different types of fractures. By extracting the "central angle-attribute" data pair of each data point in the target layer, a two-dimensional coordinate system is established to obtain the relevant attribute fitting curve; the fitting curve is used as the fracture response waveform of the point; secondly, a reasonable number of classifications is designed to classify the fracture response waveform; and the fracture strength and azimuth angle of the model trace are calculated using the relevant calculation formula; the relevant data is displayed on the plan view, and the planar distribution of different types of fracture development parameters is obtained, which can greatly improve the economic benefit of fracture reservoir exploration.

[0043] The present application is a kind of technology for predicting different types of fracture development by using the "central angle-attribute" data pair of the fracture response waveform of data points, which can accurately and effectively predict the distribution of different types of fractures. By extracting the "central angle-attribute" data pair of each data point in the target layer, a two-dimensional coordinate system is established to obtain the relevant attribute fitting curve; the fitting curve is used as the fracture response waveform of the point; secondly, a reasonable number of classifications is designed to classify the fracture response waveform; and the fracture strength and azimuth angle of the model trace are calculated using the relevant calculation formula; the relevant data is displayed on the plan view, and the planar distribution of different types of fracture development parameters is obtained, which can greatly improve the economic benefit of fracture reservoir exploration. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The present application is a kind of technology for predicting different types of fracture development by using the "central angle-attribute" data pair of the fracture response waveform of data points, which can accurately and effectively predict the distribution of different types of fractures. By extracting the "central angle-attribute" data pair of each data point in the target layer, a two-dimensional coordinate system is established to obtain the relevant attribute fitting curve; the fitting curve is used as the fracture response waveform of the point; secondly, a reasonable number of classifications is designed to classify the fracture response waveform; and the fracture strength and azimuth angle of the model trace are calculated using the relevant calculation formula; the relevant data is displayed on the plan view, and the planar distribution of different types of fracture development parameters is obtained, which can greatly improve the economic benefit of fracture reservoir exploration.

[0045] Figure 2 The present application is a kind of technology for predicting different types of fracture development by using the "central angle-attribute" data pair of the fracture response waveform of data points, which can accurately and effectively predict the distribution of different types of fractures. By extracting the "central angle-attribute" data pair of each data point in the target layer, a two-dimensional coordinate system is established to obtain the relevant attribute fitting curve; the fitting curve is used as the fracture response waveform of the point; secondly, a reasonable number of classifications is designed to classify the fracture response waveform; and the fracture strength and azimuth angle of the model trace are calculated using the relevant calculation formula; the relevant data is displayed on the plan view, and the planar distribution of different types of fracture development parameters is obtained, which can greatly improve the economic benefit of fracture reservoir exploration. DETAILED DESCRIPTION

[0046] The present application is a kind of technology for predicting different types of fracture development by using the "central angle-attribute" data pair of the fracture response waveform of data points, which can accurately and effectively predict the distribution of different types of fractures. By extracting the "central angle-attribute" data pair of each data point in the target layer, a two-dimensional coordinate system is established to obtain the relevant attribute fitting curve; the fitting curve is used as the fracture response waveform of the point; secondly, a reasonable number of classifications is designed to classify the fracture response waveform; and the fracture strength and azimuth angle of the model trace are calculated using the relevant calculation formula; the relevant data is displayed on the plan view, and the planar distribution of different types of fracture development parameters is obtained, which can greatly improve the economic benefit of fracture reservoir exploration.

[0047] Example 1

[0048] According to the present application process ( Figure 1 ), the working steps are formulated, and the example is to predict the distribution of different types of fracture reservoirs in a marine shale reservoir section of a certain three-dimensional work area.

[0049] In step S1, the seismic trace data is divided into azimuth range stacks and migration processing ( Figure 2), the central angle is 18°, 54°, 90°, 126°, 160°, etc. the total of five central angle seismic data volume - azimuth angle range is 0°-36°, 36°-72°, 72°-108°, 108°-144°, 144°-180°, according to the design of the target layer depth of 3400m, the superposition of shot range is limited to 380m-3200m, so that the maximum incidence angle of the target layer is limited within 30°, respectively, each central angle stack data for migration processing, the reflection wave group home.

[0050] Using logging data, three-dimensional post-stack seismic data for "well-seismic" calibration, determine the target layer, and the top and bottom of the target layer interface interpretation. In the example, shale target layer top and bottom interface reflection is selected as the wave trough, wave peak, and the two horizon tracking interpretation. Horizon interpretation parameters from 1280 line and 200 channels to 3100 line and 1600 channels, according to the actual situation of seismic data interpretation grid is set to 10 lines X10 channels, after interpretation, the horizon interpolation, smoothing processing, interpolation to 1 line X1 channel.

