Method, system and electronic device for automatic generation of a three-dimensional geological model

Abstract

The present application relates to the technical field of computer modeling, and particularly relates to a three-dimensional geological model automatic generation method, system and electronic equipment, the method comprises the following steps: obtaining three-dimensional geological modeling requirement data and modeling data input by a user, and processing the modeling data; generating a modeling instruction stream through a large model after reinforcement learning, and performing feasibility verification on the generated modeling instruction stream; if the verification result is feasible, the modeling instruction stream is sent to a three-dimensional geological modeling software to generate a three-dimensional geological model; obtaining feedback information of a modeling expert adjusting the three-dimensional geological model and diagnosing and analyzing the modeling instruction stream generated by the large model, and storing the feedback information in a knowledge base; if the verification result is unfeasible, problem matching is performed in the knowledge base, and the result is input to the large model, after analysis, a modeling instruction stream is regenerated to drive the three-dimensional modeling software to model; thus, the present application can automatically generate a three-dimensional geological model.

Description

Technical Field

[0001] This invention relates to the field of computer modeling technology, and in particular to a method, system and electronic device for automatically generating three-dimensional geological models. Background Technology

[0002] With the rapid development of artificial intelligence technology, it is now possible to generate 3D models from multimodal large models. However, the essence of large model technology lies in data training, and the generation process is still in a black box mode. The illusion of large models and knowledge cutoff problems cannot be avoided. In the field of model building, especially in the process of building irregular models such as 3D geological models, large models cannot be directly generated, and modeling still depends on manual labor. Summary of the Invention

[0003] The purpose of this invention is to provide a method, system, and electronic device for automatically generating three-dimensional geological models, which can solve the problem that existing model building technologies cannot automatically generate three-dimensional geological models.

[0004] To address the aforementioned technical problems, this invention provides a method for automatically generating three-dimensional geological models. The method includes: acquiring user-inputted three-dimensional geological modeling requirement data and modeling data, and processing the modeling data; inputting the three-dimensional geological modeling requirement data into a large model after reinforcement learning, so as to generate a modeling instruction stream through the large model.

[0005] The feasibility of the generated modeling instruction stream is verified using the large model. If the verification result is feasible, the modeling instruction stream generated by the large model is sent to the 3D geological modeling software, so that the 3D geological modeling software can generate the corresponding 3D geological model by calling modeling instructions and processing modeling data according to the modeling instruction stream. Feedback information from modeling experts on adjusting the 3D geological model and diagnosing and analyzing the modeling instruction stream generated by the large model is obtained and stored in a pre-created knowledge base. If the verification result is infeasible, problem matching is performed in a pre-created vector database based on problem-to-problem reverse index matching technology. The problem matching result is input into the large model, so that the large model can regenerate the modeling instruction stream according to the problem matching result.

[0006] Optionally, processing the modeling data includes: using optical character recognition technology to identify the modeling data, thereby identifying valid information and outputting the corresponding Excel format file.

[0007] Optionally, verifying the feasibility of the generated modeling instruction stream using the large model includes: using the large model to obtain the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream; based on the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream, obtaining the feasibility parameter value corresponding to the modeling instruction stream; if the feasibility parameter value corresponding to the modeling instruction stream is greater than or equal to a preset feasibility threshold, then the modeling instruction stream is determined to be feasible, otherwise the modeling instruction stream is determined to be infeasible.

[0008] Optionally, the fidelity evaluation value corresponding to the generated modeling instruction stream can be obtained using the following formula:

[0009]

[0010] In the formula, This is a loyalty assessment value, and its value ranges from [value range missing]. between;

[0011] The modeling instruction relevance evaluation value corresponding to the generated modeling instruction stream is obtained using the following formula:

[0012]

[0013] In the formula, To evaluate the relevance of modeling instructions, The total number of modeling instructions in the generated modeling instruction stream. To meet the user's modeling needs, This refers to the modeling instructions in the generated modeling instruction stream, whose values ​​range from... Between, sim This represents the correlation value between the user's modeling requirements and the modeling instructions in the generated modeling instruction stream;

[0014] The context relevance evaluation value corresponding to the generated modeling instruction stream is obtained using the following formula:

[0015]

[0016] In the formula, This is a context-dependent evaluation value, and its range is within... between.

[0017] Optionally, obtaining the feasibility parameter values ​​corresponding to the modeling instruction stream based on the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream includes:

[0018] The feasibility parameter values ​​corresponding to the modeling instruction flow are obtained using the following formula:

[0019]

[0020] in, The total number of times the modeling instruction flow is generated for the large model. As the loyalty weight value, To model the relevance weight values ​​of instructions, This is the context-related weight value. ,and .

[0021] Optionally, the method further includes: if the feasibility parameter value corresponding to the modeling instruction is less than or equal to a preset infeasibility threshold, then stop using the large model to continue generating the modeling instruction stream, and output a prompt message for supplementing modeling data.

[0022] Optionally, the method further includes: counting the number of irrelevant questions corresponding to the feedback information accumulated in the knowledge base; if the number of irrelevant questions is greater than or equal to a preset threshold for the number of irrelevant questions, then updating the large model based on the feedback information accumulated in the knowledge base.

