A Method and System for Drilling Operation Assistance and Fault Analysis Based on Multi-Agent Collaboration

Abstract

This invention relates to the field of artificial intelligence technology and discloses a drilling operation assistance and fault analysis method and system based on multi-agent collaboration. By acquiring problem text data, it jointly processes problem semantic vector data and equipment parameter status snapshot data to generate knowledge-enhanced result data. Parallel reasoning is performed using operation assistance agents and fault detection agents. Inconsistent data states are processed for consistency based on a multi-agent collaboration mechanism. Finally, a comprehensive evaluation agent performs fusion reasoning to generate drilling operation assistance and fault handling suggestions. This invention combines a vector knowledge base and a knowledge graph dual-knowledge engine. Through multi-agent collaborative consistency processing, and with division of labor and cooperation under a unified scheduling mechanism, it continuously optimizes decision quality through context sharing and memory feedback loops. This provides on-site operators with full-process, traceable intelligent support, from operation guidance to fault tracing and handling suggestions.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a drilling operation assistance and fault analysis method and system based on multi-agent collaboration. Background Technology

[0002] Oil drilling is a complex, high-risk, and costly engineering activity. Its operational procedures are intricate, its equipment systems are vast, and it involves knowledge from multiple disciplines, including geology, mechanics, electronics, and fluid mechanics. On the drilling site, both experienced engineers and novice operators frequently face two core challenges: first, forgetting or being unsure of the correct next step in a complex sequence of operations; and second, struggling to quickly pinpoint the root cause of equipment malfunctions and take effective corrective measures.

[0003] Traditional solutions rely heavily on paper-based operating procedure manuals, scattered electronic documents, and the experience and memory of operators. When problems occur, personnel need to manually consult massive amounts of unstructured technical data and fault case databases. This process is inefficient, time-consuming, and prone to misjudgment under pressure in emergency situations, which may lead to a significant increase in nonproductive time (NPT) or even safety accidents. With the development of artificial intelligence technology, large language models have demonstrated powerful capabilities in natural language understanding and generation, providing a new technical path for intelligent interaction in drilling equipment systems. However, existing research has attempted to apply large models to industrial question-answering systems or simple fault diagnosis scenarios, but these systems are mostly single-function modules, lacking effective integration of multi-source heterogeneous information (such as real-time sensor data, historical operation logs, knowledge graphs, and user dialogues), and have not built a complete intelligent agent collaboration mechanism. At the same time, considering that equipment parameters are periodically collected through network transmission protocols (e.g., initiating requests and receiving data packets at fixed intervals), when link jitter, short-term packet loss, or delays occur, multiple intelligent agents may perform inference based on equipment parameter status snapshots at different time points / versions in the same round of tasks, leading to potential issues with the real-time performance and accuracy of the inference results, as well as conflicts between operation assistance and fault analysis. Especially in the face of multi-stage task planning, cross-level fault root cause analysis, and dynamic decision support for complex drilling systems, existing technologies still have significant limitations.

[0004] Therefore, there is an urgent need for an intelligent assistance method and system that can deeply integrate drilling expertise, real-time monitoring data, and user interaction history, and achieve autonomous reasoning, dynamic decision-making, and comprehensive evaluation through a multi-agent collaborative architecture. Summary of the Invention

[0005] This invention provides a drilling operation assistance and fault analysis method and system based on multi-agent collaboration, aiming to solve at least one of the above-mentioned technical problems.

[0006] To achieve the above objectives, this invention provides a drilling operation assistance and fault analysis method based on multi-agent cooperation, comprising the following steps:

[0007] The system acquires user-input question text data, analyzes it to obtain question semantic vector data and question intent data, and sends them to the multi-agent scheduling module to generate agent invocation control data.

[0008] The drilling system detection agent calls the multi-agent scheduling module to obtain real-time operating parameter data of the drilling equipment, generate equipment parameter status snapshot data and share it to the multi-agent collaborative space;

[0009] Based on the knowledge enhancement processing flow, the semantic vector data of the problem and the snapshot data of the device parameter status are jointly processed to generate knowledge enhancement result data;

[0010] Based on the knowledge enhancement result data and the device parameter status snapshot data, the operation assistance agent and fault detection agent of the multi-agent scheduling module are invoked to perform parallel reasoning processing to generate operation step planning data and fault analysis intermediate result data.

[0011] The multi-agent coordination and consistency unit, which calls the multi-agent scheduling module, performs coordinated and consistent processing on inconsistent data states based on the multi-agent coordination mechanism.

[0012] The memory management agent of the multi-agent scheduling module is invoked to generate historical memory data;

[0013] The system uses the multi-agent scheduling module to perform fusion reasoning on the operation procedure planning data, fault analysis intermediate result data, historical memory data, and problem intent data to generate drilling operation assistance and fault handling suggestions.

[0014] Optionally, the steps of acquiring user-input question text data, analyzing it to obtain question semantic vector data and question intent data, and sending them to the multi-agent scheduling module to generate agent invocation control data include:

[0015] Receive user-inputted question text data, and perform standardization processing on the question text data to generate standardized question text data;

[0016] Semantic encoding processing is performed on the standardized question text data to generate question semantic vector data; wherein, the semantic encoding processing includes: inputting the standardized question text data into an embedding model for vectorization calculation;

[0017] The standardized question text data is subjected to intent discrimination processing to generate question intent data that characterizes whether the question text data belongs to the operation guidance category or the fault handling category;

[0018] The problem semantic vector data and the problem intent data are encapsulated into scheduling input data and sent to the multi-agent scheduling module to generate agent invocation control data.

[0019] Optionally, the steps of calling the drilling system detection agent of the multi-agent scheduling module to obtain real-time operating parameter data of the drilling equipment and generating equipment parameter status snapshot data to share to the multi-agent collaborative space specifically include:

[0020] The intelligent agent invokes control data to invoke the drilling system detection intelligent agent, and the drilling system detection intelligent agent establishes a data transmission link with the drilling equipment.

[0021] After the data transmission link is established, parameter request command data is sent to the drilling equipment, and raw real-time operating parameter data returned by the drilling equipment is received; wherein, the parameter request command data includes a predefined parameter request code for identifying the equipment parameter reading request;

[0022] The original real-time operating parameter data is snapshotted to generate device parameter status snapshot data; wherein, the snapshotting process includes: noise reduction, outlier removal, unstructured to structured conversion, and parameter field mapping processing;

[0023] The generated device parameter status snapshot data is written into the multi-agent collaborative space to form multi-agent shared status data.

[0024] Optionally, the real-time operating parameter data of the drilling equipment includes any one or more combinations of the real-time operating parameter data of the top drive status, square lever, slips, drill string assembly, blowout preventer, manifold, mud pit, pump, hydraulic tongs, hook, and drill string.

[0025] Optionally, based on the knowledge enhancement processing flow, the step of jointly processing the problem semantic vector data and the device parameter status snapshot data to generate knowledge enhancement result data specifically includes:

[0026] Using the semantic vector data of the problem as the query vector, vector similarity retrieval processing is performed in the drilling vector knowledge base to obtain the set of drilling knowledge text blocks that have the highest similarity to the query vector;

[0027] Text knowledge evidence data is extracted from the drilling knowledge text block set; wherein, the text knowledge evidence data includes: text block identifier data and its content index data;

[0028] The problem text data is subjected to key equipment extraction processing to generate key equipment set data, and the key equipment set data is mapped to the drilling equipment knowledge graph to perform adjacency retrieval processing to obtain associated equipment set data directly connected to the key equipment set data;

[0029] Extract the parameter subset data corresponding to the associated device set data from the device parameter status snapshot data to generate associated device parameter data;

[0030] The text knowledge evidence data, associated device set data, and associated device parameter data are structurally bound together to generate knowledge-enhanced result data.

