Intelligent inspection fault diagnosis method and system adopting six-dimensional judgment

Abstract

The invention discloses an intelligent inspection fault diagnosis method and system adopting six-dimensional judgment, and belongs to the technical field of oil field gathering and transportation intelligent inspection. A data access module is used for collecting information of field sensors deployed in different site devices in the oil field gathering and transportation process; protocol analysis, time alignment and semantic mapping are carried out on collected data, data collected by a single-point sensor in the oil field gathering and transportation process are packaged into a unified measuring point object UPO, and a standard data structure containing equipment ID, design attributes, real-time values, historical sequences, alarm threshold values, mechanism parameters and upstream and downstream pointers is constructed. According to the intelligent inspection fault diagnosis method and system adopting six-dimensional judgment, the accuracy of abnormal recognition, the efficiency of root cause positioning and the scientificity of regulation and control decision are improved by fusing multi-dimensional judgment, intelligent reasoning and distributed cooperative fault diagnosis, and powerful support is provided for safe and stable operation of an oil field gathering and transportation system.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent inspection of oilfield gathering and transportation, and in particular relates to an intelligent inspection fault diagnosis method and system using six-dimensional judgment. Background Art

[0002] In oilfield gathering and transportation systems, the stable operation of key facilities such as pumps, pipelines, and water blending equipment is directly related to crude oil extraction efficiency and production safety. With the accelerated advancement of "less-staffed and unmanned inspections" in the intelligent construction of oil and gas fields, station operations face multiple challenges, including complex equipment, harsh environments, data fragmentation, and high response requirements. Existing diagnosis mostly relies on single threshold judgment, and only identifies anomalies by monitoring whether the parameters exceed the preset range. It is difficult to cope with complex situations such as operating condition fluctuations and interference signals. The anti-disturbance ability is weak and it is easy to make misjudgments or missed judgments, resulting in problems such as pump group vibration degradation and pipeline leakage. There is no early warning, which poses a safety hazard. At the same time, the root cause positioning of the fault lacks systematicity. When an anomaly occurs, it is difficult to quickly trace the fault link, and it is often necessary to rely on manual experience to troubleshoot. This is not only time-consuming and labor-intensive, but may also lead to inaccurate root cause judgments due to differences in experience, affecting the effectiveness of subsequent regulatory decisions. There are problems of low practicality and functionality. Summary of the Invention

[0003] This invention aims to address technical bottlenecks in multi-source data fusion, accurate anomaly identification, intelligent root cause reasoning, predictive maintenance, and coordinated energy consumption and safety control in high-risk, remote, and unmanned scenarios such as oilfield gathering and transportation. To this end, this invention proposes an intelligent inspection and fault diagnosis method and system using six-dimensional judgment, addressing these technical issues by improving detection and processing methods.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: An intelligent inspection fault diagnosis system using six-dimensional judgment includes a data access module, an edge analysis module, an intelligent reasoning module, and an execution control model: The data access module collects information from field sensors deployed in equipment at different sites during the oilfield gathering and transportation process, performs protocol parsing, time alignment, and semantic mapping on the collected data, encapsulates the data collected by single-point sensors during the oilfield gathering and transportation process into a unified measurement point object (UPO), and constructs a standard data structure containing device ID, design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters, and upstream and downstream pointers. The edge analysis module dynamically constructs a single-point model for each UPO object, forms a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table, performs a six-dimensional judgment score calculation on the constructed single-point model, outputs a six-dimensional anomaly score vector, and performs Bayesian posterior update and logical voting fusion on the six-dimensional score vector to output an anomaly level label; The intelligent reasoning module maps abnormal measurement points to knowledge graph entities based on the abnormal level labels of the single-point model. It calculates the fault path score based on the causal chain weights, upstream and downstream impact paths, and mechanism residual information, and uses a reinforcement learning ranking model to output the top-N root cause sequence to form a priority list of issues to be addressed. It then outputs the adjustment results based on the current equipment site conditions and operating objectives. The execution control module sends control instructions to the DCS, PLC or intelligent actuator through the standard industrial protocol based on the output adjustment result, adjusts and controls the abnormal equipment, and completes the closed loop response to the abnormal state.

[0005] Furthermore, the data access module collects information from field sensors deployed in equipment at different sites during oilfield gathering and transportation, and performs protocol parsing, time alignment, and semantic mapping on the collected data, specifically including the following steps: The sensors deployed in different sites during the oilfield gathering and transportation process collect flow, pressure, temperature, vibration, and electrical quantity data, perform protocol parsing and matching on the key features of the initial data stream collected by the sensors, and calculate the protocol matching score. ,in represents the weight of the h-th feature, Represents the matching degree of the hth feature. When it matches, the value is 1 and when it does not match, the value is 0. When it is used, it means it belongs to the corresponding agreement; A timestamp is added to the data collected by each sensor, and the sensor data is time-aligned through interpolation. Semantic mapping is then performed based on the type of sensor and the meaning of the collected data.