[0051] In the example, according to the well data and expert experience, the starting attenuation frequency is selected as the attribute data. After data reconstruction of five central angle seismic data at the same data point, the starting attenuation frequency attribute value of the target layer is extracted. Form the "central angle-attribute" data pair at each data point (CDP point).

[0052] In step S2, a two-dimensional coordinate system is established, the horizontal coordinate is the angle value, and the vertical coordinate is the starting attenuation frequency attribute value. The five "central angle-attribute" data pairs of a single data point are projected into the coordinate system, and the least square fitting is performed on the five attribute data pairs to obtain the related fitting curve. The fitting curve is taken as the fracture response waveform of the data point. In this way, the fracture response waveform calculation of each data point is completed.

[0053] In step S3, the number of categories is designed and the self-organizing neural network is used to classify the crack response waveforms. According to the accuracy to be predicted and the experience of experts, the number of crack categories in the study area is set to 22, that is, different crack types in the study area can be divided into 22 model traces - different crack types have different crack development directions and crack strengths, and the crack strength and direction are calculated according to the relevant data of the model trace. In the example, the color display of different types of crack response model traces is set, that is, different types of cracks are set to different color displays, and the crack direction is calculated because the attribute is the initial decay frequency, so the "central angle - attribute" maximum value is perpendicular to the crack direction - the crack direction can be obtained by adding or subtracting 90 degrees from the corresponding central angle value. According to the crack waveform distribution plan calculated by the present application, the well data of the subsequent drilling in the study area shows that the prediction result of the present application technology is consistent with the crack development direction and strength in the subsequent drilling, and the coincidence rate is relatively better than the result of the conventional crack prediction method, which proves that the present application technology is effective.

[0054] Embodiment 2

[0055] The electronic device provided by the present application comprises a memory storing executable instructions, and a processor running the executable instructions in the memory to realize the crack prediction method based on curve waveform classification.

[0056] The electronic device according to the embodiments of the present application comprises a memory and a processor.

[0057] The memory is used to store non-transitory computer readable instructions. Specifically, the memory can include one or more computer program products, which can include various forms of computer readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM), cache memory, and the like. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, and the like.

[0058] The processor can be a central processing unit (CPU) or other forms of processing units with data processing and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions. In an embodiment of the present application, the processor is used to run the computer readable instructions stored in the memory.

[0059] Those skilled in the art should understand that, in order to solve the technical problem of how to obtain a good user experience effect, the present embodiment can also include well-known structures such as communication bus, interface, etc., which should also be included in the protection scope of the present application.

[0060] Detailed description of the present embodiment can refer to the corresponding description in the foregoing embodiments, which will not be repeated here.

[0061] Embodiment 3

[0062] The computer readable storage medium of the present embodiment stores a computer program, and the computer program is executed by a processor to implement the method for predicting cracks based on curve waveform classification.

[0063] The computer readable storage medium of the present embodiment stores a computer program, and the computer program is executed by a processor to implement the method for predicting cracks based on curve waveform classification.

[0064] The computer readable storage medium includes, but is not limited to, optical storage media (for example, CD-ROM and DVD), magneto-optical storage media (for example, MO), magnetic storage media (for example, magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (for example, memory card), and media with built-in ROM (for example, ROM cartridge).

[0065] Those skilled in the art should understand that the above description of the embodiments of the present application is only for the purpose of exemplarily illustrating the beneficial effects of the embodiments of the present application, and is not intended to limit the embodiments of the present application to any examples given.

[0066] The above has described the embodiments of the present application, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. A method for predicting fracture parameters based on waveform classification, characterized in that, It comprises the following steps: S1, calculating and data reconstruction processing of "central angle-attribute" of each data point of each central angle; the specific steps of S1 are: S1-1, dividing the corresponding azimuth range by using appropriate azimuth division parameters on the CDP gather data after the moveout correction to obtain the relevant central angle gather data; the median of the azimuth range in the actual seismic data is the central angle, and the central angle direction represents the response direction of the seismic data of the azimuth range, that is, the central angle gather data is obtained, and the maximum offset distance in the central angle gather data is required to be not greater than 30° for the incident angle of the target layer; the azimuth of the seismic data is set as 0° in the north direction, and the clockwise rotation is 360°; S1-2, using the post-stack three-dimensional seismic data to perform "well-seismic" calibration to determine the top and bottom interfaces of the target layer section; and performing artificial interpretation, interpolation and smoothing processing on the top and bottom interfaces to obtain the interpreted horizon of the relevant target layer section and facilitate the subsequent attribute extraction calculation of the target layer section; S1-3, stacking and migration processing on each central angle gather data to obtain the corresponding central angle seismic data volume, and forming the "central angle-attribute" data pair of each data point after reconstruction and attribute extraction; S2, curve fitting of the "central angle-attribute" data pair in the two-dimensional coordinate system, and the fitting curve is taken as the fracture response waveform of the data point; S3, designing the number of fracture response waveform classifications, model traces, and classifying the waveforms in the plane, and determining the fracture intensity and azimuth value of each type of model trace.

