Multi-modal knowledge graph construction method oriented to oil field equipment supervision and manufacture

Abstract

The invention discloses a multi-modal knowledge graph construction method for oilfield equipment supervision and manufacture, and the method comprises the steps: constructing a domain knowledge graph covering multi-dimensional entities such as stratum parameters, equipment components, fault modes and the like, and integrating the real-time data of the Internet of Things, equipment maintenance records, geological reports and the like; cross-database semantic mapping is realized by adopting a neural network, and data traceability credibility is guaranteed in combination with a block chain technology; real-time acquisition of operation data and generation of a local map are realized by utilizing edge nodes, global knowledge fusion and deep learning model training are completed by a cloud end, inference rules are dynamically optimized through digital simulation, an equipment-fault causal chain model is constructed to realize accurate fault positioning, and an energy efficiency optimization scheme is generated by integrating carbon footprint calculation; and on the basis of an automatic knowledge node expansion mechanism of transfer learning, VR three-dimensional visual display of the equipment disassembly process and risk early warning is supported, and the industrial pain points of difficulty in multi-source data integration, intelligent decision lag and the like are solved.

Description

Technical Field

[0001] This invention relates to the field of special equipment manufacturing supervision technology in the petroleum industry, and in particular to a method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision. Background Technology

[0002] In the process of supervising the manufacturing of oilfield equipment, there are problems such as difficulty in integrating multi-source data, insufficient correlation between data, lack of real-time monitoring and intelligent decision-making, and poor system scalability. Existing technologies struggle to efficiently process various types of data, including IoT data, equipment maintenance records, and geological parameters, leading to inaccurate fault location, low problem-solving efficiency, and a lack of data support for maintenance decisions. This makes it difficult to adapt to the new data acquisition needs brought about by the technological iteration of oilfield equipment. Therefore, a method capable of integrating multimodal data and enabling intelligent reasoning is needed. Summary of the Invention

[0003] The main objective of this invention is to provide a multimodal knowledge graph construction method for oilfield equipment supervision, in order to solve the technical problems of existing technologies such as inaccurate fault location, low problem handling efficiency, lack of data support for maintenance decisions, and difficulty in adapting to the new data collection requirements brought about by the technological iteration of oilfield equipment.

[0004] To achieve the above objectives, this invention provides a method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision, comprising the following steps: S10 is an edge computing node that collects real-time operational data of oilfield equipment during the manufacturing process and classifies it into structured data, semi-structured data and unstructured data. S20: Preprocess structured data, semi-structured data, and unstructured data separately to obtain preprocessed data that is uniformly converted into RDF triple format; S30, based on preprocessed data, performs semantic alignment and knowledge association processing of cross-regional oilfield data through entity similarity calculation and graph embedding relationship mapping. The processed data is stored in a graph database to obtain an initial domain knowledge graph. S40 generates a lightweight knowledge subgraph by receiving real-time device operation data collected from edge nodes in the cloud, combines digital twin simulation to dynamically optimize reasoning rules, and uses graph neural networks to obtain an updated domain knowledge graph. S50 integrates structured, semi-structured, and unstructured data into multimodal data to obtain multimodal feature data; S60, based on an updated domain knowledge graph and combined with multimodal feature data, classifies and summarizes the problem data that frequently occurs in the oilfield equipment during the manufacturing process, and constructs a causal reasoning model based on the cause of the problem to locate the cause of the failure and obtain the cause data of the failure, and generates energy-saving solutions by combining the carbon footprint model. S70 connects the cloud with the oilfield production management system (EMS) via an open API, and distributes the updated domain knowledge graph and causal reasoning model to edge nodes, forming a closed-loop update mechanism. S80, based on the updated domain knowledge graph, uses transfer learning technology to automate the annotation and relationship completion of knowledge nodes in the newly connected oilfield production management system (EMS). S90, based on the updated domain knowledge graph and the problem data and fault cause data in step S60, integrate AR / VR terminals to realize three-dimensional visual knowledge navigation.

