A method and device for determining an exploitation strategy based on a core image generation model

Abstract

The specification provides a method and device for determining an exploitation strategy based on a core image generation model. Obtain core description data of a target area and image quality requirement information corresponding to the target area; use a first module of a preset generation model to generate a first core image according to the core description data and the image quality requirement information; use a second module of the preset generation model to adjust and process the first core image to obtain a second core image; wherein the second module is obtained by training according to second historical core data, a first historical generated image, and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type of the target area, and the first historical generated image is a core image generated by the first module according to the second historical core data; and determine an exploitation strategy for the target area according to the second core image. Thus, the shortcomings of the existing digital core technology in terms of sample coverage are overcome.

Description

TECHNICAL FIELD

[0001] The present specification belongs to the technical field of digital core technology, and particularly relates to a mining strategy determination method and device based on core image generation model. BACKGROUND

[0002] At present, although the digital core technology realizes the digital modeling and analysis of the internal structure of the core by using high-precision imaging and computer simulation, there are still problems such as high cost, low scanning efficiency, insufficient sample coverage, and limited model extrapolation ability, and there is a lack of efficient image generation and enhancement means, which limits the comprehensive research and application of the diversity and depth characteristics of the core.

[0003] In view of the above problems, an effective solution has not been proposed at present. SUMMARY

[0004] The present specification provides a mining strategy determination method and device based on core image generation model. First, by using a first module of a preset generation model, according to core description data of a target area and image quality requirement information, a module based on a diffusion algorithm is trained in combination with first historical core data and corresponding first historical core images, so as to efficiently generate a first core image, effectively reducing the problems of high cost and low scanning efficiency of traditional high-precision imaging. Second, by using a second module of a preset generation model, the model is trained in combination with second historical core data, first historical generated images and corresponding second historical core images, and the first core image is optimized and fine-tuned to obtain a second core image that more accurately reflects the core structure characteristics of the target area, overcoming the deficiencies of existing digital core technology in terms of sample coverage, and providing fine and comprehensive data support for formulating a scientific and reasonable mining strategy.

[0005] The present specification provides a mining strategy determination method based on core image generation model, comprising:

[0006] Obtaining core description data of a target area and image quality requirement information corresponding to the target area; wherein the core description data at least includes a mineral composition proportion prompt word, and the mineral composition proportion prompt word is generated according to a preset prompt word construction rule;

[0007] Using a first module of a preset generation model, generating a first core image according to the core description data and the image quality requirement information; wherein the first module is obtained by training a module based on a diffusion algorithm according to first historical core data and corresponding first historical core images;

[0008] A second module of the preset generation model is used to perform adjustment processing on the first core image to obtain a second core image; wherein the second module is obtained by training according to second historical core data, a first historical generated image, and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target region, and the first historical generated image is a core image generated by the first module according to the second historical core data;

[0009] According to the second core image, a mining strategy for the target region is determined.

[0010] In one embodiment, the first module of the preset generation model generates a first core image according to the core description data and the image quality requirement information, including:

[0011] The first module of the preset generation model is used to perform feature extraction processing on the core description data and the image quality requirement information respectively to obtain a first feature vector corresponding to the core description data and a second feature vector corresponding to the image quality requirement information;

[0012] A conditional embedding vector is determined according to the first feature vector, the second feature vector, and a preset random noise;

[0013] The first module of the preset generation model is used to generate the first core image according to the conditional embedding vector. In one embodiment, before the first module of the preset generation model generates the first core image according to the core description data and the image quality requirement information, it further includes:

[0014] The first historical core data and the first historical core image are obtained;

[0015] The first module of the preset generation model is used to determine a second historical generated image according to the first historical core data;

[0016] The first module of the preset generation model is iteratively trained according to the first historical core image and the second historical generated image to obtain the first module of the preset generation model;

[0017] The second historical core data and a second historical core image are obtained, and the data amount of the first historical core data is greater than the data amount of the second historical core data;

[0018] The first module of the preset generation model is used to determine the first historical generated image according to the second historical core data;

[0019] A second module of the preset generation model is used to adjust the mineral color proportion of the first historical generation image to obtain a third core image.

[0020] The second module of the preset generation model is iteratively trained according to the second historical core image and the third core image to obtain the second module of the preset generation model.

[0021] In one embodiment, the core description data of the target area is obtained, including:

[0022] The first description data of the target area is obtained.

[0023] The first description data is extracted by using a preset large language model to obtain the core description data.

[0024] In one embodiment, the method further includes:

[0025] In the iterative training of the second module of the preset generation model, the preset prompt word construction rule is determined according to the mineral composition proportion parameters of the second historical core image and the third core image.

[0026] In one embodiment, the second module of the preset generation model is iteratively trained according to the second historical core image and the third core image to obtain the second module of the preset generation model, including:

[0027] A first loss value is determined according to the mineral composition proportion of the second historical core image and the mineral composition proportion of the third core image.

[0028] A second loss value is determined according to the resolution of the second historical core image and the resolution of the third core image.

[0029] The second module of the preset generation model is iteratively trained according to the first loss value and the second loss value to obtain the second module of the preset generation model.

[0030] In one embodiment, the mineral distribution of the second historical core image is obtained by using a preset extraction model to extract the distribution of the second historical core image; and the preset extraction model is a model constructed based on a preset deep learning algorithm.

[0031] The present specification provides a mining strategy determination device based on a core image generation model, including:

[0032] The data acquisition module is configured to acquire core description data of a target area and image quality requirement information corresponding to the target area. The core description data at least includes a mineral composition ratio prompt word, and the mineral composition ratio prompt word is generated according to a preset prompt word construction rule.

[0033] The first image generation module is configured to generate a first core image according to the core description data and the image quality requirement information by using a first module of a preset generation model. The first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and a corresponding first historical core image.

[0034] The second image generation module is configured to adjust and process the first core image to obtain a second core image by using a second module of the preset generation model. The second module is obtained by training according to second historical core data, a first historical generated image, and a corresponding second historical core image. The region type corresponding to the second historical core data is the same as the region type corresponding to the target area. The first historical generated image is a core image generated by using the first module according to the second historical core data.

[0035] The strategy determination module is configured to determine a mining strategy for the target area according to the second core image.

[0036] The present specification also provides an electronic device including a processor and a memory for storing processor-executable instructions, and the processor implements a mining strategy determination method based on a core image generation model when executing the instructions.

[0037] The present specification also provides a computer-readable storage medium having stored thereon computer instructions, and the instructions implement a mining strategy determination method based on a core image generation model when executed.