2. The method of claim 1, wherein, It further comprises the step of S4, displaying the waveform classification results on the plane, setting three display markers in the blank part of the graph, the first display marker is the classification model trace color, the second display marker is the fracture intensity corresponding to the classification color, and the third display marker is the fracture azimuth value corresponding to the classification color.

3. The method of claim 1, wherein, In S1-3, the specific steps are: S1-3-1, reconstructing the central angle seismic data traces of the same data point to form the gather data of the data point after reconstruction; S1-3-2, performing seismic attribute extraction of the target layer section according to the horizon data of the interpreted target layer section and the gather data of the data point after reconstruction to obtain the "central angle-attribute" data pair of each data point of the target layer section; the types of attributes include frequency, amplitude, double-layer travel time, and velocity attribute data related to anisotropy calculation caused by fractures, and one of the attribute data is selected for attribute extraction calculation.

4. The method of claim 1, wherein, The specific steps of S2 are: S2-1, establishing a two-dimensional coordinate system (θ, f), the horizontal coordinate of which is the angle value θ, and the vertical coordinate is the attribute value f; S2-2, project the "central angle - attribute" data pair (0 i -f i ) of the single data point i into the coordinate system, and perform a least squares binary curve fitting on the data pairs of the point to obtain a related fitting curve; the fitting curve serves as the fracture response waveform of the target interval of the data point; S2-3, repeating step S2-2 to complete the fracture response waveform calculation of each data point.

5. The method of claim 1, wherein, The specific steps of S3 are: S3-1, setting the number of fracture classifications and creating multiple fracture response waveform model traces, the number of which is the same as the number of fracture classifications; S3-2, classifying the fracture response waveforms of different data points according to the self-organizing neural network to distinguish different types of fractures; S3-3, fracture intensity and azimuth value are calculated according to the fracture response waveform model trace, so as to obtain the fracture intensity and direction of each fracture classification.

6. The method of claim 5, wherein, In S3-2, the calculation steps of the self-organizing neural network are as follows: Let the number of neurons be N, and the neuron vector be: The input mode is: The implementation steps are as follows: ① Initialize the weight vector The random value is selected, and the initialization learning rate a(0) and the initialization neighborhood function ω m (0) are selected. (ii) inputting the sample vector to the input layer; iii. Select the neuron whose weight vector best matches the input vector as the winning neuron, with winning neuron index m satisfying ​ (4) training the weight vector, so that the neurons in the active bubble range move to the direction of the input vector: (5) updating the learning rate a(t); (6) Reduction of neighborhood function: ω m (t); final check end condition, exit when the weights no longer change significantly, otherwise continue to step 2.

7. The method of claim 5, wherein, In S3-3, the calculation of fracture intensity and azimuth value is obtained by one of the following schemes: (1) fracture intensity and azimuth are directly obtained from well data of model trace; (2) fracture intensity is obtained by formula (1): P i = |P k | |P l | (1) where P i is the fracture intensity value for the model trace, P k is the maximum attribute value for the model trace, P l is the minimum attribute value for the model trace; Azimuth acquisition: the azimuth corresponding to the maximum value of the attribute is added or subtracted by 90° as the fracture azimuth of the model; whether to add or subtract 90° depends on the nature of the attribute response to the fracture and expert experience; (3) fracture intensity and azimuth value are obtained by fitting well data; (4) the fracture intensity and azimuth value of a certain data point in the model trace are calculated by using elliptical fitting.

8. An electronic device, the electronic device comprising: A memory storing executable instructions; The processor runs the executable instructions in the memory to implement the method for predicting fracture parameters based on waveform classification according to any one of claims 1-7.

9. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising The computer program is executed by the processor to implement the method for predicting fracture parameters based on waveform classification according to any one of claims 1-7.

Images