[0005] Optionally, step S20 includes the following steps: S210 standardizes structured data through SQL parsing and time series alignment, and uses a linear interpolation algorithm to fill in missing values; S220: Semi-structured data uses regular expressions to extract key fields and then parses them into a uniform format using XPath. S230: Unstructured data is used for text entity recognition using the BERT-CRF model, image features are extracted using ResNet-50, and speech data is converted into text using ASR technology. S240 uses Kalman filtering to eliminate noise interference in the processed structured, semi-structured, and unstructured data, and then improves the accuracy of feature extraction through image enhancement and stop word removal techniques. Finally, all data is converted into preprocessed data in RDF triple format.

[0006] Optionally, step S30 includes the following steps: S310, mark RDF triples in preprocessed data that have numerical, semantic or logical conflicts for the same entity attribute from different data sources, and denote them as contradictory data; S320 defines oilfield entity types and industry relationship models based on preprocessed data; S330, uses edit distance and word vector similarity to align entity names; S340 uses a graph embedding algorithm to map relationships with different representations to a unified dimension, thus eliminating relationship ambiguity. S350: For contradictory data, a voting mechanism or traceable blockchain records are used to determine the optimal solution; S360 stores the processed data in the graph database Neo4j to obtain the initial domain knowledge graph.

[0007] Optionally, step S330 calculates entity similarity based on the following formula:

[0008] In the formula: , These represent the names of two entities to be compared, originating from different regions or data sources. For harmonic weighting coefficients; entity name The corresponding word vector representation; entity name The corresponding word vector representation.

[0009] Optionally, in step S340, the relationships expressed differently are mapped to a unified dimension based on the following formula:

[0010] In the formula: The loss function; For boundary hyperparameters; For positive sampling triples; For negative sampling triples.

[0011] Optionally, step S50 includes the following steps: S510, extract time-series features from structured data that reflect changes in the state of oilfield equipment; S520 uses a graph structure to parse semi-structured data and uses a graph attention network to extract semantic association features of nodes and relationships. The S530 uses a combined text and image encoder to achieve cross-database semantic mapping between equipment parameters and fault cases for unstructured data.

[0012] Optionally, step S60 includes the following steps: S610, extract the temporal feature vector at time t based on the following feature extraction formula:

[0013] In the formula: The time-series feature vector extracted at time t is used for fault diagnosis or state prediction. The weight matrix maps the window data lines to the hidden feature space; This is a window of raw input data from time tw to t, with each line containing oilfield equipment operating data for one time step. This is a bias term, increasing the model's flexibility; S620 categorizes and summarizes frequently occurring problems during the supervision of oilfield equipment manufacturing. The categorization rules are shown in the following formula:

[0014] In the formula: For feature weights; The standard deviation of historical data; This represents the current monitoring value of the i-th feature; This is the normal baseline value for the i-th feature; S630 constructs a causal reasoning model based on the causes of the problem. The causal reasoning model adopts a Bayesian network structure and spatiotemporal convolution enhancement. S640 constructs a carbon footprint calculation model and generates energy-saving solutions based on equipment operating parameters.

[0015] Optionally, step S70 includes the following steps: S710 seamlessly connects the cloud with the oilfield production management system (EMS) via an open API, synchronizing equipment status and maintenance suggestions in real time, and triggering work order generation or spare parts procurement processes. The S720 automatically adjusts the allocation of equipment operation and maintenance resources based on energy efficiency optimization schemes recommended by the knowledge graph. The S730 enables the seamless connection and sharing of data throughout the entire lifecycle of the device through standardized data protocols. S740 distributes the optimized model and rules to edge nodes, forming a closed-loop update mechanism.

[0016] Optionally, step S80 includes the following steps: S810 uses labeled data from the initial domain knowledge graph to train an initial entity recognition model; The S820 uses few-shot learning to fine-tune the initial entity recognition model and automatically recognizes new entities; S830 actively learns to filter samples with high uncertainty, then hands them over to experts for a small amount of annotation before iteratively optimizing the model; The S840, based on graph embedding and knowledge graph completion technologies, automatically infers the relationships between new entities.

