Ground stress prediction method based on multi-modal deep learning

Abstract

The invention provides a crustal stress prediction method based on multi-modal deep learning, and the method comprises the steps: S11, obtaining seismic data, horizon data and logging data of an oil and gas reservoir region, and carrying out the preprocessing of the obtained seismic data and horizon data; s12, extracting deep features from the seismic data by using a convolutional neural network, performing sequence modeling on the horizon data by using a bidirectional long-short term memory network, and capturing time sequence features of horizon changes; s13, constructing a multi-modal ground stress prediction model; and S14, applying the multi-modal crustal stress prediction model to the new seismic data and horizon data for crustal stress prediction, and generating a report for displaying a prediction result. By fusing the seismic data and the horizon data, constructing the multi-modal deep learning model and comprehensively utilizing the information of the multi-source data, more comprehensive crustal stress distribution information can be provided, comparison verification with the logging data can be carried out, the reliability of the prediction result is further improved, and higher-precision crustal stress prediction is realized.

Description

Technical Field

[0001] The present invention relates to the field of geological engineering technology, and in particular to a geostress prediction method based on multimodal deep learning. Background Art

[0002] Geostress prediction plays an important role in oil and gas reservoir development, drilling design and wellbore stability assessment. Accurate prediction of geostress can effectively reduce drilling risks, optimize mining plans and improve oil and gas recovery. However, traditional geostress prediction methods mainly rely on well logging data and seismic attribute analysis, which have some limitations. In recent years, deep learning technology has made significant progress in image recognition, natural language processing and other fields, and has gradually been applied to the field of geological engineering. By constructing a multi-layer neural network, the deep learning method can automatically extract high-level features from a large amount of data, and has strong nonlinear modeling and data fusion capabilities. Therefore, geostress prediction methods based on deep learning have gradually attracted attention. Existing deep learning geostress prediction methods mainly focus on the application of a single data source, such as using seismic data or well logging data for prediction. However, the single data source method has problems such as incomplete data and insufficient feature extraction, which makes it difficult to fully reflect the underground geological characteristics, resulting in limited prediction accuracy. Summary of the invention

[0003] In view of this, the purpose of the present invention is to provide a method for predicting geostress based on multimodal deep learning. By fusing seismic attributes with stratigraphic characteristics and adopting a multimodal deep learning method, geostress can be predicted more accurately, providing important geological parameter support for oil and gas exploration and development.

[0004] To achieve the above-mentioned object of the invention, the present invention provides a method for predicting geostress based on multimodal deep learning, the method comprising:

[0005] S11, obtaining seismic data, stratigraphic data and well logging data of the oil and gas reservoir area, and preprocessing the obtained seismic data and stratigraphic data;

[0006] S12. Use convolutional neural networks to extract deep features from seismic data, and use bidirectional long short-term memory networks to perform sequence modeling on the stratigraphic data to capture the temporal characteristics of stratigraphic changes.

[0007] S13, constructing a multimodal geostress prediction model, wherein the multimodal geostress prediction model includes a stratigraphic data subnetwork and a seismic data subnetwork, wherein the stratigraphic data subnetwork is used to process temporal characteristics of stratigraphic changes, and the seismic data subnetwork is used to process spatial characteristics of seismic attributes, and the temporal characteristics and spatial characteristics output by the seismic data subnetwork and the stratigraphic data subnetwork are fused through a fusion layer to generate comprehensive characteristics;

[0008] S14. Apply the multimodal geostress prediction model to new seismic data and stratigraphic data to perform geostress prediction and generate a report showing the prediction results.

[0009] Furthermore, in step S11, the acquired seismic data and stratigraphic data are preprocessed, specifically including:

[0010] S21, using Daubechies wavelet transform method to denoise seismic data;

[0011] S22, converting the time domain seismic data into depth domain seismic data using a velocity model;

[0012] S23. Processing the stratigraphic data using an automatic extraction algorithm to generate a stratigraphic model.

