Construction method, device and equipment of logging interpretation agent, medium and product

Abstract

The embodiment of the invention provides a construction method, device and equipment of a logging interpretation agent, a medium and a product, and relates to the technical field of oil exploration and development. According to the method, a task planning and execution cooperation mechanism is introduced, a logging interpretation process originally completed step by step depending on professionals is converted into an automatic processing process, a first model understands user requirements and generates a systematic execution plan, and a second model automatically calls corresponding function modules to complete various interpretation tasks according to the plan. Through a unified interface protocol and standardized packaging, integrated deployment of various interpretation tools and models is realized, and links such as data processing, parameter calculation and result analysis can be automatically joined. According to the construction mode, the manual intervention link is reduced, and the full-process automation from data input to report generation is realized, so that the logging interpretation efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of petroleum exploration and development technology, and in particular to a method, apparatus, equipment, medium and product for constructing a well logging interpretation intelligent agent. Background Technology

[0002] Well logging interpretation refers to the process of analyzing formation lithology, reservoir distribution, fluid properties, and reservoir parameters based on formation parameters obtained from well logging equipment, combined with geological theory and engineering experience. Well logging interpretation is the core link in transforming raw well logging data into geological understanding and engineering decision-making basis, directly affecting the accuracy of resource exploration, the rationality of development plans, and the efficiency of resource extraction. Therefore, well logging interpretation plays a crucial role in the field of petroleum exploration and development technology.

[0003] In existing technologies, well logging interpretation is mainly done manually. Professionals collect various well logging curve data, conduct comprehensive analysis of the data based on theories such as rock physics and well logging geology, perform parameter calculations and fluid identification using tools such as empirical charts and theoretical models, and manually complete reservoir evaluation and write interpretation reports.

[0004] However, existing technologies suffer from low logging interpretation efficiency. Current logging interpretation processes rely on professionals to process and manually interpret massive amounts of data step-by-step. From data preprocessing and parameter calculation to comprehensive evaluation, a significant amount of time and manpower is required, resulting in a lengthy overall interpretation process that struggles to quickly respond to the high-efficiency logging interpretation needs in exploration and development. Summary of the Invention

[0005] This application provides a method, apparatus, equipment, medium, and product for constructing a well logging interpretation intelligent agent, in order to solve the problem of low well logging interpretation efficiency in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for constructing a well logging interpretation intelligent agent, including:

[0007] Obtain the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call functional modules according to the execution plan, the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process;

[0008] The interface protocols of multiple first functional modules, the first model, and the second model are unified to obtain multiple second functional modules, a third model, and a fourth model; wherein, the third model is the model obtained by unifying the interface protocol of the first model, and the fourth model is the model obtained by unifying the interface protocol of the second model.

[0009] Obtain multiple metadata of each of the second functional modules, encapsulate and publish each of the second functional modules and the multiple metadata of each of the second functional modules to obtain multiple published services; wherein, each of the published services is used for the fourth model to call to implement the tasks in each of the sub-processes;

[0010] The system acquires control logic, system prompts, and core instructions. The control logic indicates how the third model calls the fourth model and how the fourth model calls the multiple published services. The system prompts indicate the order in which the multiple published services are called. The core instructions refer to the strategies used to call and manage the multiple published services.

[0011] The multiple published services, the third model, the fourth model, the control logic, the system prompts, and the core instructions are assembled to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

[0012] Secondly, embodiments of this application provide a device for constructing a well logging interpretation intelligent agent, comprising:

[0013] The first acquisition module is used to acquire the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call the functional modules according to the execution plan, and the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process;

[0014] A unification module is used to unify the interface protocols of multiple first functional modules, the first model, and the second model to obtain multiple second functional modules, a third model, and a fourth model; wherein, the third model is a model obtained by unifying the interface protocols of the first model, and the fourth model is a model obtained by unifying the interface protocols of the second model;

[0015] The encapsulation and publishing module is used to obtain multiple metadata of each of the second functional modules, encapsulate and publish each of the second functional modules and the multiple metadata of each of the second functional modules to obtain multiple published services; wherein, each of the published services is used for the fourth model to call to implement the tasks in each of the sub-processes;

[0016] The second acquisition module is used to acquire control logic, system prompts, and core instructions; wherein, the control logic is used to indicate how the third model calls the fourth model, and how the fourth model calls the multiple published services, the system prompts are used to indicate the calling order of the multiple published services, and the core instructions refer to the strategies used to call and manage the multiple published services;

[0017] An assembly module is used to assemble the multiple published services, the third model, the fourth model, the control logic, the system prompts, and the core instructions to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

[0018] Thirdly, embodiments of this application provide an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0019] The memory stores computer-executed instructions;

[0020] When the processor executes the computer execution instructions stored in the memory, it is used to implement the method for constructing a well logging interpretation agent as described in any of the first aspects.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method for constructing a well logging interpretation agent as described in any of the first aspects.

[0022] Fifthly, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, is used to implement the method for constructing a well logging interpretation agent as described in any of the first aspects.

[0023] This application provides a method, apparatus, equipment, medium, and product for constructing a well logging interpretation intelligent agent. By introducing a mechanism for task planning and execution collaboration, the well logging interpretation process, which originally relied on professionals to complete step by step, is transformed into an automated process. The first model can understand user needs and generate a systematic execution plan, while the second model automatically calls the corresponding functional modules to complete various interpretation tasks according to the plan. Through a unified interface protocol and standardized encapsulation, the integration and allocation of various interpretation tools and models are realized, enabling automatic connection between data processing, parameter calculation, and result analysis. This construction method reduces manual intervention and realizes full-process automation from data input to report generation, thereby improving the efficiency of well logging interpretation. Attached Figure Description

[0024] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0025] Figure 1 This is a schematic diagram illustrating an application scenario of the method for constructing a well logging interpretation intelligent agent provided in the embodiments of this application;

[0026] Figure 2 A flowchart illustrating the method for constructing a well logging interpretation intelligent agent provided in the embodiments of this application. Figure 1 ;

[0027] Figure 3 A flowchart illustrating the method for constructing a well logging interpretation intelligent agent provided in the embodiments of this application. Figure 2 ;

[0028] Figure 4 The structural design diagram of the well logging interpretation intelligent agent based on a large language model is provided for the embodiments of this application;

[0029] Figure 5 A flowchart illustrating the method for constructing a well logging interpretation intelligent agent provided in the embodiments of this application. Figure 3 ;

[0030] Figure 6 A schematic diagram of the structure of the device for constructing a well logging interpretation intelligent agent provided in the embodiments of this application;

[0031] Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application.

[0032] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0034] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that "first" and "second" do not necessarily imply difference. It should be noted that in the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner. In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more.

[0035] It should be noted that the phrase "at...time" in the embodiments of this application can refer to the instant at which a certain situation occurs, or to a period of time after the occurrence of a certain situation; the embodiments of this application do not specifically limit this. Furthermore, the method, apparatus, device, medium, and product for constructing a well logging interpretation intelligent agent provided in the embodiments of this application are merely examples; a method, apparatus, device, medium, and product for constructing a well logging interpretation intelligent agent may also include more or less content.

[0036] To facilitate a clear description of the technical solutions in the embodiments of this application, some terms and technologies involved in the embodiments of this application will be briefly introduced below:

[0037] Core repositioning refers to the process of matching and correcting the depth of core samples obtained during drilling with the depth of logging curves. Core repositioning establishes a depth correspondence by identifying characteristic marker layers in the core and logging curves, and ultimately accurately calibrates the direct observation results such as rock properties and sedimentary structures from the core analysis into the logging depth system.

[0038] Lithology identification refers to the interpretation technique of using the response characteristics of well logging curves to determine the rock types of downhole formations. By analyzing the characteristic patterns of these well logging responses and combining them with regional geological knowledge, different lithologies in the formation can be effectively distinguished, providing basic geological information for reservoir evaluation, sedimentary facies analysis, and oil and gas reservoir description.

[0039] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0040] To clearly understand the technical solution of this application, the existing technical solutions will first be described in detail. Well logging interpretation is the core step in transforming raw well logging data into geological knowledge and engineering decision-making basis, directly affecting the accuracy of resource exploration, the rationality of development plans, and the efficiency of resource extraction.

[0041] In existing technologies, well logging interpretation is primarily done manually. Professionals collect various well logging curve data, evaluate the reservoir using empirical charts and theoretical models, and write interpretation reports. The current well logging interpretation process relies on professionals to process and manually interpret massive amounts of data step by step, resulting in a lengthy overall interpretation process. Therefore, existing technologies suffer from low well logging interpretation efficiency.

