Intelligent interactive system and method based on large-scale reservoir dynamic analysis scenario model

Abstract

The present invention relates to the field of artificial intelligence big model technology, and specifically to an intelligent interaction system and method based on a big model of a reservoir dynamic analysis scenario. The intelligent interaction method based on the big model of the reservoir dynamic analysis scenario includes: detailed decomposition of user questions through named entity recognition technology to identify specific needs of the questions; answering questions in oil and gas reservoir engineering in response to analysis and suggestion needs, and giving relevant measures and suggestions; analyzing and revealing the periodic change laws of reservoir production dynamic data in response to data retrieval and regularity analysis needs; and classifying and calling external tools required for answering questions in response to the needs of external auxiliary tool calls. Compared with general oil and gas engineering industry big models, the present invention is further oriented to reservoir dynamic analysis scenarios in the field of oil and gas development on the basis of the big model, and can assist engineers in analyzing and answering practical problems in related scenarios from the perspective of reservoir dynamic analysis.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence large model technology, and in particular to an intelligent interaction system and method based on a large model of oil reservoir dynamic analysis scenarios. Background Art

[0002] At present, since the large models of the oil and gas industry need to take into account their versatility in a wide range of fields such as oil and gas engineering exploration and development, their intelligence level in many specific application scenarios is not high. The proposal of scenario large models emphasizes the simplification and optimization of the human-computer interaction process between practitioners and large models in specific scenarios based on actual business needs. Traditional reservoir dynamic analysis is very time-consuming and laborious. Reservoir engineers need to use production data, rely on various engineering software, and combine expert experience to provide an effective basis for adjusting development plans and taking measures.

[0003] Based on the above reasons, in order to solve the many problems of many oil wells and producing layers in old oil fields, and the huge workload of reservoir engineers in dynamic analysis, an intelligent interactive system based on a large model of reservoir dynamic analysis scenarios is needed, so that engineers can use the web page intelligent chat question and answer window to complete intelligent auxiliary analysis functions such as well history data retrieval and analysis, well profile drawing, development and production key technical indicator calculation, and complex problem measures for reservoir dynamic analysis. It is expected to realize auxiliary reservoir dynamic analysis through simple human-computer dialogue, replace most of the manual work, and effectively improve the development efficiency of oil and gas fields. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent interactive system and method based on a large model of reservoir dynamic analysis scenarios, so that engineers can use the intelligent chat question-and-answer window on the web page to complete intelligent auxiliary analysis functions such as well history data retrieval and analysis, well profile drawing, key technical indicator calculation for development and production, and suggestions for measures to solve complex problems in reservoir dynamic analysis. It is expected that auxiliary reservoir dynamic analysis can be achieved through simple human-computer dialogue, replacing most of the manual work, thereby effectively improving the development efficiency of oil and gas fields.

[0005] To achieve the above objectives, in a first aspect, the present invention provides an intelligent interaction method based on a large model of reservoir dynamic analysis scenarios, comprising:

[0006] Through named entity recognition technology, user questions are broken down in detail, entities are extracted, and a deeper semantic analysis of the questions is performed to identify the specific needs of the questions, including analysis and suggestion needs, data retrieval and pattern analysis needs, and the need to call external auxiliary tools;

[0007] Aiming at the needs of analytical suggestions, answer questions in oil and gas reservoir engineering, and infer the possible deeper phenomena based on some superficial phenomenal statements, and give relevant measures and suggestions;

[0008] Aiming at the needs of data retrieval and law analysis, it realizes direct retrieval from natural language in the database, and analyzes and reveals the periodic change law of reservoir production dynamic data;

[0009] In response to the need to call external auxiliary tools, the external tools needed to answer questions are classified. Different questions use different tools, and similar questions use the same tools. The classification language task will classify specific questions into specific types and return the corresponding number of the type. A one-to-one association is established in advance between the label number and the function object or pointer, and the borrowing of the specified external tool is completed through function search and call.

[0010] The large model of reservoir dynamic analysis scenario adopts a separate development mode that combines algorithm API development, back-end data coupling, and front-end page display to build the software architecture.