[0038] Based on the core image generation model based on the method for determining the mining strategy provided in the specification, the core description data of the target area and the image quality requirement information corresponding to the target area are obtained; wherein the core description data at least includes a mineral composition ratio prompt word, and the mineral composition ratio prompt word is generated according to a preset prompt word construction rule; a first module of a preset generation model is used to generate a first core image according to the core description data and the image quality requirement information; wherein the first module is obtained by training a module based on a diffusion algorithm according to first historical core data and corresponding first historical core image; a second module of the preset generation model is used to adjust and process the first core image to obtain a second core image; wherein the second module is obtained by training according to second historical core data, first historical generated image and corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by the first module according to the second historical core data; the mining strategy for the target area is determined according to the second core image. In this way, first, by using the first module of the preset generation model, the first core image is efficiently generated according to the core description data and the image quality requirement information of the target area, and the first historical core data and the corresponding first historical core image are combined to train the module based on the diffusion algorithm, thereby effectively reducing the high cost and low scanning efficiency problems of traditional high-precision imaging; second, by using the second module of the preset generation model, the model is trained by combining the second historical core data, the first historical generated image and the corresponding second historical core image, and the first core image is optimized and fine-tuned to obtain a second core image that more accurately reflects the core structure characteristics of the target area, overcoming the deficiencies of existing digital core technology in sample coverage and providing fine and comprehensive data support for formulating a scientific and reasonable mining strategy. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the specification, the drawings needed in the embodiments will be briefly introduced as follows. The drawings in the following description are only some embodiments described in the specification, and those skilled in the art can also obtain other drawings according to these drawings without creating labor.

[0040] Figure 1 is a flowchart of a method for determining a mining strategy based on a core image generation model provided by an embodiment of the specification;

[0041] Figure 2 is a schematic diagram of the structure of an electronic device provided by an embodiment of the specification;

[0042] Figure 3 is a structural composition schematic diagram of a mining strategy determination device based on core image generation model provided by an embodiment of the present specification;

[0043] Figure 4 is a structural composition schematic diagram of a mining strategy determination device based on core image generation model provided by an embodiment of the present specification;

[0044] Figure 5 is a color mineral proportion schematic diagram of the original image and the generated image provided by an embodiment of the present specification. DETAILED DESCRIPTION

[0045] In order for those skilled in the art to better understand the technical solutions in the present specification, the technical solutions in the present specification will be described clearly and completely in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present specification, not all. Based on the embodiments in the present specification, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present specification.

[0046] The existing digital core technology relies on advanced imaging technology and computer simulation means to realize high-precision digital modeling and analysis of actual cores. Through CT scanning, microscope imaging and other means to obtain internal structure data of rocks, combined with computer simulation, researchers can in-depth analyze the physical and chemical properties of rocks without damaging the cores. This technology is widely used in the exploration and development of oil, gas and other resources. However, the current technical system has not introduced the text generation image (text-to-image) capability of stable diffusion model (Stable Diffusion, SD) into the field of oil and gas development, especially in the field of core image generation, which is still in a blank state.

[0047] To solve the above problems, first, by using the first module of the preset generation model, according to the core description data of the target area and the image quality requirement information, combining the first historical core data and the corresponding first historical core image, the module based on diffusion algorithm is trained, so as to efficiently generate the first core image, effectively reducing the high cost and low scanning efficiency of traditional high-precision imaging; second, by using the second module of the preset generation model, by combining the second historical core data, the first historical generated image and the corresponding second historical core image, the model is trained, the first core image is optimized and fine-tuned, and a second core image more accurately reflecting the core structure characteristics of the target area is obtained, overcoming the shortcomings of the existing digital core technology in sample coverage and providing fine and comprehensive data support for formulating scientific and reasonable mining strategies.

[0048] Reference Figure 1As shown, the embodiment of the present specification provides a method for determining mining strategy based on core image generation model, wherein the method is specifically applied to the server side. In specific implementation, the method can include the following contents:

[0049] S101: Obtain core description data of a target area and image quality requirement information corresponding to the target area; wherein the core description data at least includes a mineral composition ratio prompt word, and the mineral composition ratio prompt word is generated according to a preset prompt word construction rule;

[0050] S102: Use a first module of a preset generation model to generate a first core image according to the core description data and the image quality requirement information; wherein the first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and a corresponding first historical core image;

[0051] S103: Use a second module of the preset generation model to adjust and process the first core image to obtain a second core image; wherein the second module is obtained by training according to second historical core data, a first historical generated image and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by using the first module according to the second historical core data;

[0052] S104: Determine a mining strategy for the target area according to the second core image.

[0053] Wherein, the above core description data can include mineral information, mineral composition ratio, mineral distribution and pore structure.

[0054] The preset prompt word construction rule can be determined by statistical analysis of historical core image data of the same region type as the target region. In a specific implementation, first, the proportion information of each mineral composition in each core image is extracted from the historical data set (for example, data obtained by scanning electron microscopy (SEM) and X-ray diffraction (XRD) techniques), and then the proportion distribution of each mineral (such as calcium, quartz, feldspar, etc.) in these images is counted to determine the common content interval of each mineral. For example, it is found that the content of calcium is mostly concentrated between 0.5 and 0.6. Based on this statistical result, when constructing the positive prompt word, it is explicitly required that the calcium content in the generated image should be controlled between 0.5 and 0.6, while supplemented by negative prompt words to avoid generating results outside this range. For example, if the target region core description data requires a calcium content of 0.55, the prompt word can be expressed as the calcium content in the core image being controlled between 0.5 and 0.6 to ensure that the mineral proportion of the generated image is consistent with the historical data. This construction rule helps the model accurately capture the key features in the core during training and generation, ensuring the accuracy and stability of the generated image.

[0055] The image quality requirement information corresponding to the target region can be expressed by negative prompt words, including "low quality", "low resolution", "blur", "too many noise points", "color distortion", etc. for describing, which is used to constrain the model during generation to prevent the generated image from having low-quality features that do not meet the requirements of the target region, thereby ensuring that the final generated core image meets the expected standards in terms of clarity, detail performance, and color accuracy.

[0056] In some embodiments, the first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and corresponding first historical core images. In a specific implementation, it can include:

[0057] Specifically, a first module of a preset generation model is used to generate a first core image according to the core description data and image quality requirement information. The first module is constructed based on a diffusion algorithm and is obtained by training using first historical core data and a corresponding first historical core image. In the training process, the module fully learns key features such as mineral composition, pore structure, and color distribution in the core image. For example, by analyzing a large amount of core image data from scanning electron microscopy and X-ray diffraction (SEM-XRD) detection, typical proportions and distribution patterns of different minerals in the image are determined. Therefore, when the core description data of the target area and the corresponding image quality requirement information (such as negative prompt words requiring to avoid low quality and low resolution) are input, the module can generate a high-fidelity first core image reflecting the actual core characteristics, providing reliable data support for subsequent image optimization and core analysis. For example, if historical data show that the content of calcium in the core of a certain area is mostly concentrated between 0.5 and 0.6, the generated image will accurately restore the proportion of this mineral, thereby ensuring the consistency of the generated image with the real core data in color, structure, and composition distribution.

[0058] In some embodiments, a second module of the preset generation model is used to adjust the first core image to obtain a second core image. In specific implementation, the second module can include:

[0059] The second module of the preset generation model fully learns the subtle differences in mineral composition, pore structure, and color distribution of core images in the target area by training the second historical core data, the first historical generated image, and the corresponding second historical core image. Specifically, the second historical core data comes from the same regional type as the target area, and the first historical generated image is a preliminary core image generated by the first module, representing the basic characteristics of the region. During the training process, the second module automatically optimizes various key parameters of the image by comparing the deviations between the first historical generated image and the actual second historical core image. Then, when processing the first core image, the second module makes detailed adjustments to correct the mineral proportion, enhance the structural details, and optimize the color restoration, thereby generating a second core image that better conforms to the actual core characteristics. For example, assuming that historical data show that the content of calcium in the core of the target area should be kept between 0.5 and 0.6, and the proportion of this mineral in the first core image is slightly lower than this range, the second module will adjust the model parameters according to the corresponding second historical core image to improve the proportion of calcium and more accurately reflect the actual situation.

