Method for determining river channel width based on deep learning

By generating a river channel model and using transfer learning and convolutional neural networks, the problems of low efficiency and insufficient accuracy in existing technologies for river channel width identification are solved, and efficient and convenient automatic river channel width identification is achieved.

CN119006562APending Publication Date: 2024-11-22CHINA PETROLEUM & CHEMICAL CORP +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310568604.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-18
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing technologies are inefficient and highly subjective in identifying river widths. Quantitative analysis methods have weak spatial resolution capabilities and are difficult to apply. Furthermore, the application of deep learning in river identification has not effectively solved these problems.

Method used

By using a deep learning-based approach, multiple river channel models are generated. Convolutional neural networks are then used to transfer and train the samples to identify the river channel model that best matches the conditional data. The width of the model is then used as the most suitable value. By combining transfer learning and the powerful image feature learning capabilities of convolutional neural networks, the automatic identification of river channel width can be achieved.

Benefits of technology

It improves the efficiency and accuracy of river width identification, with high identification efficiency, simple application, and high identification accuracy, and is suitable for automatic identification of river width.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006562A_ABST
    Figure CN119006562A_ABST
Patent Text Reader

Abstract

The invention provides a method for determining the width of a river channel based on deep learning. The method for determining the width of the river channel based on deep learning comprises the following steps: step 1, inputting condition data of an actual work area; 2, designing width parameters of different river channels; 3, generating a plurality of river channel models with different widths according to different river channel width parameter combinations in the step 2; 4, training a neural network for multiple random samples of the candidate random river channel model based on deep learning; and 5, identifying the river channel model which is most matched with the condition data from the plurality of river channel models, and taking the river channel width of the model as the most appropriate value. The method for determining the river channel width based on deep learning is high in recognition efficiency, simple and convenient to apply, high in recognition precision and capable of effectively determining the river channel width value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of sedimentary phase research, and in particular to a method for determining river channel width based on deep learning. Background Art

[0002] At present, there are not many methods to identify the width of sedimentary channels based on drilling data. In the early days, geological staff basically analyzed the width and distribution characteristics of the channel through visual observation based on sedimentological knowledge and related data. This method is inefficient and highly subjective. Quantitative analysis methods include methods for determining the width of underwater distributary channels based on multi-point geostatistics. This method uses multi-point geostatistics, replaces the variogram with a "training image", and uses different channel widths to make "training images". The underwater distributary channel is described in detail, including the planar morphology of the channel and the probability statistics of the sedimentary microfacies combination encountered by different well networks, and then the training image corresponding to the drilling data is identified. The channel width of the training image is used as the channel width value of the study area. The spatial resolution comparison ability of this evaluation index is weak, and it involves multiple parameter settings, which makes it difficult to apply.

[0003] Deep learning is one of the current hot technologies in artificial intelligence and machine learning. It has been widely studied and applied in various fields. With the powerful learning and cognitive capabilities of deep artificial neural networks, deep learning is particularly suitable for complex tasks with strong nonlinear characteristics, such as personalized recommendations, image recognition and classification, and machine translation.

[0004] In the Chinese patent application with application number: CN202111212259.9, a method for generating high-precision two-dimensional topography of a one-way river channel based on deep learning is involved. According to other existing river channel measurement data and target river channel images, upstream and downstream cross-sectional data, the existing river channel measurement data is preprocessed according to the characteristics of the target river channel based on deep learning, and an image set of existing river channel data, a river channel data set and a one-to-one mapping relationship between the two are established. A method for river channel quadrant classification is proposed, and a convolutional neural network under deep learning is used to classify the existing river channel image set. The river channel shoreline image is trained using key point detection CPM, and the trained model is used to identify the key points of the target river channel, and several alternative river channel sets with similar morphology to the target river channel are matched. Finally, the two most similar rivers are matched in cross-sectional morphology in the alternative river channel set, and the terrain scatter point data corresponding to the two rivers are scaled and reconstructed using a similar ratio. After smoothing, the high-precision two-dimensional terrain scatter point data of the target river channel is obtained.