[0011] The specific steps of using named entity recognition technology to decompose user questions in detail, extract entities, and perform deeper semantic analysis on the questions to identify the specific needs of the questions include:

[0012] The key information in user questions is extracted and divided into three groups. Group 1 is general questions that can be answered without retrieval or calculation. Group 2 is variables and keywords extracted from user questions for database retrieval, and summary statements in user questions for tool selection. Group 3 is intention information extracted from user questions, including whether you want to visualize the query or calculation results, whether you require trend analysis of the query or calculation results, and whether you need to give recommended measures based on the analysis results.

[0013] The analysis reveals the periodic variation of reservoir production dynamic data using statistical methods.

[0014] Among them, the specific steps for realizing direct retrieval from natural language in the database and analyzing and revealing the periodic change patterns of reservoir production dynamic data in view of data retrieval and pattern analysis needs include: completing the conversion from natural language description to SQL query statements, converting natural language questions mixed with some key retrieval information into Select query statements in SQL language, and completing simple trend analysis and capturing periodic patterns based on given structured data.

[0015] In a second aspect, the present invention further provides an intelligent interactive system based on a large model of a reservoir dynamic analysis scenario, comprising a complex problem suggestion subsystem for reservoir engineering, a data mining and dynamic law analysis subsystem, and a tool integration calling subsystem; the complex problem suggestion subsystem for reservoir engineering, the data mining and dynamic law analysis subsystem, and the tool integration calling subsystem are connected in sequence;

[0016] The oil and gas reservoir engineering complex problem suggestion subsystem is used to analyze and suggest needs, answer questions in oil and gas reservoir engineering, and infer possible deeper phenomena based on some surface phenomenal statements, and give relevant measures and suggestions;

[0017] The data mining and dynamic law analysis subsystem is used to meet the needs of data retrieval and law analysis, realize direct retrieval from natural language in the database, and analyze and reveal the periodic change law of oil reservoir production dynamic data;

[0018] The tool integration calling subsystem is used to meet the needs of external auxiliary tool calling and classify the external tools needed to answer questions. Different questions use different tools, and similar questions use the same tools. The classification language task will classify specific questions into specific types and return the corresponding number of the type. A one-to-one association is established in advance using the label number and the function object or pointer, and the borrowing of the specified external tool is completed through function search and calling.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. Compared with the general large models of the oil and gas engineering industry, it has stronger petroleum engineering professional capabilities and can analyze and answer questions from the perspective of reservoir dynamic analysis.

[0021] 2. The interaction is simple and the command comprehension ability is strong. Through simple human-computer dialogue, auxiliary reservoir dynamic analysis can be realized, which can improve work efficiency and reduce the cost of trial and error.

[0022] 3. It has the ability to call commercial software tools in the oil and gas field, can handle complex engineering indicator calculations and drawing problems, and further enhance the analytical capabilities of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0024] Figure 1 It is a flow chart of an intelligent interaction method based on a large model of reservoir dynamic analysis scenario of the present invention.

[0025] Figure 2 It is a structural schematic diagram of an intelligent interactive system based on a large model of reservoir dynamic analysis scenarios according to the present invention.

[0026] Figure 3 It is a fine-tuning technology and functional structure diagram of an intelligent interactive system based on a large model of a reservoir dynamic analysis scenario according to the present invention.

[0027] Figure 4 It is a complex problem suggestion subsystem 101 of oil and gas reservoir engineering of an intelligent interactive system based on a large model of oil reservoir dynamic analysis scenario of the present invention.

[0028] Figure 5 It is a data mining and dynamic law analysis subsystem 102 of an intelligent interactive system based on a large model of a reservoir dynamic analysis scenario of the present invention.

[0029] Figure 6 It is a tool integration calling subsystem 103 of the intelligent interactive system based on the reservoir dynamic analysis scenario large model of the present invention.

[0030] Figure 7 It is a schematic diagram of a subsystem coupling method of an intelligent interactive system based on a large model of a reservoir dynamic analysis scenario according to the present invention.

[0031] Figure 8 The invention discloses an example diagram of the function of querying and drawing well history injection and production data by an intelligent interactive system based on a large model of a reservoir dynamic analysis scenario.

[0032] Fig. 9 The present invention is an example diagram of the function of querying and printing well history measure data of an intelligent interactive system based on a large model of a reservoir dynamic analysis scenario.

[0033] Fig.10 This is an example diagram of the function of calling Petrel2021 to draw a well-connected profile diagram based on an intelligent interactive system of the present invention that is based on a large model of a reservoir dynamic analysis scenario.

