Oil and gas geology multi-mode question and answer method

Abstract

The invention relates to an oil and gas geology multi-mode question and answer method, which comprises the following steps: S1, receiving a query text, and converting the query text into vector representation; s2, carrying out vectorization processing on the multi-modal document content in the knowledge base, wherein the multi-modal document content comprises characters, tables and graphs; s3, on the basis of a preset retrieval strategy, performing correlation calculation on the vector representation of the query text and vectorized document content in the knowledge base to obtain multi-modal answer content; and S4, integrating and outputting the multi-mode answer content.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration and development, and particularly to a multi-modal question-answering method for oil and gas geology. Background Art

[0002] All aspects of oil and gas exploration and development are becoming increasingly efficient and convenient.

[0003] However, the development of technology has also brought new challenges: the quantity of geological documents and data has increased sharply. These specialized documents usually contain multi-modal data: text, tables, and graphs, which build a knowledge barrier for non-professionals and make it difficult for them to quickly grasp the core content. At the same time, the massive reports lead to information clutter, and it is also difficult for professionals to quickly find the key information they need.

[0004] Therefore, how to achieve multi-modal data question-answering in the oil and gas field is an urgent problem to be solved. Summary of the Invention

[0005] The present invention provides an oil and gas geology question-answering system based on natural language processing and multi-modal fusion technology. Through methods such as topic setting, text vectorization processing, image data processing, similarity matching, and multi-modal fusion, it can quickly locate and query text, table, and chart content, and form a complete answer output.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present application provides a multi-modal question-answering method for oil and gas geology, including:

[0008] S1. Receive a query text and convert the query text into a vector representation;

[0009] S2. Vectorize the multi-modal document content in the knowledge base, where the multi-modal document content includes text, tables, and graphs;

[0010] S3. Based on a preset retrieval strategy, perform an association calculation between the vector representation of the query text and the vectorized document content in the knowledge base to obtain multi-modal answer content;

[0011] S4. Integrate and output the multi-modal answer content.

[0012] In one embodiment, in S1, it includes: performing semantic analysis on the query text to extract key topic words and understand the user's query intention, and performing vector representation on the key topic words.

[0013] In one embodiment, in S1, it further includes tokenizing and cleaning the query text input by the user to remove redundant content.

[0014] In one embodiment, in S3, based on the vector representation of the key topic words, a comparison is made with the subsection topics in the document content to obtain the subsection with the highest degree of approximation, and further, the text, tables, and charts that best match the query text are matched in this subsection.

[0015] In one embodiment, in the S3 topic matching and similarity analysis step, for unstructured geological cross-section diagrams, key information and feature points are extracted through image recognition technology, converted into a structured data representation, and then the similarity is calculated with the vectorized result of the query text, so as to achieve precise search for geological cross-section diagrams.

[0016] In one embodiment, in the S2 vectorization processing step of document multi-modal data, it also includes the process of constructing a knowledge graph. Entities, relationships, etc. in the text content, tables, and chart data in the knowledge base are extracted to construct a knowledge graph in the field of oil and gas geology. Searching based on this knowledge graph can more precisely match the text and charts related to the query text.

[0017] In one embodiment, in S4, the integrated multi-modal answer content is formatted and optimized to improve the readability of the answer and the user experience.

[0018] In a second aspect, an oil and gas geology multi-modal question-answering system includes:

[0019] A query processing module, configured to receive a query text and convert the query text into a vector representation;

[0020] A data storage module, configured to perform vectorization processing on the multi-modal document content in the knowledge base, where the multi-modal document content includes text, tables, and graphs;

[0021] A search module, configured to perform an association calculation between the vector representation of the query text and the vectorized document content in the knowledge base based on a preset retrieval strategy to obtain multi-modal answer content;

[0022] An answer generation module, configured to integrate and output the multi-modal answer content.

[0023] Due to the above technical solutions adopted by the present invention, the following advantages are achieved:

[0024] 1. Multi-modal data joint processing: The present invention can effectively combine multi-modal data such as text, images, and tables, extract and fuse information of different modalities through a deep learning model, and provide more comprehensive and accurate geological information query results. For example, for a query input by a user, relevant documents can be found not only according to the text description, but also the key information of geological cross-section diagrams, seismic diagrams, etc. can be extracted through the image processing module.

[0025] 2. Retrieval Enhancement Based on Graph Retrieval-Augmented Generation: Through the graph retrieval-augmented generation technology, the present invention enhances the traditional retrieval, combines the knowledge graph in the field of oil and gas geology for semantic matching and reasoning, and significantly improves the accuracy and relevance of query results. Through graph reasoning, the system can provide answers with more background knowledge support, rather than just keyword-based matching.