[0023] Optionally, the step of counting the number of irrelevant questions corresponding to the feedback information accumulated in the knowledge base includes: dividing the feedback information accumulated in the knowledge base into multiple groups of documents, wherein each group of documents contains multiple groups of feedback questions and solutions; for each group of documents, calculating the correlation value between different groups of feedback questions and solutions in that document, and considering feedback questions in different groups of feedback questions and solutions with a correlation value greater than or equal to a preset threshold as relevant questions, and vice versa, as irrelevant questions, thereby obtaining the number of irrelevant questions corresponding to that document; and obtaining the number of irrelevant questions corresponding to the feedback information accumulated in the knowledge base based on the sum of the number of irrelevant questions corresponding to each group of documents.

[0024] To address the aforementioned technical problems, this invention also provides a method and system for automatically generating three-dimensional geological models, comprising: a requirement input module configured to acquire three-dimensional geological modeling requirement data input by a user; a data processing module configured to acquire modeling data input by the user and process the modeling data; an instruction generation module configured to input the three-dimensional geological modeling requirement data into a large model after reinforcement learning, so as to generate a modeling instruction stream through the large model; a verification module configured to verify the feasibility of the generated modeling instruction stream using the large model; and a model generation module configured to, when the feasibility verification result of the modeling instruction stream is feasible, send the modeling instruction stream generated by the large model to the three-dimensional geological model. The modeling software enables the 3D geological modeling software to generate a corresponding 3D geological model based on the modeling instruction flow, calling modeling instructions and the processed modeling data. When the feasibility verification result of the modeling instruction flow is infeasible, it performs problem matching in a pre-created vector database based on problem-to-problem reverse index matching technology and inputs the problem matching result into the large model, so that the large model can regenerate the modeling instruction flow based on the problem matching result. The adjustment and feedback module is configured to obtain feedback information from modeling experts regarding adjustments to the 3D geological model and diagnosis and analysis of the modeling instruction flow generated by the large model, and to store the feedback information in a pre-created knowledge base.

[0025] To address the aforementioned technical problems, the present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it implements the above-described method for automatically generating three-dimensional geological models.

[0026] Compared with existing technologies, the automatic generation method, system, and electronic device for three-dimensional geological models provided by this invention have the following advantages:

[0027] The present invention provides an automatic three-dimensional geological model generation method, comprising: acquiring user-input three-dimensional geological modeling requirement data and modeling data, and processing the modeling data; inputting the three-dimensional geological modeling requirement data into a large model after reinforcement learning to generate a modeling instruction stream through the large model; verifying the feasibility of the generated modeling instruction stream using the large model; if the verification result is feasible, sending the modeling instruction stream generated by the large model to three-dimensional geological modeling software, so that the three-dimensional geological modeling software can generate a corresponding three-dimensional geological model by calling modeling instructions and processing the modeling data according to the modeling instruction stream; acquiring feedback information from modeling experts on adjusting the three-dimensional geological model and diagnosing and analyzing the modeling instruction stream generated by the large model, and storing the feedback information in a pre-created knowledge base; if the verification result is infeasible, performing problem matching in a pre-created vector database based on problem-to-problem reverse index matching technology; inputting the problem matching result into the large model, so that the large model can regenerate the modeling instruction stream according to the problem matching result.

[0028] Therefore, the automatic generation method for 3D geological models provided by this invention innovatively solves the problems of black-box generation, model illusion, and knowledge solidification in traditional large-scale 3D geological modeling by introducing the Agentic RAG paradigm. At the same time, it continuously optimizes the large-scale model, enabling it to autonomously generate and verify the modeling instruction flow, significantly improving the accuracy and efficiency of instruction generation. In addition, the automatic generation method for 3D geological models provided by this invention uses the modeling instruction flow to drive 3D modeling software, realizing full automation from requirement input to 3D geological model generation, greatly reducing the cost of manual intervention, and achieving efficient generation of complex and irregular 3D geological models.

[0029] The automatic 3D geological model generation system provided by this invention, through a requirement input module, a data processing module, an instruction generation module, a verification module, a model generation module, and an adjustment and feedback module, can completely realize the automatic modeling process based on Agentic RAG. It can automatically process modeling data and autonomously complete the generation of modeling instruction flow and feasibility verification with the help of a large model after reinforcement learning. When verification fails, it can achieve problem matching and instruction flow reconstruction through reverse indexing of vector database. By continuously accumulating expert feedback data into the knowledge base and triggering large model updates, the automatic 3D geological model generation system provided by this invention has the ability to self-evolve, effectively overcoming the problems of large model illusion and time deadline limitations, and significantly improving the automation level and modeling accuracy of 3D geological model generation.

[0030] The electronic device for automatically generating three-dimensional geological models provided by this invention realizes an automatic modeling process based on Agentic RAG by executing the automatic three-dimensional geological model generation method. It can autonomously complete the requirements understanding, modeling instruction flow generation, feasibility verification and three-dimensional geological model construction using a large model optimized by reinforcement learning. Through the built-in vector database, it supports retrieval and continuous learning, effectively solving the problems of existing three-dimensional geological model modeling relying on manual modeling, large model illusion and knowledge update lag. Attached Figure Description

[0031] Figure 1 A flowchart of an automatic three-dimensional geological model generation method provided in one embodiment of the present invention.

[0032] Figure 2 The overall flowchart of the automatic generation method for three-dimensional geological models provided in one embodiment of the present invention is shown.

[0033] Figure 3 This is a flowchart illustrating the generation of a large model reinforcement learning dataset in an automatic three-dimensional geological model generation method provided by an embodiment of the present invention.

[0034] Figure 4 This is a flowchart of the problem matching process in the automatic generation method for three-dimensional geological models provided in one embodiment of the present invention.

[0035] Figure 5 A schematic diagram of the structure of an automatic three-dimensional geological model generation system provided in one embodiment of the present invention.