[0034] Fig.11 The invention is an intelligent interactive system call tool based on a large model of reservoir dynamic analysis scenario to calculate key technical indicators for development and propose analysis of effective water injection layer measures. DETAILED DESCRIPTION

[0035] In the first aspect, the present invention provides an intelligent interaction method based on a large model of reservoir dynamic analysis scenarios, see Figure 1 The present invention provides an intelligent interaction method based on a large model of reservoir dynamic analysis scenarios, comprising:

[0036] S1 uses named entity recognition technology to break down user questions in detail, extract entities, and conduct a deeper semantic analysis of the questions to identify the specific needs of the questions, including analysis and suggestion needs, data retrieval and pattern analysis needs, and the need to call external auxiliary tools;

[0037] The key information in user questions is extracted and divided into three groups. Group 1 is general questions that can be answered without retrieval or calculation. Group 2 is variables and keywords extracted from user questions for database retrieval, and summary statements in user questions for tool selection. Group 3 is intention information extracted from user questions, including whether you want to visualize the query or calculation results, whether you require trend analysis of the query or calculation results, and whether you need to give recommended measures based on the analysis results.

[0038] S2 is oriented to the analysis and suggestion needs, answering the questions in oil and gas reservoir engineering, and reasoning about the possible deeper phenomena based on some superficial phenomenal statements, and giving relevant measures and suggestions;

[0039] S3 is designed for data retrieval and law analysis needs, enabling direct retrieval from natural language in the database and analyzing and revealing the periodic change laws of reservoir production dynamic data;

[0040] The statistical method is used to reveal the law of production data and analyze the changes of reservoir production dynamic data. The specific steps include: completing the conversion from natural language description to SQL query statement, converting the natural language question sentence mixed with some key retrieval information into Select query statement in SQL language, and completing simple trend analysis and capturing periodic laws based on given structured data.

[0041] S4 is oriented to the needs of external auxiliary tool calls. It classifies the external tools needed to answer questions. Different questions use different tools, and similar questions use the same tools. The classification language task will classify specific questions into specific types and return the corresponding number of the type. It uses the label number to establish a one-to-one association with the function object or pointer in advance, and completes the borrowing of the specified external tool through function search and call.

[0042] The S5 reservoir dynamic analysis scenario large model uses a separate development model that combines algorithm API development, back-end data coupling, and front-end page display to build the software architecture.

[0043] At the software architecture level, the reservoir dynamic analysis scenario model (Reservoir Dynamic Analysis Toolkit) adopts a separate development mode of algorithm API development + back-end data coupling + front-end page display. Among them, the algorithm part is based on the Pytorch framework and the Transformers function package for incremental pre-training, fine-tuning and model calling, and forms an algorithm call API for functional subsystem coupling. After the subsystem algorithm embeds the processed data into a unified JSON format, the data is sent to the front-end user interface in real time through SocketIO, and the data is sent to the back-end database for storage through Http RESETfulAPI to form historical record data. When the front-end queries the historical records, the back-end database is queried through http and displayed on the front-end. The back-end builds the subsystem coupling logic based on Http RESETfulAPI and SocketIO network protocols, and completes the response to the front-end question request in combination with user data, while recording historical data for user query. Based on the JSON data format, it provides developers with a simple and easy-to-use API that supports cross-platform and cross-browser real-time communication. The front-end builds a user question-and-answer interface based on the Vue3 framework in NodeJS to facilitate interactive data visualization, and builds the user interface based on standard HTML, CSS, and JavaScript scripting language rules in Vue3. It uses component-based programming mode to directly embed mature chart generation modules such as Echart in the user UI.

[0044] The large model of reservoir dynamic analysis scenario is realized by the construction method of hybrid expert system. This method of mixing and giving full play to the capabilities of multiple models through program design has become more common in the application of large models in specific scenarios. The main idea of ​​the construction is to use the fine-tuning models A, B, and C, which are good at three specific task modes of analysis suggestions, data retrieval, and tool classification, respectively, and use the named entity recognition model D to obtain multi-dimensional entity information from user input to plan the specific calling scheme of A, B, and C, so as to realize the following comprehensive scenario application functions: 1. Analyze and answer reservoir dynamic analysis scenario problems; 2. Query the dynamic and static database of the oil field and explore the law of data changes; 3. Call the tool to calculate the oil field field development indicators; 4. Linkage analysis involves complex mixed problems of scenarios 1 to 3.