[0042] Therefore, addressing the low efficiency of existing well logging interpretation technologies, this research found that a well logging interpretation technology solution can be constructed that automatically plans the well logging interpretation task process and automatically calls relevant functional modules to perform data processing and interpretation analysis. This solution includes: ① Constructing an integrated automated well logging interpretation tool platform that integrates functional modules for well logging data import, parameter calculation, reservoir evaluation, and report generation. By pre-setting process logic, it achieves automatic connection of each step, reducing the time cost of manual tool switching and shortening the overall interpretation cycle. ② Utilizing historical well logging data to train data-driven intelligent models, allowing the models to learn the well logging response patterns corresponding to different formation lithologies and reservoir characteristics. This enables the models to automatically identify data anomalies, calculate reservoir parameters, and determine fluid properties, replacing manual step-by-step data analysis and interpretation. ③ Leveraging the understanding and planning capabilities of large language models, an intelligent interactive system supporting natural language input can be developed. Users can issue tasks via voice or text commands, and the system automatically parses the requirements and generates a standardized interpretation process, while simultaneously optimizing the task execution path through intelligent planning.

[0043] Specifically, an intelligent agent system based on a large language model can be constructed to achieve intelligent processing of the entire process through a collaborative mechanism of task planning and automatic execution. This system breaks down well logging interpretation into standardized sub-processes. The planning module parses user instructions and generates execution sequences, while the execution module automatically allocates professional tools and computational models according to the plan. Finally, a unified interface protocol is used to achieve seamless connection and automated operation of each link, thereby improving interpretation efficiency.

[0044] This application discloses a method, apparatus, device, medium, and product for constructing a well logging interpretation intelligent agent. By introducing a mechanism for task planning and execution collaboration, it transforms the well logging interpretation process, which originally relied on professionals to complete step by step, into an automated process. The first model can understand user needs and generate a systematic execution plan, while the second model automatically calls the corresponding functional modules to complete various interpretation tasks according to the plan. Through a unified interface protocol and standardized encapsulation, it achieves the integration and allocation of various interpretation tools and models, enabling automatic connection between data processing, parameter calculation, and result analysis. This construction method reduces manual intervention and achieves full automation from data input to report generation, thereby improving the efficiency of well logging interpretation.

[0045] Based on the above-mentioned inventive discovery, the technical solution of this application is proposed.

[0046] The following describes the application scenarios of the well logging interpretation intelligent agent construction method provided in the embodiments of the present invention. Figure 1 This is a schematic diagram illustrating an application scenario for the method of constructing a well logging interpretation intelligent agent provided in an embodiment of this application. For example... Figure 1 As shown, this application scenario includes a user terminal 101 and a server 102. The server 102 obtains the first model, the second model, and the first functional modules of each sub-process. The server 102 unifies the interface protocols of multiple first functional modules, the first model, and the second model to obtain multiple second functional modules, the third model, and the fourth model. The server 102 obtains multiple metadata of each second functional module, encapsulates and publishes each second functional module and its metadata to obtain multiple published services. The server 102 obtains control logic, system prompts, and core instructions. The server 102 assembles the multiple published services, the third model, the fourth model, the control logic, the system prompts, and the core instructions to obtain a well logging interpretation agent. The user terminal 101 uses the well logging interpretation agent to perform well logging interpretation.

[0047] Figure 2 A flowchart illustrating the method for constructing a well logging interpretation intelligent agent provided in the embodiments of this application. Figure 1 .like Figure 2 As shown, in this embodiment, the execution entity of this invention is a server. Therefore, the method for constructing a well logging interpretation intelligent agent provided in this embodiment includes the following steps:

[0048] S201. Obtain the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call functional modules based on the execution plan, the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process.

[0049] Specifically, models with natural language understanding and task planning capabilities can be selected from existing algorithm libraries or development frameworks as planning models, and models with tool invocation and execution capabilities can be selected as execution models. Each stage of the well logging interpretation process, including data preprocessing, parameter calculation, and fluid identification, is configured as an independent task module. This step provides the basic core components for the subsequent assembly of the well logging interpretation intelligent agent, clarifying in advance the key components responsible for requirement understanding and planning, module allocation, and execution of specific sub-process tasks, thus laying the component foundation for automating the well logging interpretation process.

[0050] The process includes several sub-processes, such as data collection, quality control, environmental correction, logging curve alignment, core repositioning, missing value handling, lithology identification, reservoir parameter prediction, fluid identification, reservoir evaluation and classification, and interpretation report generation.

[0051] S202. Unify the interface protocols of multiple first functional modules, first models and second models to obtain multiple second functional modules, third models and fourth models; wherein, the third model is the model obtained by unifying the interface protocols of the first model, and the fourth model is the model obtained by unifying the interface protocols of the second model.

[0052] Specifically, a unified interface protocol specification can be formulated, including data input and output format standards, call triggering conditions, and interactive response mechanisms. Subsequently, the interfaces of these modules and models are modified according to this specification, adjusting their data transmission format, parameter definitions, and call logic to ensure they all follow unified communication rules. This results in multiple second functional modules, third models, and fourth models. The purpose of this step is to eliminate interface barriers between different components, ensuring they can recognize each other, communicate smoothly, and coordinate calls. This provides a compatibility foundation for subsequently encapsulating functional modules into services, implementing orderly calls between models and services, and ultimately assembling them into an intelligent agent.

[0053] Among them, interface protocol unification refers to setting a common data exchange and communication standard for all functional modules and models. It can adopt an HTTP-based RESTful API architecture and uniformly use JSON as the data serialization format. At the same time, it stipulates that all service calls follow the predefined OpenAPI specification and clearly define the endpoint, request method, parameter structure and status code meaning of each service.

[0054] S203. Obtain multiple metadata of each second functional module, encapsulate and publish each second functional module and its multiple metadata to obtain multiple published services; wherein, each published service is used for the fourth model to call to implement the tasks in each sub-process.

[0055] Specifically, metadata can be extracted from the design specifications, operating parameters, and interaction records of each second functional module, including descriptions of its functional scope, input / output requirements, operating dependencies, processing capabilities, and version information. Then, using a unified service encapsulation format, the program logic of each second functional module and its corresponding metadata are integrated into an independent service unit. Subsequently, these units are registered in a callable resource pool through a pre-defined publishing mechanism, forming multiple published services. This step is used to transform each functional module into a service entity with clear attributes and calling standards, making it easier for the subsequent model to accurately identify service capabilities and make precise calls based on the metadata.

[0056] S204. Obtain control logic, system prompts, and core instructions; whereby the control logic indicates how the third model calls the fourth model, and how the fourth model calls multiple published services; the system prompts indicate the order in which multiple published services are called; and the core instructions refer to the strategies used to call and manage multiple published services.

[0057] Specifically, based on the collaborative requirements of the entire well logging interpretation process and the functional positioning of each component, the conditions for the third model to trigger the fourth model's call and the criteria for the fourth model to select published services can be identified to form control logic. By analyzing the typical execution order of each sub-process in historical interpretation tasks, service call sequence rules that ensure the continuity of interpretation steps can be extracted as system prompts. Referring to common scenarios in service calls, such as abnormal retries and resource allocation, specific strategies for managing service operation can be summarized to obtain core instructions. This step is used to clarify the collaboration rules and operating principles between components, providing clear guidance for model service calls and service collaborative execution in the agent, ensuring that the subsequently assembled agent can complete the well logging interpretation task in sequence and efficiently.

[0058] The core commands include: data processing commands, model calling commands, tool calling commands, result analysis commands, and report generation commands. For data processing commands, when the data quality is poor, the well logging data quality intelligent evaluation model is called first for evaluation, and then a data correction tool or model completion tool is selected based on the evaluation results. For model calling commands, when performing lithology identification tasks, if the amount of data is sufficient, the well logging intelligent lithology identification model is called; if the amount of data is insufficient, a lithology identification tool based on natural gamma curves is used. A conflict resolution strategy between tools and models can be formulated. Based on task priority, historical calling effects, and resource consumption, the calling order of multiple callable models or tools for the same task can be determined. A command execution feedback mechanism can also be established, whereby the executor returns the execution status and results after executing the command. If the execution fails, the command retry or switching strategy is triggered.

[0059] The control logic is a rule system that clarifies the calling relationships between the third and fourth models and between the fourth model and multiple published services in the intelligent agent. Its core includes the specific conditions under which the third model triggers the fourth model to call, such as task type and data readiness status, as well as the judgment criteria for the fourth model to select different published services, such as service function matching degree and task requirement scenario. Ultimately, it provides clear interaction guidance for the model to call services and for services to perform collaborative execution.

[0060] The system prompt is a set of rules that standardize the order of calling multiple published services. Its purpose is to clarify the order of service calls. For example, if the data preprocessing service needs to be completed first, and then the lithology identification service is executed, this will ensure that the well logging interpretation steps are coherent and uninterrupted, and avoid interruption of the interpretation process or error in the results due to disordered service call order.

[0061] The core instructions are specific strategies and operational guidelines for calling and managing multiple published services. They not only clarify the execution logic of various instructions, such as prioritizing the use of the intelligent evaluation model for well logging data quality when the data quality is poor, and switching to the lithology identification tool when the data volume is insufficient, but also cover strategies for resolving conflicts between tools and models. They determine the order of calls and the instruction execution feedback mechanism based on task priority, historical results, etc. Overall, they are used to ensure efficient service calls, handle anomalies, and support the smooth completion of well logging interpretation tasks.