[0006] Furthermore, the data collected by single-point sensors in the oilfield gathering and transportation process is encapsulated into a unified measurement point object UPO, and a standard data structure including design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters and upstream and downstream pointers is constructed, specifically including the following steps: Define the structure of the unified measurement point object UPO, including device ID, design attributes, real-time value, historical sequence, alarm threshold, mechanism parameters and upstream and downstream pointers; Encapsulate the data after protocol parsing, time alignment and semantic mapping into UPO to build a standard data structure ,in It is a combination code of equipment number and measuring point sensor type. Represents the real-time value collected, Store the historical sequence formed by the latest N real-time values ​​for the time window, Represents the rated value in the design properties of the current measurement point object. are the high-alarm threshold and the low-alarm threshold, It is the ID of the upstream and downstream pointers, i.e. the predecessor node and the successor node. Represents the parameters of the mechanism equation.

[0007] Furthermore, the edge analysis module dynamically constructs a single-point model for each UPO object, forming a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table, specifically including the following steps: Based on the unified measurement point object UPO encapsulated by the data access module, a single-point model is dynamically constructed for the standard data structure, and the static parameters in the UPO are constructed as static attribute sub-tables, including design attributes and alarm thresholds; By using the real-time values ​​and historical sequence data in the UPO object, a dynamic state table is constructed, and by calculating the statistical characteristics of the data, the trend slope of the measurement point is calculated in real time using the linear regression equation; According to the upstream and downstream association pointers in the UPO object, an upstream and downstream linked list is constructed so that each node in the upstream and downstream linked list corresponds to a measuring point, and the upstream and downstream relationships are represented by pointer connections, that is, other measuring points associated with the current measuring point; The mechanism parameters are extracted from the UPO object to construct the mechanism parameter sub-table, and the static attribute sub-table, dynamic state sub-table, mechanism parameter table and upstream and downstream link tables are integrated to form a decidable multi-dimensional model structure.

[0008] Furthermore, the six-dimensional judgment score calculation is performed on the constructed single-point model to output a six-dimensional anomaly score vector, and the six-dimensional score vector is fused with Bayesian posterior update and logical voting to output an anomaly level label, including the following steps: Based on the single-point models constructed at different measuring points, six-dimensional judgment scores are calculated for different measuring point models and a six-dimensional abnormality score vector is output. The six dimensions include design boundaries, operating boundaries, upstream and downstream coupling, mechanism residuals, historical evolution, high and low reporting thresholds, to form a normalized six-dimensional score vector. ,in , specifically including the following steps: Design boundary scoring , determine the normal range of design parameters through the static property table , based on the real-time value of the current measuring point , calculate the design boundary deviation : ; Deviation from the design boundary Map to Get design boundary score within the interval ; Determine the operation boundary by the statistical characteristics of historical data in the dynamic state table, based on the mean value in the historical data and standard deviation Get the operating range ,in is a constant to calculate the deviation of the operating boundary , get the running boundary score ; According to the current measuring point and the upstream and downstream related measuring points, calculate the Pearson correlation coefficient between the real-time value of the current measuring point and the real-time value of the upstream and downstream related measuring points ,in is the current measuring point, For other measurement points in the upstream and downstream associated measurement point set, based on the normal correlation range , calculate the upstream and downstream coupling deviation : ; in, The average value of the real-time value of the current measuring point and the correlation coefficient of the upstream and downstream measuring points is used to obtain the upstream and downstream coupling score. , according to the parameters in the mechanism equation parameter subtable, use the mechanism equation to calculate the theoretical value of the measuring point And calculate the residual , according to the upper limit of the normal range of residuals , and obtain the mechanism residual score ,when hour, ; The trend slope obtained from the historical data series , according to the normal trend slope range , the calculation can obtain the historical evolution deviation , get the historical evolution score : ; Based on high and low alarm thresholds, when If it falls within the high and low reporting threshold range, the high and low reporting threshold score ,when If it does not fall within the high or low reporting threshold range, the high or low reporting threshold score ; Based on the abnormality level normal, attention, abnormal, severe, respectively , according to historical data statistics, the prior probability of each abnormal level is obtained ,and , for each abnormal level , respectively calculate the six-dimensional scoring vectors under different abnormal levels The likelihood probability , calculate the posterior probability according to the Bayesian formula ,in Represents the observed six-dimensional rating vector After that, the current measurement point belongs to the abnormal level At the same time, when the score of any of the four dimensions is greater than 0.7, a response is triggered synchronously, the abnormality level is judged to be severe, and the abnormality level label is output to enter the subsequent module.