[0026] 3. Precise Image and Table Data Retrieval: A dedicated image processing module (such as a convolutional neural network) is used to extract deep features of geological images and fuse them with text information, achieving precise joint retrieval of images, tables, and text. This enables the system to understand and process unstructured data such as geological cross-sections and experimental result charts, filling the gap in the processing of unstructured data in the prior art.

[0027] 4. Deep Application of Knowledge Graph: The present invention constructs a dedicated knowledge graph in the field of oil and gas geology to support semantic reasoning and deep retrieval based on entities and relationships. Through the semantic enhancement of the knowledge graph, the system can provide more accurate answers to complex queries.

[0028] 5. Efficient Cross-Modal Reasoning Ability: By combining multi-modal learning and the reasoning ability of the knowledge graph, the present invention can not only understand the surface meaning of text but also capture potential information in image and table data, providing users with more in-depth geological analysis and decision-making support.

[0029] The advantage of this patented technology is that it can effectively combine multi-modal data for joint analysis, improving the accuracy and practicality of the question-answering system in the field of oil and gas geology in actual applications. Especially when facing complex queries and unstructured data, it demonstrates its unique technical advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flowchart of the multi-modal question-answering method for oil and gas geology;

[0031] Figure 2 is a flowchart of a more detailed embodiment;

[0032] Figure 3 is a flowchart of a numerical example. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention fall within the scope of protection of the present invention.

[0034] In view of the deficiencies and problems of the prior art, the present application provides a multi-modal question-answering method for oil and gas geology, including:

[0035] S1. Receive a query text and convert the query text into a vector representation;

[0036] S2. Vectorize the multi-modal document content in the knowledge base, where the multi-modal document content includes text, tables, and graphs;

[0037] S3. Based on a preset retrieval strategy, perform an association calculation between the vector representation of the query text and the vectorized document content in the knowledge base to obtain multi-modal answer content;

[0038] S4. Integrate and output the multi-modal answer content.

[0039] The above method will be described in a more detailed embodiment with reference to more drawings. After a detailed embodiment of the present application, a numerical example will be described to illustrate the practical effect of the method.

[0040] Detailed Embodiment

[0041] The present invention provides a method and system for storing and querying unstructured data in exploration and development of oil and gas geology based on a knowledge graph and a large language model.

[0042] Among them, the system mainly includes the following functions:

[0043] (1) Vectorize the query text, and combine the graph retrieval enhancement generation technology to match the relevant text and charts in the geological documents, including the process of constructing a knowledge graph (KG) in the field of oil and gas geology and searching based on the knowledge graph;

[0044] (2) Perform a similarity analysis on the content of the charts and output information in combination with the charts related to the query topic;

[0045] (3) Through text cleaning and semantic analysis, integrate the relevant charts and text content of the query, and finally form a complete answer.

[0046] In this detailed embodiment, the process of the method includes:

[0047] Step 1: Query text parsing and processing;

[0048] Specifically, the user inputs a query text through the question-answering system. The system first performs word segmentation and cleaning on the input text, removes redundant content, and extracts key topic words through natural language processing technology.

[0049] Step 2: Construct a knowledge graph (KG);

[0050] Specifically, through natural language processing (NLP) technology, domain-related entities are identified from text data, such as oil and gas field locations, rock formation types, oil and gas reserves, exploration methods, etc. Through relation extraction technology, the relationships between different entities are identified and constructed. For example, the relationship between the exploration location and the oil and gas layer, the correlation between the exploration results and the experimental methods, etc. The extracted entities and relationships are transformed into a graph structure to form a knowledge graph in the field of oil and gas geology.

[0051] Step 3: Vectorization processing of document multi-modal data;

[0052] Specifically, the cleaned query text is transformed into a vector representation and compared with the vectorized document content in the knowledge base. The specific implementation includes using graph retrieval enhanced generation technology to calculate the relevance between the query text and the document content; vectorizing the text content, tables, and chart data in the knowledge base respectively to form a unified vector space representation.

[0053] Step 4: Search based on the knowledge graph;

[0054] Specifically, the constructed knowledge graph is used as background knowledge for the query to enhance the semantic understanding ability of the query. The query text is matched with the entities and relationships in the knowledge graph, and more accurate retrieval results are provided through graph reasoning. The system searches for nodes related to the entities in the query text in the knowledge graph according to the vectorized query text. Through graph reasoning, the system can automatically infer more potential matching items based on the context information of the query. On the basis of performing graph queries, the system will also combine the image processing module and table data, and further refine the retrieval results through the vectorized representation of image features and table data.

[0055] Step 5: Topic matching and similarity analysis;

[0056] Specifically, based on the vectorized result of the query text, combined with the subsection topic words and the vectorized content, the system retrieves the most relevant texts and charts. The specific implementation includes using a semantic matching algorithm for the text content to extract the subsection most relevant to the query text; calculating the similarity of the chart data to select the chart most relevant to the query topic.