[0060] Based on the above embodiment, through the fine adjustment of the second module to the preliminary generated image, the final generated second core image is closer to the real core data in key features such as mineral proportion, color distribution and pore structure, thereby significantly improving the accuracy of core physical property analysis and the reliability of mining strategy development, and reducing the cost of data acquisition and analysis.

[0061] In some embodiments, the method for determining a mining strategy for the target region according to the second core image can include:

[0062] The key features such as mineral composition, pore structure and fracture distribution in the second core image are quantitatively analyzed, combined with regional geological data and mining history, to evaluate the reservoir development potential and production risk, and accordingly to develop the optimal mining scheme. For example, if the second core image shows that the main mineral proportion and porosity in the target region core are in a favorable range, a high recovery strategy is recommended; otherwise, a more conservative development scheme is recommended. Such a strategy determination method not only ensures the close correspondence between the core image data and the actual geological conditions, but also provides data support and decision basis for scientific and reasonable oil and gas resource mining.

[0063] Based on the above embodiment, first, by using the first module of the preset generation model, according to the core description data and image quality requirement information of the target region, the first core image is generated efficiently by training the module based on the diffusion algorithm combined with the first historical core data and the corresponding first historical core image, thereby effectively reducing the high cost and low scanning efficiency problems of traditional high-precision imaging; second, by using the second module of the preset generation model, the first core image is optimized and fine-tuned by training the model combined with the second historical core data, the first historical generated image and the corresponding second historical core image, to obtain a second core image that more accurately reflects the core structure characteristics of the target region, thereby overcoming the shortcomings of the existing digital core technology in terms of sample coverage, and providing fine and comprehensive data support for developing a scientific and reasonable mining strategy.

[0064] In some embodiments, the first module of the preset generation model generates the first core image according to the core description data and the image quality requirement information, and the method can further include the following content when implemented:

[0065] S1: using the first module of the preset generation model, respectively performing feature extraction processing on the core description data and the image quality requirement information to obtain a first feature vector corresponding to the core description data and a second feature vector corresponding to the image quality requirement information;

[0066] S2: determining a conditional embedding vector according to the first feature vector, the second feature vector, and a preset random noise;

[0067] S3: generating the first core image according to the conditional embedding vector by using a first module of the preset generative model.

[0068] In some embodiments, the first module of the preset generative model respectively performs feature extraction processing on the core description data and the image quality requirement information to obtain a first feature vector corresponding to the core description data and a second feature vector corresponding to the image quality requirement information. Specifically, the process can include:

[0069] In some embodiments, in the processing process of the first module of the preset generative model, the core description data is first preprocessed, and a pre-trained text encoder is used to extract key information such as mineral composition ratio, pore structure, and color distribution, and generate a corresponding first feature vector. At the same time, the image quality requirement information (such as negative prompt words “low resolution”, “blur”, etc.) is analyzed, and features related to image clarity and detail performance are extracted through a specific encoding algorithm to form a second feature vector. Specifically, this process can include data cleaning, text standardization, vectorization, and feature dimension reduction steps, so as to ensure that the two feature vectors can fully represent the core description information and the image quality requirement, respectively. For example, if the core description data contains information such as “calcium content is about 0.55, porosity is about 15%, and mineral distribution is uniform”, the first feature vector will capture these numerical features and distribution patterns; and “avoid low resolution and ensure high definition and detail” in the image quality requirement information will reflect the requirements for image resolution and detail clarity through the second feature vector. In this way, the first module can provide accurate conditional information for subsequent image generation and optimization, ensuring that the generated results not only conform to the geological characteristics of the core, but also meet the expected visual quality standards.

[0070] In some embodiments, the determination of the conditional embedding vector according to the first feature vector, the second feature vector, and the preset random noise can include:

[0071] The first feature vector (capturing key information such as mineral composition and pore structure in the core description data) and the second feature vector (reflecting image quality requirements such as high resolution and clarity requirements) are combined with random noise preset sampled from a standard Gaussian distribution to determine the conditional embedding vector through a specific fusion algorithm.

[0072] The above conditional embedding vector synthesizes the text description and the image quality requirement, and guides the diffusion model to gradually denoise and restore a high-fidelity core image in the generation process.

[0073] The above random noise provides initial diversity for the generation process, and different random seeds can generate different noise samples, thereby achieving diversification of the image generation results. Specifically, the random noise is usually sampled from a standard Gaussian distribution, but its role will be different when generating the conditional embedding vector depending on the combination method of the first feature vector and the second feature vector. For the first feature vector, i.e., the vector reflecting the core description data such as mineral composition, pore structure and other core geological features, the random noise is mainly used to introduce structural diversity in the latent space, so that the model can capture the natural changes of the real core when restoring the core structure. For the second feature vector, i.e., the vector representing the image quality requirements (such as high resolution, clarity requirements), the random noise usually needs to be controlled within a relatively low fluctuation range to ensure consistency in visual details and quality of the generated image.

[0074] In addition, the above preset random noise is not just a simple random disturbance, but is obtained by extraction and statistical analysis according to the inherent noise and error characteristics of the first historical core image. Specifically, the first historical core image is analyzed for noise, the deviations in mineral distribution, color ratio and structural details are quantified, and then parameterized design is performed in combination with the Gaussian distribution to generate random noise closely related to the characteristics of the core image.

[0075] For example, the mineral composition ratio determined by image analysis of the first historical core image is: quartz 40%, feldspar 35%, clay 25%. However, further noise analysis shows that the first historical core image has about +5% deviation in the quartz part and about -3% deviation in the feldspar part, i.e., the quartz is often overestimated and the feldspar is underestimated relative to the true statistical data. Therefore, these deviations are parameterized and adjusted in combination with the Gaussian distribution to construct a preset random noise. The preset random noise will apply appropriate negative disturbance to the quartz part and positive compensation to the feldspar part in the generation process, so that the random noise not only provides initial disturbance, but also closely relates to the inherent noise characteristics of the first historical core image.

[0076] Based on the above embodiment, by fusing the feature vectors extracted from the core description data and the image quality requirement information with the random noise sampled from the standard Gaussian distribution, a comprehensive conditional embedding vector is generated, which effectively guides the diffusion model to restore a high-fidelity core image from random noise. This scheme not only ensures that the generated image is consistent with the actual core data in terms of mineral composition, pore structure, color distribution and other aspects.

[0077] In some embodiments, before the first module of the preset generation model is used to generate the first core image according to the core description data and the image quality requirement information, the method can further include the following content when implemented:

[0078] S1: obtaining the first historical core data and the first historical core image;

[0079] S2: using the first module of the preset generation model to determine a second historical generated image according to the first historical core data;

[0080] S3: iteratively training the first module of the preset generation model according to the first historical core image and the second historical generated image to obtain the first module of the preset generation model;

[0081] S4: obtaining the second historical core data and a second historical core image, the data amount of the first historical core data being greater than that of the second historical core data;

[0082] S5: using the first module of the preset generation model to determine the first historical generated image according to the second historical core data;

[0083] S6: using the second module of the preset generation model to adjust the mineral color proportion of the first historical generated image to obtain a third core image;

[0084] S7: iteratively training the second module of the preset generation model according to the second historical core image and the third core image to obtain the second module of the preset generation model.