[0031] Optionally, the operation-assistant agent and fault detection agent of the multi-agent scheduling module are invoked to perform parallel reasoning processing based on the knowledge enhancement result data and the device parameter status snapshot data to generate operation step planning data and fault analysis intermediate result data. Specifically, this includes:

[0032] Based on the intelligent agent calling control data to call the operation auxiliary intelligent agent, the knowledge enhancement result data and the device parameter status snapshot data are used as joint input data to perform operation stage identification and step planning processing to generate operation step planning data;

[0033] Based on the intelligent agent calling control data, the fault detection intelligent agent is called, and the knowledge enhancement result data and the equipment parameter status snapshot data are used as fault reasoning input data to perform fault analysis and root cause reasoning processing to generate intermediate fault analysis result data.

[0034] Knowledge evidence reference data and parameter snapshot reference data are added to the operation step planning data and the fault analysis intermediate result data respectively to form traceable parallel inference output data, and the traceable parallel inference output data is sent to the comprehensive evaluation agent.

[0035] Optionally, the multi-agent collaborative consistency unit performs collaborative consistency processing on inconsistent data states based on a multi-agent collaborative mechanism, including the following execution process:

[0036] Assign status version identifier data to each generated device parameter status snapshot data, and write the status version identifier data and the device parameter status snapshot data together into the multi-agent collaborative space;

[0037] When the operation assistance agent and the fault detection agent generate operation step planning data or fault analysis intermediate result data, they attach the state version identifier data used to their respective outputs and send the attached outputs to the comprehensive evaluation agent.

[0038] The system drives the comprehensive evaluation agent to perform consistency verification processing on the received outputs of multiple agents, and generates conflict detection marker data based on the semantic constraint relationship between the operation step planning data and the intermediate result data of the fault analysis; wherein, the consistency verification processing includes: determining whether there is cross-version inconsistency in the outputs of multiple agents based on the state version identifier data;

[0039] When the conflict detection marker data indicates a conflict, a collaborative arbitration process is triggered; wherein, the collaborative arbitration process includes: sending parameter refresh request data to the drilling system detection agent to obtain new equipment parameter status snapshot data and its status version identifier data, and recalculating or selecting the conflict output based on the updated knowledge enhancement result data;

[0040] The final operational assistance and fault handling suggestions determined after collaborative arbitration are written into the memory management module, and the conflict detection mark data, status version identifier data and arbitration result data are written into the memory tuple set to form traceable data.

[0041] Optionally, the step of invoking the memory management agent of the multi-agent scheduling module to generate historical memory data specifically includes:

[0042] Based on the aforementioned intelligent agent call control data, the memory management intelligent agent is invoked to read historical dialogue data, historical device parameter data, and historical user operation data.

[0043] The historical dialogue data, historical device parameter data, and historical user operation data are uniformly modeled into a set of memory tuples to generate historical memory data.

[0044] Optionally, the comprehensive evaluation agent of the multi-agent scheduling module is invoked to perform fusion reasoning processing on the operation step planning data, fault analysis intermediate result data, historical memory data, and problem intent data to generate drilling operation assistance and fault handling suggestion data steps, specifically including:

[0045] Based on the intelligent agent calling control data, a comprehensive evaluation intelligent agent is called to receive the operation step planning data, the fault analysis intermediate result data, the historical memory data, and the problem intent data, and to form a fusion reasoning input data set;

[0046] A consistency alignment process is performed on the fusion reasoning input data set; wherein, the consistency alignment process includes: semantically aligning the operation step planning data and the fault analysis intermediate result data according to the problem intent data, and performing context pruning on the historical memory data to generate context constraint data;

[0047] Based on the aligned fusion inference input data set, multi-source information comprehensive inference processing is performed to generate structured drilling operation assistance and fault handling suggestion data;

[0048] Output the drilling operation assistance and fault handling suggestion data, and write the drilling operation assistance and fault handling suggestion data and problem text data into the memory tuple set to update the historical memory data.

[0049] Furthermore, to achieve the above objectives, the present invention also provides a drilling operation assistance and fault analysis system based on multi-agent collaboration. The drilling operation assistance and fault analysis system based on multi-agent collaboration includes: a memory, a processor, and a drilling operation assistance and fault analysis program based on multi-agent collaboration stored in the memory and executable on the processor. When the drilling operation assistance and fault analysis program based on multi-agent collaboration is executed by the processor, it implements the steps of the drilling operation assistance and fault analysis method based on multi-agent collaboration as described above.

[0050] The beneficial effects of this invention are as follows: It proposes a drilling operation assistance and fault analysis method and system based on multi-agent collaboration. By acquiring user-input question text data, it jointly processes the question semantic vector data and equipment parameter status snapshot data to generate knowledge-enhanced result data. Parallel reasoning processing is performed using operation assistance agents and fault detection agents to generate operation step planning data and fault analysis intermediate result data. Inconsistent data states are processed for consistency based on a multi-agent collaboration mechanism. Finally, a comprehensive evaluation agent from a multi-agent scheduling module performs fusion reasoning processing to generate drilling operation assistance and fault handling suggestion data. Therefore, this invention combines a vector knowledge base and a knowledge graph dual knowledge engine, leverages multi-agent collaborative consistency processing, and collaborates under a unified scheduling mechanism. Through context sharing and memory feedback loops, it continuously optimizes decision quality, thereby providing on-site operators with full-process, traceable intelligent support from operation guidance to fault tracing and handling suggestions. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating an embodiment of the drilling operation assistance and fault analysis method based on multi-agent collaboration of the present invention.

[0052] Figure 2 This is a schematic diagram illustrating the principle of an embodiment of the drilling operation assistance and fault analysis system based on multi-agent collaboration of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0054] This invention provides a drilling operation assistance and fault analysis method based on multi-agent cooperation, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the drilling operation assistance and fault analysis method based on multi-agent collaboration of the present invention.

[0055] In this embodiment, a drilling operation assistance and fault analysis method based on multi-agent collaboration includes the following steps:

[0056] S1: Obtain the user-input question text data, analyze it to obtain question semantic vector data and question intent data, and send them to the multi-agent scheduling module to generate agent call control data;

[0057] S2: Call the drilling system detection agent of the multi-agent scheduling module to obtain real-time operating parameter data of the drilling equipment, generate equipment parameter status snapshot data and share it to the multi-agent collaborative space;

[0058] S3: Based on the knowledge enhancement processing flow, the semantic vector data of the problem and the snapshot data of the device parameter status are jointly processed to generate knowledge enhancement result data;

[0059] S4: Based on the knowledge enhancement result data and the device parameter status snapshot data, call the operation assistance agent and fault detection agent of the multi-agent scheduling module to perform parallel reasoning processing, and generate operation step planning data and fault analysis intermediate result data;

[0060] S5: The multi-agent coordination and consistency unit calls the multi-agent scheduling module to perform coordinated and consistent processing on inconsistent data states based on the multi-agent coordination mechanism;

[0061] S6: Call the memory management agent of the multi-agent scheduling module to generate historical memory data;

[0062] S7: The system calls upon the comprehensive evaluation agent of the multi-agent scheduling module to perform fusion reasoning processing on the operation step planning data, fault analysis intermediate result data, historical memory data, and problem intent data, and generates drilling operation assistance and fault handling suggestion data.