[0009] Furthermore, the intelligent reasoning module includes the following steps: By deploying an embedded Neo4j graph database, we constructed a five-element graph of "equipment-measurement point-process-fault-measurement." Based on the anomaly level labels of the single-point model, we mapped the abnormal measurement points to knowledge graph entities. Using causal chain weights, upstream and downstream impact paths, and mechanism residual information, we calculated the fault path score and used a reinforcement learning ranking model to output the top-N root cause sequence, forming a prioritized list of issues to be addressed. The specific steps are as follows: Collect historical data related to equipment, measurement points, processes, faults, and measures, including basic equipment information, real-time data from measurement points, process flow and parameters, historical records of faults, and corresponding control measures. Create nodes and relationships in the embedded Neo4j graph database to construct a five-element graph of "equipment-measurement-process-fault-measurement". Determine the causal relationship between measurement points, processes, and faults based on historical data, and use the hierarchical analysis method to evaluate the importance of the causal relationship to obtain the weight of each causal chain. Based on the physical principles of the system and process knowledge, establish the mechanism equation between measurement points and process parameters. ,in represents the theoretical value of the process parameter, Represents the measured value of the measuring point, the actual measured value of the process parameter Compared with the theoretical calculated value Perform difference calculation to obtain residuals, distribute the residuals to each measuring point, and obtain the residual contribution value of each measuring point; Match the identified abnormal measurement points with the measurement point nodes in the knowledge graph, map the abnormal information to the knowledge graph, use breadth-first search to search all possible paths from the abnormal measurement point to the fault node in the knowledge graph, and calculate the score of each fault path by combining the causal chain weight, upstream and downstream impact paths and mechanism residual information. : ; in, Represents the fault path contains nodes and edges, Representative Nodes to The causal chain weight of each node, represent The residual contribution value of each node is calculated, and an offline Q-learning architecture reinforcement learning ranking model is used to output the Top-N root cause sequence to form a priority list of problems to be handled.

[0010] Furthermore, the intelligent reasoning module outputs adjustment results based on the priority list of issues to be addressed, combined with the current equipment site conditions and operation goals, including the following steps: Each root cause in the root cause list and the fault context information of the historical case are converted into a semantic vector. The text information is encoded into a vector using the BERT model. The cosine similarity is used to calculate the similarity between the root cause list and the historical case. The semantic vector of , historical cases The fault context semantic vector is , cosine similarity for: ; Based on the similarity threshold The root cause similarity in the root cause list is greater than The historical cases of similar fault scenarios are combined to form a case set of similar fault scenarios. The corresponding historical disposal paths are extracted from the case set of similar fault scenarios. The on-site working condition information of the current equipment is recorded as a state vector. The operation goals of the equipment are clarified, including safety goals, energy consumption goals, and response efficiency goals. The historical disposal paths are used as reinforcement learning actions. The reward function is used to evaluate the pros and cons of each action. The optimal strategy is obtained through Q-learning to output the adjustment results.

[0011] Furthermore, the execution control module sends control instructions to the DCS, PLC or intelligent actuator through the standard industrial protocol based on the output adjustment result, adjusts and controls the abnormal equipment, and completes the closed loop response to the abnormal state, including the following steps: For measurement points with severe abnormality levels, control instructions are sent to the DCS, PLC, or intelligent actuator based on the upstream and downstream indicators of the measurement point. Safety interlocks are performed on the measurement point and its upstream and downstream measurement points, and the equipment is stopped and started. Based on the adjustment results output by the intelligent reasoning module, the abnormal equipment is adjusted and controlled according to the control instructions in the adjustment results. At the same time, the status feedback data of the measuring points are collected in real time and transmitted to the edge analysis module and the intelligent reasoning module.