[0062] S205. Assemble multiple published services, third models, fourth models, control logic, system prompts, and core instructions to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

[0063] Specifically, a unified coordination framework can be built to connect the third model, the fourth model, and multiple published services through pre-defined interfaces. At the same time, control logic, system prompts, and core instructions are embedded in the framework, clarifying the interaction paths and operating rules of each part. This allows the execution plan generated by the third model to call published services according to the rules through the fourth model, and the execution of each service proceeds in an orderly manner according to system prompts. The core instructions play a role in service invocation and management. This integration forms a well logging interpretation agent. This step is used to integrate the scattered components and rules into an organic whole, enabling it to autonomously plan and execute complete well logging interpretation tasks based on user input.

[0064] One of the multiple large language models can be defined as the planner, used to plan the overall well logging interpretation process and issue instructions to the executors. Other large language models can be defined as executors, used to call computational models and tools. The planner large language model and the executor large language model can be connected through a unified interface protocol. It also integrates the encapsulated small-scale intelligent well logging interpretation scientific computing model and tools. Multiple sets of test cases are designed to cover different types of well logging data, complex geological scenarios, and abnormal data conditions to verify the accuracy of the agent in planning tasks, calling tools and models, processing data, and generating interpretation results. Stress tests and stability tests are used to evaluate the agent's response speed and long-term reliability under high-concurrency tasks. If problems such as tool call failure or interpretation result errors occur during the test, the root cause of the problem is located, and the interface protocol, model parameters, or control logic are adjusted until the test is passed. The qualified agent is then packaged into a deployable software package, released to the specified production environment, and accompanied by usage instructions and maintenance documentation.

[0065] This embodiment provides a caching optimization method that transforms the well logging interpretation process, which relies on professionals to complete step-by-step, into an automated process by introducing a task planning and execution coordination mechanism. The first model understands user needs and generates a systematic execution plan, while the second model automatically calls the corresponding functional modules to complete various interpretation tasks according to the plan. Through a unified interface protocol and standardized encapsulation, it achieves the integrated deployment of various interpretation tools and models, enabling automatic connection between data processing, parameter calculation, and result analysis. This approach reduces manual intervention, automates the entire process from data input to report generation, and improves well logging interpretation efficiency.

[0066] In one possible design, S205 assembles multiple published services, a third model, a fourth model, control logic, system hints, and core instructions to obtain a logging interpretation agent, including:

[0067] S2051. Obtain the deployment environment for the third model.

[0068] Specifically, by investigating the hardware conditions required for the actual operation of the third model, such as the processing power and storage capacity of computing devices, as well as software support, such as dependent program versions, runtime library requirements, network conditions, and resource usage limitations, these environmental factors affecting the operation of the model can be identified, thereby obtaining the deployment environment of the third model. The purpose of this step is to clarify the specific conditions required for the operation of the third model, providing a basis for subsequent adjustments to the model based on this environment, and ensuring that the model can play a stable and efficient role in actual deployment scenarios.

[0069] S2052. Based on the deployment environment of the third model and multiple adjustment data, the third model is adjusted to obtain the fifth model.

[0070] Specifically, based on the deployment environment of the third model, such as hardware processing capabilities and software dependency requirements, the structural complexity and operating parameters of the model can be adjusted for adaptability. For example, the model architecture can be simplified to adapt to limited computing resources, or the program interface can be modified to match the software version of the operating environment. Then, combined with multiple adjustment data, historical logging data and logging interpretation data can be used to supplement the training or parameter calibration of the adjusted model, so that the execution plan generated is more in line with the actual logging interpretation scenario. Finally, the fifth model is obtained. The purpose of this step is to enable the model to adapt to the actual deployment environment conditions and to optimize performance by combining specific logging data and interpretation rules, so as to ensure that it can stably and accurately generate the required execution plan in the subsequent intelligent agent.

[0071] For example, for large models with available computing resources and corpora, deployed on private servers or private clouds, well logging domain knowledge can be used to fine-tune the planner's large language model. Combined with cue word engineering, control logic can be developed to guide the planner in planning the well logging interpretation process in a logical order. Based on information from smaller models or tools, system prompts can be used to restrict the executor's access to models and tools. For large models that cannot be trained on large language models or are deployed on public platforms, knowledge base retrieval enhancement technology can be used to connect to a well logging domain knowledge base, allowing planners to consult materials for task planning. Combined with cue word engineering, control logic can be developed to guide the planner in planning the well logging interpretation process in a logical order. Based on information from smaller models or tools, system prompts can be used to restrict the executor's access to models and tools.

[0072] S2053. Assemble multiple published services, the fifth model, the fourth model, control logic, system prompts, and core instructions to obtain the well logging interpretation agent.

[0073] Specifically, an integrated framework can be built to connect the fifth model, the fourth model, and multiple published services through a unified interface. At the same time, control logic, system prompts, and core instructions are integrated into the framework's operating mechanism. This clarifies how the fourth model is triggered after the fifth model generates the plan, how the fourth model calls published services according to rules, how each service executes in an orderly manner according to prompts, and how the core instructions regulate the operation during the process. Through such integration, a well logging interpretation agent is formed. This step is used to integrate the environment-adapted model with other components and rules into a whole, enabling it to adapt to the deployment environment and efficiently complete well logging interpretation tasks according to user needs.

[0074] The technical effect of this solution in this embodiment is that by optimizing the planning model according to the deployment environment, the adaptability of the well logging interpretation agent under different technical conditions is improved. This method identifies the model's deployment environment and implements corresponding adjustment strategies, enabling the planning model to better adapt to the needs of specific operational scenarios, thereby ensuring that the agent maintains a high level of task planning capability and professional judgment accuracy under various deployment conditions.

[0075] In one possible design, the deployment environment includes private servers and public platforms. S2052, based on the deployment environment of the third model and multiple adjustment data, the third model is adjusted to obtain a fifth model, including:

[0076] S20521. In response to the deployment environment being a public platform, perform permission verification on the public platform and obtain the verification results; among which, the verification results include allow adjustment and disallow adjustment. Allow adjustment is used to indicate that the public platform allows adjustment of the third model, and disallow adjustment is used to indicate that the public platform does not allow adjustment of the third model.

[0077] Specifically, one can consult the publicly available application programming interface (API) documentation of the public cloud platform and attempt to perform model fine-tuning operations. The returned authorization status or error message will then determine whether permission to modify the model is granted. This step is used to clarify the rule boundaries when utilizing public resources in advance, ensuring that any subsequent customization operations on the planning model comply with platform regulations and avoid technical violations or service interruptions.

[0078] S20522. In response to the verification result indicating that adjustment is permitted, the third model is adjusted based on multiple adjustment data to obtain the fifth model.

[0079] Specifically, historical logging data and logging interpretation data can be used as training materials to fine-tune the planning model, enabling it to deeply understand the professional processes and decision-making logic in the field of logging interpretation. This step is used to improve the professionalism and accuracy of the planning model when facing actual logging interpretation tasks, ensuring that the generated execution plan not only conforms to domain standards but also effectively utilizes existing expert experience and data patterns.

[0080] S20523. In response to the verification result indicating that adjustment is not allowed, the third model is linked to the preset well logging domain knowledge base so that the third model generates an execution plan that conforms to the preset well logging interpretation specifications based on the domain knowledge in the well logging domain knowledge base.

[0081] Specifically, by configuring external knowledge base links and setting prompt word rules for the planning model, it can proactively query and construct task sequences based on the well logging interpretation standard procedures, key parameters, and decision rules stored in the knowledge base when generating plans. This step is used to constrain its output within the framework of professional standards through external guidance without modifying the model's internal parameters, thereby ensuring that feasible solutions with domain expertise can still be generated even in constrained environments.

[0082] The "attachment" process refers to establishing a stable and interactive connection between the third-party model and a pre-defined well logging domain knowledge base. In terms of interface configuration, a RESTful API or JSON-RPC protocol is used to build the communication interface between the model and the knowledge base, clearly defining the model request format and the knowledge base return format. Logically, pre-defined trigger rules are used for the model to generate execution plans. When the model enters a critical stage, it automatically initiates a knowledge base query request, prioritizing the extraction of well logging interpretation specifications from the database as the basis for planning. After attachment, each step of the model's execution plan generation requires real-time calls to the knowledge base and reference to the specifications. If the planned content conflicts with the specifications in the knowledge base, the model is triggered to readjust the plan. Simultaneously, the knowledge base supports periodic updates of specification data; simply synchronizing the interface data ensures that the model always generates plans based on the latest specifications. This clearly defines the specific operation path of the attachment and guarantees that the execution plan conforms to the well logging interpretation specifications.