[0063] It should be noted that traditional solutions rely heavily on paper-based operating procedure manuals, scattered electronic documents, and the experience and memory of operators. When problems occur, personnel need to manually consult massive amounts of unstructured technical data and fault case databases. This process is inefficient, time-consuming, and prone to misjudgment under pressure in emergency situations, which may lead to a significant increase in nonproductive time (NPT) or even safety accidents. With the development of artificial intelligence technology, large language models have demonstrated powerful capabilities in natural language understanding and generation, providing a new technical path for intelligent interaction in drilling equipment systems. However, existing research has attempted to apply large models to industrial question-answering systems or simple fault diagnosis scenarios, but these systems are mostly single-function modules, lacking effective integration of multi-source heterogeneous information (such as real-time sensor data, historical operation logs, knowledge graphs, and user dialogues), and have not built a complete intelligent agent collaboration mechanism. At the same time, considering that equipment parameters are periodically collected through network transmission protocols (e.g., initiating requests and receiving data packets at fixed intervals), when link jitter, short-term packet loss, or delays occur, multiple intelligent agents may perform inference based on equipment parameter status snapshots at different time points / versions in the same round of tasks, leading to potential issues with the real-time performance and accuracy of the inference results, as well as conflicts between operation assistance and fault analysis. Especially in the face of multi-stage task planning, cross-level fault root cause analysis, and dynamic decision support for complex drilling systems, existing technologies still have significant limitations.

[0064] To address the aforementioned issues, this embodiment acquires user-input question text data, jointly processes question semantic vector data and equipment parameter status snapshot data to generate knowledge-enhanced result data. Parallel reasoning is then performed using operation assistance agents and fault detection agents to generate operation step planning data and fault analysis intermediate result data. Inconsistent data states are processed for consistency based on a multi-agent collaborative mechanism. Finally, a comprehensive evaluation agent from a multi-agent scheduling module performs fusion reasoning to generate drilling operation assistance and fault handling suggestions. Thus, this invention combines a vector knowledge base and a knowledge graph dual-knowledge engine, leverages multi-agent collaborative consistency processing, and collaborates under a unified scheduling mechanism. Through context sharing and memory feedback loops, it continuously optimizes decision quality, providing on-site operators with end-to-end, traceable intelligent support from operation guidance to fault tracing and handling suggestions.

[0065] In a preferred embodiment, the steps of acquiring user-input question text data, analyzing it to obtain question semantic vector data and question intent data, and sending them to the multi-agent scheduling module to generate agent invocation control data specifically include:

[0066] S11: Receive question text data input by the user, and perform standardization processing on the question text data to generate standardized question text data;

[0067] S12: Perform semantic encoding processing on the standardized question text data to generate question semantic vector data; wherein, the semantic encoding processing includes: inputting the standardized question text data into an embedding model for vectorization calculation;

[0068] S13: Perform intent discrimination processing on the standardized problem text data to generate problem intent data that characterizes whether the problem text data belongs to the operation guidance category or the fault handling category;

[0069] S14: Encapsulate the problem semantic vector data and the problem intent data into scheduling input data and send them to the multi-agent scheduling module to generate agent call control data.

[0070] In this embodiment, the "question text data" input by the user in natural language first needs to be converted into a structured data representation that can be directly used by the subsequent knowledge enhancement process and the multi-agent collaborative reasoning process. To this end, this embodiment standardizes, semantically encodes, and discriminates the intent of the question text data to form two types of core data outputs: question semantic vector data and question intent data. These are then sent as scheduling input data to the multi-agent scheduling module to generate the agent invocation control data required for subsequent agent collaborative execution.

[0071] In one implementation of the standardization process, the standardization process may include at least one or more of the following data processing actions: performing character normalization on the text to ensure that synonymous full-width / half-width characters and uppercase / lowercase forms are consistent at the data layer; filtering out meaningless redundant whitespace characters, repeated punctuation marks, and control characters to reduce noise interference with semantic representation; and performing unified encoding format processing on the text to ensure the consistency of character encoding in subsequent model inputs.

[0072] For semantic encoding processing, standardized question text data needs to be input into the embedding model for vectorization computation, and the output should be a vector representation corresponding to the semantics of the standardized question text data, serving as the question semantic vector data. It should be noted that the goal of semantic encoding processing is to map natural language text into a high-dimensional vector space, enabling it to have computable similarity measurement capabilities, thereby supporting subsequent similarity retrieval and knowledge recall in the drilling vector knowledge base.

[0073] Furthermore, the embedding model can adopt the same text embedding model as the vector knowledge base construction stage to ensure that the question vector and the knowledge text block vector are in the same vector space, thus enabling direct similarity calculation and matching retrieval. The question semantic vector data generated in this way will be used as the retrieval query vector in the knowledge enhancement processing flow in subsequent steps to recall the set of drilling knowledge text blocks most relevant to the user's question from the drilling vector knowledge base.

[0074] For semantic discrimination processing, the goal is to identify the "task type" of the user's question, at least determining whether the question belongs to: operation guidance (e.g., asking "What's the next step?" or "What should I do at this stage?") or fault handling (e.g., asking "Why is the device malfunctioning?" or "How do I troubleshoot the fault?"). In one implementation, intent discrimination processing can output intent labels using data processing methods such as rule-based feature extraction and classification, or model-based text classification inference. It should be noted that regardless of the implementation method used, the output is represented in the form of structured question intent data. This question intent data at least includes an intent category identifier field, which is used by the subsequent multi-agent scheduling module to select which agents to call and to determine the focus of fusion inference during the comprehensive evaluation phase.

[0075] For the encapsulation scheduling process, the question semantic vector data generated in step S12 and the question intent data generated in step S13 need to be encapsulated into scheduling input data and sent to the multi-agent scheduling module to generate agent invocation control data. In one embodiment, the scheduling input data includes at least: question semantic vector data (used to drive subsequent knowledge enhancement retrieval and related task decomposition); question intent data (used to drive agent selection, execution priority, and collaborative path); and may further carry index fields such as session identifier and timestamp to support association in multi-round interactions.

[0076] After receiving the scheduling input data, the multi-agent scheduling module generates agent invocation control data. This data clarifies the invocation relationships and execution order of each agent in subsequent steps. For example, it might first call the drilling system detection agent to obtain equipment parameter status snapshot data; then, after generating knowledge enhancement result data in the knowledge enhancement process, it might call the operation assistance agent and the fault detection agent in parallel; finally, it might call the comprehensive evaluation agent to perform fusion reasoning and output suggestions. In this way, the output agent invocation control data provides a unified scheduling entry point for the multi-agent collaborative process, offering consistent control criteria and data entry for subsequent steps.

[0077] In a preferred embodiment, the steps of invoking the drilling system detection agent of the multi-agent scheduling module to obtain real-time operating parameter data of the drilling equipment and generating equipment parameter status snapshot data to be shared to the multi-agent collaborative space specifically include:

[0078] S21: Based on the intelligent agent calling control data, the drilling system detection intelligent agent is called, and the drilling system detection intelligent agent establishes a data transmission link with the drilling equipment;

[0079] S22: After the data transmission link is established, send parameter request instruction data to the drilling equipment and receive the raw real-time operating parameter data returned by the drilling equipment; wherein, the parameter request instruction data includes a predefined parameter request code for identifying the equipment parameter reading request;

[0080] S23: Perform snapshot processing on the original real-time operating parameter data to generate device parameter status snapshot data; wherein, the snapshot processing includes: noise reduction, outlier removal, unstructured to structured conversion, and parameter field mapping processing;

[0081] S24: Write the generated device parameter status snapshot data into the multi-agent collaborative space to form multi-agent shared status data.