[0012] An intelligent inspection fault diagnosis method using six-dimensional judgment includes the following steps: S1. Collect information from field sensors deployed in equipment at different sites during oilfield gathering and transportation, perform protocol parsing, time alignment, and semantic mapping on the collected data, encapsulate the data collected by single-point sensors during oilfield gathering and transportation into a unified measurement point object (UPO), and construct a standard data structure containing device ID, design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters, and upstream and downstream pointers. S2. Dynamically construct a single-point model for each UPO object, forming a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table. Perform a six-dimensional judgment score calculation on the constructed single-point model, output a six-dimensional anomaly score vector, and perform Bayesian posterior update and logical voting fusion on the six-dimensional score vector to output an anomaly level label. S3. Based on the abnormality level label of the single-point model, the abnormal measurement point is mapped to the knowledge graph entity. Through the causal chain weight, upstream and downstream impact path and mechanism residual information, the fault path score is calculated and the reinforcement learning sorting model is used to output the Top-N root cause sequence to form a priority list of problems to be handled. Combined with the current equipment site conditions and operating goals, the adjustment results are output. Based on the output adjustment results, the control instructions are sent to the DCS, PLC or intelligent actuator through the standard industrial protocol to adjust and control the abnormal equipment, completing the response closed loop to the abnormal state.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. In this invention, a six-dimensional judgment link of "design boundary + operation boundary + upstream and downstream coupling + mechanism residual + historical evolution + high and low reporting thresholds" is constructed through the edge analysis module to form a high-dimensional scoring matrix for each measuring point. Compared with the traditional judgment method based on a single threshold, it has stronger anti-disturbance ability and state discrimination accuracy. It can realize the early identification of oilfield gathering and transportation problems such as pump group vibration degradation, water injection abnormalities, and pipeline leakage, thereby forming a more comprehensive risk prevention and control system; 2. In this invention, a knowledge graph reasoning engine for fault links is constructed through an intelligent reasoning module. Abnormal measurement points are mapped to fault modes and path sorting is performed based on causal strength scores. A reinforcement learning algorithm is used to output a Top-N root cause sequence, which can independently output high-confidence root cause location results. Furthermore, through reinforcement learning sorting and a multi-objective optimization algorithm, the risk level, energy consumption cost, and response time are automatically evaluated and weighed, ultimately recommending the optimal control instructions, thus enhancing practicality and functionality. 3. In the present invention, by building a distributed collaborative system, the six-dimensional judgment and control logic are completed at the edge to perform abnormality identification and linkage control chain, meeting the needs of high-risk scenarios such as oil field well sites, remote unmanned stations, and field mobile operations, and enhancing the system's risk resistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1This is a block diagram of an intelligent inspection fault diagnosis system using six-dimensional judgment according to the present invention; Figure 2 This is a diagram of a single-point model structure dynamically constructed for a UPO object in an intelligent inspection fault diagnosis system using six-dimensional judgment according to the present invention. DETAILED DESCRIPTION

[0015] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0016] Example 1: Figure 1 As shown in the figure, an intelligent inspection fault diagnosis system using six-dimensional judgment includes a data access module, an edge analysis module, an intelligent reasoning module, and an execution control model: The data access module collects information from field sensors deployed in equipment at different sites during oilfield gathering and transportation, performs protocol parsing, time alignment, and semantic mapping on the collected data, encapsulates the data collected by single-point sensors during oilfield gathering and transportation into a unified measurement point object (UPO), and constructs a standard data structure that includes device ID, design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters, and upstream and downstream pointers. The data access module collects information from field sensors deployed in equipment at different sites during oilfield gathering and transportation, and performs protocol parsing, time alignment, and semantic mapping on the collected data. The specific steps include: The sensors deployed in different sites during the oilfield gathering and transportation process collect flow, pressure, temperature, vibration, and electrical quantity data, perform protocol parsing and matching on the key features of the initial data stream collected by the sensors, and calculate the protocol matching score. ,in represents the weight of the h-th feature, Represents the matching degree of the hth feature. When it matches, the value is 1 and when it does not match, the value is 0. When it is used, it means it belongs to the corresponding agreement; It should be noted that The value is the average number of h features, It can also be divided unequally based on the representativeness of features in different protocols, but the sum of all weights must be 1. Key features include frame structure features and transport layer features, specifically frame start character, end character, frame length range, port number, and message header identifier.

[0017] A timestamp is added to the data collected by each sensor, and the sensor data is time-aligned through interpolation. Semantic mapping is then performed based on the type of sensor and the meaning of the collected data.

[0018] It should be noted that when semantic mapping is performed based on the sensor type and the meaning of the collected data, the corresponding sensor collected data is mapped to the corresponding data according to the sensor type, such as the data from the temperature sensor is identified as "temperature", and the data from the flow sensor is identified as "flow". At the same time, the device ID, sensor measurement point type, and sensor value are mapped and associated. 、 ,in Represents timestamp and data value respectively. Each time point in the sequence Find the closest two time points in a sequence , making Use linear interpolation formula to calculate the aligned value .