[0017] Optionally, step S820 identifies new entities based on the following formula:

[0018]

[0019] In the formula: This refers to the original input text, such as maintenance records. The template constructor can convert raw input text into text containing... The prompt text; The set of candidate entities, i.e., the target types that the model needs to predict; For entities Corresponding word embeddings; Representing entities yes The probability of the correct answer being located.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Based on multi-source heterogeneous data such as IoT data, equipment maintenance records, and geological parameters, this invention constructs a domain knowledge graph for oilfield supervision, which includes oilfield entity types and industry relationship models. It adopts multilingual knowledge fusion technology to support cross-regional oilfield data semantic alignment and knowledge association, and integrates a dynamic update mechanism. It collects equipment operation data in real time through edge computing nodes, iteratively optimizes the graph reasoning rules by combining digital twin simulation results to update the global knowledge graph, and constructs a system network to locate fault causes and generate energy-saving solutions by combining carbon footprint models. This solves the problems of inaccurate fault cause location, low problem handling efficiency, and lack of data support for maintenance decisions in existing technologies. (2) This invention uses a causal reasoning model to associate the vibration spectrum of equipment with historical failure cases to locate the root cause of failure, and combines a carbon footprint calculation model with equipment operating parameters to generate energy-saving solutions and provide energy efficiency optimization suggestions. It generates maintenance path recommendations with confidence based on knowledge graphs, and associates spare parts inventory with expert experience base maintenance decisions as data support, which can adapt to the new data collection needs brought about by the technological iteration of oilfield equipment. (3) The present invention is based on the automated knowledge node expansion mechanism of transfer learning, which supports VR three-dimensional visualization of equipment disassembly process and risk warning, effectively solves industry pain points such as difficulty in integrating multi-source data and lagging intelligent decision-making, and significantly improves the intelligent level and full life cycle management capability of oilfield equipment supervision process. Attached Figure Description

[0021] Figure 1 A flowchart illustrating the multimodal knowledge graph construction method for oilfield equipment supervision provided by this invention; Figure 2 This is another flowchart of the multimodal knowledge graph construction method for oilfield equipment supervision provided by the present invention; Figure 3 Another flowchart of the multimodal knowledge graph construction method for oilfield equipment supervision provided by the present invention.

[0022] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the application. Rather, these embodiments are provided to make the disclosure more thorough and complete, and to fully convey the scope of the disclosure to those skilled in the art.

[0024] To address the technical challenges of existing technologies in efficiently processing diverse data types, such as IoT data, equipment maintenance records, and geological parameters, which leads to inaccurate fault location, low problem-solving efficiency, and a lack of data support for maintenance decisions, thus failing to meet the new data acquisition requirements brought about by the technological iteration of oilfield equipment, this paper refers to... Figures 1-3 This application provides a multimodal knowledge graph for oilfield equipment manufacturing supervision, the method including the following steps: S10 collects real-time operational data of oilfield equipment during the monitoring process through edge nodes. The operational data includes three main categories: IoT sensor data, equipment maintenance records, and geological parameters. It is further classified according to whether the data format is standardized. Data with standardized format that can be directly quantified and parsed is classified as structured data. Data with semi-standard format that requires extraction of key fields for parsing is classified as semi-structured data. Data without a fixed format that requires feature extraction or entity recognition processing is classified as unstructured data.

[0025] Network sensor data refers to dynamic data collected in real time by various sensors deployed on oilfield equipment and in the field. The core is to reflect the real-time quantitative information of equipment operating status and environmental parameters, such as: the speed, torque, and suspension load of the pumping unit; the inlet and outlet pressure, flow rate, and motor current of the oil pump; the temperature, pressure, and casing pressure at the wellhead; the ambient temperature, humidity, and wind speed at the well site; and the vibration frequency of the pipeline, etc. Equipment maintenance records refer to the operation, maintenance, and fault records of oilfield equipment throughout its entire life cycle. The core of these records is event-based data that reflects the historical status of the equipment and maintenance activities, such as: equipment periodic inspection reports, lubrication and maintenance records, fault occurrence time, fault location, fault cause and fault phenomenon, maintenance measures, and post-maintenance inspection results. Geological parameters refer to the basic data of oilfield reservoirs and geological environment. The core is to reflect the static / quasi-static information of reservoir characteristics and geological structure, such as: porosity, permeability, and gas saturation of reservoir rocks; composition of rock minerals; reservoir depth, formation thickness, and formation dip angle; salinity, density, and viscosity of formation water; and composition, density, and critical pressure of natural gas.

[0026] S20, structured data, semi-structured data, and unstructured data are preprocessed separately to obtain preprocessed data that is uniformly converted into RDF triple format, such as... Figure 2 As shown.