[0013] Further, in step S22, the time domain seismic data is converted into depth domain seismic data using the velocity model, specifically including:

[0014] S31, using geostatistical inversion technology, combining seismic data, stratigraphic data and well logging data to perform inversion and obtain a velocity model;

[0015] S32, using a velocity model to sequentially perform time-to-depth conversion on the seismic data in equally spaced time domains to obtain depth-domain seismic data;

[0016] S33. Fusing the depth domain seismic data with the logging data to form a cross-sectional fusion diagram that can reflect the seismic data.

[0017] Furthermore, in step S12, a convolutional neural network is used to extract deep features from the seismic data, specifically including:

[0018] S41. Introducing residual connections and multi-scale structures into the convolutional neural network ResNet-50 to extract tiny features from seismic data;

[0019] S42, adding attention mechanism after residual connection to extract deep features in seismic data;

[0020] S43. Through three 3*3 convolutions, tiny features and deep features are integrated to capture the spatial characteristics of seismic data.

[0021] Furthermore, in step S13, the seismic attribute characteristics are processed by the seismic data subnetwork, specifically including:

[0022] S51, performing global average pooling on the spatial features in step S43, and aggregating the input features in the spatial dimension to obtain global information;

[0023] S52, adaptively calculating the coverage of local cross-channel interaction according to the number of channels;

[0024] S53. The interaction between each channel and its k neighboring channels is realized through convolution with a one-dimensional convolution kernel of size k.

[0025] Furthermore, in step S12, a bidirectional long short-term memory network is used to perform sequence modeling on the layer data to capture the temporal characteristics of layer changes, specifically including:

[0026] S61, inputting the original stratigraphic data into a bidirectional long short-term memory network according to the chronological order to extract forward and reverse information;

[0027] S62, inputting the information extracted in step S61 into the long short-term memory network to further extract the time dependency;

[0028] S63. Add the temporal attention mechanism to achieve adaptive weighting of deep time series features and capture the temporal characteristics of layer changes.

[0029] Furthermore, in step S13, the temporal characteristics of the layer change are processed by the layer data subnetwork, specifically including:

[0030] S71, normalize the temporal features outputted in step S63, use them as input features of the attention layer, and activate them using the tanh function;

[0031] S72, multiplying the output result of step S71 by the attention parameter matrix, adding the bias weight and performing nonlinear activation via the Softmax function to obtain the weight score of the time series feature in the window;

[0032] S73. Through the matrix broadcast mechanism, perform Hadamard product on the weight score in step S72 and the time series feature input in step S71 to weight the time series feature input in S71.

[0033] Further, in step S13, the temporal features and spatial features output by the seismic data subnetwork and the stratigraphic data subnetwork are fused through a fusion layer, specifically including:

[0034] S81, the fusion layer uses the self-attention mechanism to learn temporal features and spatial features respectively;

[0035] S82, connect the output of S81 to the spatial feature through residual connection, perform layer normalization, and assign weights according to the importance of temporal features and spatial features;

[0036] S83, fuse the output of S82 into comprehensive features through a feedforward neural network.

[0037] Furthermore, in step S14, the in-situ stress prediction is specifically to map the comprehensive features fused in step S83 into in-situ stress prediction values ​​through a fully connected layer.

[0038] Furthermore, in step S14, the report includes a geostress prediction profile and a geostress distribution map.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a method for predicting ground stress based on multimodal deep learning and fusion of seismic attributes and stratigraphic features. By constructing a multimodal deep learning model and fusing seismic data and stratigraphic data, a high-precision prediction of ground stress is achieved, and the prediction results are compared and verified with well logging data, achieving remarkable results, including:

[0041] (1) Improving the accuracy of geostress prediction: By fusing seismic data and stratigraphic data, the method of the present invention can automatically extract and fuse high-level features from multiple data. Seismic data provides extensive spatial information, and stratigraphic data provides detailed geological stratigraphic change characteristics. The multimodal deep learning model organically combines the two and enhances the model's prediction ability.

[0042] (2) Expanding the scope of application of geostress prediction: The method of the present invention can be applied to geostress prediction under different geological conditions. By integrating seismic data and stratigraphic data, effective geostress prediction can be performed in undrilled areas, providing geostress distribution information in a larger range, and providing comprehensive geological parameter support for oil and gas exploration and development.