[0083] S20524. In response to the deployment environment being a private server, obtain the amount of computing resources and domain corpus of the private server; wherein, the amount of domain corpus is used to represent the scale of knowledge data for well logging interpretation task planning stored on the private server.

[0084] Specifically, resource monitoring commands can be run to investigate the server's processor, memory, and video memory capacity, while simultaneously evaluating the data scale and quality of well logging interpretation documents, case libraries, and historical task data stored locally on the server. This step is used to accurately assess whether the current private environment possesses the necessary hardware computing power and domain knowledge reserves to fully fine-tune the planning model, providing crucial decision-making basis for subsequently selecting the most suitable model optimization path.

[0085] S20525. In response to the amount of computing resources being greater than or equal to a preset resource threshold, and the amount of domain corpus being greater than or equal to a preset corpus threshold, the third model is adjusted based on multiple adjustment data to obtain the fifth model.

[0086] Specifically, once it's confirmed that the local server has sufficient computing power and stores a adequate amount of well logging data, this historical data and interpretation rules can be used as a training set to fine-tune the planning model. This step fully utilizes the superior local environment, injecting domain knowledge into the model to improve the accuracy and professionalism of its well logging interpretation task planning, making the generated solutions more aligned with actual business needs.

[0087] S20526. In response to the amount of computing resources being less than the resource threshold, or the amount of domain corpus being less than the corpus threshold, the third model is connected to the well logging domain knowledge base so that the third model generates an execution plan that conforms to the well logging interpretation specification based on the domain knowledge in the well logging domain knowledge base.

[0088] Specifically, the planning model can be connected to a pre-built well logging expertise base by configuring the application programming interface (API), and rules can be set so that the model must refer to the process specifications and decision logic in this knowledge base during planning. This step is used to ensure that the task flow output by the planning model still meets the professional standards and technical requirements of well logging interpretation when local environmental resources are insufficient to support in-depth model optimization, through external guidance.

[0089] The technical effect of this solution in this embodiment is as follows: differentiated adjustment strategies are formulated for public platforms and private servers respectively: on the public platform, the model fine-tuning or knowledge base integration scheme is intelligently selected through permission verification; on the private server, the optimization path is dynamically selected based on the sufficiency of computing resources and domain corpus. This flexible processing method not only fully leverages the performance advantages of model fine-tuning, but also ensures basic professional capabilities through knowledge base integration, thereby ensuring that the planning model has the task planning capability that conforms to well logging interpretation specifications under various technical constraints.

[0090] Figure 3 A flowchart illustrating the method for constructing a well logging interpretation intelligent agent provided in the embodiments of this application. Figure 2 In this embodiment, in Figure 2 Based on the provided embodiments, the method for constructing a well logging interpretation intelligent agent is further explained. The method for constructing a well logging interpretation intelligent agent includes:

[0091] S301. Acquire multiple adjustment data and multiple models; wherein, the multiple adjustment data includes multiple historical logging data and multiple logging interpretation data. The multiple historical logging data refers to the production data of the preset logging within a preset time period in the past. The multiple logging interpretation data is used to represent the verified deterministic rules and decision logic in each sub-process. The multiple models are preset multiple logging interpretation domain models.

[0092] Specifically, well logging production data from past well logging operations and exploration operation archives can be collected as historical well logging data. Deterministic rules and decision-making logic that have been confirmed and are valid in each sub-process can be extracted from completed and verified accurate well logging interpretation reports and experience summaries accumulated by professionals, serving as well logging interpretation data. Pre-defined well logging interpretation models can be selected from commonly used technical literature in the field of well logging interpretation, industry-application model libraries, or relevant research results. This process provides necessary basic data support and initial model resources for subsequent construction of computational models based on historical well logging data, tool development based on well logging interpretation data, and the selection of the first and second models, ensuring that subsequent component construction and selection work have reliable data and model sources.

[0093] Among them, multiple models refer to a pre-selected set of specialized models focused on the field of well logging interpretation. These models can be selected from commonly used technical literature in the field of well logging interpretation and model libraries that have been practically applied in the industry. Their core purpose is to provide initial model resources for subsequent processes. They are used to support the construction of computational models from historical well logging data and the construction of tools from well logging interpretation data. More importantly, they serve as a foundational pool for selecting the first and second models, ensuring that there are reliable initial models that meet the professional needs of well logging interpretation available for subsequent model selection.

[0094] S302. A calculation model is constructed based on multiple historical logging data; the calculation model is used to execute data-driven tasks in each sub-process.

[0095] Specifically, the collected historical logging data can be preprocessed first to remove outliers and fill in missing information, ensuring the data meets the standards for model training. Then, based on the needs of data-driven tasks in each sub-process, such as reservoir parameter calculation and logging curve correction, suitable algorithms, such as regression analysis and deep learning methods, can be selected. Subsequently, the selected algorithm is trained using the preprocessed historical logging data. During the process, the model output results are continuously verified using historical data that was not used in the training. The model parameters are adjusted based on the verification results until the model accuracy reaches the preset requirements, resulting in a computational model. This step is used to build a component that can autonomously handle data-driven tasks in each sub-process, replacing the manual data calculation method that relies on experience, thereby improving the accuracy and efficiency of data processing.

[0096] S303. Formulas are constructed based on multiple well logging interpretation data to obtain tools; the tools are used to execute tasks in each sub-process that rely on empirical rules.

[0097] Specifically, the collected well logging interpretation data can be sorted out first, and the verified empirical rules and decision-making logic can be extracted, such as the correlation between well logging parameters corresponding to different lithologies and the threshold conditions for judging fluid properties. Then, these empirical rules can be transformed into quantifiable mathematical formulas, clarifying the definition and calculation logic of each parameter in the formula. Subsequently, the formulas can be tested using previously verified interpretation cases, and the calculation results of the formulas can be compared with the actual interpretation conclusions. If there are deviations, the parameters or logic in the formulas can be adjusted until the formulas can accurately reflect the empirical rules, and finally, a tool can be obtained. The purpose of this step is to transform the scattered empirical rules into standardized executable tools, replacing the way of manually relying on personal experience to judge and process sub-process tasks, improving the processing efficiency of experience-based tasks, and providing the task execution capability based on empirical rules for the subsequent formation of the first functional module.

[0098] Among them, the computational model is the intelligent logging interpretation scientific computational model, which includes: traditional machine learning model and deep learning model, specifically the intelligent evaluation model for logging data quality, the intelligent correction model for logging data, the intelligent alignment model for logging curves, the intelligent repositioning model for core data, the intelligent completion model for logging data, the intelligent lithology identification model for logging, the prediction model for logging reservoir parameters, the identification model for logging fluids, and the classification model for logging reservoirs.

[0099] The tools are code libraries or toolkits that do not use machine learning algorithms. They complete well logging interpretation tasks through empirical formulas, including well logging data collection tools, well logging data quality control tools based on cross-plot analysis, well logging data environment correction tools based on charts, well logging curve alignment tools based on peak alignment, core repositioning tools based on intuitive comparison methods, missing value mean or minimum value filling tools, lithology identification tools based on natural gamma curves, well logging reservoir parameter prediction tools based on mechanistic models, fluid identification tools based on reservoir parameters, reservoir evaluation and classification tools based on expert experience, and well logging interpretation report generation tools.

[0100] The computational model and tool modules can be specifically used for: retrieving well logging data files as needed using machine learning models or various code tools; evaluating and analyzing the quality of well logging data; performing environmental correction on well logging data; aligning curves in well logging data; repositioning core samples; filling in missing values ​​in the data; identifying lithology based on well logging data; predicting reservoir parameters based on well logging data; identifying fluids based on reservoir parameter prediction results; evaluating and classifying reservoirs based on reservoir parameter prediction results and fluid identification results; and writing well logging interpretation reports in a prescribed format based on the results obtained from various calculations or analyses.

[0101] S304. Filter through multiple models to obtain the first model and the second model.

[0102] Specifically, we can first clarify the core capabilities required by the first and second models. For example, the first model needs to be able to understand user input and generate reasonable execution plans, while the second model needs to be able to accurately call functional modules according to the plan. Then, based on these capability requirements, we can test multiple preset well logging interpretation domain models, evaluate the performance of each model in understanding requirements, generating plans, and allocating modules, and select the two models that perform best in the corresponding capabilities as the first and second models, respectively. This step is used to determine the suitable core model from the existing domain models, providing suitable components for the key parts of the intelligent agent responsible for task planning and module calling in the subsequent construction, ensuring that the intelligent agent can effectively realize the connection from requirements to execution.

[0103] Among them, multiple models can be large language models, using two or more large language models working together. The large language models are those published on public platforms and available for use. These models cannot be fine-tuned, and the versions can be the same or different. One of the large language models is defined as the planner, used to plan the overall well logging interpretation process and issue instructions to the executors. The other large language models are defined as executors, used to translate the planner's commands into the calling language of the models and tools, and automatically select the appropriate tools or models for invocation. According to the platform where the models are located and the requirements of the large language models themselves, a unified interface protocol is used to realize the invocation between the executors and the scientific computing models and tools. The large language models are those built on private servers or private servers, and can be trained using corpora. The versions can be the same or different, depending on the platform where the models are located and the requirements of the large language models themselves.