[0082] In this embodiment, the drilling system detection agent, acting as a client, actively initiates a connection request, establishes a stable data transmission link with the server established by the drilling equipment, and then sends a predefined parameter request code to the server established by the drilling equipment to identify this communication as a device parameter reading request. After receiving the parameter request code, the server established by the drilling equipment verifies the validity of the request and responds immediately, packaging the current device operating status parameters and sending them back to the drilling system detection agent. After receiving the current drilling equipment operating status parameters, the drilling system detection agent performs noise reduction and outlier removal, converts unstructured data into structured data, and converts it into the corresponding Chinese parameter format.

[0083] Specifically, in one particular embodiment of the present invention, the drilling system detects the drilling equipment operating status parameters processed by the intelligent agent as follows:

[0084] "Top Drive": {"Height": 2.500001,"Spindle Speed": 0,"Torque": 0,"Connection Status with Drill String": false,"Connection Status with Internal Blowout Preventer Switch": On},"Drill Pipe": {"Spindle Speed": 0,"Torque": 0,"Connection Status with Drill String": No},"Slips": {"Wellhead Holding Tool Type": "Safety Lock"}, Drill String Assembly: {"Drill String Data": [{"Type": "Drill Collar", "Number of Drill Strings": 3, "Drill String Connected": false}, {"Type": "Drill Bit Adapter", "Number of Drill Strings": 1, "Drill String Connected": false}, {"Type": "Tricone Bit", "Number of Drill Strings": 1, "Drill String Connected": false}], "Plug Status": false, "Back Pressure Valve Top Pass Status": false, "Total Drill String Length": 91.50858, "Drill Bit Position": 88.7198639, "Drill Pipe Height on Drill Platform": 0.850001335, "Drill Pressure": 0.0, "Drill Speed": 0.0}.

[0085] The parameters mentioned above together constitute a complete snapshot of the current drilling equipment status. Other equipment parameters not listed individually are updated synchronously based on actual monitoring results, and their specific values ​​are dynamically determined according to the real-time operating status. They will not be elaborated here.

[0086] In practical applications, the real-time operating parameter data of the drilling equipment includes any one or more combinations of the real-time operating parameter data of the top drive status, square lever, slips, drill string assembly, blowout preventer, manifold, mud pit, pump, hydraulic tongs, hook, and drill string.

[0087] In a preferred embodiment, the step of jointly processing the question semantic vector data and the device parameter status snapshot data based on the knowledge enhancement processing flow to generate knowledge enhancement result data specifically includes:

[0088] S31: Using the semantic vector data of the problem as the query vector, perform vector similarity retrieval processing in the drilling vector knowledge base to obtain the set of drilling knowledge text blocks with the highest similarity to the query vector;

[0089] S32: Extract text knowledge evidence data from the drilling knowledge text block set; wherein, the text knowledge evidence data includes: text block identifier data and its content index data;

[0090] S33: Perform key equipment extraction processing on the question text data to generate key equipment set data, and map the key equipment set data to the drilling equipment knowledge graph to perform adjacency retrieval processing to obtain associated equipment set data directly connected to the key equipment set data;

[0091] S34: Extract the parameter subset data corresponding to the associated device set data from the device parameter status snapshot data to generate associated device parameter data;

[0092] S35: Structure and bind the text knowledge evidence data, associated device set data, and associated device parameter data to generate knowledge enhancement result data.

[0093] In this embodiment, the top K vectors with the highest similarity to the question semantic vector are first retrieved from the drilling vector knowledge base. The vector similarity calculation formula for each drilling knowledge text block is as follows:

[0094]

[0095] in: For query vector, For knowledge vectors, This is the dot product of vectors.

[0096] Therefore, the resulting set of drilling knowledge text blocks will serve as a source of knowledge evidence in subsequent steps, providing referable domain knowledge support for operation planning and fault reasoning, and avoiding insufficient interpretability or unreliable recommendations due to relying solely on real-time parameters.

[0097] Preferably, the specific implementation method for constructing the drilling vector knowledge base is as follows: For the source document... Perform preprocessing (divide the blocks according to...) After separating them in this way, we get ,according to The result is obtained by dividing the block using the delimiter. , Representing a text block, the block segmentation result It is expressed as follows:

[0098]

[0099] Then the block results The drilling vector knowledge base is obtained by converting the data into high-dimensional vectors using the embedding model bge-large-zh-v1.5. , A vector block is represented as follows:

[0100]

[0101] The drilling vector knowledge base has been built by following the steps above.

[0102] Then, textual knowledge evidence data is extracted from the drilling knowledge text block set; wherein the textual knowledge evidence data includes: text block identification data and its content index data. In one embodiment, the text block identification data is used to uniquely identify each recalled knowledge text block (e.g., a combination of text block number, document source identifier, and block sequence number); the content index data is used to indicate the position of the text block in the source document or its index position in the knowledge base (e.g., start and end offsets, paragraph sequence number, or key sentence index within the block), to support accurate referencing and tracing of knowledge fragments in subsequent reasoning processes.

[0103] Therefore, the obtained text knowledge evidence data can transform the recalled text content into computable and traceable structured evidence, enabling subsequent agents to bind corresponding evidence sources when outputting suggestions or conclusions, thereby improving interpretability and consistency.

[0104] Next, key equipment extraction processing is performed on the question text data to generate key equipment set data. This key equipment set data is then mapped to a drilling equipment knowledge graph to perform adjacency retrieval processing, obtaining associated equipment set data directly connected to the key equipment set data. In one embodiment, large model prompt word engineering technology is employed to extract key equipment from unstructured user questions. Accurately acquire key equipment sets assemble key equipment Mapping this onto a drilling equipment knowledge graph yields a set of key equipment. All connected devices , represented as:

[0105]

[0106] in, This means retrieving all adjacent devices directly connected to the target device in the drilling equipment knowledge graph.

[0107] Subsequently, a subset of parameters corresponding to the associated device set data is extracted from the device parameter status snapshot data to generate associated device parameter data. In one implementation, the device parameter status snapshot data contains parameter items for multiple devices or subsystems, each parameter item being associated with a corresponding device identifier or device category field. By using the associated device set data as a filter key, parameter filtering and field aggregation are performed on the device parameter status snapshot data to extract parameter items matching the associated device set, forming the associated device parameter data.

[0108] It is easy to understand that the associated device parameter data can be represented in a structured record format, such as a record set of "device identifier, parameter name, parameter value, and timestamp", so that the subsequent intelligent agent reasoning process can directly use the data to perform data processing such as state comparison, abnormal feature extraction, or rule verification.

[0109] Finally, the textual knowledge evidence data, associated device set data, and associated device parameter data are structurally bound together to generate knowledge-enhanced result data. Thus, the generated knowledge-enhanced result data includes at least a joint data structure of knowledge evidence (text block identifiers and indexes), device associations (key devices / associated device sets), and associated device parameters (parameter subsets). This enables subsequent operation-assisting agents and fault detection agents to reason within the same enhanced context, avoiding inconsistencies in recommendations or incomplete analysis chains caused by the fragmentation of knowledge, topology, and parameters.