[0043] (3) Realize automation and efficiency of the prediction process: The method of the present invention adopts deep learning technology to realize the automation and efficiency of the geostress prediction process. The multimodal deep learning model can automatically extract features from seismic data and stratigraphic data, reducing manual intervention, simplifying the prediction process, and improving work efficiency. In addition, the high parallel computing capability of the deep learning model significantly shortens the prediction time.

[0044] (4) Comparison and verification with well logging data to improve the reliability of results: The method of the present invention can not only independently predict geostress, but also compare and analyze the predicted results with actual well logging data to evaluate and verify the prediction accuracy and reliability of the model. By comparing and verifying with well logging data, the credibility of the geostress prediction results is further improved, providing more reliable geological parameter support for oil and gas reservoir development and wellbore stability assessment.

[0045] (5) Providing comprehensive geological information support: The method of the present invention provides more comprehensive and detailed geological information by fusing seismic data and stratigraphic data. The combination of seismic data and stratigraphic data not only makes the ground stress prediction results highly accurate, but also can reflect the changes in geological stratigraphic conditions, thus providing important geological information support for various applications such as oil and gas reservoir development, drilling design, and geological disaster prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 It is a flow chart of a method for predicting geostress based on multimodal deep learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The enumerated embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0049] Reference Figure 1 This embodiment provides a method for predicting geostress based on multimodal deep learning, the method comprising:

[0050] S11, obtaining seismic data, stratigraphic data and well logging data of the oil and gas reservoir area, and preprocessing the obtained seismic data and stratigraphic data.

[0051] In this embodiment, seismic data includes the time domain and depth domain information of seismic reflection waves, and the stratigraphic data includes the thickness, dip, lithology and other attributes of the strata. The seismic reflection profile and velocity model data are denoised, and the time domain seismic data are converted into depth domain data. The top and bottom interfaces of different strata are determined by combining manual interpretation and automatic extraction, and a stratigraphic model is generated. The well logging data is standardized and key parameters such as acoustic wave velocity, resistivity, density, etc. are extracted for model verification.

[0052] In step S11, the acquired seismic data and horizon data are preprocessed, specifically including:

[0053] S21. De-noising the seismic data using the Daubechies wavelet transform method. De-noising the seismic data using the Daubechies wavelet transform method (db10) has good de-noising effect and protection capabilities for seismic signals.

[0054] S22. Use the velocity model to convert the time domain seismic data into depth domain seismic data. Use seismic interpretation software, combined with seismic reflection characteristics and geological knowledge, to manually interpret the top and bottom interfaces of the strata.

[0055] S23. Use the automatic extraction algorithm to process the layer data and generate a layer model. Collect well logging data, including parameters such as acoustic wave velocity, resistivity and density.

[0056] In step S22, the time domain seismic data is converted into depth domain seismic data using the velocity model, which specifically includes:

[0057] S31. Use geostatistical inversion technology to combine seismic data, stratigraphic data and logging data for inversion to obtain a velocity model.

[0058] S32. Using a velocity model, the seismic data is sequentially converted into time-depth in a time domain at equal intervals to obtain seismic data in a depth domain.

[0059] S33. Fusing the depth domain seismic data with the logging data to form a cross-sectional fusion diagram that can reflect the seismic data.

[0060] In this embodiment, the application of geostatistical inversion technology can not only directly obtain the layer velocity body of longitudinal wave and shear wave, but also the inversion result has a higher resolution. At the same time, each sampling point of the same sampling interval can correspond one to one, which can ensure the accuracy of time-to-depth conversion. Geostatistical inversion not only effectively integrates seismic data, stratigraphic data and logging data, but also integrates existing geological knowledge information through means such as histograms and variograms, and obtains high-resolution inversion results through inversion. There will be certain differences in details between each inversion result. These differences in details reflect the uncertainty of underground geological conditions and can provide a variety of input information for subsequent multi-modal geostress prediction models.

[0061] S12. Use convolutional neural networks to extract deep features from seismic data, and use bidirectional long short-term memory networks to perform sequence modeling on stratigraphic data to capture the temporal characteristics of stratigraphic changes.