[0027] In an exemplary embodiment, step S20 includes the following steps: S210, structured data such as SCADA (Supervisory and Data Acquisition System) system logs and equipment parameter tables are standardized through SQL parsing and time series alignment, and missing values ​​are filled using a linear interpolation algorithm, as shown in the following formula:

[0028] In the formula: For missing points, For timestamps, and for The effective values ​​before and after; S220, semi-structured data such as maintenance work orders (XML) and sensor configuration (JSON) are extracted using regular expressions and parsed into a unified format using XPath / JSONPath; S230, unstructured data such as geological report PDFs, maintenance videos, and engineer notes, are processed using the BERT-CRF model for text entity recognition, ResNet-50 for image feature extraction, and speech data for conversion to text via ASR technology, as shown in the following formula:

[0029] In the formula: The input word sequence; To be related to the input sequence The corresponding actual label sequence; For a candidate sequence from the set of all possible label sequences; The hidden state of the i-th token output by BERT; The i-th token is predicted as a label The score; The i-th token is predicted as a label The score; and The CRF transition score.

[0030] S240 employs Kalman filtering to eliminate noise interference in the processed structured, semi-structured, and unstructured data. The following formula is used to improve the data quality for oilfield scenarios:

[0031]

[0032] In the formula: for The process noise covariance matrix at time step; This is the scaling factor for the process noise; From Time's up The system state sequence at time 10:00; for The observation noise covariance matrix at time step; is the scaling factor for observation noise; MAD is the median absolute deviation, used to combat outlier interference; The observation sequence window is defined by time. For text and image data, techniques such as image enhancement and stop word removal are used to improve the accuracy of feature extraction. Finally, all data is converted into RDF triple format, laying the foundation for knowledge graph construction.

[0033] S30, based on preprocessed data, performs semantic alignment and knowledge association processing of cross-regional oilfield data through entity similarity calculation and graph embedding relationship mapping. The processed data is stored in a graph database to obtain an initial domain knowledge graph.

[0034] In an exemplary embodiment, step S30 includes the following steps: S310, mark RDF triples in preprocessed data that have numerical, semantic or logical conflicts for the same entity attribute from different data sources, and denote them as contradictory data; S320, Based on the obtained preprocessed data in RDF triplet format, define oilfield entity types and industry relationship models. The oilfield entity types include reservoir and geological entities, oilfield equipment entities, supervision and fault entities, and sensing and operating condition entities. S330 uses edit distance and word vector similarity to align entity names; the formula for calculating entity similarity is as follows:

[0035] In the formula: and These represent the names of two entities to be compared, originating from different regions or data sources. For harmonic weighting coefficients; entity name The corresponding word vector representation; entity name The corresponding word vector representation; S340 uses a graph embedding algorithm to map relationships with different representations to a unified dimension, eliminating ambiguity in relationships; the formula is as follows:

[0036] In the formula: The loss function; is the boundary hyperparameter; (h,r,t) is the positive sampling triple; For negative sampling triples; S350: For contradictory data, a voting mechanism or traceable blockchain records are used to determine the optimal solution; S360 stores the processed data in the graph database Neo4j to obtain the initial domain knowledge graph.

[0037] S40 generates a lightweight knowledge subgraph by receiving real-time device operation data collected from edge nodes in the cloud. It combines digital twin simulation to dynamically optimize reasoning rules, integrates edge node data and knowledge subgraphs in the cloud, performs global knowledge fusion, and uses graph neural networks to obtain an updated domain knowledge graph. The inference rules are optimized using the following formula:

[0038]

[0039] In the formula: Causal weights; This represents the number of iterations. The learning rate; The loss function; The actual type of fault that occurred; The set of observed state data.

[0040] S50 integrates structured, semi-structured, and unstructured data into multimodal data to obtain multimodal feature data.