[0045] The above includes a total of 4 fine-tuning models. The 4 models are fine-tuned by the LoRA method based on ChiPetro-Alpaca-2.0 using different labeled data sets. Among them, 3 are fine-tuned for specific scenarios: A. Analysis suggestion model, B. Data retrieval model, C. Tool classification model, and the other fine-tuning model D. Named entity recognition model, which is mainly responsible for feature extraction of user input. Since models A, B, and C all receive natural language input and output in Markdown lightweight text markup format, and the coupling of computer program modules is usually through function calls, we encapsulate A, B, and C into the complex problem suggestion subsystem 101 of oil and gas reservoir engineering, the data mining and dynamic law analysis subsystem 102, and the tool integration call subsystem 103, respectively, so as to specify different parameter call interfaces and return values ​​for different systems to ensure programmable coupling between systems. In the system, the function input parameters and variables can be mixed into the JSON string format as the prompt word of the large model. This method of mixed prompt words allows each system to use the model more flexibly and achieve the effect of one model for multiple purposes.

[0046] The intelligent interactive method based on the reservoir dynamic analysis scenario big model of the present invention has preliminarily realized the relevant functions of the reservoir dynamic analysis assistant, and verified its practical value and potential in assisting reservoir engineers in reservoir dynamic analysis through tasks such as well history combing query, well group analysis, and calling tool mapping in actual reservoir dynamic analysis cases. In terms of the construction of professional field big models, the 1,800 Chinese documents and 8 books used in the present invention as the data volume for incremental fine-tuning are difficult to fully support the construction needs of a professional field big model. Considering that the analysis capability of the reservoir dynamic analysis scenario big model (Reservoir Dynamic Analysis Toolkit) still needs to be based on the professional field big model's in-depth reading and comprehension capabilities of the oil and gas development professional field case prompt words or context scenarios, in the future, the high-quality oil and gas industry corpus can be expanded to further enhance the analysis and comprehension capabilities or directly conduct scene big model function research based on mature oil and gas professional field big models. In terms of reservoir dynamic analysis scenario functions, diversified test cases or data sets simulating different reservoir characteristics can be added in the future to improve the accuracy and adaptability of the Reservoir Dynamic Analysis Toolkit in various reservoir dynamic analysis scenarios, especially for the analysis of different geological conditions and reservoir types. In order to adapt to the continuous accumulation of oilfield data and changes in reservoir conditions, the Reservoir Dynamic Analysis Toolkit can establish a semi-automatic or automatic update mechanism according to whether the database structure has changed, so as to achieve data iterative update. At the same time, consider adding a human-computer collaborative workflow to allow reservoir engineers to intervene, correct or supplement the model output, and effectively integrate their feedback into the learning and optimization process of the model, thereby achieving deep integration and complementarity of human-computer collaboration. The intelligent interaction method based on the reservoir dynamic analysis scenario large model provided by the present invention provides a feasible technical route for the application of large models in reservoir dynamic analysis. In the future, by further expanding the pre-trained literature, fine-tuning the data set, and especially combining the field data of oil and gas field development, the performance of the reservoir dynamic analysis scenario large model (Reservoir Dynamic Analysis Toolkit) will be improved, and its application in auxiliary reservoir dynamic analysis will be further deepened. It is expected that auxiliary reservoir dynamic analysis can be realized through simple human-computer dialogue, replacing most of the basic human work, and greatly improving the efficiency of oil and gas field development.

[0047] The second embodiment of the present application is:

[0048] Second, see Figure 2-Figure 11The present invention also provides an intelligent interactive system based on a large model of oil reservoir dynamic analysis scenarios, comprising: a complex problem suggestion subsystem 101 for oil and gas reservoir engineering, a data mining and dynamic law analysis subsystem 102, and a tool integration calling subsystem 103; the complex problem suggestion subsystem 101 for oil and gas reservoir engineering, the data mining and dynamic law analysis subsystem 102, and the tool integration calling subsystem 103 are connected in sequence;

[0049] The oil and gas reservoir engineering complex problem suggestion subsystem 101 is used to analyze suggestion needs, answer questions in oil and gas reservoir engineering, and infer possible deeper phenomena based on some surface phenomenal statements, and give relevant measures and suggestions;