[0036] Figure 6 A schematic diagram of the structure of an electronic device for automatically generating three-dimensional geological models provided in one embodiment of the present invention.

[0037] The reference numerals in the attached figures are explained as follows:

[0038] 1-Requirement input module, 2-Data processing module, 3-Instruction generation module, 4-Verification module, 5-Model generation module, 6-Adjustment and feedback module, 7-Processor, 8-Memory, 9-Data storage module. Detailed Implementation

[0039] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, further illustrates the automatic generation method, system, and electronic device for three-dimensional geological models proposed in this invention. The advantages and features of this invention will become clearer from the following description. Please refer to the accompanying drawings for a more apparent understanding of the objectives, features, and advantages of this invention.

[0040] The core idea of ​​this invention is to provide a method, system, and electronic device for automatically generating three-dimensional geological models, which can solve the problem that existing model building technologies cannot automatically generate three-dimensional geological models.

[0041] To achieve the above ideas, this invention provides a method for automatically generating three-dimensional geological models. Please refer to [the relevant documentation]. Figure 1 ,like Figure 1 As shown, the present invention provides an automatic three-dimensional geological model generation method, the method comprising: step S100, acquiring user-inputted three-dimensional geological modeling requirement data and modeling data, and processing the modeling data; step S200, inputting the three-dimensional geological modeling requirement data into a large model after reinforcement learning, so as to generate a modeling instruction stream through the large model; step S300, using the large model to verify the feasibility of the generated modeling instruction stream; step S400, if the verification result is feasible, then sending the modeling instruction stream generated by the large model to three-dimensional geological modeling software, so that the three-dimensional geological modeling software can... The modeling instruction flow is used to call modeling instructions and process the modeling data to generate a corresponding 3D geological model; Step S500: Obtain feedback information from modeling experts on adjusting the 3D geological model and diagnosing and analyzing the modeling instruction flow generated by the large model, and store the feedback information in a pre-created knowledge base; Step S600: If the verification result is infeasible, perform problem matching in a pre-created vector database based on problem-to-problem reverse index matching technology; Step S700: Input the problem matching result into the large model so that the large model can regenerate the modeling instruction flow according to the problem matching result.

[0042] Therefore, the automatic generation method for 3D geological models provided by this invention innovatively solves the problems of black-box generation, model illusion, and knowledge solidification in traditional large models in 3D geological modeling by introducing Agentic RAG (Agent-Driven Retrieval Enhanced Generation) and LLM (Large Language Model). At the same time, it continuously optimizes the large model, enabling it to autonomously generate and verify modeling instruction streams, significantly improving the accuracy and efficiency of modeling instruction stream generation. In addition, the automatic generation method for 3D geological models provided by this invention uses modeling instruction streams to drive 3D modeling software, realizing full automation from requirement input to 3D geological model generation, greatly reducing the cost of manual intervention, and achieving efficient generation of complex and irregular 3D geological models.

[0043] For details, please refer to Figure 2 ,like Figure 2As shown, the user first inputs their 3D geological modeling requirements, and then uses a professional large model that has undergone reinforcement learning to generate a modeling command flow. Considering that the large model takes a long time to run, it is necessary to minimize the number of times the large model is used. Therefore, the large model is trained with reinforcement learning to generate a modeling command flow without calling the knowledge base, thus reducing its generation time. The large model after reinforcement learning is then used to verify the feasibility of the generated modeling command flow. Since the large model's reinforcement learning dataset contains both correct and incorrect command flow data, the large model has this capability. Here, two branches are executed depending on whether the modeling command flow is feasible or not.

[0044] The first branch is as follows: If the large model verifies that the modeling instruction flow is feasible and reaches the feasibility threshold, the modeling instruction flow will be sent to the 3D geological modeling software. The 3D geological modeling software will call the modeling instructions and the organized modeling data according to the modeling instruction flow to create the 3D geological model. Then, the modeling experts will adjust the generated 3D geological model and provide feedback on the generated modeling instruction flow. The process is divided according to the amount of feedback information accumulated. If a lot of feedback information is accumulated, the data will be used to fine-tune and update the large model to improve the accuracy of the modeling instruction flow generation of the large model. If a little feedback information is accumulated, it will be directly entered into the vector database.

[0045] The second branch is as follows: If the generated instruction stream is infeasible or has not reached the infeasibility threshold, then question matching is performed based on QulM-RAG (Question Inverted Matching RAG). The large model finds the questions with the highest matching degree based on the relevance of the matched questions, finds the corresponding text blocks and their positions using the inverted index method, and uses the large model to generate a new modeling instruction stream based on the found content. If the generated modeling instruction stream result is ultimately determined to be feasible, then the 3D geological model generation operation continues. If the generated instruction stream is infeasible and has reached the infeasibility threshold, then the system is prompted that it cannot generate this 3D geological model and that supplementary modeling data is required.

[0046] Furthermore, the large model identifies the most relevant matching questions, finds the corresponding text blocks and their positions using an inverted index method, and then generates a new modeling instruction flow based on the found content.

[0047] Specifically, this includes calculating the relevance between the user's modeling requirements and the questions generated by the vector library using the following formula:

[0048]

[0049] in, To meet the user's modeling needs, The problem generated for the vector library, The relevance of user modeling needs to the problems generated by the vector library.

[0050] Furthermore, the processing of the modeling data includes: using optical character recognition technology to identify the modeling data, thereby identifying valid information and outputting the corresponding Excel format file.