[0110] In a preferred embodiment, the steps of invoking the operation-assistant agent and fault detection agent of the multi-agent scheduling module to perform parallel reasoning processing based on the knowledge enhancement result data and the device parameter status snapshot data to generate operation step planning data and fault analysis intermediate result data specifically include:

[0111] S41: Based on the intelligent agent calling control data, the operation-assisted intelligent agent is called, and the knowledge enhancement result data and the device parameter status snapshot data are used as joint input data to perform operation stage identification and step planning processing to generate operation step planning data;

[0112] S42: Based on the intelligent agent calling control data, the fault detection intelligent agent is called, and the knowledge enhancement result data and the equipment parameter status snapshot data are used as fault reasoning input data. Fault analysis and root cause reasoning processing are performed to generate intermediate fault analysis result data.

[0113] S43: Add knowledge evidence reference identification data and parameter snapshot reference identification data to the operation step planning data and the fault analysis intermediate result data respectively to form traceable parallel reasoning output data, and send the traceable parallel reasoning output data to the comprehensive evaluation agent.

[0114] In this embodiment, the operation-assisted intelligent agent is invoked based on the intelligent agent's call control data. The knowledge enhancement result data and the device parameter status snapshot data are used as joint input data to perform operation stage identification and step planning processing to generate operation step planning data.

[0115] Specifically, the operation-assisted intelligent agent combines the knowledge-enhanced result data with drilling knowledge text blocks in the equipment parameter status snapshot data. and the parameter status of drilling equipment Extract the data and then plan the user's next steps. Operation prompts The results are as follows:

[0116]

[0117] in, Prompt words indicating how to guide the operation of the intelligent agent in planning user questions. For large models.

[0118] For example, in a specific instance, where for:

[0119] You are an operator's assistant for a drilling equipment system, specifically tasked with helping users complete drilling operations. Your job is to accurately answer users' operational questions based on the information they provide. Users will describe the current operational steps or status; you need to retrieve relevant information from drilling knowledge and generate instructions for the next step.

[0120] 1. Understand the user's current step: Users will clearly state which stage of the operation they are currently in.

[0121] 2. Retrieve relevant text passages: Following the steps provided by the user, locate the corresponding parts within the provided text paragraphs.

[0122] 3. Generate next step guidance: Based on the retrieved information, guide users on what to do next in a clear and concise manner, which may include specific operation steps, precautions, etc.

[0123] Given a user question and a text, please analyze which operation process the user is in according to the above guidelines, and suggest the user's next steps.

[0124] Output format requirements:

[0125] 1. Operation Phase: Indicates the operation phase that the user should be in.

[0126] 2. Operation steps: Give the next operation instructions.

[0127] 3. Precautions: List the precautions or key points for the next step.

[0128] For example, the operator-assisted intelligent agent responds to the following user questions Operation prompts:

[0129] I have just finished shutting down the pump. What should I do next?

[0130] Operation prompts:

[0131] Operational phase: Adjusting the performance of the well control mud.

[0132] Operating steps: Enter the “Mud Circulation System” interface, click “Suction Tank” or “Standby Tank”, and adjust the mud density to the well control mud density specified on the well control operation order; then press the “Cumulative Pump Flush Reset” button on the throttling control box to reset the cumulative pump flush count to zero.

[0133] Precautions:

[0134] 1. Ensure the mud density is accurately adjusted to the value required by the work order; there must be no errors.

[0135] 2. Before adjusting the mud properties, confirm that the pump has been completely stopped and the system is in a stable state;

[0136] 3. The “cumulative pump flush reset” operation must be completed before restarting the pump to ensure accurate pump flush count during well control.

[0137] In this embodiment, the fault detection agent is invoked based on the intelligent agent's call control data. The knowledge enhancement result data and the device parameter status snapshot data are used as fault reasoning input data to perform fault analysis and root cause reasoning processing to generate intermediate fault analysis result data.

[0138] Specifically, the fault detection agent will compare the knowledge-enhanced result data with all devices in the device parameter status snapshot data. Specific parameter values ​​are extracted and combined with information related to the current problem transmitted by the operation-assisting intelligent agent. A drilling knowledge text block Generate fault analysis results Fault analysis results It is expressed as follows:

[0139]

[0140] in, This indicates that the fault analysis results have been generated.

[0141] For example, in a specific instance, where for:

[0142] You are a fault detection agent. Please complete your three tasks in the following order:

[0143] 1. Identify the devices mentioned in the user's query. The following is a list of devices:

[0144] Equipment list: Top drive; Square drill pipe; Slips; Drill string assembly; Blowout preventer; Manifold; Mud pit; Pump; Well condition data; Hydraulic tongs; Hook; Rotary table.

[0145] 2. Based on the device identified by the query, invoke the tool. It completes the retrieval of the knowledge graph and retrieves all related devices.

[0146] 3. Based on drilling knowledge, the status of all relevant equipment, and user issues, generate a logically clear inspection plan.

[0147] Output format requirements:

[0148] Inspection plan: Generate a logically clear list of inspection steps, each of which should include instructions and expected results.

[0149] For example, all devices currently associated with [] (devices identified in the question query) are [] (results returned after retrieving the knowledge graph).

[0150] Inspection plan:

[0151] 1. [Step 1]:

[0152] - Operation: [Operation Instructions];

[0153] - Expected results: [Expected results];

[0154] 2. [Step 2]:

[0155] - Operation: [Operation Instructions];

[0156] - Expected results: [Expected results];

[0157] For example, the fault detection agent addresses the following user questions Perform fault analysis:

[0158] Why can't I turn off the pump now?

[0159] Fault Analysis:

[0160] Currently, all equipment associated with the [pump] is the [top drive, screw drill pipe, riser manifold];

[0161] Inspection plan:

[0162] 1. Check that the top drive and angular drill pipe are connected;

[0163] Operation: Check that the top drive and screed are connected;

[0164] Correct status: Top drive and screw are connected;

[0165] 2. Check if the riser manifold is unobstructed;

[0166] Procedure: Check if the riser manifold is unobstructed;

[0167] Correct status: Riser manifold is open;

[0168] 3. Check the status of the pump's safety valve;

[0169] Operation: Check the status of the pump's safety valve;

[0170] Correct status: The pump's safety valve is activated.

[0171] In a preferred embodiment, the multi-agent collaborative consistency unit performs collaborative consistency processing on inconsistent data states based on a multi-agent collaborative mechanism, comprising the following execution process:

[0172] S51: Assign status version identifier data to each generated device parameter status snapshot data, and write the status version identifier data and the device parameter status snapshot data together into the multi-agent collaborative space;

[0173] S52: When the operation assistance agent and the fault detection agent generate operation step planning data or fault analysis intermediate result data, they attach the state version identifier data used to their respective outputs and send the attached outputs to the comprehensive evaluation agent.

[0174] S53: Drive the comprehensive evaluation agent to perform consistency verification processing on the received outputs of multiple agents, and generate conflict detection marker data based on the semantic constraint relationship between the operation step planning data and the intermediate result data of the fault analysis; wherein, the consistency verification processing includes: judging whether there is cross-version inconsistency in the outputs of multiple agents based on the state version identifier data.

[0175] S54: When the conflict detection marker data indicates a conflict, a collaborative arbitration process is triggered; wherein, the collaborative arbitration process includes: sending parameter refresh request data to the drilling system detection agent to obtain new equipment parameter status snapshot data and its status version identifier data, and recalculating or selecting the conflict output based on the updated knowledge enhancement result data;

[0176] S55: Write the final operation assistance and fault handling suggestions data determined after collaborative arbitration into the memory management module, and write the conflict detection mark data, status version identifier data and arbitration result data into the memory tuple set to form traceable data.