[0085] In some embodiments, the first historical core data includes historical core description data and historical image quality requirement information.

[0086] In some embodiments, the second module of the preset generation model is used to adjust the mineral color proportion of the first historical generated image to obtain a third core image, and when implemented, can include:

[0087] First, this module performs local color feature analysis on the first historical generated image, extracts the color distribution and proportion of each mineral in the image; then, compares the extracted color proportion with the target mineral color proportion in the second historical core data to calculate the deviation; then, dynamically adjusts the conditional embedding vector in the back propagation process by optimizing the loss function (such as based on color histogram matching and mean square error), thereby finely correcting the color proportion in the image; finally, the third core image obtained is closer to the target core data in terms of mineral color proportion, ensuring that the generated image has high authenticity and accuracy.

[0088] In some embodiments, the second module of the preset generation model is iteratively trained according to the second historical core image and the third core image, and the second module of the preset generation model is obtained. In actual implementation, the method can include the following steps:

[0089] The second module of the preset generation model is iteratively optimized by combining and training a plurality of third core images and corresponding second historical core images. First, a preliminary first historical generation image is generated by using the first module, and the mineral color ratio is adjusted by the second module to obtain a plurality of third core images. The third core images and the corresponding second historical core images jointly constitute a new training data set, which is input into the second module for further training.

[0090] Specifically, according to the first historical core image, a plurality of third core images are generated by using the preset generation model through multiple iterations, and the third core images show diversity and randomness in mineral distribution and color ratio. Subsequently, the first generated third core image is selected and fused with the first historical core image, which can include image alignment, color matching, structure fusion and other technologies, so as to compare the differences in mineral color boundary, color saturation and overall structure between the two images. By calculating the difference index, the deviation between the generated third core image and the first historical core image is identified. For example, assuming that the color ratio of mineral A in the original first historical core image is 55%, and the color ratio of mineral B is 30%, while the ratio of mineral A in one of the preliminary generated third core images is 60%, and the ratio of mineral B is 25%, it will automatically detect that mineral A is overestimated and mineral B is underestimated. Based on the deviation between the generated third core image and the first historical core image, the parameters of the generation model are adjusted, and the iteration process is entered again to generate a new third core image, so that the mineral ratio and color distribution gradually tend to the characteristics of the real core.

[0091] In this way, the model can learn the variation rule and optimization direction of core mineral composition under a more abundant sample combination, thereby continuously improving the adaptability and generation accuracy of different mineral ratios, and ultimately obtaining a more perfect second module to ensure that the generated core image is highly consistent with the real data in terms of mineral composition, texture distribution and the like.

[0092] Further, in the iterative training process, various optimization strategies are employed, such as adjustment based on adversarial loss function, gradient descent optimization or regularization method, to dynamically optimize the generation ability of the second module, so that it can more accurately reproduce the real mineral composition characteristics of the target area core. In addition, by introducing data enhancement strategies (such as random noise disturbance, illumination change or small-scale morphological transformation), the generalization ability of the model can be further improved to ensure that it can still generate high-quality synthetic images when facing different types of core data.

[0093] The first historical core data further includes corresponding prompt words.

[0094] In some embodiments, during the training of the preset generation model, the method can further include the following.

[0095] First, adjust the prompt words in the corresponding first historical core data to obtain a corresponding second historical core data set. Each second historical core data set includes multiple second historical core data, and the multiple second historical core data in the same set at least include one historical core data containing a positive prompt word and one historical core data containing a negative prompt word, and the multiple second historical core data in the same set correspond to the same first historical core image.

[0096] The positive prompt word can include a prompt word indicating a high-quality core image, such as high resolution, high precision, etc. The negative prompt word can include a prompt word indicating a low-quality core image, such as low resolution, low precision, etc.

[0097] Then, the preset generation model can be called to process the multiple second historical core data in each second historical core data set respectively to obtain a corresponding multiple image sets. Each image set corresponds to a second historical core data set and contains multiple core images corresponding to the multiple second historical core data in the second historical core data set. From each image set, core images corresponding to the positive prompt word and meeting the quality requirements, and core images corresponding to the negative prompt word and not meeting the quality requirements are selected and combined, and corresponding labels are set to obtain multiple sample image sets.

[0098] Finally, the preset generation model is trained using the multiple sample image sets, which continuously guides and trains the preset generation model to optimize learning in the direction of the positive prompt word, while avoiding degradation learning in the direction of the negative prompt word, so that a preset generation model suitable for the core image generation scene, with relatively higher precision and relatively better effect, can be obtained.

[0099] In some embodiments, the method for obtaining core description data of the target area may further include the following:

[0100] S1: Obtain the first description data of the target region;

[0101] S2: Using a preset large language model, the first description data is extracted and processed to obtain the core description data.

[0102] Specifically, the process begins by collecting primary descriptive data from the target area. This data can include geological survey reports, core collection records, and related field observations, covering mineral composition, pore structure, stratigraphy, and other key core features. Subsequently, a pre-defined large language model is used to process this primary descriptive data. This model extracts the core features of the core descriptive data through natural language processing and semantic analysis, converting scattered text information into structured data, such as mineral proportions, porosity values, color, and texture features. This provides accurate geological parameter support for subsequent core image generation and analysis.

[0103] Specifically, firstly, primary descriptive data is collected from the target area. This data may include geological survey reports, core collection records, and field observation notes. This information details the mineral composition, pore structure, stratigraphy, fracture conditions, and other key core characteristics. For example, a survey report might record that in a certain area, quartz accounts for approximately 40%, feldspar approximately 35%, and clay minerals approximately 25% of the core, indicating a porosity between 12% and 15% and a stratigraphy of upper sandstone. Subsequently, this primary descriptive data is processed using a pre-defined large language model. This model extracts the core descriptive features from a large amount of scattered text through natural language processing and semantic analysis, converting unstructured text into structured data. Specifically, the model identifies key terms and numerical information, such as "mineral composition," "porosity," and "stratigraphy," and then automatically organizes parameters such as mineral proportions, porosity values, color, and texture characteristics. For example, if the report states that "the core is mainly composed of quartz, feldspar, and clay minerals, with a high quartz content and a porosity of approximately 12%", the pre-defined large-scale language model will organize these descriptions as follows: quartz content approximately 40%-45%, feldspar content approximately 30%-35%, clay content approximately 20%-25%, and porosity approximately 12%, while also indicating the corresponding stratigraphic position and texture characteristics of the core. In this way, the system transforms the initial descriptive data into a set of structured core description data, which provides accurate and systematic foundational information for subsequent core image generation, quality control, and geological parameter analysis. Ultimately, this process not only improves the efficiency of data acquisition and processing but also ensures that the generated model is based on real and accurate geological parameters in image generation and subsequent analysis.