[0051] Therefore, by converting the original modeling data into Excel format files, 3D modeling software can directly and efficiently read and parse the modeling parameters automatically without additional data conversion or manual processing, which greatly improves the automation level and overall efficiency of the modeling process.

[0052] Specifically, modeling data includes borehole diagrams, stratigraphic information, and other data used for modeling. After processing, the modeling data is generated into an Excel-formatted modeling data file and stored in a relational database for use by 3D modeling software.

[0053] For further details, please refer to... Figure 3 ,like Figure 3 As shown, the user-input modeling instruction data includes commonly used modeling instructions, modeling instruction streams, modeling parameters, and other data used to train the large model. During processing, the modeling instruction data is first categorized into modeling instruction files, correct modeling instruction stream files, and incorrect modeling instruction stream files. The large model then generates question-and-answer pairs based on these files, and after being organized by modeling experts, forms the large model reinforcement learning dataset. The instruction generation module, based on the Agentic RAG paradigm, uses the large model as a proxy to execute queries and tool calls. The large model reinforcement learning dataset is used to train the large model, enabling it to develop a certain understanding of commonly used instructions and instruction streams, allowing it to autonomously generate modeling instructions and determine whether the instructions are correct and whether a knowledge base call is needed for secondary instruction generation.

[0054] For modeling instruction data that cannot be used for training large models (including newly generated correct modeling instruction stream files and newly generated incorrect modeling instruction stream files, etc.), the RAG technology processing method based on QulM-RAG is used to process it and store it in a vector database. For details, please refer to [link to relevant documentation]. Figure 4 ,like Figure 4 As shown, the modeling instruction data is first cleaned, classified, segmented, and vectorized. Then, a modeling instruction corpus is built. Next, a set of questions is generated based on the corpus content using a large model and RAG technology. Then, the generated questions are segmented and vectorized using an inverted index and stored in a vector database for easy question querying and matching.

[0055] Furthermore, the verification of the feasibility of the generated modeling instruction stream using the large model includes: using the large model to obtain the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream; based on the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream, obtaining the feasibility parameter value corresponding to the modeling instruction stream; if the feasibility parameter value corresponding to the modeling instruction stream is greater than or equal to a preset feasibility threshold, then the modeling instruction stream is determined to be feasible, otherwise the modeling instruction stream is determined to be infeasible.

[0056] Therefore, by constructing a quantitative evaluation based on three dimensions—fidelity, modeling instruction relevance, and context relevance—the feasibility of automating and objectively verifying the generation of modeling instruction flows from large models can be achieved, which can significantly improve the efficiency and accuracy of quality control over modeling instruction flows.

[0057] Furthermore, the fidelity evaluation value corresponding to the generated modeling instruction stream is obtained using the following formula:

[0058]

[0059] In the formula, This is a loyalty assessment value, and its value ranges from [value range missing]. between;

[0060] Furthermore, the modeling instruction relevance evaluation value corresponding to the generated modeling instruction stream is obtained using the following formula:

[0061]

[0062] In the formula, To evaluate the relevance of modeling instructions, The total number of modeling instructions in the generated modeling instruction stream. To meet the user's modeling needs, This refers to the modeling instructions in the generated modeling instruction stream, whose values ​​range from... Between, sim This represents the correlation value between the user's modeling requirements and the modeling instructions in the generated modeling instruction stream. Specifically, the correlation value sim between the user's modeling requirements and the modeling instructions in the generated modeling instruction stream... The calculation formula is as follows:

[0063]

[0064] Furthermore, the context relevance evaluation value corresponding to the generated modeling instruction flow is obtained using the following formula:

[0065]

[0066] In the formula, This is a context-dependent evaluation value, and its range is within... between.

[0067] Therefore, by using the RAGAs (Retrieval-Augmented Generation Assessment System) framework to conduct multi-dimensional quantitative evaluation of the generated modeling instruction stream in terms of fidelity, relevance, and contextual relevance, it is possible to comprehensively measure the factual consistency, demand matching degree, and knowledge utilization effectiveness of the modeling instruction stream, and avoid the generation of low-quality modeling instruction streams.

[0068] Specifically, faithfulness measures the consistency between the generated modeling instructions and the facts of the given context. It is calculated based on the total number of modeling instructions in the generated instruction stream and the number of modeling instructions inferred from the context. The faithfulness evaluation value ranges from... Between these values, the higher the fidelity rating, the better the factual consistency.

[0069] The modeling instruction relevance assessment focuses on evaluating the relevance between generated modeling instructions and user modeling requirements. Incomplete or redundant modeling instructions will receive lower scores. It is based on the user's modeling needs. and the modeling instructions in the generated modeling instruction stream The calculated value of the modeling instruction relevance assessment ranges from [value range missing]. Between these values, a higher value for the modeling instruction relevance evaluation indicates a better relevance.

[0070] Context relevance assessment measures the relevance of the retrieved context. It is calculated based on the total number of modeling instructions in the generated modeling instruction stream and the number of modeling instructions relevant to the user's needs. The value of the context relevance assessment ranges from [value missing]. Between these parameters, the higher the value of the context relevance assessment, the better the factual consistency; ideally, the retrieved context should only contain information that answers the user's modeling needs.

[0071] Furthermore, the step of obtaining the feasibility parameter values ​​corresponding to the modeling instruction stream based on the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream includes:

[0072] The feasibility parameter values ​​corresponding to the modeling instruction flow are obtained using the following formula:

[0073]

[0074] in, The total number of times the modeling instruction flow is generated for the large model. As the loyalty weight value, To model the relevance weight values ​​of instructions, This is the context-related weight value. ,and .