[0062] In this embodiment, an improved version of the convolutional neural network (ResNet-50) is used to extract deep abstract seismic attribute features from seismic data. By adding residual connections or dense connections, the model's ability to capture complex seismic patterns is improved. Through multiple levels of abstract processing of the deep network structure, the complex patterns and intrinsic structures in the seismic data can be captured, thereby providing richer and more detailed information for ground stress prediction. The convolutional neural network consists of multiple convolutional layers and pooling layers to capture the spatial characteristics of seismic data. A bidirectional long short-term memory network (Bi-LSTM) is used to perform sequence modeling on the stratigraphic data to capture the temporal characteristics of stratigraphic changes. The bidirectional long short-term memory network can effectively extract contextual information and enhance the understanding of dynamic changes in stratigraphic layers.

[0063] In step S12, deep features are extracted from seismic data using a convolutional neural network, specifically including:

[0064] S41. Introduce residual connections and multi-scale structures into the convolutional neural network ResNet-50 to extract tiny features in seismic data. Introduce multi-scale structures into the ResNet-50 network model for feature fusion to improve the network's ability to extract tiny features in seismic data.

[0065] S42. Add an attention mechanism after the residual connection to extract deep features in seismic data. The attention mechanism is added to the model to assign different weights to the feature information of different channels, enhance the feature expression ability of the network, and enable the model to better learn the deep features in the image.

[0066] S43, through three 3*3 convolutions, the tiny features and deep features are integrated to capture the spatial features of the seismic data. It is composed of three 3*3 convolutions stacked vertically, and the standardization and Hswish activation function are applied between the convolutions. The integrated features obtained by the convolution block are processed by maximum pooling, which can better obtain the detailed features in the seismic data, and with fewer parameters, the model can capture higher-level spatial features and increase the expressive power of the model.

[0067] In step S12, a bidirectional long short-term memory network is used to perform sequence modeling on the layer data, and the features specifically include:

[0068] S61. Input the original stratigraphic data into a bidirectional long short-term memory network in chronological order for forward and reverse information extraction.

[0069] S62. Input the information extracted in step S61 into the long short-term memory network to further extract the time dependency.

[0070] S63. Add the temporal attention mechanism to achieve adaptive weighting of deep time series features and capture the temporal characteristics of layer changes.

[0071] In this embodiment, since the long short-term memory network has a strong time series capture capability, the bidirectional long short-term memory network is selected to capture the time series characteristics of the layer change, and the information on the layer change time series is further captured by embedding the long short-term memory network to obtain a deeper level of time series dependency. At the same time, in order to enhance the significant expression of different time series characteristics on the current layer change prediction sequence, a time attention mechanism is designed to be added to the above-mentioned bidirectional long short-term memory network to achieve adaptive weighting of deep time series features.

[0072] S13. Construct a multimodal geostress prediction model, which includes a stratigraphic data subnetwork and a seismic data subnetwork. The stratigraphic data subnetwork is used to process the temporal characteristics of stratigraphic changes, and the seismic data subnetwork is used to process the spatial characteristics of seismic attributes. The temporal characteristics and spatial characteristics output by the seismic data subnetwork and the stratigraphic data subnetwork are fused through a fusion layer to generate comprehensive characteristics.

[0073] In this embodiment, the stratigraphic data subnetwork and the seismic data subnetwork are two parallel subnetworks, processing seismic data and stratigraphic data respectively. The seismic data subnetwork adopts an improved version of the convolutional neural network (ResNet-50), and the stratigraphic data subnetwork adopts a bidirectional long short-term memory network (Bi-LSTM). The output features of the seismic data subnetwork and the stratigraphic data subnetwork are fused through the fusion layer to generate a comprehensive feature representation. This layer uses a self-attention mechanism (Self-Attention) for weighted fusion in order to better combine the features of seismic data and stratigraphic data.

[0074] In step S13, the seismic attribute characteristics are processed by the seismic data subnetwork, specifically including:

[0075] S51, the spatial features in step S43 are pooled by global average, and the input features are aggregated in the spatial dimension to obtain global information.

[0076] S52. Adaptively calculate the coverage of local cross-channel interaction according to the number of channels.

[0077] S53. The interaction between each channel and its k neighboring channels is realized through convolution with a one-dimensional convolution kernel of size k.