[0104] The planner's large language model can be used to: receive natural language information from users, clarify users' needs, plan the process to complete the task based on users' needs and knowledge of the logging field, issue instructions to the executors in sequence, and the executors will return information to the planner after each instruction. The planner will revise the previously planned process based on the returned information and issue the next instruction. The task is completed when the logging interpretation report generation tool or model is called, without waiting for feedback from the executors.

[0105] The executor's large language model can be used to: receive instructions from the planner, analyze the small-scale models or tools required by the instructions, call the corresponding models or tools according to the calling method of the small-scale models or tools, integrate the results returned by the small-scale models or tools and return them to the planner in the form of natural language. When calling the well logging interpretation report generation tool or model, there is no need to integrate the information of the small-scale models or tools, nor is there a need to return information to the planner.

[0106] For example, GPT-4 can be selected as the planner, and Claude-3.5 as the executor. The planner is responsible for receiving the user's instruction to "perform reservoir evaluation on well logging data of a certain block," formulating a task flow of "data quality check → environmental correction → reservoir parameter prediction → reservoir evaluation," and issuing instructions to the executor. The executor translates the instructions into code for calling models and tools. For example, when calling the intelligent well logging data quality evaluation model, the interface protocol specifies that the input is the path to the original well logging data file, and the output is a quality score and anomaly report.

[0107] GPT-4 as the planner and Claude-3.5 as the executor are just one example among many feasible options. The planner can also be other general-purpose models with strong reasoning capabilities, such as Gemini Ultra or Claude 3 Opus, and the executor can be replaced with other models that are good at code and tool calls, such as CodeLlama. These choices should be determined based on the specific task's requirements for planning accuracy, execution efficiency, or domain adaptability.

[0108] Docker containers can be used to encapsulate the intelligent evaluation model for well logging data quality and the quality control tool based on cross-plot analysis. The container configuration file specifies the model name "LogDataQualityModel", its function "evaluating well logging data quality", the calling method "data is passed via HTTP POST request", and the input mode "JSON format data file". The encapsulated model is then deployed to a Kubernetes cluster for deployment by a large language model.

[0109] The GPT-4 planner can be fine-tuned using 100,000 sets of well logging interpretation case data accumulated in a certain oilfield. A prompt can be set to "prioritize data quality checks; if the data quality score is below 80, initiate the data correction process." For the Claude-3.5 executor, a system prompt can be set to restrict it to using lithology identification tools only if the data volume is greater than 100 sets; if insufficient, a lithology identification tool based on natural gamma curves should be used.

[0110] A core instruction set can be defined. When the executor receives the instruction to "process logging data," it invokes the intelligent logging data quality evaluation model. If the score is below 70, it invokes the intelligent logging data correction model. For lithology identification tasks, if the data volume is sufficient (>200 sets), the intelligent logging lithology identification model (based on the Transformer architecture) is invoked; if the data volume is insufficient, a lithology identification tool based on natural gamma curves is invoked. An instruction execution feedback mechanism is established. If the model invocation fails, the executor reports an error code to the planner, who then readjusts the instructions or switches tools.

[0111] The planner GPT-4 and executor Claude-3.5 can be connected via a RESTful API interface, integrating pre-packaged scientific computing models and tools. Test cases are designed, including scenarios such as normal logging data, anomalous data with 50% missing data, and complex data with noise interference, to verify the agent's processing capabilities under different conditions. Stress testing is conducted using JMeter to simulate 500 concurrent logging interpretation tasks, ensuring the agent's response time is within 10 seconds. After successful testing, the agent is packaged into a Docker image, deployed to the oilfield's internal server, and an operation manual and API call documentation are provided.

[0112] Figure 4 The well logging interpretation agent structure design diagram based on a large language model provided in the embodiments of this application is as follows: Figure 4 As shown, there is one and only one planner, while there may be one or more executors. The planner interacts with the user and issues instructions to the executors. The planner needs to possess well logging expertise. When deployed locally, a well logging knowledge corpus can be used to fine-tune and train the planner. When the planner cannot be fine-tuned or trained, knowledge base retrieval enhancement techniques can be used to ensure that the overall process specified by the planner meets the requirements of well logging interpretation. After receiving instructions from the planner, the executor calls the corresponding model and / or tool according to the scientific computing model and tool usage instructions, following the prescribed format. After the model and / or tool is successfully called, the executor returns the call result to the planner.

[0113] S305. Obtain the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call functional modules based on the execution plan, the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process.

[0114] S306. Unify the interface protocols of multiple first functional modules, first models and second models to obtain multiple second functional modules, third models and fourth models; wherein, the third model is the model obtained by unifying the interface protocols of the first model, and the fourth model is the model obtained by unifying the interface protocols of the second model.

[0115] S307. Obtain multiple metadata of each second functional module, encapsulate and publish each second functional module and its multiple metadata to obtain multiple published services; wherein, each published service is used for the fourth model to call to implement the tasks in each sub-process.

[0116] S308, obtain control logic, system prompts and core instructions; where the control logic is used to indicate how the third model calls the fourth model, and how the fourth model calls multiple published services, the system prompts are used to indicate the order in which multiple published services are called, and the core instructions refer to the strategies used to call and manage multiple published services.

[0117] S309. Assemble multiple published services, third models, fourth models, control logic, system prompts, and core instructions to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

[0118] S305-S309 are similar to S201-S205, and will not be described again in this embodiment.

[0119] The technical advantages of this solution in this embodiment are as follows: the computational model built based on historical logging data enables automated processing of data-driven tasks, while the tools built based on knowledge in the logging interpretation domain encapsulate mature expert experience and rules. This modular construction approach not only ensures the professionalism and accuracy of each sub-process task execution, but also matches core models suitable for planning and execution roles to subsequent processes through a model selection mechanism, thereby ensuring the reliability and effectiveness of the agent in completing the logging interpretation task from the source.

[0120] In one possible design, S304, multiple models are selected to obtain a first model and a second model, including:

[0121] S3041. Based on preset evaluation rules, evaluate multiple models to obtain the planning capability score and execution capability score of each model; wherein, the planning capability score of each model is used to represent the ability of each model in process planning, and the execution capability score of each model is used to represent the ability of each model in calling functional modules.

[0122] Specifically, we can first set specific criteria for evaluating planning capabilities, such as the rationality, completeness, and matching degree of the generated execution plan with user needs. Then, we can set specific criteria for evaluating execution capabilities, such as the accuracy, efficiency, and exception handling effect of calling functional modules, forming a preset evaluation rule. Subsequently, for each model, we test its generated plan performance by inputting typical user needs, and score it according to the planning capability criteria to obtain a planning capability score. We then test its calling module performance by simulating the execution plan, and score it according to the execution capability criteria to obtain an execution capability score. This step is used to quantify the scores and intuitively reflect the differences in planning and execution capabilities of each model, providing an objective basis for subsequently determining the first and second models.

[0123] The evaluation rules can be a quantitative, multi-dimensional indicator system developed in conjunction with the well logging interpretation task scenario. This system includes: prediction accuracy, using the deviation rate between the predicted values ​​of core well logging parameters such as clay content and porosity and the actual formation data as the standard; a deviation rate ≤5% is considered full marks, and >15% is considered unqualified; response efficiency, requiring a response time ≤30 seconds for processing 1000 sets of historical well logging data to ensure task timeliness; robustness, simulating scenarios with missing data and noise interference, requiring a model's effective output probability ≥80%; and interface compatibility, verifying the model's adaptation to the unified protocol, requiring an interface call success rate ≥99%. Simultaneously, weights are assigned to each indicator based on task priority, such as 40% for accuracy, 30% for response efficiency, 20% for robustness, and 10% for compatibility, and a comprehensive score is obtained through weighted calculation.

[0124] S3042. The model corresponding to the maximum planning capability score among multiple models is determined as the first model, and the model corresponding to the maximum execution capability score among multiple models is determined as the second model.

[0125] Specifically, the planning capability scores of all models can be compared to identify the model with the highest score, which is then designated as the first model. Next, the execution capability scores of all models can be compared to identify the model with the highest score, which is then designated as the second model. This step is used to accurately select the two models with the best performance in planning and execution capabilities from multiple models. These two models are then responsible for generating execution plans and calling functional modules, respectively. This ensures that the intelligent agent built subsequently has the strongest processing capabilities in key areas, providing core support for efficiently completing well logging interpretation tasks.

[0126] The technical effect of this solution in this embodiment is as follows: Based on preset evaluation rules, the capabilities of candidate models are evaluated, examining their process planning and function invocation capabilities respectively. Suitable models are then assigned to corresponding roles based on the scoring results. This screening mechanism ensures that planners possess task decomposition and process design capabilities, while executors have tool allocation and instruction execution capabilities, thereby guaranteeing the overall coordination and efficiency of the intelligent agent's operation at the architectural level.