[0019] The data collected by single-point sensors in the oilfield gathering and transportation process is encapsulated into a unified measurement point object (UPO). A standard data structure is constructed that includes design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters, and upstream and downstream pointers. The specific steps include: Define the structure of the unified measurement point object UPO, including device ID, design attributes, real-time value, historical sequence, alarm threshold, mechanism parameters and upstream and downstream pointers; Encapsulate the data after protocol parsing, time alignment and semantic mapping into UPO to build a standard data structure ,in It is a combination code of equipment number and measuring point sensor type. Represents the real-time value collected, Store the historical sequence formed by the latest N real-time values ​​for the time window, Represents the rated value in the design properties of the current measurement point object. are the high-alarm threshold and the low-alarm threshold, It is the ID of the upstream and downstream pointers, i.e. the predecessor node and the successor node. Represents the parameters of the mechanism equation.

[0020] It should be noted that The historical sequence formed by the most recent N real-time values ​​is stored for the time window, where the value of N is 10 and can also be adjusted according to actual conditions. The device number and the measurement point sensor type combination code are unique to distinguish different measurement points. This can be determined through the equipment's design documentation or instructions. The high alarm threshold and low alarm threshold need to be set based on the experience method in combination with the device characteristics of the measurement point and the actual application requirements. Represents the mechanism equation parameters. Based on the physical model or empirical formula, the mechanism equation parameters of different measuring points are determined. For example, for a temperature measuring point of a heat exchanger, the mechanism equation parameters include heat exchange coefficient, fluid specific heat capacity, etc.

[0021] Example 2: The edge analysis module dynamically constructs a single-point model for each UPO object, forms a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table, performs a six-dimensional judgment score calculation on the constructed single-point model, outputs a six-dimensional anomaly score vector, and performs Bayesian posterior update and logical voting fusion on the six-dimensional score vector to output an anomaly level label;\ The edge analysis module dynamically builds a single-point model for each UPO object, forming a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table. Specifically, it includes the following steps: Based on the unified measurement point object UPO encapsulated by the data access module, a single-point model is dynamically constructed for the standard data structure, and the static parameters in the UPO are constructed as static attribute sub-tables, including design attributes and alarm thresholds; By using the real-time values ​​and historical sequence data in the UPO object, a dynamic state table is constructed, and by calculating the statistical characteristics of the data, the trend slope of the measurement point is calculated in real time using the linear regression equation; According to the upstream and downstream association pointers in the UPO object, an upstream and downstream linked list is constructed so that each node in the upstream and downstream linked list corresponds to a measuring point, and the upstream and downstream relationships are represented by pointer connections, that is, other measuring points associated with the current measuring point; The mechanism parameters are extracted from the UPO object to construct the mechanism parameter sub-table, and the static attribute sub-table, dynamic state sub-table, mechanism parameter table and upstream and downstream link tables are integrated to form a decidable multi-dimensional model structure.

[0022] It should be noted that the static attribute table is used to describe the design boundaries of different measuring point equipment, such as the measuring range, the parameter interval for normal operation, etc. In the dynamic state table, the statistical characteristics of the data, such as the mean and standard deviation, are calculated in combination with the historical sequence data. The real-time values ​​are dynamically analyzed based on the 3sigma principle to assist in subsequent abnormality determination. The mechanism parameter subtable is used to store the parameters in the physical model or empirical formula related to the measuring point.

[0023] Perform six-dimensional judgment scoring calculation on the constructed single-point model, output a six-dimensional anomaly score vector, and perform Bayesian posterior update and logical voting fusion on the six-dimensional score vector to output an anomaly level label. The steps include: Based on the single-point models constructed at different measuring points, six-dimensional judgment scores are calculated for different measuring point models and a six-dimensional abnormality score vector is output. The six dimensions include design boundaries, operating boundaries, upstream and downstream coupling, mechanism residuals, historical evolution, high and low reporting thresholds, to form a normalized six-dimensional score vector. ,in , specifically including the following steps: Design boundary scoring , determine the normal range of design parameters through the static property table , based on the real-time value of the current measuring point , calculate the design boundary deviation : ; Deviation from the design boundary Map to Get design boundary score within the interval ; It should be noted that the normal range of design parameters is obtained through the design properties in the static property table. hour, ,when hour, .

[0024] Determine the operation boundary by the statistical characteristics of historical data in the dynamic state table, based on the mean value in the historical data and standard deviation Get the operating range ,in is a constant to calculate the deviation of the operating boundary , get the running boundary score ; It should be noted that It is the operating range constant, with a value range of 1-3, usually set to 2. It can also be set according to actual usage to control the size of the operating range.