[0075] Therefore, by introducing dynamically configurable weight coefficients, fidelity, instruction relevance, and context relevance are weighted, fused, and quantitatively evaluated. This allows for accurate and automatic determination of the effectiveness and reliability of the modeling instruction flow, greatly reducing the risk of subsequent 3D modeling software failures or the generation of incorrect models due to large model illusions or logical errors.

[0076] Furthermore, the method also includes: if the feasibility parameter value corresponding to the modeling instruction is less than or equal to a preset infeasibility threshold, then stop using the large model to continue generating the modeling instruction stream, and output a prompt message for supplementing modeling data.

[0077] Therefore, by calculating feasibility parameters with weights and setting iteration termination conditions, the system can stop in time and trigger the supplementation of modeling data when it is not feasible, and output a high-quality modeling instruction stream when it is feasible, which can effectively prevent resource waste caused by invalid iterations.

[0078] Furthermore, the method also includes: counting the number of irrelevant questions corresponding to the feedback information accumulated in the knowledge base; if the number of irrelevant questions is greater than or equal to a preset threshold for the number of irrelevant questions, then updating the large model based on the feedback information accumulated in the knowledge base.

[0079] Therefore, by quantifying the number of irrelevant questions in the feedback information database and comparing it with preset expert experience thresholds, the feedback data can be automatically, accurately sorted, and efficiently utilized. This ensures that when there is enough accumulated feedback information, the incremental training of the large model can be effectively triggered to continuously improve its core capabilities and avoid ineffective or redundant training. At the same time, when the accumulation of feedback information is insufficient, new knowledge can be added to the vector database in a timely manner to enrich the system's real-time reference.

[0080] Furthermore, the step of counting the number of irrelevant questions corresponding to the feedback information accumulated in the knowledge base includes: dividing the feedback information accumulated in the knowledge base into multiple groups of documents, wherein each group of documents contains multiple groups of feedback questions and solutions; for each group of documents, calculating the correlation value between different groups of feedback questions and solutions in that document, and considering feedback questions in different groups of feedback questions and solutions with a correlation value greater than or equal to a preset threshold as relevant questions, and vice versa, as irrelevant questions, thereby obtaining the number of irrelevant questions corresponding to that document; and obtaining the number of irrelevant questions corresponding to the feedback information accumulated in the knowledge base based on the sum of the number of irrelevant questions corresponding to each group of documents.

[0081] This allows for the accurate detection of logically contradictory, redundant, or isolated entries in the knowledge base, preventing the generation of low-quality modeling instruction flows caused by superficially similar but essentially unrelated issues being misjudged as related.

[0082] Specifically, the document ={ , ,..., }, ( =1,2,… ) indicates a document The first in A data set of feedback issues and solutions. Further, the document is calculated using the following formula. Feedback issues and solutions data group and Correlation value between :

[0083]

[0084] in, , ,and ≠ , The total number of data sets containing feedback questions and solutions included in each document. The relevance of different feedback issues and solutions in different documents.

[0085] If you provide feedback on issues and solutions, please contact the data group. and Correlation value between If the value is less than the preset threshold, a data set of feedback problems and solutions will be provided. Feedback issues and solutions data set The feedback questions are considered irrelevant (that is, the two feedback questions in these two data sets are considered two irrelevant questions), and the feedback questions and solutions data sets are considered irrelevant. and Correlation value between If the value is greater than or equal to a preset threshold, a data set of problems and solutions will be provided. Feedback issues and solutions data set The feedback questions are considered relevant (that is, the two feedback questions for the two data sets are considered as the same unrelated question).

[0086] The number of irrelevant questions corresponding to the accumulated feedback information stored in the knowledge base is calculated using the following formula:

[0087]

[0088] in, This represents the total number of irrelevant questions corresponding to the feedback information accumulated and stored in the knowledge base. The number of document groups into which the feedback information accumulated and stored in the knowledge base is divided. ) is the first The number of irrelevant issues corresponding to a group of documents.

[0089] Furthermore, the preset threshold for the number of irrelevant issues The accuracy of the large model's capabilities and the modeling instruction flow generated by its reliance on the knowledge base is determined by modeling experts; if Greater than or equal to If there is too much information, it is considered that incremental data training is needed for the large model. Less than The feedback information will then be entered into the knowledge base.

[0090] Based on the same inventive concept, this invention also provides a method and system for automatically generating three-dimensional geological models. Please refer to [link / reference]. Figure 5 ,like Figure 5As shown, it includes: a requirement input module 1, configured to acquire 3D geological modeling requirement data input by the user; a data processing module 2, configured to acquire modeling data input by the user and process the modeling data; an instruction generation module 3, configured to input the 3D geological modeling requirement data into a large model after reinforcement learning, so as to generate a modeling instruction stream through the large model; a verification module 4, configured to verify the feasibility of the generated modeling instruction stream using the large model; and a model generation module 5, configured to, when the feasibility verification result of the modeling instruction stream is feasible, send the modeling instruction stream generated by the large model to the 3D geological modeling software, so that the 3D geological modeling requirement data can be generated. The geological modeling software can generate a corresponding 3D geological model by calling modeling instructions and processing modeling data according to the modeling instruction flow. When the feasibility verification result of the modeling instruction flow is infeasible, it performs problem matching in a pre-created vector database based on problem-to-problem reverse index matching technology and inputs the problem matching result into the large model so that the large model can regenerate the modeling instruction flow according to the problem matching result. The adjustment and feedback module 6 is configured to obtain feedback information from modeling experts on adjusting the 3D geological model and diagnosing and analyzing the modeling instruction flow generated by the large model, and store the feedback information in a pre-created knowledge base.