[0078] In this embodiment, global average pooling and adaptive weight adjustment are introduced to significantly reduce the amount of channel attention calculation, thereby improving the perception of features at each stage and maintaining good model performance. Finally, one-dimensional convolution is used to achieve cross-channel dimensional interaction and avoid dimensionality reduction.

[0079] In step S13, the temporal characteristics of the layer change are processed by the layer data subnetwork, specifically including:

[0080] S71. Normalize the temporal features outputted in step S63, use them as input features of the attention layer, and activate them using the tanh function.

[0081] S72: Multiply the output result of step S71 by the attention parameter matrix, add the bias weight, and activate it nonlinearly through the Softmax function to obtain the weight score of the time series feature in the window.

[0082] S73. Through the matrix broadcast mechanism, perform Hadamard product on the weight score in step S72 and the time series feature input in step S71 to weight the time series feature input in S71.

[0083] In this embodiment, considering the correlation of time series in time series, the output features after passing through the bidirectional long short-term memory network and the long short-term memory network are regarded as the input features of the attention layer, and the tanh function is used for activation to make the input sequence information more stable to ensure a more reasonable distribution of weights. Different weight scores weight the time series information extracted by the long short-term memory network, and adaptively enhance the significant expression of different time series features for the current layer change prediction sequence.

[0084] In step S13, the temporal features and spatial features output by the seismic data subnetwork and the stratigraphic data subnetwork are fused through a fusion layer, specifically including:

[0085] S81, the fusion layer uses the self-attention mechanism to learn temporal features and spatial features respectively;

[0086] S82, connect the output of S81 to the spatial feature through residual connection, perform layer normalization, and assign weights according to the importance of temporal features and spatial features;

[0087] S83, fuse the output of S82 into comprehensive features through a feedforward neural network.

[0088] In this embodiment, the temporal features and spatial features are learned separately through the self-attention mechanism to capture the complex relationship between the temporal features and the spatial features, and weights are assigned according to the importance of the two for fusion. The purpose of using residual connection to connect with the spatial features is to retain more spatial information, because the spatial features are more important than the temporal features for the final prediction of geostress. The temporal features and spatial features are fused through layer normalization and feedforward neural network sub-modules to obtain more accurate comprehensive features that can map the predicted geostress values.

[0089] S14. Apply the multimodal geostress prediction model to new seismic data and stratigraphic data to perform geostress prediction and generate a report showing the prediction results.

[0090] In step S14, the in-situ stress prediction is specifically to map the integrated features fused in step S83 into in-situ stress prediction values ​​through a fully connected layer. The fully connected layer uses the Dropout technology to reduce overfitting, and uses BatchNormalization to speed up the training and improve the model stability.

[0091] In step S14, the report includes a geostress prediction profile and a geostress distribution map.

[0092] In this embodiment, the trained model is applied to new seismic data and stratigraphic data to predict geostress, generate geostress prediction profiles and stress distribution diagrams, and intuitively display the prediction results. The geostress prediction results are compared and analyzed with the actual logging data to evaluate the prediction accuracy and reliability of the model, and the accuracy of the prediction results is quantified using statistical indicators (such as mean square error, correlation coefficient, etc.).

[0093] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting ground stress based on multimodal deep learning, characterized in that: The method comprises: S11, obtaining seismic data, stratigraphic data and well logging data of the oil and gas reservoir area, and preprocessing the obtained seismic data and stratigraphic data; S12. Use convolutional neural networks to extract deep features from seismic data, and use bidirectional long short-term memory networks to perform sequence modeling on the stratigraphic data to capture the temporal characteristics of stratigraphic changes. S13, constructing a multimodal geostress prediction model, wherein the multimodal geostress prediction model includes a stratigraphic data subnetwork and a seismic data subnetwork, wherein the stratigraphic data subnetwork is used to process temporal characteristics of stratigraphic changes, and the seismic data subnetwork is used to process spatial characteristics of seismic attributes, and the temporal characteristics and spatial characteristics output by the seismic data subnetwork and the stratigraphic data subnetwork are fused through a fusion layer to generate comprehensive characteristics; S14. Apply the multimodal geostress prediction model to new seismic data and stratigraphic data to perform geostress prediction and generate a report showing the prediction results.