[0025] According to the current measuring point and the upstream and downstream related measuring points, calculate the Pearson correlation coefficient between the real-time value of the current measuring point and the real-time value of the upstream and downstream related measuring points ,in is the current measuring point, For other measurement points in the upstream and downstream associated measurement point set, based on the normal correlation range , calculate the upstream and downstream coupling deviation : ; in, The average value of the real-time value of the current measuring point and the correlation coefficient of the upstream and downstream measuring points is used to obtain the upstream and downstream coupling score. , according to the parameters in the mechanism equation parameter subtable, use the mechanism equation to calculate the theoretical value of the measuring point And calculate the residual , according to the upper limit of the normal range of residuals , and obtain the mechanism residual score ,when hour, ; The trend slope obtained from the historical data series , according to the normal trend slope range , the calculation can obtain the historical evolution deviation , get the historical evolution score : ; Based on high and low alarm thresholds, when If it falls within the high and low reporting threshold range, the high and low reporting threshold score ,when If it does not fall within the high or low reporting threshold range, the high or low reporting threshold score ; It should be noted that the correlation coefficient range is obtained by clustering the correlation coefficients in the historical data through a clustering algorithm, dividing the correlation coefficients into different clusters, and selecting the cluster containing most of the normal data as the normal correlation range. The mechanism equation needs to be determined according to the type of equipment at different measuring points and the type of related parameters involved. For the residual, it is necessary to measure and calculate the system multiple times during the experiment, record the residual data to analyze the distribution of the residual and set the upper limit value. It can also be achieved by using computer simulation to generate a large amount of simulation data, and then calculate the residual and analyze its distribution. It needs to be set according to the specific actual situation of the system deployment. The slope range needs to be determined according to the physical characteristics, operating requirements and operating conditions of the system's application environment to determine the reasonable range of the slope. For example, in the temperature control system of oil field gathering and transportation, the temperature rise or fall rate is limited by the power of the heating or cooling equipment. Therefore, the reasonable trend slope range of the temperature change needs to be set according to the power of the related equipment. If the slope exceeds this range, it means that the current measuring point has a fault.

[0026] Based on the abnormality level normal, attention, abnormal, severe, respectively , according to historical data statistics, the prior probability of each abnormal level is obtained ,and , for each abnormal level , respectively calculate the six-dimensional scoring vectors under different abnormal levels The likelihood probability , calculate the posterior probability according to the Bayesian formula ,in Represents the observed six-dimensional rating vector After that, the current measurement point belongs to the abnormal level At the same time, when the score of any of the four dimensions is greater than 0.7, a response is triggered synchronously, the abnormality level is judged to be severe, and the abnormality level label is output to enter the subsequent module.

[0027] It should be noted that The product of the likelihood probability and the prior probability of all abnormal levels is summed, and normalization is achieved through the summation term to ensure that the sum of all posterior probabilities is 1. By constructing a dynamic modeling instance at each measuring point, the abnormality level results are generated through parallel scoring and Bayesian fusion algorithm to overcome the problems of high false alarm rate and serious missed alarm rate of traditional single threshold judgment, and realize high-accuracy identification and early warning of typical oilfield gathering and transportation station faults such as pump vibration exceeding the limit, heating abnormality, water injection abnormality and other working conditions. The method and device structure of the Bayesian fusion algorithm output abnormality level are suitable for the status identification of equipment such as gathering and transportation, heating, and water injection.

[0028] Example 3: The intelligent reasoning module maps abnormal measurement points to knowledge graph entities based on the anomaly level labels of the single-point model. It calculates the fault path score based on the causal chain weights, upstream and downstream impact paths, and mechanism residual information. It then uses a reinforcement learning ranking model to output a top-N root cause sequence, forming a prioritized list of issues to be addressed. It then outputs the adjustment results based on the current equipment site conditions and operational objectives. The intelligent reasoning module includes the following steps: By deploying an embedded Neo4j graph database, we constructed a five-element graph of "equipment-measurement point-process-fault-measurement." Based on the anomaly level labels of the single-point model, we mapped the abnormal measurement points to knowledge graph entities. Using causal chain weights, upstream and downstream impact paths, and mechanism residual information, we calculated the fault path score and used a reinforcement learning ranking model to output the top-N root cause sequence, forming a prioritized list of issues to be addressed. The specific steps are as follows: Collect historical data related to equipment, measurement points, processes, faults, and measures, including basic equipment information, real-time data from measurement points, process flow and parameters, historical records of faults, and corresponding control measures. Create nodes and relationships in the embedded Neo4j graph database to construct a five-element graph of "equipment-measurement-process-fault-measurement". Determine the causal relationship between measurement points, processes, and faults based on historical data, and use the hierarchical analysis method to evaluate the importance of the causal relationship to obtain the weight of each causal chain. Based on the physical principles of the system and process knowledge, establish the mechanism equation between measurement points and process parameters. ,in represents the theoretical value of the process parameter, Represents the measured value of the measuring point, the actual measured value of the process parameter Compared with the theoretical calculated value Perform difference calculation to obtain residuals, distribute the residuals to each measuring point, and obtain the residual contribution value of each measuring point; It should be noted that the corresponding control measures include the parameter adjustment range and the adjustment parameters involved. Equipment, measuring points, processes, faults and measures are respectively regarded as different types of nodes, and the association relationships between them are regarded as edges. For example, there is a "belonging to" relationship between the measuring point and the equipment, and an "influence" relationship between the measuring point and the process. When the hierarchical analysis method is used to evaluate the importance of causal relationships, the target layer, criterion layer and solution layer are constructed respectively. For each criterion in the criterion layer, the elements of the solution layer are compared pairwise. The 1-9 scaling method is used to determine the proportional scale of their relative importance and the judgment matrix is ​​obtained by pairwise comparison. For each judgment matrix, the sum-product method is used to solve its maximum eigenvalue and its corresponding eigenvector. The obtained eigenvector is the weight vector of the solution layer element relative to the criterion. When there are multiple criteria in the criterion layer, the weights of the solution layer elements relative to each criterion need to be summarized and the combined weight is calculated.