[0091] Therefore, through the demand input module 1, data processing module 2, instruction generation module 3, verification module 4, model generation module 5, and adjustment and feedback module 6, the automatic modeling process based on Agentic RAG can be fully realized. It can automatically process modeling data and autonomously complete the generation of modeling instruction flow and feasibility verification with the help of a large model after reinforcement learning. When verification fails, it can achieve problem matching and reconstruction of modeling instruction flow through reverse indexing of vector database. By continuously accumulating expert feedback data into the knowledge base and triggering large model updates, the automatic generation system of 3D geological models provided by this invention has the ability to self-evolve, effectively overcome the illusion of large models and time deadlines, and significantly improve the automation level and modeling accuracy of 3D geological model generation.

[0092] Specifically, the system first receives the 3D geological modeling requirement information submitted by the user through the requirement input module 1. Then, the instruction generation module 3 transmits the 3D geological modeling requirement information text to the large model that has been fine-tuned by reinforcement learning. The professional large model after reinforcement learning is used to generate a modeling instruction stream. After the modeling instruction stream is generated, the verification module 4 uses the large model to verify the feasibility of the generated modeling instruction stream. If the score is lower than the preset threshold, a self-feedback loop mechanism is automatically triggered, and the large model is iteratively optimized until a qualified modeling instruction stream is generated and output to the model generation module 5 to generate the 3D geological model.

[0093] Furthermore, the requirement input module 1 includes a requirement input window; the instruction generation module 3 includes a large model interface and a vector database interface, wherein the large model interface is used to communicate with the large model, and the vector database interface is used to communicate with the vector database.

[0094] Therefore, by setting up a requirement input window, users can be provided with an entry point to submit their 3D geological modeling requirements, ensuring that the 3D geological modeling requirements can be completely and accurately transmitted to the instruction generation module 3; by setting up a large model interface, bidirectional communication between the instruction generation module 3 and the large model can be realized; by setting up a vector database interface, retrieval and real-time data calling between the large model and the knowledge base can be realized, enabling the large model to generate a more accurate and reliable modeling instruction flow.

[0095] Specifically, after the user submits 3D geological modeling requirements through the requirements input window of the requirements input module 1, the requirements input module 1 transmits the 3D geological modeling requirements to the instruction generation module 3. After receiving the 3D geological modeling requirements, the instruction generation module 3 first initiates a communication request to the large model through the large model interface, sending the text of the 3D geological modeling requirements to the large model and waiting for its response. At the same time, it establishes a connection with the vector database through the vector database interface, preparing to perform knowledge retrieval operations. When the large model self-assesses the need to supplement knowledge base data during the processing of requirements, the instruction generation module 3 then sends a semantic similarity query request to the vector database through the vector database interface, retrieves relevant vectorized knowledge base data, and returns the results to the large model interface. The large model interface continuously receives the modeling instruction stream generated by the large model.

[0096] Furthermore, the data processing module 2 includes a data input window and a data processing window. The data input window is used to receive the model modeling data and the model modeling instruction data. The data processing window is used to process the model modeling data using OCR technology and process the model modeling instructions using a QulM-RAG-based RAG technology processing method.

[0097] Therefore, by setting up a data entry window, users are provided with an entry point for submitting modeling data, which enables efficient reception of modeling data and modeling instruction data. By setting up a data processing window, key information in the modeling data can be automatically identified and extracted using OCR (Optical Character Recognition) technology. Furthermore, modeling instruction data can be transformed into semantic vectors that large models can understand using RAG technology based on QulM-RAG (Question Inverted Matching RAG).

[0098] Specifically, modeling data includes borehole diagrams, stratigraphic information data, and other data used for modeling, while modeling instruction data includes commonly used modeling instructions, instruction streams, and parameters used for training large models. Users upload modeling data and instruction data to data processing module 2 through the data entry window. The data entry window receives and transmits the data to the data processing window. The data processing window first uses an OCR engine to recognize text, symbols, and geometric information in the modeling data, extracting structured parameters such as borehole number, coordinates, depth, and lithological description. At the same time, the modeling instruction data is processed using RAG technology based on QulM-RAG before being entered into the knowledge base.

[0099] Furthermore, the data processing module 2 outputs the processed modeling data and modeling instruction data to the data storage module 9 for storage. The data storage module 9 includes a modeling software data input window, a modeling software data processing window, a knowledge base data input window, and a knowledge base data processing window. The modeling software data input window is used to receive the modeling data file output by the data processing module 2. The modeling software data processing window is used to store the modeling data file according to the data calling rules of the 3D modeling software for use by the 3D modeling software. The knowledge base data input window is used to receive the knowledge base data output by the data processing module 2. The knowledge base data processing window is used to store the knowledge base data in the vector database for use by the large model.

[0100] Therefore, by setting up a data entry window for the modeling software, a channel for receiving modeling data files can be provided, ensuring that the modeling data files can be accurately entered into the data storage module 9. In addition, by setting up a data processing window for the modeling software, the modeling data files can be processed, allowing them to be directly recognized and called by the 3D modeling software. By setting up a knowledge base data entry window as the entry point for receiving knowledge base data, and by setting up a knowledge base data processing window, knowledge base data can be stored in the vector database, enabling large models to efficiently acquire knowledge through semantic retrieval.