[0177] In this embodiment, to address the problem of multiple agents reasoning based on inconsistent device states due to issues such as periodic acquisition, jitter, short-term delays, or packet loss in the real-time parameter link at the drilling site, a multi-agent collaborative consistency unit is established. This unit is used to version-identify and verify the consistency of the device parameter states upon which the output is based during parallel reasoning by multiple agents. When cross-version reasoning or semantic conflict is detected, collaborative arbitration is triggered, and the suggestion is recalculated or selected, thereby outputting a single, consistent, and executable final suggestion, while simultaneously forming a traceable collaborative evidence chain. Specifically, the multi-agent collaborative consistency unit includes the following execution process:

[0178] First, assign a status version identifier to each generated device parameter status snapshot data, and write the status version identifier data and the device parameter status snapshot data together into the multi-agent collaborative space.

[0179] In one implementation, a status version identifier is assigned to each snapshot for subsequent verification of the inference basis. The status version identifier can be generated using at least one of the following methods: generating a version number based on the snapshot generation timestamp; generating an incrementing version number based on a collection period counter; calculating a hash value based on the snapshot content and using it as a version identifier; or combining the above methods to form a composite version identifier.

[0180] After writing the status version identifier data and device parameter status snapshot data together into the multi-agent collaborative space, any agent can synchronously read its corresponding version identifier when reading the snapshot data in the collaborative space, thus providing basic data for subsequent consistency verification.

[0181] Then, when the operation assistance agent and the fault detection agent generate operation step planning data or fault analysis intermediate result data, they attach the state version identification data used to their respective outputs and send the attached outputs to the comprehensive evaluation agent.

[0182] In one implementation, the operation-assisting agent and the fault detection agent read device parameter status snapshot data from the multi-agent collaborative space at the start of inference, and simultaneously read the corresponding status version identifier data; when the inference is completed and the output is generated, the output data and the version identifier data are bound and encapsulated to form agent output data with version information.

[0183] For example, operational procedure planning data includes the state version identifier data used for its inference; fault analysis intermediate result data includes the state version identifier data used for its inference. In this way, the comprehensive evaluation agent receives not only the inference conclusion itself, but also the state version information on which the conclusion is based. This allows for the determination of whether the outputs of multiple agents are based on the same device state, avoiding situations where each agent's conclusion is correct but based on different snapshots, leading to overall conflicts.

[0184] Next, the comprehensive evaluation agent is driven to perform consistency verification processing on the received outputs of multiple agents, and generate conflict detection label data based on the semantic constraint relationship between the operation step planning data and the intermediate result data of the fault analysis; wherein, the consistency verification processing includes determining whether there is cross-version inconsistency in the outputs of multiple agents based on the state version identifier data.

[0185] In one implementation, the consistency verification process includes at least the following:

[0186] Version consistency judgment: Compare the status version identifier data attached to the operation step planning data with the status version identifier data attached to the fault analysis intermediate result data; if the two are different, it is determined that there is a cross-version reasoning inconsistency, and a version inconsistency mark is generated.

[0187] Semantic consistency judgment: Under the condition that the versions are consistent or the version differences are acceptable, conflict detection is performed on the operation step planning data and the intermediate result data of fault analysis according to the preset semantic constraint relationship.

[0188] Semantic constraints can be used to characterize the constraint rules that cannot be met simultaneously between suggestions. For example, if the intermediate result data of the fault analysis indicates that the system needs to be shut down / investigated / prohibited from continuing operation, and the operation plan data still outputs the instruction to continue to execute the critical operation, then a conflict exists. If the fault analysis conclusion requires that a certain device be inspected and isolated, and the operation plan requires that the device be immediately activated to perform a high-risk action, then a conflict exists.

[0189] Therefore, after the intelligent agents complete the above verification and constraint detection, they generate conflict detection label data to indicate whether the current outputs of the multiple intelligent agents can be directly merged or need to enter the collaborative arbitration process.

[0190] Subsequently, when the conflict detection marker data indicates a conflict, a collaborative arbitration process is triggered; the collaborative arbitration process includes: sending parameter refresh request data to the drilling system detection agent to obtain new equipment parameter status snapshot data and its status version identifier data, and recalculating or selecting the conflict output based on the updated knowledge enhancement result data.

[0191] In one implementation, when the comprehensive evaluation agent detects inconsistencies or semantic conflicts across versions, a collaborative arbitration process is triggered to avoid erroneous operations caused by directly outputting conflicting suggestions. This process includes at least the following data processing actions:

[0192] Send parameter refresh request data: The integrated evaluation agent or the unification unit initiates a refresh request to the drilling system detection agent, requesting to reacquire the latest equipment parameter status snapshot data;

[0193] Obtain new snapshots and new versions: The detection agent returns new device parameter status snapshot data, and the consistency unit assigns new status version identifier data to it;

[0194] Update knowledge enhancement result data: Using the new device parameter status snapshot data as input, re-execute or incrementally update the knowledge enhancement process to obtain the updated knowledge enhancement result data;

[0195] It should be noted that the following rules apply when recalculating or selecting conflicting outputs:

[0196] Recalculation method: Drive the relevant intelligent agent to regenerate operation step planning data and / or fault analysis intermediate result data based on the new knowledge enhancement result data; Selection method: When a complete recalculation is not required, select the existing output based on the preset arbitration rules, such as giving priority to the data output based on the latest version identifier, or giving priority to the disposal suggestion output with lower risk and stricter constraints.

[0197] Through the aforementioned collaborative arbitration process, multi-agent reasoning can be unified back to the same latest state, or conflicts can be resolved through rules, thereby ensuring that the final recommendations are consistent and executable.

[0198] Finally, the final operational assistance and fault handling suggestions determined after collaborative arbitration are written into the memory management module, and the conflict detection mark data, status version identifier data and arbitration result data are written into the memory tuple set to form traceable data.

[0199] In one implementation, the memory management module maintains a set of memory tuples to store information such as historical dialogues, device parameter states, and historical operations. After the final suggestion data is written back, it can be used as contextual constraints for the next round of interaction or subsequent reasoning, forming a closed-loop learning and continuous optimization. Simultaneously, writing conflict detection marker data, state version identifier data, and arbitration result data into the memory tuple set enables the traceability and recording of the following information:

[0200] Did this recommendation go through conflict and arbitration?

[0201] The state version that each agent outputs when a conflict occurs;

[0202] Ultimately, the arbitration strategy of recalculation or selection was adopted, and the result was determined.

[0203] Thus, the traceable data chain formed not only facilitates on-site review and auditing, but also provides data support for subsequent optimization of multi-agent scheduling strategies, improvement of semantic constraint rules and arbitration rules, thereby further enhancing the robustness and reliability of the system under complex working conditions.

[0204] In a preferred embodiment, the step of invoking the memory management agent of the multi-agent scheduling module to generate historical memory data specifically includes:

[0205] S61: Based on the intelligent agent's call control data, call the memory management intelligent agent to read historical dialogue data, historical device parameter data, and historical user operation data;

[0206] S62: The historical dialogue data, historical device parameter data and historical user operation data are uniformly modeled into a set of memory tuples to generate historical memory data.

[0207] In this embodiment, the memory management agent is used to manage user dialogues, device parameter states, and user historical operations, that is, at any given time... The memory management agent spontaneously acquires three types of heterogeneous information: user dialogue, device parameter status, and user historical operations. Specifically, user dialogue: ,in For users at any time Input, For system response; device parameter status: ,in Indicates the first Each device parameter at time Specific values; User history operations: ,in For the operation type (e.g., "pump start"), This is the operation timestamp.