[0029] Match the identified abnormal measurement points with the measurement point nodes in the knowledge graph, map the abnormal information to the knowledge graph, use breadth-first search to search all possible paths from the abnormal measurement point to the fault node in the knowledge graph, and calculate the score of each fault path by combining the causal chain weight, upstream and downstream impact paths and mechanism residual information. : ; in, Represents the fault path contains nodes and edges, Representative Nodes to The causal chain weight of each node, represent The residual contribution value of each node is calculated, and an offline Q-learning architecture reinforcement learning ranking model is used to output the Top-N root cause sequence to form a priority list of problems to be handled.

[0030] It should be noted that the specific steps of using the offline Q-learning architecture reinforcement learning ranking model to output the Top-N root cause sequence are as follows: the score of the fault path, the causal chain weight, and the upstream and downstream impact path are used as the state of the reinforcement learning model, different fault paths are used as the actions of the reinforcement learning model, and the score of the fault path is used as the reward value. The higher the score, the greater the reward. The offline Q-learning reinforcement learning algorithm is used to ensure that the action selected in each state can obtain the maximum cumulative reward. According to the strategy learned by the reinforcement learning model, all fault paths are sorted, and the top N fault paths with the highest scores are selected as the root cause sequence to form a priority list of problems to be handled.

[0031] Example 4: The intelligent reasoning module outputs adjustment results based on the priority list of issues to be addressed, combined with the current equipment site conditions and operation goals, including the following steps: Each root cause in the root cause list and the fault context information of the historical case are converted into a semantic vector. The text information is encoded into a vector using the BERT model. The similarity between the root cause list and the historical case is calculated using cosine similarity. The semantic vector of the root cause is , the semantic vector of the fault context of the historical case is , and the cosine similarity is: According to the similarity threshold, historical cases with root cause similarity greater than 0.05 in the root cause list are combined to form a set of similar fault scenario cases. The corresponding historical disposal paths are extracted from the set of similar fault scenario cases. The on-site working condition information of the current equipment is recorded as a state vector. The operation goals of the equipment are clarified, including safety goals, energy consumption goals, and response efficiency goals. The historical disposal paths are used as reinforcement learning actions. The reward function is used to evaluate the pros and cons of each action. The optimal strategy is obtained through Q-learning to output the adjustment results.

[0032] It should be noted that when obtaining the optimal strategy through Q-learning to output the adjustment results, it is also possible to define a multi-objective function F(a)=(S(a), E(a), T(a)). The goal is to optimize safety, energy consumption and response efficiency while satisfying certain constraints. The multi-objective optimization algorithm NSGA-II is used to solve the Pareto optimal solution set, that is, through iterative search, a set of non-dominated solutions is found in the solution space. According to the reinforcement learning strategy and the results of multi-objective optimization, multiple sets of control schemes are generated. Each set of schemes corresponds to a historical disposal path. By controlling the weight ratio of S(a), E(a), and T(a), the comprehensive score of each set of control schemes is calculated according to the reward function, and the control scheme with the highest comprehensive score is selected as the recommended output of the optimal control instruction.

[0033] The execution control module sends control instructions to the DCS, PLC, or intelligent actuator through standard industrial protocols based on the output adjustment results, adjusts and controls the abnormal equipment, and completes the closed-loop response to the abnormal state. The steps include: For measurement points with severe abnormality levels, control instructions are sent to the DCS, PLC, or intelligent actuator based on the upstream and downstream indicators of the measurement point. Safety interlocks are performed on the measurement point and its upstream and downstream measurement points, and the equipment is stopped and started. Based on the adjustment results output by the intelligent reasoning module, the abnormal equipment is adjusted and controlled according to the control instructions in the adjustment results. At the same time, the status feedback data of the measuring points are collected in real time and transmitted to the edge analysis module and the intelligent reasoning module.