[0101] Specifically, after receiving the modeling data file output by the data processing module 2, the modeling software data entry window of the data storage module 9 transmits it to the modeling software data processing window. The modeling data file is format-adapted and stored according to the data calling rules of the 3D modeling software, so that it can be directly called by the 3D modeling software. At the same time, the knowledge base data entry window receives the knowledge base data output by the data processing module 2 and transmits it to the knowledge base data processing window. The knowledge base data is written into the vector database for semantic retrieval and calling by the large model when the modeling instruction stream is generated.

[0102] Furthermore, the model generation module 5 includes a 3D modeling software interface, which is used to accept the modeling instruction stream output by the instruction generation module 3 to drive the 3D modeling software to generate a 3D geological model.

[0103] Therefore, by setting up a 3D modeling software interface, a standardized channel for issuing commands and transmitting results can be provided, ensuring stable and reliable communication between the command generation module 3 and the 3D modeling software, and guaranteeing the accuracy of the generated 3D geological model.

[0104] Specifically, the 3D modeling software interface built into the model generation module 5 first receives the modeling command stream output by the command generation module 3, then parses the modeling command stream into an API (Application Programming Interface) request sequence that conforms to the 3D modeling software calling specification, and sends it to the 3D modeling software through inter-process communication or network protocol to drive it to perform modeling operations such as stratigraphic construction, mesh generation and attribute assignment. After the 3D modeling software completes the calculation and generates the 3D geological model, the 3D modeling software interface receives the 3D geological model file path or memory handle and returns it to the model generation module 5 for loading and visualization.

[0105] Based on the same inventive concept, the present invention provides an electronic device, please refer to [reference needed]. Figure 6 ,like Figure 6 As shown, it includes a processor 7 and a memory 8. The memory 8 stores a computer program. When the computer program is executed by the processor 7, it implements the automatic generation method of the three-dimensional geological model described above.

[0106] Therefore, by executing the aforementioned automatic generation method for three-dimensional geological models, an automatic modeling process based on the Agentic RAG paradigm can be realized. It can autonomously complete the requirements understanding, modeling instruction flow generation, feasibility verification, and three-dimensional geological model construction using a large model optimized by reinforcement learning. Through the built-in vector database and knowledge base, it supports retrieval and continuous learning, effectively solving the problems of reliance on manual labor, large model illusion, and lagging knowledge updates in traditional modeling techniques.

[0107] Compared with existing technologies, the automatic generation method, system, and electronic device for three-dimensional geological models provided by this invention have the following advantages:

[0108] The automatic generation method for 3D geological models provided by this invention innovatively solves the problems of black-box generation, model illusion, and knowledge solidification in traditional large-scale 3D geological modeling by introducing the Agentic RAG paradigm. At the same time, it continuously optimizes the large model, enabling it to autonomously generate and verify the modeling instruction flow, significantly improving the accuracy and efficiency of modeling instruction flow generation. In addition, the automatic generation method for 3D geological models provided by this invention uses the modeling instruction flow to drive 3D modeling software, realizing full automation from requirement input to model generation, greatly reducing the cost of manual intervention, and achieving efficient generation of complex and irregular 3D geological models.

[0109] The automatic 3D geological model generation system provided by this invention fully realizes the automatic modeling process based on AgenticRAG through a requirement input module 1, a data processing module 2, an instruction generation module 3, a verification module 4, a model generation module 5, and an adjustment and feedback module 6. It can automatically process modeling data and autonomously complete instruction flow generation and feasibility verification with the help of a large model after reinforcement learning. When verification fails, it can achieve problem matching and instruction flow reconstruction through reverse indexing of the vector database. By continuously accumulating expert feedback data into the knowledge base and triggering large model updates, the automatic 3D geological model generation system provided by this invention has self-evolution capabilities, effectively overcoming the problems of large model illusion and time deadlines, and significantly improving the automation level and modeling accuracy of 3D geological model generation.

[0110] The electronic device for automatically generating three-dimensional geological models provided by this invention can realize an automatic modeling process based on the Agentic RAG paradigm by executing the automatic generation method of the three-dimensional geological models. It can autonomously complete the requirements understanding, modeling instruction flow generation, feasibility verification and three-dimensional geological model construction using a large model optimized by reinforcement learning. Through the built-in vector database and knowledge base, it supports retrieval and continuous learning, which can effectively solve the problems of traditional modeling relying on manual labor, large model illusion and knowledge update lag.

[0111] The above description is merely a description of preferred embodiments of the present invention and is not intended to limit the scope of the present invention in any way. Any changes or modifications made by those skilled in the art based on the above disclosure shall fall within the protection scope of the present invention.