[0050] The data mining and dynamic law analysis subsystem 102 is used to meet the needs of data retrieval and law analysis, realize direct retrieval from natural language in the database, and analyze and reveal the periodic change law of oil reservoir production dynamic data;

[0051] The tool integration calling subsystem 103 is used to meet the needs of external auxiliary tool calling and classify the external tools needed to answer questions. Different questions use different tools, and similar questions use the same tools. The classification language task will classify specific questions into specific types and return the corresponding number of the type. A one-to-one association is pre-established with the label number and the function object or pointer, and the borrowing of the specified external tool is completed through function search and calling.

[0052] The present invention has developed three supporting fine-tuning technologies for the oil and gas reservoir engineering complex problem suggestion subsystem 101, the data mining and dynamic law analysis subsystem 102 and the tool integration calling subsystem 103: named entity recognition technology, Text-to-SQL statement conversion technology, classification task and tool calling technology. The named entity recognition technology extracts the key information in the question and divides it into three groups. Group one is a general question that can be answered without retrieval or calculation. Group two is the variables and keywords extracted from the user's question for retrieval from the database, and the summary statement in the user's question for selecting the tool. Group three is the intention information extracted from the user's question, including whether to visualize the query or calculation results, whether to require trend analysis of the query or calculation results, and whether to give suggestions based on the analysis results. The Text-to-SQL statement conversion technology is used to open up the SQL database, realize the conversion from natural language to SQL statements, and retrieve the data in the database at any time for viewing and analysis. The classification task and tool calling technology can accurately call the tool to complete the calculation of the corresponding indicators, open up the calling of commercial software functions (such as calling Petrel to draw the well profile), and further enhance the model's ability to analyze a single well. In addition, the large model can score the called tools based on the classifier classification task and return the tool label with the highest score. Then, based on the Callback function, the tool corresponding to the label is called for calculation, and finally the results of the relevant indicators are returned.

[0053] The oil and gas reservoir engineering complex problem suggestion subsystem 101 mainly answers complex problems in oil and gas reservoir engineering, and infers possible deeper phenomena based on some superficial phenomenal statements, and gives relevant measures and suggestions. Figure 4 Analysis, the oil and gas reservoir engineering complex problem suggestion subsystem 101 is based on the model A's own ability to read and understand the professional context, and relies on the dynamic and static data retrieved by the data mining and dynamic law analysis subsystem 102 and the tool integration calling subsystem 103 to output the key indicator calculation results or other tool calling results and the prompts in the chat page context to complete the functions of capacity evaluation, single well optimization production allocation, and measure effect evaluation.

[0054] The data mining and dynamic law analysis subsystem 102 has two main tasks: 1. Complete the conversion from natural language description to SQL query statements, mainly converting natural language questions mixed with some key search information into Select query statements in SQL language; 2. Complete simple trend analysis and capture of periodic laws based on given structured data. Figure 5Analysis, data mining and dynamic law analysis subsystem 102 mainly relies on the Text-to-SQL function of model B to complete the conversion of natural language to SQL search statements, uses the converted SQL statements to complete the query in MySql, and outputs the query results to output interfaces 2 and 3. The system will also mix the data results in 2 and 3 into a string in JSON format according to the trend analysis requirements, hand it over to model A to complete the trend analysis, and output the results to output port 1.