[0005] In the Chinese patent application with application number: CN201410844281.9, a method for identifying river channel sandstone reservoirs is involved, including: obtaining the seismic response characteristics of the river channel sand body by forward modeling the reservoir, determining the seismic identification marks of the river channel sand body of different geological bodies; extracting the corresponding seismic attributes according to the seismic identification marks of the river channel sand body, and determining the river channel boundary; determining the sedimentary period of the river channel sand by using the stratigraphic slicing technology under the isochronous stratigraphic framework; determining the thickness of the river channel sand body according to the statistical fitting relationship between the actual drilling sand body thickness and the seismic attributes. According to this method, in-depth seismic attribute extraction and analysis, using the reservoir of known wells, oil and gas layer calibration results, drilling statistical analysis, and combined with the model forward analysis results, a river channel sand body identification and description technology was established. It can accurately identify and describe river channel sandstone reservoirs with ultra-deep layers, thin sand layers, and narrow channels, and is less affected by seismic resolution.

[0006] In the Chinese patent application with application number: CN202010022741.5, a channel sand modeling method, device and readable storage medium are involved. The channel sand modeling method includes the following steps: obtaining the basic parameters of the reservoir to be modeled; generating a number of first channel sand body models and training condition data corresponding to the first channel sand body models according to the obtained basic parameters; establishing a generation network; establishing a discriminant network; training the discriminant network and the generation network to obtain a generator; inputting the well data into the generator to obtain the sand body model of the reservoir to be modeled. The beneficial effect of the invention is: a large number of first channel sand body models and corresponding training condition data are generated by a target-based simulation method, and then the generation network and the discriminant network are trained by these first channel sand body models and training condition data. After a large amount of training, the generation network obtains a generator that can generate a real channel sand body model, and then the well data of the study area is input into the generator to obtain the sand body model of the reservoir to be modeled in the study area.

[0007] The above existing technologies are significantly different from the present invention and fail to solve the technical problem we want to solve. For this reason, we have invented a new method for determining the width of a river channel based on deep learning. Summary of the invention

[0008] The purpose of the present invention is to provide a method for determining river width based on deep learning with high recognition efficiency, simple application and high recognition accuracy.

[0009] The object of the present invention can be achieved by the following technical measures: a method for determining the width of a river channel based on deep learning, the method for determining the width of a river channel based on deep learning includes:

[0010] The purpose of the present invention can also be achieved by the following technical measures:

[0011] In step 1, the condition data C of the actual work area is input, where C is a set of k points.

[0012] In step 1, the scope of the study area is defined according to the distribution characteristics of the drilling data, and the study area must include all well data.

[0013] In step 2, different channel width parameters are designed, including channel width, aspect ratio, and main direction.

[0014] In step 3, based on the fluvsim algorithm, M river channel models with different widths are generated according to different parameter combinations in step 2. m Represents the width of the mth river channel model, where m=1,…,M.

[0015] In step 4, we first define the number of samples from each river model as n, the training data set PS for convolutional deep learning, and then randomly select k points from the mth river model to obtain the point set P m (i) to P m (i) Add the text tag {m} and change P m (i) Add to PS, and finally use CNN to train PS based on transfer learning to obtain the trained model CNN PS .

[0016] Step 4 includes:

[0017] (4.1) Define the training data set PS, which is used to store all random sampling data of all river reference models;

[0018] (4.2) Sample n times from the m-th river width reference model, and each time extract k points from the m-th reference model to obtain the point set P m (i);

[0019] (4.3) Give P m (i) Add the text tag {m} and change P m (i) and the label {m} are added to PS;

[0020] (4.4) Based on transfer learning, CNN is used to train PS and obtain the trained model CNN PS .

[0021] In step 5, use the trained model CNN PS Detect condition data C, identify the river channel model that best matches C from M river channel models, and use the river channel width of this model as the most appropriate value.

[0022] The method for determining the width of a river channel based on deep learning also includes, after step 5, step 6, checking the accuracy of river channel width recognition.

[0023] In step 6, a random sample is taken from the virtual model of the narrow river channel as conditional data to identify the reference model of the equally narrow river channel, and the recognition accuracy of 100 tests is counted to obtain a statistical histogram of the river channel width recognition accuracy, thereby detecting the river channel width recognition accuracy.

[0024] The method for determining the width of a river channel based on deep learning in the present invention first performs multiple random sampling of reference models of different widths, then uses a convolutional neural network to perform migration training on the samples, and finally uses the trained neural network model to select the reference model that best matches the well condition data, and uses the river channel width of the reference model as the most reasonable river channel width value. Based on the powerful learning ability of deep convolutional neural networks in image features, the new method can quickly and accurately automatically identify the spatial feature differences between conditional data and reference model sample points. Compared with traditional optimization indicators, the present invention has high recognition efficiency, simple application, and high recognition accuracy. The case proves that the new method can effectively determine the value of the river channel width. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the river channel width identification method of the river phase study area based on the convolutional neural network of the present invention;

[0026] Figure 2 It is a schematic diagram of three fluvs im river phase models with different river channel widths and corresponding well condition data in a specific embodiment of the present invention;

[0027] Figure 3 It is a schematic diagram of the recognition rate results of three river phase well data with different river channel widths in a specific embodiment of the present invention;

[0028] Figure 4 This is a flowchart of a specific embodiment of the method for determining river channel width based on deep learning of the present invention. DETAILED DESCRIPTION

[0029] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0030] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations and / or combinations thereof.