Claims

1. A method for automatically generating three-dimensional geological models, characterized in that, The method includes: Obtain the user-inputted 3D geological modeling requirements data and modeling data, and process the modeling data; The three-dimensional geological modeling requirement data is input into a large model after reinforcement learning, so as to generate a modeling instruction flow through the large model; The feasibility of the generated modeling instruction stream is verified using the large model. If the verification result is feasible, the modeling instruction stream generated by the large model is sent to the 3D geological modeling software so that the 3D geological modeling software can call the modeling instructions and the processed modeling data according to the modeling instruction stream to generate the corresponding 3D geological model. Obtain feedback information from modeling experts regarding adjustments to the 3D geological model and the diagnosis and analysis of the modeling instruction flow generated by the large model, and store the feedback information in a pre-created knowledge base; If the verification result is not feasible, then problem matching is performed in a pre-created vector database based on the problem-to-problem reverse index matching technology; The problem matching result is input into the large model so that the large model can regenerate the modeling instruction flow based on the problem matching result; The verification of the feasibility of using the large model to generate the modeling instruction stream includes: The large model is used to obtain the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream; Based on the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value corresponding to the generated modeling instruction stream, obtain the feasibility parameter value corresponding to the modeling instruction stream; If the feasibility parameter value corresponding to the modeling instruction flow is greater than or equal to the preset feasibility threshold, the modeling instruction flow is determined to be feasible; otherwise, the modeling instruction flow is determined to be infeasible. The method further includes: The number of irrelevant questions corresponding to the accumulated feedback information stored in the knowledge base is statistically analyzed. If the number of irrelevant questions is greater than or equal to a preset threshold for the number of irrelevant questions, the large model is updated based on the feedback information accumulated in the knowledge base. The step of counting the number of irrelevant questions corresponding to the accumulated feedback information stored in the knowledge base includes: The feedback information accumulated in the knowledge base is divided into multiple sets of documents, each set of documents containing multiple sets of feedback questions and solutions. For each set of documents, the correlation value between different feedback issues and solutions data groups in the document is calculated. Feedback issues in different feedback issues and solutions data groups with a correlation value greater than or equal to a preset threshold are considered as relevant issues, and those with a correlation value less than or equal to a preset threshold are considered as irrelevant issues. This allows us to obtain the number of irrelevant issues corresponding to the document. The number of irrelevant questions corresponding to the accumulated feedback information in the knowledge base is obtained by summing the number of irrelevant questions corresponding to each group of documents.

2. The method for automatically generating a three-dimensional geological model according to claim 1, characterized in that, The processing of the modeling data includes: Optical character recognition technology is used to identify the modeling data in order to identify valid information and output the corresponding Excel format file.

3. The method for automatically generating a three-dimensional geological model according to claim 1, characterized in that, The fidelity evaluation value corresponding to the generated modeling instruction stream is obtained using the following formula: In the formula, This is a loyalty assessment value, and its value ranges from [value range missing]. between; The modeling instruction relevance evaluation value corresponding to the generated modeling instruction stream is obtained using the following formula: In the formula, To evaluate the relevance of modeling instructions, The total number of modeling instructions in the generated modeling instruction stream. To meet the user's modeling needs, This refers to the modeling instructions in the generated modeling instruction stream, whose values ​​range from... Between, sim This represents the correlation value between the user's modeling requirements and the modeling instructions in the generated modeling instruction stream; The context relevance evaluation value corresponding to the generated modeling instruction stream is obtained using the following formula: In the formula, This is a context-dependent evaluation value, and its range is within... between.

4. The method for automatically generating a three-dimensional geological model according to claim 1, characterized in that, The process of obtaining feasibility parameter values ​​corresponding to the generated modeling instruction stream based on the fidelity evaluation value, modeling instruction relevance evaluation value, and context relevance evaluation value includes: The feasibility parameter values ​​corresponding to the modeling instruction flow are obtained using the following formula: in, The total number of times the modeling instruction flow is generated for the large model. As the loyalty weight value, To model the relevance weight values ​​of instructions, This is the context-related weight value. ,and .

5. The method for automatically generating a three-dimensional geological model according to claim 1, characterized in that, The method further includes: If the feasibility parameter value corresponding to the modeling instruction is less than or equal to the preset infeasibility threshold, then the generation of modeling instruction stream using the large model will be stopped, and a prompt message for supplementing modeling data will be output.

6. A method system for automatically generating three-dimensional geological models, characterized in that, include: The requirement input module is configured to obtain the 3D geological modeling requirement data input by the user. The data processing module is configured to acquire modeling data input by the user and process the modeling data. The instruction generation module is configured to input the three-dimensional geological modeling requirement data into a large model after reinforcement learning, so as to generate a modeling instruction flow through the large model; The verification module is configured to use the large model to verify the feasibility of the generated modeling instruction stream; The model generation module is configured to, when the feasibility verification result of the modeling instruction flow is feasible, send the modeling instruction flow generated by the large model to the 3D geological modeling software, so that the 3D geological modeling software can generate the corresponding 3D geological model by calling the modeling instructions and the processed modeling data according to the modeling instruction flow; and when the feasibility verification result of the modeling instruction flow is infeasible, perform problem matching in a pre-created vector database based on problem-to-problem reverse index matching technology, and input the problem matching result into the large model, so that the large model can regenerate the modeling instruction flow according to the problem matching result. The adjustment and feedback module is configured to obtain feedback information from modeling experts on adjusting the three-dimensional geological model and diagnosing and analyzing the modeling instruction flow generated by the large model, and to store the feedback information in a pre-created knowledge base; The adjustment and feedback module is further configured to divide the feedback information accumulated in the knowledge base into multiple groups of documents, wherein each group of documents contains multiple groups of feedback questions and solutions; for each group of documents, the correlation value between different groups of feedback questions and solutions in that document is calculated, and feedback questions in different groups of feedback questions and solutions with a correlation value greater than or equal to a preset threshold are considered as related questions, and vice versa, thereby obtaining the number of irrelevant questions corresponding to that document; the number of irrelevant questions corresponding to the feedback information accumulated in the knowledge base is obtained based on the sum of the number of irrelevant questions corresponding to each group of documents; if the number of irrelevant questions is greater than or equal to a preset threshold for the number of irrelevant questions, the large model is updated based on the feedback information accumulated in the knowledge base.

7. An electronic device, characterized in that, The system includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, implements the automatic generation method for three-dimensional geological models according to any one of claims 1 to 5.

Images