[0041] In an exemplary embodiment, step S50 includes the following steps: S510 extracts time-series features from structured data that reflect changes in the state of oilfield equipment, using the following formula:

[0042] In the formula: h is the feature vector extracted from the time series data; t represents the current time step; k is the size of the sliding window, which determines the range of historical data analyzed each time; It is a dimension A matrix containing time The original monitoring parameters up to t; It is a weight matrix used to extract temporal feature patterns; This serves as a bias vector, enhancing model flexibility. It is an activation function; S520 uses a graph attention network to parse semi-structured data, first constructing a work order relationship graph: , where nodes Includes device ID, fault code, and responsible personnel. Includes report-problem, problem-solution; the graph attention network formula used is as follows:

[0043] In the formula: It is a scalar, and its value range is... Between, represents the normalized attention weight of node i to node j, reflecting the importance of j to i; node features and An embedded representation of original features such as device ID and fault code; To share the weight matrix, a uniform linear transformation is performed on all nodes to extract higher-order features; This is the attention parameter vector, used to calculate the correlation score between nodes; The activation function is used to alleviate the gradient vanishing problem while preserving nonlinear features; This is an exponential function used to map attention scores to a positive range so that normalization can be achieved through the Softmax function; S530 employs a graph-text joint encoder to achieve cross-database semantic mapping between equipment parameters and fault cases for unstructured data. It records data traceability information based on blockchain technology. Utilizing the decentralized, immutable, and traceable characteristics of blockchain, it can be used to record data traceability information, thereby ensuring the trustworthiness and integrity of knowledge nodes. The graph-text joint encoder formula is as follows:

[0044] In the formula: The fused multimodal feature vector is used for entity or relation representation in knowledge graphs; For a fully connected neural network, the concatenated features are mapped to a unified multimodal semantic space; For the input text information; The input image information.

[0045] S60, based on an updated domain knowledge graph and combined with multimodal feature data, categorizes and summarizes frequently occurring problem data of oilfield equipment during the manufacturing supervision process. It then constructs a causal reasoning model based on the causes of these problems to locate the root causes of failures and generate energy-saving solutions by combining this with a carbon footprint model. Figure 3 As shown.

[0046] In an exemplary embodiment, step S60 includes the following steps: S610, extract the temporal feature vector at time t based on the following feature extraction formula:

[0047] In the formula: The time-series feature vector extracted at time t is used for fault diagnosis or state prediction. The weight matrix maps the window data lines to the hidden feature space; This is a window of raw input data from time tw to t, with each line containing oilfield equipment operating data for one time step. This is a bias term, increasing the model's flexibility; S620 categorizes and summarizes frequently occurring problems during the supervision of oilfield equipment manufacturing. The categorization rules are shown in the following formula:

[0048] In the formula: For feature weights, The standard deviation of historical data. Let i be the current monitoring value of the i-th feature. This is the normal baseline value for the i-th feature; S630 constructs a causal reasoning model based on the causes of the problem. The causal reasoning model adopts a Bayesian network structure and spatiotemporal convolution enhancement. The formula for the Bayesian network structure is as follows:

[0049] In the formula: Indicates a target failure event; Indicates that during observation Under these conditions, the equipment malfunctions. The probability of; Indicates the observed equipment status indicators; Indicates the state of a given parent node The probability of occurrence; Indicates the initial probability of a failure event; The formula for enhancing spatiotemporal convolution is as follows:

[0050] In the formula: Indicates the first The layer's output features capture node representations after high-order topological relationships; Represents a non-linear activation function; Degree matrix; This represents the adjacency matrix, indicating the connection relationships between devices and sensors; For the first The trainable weight matrix of the layer is used for feature transformation.

[0051] S640: Constructs a carbon footprint calculation model and generates energy-saving solutions based on equipment operating parameters; Carbon footprint calculation uses a real-time emission factor model, as shown in the following formula:

[0052] In the formula: Fuel coefficient ( ); For equipment power ( ); For equipment uptime; To produce emission factors; For the quality of consumables; For direct emissions; It is an indirect emission.

[0053] S70 connects the cloud with the oilfield production management system (EMS) via an open API, and distributes optimized models and rules to edge nodes to form a closed-loop update mechanism.

[0054] In an exemplary embodiment, step S70 includes the following steps: S710 seamlessly integrates with the oilfield production management system (EMS) via an open API, synchronizing equipment status and maintenance suggestions in real time, and triggering work order generation or spare parts procurement processes. The S720 automatically adjusts the allocation of equipment operation and maintenance resources based on energy efficiency optimization schemes recommended by the knowledge graph. The S730 enables the seamless connection and sharing of data throughout the entire lifecycle of the device through standardized data protocols. S740 distributes the optimized model and rules to edge nodes, forming a closed-loop update mechanism.