[0208] Meanwhile, the memory management agent unifies the three types of heterogeneous information—user dialogue, device parameter status, and user historical operations—into memory tuples. memory tuple It is expressed as follows:

[0209] .

[0210] In a preferred embodiment, the comprehensive evaluation agent of the multi-agent scheduling module is invoked to perform fusion reasoning processing on the operation step planning data, fault analysis intermediate result data, historical memory data, and problem intent data to generate drilling operation assistance and fault handling suggestion data steps, specifically including:

[0211] S71: Based on the intelligent agent calling control data, the comprehensive evaluation intelligent agent is called to receive the operation step planning data, the fault analysis intermediate result data, the historical memory data and the problem intent data, and form a fusion reasoning input data set;

[0212] S72: Perform consistency alignment processing on the fusion reasoning input data set; wherein, the consistency alignment processing includes: performing semantic constraint alignment on the operation step planning data and the fault analysis intermediate result data according to the problem intent data, and performing context pruning on the historical memory data to generate context constraint data;

[0213] S73: Perform multi-source information integrated reasoning processing based on the aligned fusion inference input data set to generate structured drilling operation assistance and fault handling suggestion data;

[0214] S74: Output the drilling operation assistance and fault handling suggestion data, and write the drilling operation assistance and fault handling suggestion data and problem text data into the memory tuple set to update the historical memory data.

[0215] In this embodiment, the comprehensive evaluation agent comprises two unit modules: a memory unit and a decision suggestion generation unit. The memory unit enables bidirectional interaction with the memory management agent, reads historical memories to support the current comprehensive evaluation, and writes new dialogue content as memory tuples into the system. The decision suggestion generation unit is responsible for processing multi-source input data (current user question). Memory tuple Operation prompts Fault analysis results This integrates various methods and utilizes a built-in thought process to generate precise operational guidance or troubleshooting suggestions. Operating instructions or troubleshooting suggestions It is expressed as follows:

[0216]

[0217] in, This indicates the generation of operation instructions or troubleshooting suggestions.

[0218] For example, in a specific instance, where for:

[0219] You are a comprehensive evaluation agent in a professional oil drilling operation assistance and fault diagnosis system. Your task is to perform comprehensive reasoning based on multi-source information to generate accurate, safe, and actionable operation guidance or fault handling suggestions.

[0220] Please strictly follow the Chain-of-Thought (COT) steps below for step-by-step reasoning, without skipping any steps. The final output must be a clear and structured suggestion.

[0221] [Steps of Reasoning in the Mind Chain]:

[0222] Step 1: Understand the core needs of users:

[0223] - Analyze the keywords and intent of user questions (such as "How to deal with abnormal pump pressure?", "What should I do next?").

[0224] - Determine whether the issue pertains to normal operating procedure guidance or handling of anomalies / faults.

[0225] - Output: Clearly identify the problem type and the device / system of concern.

[0226] Step 2: Review the historical context and memories:

[0227] - Check past conversations to see if there have been similar questions or if any actions have been taken.

[0228] - Check the user's recent operation history to determine if the current state was caused by a previous operation.

[0229] - Identify potential risk points by combining the trend of equipment parameter changes (if any).

[0230] - Output: Contextual conclusions and potential risk warnings.

[0231] Step 3: Integrate operational suggestions and troubleshooting:

[0232] - Compare the suggestions of the operation assistance agent with the analysis results of the fault detection agent:

[0233] Are they consistent? Are there any conflicts?

[0234] - If a fault signal is present, are the original operating recommendations still applicable?

[0235] - In case of conflict, the fault analysis shall take precedence and the operation path shall be adjusted accordingly.

[0236] - Output: Preliminary decision direction after fusion.

[0237] Step 4: Generate final recommendations

[0238] - Based on the above reasoning, generate one or more specific and actionable suggestions.

[0239] - The suggested format is as follows:

[0240] [Suggestion Type] {{Operation Guide | Troubleshooting}};

[0241] [Parameters involved] {{e.g.: mud pit, top drive, blowout preventer, etc.}};

[0242] [Suggested Content]:

[0243] ;

[0244] ;

[0245] ...

[0246] [Risk Warning] {{Please specify any potential risks}};

[0247] Output Requirements:

[0248] - Use Chinese;

[0249] - Vague words such as "maybe" or "probably" must not be used;

[0250] - All suggestions must be derived from the input information;

[0251] - It is forbidden to fabricate knowledge or parameters that are not provided.

[0252] Therefore, this invention combines a vector knowledge base and a knowledge graph dual knowledge engine, and with the help of multi-agent collaborative consistency processing, it divides tasks and collaborates under a unified scheduling mechanism. Through context sharing and memory feedback closed loop, it continuously optimizes decision quality, thereby providing on-site operators with full-process, traceable intelligent support from operation guidance to fault tracing and handling suggestions.

[0253] Furthermore, this invention also proposes a drilling operation assistance and fault analysis system based on multi-agent collaboration. The drilling operation assistance and fault analysis device based on multi-agent collaboration includes: a memory, a processor, and a drilling operation assistance and fault analysis program based on multi-agent collaboration stored in the memory and executable on the processor. When the drilling operation assistance and fault analysis program based on multi-agent collaboration is executed by the processor, it implements the steps of the drilling operation assistance and fault analysis method based on multi-agent collaboration as described above.

[0254] like Figure 2 As shown, the present invention proposes a drilling operation assistance and fault analysis system based on multi-agent collaboration. Through the interaction between drilling equipment and drilling system detection agents, memory management agents, operation assistance agents, fault detection agents, and comprehensive evaluation agents, it provides on-site operators with full-process, traceable intelligent support from operation guidance to fault tracing and handling suggestions.

[0255] As is easily understood, the operation assistance agent, fault detection agent, and comprehensive evaluation agent are all functional agents obtained by fine-tuning the corresponding base model with corpus text from relevant domains and scenarios. It should be noted that the core of this invention lies in the collaborative operation control of drilling operation assistance and fault analysis based on multi-agent collaboration, and this invention does not limit the fine-tuning method of the base model.

[0256] The specific implementation of the drilling operation assistance and fault analysis device based on multi-agent collaboration in this application is basically the same as the embodiments of the drilling operation assistance and fault analysis method based on multi-agent collaboration described above, and will not be repeated here.