[0127] Figure 5 A flowchart illustrating the method for constructing a well logging interpretation intelligent agent provided in the embodiments of this application. Figure 3 In this embodiment, in Figure 2 Based on the provided embodiments, the method for constructing a well logging interpretation intelligent agent is further explained. The method for constructing a well logging interpretation intelligent agent includes:

[0128] S501. Obtain the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call functional modules based on the execution plan, the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process.

[0129] S502. Unify the interface protocols of multiple first functional modules, first models and second models to obtain multiple second functional modules, third models and fourth models; wherein, the third model is the model obtained by unifying the interface protocols of the first model, and the fourth model is the model obtained by unifying the interface protocols of the second model.

[0130] S503. Obtain multiple metadata of each second functional module, encapsulate and publish each second functional module and its multiple metadata to obtain multiple published services; wherein, each published service is used for the fourth model to call to implement the tasks in each sub-process.

[0131] S504, Obtain control logic, system prompts, and core instructions; wherein, the control logic is used to indicate how the third model calls the fourth model, and how the fourth model calls multiple published services, the system prompts are used to indicate the order in which multiple published services are called, and the core instructions refer to the strategies used to call and manage multiple published services.

[0132] S505. Multiple published services, third models, fourth models, control logic, system prompts, and core instructions are assembled to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

[0133] S501-S505 are similar to S201-S205, and will not be described again in this embodiment.

[0134] S506. Obtain user input statements and multiple test cases; each test case includes multiple test input data and multiple test output data.

[0135] Specifically, well logging interpretation requirements can be collected from actual scenarios related to well logging interpretation, and these requirements can be used as user input statements. Input information containing different formation conditions and well logging data, along with corresponding correct interpretation results, can be extracted from previously completed and verified accurate well logging interpretation data. Each set of input information and its corresponding correct result can be organized into a set of test cases to obtain multiple sets of test cases. This step is used to provide verification basis for subsequent testing of the accuracy of the well logging interpretation agent, and at the same time, prepare real input content for the agent to finally receive actual requirements and generate interpretation reports, ensuring that subsequent testing and actual application can be effectively connected.

[0136] S507. Input multiple test input data into the well logging interpretation agent to obtain multiple actual output data.

[0137] Specifically, we can first check whether the format of multiple test input data matches the receiving requirements of the well logging interpretation agent. If there is a mismatch, we can adjust the data format to an appropriate state and input the processed test input data into the well logging interpretation agent. The agent completes operations such as execution plan generation, function module calling and data processing according to its internal logic. Then, we collect the results output by the agent after processing each test input data to obtain multiple actual output data. This step is used to obtain the actual processing results of the agent in the test scenario, which provides a practical result basis for comparing the actual output data with the test output data and calculating the interpretation accuracy, and to determine whether the interpretation capability of the agent meets the standard.

[0138] S508. Calculate the interpretation accuracy of the well logging interpretation agent based on multiple test output data and multiple actual output data.

[0139] Specifically, the test output data corresponding to each test input data can be compared with the actual output data one by one to determine whether the two are consistent in key interpretation results such as reservoir parameters and fluid properties. The proportion of test cases with consistent results to the total number of test cases is counted to obtain the interpretation accuracy of the logging interpretation agent. This step is used to quantitatively evaluate whether the interpretation effect of the agent meets expectations, and provides objective numerical basis for judging whether the agent needs to be optimized and whether it can be put into practical application.

[0140] S509. In response to the fact that the explanation accuracy of the interpreting agent is less than the preset accuracy threshold, root cause analysis is performed based on multiple actual output data to obtain a problem location report.

[0141] Specifically, the differences between the actual output data and the test output data can be categorized. By combining the execution trajectory of the agent when processing this data, the specific steps that caused the differences can be traced. For example, it can be determined whether the execution plan generated by the third model is unreasonable, whether the fourth model's service call is biased, or whether the processing result of a certain published service is incorrect. Then, the specific reasons for these differences and the corresponding steps are compiled into a report to obtain a problem localization report. The purpose of this step is to accurately identify the root cause of the agent's inaccurate explanation, providing a clear direction for subsequent targeted optimization of the agent and ensuring that the optimization work directly addresses the core of the problem.

[0142] S510. Optimize the well logging interpretation agent based on the problem location report to obtain the optimized well logging interpretation agent, and calculate the accuracy of the optimized well logging interpretation agent.

[0143] Specifically, based on the identified problem areas in the problem localization report, if the problem lies in the model's planning capability, the training data or parameters of the third model are adjusted; if the problem lies in service calls, the calling logic of the fourth model is optimized; if the problem lies in the service itself, the corresponding published service is improved. Through these targeted adjustments, an optimized logging interpretation agent is obtained. The original test input data is then input into the optimized agent to obtain new actual output data, which is compared with the test output data to calculate its accuracy. The purpose of this step is to improve the interpretation capability of the agent by solving the identified problems, so that its accuracy gradually approaches or reaches the preset threshold, providing a guarantee for the subsequent generation of reliable interpretation reports and practical application.

[0144] S511. In response to the fact that the accuracy of the optimized well logging interpretation agent is greater than or equal to the accuracy threshold, the user input statement is input into the optimized well logging interpretation agent to obtain an interpretation report; wherein, the interpretation report is a file generated by the optimized well logging interpretation agent when performing well logging interpretation tasks based on the user input statement.

[0145] Specifically, the completeness and clarity of the user's input statement can be confirmed first. If there are any ambiguities, the necessary information can be supplemented through preset interaction logic. Then, the sorted user input statement is input into the optimized well logging interpretation agent. The agent completes the entire process of well logging interpretation task based on internal model planning, service calls, and rule control. Finally, a document containing interpretation conclusions, analysis basis, and relevant data is generated, resulting in an interpretation report. The purpose of this step is to transform the user's actual needs into specific well logging interpretation results, providing directly usable professional conclusions for exploration and development decisions, and realizing the transformation of the agent from testing and verification to practical application.

[0146] S512. The optimized logging interpretation agent is packaged to obtain a software package, and the software package is published to the preset production environment.

[0147] Specifically, the optimized well logging interpretation agent can first integrate its components such as models, services, and rules, sort out and package its required dependency files, configure startup parameters and runtime environment adaptation settings to form an independently runnable software package, confirm the deployment requirements of the preset production environment, apply for the corresponding deployment permissions, transfer the software package to the designated location in the production environment, and perform simple functional tests after deployment to confirm that it can start and respond normally. Finally, the release is completed. This step is used to transform the optimized agent into a software form that can be directly used in actual business scenarios, so that it can be stably integrated into the production process and provide automated support for daily well logging interpretation work.

[0148] The explanatory report may include:

[0149] Geological Background: This report focuses on an oilfield in a basin, where the main exploration target formations are sandstone and mudstone. Based on existing data, the sandstone and mudstone in this area exhibit good sedimentary continuity, with moderate porosity and permeability, making them ideal oil and gas reservoirs.

[0150] Well logging data processing: Calculation of VSH (Volume of Shale): The VSH content is mainly calculated using natural gamma ray (GR) and photoelectric factor (PEF), and corrected for based on the wellbore diameter (CAL). This report uses the logging_test model to predict the VSH content based on conventional well logging curves.

[0151] Porosity calculation: Porosity is mainly calculated using density logging (DEN) and neutron logging, and then corrected for by clay content. This report uses the logging_test model to predict porosity based on conventional logging curves.

[0152] Water saturation (SW) calculation: Water saturation is mainly calculated using deep and medium resistivity, and then corrected for by porosity and clay content. This report uses the logging_test model to predict water saturation based on conventional well logging curves.

[0153] Oil, gas, and water evaluation based on well logging data: Based on the characteristics of the logging curve of Well No. 100, combined with geological logging data and data from adjacent wells, the target interval was processed and evaluated, identifying all oil-bearing layers and some representative water layers, totaling 6 layers. The lithology, physical properties, electrical properties, and hydrocarbon-bearing characteristics of the oil (gas) and water layers are described and analyzed below:

[0154] 1. Layer 1: 15912.5 feet (ft) to 16207.5 ft, thickness: 295.0 ft. Analysis of well logging curves shows a deep lateral resistivity of 5.435895, a mid-depth lateral resistivity of 5.359395, a natural gamma ray of 14.183899, and a neutron count of 0.100574. Calculated results: clay content 7.785168%, porosity 10.601635%, water saturation 64.560509%. Overall assessment: water-bearing layer.

[0155] 2. Layer 2: 16207.5ft to 16477.5ft, thickness: 270.0ft. Analysis of the logging curves shows an average deep lateral resistivity of 3.829558, a mid-depth lateral resistivity of 3.672414, an average natural gamma ray of 29.835674, and an average neutron count of 0.117145. Calculation results show a clay content of 20.496523%, porosity of 9.808496%, and water saturation of 79.173593%. The overall assessment is: dry layer.