[0104] In some embodiments, the method, when implemented, can further include the following:

[0105] In the iterative training of the second module of the preset generation model, the mineral composition ratio parameters of the second historical core image and the third core image are determined to determine the preset prompt word construction rule.

[0106] Specifically, in the iterative training process of the second module of the preset generation model, the mineral composition ratio parameters in the second historical core image and the third core image are compared and statistically analyzed in depth, the distribution deviation of each mineral component in different images is calculated, and an optimization algorithm (such as a color histogram matching or mean square error loss function based strategy) is used to dynamically adjust the model parameters, so as to determine a set of preset prompt word construction rules. The preset prompt word construction rule clearly specifies the ideal proportion range of each mineral component (such as calcium, quartz, feldspar, etc.), for example, if the statistical result shows that the content of calcium in the target area should be controlled between 0.5 and 0.6, the prompt word will contain this numerical requirement, so as to always maintain the consistency of the mineral proportion in the image generation process.

[0107] When implemented, the preset prompt word construction rule can also be constructed in the following manner: determining the mineral component ratio difference of the mineral component ratio of the corresponding third core image and the mineral component ratio of the first historical core image; according to the mineral component ratio difference, screening the images with a mineral component ratio difference less than a preset ratio difference threshold from the plurality of first historical core images as reference historical core images; extracting the corresponding prompt words from the first historical core data corresponding to the reference historical core images as reference prompt words; clustering the reference prompt words to determine the common features of the reference prompt words; and constructing the preset prompt word construction rule according to the common features.

[0108] Thus, the preset prompt word construction rule suitable for the core image generation scene and adapted to the preset generation model can be automatically constructed.

[0109] In this way, based on the preset prompt word construction rule obtained by the iterative training of the second module, not only the accuracy and stability of the generated image in the mineral composition ratio are improved, but also the real geological information in the historical data can be effectively transmitted to the generation process, so that the generated core image is closer to the actual core characteristics.

[0110] In some embodiments, the second module of the preset generation model is obtained by iteratively training the second module of the preset generation model according to the second historical core image and the third core image, and the method, when implemented, can further include the following:

[0111] S1: determining a first loss value according to a mineral composition ratio of the second historical core image and a mineral composition ratio of the third core image;

[0112] S2: determining a second loss value according to a resolution of the second historical core image and a resolution of the third core image;

[0113] S3: iteratively training the second module of the preset generation model according to the first loss value and the second loss value to obtain the second module of the preset generation model.

[0114] The first loss value measures the difference between the generated image and the real core image in the mineral composition ratio, and can be calculated by using mean squared error (MSE) or KL divergence (Kullback-Leibler Divergence).

[0115] The second loss value measures the difference between the generated image and the real core image in the resolution, and common methods can include structural similarity (SSIM) and perceptual loss.

[0116] Specifically, first, the deviation of the generated third core image and the corresponding second historical core image in the mineral composition ratio is calculated to obtain the first loss value, so as to measure the accuracy of the generated image in the mineral composition. At the same time, the difference between the two in the resolution is calculated to obtain the second loss value, so as to ensure that the image quality meets the requirements of fine analysis. Subsequently, the first loss value and the second loss value are used together for model optimization to guide the second module to adjust the generation strategy in subsequent training, so that it can more accurately reproduce the core characteristics of the target area. Through multiple rounds of iterative training, the model gradually improves the optimization ability of the mineral composition ratio and the resolution of the core image.

[0117] For example, during training, assume that in the mineral composition ratio of the second historical core image, the proportion of quartz is 40%, the proportion of feldspar is 35%, and the proportion of clay minerals is 25%. However, in the third core image generated by the second module of the preset generation model, the proportion of quartz is 45%, the proportion of feldspar is 30%, and the proportion of clay minerals is 25%, which deviates from the true data. Therefore, when calculating the first loss value, the difference in mineral composition ratio between the two is used as a constraint to guide the model to adjust the distribution of mineral components in subsequent iterative training, so that it is closer to the real core data. At the same time, in terms of resolution, assume that the resolution of the second historical core image is 512x512 pixels, while the resolution of the generated third core image is 480x480, resulting in the loss of some core structure details. At this time, the second loss value is calculated, and the resolution error is used as an optimization target to make the model improve the image clarity and detail retention ability in subsequent training. By continuously adjusting the model parameters and combining multiple third core images with the second historical core image for combined training, the core image generated by the second module is eventually closer to the actual situation in terms of mineral composition and resolution.

[0118] In some embodiments, the second module of the preset generation model is iteratively trained according to the first loss value and the second loss value to obtain the second module of the preset generation model. In specific implementation, it can include:

[0119] The first loss value and the second loss value are integrated according to a preset weighting strategy to obtain a final loss function. During training, through optimization algorithms such as backpropagation and gradient descent, model parameters are continuously adjusted, so that the generated core image gradually approaches the real core data in terms of mineral composition and resolution.

[0120] Specifically, the second module of the preset generation model is continuously optimized through iterative training, so that the generated core image gradually approaches the real data in terms of mineral composition and resolution. First, a first loss value is calculated, which measures the difference between the generated image and the real core image in terms of the proportion of minerals (for example, the proportion of quartz, feldspar, clay and other minerals in the generated image is compared using mean square error), and a second loss value is calculated, which reflects the similarity between the generated image and the real image in terms of resolution and structural details (for example, the structural similarity index SSIM is used for evaluation). Subsequently, the two loss values are integrated according to a preset weighting strategy (for example, the weights are 0.7 and 0.3 respectively), to obtain the final loss function. During the training process, the parameters of the second module are continuously adjusted through optimization algorithms such as back propagation and gradient descent. For example, assuming that in a certain iteration, the proportion of quartz in the generated image is 50%, while the real core image is 45%, and the SSIM value is 0.85 (the target is 0.95), then the integrated loss function will respond to these two deviations, prompting the model to adjust the parameters in the next iteration, so that the proportion of quartz gradually decreases to about 45%, while the image clarity and structural consistency are improved. Through multiple rounds of iterative training, the preset generation model gradually converges, and the final generated core image can achieve a high degree of consistency with the real core data in terms of mineral composition and image quality.

[0121] In some embodiments, when the method is implemented, it can also include the following content:

[0122] The mineral distribution of the second historical core image is obtained by performing distribution extraction processing on the second historical core image using a preset extraction model.

[0123] Specifically, the distribution extraction processing of the second historical core image using the preset extraction model can effectively identify and quantify the mineral distribution in the rock sample. The preset extraction model is constructed based on a deep learning algorithm, for example, a convolutional neural network (CNN) or a variational autoencoder (VAE), which is trained on a large-scale core data set to have the ability to accurately extract mineral distribution features. In specific implementation, the model can perform layered analysis on the input core image, extract the mineral types, spatial distribution and microstructure information of the rock, and convert them into numerical mineral distribution data in a high-precision manner.

[0124] For example, the preset extraction model can identify that the proportion of quartz is 45%, the proportion of feldspar is 30%, and the proportion of calcite is 25%, and further mark the spatial distribution of the minerals. In the upper left corner area of the image, the content of quartz is high, and in the lower right corner area, calcite is more concentrated. In order to improve the identification accuracy, the preset extraction model can also combine a variational autoencoder (VAE) to optimize the extraction result through a reconstruction loss, so that the preset extraction model is more accurate in identifying the boundaries of mineral distribution. In addition, multi-modal learning combined with spectral analysis data (such as XRD or XRF data) can further enhance the ability to distinguish mineral types.