[0057] Step 6: Multi-modal integration output of charts and texts.

[0058] Specifically, for the subsections and charts related to the query text, the system integrates the chart content and the corresponding text content into a multi-modal answer through semantic analysis technology. The specific implementation includes extracting the charts related to the query topic and generating descriptive text; fusing the text answer and the chart description and outputting the complete answer in a fixed format.

[0059] Some calculation examples are provided below to illustrate the effect of the present invention.

[0060] Example 1: Q&A about Reservoir Characteristics Based on Text and Charts

[0061] The user inputs the query text: "What are the main characteristics of the reservoir?" The system processes as follows:

[0062] (1) Extract the topic "reservoir characteristics" from the query text;

[0063] (2) Retrieve relevant chapters and charts in the knowledge base, and extract the reservoir lithology table and pressure-temperature diagram;

[0064] (3) Integrate the relevant subsection text and chart content and output to form a multimodal answer.

[0065] Output example:

[0066] Text answer:

[0067] The main characteristics of the reservoir include lithology, porosity, permeability, etc. The following charts show the relevant characteristic parameters:

[0068] Relevant charts:

[0069] (1) Lithology data table

[0070] (2) Pressure-temperature curve diagram

[0071] Example 2: Q&A Based on Geological Section

[0072] The user inputs the query text: "Please show the geological section characteristics of XX area." The system processes as follows:

[0073] (1) Extract the topic "geological section characteristics";

[0074] (2) Extract entity, relationship, and attribute information from the geological documents in the knowledge base to construct a knowledge graph related to the geological section. Process the geological section diagrams in the knowledge base through image recognition technology, extract key information and feature points such as stratigraphic boundaries and lithology changes, convert them into structured data representations, and integrate them into the knowledge graph;

[0075] (3) Use graph retrieval enhanced generation technology to search for entities and relationship paths related to "geological section characteristics" in the knowledge graph, calculate the similarity between the query text and the geological section diagrams in the knowledge base, and retrieve the geological section diagram most relevant to the topic;

[0076] (4) Integrate the subsection text related to the topic and the retrieved geological section diagram and output the answer.

[0077] In several embodiments provided by the present invention, it should be understood that the disclosed method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the above division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple 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 displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces, and the indirect coupling or communication connection of devices or units can be in electrical, mechanical or other forms.

[0078] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An oil and gas geology multi-modal question answering method, characterized in that, including: S1. Receive a query text and convert the query text into a vector representation; S2. Vectorize the multi-modal document content in the knowledge base, where the multi-modal document content includes text, tables, and figures; S3. Based on a preset retrieval strategy, perform an association calculation between the vector representation of the query text and the vectorized document content in the knowledge base to obtain multi-modal answer content; S4. Integrate and output the multi-modal answer content.

2. The oil and gas geology multimodal question-answering method according to claim 1, wherein In S1, it includes: performing semantic analysis on the query text to extract key topic words and understand the user's query intention, and performing vector representation on the key topic words.

3. The multi-modal question-answering method for oil and gas geology according to claim 2, wherein, In S1, it also includes tokenizing and cleaning the query text input by the user to remove redundant content.

4. The oil and gas geology multimodal question-answering method according to claim 2, characterized in that In S3, based on the vector representation of the key topic words, compare with the subsection topics in the document content to obtain the subsection with the highest similarity, and further match the text, table, and chart that best match the query text in this subsection.

5. The multi-modal question-answering method for oil and gas geology according to claim 4, characterized in that In the S3 topic matching and similarity analysis step, for unstructured geological section diagrams, extract the key information and feature points therein through image recognition technology, convert them into a structured data representation, and then calculate the similarity with the vectorized result of the query text, so as to achieve precise search for geological section diagrams.

6. The multi-modal question-answering method for oil and gas geology according to claim 4, wherein, In the S2 vectorization processing step of multi-modal data of documents, it also includes the process of constructing a knowledge graph, extracting the entity, relationship, etc. information in the text content, tables, and chart data in the knowledge base, constructing a knowledge graph in the field of oil and gas geology, and based on this knowledge graph for searching, it can more precisely match the text and charts related to the query text.

7. The multi-modal Q&A method for oil and gas geology according to claim 4, wherein In S4, format and optimize the integrated multi-modal answer content to improve the readability and user experience of the answer.

8. An oil and gas geology multi-modal question answering system, characterized in that, including: A query processing module for receiving a query text and converting the query text into a vector representation; A data storage module for vectorizing the multi-modal document content in the knowledge base, where the multi-modal document content includes text, tables, and figures; A search module for performing an association calculation between the vector representation of the query text and the vectorized document content in the knowledge base based on a preset retrieval strategy to obtain multi-modal answer content; An answer generation module for integrating and outputting the multi-modal answer content.

Images