[0055] S80, based on the updated domain knowledge graph, uses transfer learning technology to automate the annotation and relationship completion of knowledge nodes in the newly connected oilfield production management system (EMS).

[0056] In an exemplary embodiment, step S80 includes the following steps: S810 uses labeled data from the initial domain knowledge graph to train the initial entity recognition model; The S820 uses few-shot learning to fine-tune the initial entity recognition model and automatically recognizes new entities; New entities are identified based on the following formula:

[0057]

[0058] In the formula: This refers to the original input text, such as maintenance records. The template constructor can convert raw input text into text containing... The prompt text; The set of candidate entities, i.e., the target types that the model needs to predict; For entities Corresponding word embeddings; Representing entities yes The probability of the correct answer being located.

[0059] S830 actively learns to filter samples with high uncertainty, then hands them over to experts for a small amount of annotation before iteratively optimizing the model; The S840, based on graph embedding and knowledge graph completion technologies, automatically infers the relationships between new entities.

[0060] S90, based on the updated domain knowledge graph and the problem data and fault cause data in step S60, integrate AR / VR terminals to realize three-dimensional visual knowledge navigation.

[0061] In an exemplary embodiment, a VR environment is built based on the Unity engine and knowledge graph data is integrated to enable interactive disassembly of the device's 3D model. Users can simulate the disassembly and assembly process using VR controllers, and the system displays step prompts in real time (such as "remove the flange bolts first"). High-risk operations (such as "overpressure warning") trigger AR overlay warning information (red flashing prompt). Combined with digital twin technology, the real-time status of the device is dynamically mapped to the VR scene (such as highlighting abnormal vibration areas). VR headsets are used for immersive training, AR glasses support on-site repair guidance, and mobile devices provide lightweight knowledge query functions.

[0062] In the description of this application, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0063] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0064] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0067] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the technical scope disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

[0069] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

Claims

1. A method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision, characterized in that, Includes the following steps: S10 is an edge computing node that collects real-time operational data of oilfield equipment during the manufacturing process and classifies it into structured data, semi-structured data and unstructured data. S20: Preprocess structured data, semi-structured data, and unstructured data separately to obtain preprocessed data that is uniformly converted into RDF triple format; S30, based on preprocessed data, performs semantic alignment and knowledge association processing of cross-regional oilfield data through entity similarity calculation and graph embedding relationship mapping. The processed data is stored in a graph database to obtain an initial domain knowledge graph. S40 generates a lightweight knowledge subgraph by receiving real-time device operation data collected from edge nodes in the cloud, combines digital twin simulation to dynamically optimize reasoning rules, and uses graph neural networks to obtain an updated domain knowledge graph. S50 integrates structured, semi-structured, and unstructured data into multimodal data to obtain multimodal feature data; S60, based on an updated domain knowledge graph and combined with multimodal feature data, classifies and summarizes the problem data that frequently occurs in the oilfield equipment during the manufacturing process, and constructs a causal reasoning model based on the cause of the problem to locate the cause of the failure and obtain the cause data of the failure, and generates energy-saving solutions by combining the carbon footprint model. S70 connects the cloud with the oilfield production management system (EMS) via an open API, and distributes the updated domain knowledge graph and causal reasoning model to edge nodes, forming a closed-loop update mechanism. S80, based on the updated domain knowledge graph, uses transfer learning technology to automate the annotation and relationship completion of knowledge nodes in the newly connected oilfield production management system (EMS). S90, based on the updated domain knowledge graph and the problem data and fault cause data in step S60, integrate AR / VR terminals to realize three-dimensional visual knowledge navigation.

2. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 1, characterized in that, Step S20 includes the following steps: S210 standardizes structured data through SQL parsing and time series alignment, and uses a linear interpolation algorithm to fill in missing values; S220: Semi-structured data uses regular expressions to extract key fields and then parses them into a uniform format using XPath. S230: Unstructured data is used for text entity recognition using the BERT-CRF model, image features are extracted using ResNet-50, and speech data is converted into text using ASR technology. S240 uses Kalman filtering to eliminate noise interference in the processed structured, semi-structured, and unstructured data, and then improves the accuracy of feature extraction through image enhancement and stop word removal techniques. Finally, it converts all data into preprocessed data in RDF triple format.

3. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 1, characterized in that, Step S30 includes the following steps: S310, mark RDF triples in preprocessed data that have numerical, semantic or logical conflicts for the same entity attribute from different data sources, and denote them as contradictory data; S320 defines oilfield entity types and industry relationship models based on preprocessed data; S330, uses edit distance and word vector similarity to align entity names; S340 uses a graph embedding algorithm to map relationships with different representations to a unified dimension, thus eliminating relationship ambiguity. S350: For contradictory data, a voting mechanism or traceable blockchain records are used to determine the optimal solution; S360 stores the processed data in the graph database Neo4j to obtain the initial domain knowledge graph.

4. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 3, characterized in that, Step S330 calculates entity similarity based on the following formula: In the formula: , These represent the names of two entities to be compared, originating from different regions or data sources. For harmonic weighting coefficients; For entity name The corresponding word vector representation; For entity name The corresponding word vector representation.

5. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 3, characterized in that, In step S340, the relationships expressed differently are mapped to a unified dimension based on the following formula: In the formula: The loss function; For boundary hyperparameters; For positive sampling triples; For negative sampling triples.

6. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 1, characterized in that, Step S50 includes the following steps: S510, extract time-series features from structured data that reflect changes in the state of oilfield equipment; S520 uses a graph structure to parse semi-structured data and uses a graph attention network to extract semantic association features of nodes and relationships. The S530 uses a combined text and image encoder to achieve cross-database semantic mapping between equipment parameters and fault cases for unstructured data.

7. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 1, characterized in that, Step S60 includes the following steps: S610, extract the temporal feature vector at time t based on the following feature extraction formula: In the formula: The time-series feature vector extracted at time t is used for fault diagnosis or state prediction. The weight matrix maps the window data lines to the hidden feature space; From time arrive The raw input data window contains oilfield equipment operation data for one time step in each row; This is a bias term, increasing the model's flexibility; S620 categorizes and summarizes frequently occurring problems during the supervision of oilfield equipment manufacturing. The categorization rules are shown in the following formula: In the formula: For feature weights; The standard deviation of historical data; This represents the current monitoring value of the i-th feature; This is the normal baseline value for the i-th feature; S630 constructs a causal reasoning model based on the causes of the problem. The causal reasoning model adopts a Bayesian network structure and spatiotemporal convolution enhancement. S640 constructs a carbon footprint calculation model and generates energy-saving solutions based on equipment operating parameters.

8. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 1, characterized in that, Step S70 includes the following steps: S710 seamlessly connects the cloud with the oilfield production management system (EMS) via an open API, synchronizing equipment status and maintenance suggestions in real time, and triggering work order generation or spare parts procurement processes. The S720 automatically adjusts the allocation of equipment operation and maintenance resources based on energy efficiency optimization schemes recommended by the knowledge graph. The S730 enables the seamless connection and sharing of data throughout the entire lifecycle of the device through standardized data protocols. S740 distributes the optimized model and rules to edge nodes, forming a closed-loop update mechanism.

9. The method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 1, characterized in that, Step S80 includes the following steps: S810 uses labeled data from the initial domain knowledge graph to train an initial entity recognition model; The S820 uses few-shot learning to fine-tune the initial entity recognition model and automatically recognizes new entities; S830 actively learns to filter samples with high uncertainty, then hands them over to experts for a small amount of annotation before iteratively optimizing the model; The S840, based on graph embedding and knowledge graph completion technologies, automatically infers the relationships between new entities.

10. A method for constructing a multimodal knowledge graph for oilfield equipment manufacturing supervision according to claim 9, characterized in that, Step S820 identifies new entities based on the following formula: In the formula: This refers to the original input text, such as maintenance records. The template constructor can convert the original input text into a format containing... The prompt text; The set of candidate entities, i.e., the target types that the model needs to predict; Candidate entities A specific entity in; For entities The corresponding word embeddings; For prompt text The context semantic vector corresponding to the position; To The vector after performing the transpose operation; To The vector after performing the transpose operation; Representing entities yes The probability of the correct answer being located.

Images