[0125] As can be seen from the above, the core image generation model-based mining strategy determination method provided by the embodiments of the present specification acquires core description data of a target area and image quality requirement information corresponding to the target area; wherein the core description data at least includes a mineral composition proportion prompt word, and the mineral composition proportion prompt word is generated according to a preset prompt word construction rule; a first module of a preset generation model is used to generate a first core image according to the core description data and the image quality requirement information; wherein the first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and a corresponding first historical core image; a second module of the preset generation model is used to adjust and process the first core image to obtain a second core image; wherein the second module is obtained by training according to second historical core data, a first historical generated image, and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by the first module according to the second historical core data; and a mining strategy for the target area is determined according to the second core image. In this way, first, the first core image is efficiently generated by using the first module of the preset generation model to train the module based on the diffusion algorithm according to the core description data and the image quality requirement information of the target area in combination with the first historical core data and the corresponding first historical core image, thereby effectively reducing the high cost and low scanning efficiency problems of traditional high-precision imaging; second, the second core image more accurately reflecting the core structure characteristics of the target area is obtained by using the second module of the preset generation model to train the model by combining the second historical core data, the first historical generated image, and the corresponding second historical core image to optimize and fine-tune the first core image, thereby overcoming the deficiencies of existing digital core technology in terms of sample coverage, and providing fine and comprehensive data support for formulating a scientific and reasonable mining strategy.

[0126] Referring to Figure 2As shown, the embodiment of the present specification further provides a specific electronic device, wherein the electronic device comprises a network communication port 201, a processor 202 and a memory 203, and the above structures are connected through internal cables so that each structure can perform specific data interaction.

[0127] The network communication port 201 can be specifically used to obtain core description data of a target area and image quality requirement information corresponding to the target area, wherein the core description data at least comprises a mineral composition ratio prompt word, and the mineral composition ratio prompt word is generated according to a preset prompt word construction rule.

[0128] The processor 202 can be specifically used to generate a first core image according to the core description data and the image quality requirement information by using a first module of a preset generation model, wherein the first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and a corresponding first historical core image; adjust the first core image by using a second module of the preset generation model to obtain a second core image, wherein the second module is obtained by training according to second historical core data, a first historical generated image and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by using the first module according to the second historical core data; and determine a mining strategy for the target area according to the second core image.

[0129] The memory 203 can be specifically used to store corresponding instruction programs.

[0130] Based on the above method, the related structure performance of the electronic device can be effectively utilized, the data processing speed of the electronic device can be improved, and the mining strategy determination method based on the core image generation model can be efficiently realized.

[0131] In the present embodiment, the network communication port 201 can be a virtual port that can send or receive different data by binding with different communication protocols. For example, the network communication port can be a port responsible for web data communication, can be a port responsible for FTP data communication, and can be a port responsible for mail data communication. In addition, the network communication port can also be an entity communication interface or a communication chip. For example, it can be a wireless mobile network communication chip such as GSM, CDMA, etc.; it can also be a Wifi chip; and it can also be a Bluetooth chip.

[0132] In the present embodiment, the processor 202 can be implemented in any suitable manner. For example, the processor can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code (e.g. software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, etc. The present specification does not limit the form of the processor.

[0133] In the present embodiment, the memory 203 can comprise a hierarchy, and in a digital system, anything that can hold binary data is a storage medium; in an integrated circuit, a circuit that has no physical form and has a storage function is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, a TF card, etc.

[0134] The embodiment of the present specification also provides a computer readable storage medium based on the above-mentioned core image generation model based mining strategy determination method, core description data of a target area and image quality requirement information corresponding to the target area are obtained; wherein the core description data at least includes a mineral composition ratio prompt word, the mineral composition ratio prompt word is generated according to a preset prompt word construction rule; a first module of a preset generation model is used to generate a first core image according to the core description data and the image quality requirement information; wherein the first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and a corresponding first historical core image; a second module of the preset generation model is used to adjust and process the first core image to obtain a second core image; wherein the second module is obtained by training according to second historical core data, a first historical generated image and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by the first module according to the second historical core data; a mining strategy for the target area is determined according to the second core image.

[0135] In the present embodiment, the above-mentioned storage medium includes but is not limited to a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk drive (HDD) or a memory card. The memory can be used to store computer program instructions. The network communication unit can be an interface set according to the standard of the communication protocol, used for network connection communication.

[0136] In the embodiment, the functions and effects realized by the program instructions stored in the computer readable storage medium can be explained in comparison with other embodiments, and will not be described here.

[0137] Referring to Figure 3 At the software level, the embodiment of the present specification also provides a core image model-based mining strategy determination device, which can specifically include the following structure modules:

[0138] The data acquisition module 301 is configured to acquire core description data of a target area and image quality requirement information corresponding to the target area, wherein the core description data at least includes a mineral composition ratio prompt word, and the mineral composition ratio prompt word is generated according to a preset prompt word construction rule.

[0139] The first image generation module 302 is configured to generate a first core image according to the core description data and the image quality requirement information by using a first module of a preset generation model, wherein the first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and a corresponding first historical core image.

[0140] The second image generation module 303 is configured to adjust and process the first core image to obtain a second core image by using a second module of the preset generation model, wherein the second module is obtained by training according to second historical core data, a first historical generated image, and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by using the first module according to the second historical core data.

[0141] The strategy determination module 304 is configured to determine a mining strategy for the target area according to the second core image.

[0142] In some embodiments, the first image generation module 302 described above, when implemented, uses the first module of the preset generation model to respectively perform feature extraction processing on the core description data and the image quality requirement information to obtain a first feature vector corresponding to the core description data and a second feature vector corresponding to the image quality requirement information; determines a conditional embedding vector according to the first feature vector, the second feature vector, and a preset random noise; and generates the first core image according to the conditional embedding vector by using the first module of the preset generation model.

[0143] In some embodiments, before the first image generation module 302 described above, in specific implementation, the first historical core data and the first historical core image are obtained; the first module of the preset generation model is used to determine a second historical generated image according to the first historical core data; the first module of the preset generation model is iteratively trained according to the first historical core image and the second historical generated image, to obtain the first module of the preset generation model; the second historical core data and a second historical core image are obtained, and the data quantity of the first historical core data is greater than that of the second historical core data; the first module of the preset generation model is used to determine the first historical generated image according to the second historical core data; the second module of the preset generation model is used to adjust and process the mineral color proportion of the first historical generated image to obtain a third core image; the second module iterative module is used to iteratively train the second module of the preset generation model according to the second historical core image and the third core image, to obtain the second module of the preset generation model.

[0144] In some embodiments, the data acquisition module 301 described above, in specific implementation, acquires first description data of the target area; a preset large language model is used to extract and process the first description data to obtain the core description data.

[0145] In some embodiments, in specific implementation, in the iterative training of the second module of the preset generation model, the preset prompt word construction rule is determined according to the mineral composition proportion parameters of the second historical core image and the third core image.