[0156] 3. Layer 3: 16477.5ft to 16567.5ft, thickness: 90.0ft. Analysis of the logging curves shows an average deep lateral resistivity of 2.752625, a mid-depth lateral resistivity of 2.641979, an average natural gamma ray of 51.972068, and an average neutron count of 0.200535. Calculated results: clay content 39.661270%, porosity 10.092776%, water saturation 91.715800%. Overall assessment: water-bearing layer.

[0157] 4. Layer 4: 16567.5ft to 17057.5ft, thickness: 490.0ft. Analysis of the logging curves shows an average deep lateral resistivity of 3.843370, a mid-depth lateral resistivity of 3.521115, an average natural gamma ray of 150.271262, and an average neutron count of 0.332561. Calculated results show a clay content of 93.886018%, porosity of 5.774355%, and water saturation of 76.857453%. Overall assessment: non-reservoir.

[0158] 5. Layer 5: 17057.5ft to 17182.5ft, thickness: 125.0ft. Analysis of the logging curves shows an average deep lateral resistivity of 0.616488, an average mid-depth lateral resistivity of 0.572041, an average natural gamma ray of 37.777756, and an average neutron count of 0.223414. Calculated results: clay content 25.586128%, porosity 21.469674%, water saturation 87.297367%. Overall assessment: water-bearing layer.

[0159] 6. Layer 6: 17182.5ft to 17237.5ft, thickness: 55.0ft. Analysis of the logging curves shows an average deep lateral resistivity of 273.624193, an average mid-depth lateral resistivity of 1046.093854, an average natural gamma ray of 17.274352, and an average neutron count of 0.212187. Calculated results: clay content 9.808295%, porosity 24.831443%, water saturation 62.645295%. Overall assessment: water-bearing layer.

[0160] The technical effect of this solution in this embodiment is that it evaluates the accuracy of the agent's interpretation through systematic testing, performs root cause analysis and targeted optimization on substandard aspects, and forms a closed-loop improvement process. This quality assurance system not only ensures that the final released agent can generate interpretation reports that meet professional requirements, but also improves the stability and usability of the system through optimization, providing reliable technical support for the automated processing of well logging interpretation tasks.

[0161] Figure 6 This is a schematic diagram of the structure of the well logging interpretation intelligent agent construction device provided in the embodiments of this application. Figure 6 As shown, the apparatus for constructing the well logging interpretation agent includes:

[0162] The first acquisition module 601 is used to acquire the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call functional modules based on the execution plan, the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process.

[0163] The unification module 602 is used to unify the interface protocols of multiple first functional modules, first models and second models to obtain multiple second functional modules, third models and fourth models; wherein, the third model is the model obtained by unifying the interface protocols of the first model, and the fourth model is the model obtained by unifying the interface protocols of the second model.

[0164] The encapsulation and release module 603 is used to obtain multiple metadata of each second functional module, encapsulate and release each second functional module and its multiple metadata to obtain multiple released services; among them, each released service is used for the fourth model to call to implement the tasks in each sub-process.

[0165] The second acquisition module 604 is used to acquire control logic, system prompts, and core instructions. The control logic is used to indicate how the third model calls the fourth model and how the fourth model calls multiple published services. The system prompts are used to indicate the order in which multiple published services are called. The core instructions refer to the strategies used to call and manage multiple published services.

[0166] Assembly module 605 is used to assemble multiple published services, third models, fourth models, control logic, system prompts and core instructions to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

[0167] In one possible design, the first functional module of each sub-process includes a computational model and tools, a device for constructing the logging interpretation agent, and also includes:

[0168] The third acquisition module is used to acquire multiple adjustment data and multiple models. The multiple adjustment data includes multiple historical logging data and multiple logging interpretation data. The multiple historical logging data refers to the production data of the logging in the past within a preset time period. The multiple logging interpretation data is used to represent the verified deterministic rules and decision logic in each sub-process. The multiple models are preset multiple logging interpretation domain models.

[0169] The first construction module is used to build a model based on multiple historical logging data to obtain a computational model; the computational model is used to execute data-driven tasks in each subprocess.

[0170] The second construction module is used to construct formulas based on multiple well logging interpretation data to obtain tools; the tools are used to execute tasks in each subprocess that rely on empirical rules.

[0171] The filtering module is used to filter multiple models to obtain the first model and the second model.

[0172] In one possible design, the filtering module includes:

[0173] The evaluation unit is used to evaluate multiple models based on preset evaluation rules to obtain the planning capability score and the execution capability score of each model. The planning capability score of each model is used to represent the ability of each model in process planning, and the execution capability score of each model is used to represent the ability of each model in calling functional modules.

[0174] The determination unit is used to determine the model corresponding to the maximum planning capability score among multiple models as the first model, and the model corresponding to the maximum execution capability score among multiple models as the second model.

[0175] In one possible design, assembly module 605 includes:

[0176] The acquisition unit is used to acquire the deployment environment of the third model.

[0177] The adjustment unit is used to adjust the third model based on the deployment environment of the third model and multiple adjustment data to obtain the fifth model.

[0178] The assembly unit is used to assemble multiple published services, the fifth model, the fourth model, control logic, system prompts, and core instructions to obtain a well logging interpretation agent.

[0179] In one possible design, the deployment environment includes private servers and a public platform, and the adjustment unit includes:

[0180] The permission verification component is used to perform permission verification on the public platform in response to the deployment environment being a public platform, and obtain the verification results. The verification results include "allowed adjustment" and "disallowed adjustment". "Allowed adjustment" means that the public platform allows the third-party model to be modified, and "disallowed adjustment" means that the public platform does not allow the third-party model to be modified.

[0181] The first adjustment component is used to adjust the third model based on multiple adjustment data in response to the verification result indicating that adjustment is allowed, to obtain the fifth model.

[0182] The first docking component is used to dock the third model with the preset well logging domain knowledge base in response to the verification result that adjustment is not allowed, so that the third model generates an execution plan that conforms to the preset well logging interpretation specification based on the domain knowledge in the well logging domain knowledge base.

[0183] The acquisition component is used to acquire the computing resources and domain corpus of the private server in response to the deployment environment being a private server; wherein, the domain corpus is used to represent the scale of the knowledge data of the well logging interpretation task planning stored on the private server.

[0184] The second adjustment component is used to adjust the third model based on multiple adjustment data in response to the amount of computing resources being greater than or equal to a preset resource threshold, and the amount of domain corpus being greater than or equal to a preset corpus threshold, to obtain the fifth model.

[0185] The second connection component is used to connect the third model to the well logging domain knowledge base in response to the amount of computing resources being less than the resource threshold or the amount of domain corpus being less than the corpus threshold, so that the third model can generate an execution plan that conforms to the well logging interpretation specification based on the domain knowledge in the well logging domain knowledge base.

[0186] In one possible design, the apparatus for constructing the well logging interpretation agent also includes:

[0187] The fourth acquisition module is used to acquire user input statements and multiple test cases; each test case includes multiple test input data and multiple test output data.

[0188] The first calculation module is used to input multiple test input data into the well logging interpretation agent to obtain multiple actual output data.

[0189] The second calculation module is used to calculate the interpretation accuracy of the well logging interpretation agent based on multiple test output data and multiple actual output data.

[0190] The root cause analysis module is used to perform root cause analysis based on multiple actual output data when the explanation accuracy of the explanation agent is less than a preset accuracy threshold, and obtain a problem location report.

[0191] The optimization module is used to optimize the well logging interpretation agent based on the problem location report, obtain the optimized well logging interpretation agent, and calculate the accuracy of the optimized well logging interpretation agent.

[0192] The report generation module is used to respond to the fact that the accuracy of the optimized well logging interpretation agent is greater than or equal to the accuracy threshold, by inputting the user input statement into the optimized well logging interpretation agent to obtain an interpretation report; wherein, the interpretation report is a file generated by the optimized well logging interpretation agent in performing well logging interpretation tasks based on the user input statement.

[0193] The publishing module is used to encapsulate the optimized well logging interpretation agent into a software package and publish the software package to the preset production environment.

[0194] The well logging interpretation intelligent agent construction device provided in this embodiment can execute... Figure 2 , Figure 3 and Figure 5 The technical solution of the embodiment of the well logging interpretation intelligent agent construction method shown herein, its implementation principle and technical effect are similar to Figure 2 , Figure 3 and Figure 5 The embodiment of the method for constructing a well logging interpretation agent shown is similar and will not be described in detail here.

[0195] Figure 7 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device includes at least one processor 710 and a memory 720. The electronic device also includes a communication component 730. The processor 710, memory 720, and communication component 730 are connected via a bus 740.

[0196] In a specific implementation, at least one processor 710 executes computer execution instructions stored in memory 720, causing at least one processor 710 to implement a method for constructing a well logging interpretation intelligent agent according to the above embodiment.