2. The method for predicting geostress based on multimodal deep learning according to claim 1, characterized in that: In step S11, the acquired seismic data and horizon data are preprocessed, specifically including: S21, using Daubechies wavelet transform method to denoise seismic data; S22, converting the time domain seismic data into depth domain seismic data using a velocity model; S23. Processing the stratigraphic data using an automatic extraction algorithm to generate a stratigraphic model.

3. The method for predicting geostress based on multimodal deep learning according to claim 2, characterized in that: In step S22, the time domain seismic data is converted into depth domain seismic data using the velocity model, which specifically includes: S31, using geostatistical inversion technology, combining seismic data, stratigraphic data and well logging data to perform inversion and obtain a velocity model; S32, using a velocity model to sequentially perform time-to-depth conversion on the seismic data in equally spaced time domains to obtain depth-domain seismic data; S33. Fusing the depth domain seismic data with the logging data to form a cross-sectional fusion diagram that can reflect the seismic data.

4. The method for predicting geostress based on multimodal deep learning according to claim 1, characterized in that: In step S12, deep features are extracted from seismic data using a convolutional neural network, specifically including: S41. Introducing residual connections and multi-scale structures into the convolutional neural network ResNet-50 to extract tiny features from seismic data; S42, adding attention mechanism after residual connection to extract deep features in seismic data; S43. Through three 3*3 convolutions, tiny features and deep features are integrated to capture the spatial characteristics of seismic data.

5. The method for predicting geostress based on multimodal deep learning according to claim 4, characterized in that: In step S13, the seismic attribute characteristics are processed by the seismic data subnetwork, specifically including: S51, performing global average pooling on the spatial features in step S43, and aggregating the input features in the spatial dimension to obtain global information; S52, adaptively calculating the coverage of local cross-channel interaction according to the number of channels; S53. The interaction between each channel and its k neighboring channels is realized through convolution with a one-dimensional convolution kernel of size k.

6. The method for predicting geostress based on multimodal deep learning according to claim 1, characterized in that: In step S12, a bidirectional long short-term memory network is used to perform sequence modeling on the layer data to capture the temporal characteristics of layer changes, specifically including: S61, inputting the original stratigraphic data into a bidirectional long short-term memory network according to the chronological order to extract forward and reverse information; S62, inputting the information extracted in step S61 into the long short-term memory network to further extract the time dependency; S63. Add the temporal attention mechanism to achieve adaptive weighting of deep time series features and capture the temporal characteristics of layer changes.

7. The method for predicting geostress based on multimodal deep learning according to claim 6, characterized in that: In step S13, the temporal characteristics of the layer change are processed by the layer data subnetwork, specifically including: S71, normalize the temporal features outputted in step S63, use them as input features of the attention layer, and activate them using the tanh function; S72, multiplying the output result of step S71 by the attention parameter matrix, adding the bias weight and performing nonlinear activation via the Softmax function to obtain the weight score of the time series feature in the window; S73. Through the matrix broadcast mechanism, perform Hadamard product on the weight score in step S72 and the time series feature input in step S71 to weight the time series feature input in S71.

8. The method for predicting geostress based on multimodal deep learning according to claim 1, characterized in that: In step S13, the temporal features and spatial features output by the seismic data subnetwork and the stratigraphic data subnetwork are fused through a fusion layer, specifically including: S81, the fusion layer uses the self-attention mechanism to learn temporal features and spatial features respectively; S82, connect the output of S81 to the spatial feature through residual connection, perform layer normalization, and assign weights according to the importance of temporal features and spatial features; S83, fuse the output of S82 into comprehensive features through a feedforward neural network.

9. The method for predicting geostress based on multimodal deep learning according to claim 8, characterized in that: In step S14, the in-situ stress prediction is specifically to map the comprehensive features fused in step S83 into in-situ stress prediction values ​​through a fully connected layer.

10. The method for predicting geostress based on multimodal deep learning according to claim 1, characterized in that: In step S14, the report includes a geostress prediction profile and a geostress distribution map.

Images