[0257] It is understood that in the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Nth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0258] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0259] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0260] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0261] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A drilling operation assistance and fault analysis method based on multi-agent collaboration, characterized in that, Includes the following steps: The system acquires user-input question text data, analyzes it to obtain question semantic vector data and question intent data, and sends them to the multi-agent scheduling module to generate agent invocation control data. The drilling system detection agent calls the multi-agent scheduling module to obtain real-time operating parameter data of the drilling equipment, generate equipment parameter status snapshot data and share it to the multi-agent collaborative space; Based on the knowledge enhancement processing flow, the semantic vector data of the problem and the snapshot data of the device parameter status are jointly processed to generate knowledge enhancement result data; specifically including: Using the semantic vector data of the problem as the query vector, vector similarity retrieval processing is performed in the drilling vector knowledge base to obtain the set of drilling knowledge text blocks that have the highest similarity to the query vector; Text knowledge evidence data is extracted from the drilling knowledge text block set; wherein, the text knowledge evidence data includes: text block identifier data and its content index data; The problem text data is subjected to key equipment extraction processing to generate key equipment set data, and the key equipment set data is mapped to the drilling equipment knowledge graph to perform adjacency retrieval processing to obtain associated equipment set data directly connected to the key equipment set data; Extract the parameter subset data corresponding to the associated device set data from the device parameter status snapshot data to generate associated device parameter data; The text knowledge evidence data, associated device set data, and associated device parameter data are structurally bound together to generate knowledge-enhanced result data; Based on the knowledge enhancement result data and the device parameter status snapshot data, the operation assistance agent and fault detection agent of the multi-agent scheduling module are invoked to perform parallel reasoning processing to generate operation step planning data and fault analysis intermediate result data. The multi-agent coordination and consistency unit, which invokes the multi-agent scheduling module, performs coordinated consistency processing on inconsistent data states based on the multi-agent coordination mechanism; specifically, it includes: Assign status version identifier data to each generated device parameter status snapshot data, and write the status version identifier data and the device parameter status snapshot data together into the multi-agent collaborative space; When the operation assistance agent and the fault detection agent generate operation step planning data or fault analysis intermediate result data, they attach the state version identifier data used to their respective outputs and send the attached outputs to the comprehensive evaluation agent. The system drives the comprehensive evaluation agent to perform consistency verification processing on the received outputs of multiple agents, and generates conflict detection marker data based on the semantic constraint relationship between the operation step planning data and the intermediate result data of the fault analysis; wherein, the consistency verification processing includes: determining whether there is cross-version inconsistency in the outputs of multiple agents based on the state version identifier data; When the conflict detection marker data indicates a conflict, a collaborative arbitration process is triggered; wherein, the collaborative arbitration process includes: sending parameter refresh request data to the drilling system detection agent to obtain new equipment parameter status snapshot data and its status version identifier data, and recalculating or selecting the conflict output based on the updated knowledge enhancement result data; The final operational assistance and fault handling suggestions determined after collaborative arbitration are written into the memory management module, and the conflict detection mark data, status version identifier data and arbitration result data are written into the memory tuple set to form traceable data. The memory management agent of the multi-agent scheduling module is invoked to generate historical memory data; The system uses the multi-agent scheduling module to perform fusion reasoning on the operation procedure planning data, fault analysis intermediate result data, historical memory data, and problem intent data to generate drilling operation assistance and fault handling suggestions.

2. The drilling operation assistance and fault analysis method based on multi-agent collaboration as described in claim 1, characterized in that, The steps involved in acquiring user-input question text data, analyzing it to obtain question semantic vector data and question intent data, and sending these to the multi-agent scheduling module to generate agent invocation control data include: Receive user-inputted question text data, and perform standardization processing on the question text data to generate standardized question text data; Semantic encoding processing is performed on the standardized question text data to generate question semantic vector data; wherein, the semantic encoding processing includes: inputting the standardized question text data into an embedding model for vectorization calculation; The standardized question text data is subjected to intent discrimination processing to generate question intent data that characterizes whether the question text data belongs to the operation guidance category or the fault handling category; The problem semantic vector data and the problem intent data are encapsulated into scheduling input data and sent to the multi-agent scheduling module to generate agent invocation control data.

3. The drilling operation assistance and fault analysis method based on multi-agent collaboration as described in claim 1, characterized in that, The steps of calling the drilling system detection agent of the multi-agent scheduling module to obtain real-time operating parameter data of the drilling equipment, and generating equipment parameter status snapshot data to share to the multi-agent collaborative space specifically include: The intelligent agent invokes control data to invoke the drilling system detection intelligent agent, and the drilling system detection intelligent agent establishes a data transmission link with the drilling equipment. After the data transmission link is established, parameter request command data is sent to the drilling equipment, and raw real-time operating parameter data returned by the drilling equipment is received; wherein, the parameter request command data includes a predefined parameter request code for identifying the equipment parameter reading request; The original real-time operating parameter data is snapshotted to generate device parameter status snapshot data; wherein, the snapshotting process includes: noise reduction, outlier removal, unstructured to structured conversion, and parameter field mapping processing; The generated device parameter status snapshot data is written into the multi-agent collaborative space to form multi-agent shared status data.

4. The drilling operation assistance and fault analysis method based on multi-agent collaboration as described in claim 3, characterized in that, The real-time operating parameter data of the drilling equipment includes any one or more combinations of the following: top drive status, square lever, slips, drill string assembly, blowout preventer, manifold, mud pit, pump, hydraulic tongs, hook, and drill string real-time operating parameter data.

5. The drilling operation assistance and fault analysis method based on multi-agent collaboration as described in claim 1, characterized in that, The operation-assistant agent and fault detection agent of the multi-agent scheduling module are invoked to perform parallel reasoning processing based on the knowledge enhancement result data and the equipment parameter status snapshot data to generate operation step planning data and fault analysis intermediate result data. Specifically, this includes: Based on the intelligent agent calling control data to call the operation auxiliary intelligent agent, the knowledge enhancement result data and the device parameter status snapshot data are used as joint input data to perform operation stage identification and step planning processing to generate operation step planning data; Based on the intelligent agent calling control data, the fault detection intelligent agent is called, and the knowledge enhancement result data and the equipment parameter status snapshot data are used as fault reasoning input data to perform fault analysis and root cause reasoning processing to generate intermediate fault analysis result data. Knowledge evidence reference data and parameter snapshot reference data are added to the operation step planning data and the fault analysis intermediate result data respectively to form traceable parallel inference output data, and the traceable parallel inference output data is sent to the comprehensive evaluation agent.

6. The drilling operation assistance and fault analysis method based on multi-agent collaboration as described in claim 1, characterized in that, The steps involved in calling the memory management agent of the multi-agent scheduling module to generate historical memory data include: Based on the aforementioned intelligent agent call control data, the memory management intelligent agent is invoked to read historical dialogue data, historical device parameter data, and historical user operation data. The historical dialogue data, historical device parameter data, and historical user operation data are uniformly modeled into a set of memory tuples to generate historical memory data.

7. The drilling operation assistance and fault analysis method based on multi-agent cooperation as described in claim 1, characterized in that, The comprehensive evaluation agent, invoked by the multi-agent scheduling module, performs fusion reasoning processing on the operation step planning data, fault analysis intermediate result data, historical memory data, and problem intent data to generate drilling operation assistance and fault handling suggestion data steps, specifically including: Based on the intelligent agent calling control data, a comprehensive evaluation intelligent agent is called to receive the operation step planning data, the fault analysis intermediate result data, the historical memory data, and the problem intent data, and to form a fusion reasoning input data set; A consistency alignment process is performed on the fusion reasoning input data set; wherein, the consistency alignment process includes: semantically aligning the operation step planning data and the fault analysis intermediate result data according to the problem intent data, and performing context pruning on the historical memory data to generate context constraint data; Based on the aligned fusion inference input data set, multi-source information comprehensive inference processing is performed to generate structured drilling operation assistance and fault handling suggestion data; Output the drilling operation assistance and fault handling suggestion data, and write the drilling operation assistance and fault handling suggestion data and problem text data into the memory tuple set to update the historical memory data.

8. A drilling operation assistance and fault analysis system based on multi-agent collaboration, characterized in that, The drilling operation assistance and fault analysis system based on multi-agent collaboration includes: a memory, a processor, and a drilling operation assistance and fault analysis program based on multi-agent collaboration stored in the memory and executable on the processor. When the drilling operation assistance and fault analysis program based on multi-agent collaboration is executed by the processor, it implements the steps of the drilling operation assistance and fault analysis method based on multi-agent collaboration as described in any one of claims 1 to 7.

Images