[0197] The specific implementation process of processor 710 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0198] In the above embodiments, it should be understood that the processor 710 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0199] The memory 720 may include high-speed RAM memory, and may also include non-volatile memory NVM, such as at least one disk storage.

[0200] Bus 740 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Bus 740 can be divided into address bus, data bus, control bus, etc. For ease of illustration, the bus 740 in the accompanying drawings of this application is not limited to only one bus or one type of bus.

[0201] The above description of the functions implemented by electronic devices and main control devices has introduced the solutions provided by the embodiments of the present invention. It is understood that, in order to implement the above functions, the electronic device or main control device includes hardware structures and / or software modules corresponding to the execution of each function. By combining the units and algorithm steps of the various examples described in the embodiments of the present invention, the embodiments of the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solutions of the embodiments of the present invention.

[0202] This application also provides a computer-readable storage medium storing computer-executable instructions. When executed by a processor, these instructions are used to implement a method for constructing a well logging interpretation intelligent agent as described in the above embodiments. In the specific implementation of the aforementioned method for constructing a well logging interpretation intelligent agent, each module can be implemented as a processor.

[0203] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0204] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, the processor and the readable storage medium can exist as discrete components in an electronic device or a host device.

[0205] This application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement a method for constructing a well logging interpretation intelligent agent as described in the above embodiments.

[0206] The computer program is stored in a readable storage medium, and at least one processor can read the computer program from the readable storage medium and execute the computer program to perform the scheme provided in any of the above embodiments.

[0207] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0208] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for constructing a well logging interpretation intelligent agent, characterized in that, include: Obtain the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call functional modules according to the execution plan, the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process; The interface protocols of multiple first functional modules, the first model, and the second model are unified to obtain multiple second functional modules, a third model, and a fourth model; wherein, the third model is the model obtained by unifying the interface protocol of the first model, and the fourth model is the model obtained by unifying the interface protocol of the second model. Obtain multiple metadata of each of the second functional modules, encapsulate and publish each of the second functional modules and the multiple metadata of each of the second functional modules to obtain multiple published services; wherein, each of the published services is used for the fourth model to call to implement the tasks in each of the sub-processes; The system acquires control logic, system prompts, and core instructions. The control logic indicates how the third model calls the fourth model and how the fourth model calls the multiple published services. The system prompts indicate the order in which the multiple published services are called. The core instructions refer to the strategies used to call and manage the multiple published services. The multiple published services, the third model, the fourth model, the control logic, the system prompts, and the core instructions are assembled to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

2. The method for constructing a well logging interpretation intelligent agent according to claim 1, characterized in that, The first functional module of each sub-process includes a computational model and tools. Before obtaining the first model, the second model, and the first functional module of each sub-process, the process further includes: Acquire multiple adjustment data and multiple models; wherein, the multiple adjustment data includes multiple historical logging data and multiple logging interpretation data, the multiple historical logging data refers to the production data of the preset logging within a preset time period in the past, the multiple logging interpretation data is used to represent the verified deterministic rules and decision logic in each of the sub-processes, and the multiple models are preset multiple logging interpretation domain models; The computational model is constructed based on the multiple historical logging data; wherein, the computational model is used to execute the data-driven tasks in each of the sub-processes. The tool is obtained by constructing formulas based on the multiple well logging interpretation data; wherein, the tool is used to execute tasks that rely on empirical rules in each of the sub-processes; The multiple models are filtered to obtain the first model and the second model.

3. The method for constructing a well logging interpretation intelligent agent according to claim 2, characterized in that, The step of filtering the multiple models to obtain the first model and the second model includes: Based on preset evaluation rules, the multiple models are evaluated to obtain a planning capability score and an execution capability score for each model; wherein, the planning capability score of each model is used to represent the ability of each model in process planning, and the execution capability score of each model is used to represent the ability of each model in calling functional modules. The model corresponding to the maximum planning capability score among the multiple models is determined as the first model, and the model corresponding to the maximum execution capability score among the multiple models is determined as the second model.

4. The method for constructing a well logging interpretation intelligent agent according to claim 2, characterized in that, The assembly of the multiple published services, the third model, the fourth model, the control logic, the system prompts, and the core instructions to obtain the well logging interpretation agent includes: Obtain the deployment environment of the third model; Based on the deployment environment of the third model and multiple adjustment data, the third model is adjusted to obtain the fifth model; The well logging interpretation agent is obtained by assembling the multiple published services, the fifth model, the fourth model, the control logic, the system prompts, and the core instructions.

5. The method for constructing a well logging interpretation intelligent agent according to claim 4, characterized in that, The deployment environment includes private servers and public platforms. Based on the deployment environment of the third model and multiple adjustment data, the third model is adjusted to obtain a fifth model, including: In response to the deployment environment being the public platform, an access control check is performed on the public platform to obtain a check result; wherein, the check result includes "allow adjustment" and "disallow adjustment", "allow adjustment" indicates that the public platform allows adjustments to the third model, and "disallow adjustment" indicates that the public platform does not allow adjustments to the third model; In response to the verification result indicating that adjustment is permitted, the third model is adjusted based on the multiple adjustment data to obtain the fifth model; In response to the verification result indicating that adjustment is not allowed, the third model is linked to a preset well logging domain knowledge base, so that the third model generates an execution plan that conforms to the preset well logging interpretation specifications based on the domain knowledge in the well logging domain knowledge base; In response to the deployment environment being the private server, the computing resources and domain corpus of the private server are obtained; wherein, the domain corpus is used to represent the scale of the knowledge data of well logging interpretation task planning stored on the private server; In response to the amount of computing resources being greater than or equal to a preset resource threshold, and the amount of domain corpus being greater than or equal to a preset corpus threshold, the third model is adjusted according to the multiple adjustment data to obtain the fifth model; In response to the amount of computing resources being less than the resource threshold, or the amount of domain corpus being less than the corpus threshold, the third model is connected to the well logging domain knowledge base, so that the third model generates an execution plan that conforms to the well logging interpretation specification based on the domain knowledge in the well logging domain knowledge base.

6. The method for constructing a well logging interpretation agent according to claim 1, characterized in that, After assembling the multiple published services, the third model, the fourth model, the control logic, the system prompts, and the core instructions to obtain the well logging interpretation agent, the process further includes: Obtain user input statements and multiple test cases; wherein each test case includes multiple test input data and multiple test output data; The multiple test input data are input into the well logging interpretation agent to obtain multiple actual output data. The interpretation accuracy of the well logging interpretation agent is calculated based on the multiple test output data and the multiple actual output data. In response to the fact that the explanation accuracy of the explanation agent is less than a preset accuracy threshold, root cause analysis is performed based on the multiple actual output data to obtain a problem location report; The well logging interpretation agent is optimized based on the problem location report to obtain an optimized well logging interpretation agent, and the accuracy of the optimized well logging interpretation agent is calculated. In response to the fact that the accuracy of the optimized well logging interpretation agent is greater than or equal to the accuracy threshold, the user input statement is input into the optimized well logging interpretation agent to obtain an interpretation report; wherein, the interpretation report is a file generated by the optimized well logging interpretation agent when performing well logging interpretation tasks based on the user input statement; The optimized logging interpretation agent is packaged to obtain a software package, and the software package is published to a preset production environment.

7. A device for constructing a well logging interpretation intelligent agent, characterized in that, include: The first acquisition module is used to acquire the first model, the second model, and the first functional modules of each sub-process; wherein, the first model is used to generate an execution plan based on user input, the second model is used to call the functional modules according to the execution plan, and the preset well logging interpretation process includes multiple sub-processes, and the first functional module of each sub-process is used to implement the preset tasks in each sub-process; A unification module is used to unify the interface protocols of multiple first functional modules, the first model, and the second model to obtain multiple second functional modules, a third model, and a fourth model; wherein, the third model is a model obtained by unifying the interface protocols of the first model, and the fourth model is a model obtained by unifying the interface protocols of the second model; The encapsulation and publishing module is used to obtain multiple metadata of each of the second functional modules, encapsulate and publish each of the second functional modules and the multiple metadata of each of the second functional modules to obtain multiple published services; wherein, each of the published services is used for the fourth model to call to implement the tasks in each of the sub-processes; The second acquisition module is used to acquire control logic, system prompts, and core instructions; wherein, the control logic is used to indicate how the third model calls the fourth model, and how the fourth model calls the multiple published services, the system prompts are used to indicate the calling order of the multiple published services, and the core instructions refer to the strategies used to call and manage the multiple published services; An assembly module is used to assemble the multiple published services, the third model, the fourth model, the control logic, the system prompts, and the core instructions to obtain a well logging interpretation agent; wherein, the well logging interpretation agent is used to plan and execute well logging interpretation tasks according to user input.

8. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; When the processor executes the computer execution instructions stored in the memory, it is used to implement the method for constructing a well logging interpretation agent as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method for constructing a well logging interpretation agent as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The system includes a computer program, which, when executed by a processor, is used to implement the method for constructing a well logging interpretation agent as described in any one of claims 1 to 6.

Images