[0031] like Figure 4 As shown, Figure 4 This is a flow chart of a method for determining a river channel width based on deep learning according to the present invention. The method for determining a river channel width based on deep learning includes:

[0032] Step 101, inputting condition data of the actual work area;

[0033] Step 102, designing different channel width parameters, including channel width, width-to-thickness ratio, main direction, etc.;

[0034] Step 103, generating a plurality of river channel models with different widths according to different parameter combinations in step 102;

[0035] Step 104, training a neural network based on multiple random samples of the candidate random river channel model based on deep learning;

[0036] Step 105, using the trained model to detect the conditional data C, identifying the river channel model that best matches the conditional data from multiple river channel models, and taking the river channel width of the model as the most appropriate value.

[0037] The following are several specific embodiments of the present invention.

[0038] Example 1

[0039] In a specific embodiment 1 of the present invention, the method for determining the width of a river channel based on deep learning includes the following steps:

[0040] (1) Input the conditional data C of the actual work area, where C is a set of k points. Define the scope of the study area based on the distribution characteristics of the drilling data. Usually, the study area must include all well data;

[0041] (2) Design different channel width parameters, including channel width, width-to-thickness ratio, main direction, etc.;

[0042] (3) Based on the fluvs im algorithm, M river channel models with different widths are generated according to different parameter combinations in step (2), L m Represents the width of the mth river channel model, where m = 1,…,M

[0043] (4) Then, based on deep learning, the neural network is trained on multiple random samples of the candidate random river channel models. The specific training process is as follows: first, the number of samples from each river channel model is defined as n, the training data set PS of convolutional deep learning is defined, and then k points are randomly selected from the mth river channel model to obtain the point set P m (i) to P m (i) Add the text tag {m} and change P m (i) Add to PS, and finally use CNN to train PS based on transfer learning to obtain the trained model CNNPS .

[0044] (5) Use the trained model CNN PS Detect condition data C, identify the channel model that best matches C from M channel models, and use the channel width of the model as the most appropriate value. This enables the trained network model to identify the channel model that best matches the drilling data.

[0045] Example 2

[0046] In a specific embodiment 2 of the present invention, the essence of analyzing the width of the river channel based on the well condition data is to quantitatively evaluate the similarity of the spatial characteristics of discrete points and regular grids.

[0047] like Figure 1 The schematic diagram of the river channel width identification method for the river phase study area based on the convolutional neural network of the present invention is shown, wherein the well condition data C of the actual work area contains k data points, and the random models A and B are reference models of different river widths, and the river channel widths of the two models are significantly different. In order to find the model that best matches C from A and B, firstly, n samplings are made from each of the models A and B, and k points are randomly selected each time, and the sampling point sets are recorded as A_i and B_i, where i∈{1,…,n}. Then, labels are added to the sampling point sets A_i and B_i, and the discrete point set label from the model A is "A", and similarly, the discrete point label from the model B is "B". All the sampling point sets A_i and B_i are used as training sets, and the deep convolutional neural network Inception-Resnet-v2 is used for migration training. Finally, the trained convolutional network model is used to identify the model A or B that best matches C, and the width of the river channel inside the best matching model is used as the river channel width value of the actual work area.

[0048] Example 3

[0049] In a specific embodiment 3 of the present invention, the method for determining the width of a river channel based on deep learning of the present invention includes:

[0050] (1) Figure 2 As shown, it is the data of a specific embodiment, wherein Figure 2 AC is a reference model with M (M=3) river channel widths of different types. The river channel widths of the reference models gradually change from narrow to wide. Figure 2 DF and Figure 2 AC corresponds to Figure 2 The M virtual models of DF are used as the sedimentary facies models of the real work area, and a discrete point set of k (k / total number of model points = 10%) points is randomly selected from them as the actual well condition data C (see Figure 2 GI).