[0055] The tool integration calling subsystem 103 mainly completes the classification of problems. Different problems correspond to different labels, and similar problems have the same label. The classification language task will classify a specific problem into a specific type and return the preset number of the type. A one-to-one correspondence is established between the label number and the function object or pointer to achieve classification mapping. Figure 6 Analysis shows that in the tool integration calling subsystem 103, in order to ensure the independence and scalability of tool calling, it is recommended to build each tool independently as a service, and implement API calling through the RPC protocol and RESTful interface. For example, when developing a Petrel plug-in service, specific functions can be implemented through the Petrel SDK, such as the well profile drawing service. The service accepts the model name and the well numbers of the wells A and B to be connected as request parameters, calls the Petrel SDK interface to obtain the well coordinates through the well name, and then draws the well profile based on the coordinates of the two wells. The generated image is saved to the file server, and the URL of the file server where the image is located is passed in the returned JSON. Therefore, the API calling method of each tool service needs to be pre-written in different functions, and the call to the specified function is implemented by querying the function number. The function number is obtained by converting the relevant description containing the tool type information into a specific classification number through model C. In order to realize the flexible requirements for input and output data when calling the tool and the readability of the large model, the input and output streams of the plug-in tool service in this study are unified in JSON string format. Combined with Figure 7 analyze, Figure 7 The I. subsystem represents the proposed subsystem 101 for the complex problem of oil and gas reservoir engineering, Figure 7 The II. subsystem in the figure represents the data mining and dynamic law analysis subsystem 102, Figure 7 The III. subsystem in it represents the tool integration calling subsystem 103; the reservoir dynamic analysis scenario large model firstly decomposes the user's problem in detail through the named entity recognition technology, extracts entities and performs a deeper semantic analysis on the problem to identify the specific requirements of the problem such as data retrieval, computing requirements or analysis intentions.

[0056] Named entity recognition technology extracts key information from questions and divides them into three groups.

[0057] Group 1 is general questions that can be answered without retrieval or calculation (denoted as ①).

[0058] Group 2 is the variables and keywords extracted from the user's questions for searching the database (denoted as ②), and the summary statement in the user's questions for selecting tools (denoted as ③).

[0059] Group 3 is the intention information extracted from the user's question, including whether the query or calculation results want to be visualized (denoted as ④), whether trend analysis of the query or calculation results is required (denoted as ⑤), and whether recommended measures need to be given based on the analysis results (denoted as ⑥).

[0060] For data retrieval needs, the relevant data in the database is queried through the Text-to-SQL function in subsystem II. For calculation needs, the relevant technical indicators are calculated through the tool call function in subsystem III.

[0061] This method not only ensures the independence of each subsystem, but also efficiently shares information and resources, greatly improving the overall flexibility and scalability of the system, and providing operational space for the functional expansion of subsequent large-scale reservoir dynamic analysis scenario models.

[0062] The general questions are directly answered by the oil and gas reservoir engineering complex problem suggestion subsystem 101 (denoted as ⑦).

[0063] The tool integration calling subsystem 103 selects tools according to the summary statement in ⑦, sends a data retrieval request to the data mining and dynamic law analysis subsystem 102 and obtains the retrieval results (denoted as ⑧), and further calculates the key indicators of oilfield development based on the retrieval results (denoted as ⑨).

[0064] Then, the calculation results of oilfield development indicators and the results of trend analysis are combined to provide relevant suggestions and auxiliary measures (denoted as ⑩) for complex problems in oil and gas reservoir engineering. The subsystem combines the general answers and the suggested measures to output the results (corresponding to ).

[0065] In addition, you can control whether to visualize data and calculation results (denoted as ). The final result data transmission and data transfer between subsystems use JSON format, and data is transmitted within the subsystem in the form of function calls.

[0066] At the software architecture level, the large model of reservoir dynamic analysis scenarios adopts a separate development model of algorithm API development + back-end data coupling + front-end page display.

[0067] Among them, the algorithm part is based on the Pytorch framework and Transformers function package for incremental pre-training, fine-tuning and model calling, and forms an algorithm calling API for functional subsystem coupling.

[0068] After the subsystem algorithm embeds the processed data into a unified JSON format, the data is sent to the front-end user interface in real time through SocketIO, and the data is sent to the back-end database for storage through Http RESETfulAPI to form historical record data. When the front-end queries the historical records, the back-end database is queried through http and displayed on the front-end.

[0069] The backend builds subsystem coupling logic based on Http RESETfulAPI and SocketIO network protocol, and completes the response to frontend question requests in combination with user data, while recording historical data for user query. Based on the JSON data format, it provides developers with a simple and easy-to-use API that supports cross-platform and cross-browser real-time communication.

[0070] The front-end builds a user question-and-answer interface based on the Vue3 framework in NodeJS to facilitate interactive data visualization, and builds the user interface based on standard HTML, CSS, and JavaScript scripting language rules in Vue3. It uses component-based programming mode to directly embed mature chart generation modules such as Echart in the user UI.