[0146] In some embodiments, the second module iterative module described above, in specific implementation, determines a first loss value according to the mineral composition proportion of the second historical core image and the mineral composition proportion of the third core image; determines a second loss value according to the resolution of the second historical core image and the resolution of the third core image; iteratively trains the second module of the preset generation model according to the first loss value and the second loss value, to obtain the second module of the preset generation model.

[0147] In some embodiments, in specific implementation, the mineral distribution of the second historical core image is obtained by using a preset extraction model to extract and process the distribution of the second historical core image; the preset extraction model is a model constructed based on a preset deep learning algorithm.

[0148] It should be noted that the units, devices or modules and the like illustrated in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described as various modules with functions. Of course, in the implementation of the present specification, the functions of each module can be implemented in the same software and / or hardware, or the modules implementing the same function can be implemented by a combination of sub-modules or sub-units, etc. The above described device embodiments are only illustrative, for example, the division of the units is only a logical functional division, and in actual implementation, there can be another division method, for example, the units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the coupling or direct coupling or communication connection between the units or devices shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0149] As can be seen from the above, based on the core image generation model provided by the embodiment of the present specification, first, by using the first module of the preset generation model, according to the core description data of the target area and the image quality requirement information, the first historical core data and the corresponding first historical core image are combined to train the module based on the diffusion algorithm, so as to efficiently generate the first core image, effectively reducing the problem of high cost and low scanning efficiency of traditional high-precision imaging; secondly, by using the second module of the preset generation model, the model is trained by combining the second historical core data, the first historical generated image and the corresponding second historical core image, the first core image is fine-tuned, and a second core image more accurately reflecting the core structure characteristics of the target area is obtained, overcoming the deficiencies of existing digital core technology in sample coverage and the like, and providing fine and comprehensive data support for formulating a scientific and reasonable exploitation strategy.

[0150] In one specific scenario example, the core image generation model-based mining strategy determination method and device provided in the present specification can be applied. First, by using the first module of the preset generation model, the core description data and image quality requirement information of the target area are combined with the first historical core data and the corresponding first historical core image to train the diffusion algorithm-based module, thereby efficiently generating the first core image and effectively reducing the high cost and low scanning efficiency problems of traditional high-precision imaging. Second, by using the second module of the preset generation model, the second historical core data, the first historical generated image and the corresponding second historical core image are combined to train the model, and the first core image is optimized and fine-tuned to obtain a second core image that more accurately reflects the core structure characteristics of the target area, overcoming the deficiencies of existing digital core technology in sample coverage and providing fine and comprehensive data support for formulating a scientific and reasonable mining strategy. The specific implementation process can include the following.

[0151] S1: Setting and optimization of Stable Diffusion (SD) model.

[0152] First, deploy and configure Stable Diffusion (SD), and apply diffusion model technology to core image generation tasks. To improve image generation quality, use tiled diffusion variational autoencoder (VAE) for data compression and decoding to reduce information loss and improve model generation accuracy. In addition, RealisticVision V2.0 is selected as the base model, which performs well in image generation quality, detail restoration ability and stability.

[0153] To meet the special needs of core images, further use Low-Rank Adaptation (LoRA) for model training. LoRA can efficiently adjust parameters without affecting the original capabilities of the base model, enabling the model to have specific capabilities for core image generation. During training, a large number of real core images and their corresponding mineral composition, porosity, permeability and other label data are input, and LoRA adapter is trained to learn the structure characteristics of core images through supervised learning, improving its adaptability in different lithology conditions.

[0154] The key parameters of the above LoRA training are as follows: Rank: controls the degrees of freedom of the adapter to balance the computational efficiency and fitting ability. Learning Rate: used to optimize the parameter update speed to avoid overfitting or undertraining. Training Steps: ensures that the adapter can fully learn the core image features while avoiding the model from overfitting to a specific data set. For details, see Table 1.

[0155] Table 1

[0156]

[0157]

[0158] S2: Text encoding and conditional embedding generation

[0159] In the generation process of core images, first, the core descriptive text is taken as input, such as mineral composition, pore characteristics, formation depth, lithology type, etc., and a Stable Diffusion pre-trained text encoder (such as Contrastive Language-Image Pre-training, CLIP) is used for text processing. The text encoder will convert the input text into a fixed-length vector representation, which captures the key information of the core features, so that the model can generate matching core images based on these information.

[0160] On this basis, Stable Diffusion generates conditional embedding (Conditional Embedding) through random noise combined with text vectors, and uses it as a guidance signal for diffusion model to generate images. Conditional embedding fuses text information with noise, so that the generated image not only meets the physical constraints of the text description, but also has a certain randomness to ensure the diversity and authenticity of the generated results.

[0161] For example, the input text is: The core sample comes from a sandstone layer at a depth of 3200m, the main minerals are quartz (70%), feldspar (20%), and clay minerals (10%), the porosity is 15%, and the permeability is 120mD. The text encoder can take the key features and convert them into vector form. Combined with random noise to generate conditional embedding, so that the diffusion process is constrained by the text content.

[0162] S3: Iterative diffusion and final image generation.

[0163] Throughout the iteration process, the diffusion model gradually decodes the conditional embedding, i.e., gradually optimizes the initial random noise to an image consistent with the core description information through the denoising step. The core of the diffusion process is to gradually remove noise and enhance the target features in each iteration, and finally generate high-quality core images.

[0164] In some embodiments, the method can further include, when implemented:

[0165] S1: Load a large number (2000 groups) of core images and their corresponding label data in Stable Diffusion (SD), and train the pre-trained model using LoRA (Low Rank Adapter) technology.

[0166] S2: Use descriptive text as input, such as "mineral A accounts for 0.00072800 of the entire image", to guide the model to draw specific mineral proportions in the image.

[0167] Specifically, when a series of negative prompt words (such as "low quality", "low resolution", "worst quality") are added during the generation process, it can avoid producing blurred or undesirable results, thereby ensuring the quality and consistency of the output image.

[0168] Referring to Figure 4 The left side Original image represents the original image, showing a variety of mineral particles and pore space; the middle Prompts+Trained SD represents the process of generating simulated core images under the joint action of text prompts (Prompts) and LoRA-trained SD model; and the right side three Generated images are generated images obtained using different random seeds or generation parameters, where different color blocks represent mineral types or component proportions, and their distribution has a high similarity in color proportion and mineral distribution with the original image.

[0169] S3: Compare the color proportion in the original core image with the color proportion in the generated image.

[0170] Referring to Figure 5 The color mineral proportion comparison between the original core image and the generated image is shown, which is used to evaluate the accuracy and consistency of the generated model in the distribution of mineral components.

[0171] Specifically, Figure 5The proportions of different colors of minerals in the original and generated images of the three core samples (Images 1-3) are visually compared through bar charts. Each subgraph shows two sets of bar charts: the left side is the mineral proportion of the original image, and the right side is the generated image result. The mineral types are distinguished by color and letter code, and the proportion values are marked at the top of the bar chart.

[0172] The minerals in the legend and their corresponding English and Chinese explanations are as follows: Quartz (A) - Quartz, Albite (B) - Sodium Feldspar, Organic (C) - Organic Matter, Calcite (D) - Calcite, Apatite (E) - Apatite, Chlorite (F) - Chlorite, Unknown (G) - Unknown Mineral, Pyrite (H) - Pyrite, Illite (I) - Illite, Rutile (J) - Rutile, Dolomite (K) - Dolomite, Pores (L) - Pores.