[0051] (2) Based on deep learning, a convolutional neural network is trained by multiple random sampling samples of candidate reference river models. The specific training process is as follows:

[0052] (2.1) Define the training data set PS, which is used to store all random sampling data of all river reference models;

[0053] (2.2) From Figure 2 The m-th channel width reference model of AC is sampled n times (n=100), and each time k points are extracted from the m-th reference model to obtain the point set P m (i);

[0054] (2.3) Give P m (i) Add the text tag {m} and change P m (i) and the label {m} are added to PS;

[0055] (2.4) Based on transfer learning, CNN is used to train PS and obtain the trained model CNN PS .

[0056] (3) Use the trained model CNN PS The conditional data C is identified, and the river channel reference model that best matches C is identified from the M river channel models, and the river channel width of the reference model is taken as the most appropriate value.

[0057] (4) Check the accuracy of river width recognition. Figure 3 As shown, from Figure 2 The random sampling in the virtual model of the narrow river in D is used as the conditional data C for the reference model of the same narrow river ( Figure 2 A) Perform recognition, count the recognition accuracy of 100 tests, and obtain the statistical histogram of the recognition accuracy of river width. The recognition accuracy of narrow river can reach about 85%. The corresponding recognition accuracy of medium river and wide river models is close to 100%, indicating that the proposed method is more suitable for the recognition task of river width.

[0058] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions recorded in the aforementioned embodiments or replace some of the technical features therein with equivalents. 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.

[0059] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. A method for determining river width based on deep learning, characterized in that: The method for determining the width of a river channel based on deep learning includes: Step 1, input the condition data of the actual work area; Step 2, design different river channel width parameters; Step 3, generating a plurality of river channel models with different widths according to the different river channel width parameter combinations in step 2; Step 4, training a neural network based on multiple random samples of the candidate random river channel model based on deep learning; Step 5, identifying the river channel model that best matches the conditional data from multiple river channel models, and taking the river channel width of the model as the most appropriate value.

2. The method for determining river channel width based on deep learning according to claim 1, characterized in that: In step 1, the condition data C of the actual work area is input, where C is a set of k points.

3. The method for determining river channel width based on deep learning according to claim 2, characterized in that: In step 1, the scope of the study area is defined according to the distribution characteristics of the drilling data, and the study area must include all well data.

4. The method for determining river channel width based on deep learning according to claim 1, characterized in that: In step 2, different channel width parameters are designed, including channel width, aspect ratio, and main direction.

5. The method for determining river channel width based on deep learning according to claim 1, characterized in that: In step 3, based on the fluvsim algorithm, M river channel models with different widths are generated according to different parameter combinations in step 2. m Represents the width of the mth river channel model, where m=1,…,M.

6. The method for determining river channel width based on deep learning according to claim 1, characterized in that: In step 4, we first define the number of samples from each river model as n, the training data set PS for convolutional deep learning, and then randomly select k points from the mth river model to obtain the point set P m (i) to P m (i) Add the text tag {m} and change P m (i) Add to PS, and finally use CNN to train PS based on transfer learning to obtain the trained model CNN PS .

7. The method for determining river channel width based on deep learning according to claim 1, characterized in that: Step 4 includes: (4.1) Define the training data set PS, which is used to store all random sampling data of all river reference models; (4.2) Sample n times from the m-th river width reference model, and each time extract k points from the m-th reference model to obtain the point set P m (i); (4.3) Give P m (i) Add the text tag {m} and change P m (i) and the label {m} are added to PS; (4.4) Based on transfer learning, CNN is used to train PS and obtain the trained model CNN PS .

8. The method for determining river channel width based on deep learning according to claim 1, characterized in that: In step 5, use the trained model CNN PS Detect condition data C, identify the river channel model that best matches C from M river channel models, and use the river channel width of this model as the most appropriate value.

9. The method for determining river channel width based on deep learning according to claim 1, characterized in that: The method for determining the width of a river channel based on deep learning also includes, after step 5, step 6, checking the accuracy of river channel width recognition.

10. The method for determining river channel width based on deep learning according to claim 1, characterized in that: In step 6, a random sample is taken from the virtual model of the narrow river channel as conditional data to identify the reference model of the equally narrow river channel, and the recognition accuracy of 100 tests is counted to obtain a statistical histogram of the river channel width recognition accuracy, thereby detecting the river channel width recognition accuracy.

Citation Information

Patent Citations

  • Three-dimensional geologic model generation method and system and computer storage medium

    CN111612906A

  • Method for generating one-way river channel high-precision two-dimensional terrain based on deep learning

    CN113888670A