[0071] The intelligent interactive system based on the large model of reservoir dynamic analysis scenario of the present invention provides a feasible technical route for the application of large models in reservoir dynamic analysis. In the future, by further expanding the pre-trained literature and fine-tuning the data set, especially combining the field data of oil and gas field development, the performance will be improved and the application in auxiliary reservoir dynamic analysis will be further deepened. It is expected that auxiliary reservoir dynamic analysis can be realized through simple human-computer dialogue, replacing most of the basic work of human beings, and greatly improving the efficiency of oil and gas field development.

[0072] What is disclosed above is only one or more preferred embodiments of the present application, and cannot be used to limit the scope of rights of the present application. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. An intelligent interactive method based on a large model of reservoir dynamic analysis scenarios, characterized in that: include: Through named entity recognition technology, user questions are broken down in detail, entities are extracted, and a deeper semantic analysis of the questions is performed to identify the specific needs of the questions, including analysis and suggestion needs, data retrieval and pattern analysis needs, and the need to call external auxiliary tools; Aiming at the needs of analytical suggestions, answer questions in oil and gas reservoir engineering, and infer the possible deeper phenomena based on some superficial phenomenal statements, and give relevant measures and suggestions; Aiming at the needs of data retrieval and law analysis, it realizes direct retrieval from natural language in the database, and analyzes and reveals the periodic change law of reservoir production dynamic data; In response to the need to call external auxiliary tools, the external tools needed to answer questions are classified. Different questions use different tools, and similar questions use the same tools. The classification language task will classify specific questions into specific types and return the corresponding number of the type. A one-to-one association is established in advance between the label number and the function object or pointer, and the borrowing of the specified external tool is completed through function search and call. The large model of reservoir dynamic analysis scenario adopts a separate development mode that combines algorithm API development, back-end data coupling, and front-end page display to build the software architecture.

2. The intelligent interaction method based on the reservoir dynamic analysis scenario large model according to claim 1, characterized in that: The specific steps of using named entity recognition technology to decompose user questions in detail, extract entities, and conduct deeper semantic analysis on the questions to identify the specific needs of the questions include: The key information in user questions is extracted and divided into three groups. Group 1 is general questions that can be answered without retrieval or calculation. Group 2 is variables and keywords extracted from user questions for database retrieval, and summary statements in user questions for tool selection. Group 3 is intention information extracted from user questions, including whether you want to visualize the query or calculation results, whether you require trend analysis of the query or calculation results, and whether you need to give recommended measures based on the analysis results.

3. The intelligent interaction method based on the reservoir dynamic analysis scenario large model according to claim 1, characterized in that: The analysis reveals the periodic variation patterns of reservoir production dynamic data using statistical methods.

4. The intelligent interaction method based on the reservoir dynamic analysis scenario large model according to claim 1, characterized in that: The specific steps for realizing direct retrieval from natural language in a database and analyzing and revealing the periodic change patterns of reservoir production dynamic data in accordance with the needs of data retrieval and pattern analysis include: completing the conversion from natural language description to SQL query statements, converting natural language questions mixed with some key retrieval information into Select query statements in SQL language, and completing simple trend analysis and capturing periodic patterns based on given structured data.

5. An intelligent interactive system based on a large model of reservoir dynamic analysis scenarios, using an intelligent interactive method based on a large model of reservoir dynamic analysis scenarios as claimed in any one of claims 1 to 4, characterized in that: It includes a complex problem suggestion subsystem for oil and gas reservoir engineering, a data mining and dynamic law analysis subsystem and a tool integration calling subsystem; the complex problem suggestion subsystem for oil and gas reservoir engineering, the data mining and dynamic law analysis subsystem and the tool integration calling subsystem are connected in sequence; The oil and gas reservoir engineering complex problem suggestion subsystem is used to analyze and suggest needs, answer questions in oil and gas reservoir engineering, and infer possible deeper phenomena based on some surface phenomenal statements, and give relevant measures and suggestions; The data mining and dynamic law analysis subsystem is used to meet the needs of data retrieval and law analysis, realize direct retrieval from natural language in the database, and analyze and reveal the periodic change law of oil reservoir production dynamic data; The tool integration calling subsystem is used to meet the needs of external auxiliary tool calling and classify the external tools needed to answer questions. Different questions use different tools, and similar questions use the same tools. The classification language task will classify specific questions into specific types and return the corresponding number of the type. A one-to-one association is established in advance using the label number and the function object or pointer, and the borrowing of the specified external tool is completed through function search and calling.

Images