[0173] The comparison results show that when the proportion of the most important color (corresponding to mineral C, sodium feldspar) in the generated image is set between 0.5 and 0.6, the color distribution of the generated image has high consistency with the original core image. However, when this proportion is adjusted to 0.4 or 0.7, the color proportion between the generated image and the original image begins to deviate significantly. This phenomenon may be closely related to the distribution of the color proportion in the training data set.

[0174] Therefore, only when the color proportion distribution in the test data set is fully covered by the training data set, can the accuracy of the generated image in mineral proportion be ensured. This research result further shows that in the training process, the diversity and coverage of data play a crucial role in the accuracy of the generated model, and optimizing the distribution of training data can effectively improve the adaptability of the model to different core samples.

[0175] Based on the above embodiments, by providing a text-to-image method, mainly applied to the generation of core pictures, a powerful tool is provided for digital core analysis of reservoir conditions. Because in rock physical property analysis, the results of text-to-image can be used as samples, effectively avoiding the problem caused by high scanning cost.

[0176] Although the present specification provides method operational steps in the order in which the steps are performed, the order of the steps can be changed based on the underlying logic of the method. The steps of the embodiments recited in the claims can be executed in any order that is practicable and / or desirable. The order of the steps can be varied in actual implementation of the method. The steps recited in the embodiments or the figures can be executed in parallel or in series (for example, in a parallel processor or multi-threaded processing environment, or even in a distributed data processing environment). The terms "comprises", "comprising", or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical or similar elements in the process, method, article, or apparatus that comprises the element. The terms "first", "second", and the like, define names of particular elements, and do not necessarily limit the elements to these two. The terms "including", "containing", or any other similar terms are intended to be inclusive in a manner similar to the term "comprising", as constituting a non- limiting inclusion.

[0177] Those skilled in the art will also appreciate that, in addition to being embodied in a purely computer readable program code manner, the controller can be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc. to achieve the same functions by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the means for achieving various functions included therein can also be considered as structures within the hardware component. Alternatively, the means for achieving various functions can be considered as both a software module implementing a method and a structure within a hardware component.

[0178] From the above description of the embodiments, those skilled in the art can clearly understand that the present specification can be implemented by means of software in conjunction with a necessary general hardware platform. Based on such an understanding, the technical solutions of the present specification can essentially be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0179] Although the present specification is described by means of embodiments, those skilled in the art will know that there are many modifications and variations of the present specification without departing from the spirit of the present specification, and it is intended that the appended claims encompass these modifications and variations without departing from the spirit of the present specification.

Claims

1. A method for determining a production strategy based on a model generated from a core image, characterized by, The method comprises the following steps: obtaining core description data of a target area and image quality requirement information corresponding to the target area; wherein the core description data at least comprises a mineral composition proportion prompt word, and the mineral composition proportion prompt word is generated according to a preset prompt word construction rule; generating a first core image according to the core description data and the image quality requirement information by using a first module of a preset generation model; wherein the first module is obtained by training a module constructed based on a diffusion algorithm according to first historical core data and a corresponding first historical core image; adjusting and processing the first core image by using a second module of the preset generation model to obtain a second core image; wherein the second module is obtained by training according to second historical core data, a first historical generated image and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by using the first module according to the second historical core data; determining a mining strategy for the target area according to the second core image; wherein the first module of the preset generation model generates the first core image according to the core description data and the image quality requirement information, comprising: the first module of the preset generation model respectively performs feature extraction processing on the core description data and the image quality requirement information to obtain a first feature vector corresponding to the core description data and a second feature vector corresponding to the image quality requirement information; determining a conditional embedding vector according to the first feature vector, the second feature vector and a preset random noise; the first module of the preset generation model generates the first core image according to the conditional embedding vector.

2. The method of claim 1, wherein, Before the first module of the preset generation model generates the first core image according to the core description data and the image quality requirement information, the method further comprises the following steps: obtaining the first historical core data and the first historical core image; determining a second historical generated image according to the first historical core data by using the first module of the preset generation model; iteratively training the first module of the preset generation model according to the first historical core image and the second historical generated image to obtain the first module of the preset generation model; obtaining the second historical core data and a second historical core image, and the data amount of the first historical core data is greater than the data amount of the second historical core data; determining the first historical generated image according to the second historical core data by using the first module of the preset generation model; adjusting and processing the mineral color proportion of the first historical generated image by using the second module of the preset generation model to obtain a third core image; iteratively training the second module of the preset generation model according to the second historical core image and the third core image to obtain the second module of the preset generation model.

3. The method of claim 2, wherein, The method for obtaining the core description data of the target area comprises the following steps: Acquire first description data of the target area; Utilize a preset large language model to perform extraction processing on the first description data to obtain the core description data.

4. The method of claim 3, wherein, The method further comprises: In the iterative training of the second module of the preset generation model, the preset prompt word construction rule is determined according to the mineral composition ratio parameters of the second historical core image and the third core image.

5. The method of claim 4, wherein, The iterative training of the second module of the preset generation model according to the second historical core image and the third core image comprises: Determine a first loss value according to the mineral composition ratio of the second historical core image and the mineral composition ratio of the third core image; Determine a second loss value according to the resolution of the second historical core image and the resolution of the third core image; Iteratively train the second module of the preset generation model according to the first loss value and the second loss value to obtain the second module of the preset generation model.

6. The method of claim 5, wherein, The mineral distribution of the second historical core image is obtained by performing distribution extraction processing on the second historical core image using a preset extraction model.

7. A core image-based model generation mining strategy determination device, characterized in that, Comprise: The data acquisition module is used for acquiring core description data of a target area and image quality requirement information corresponding to the target area; wherein the core description data at least includes a mineral composition ratio prompt word, and the mineral composition ratio prompt word is generated according to a preset prompt word construction rule; The first image generation module is used for generating a first core image according to the core description data and the image quality requirement information by using a first module of a preset generation model; wherein the first module is obtained by training a module based on a diffusion algorithm according to first historical core data and a corresponding first historical core image; The second image generation module is used for adjusting the first core image by using a second module of the preset generation model to obtain a second core image; wherein the second module is obtained by training according to second historical core data, a first historical generated image and a corresponding second historical core image, the region type corresponding to the second historical core data is the same as the region type corresponding to the target area, and the first historical generated image is a core image generated by using the first module according to the second historical core data; The strategy determination module is used for determining a mining strategy for the target area according to the second core image. The first module of the preset generation model is used for performing feature extraction processing on the core description data and the image quality requirement information respectively to obtain a first feature vector corresponding to the core description data and a second feature vector corresponding to the image quality requirement information. ​ determining a conditional embedding vector according to the first feature vector, the second feature vector, and a preset random noise; generating the first core image according to the conditional embedding vector by using a first module of the preset generation model.

8. An electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the steps of the method for determining an exploitation strategy based on a core image generation model according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A computer instruction is stored thereon, and the instruction is executed by a processor to implement the steps of the method for determining an exploitation strategy based on a core image generation model according to any one of claims 1 to 6.

Images