[0034] It should be noted that after the control module executes the control instruction, the system collects device status feedback in real time to determine whether the execution result meets expectations, and the result will be automatically written back to the edge analysis module and the intelligent reasoning module to update the operation boundary, residual threshold and reinforcement model parameters to form a closed-loop self-learning path. The diagnostic data, root cause reasoning, strategy execution and response feedback of the entire process are further structured and archived to facilitate subsequent model training sample extraction, event backtracking and rule base evolution. The system returns to the next monitoring cycle to continue running, realizing a complete closed loop of "execution → feedback → correction → evolution". The intelligent actuator intercommunication interface supports point-to-point command issuance and status reading with various electric valves, frequency converters, pump start-stop units, burner control modules, etc.

[0035] Example 5: An intelligent inspection fault diagnosis method using six-dimensional judgment, comprising the following steps: S1. Collect information from field sensors deployed in equipment at different sites during oilfield gathering and transportation, perform protocol parsing, time alignment, and semantic mapping on the collected data, encapsulate the data collected by single-point sensors during oilfield gathering and transportation into a unified measurement point object (UPO), and construct a standard data structure containing device ID, design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters, and upstream and downstream pointers. S2. Dynamically construct a single-point model for each UPO object, forming a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table. Perform a six-dimensional judgment score calculation on the constructed single-point model, output a six-dimensional anomaly score vector, and perform Bayesian posterior update and logical voting fusion on the six-dimensional score vector to output an anomaly level label. S3. Based on the abnormality level label of the single-point model, the abnormal measurement point is mapped to the knowledge graph entity. Through the causal chain weight, upstream and downstream impact path and mechanism residual information, the fault path score is calculated and the reinforcement learning sorting model is used to output the Top-N root cause sequence to form a priority list of problems to be handled. Combined with the current equipment site conditions and operating goals, the adjustment results are output. Based on the output adjustment results, the control instructions are sent to the DCS, PLC or intelligent actuator through the standard industrial protocol to adjust and control the abnormal equipment, completing the response closed loop to the abnormal state.

[0036] In the embodiments provided herein, it should be understood that the disclosed devices, apparatuses, and methods may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the module division is merely a logical functional division, and actual implementation may employ other division methods. Modules described as separate components may or may not be physically separate, and components displayed as modules may or may not be physical units, i.e., they may be located in one location or distributed across multiple network elements. Some or all of the modules may be selected to achieve the objectives of the methods of this embodiment based on actual needs.

[0037] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent inspection fault diagnosis system using six-dimensional judgment, characterized by: Including data access module, edge analysis module, intelligent reasoning module, and execution control model: The data access module collects information from field sensors deployed in equipment at different sites during the oilfield gathering and transportation process, performs protocol parsing, time alignment, and semantic mapping on the collected data, encapsulates the data collected by single-point sensors during the oilfield gathering and transportation process into a unified measurement point object (UPO), and constructs a standard data structure containing device ID, design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters, and upstream and downstream pointers. The edge analysis module dynamically constructs a single-point model for each UPO object, forms a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table, performs a six-dimensional judgment score calculation on the constructed single-point model, outputs a six-dimensional anomaly score vector, and performs Bayesian posterior update and logical voting fusion on the six-dimensional score vector to output an anomaly level label; The intelligent reasoning module maps abnormal measurement points to knowledge graph entities based on the abnormal level labels of the single-point model. It calculates the fault path score based on the causal chain weights, upstream and downstream impact paths, and mechanism residual information, and uses a reinforcement learning ranking model to output the top-N root cause sequence to form a priority list of issues to be addressed. It then outputs the adjustment results based on the current equipment site conditions and operating objectives. The execution control module sends control instructions to the DCS, PLC or intelligent actuator through the standard industrial protocol based on the output adjustment result, adjusts and controls the abnormal equipment, and completes the closed loop response to the abnormal state.

2. The intelligent inspection fault diagnosis system using six-dimensional judgment according to claim 1 is characterized in that: The data access module collects information from field sensors deployed in equipment at different sites during oilfield gathering and transportation, and performs protocol parsing, time alignment, and semantic mapping on the collected data. Specifically, the following steps are included: The sensors deployed in different sites during the oilfield gathering and transportation process collect flow, pressure, temperature, vibration, and electrical quantity data, perform protocol parsing and matching on the key features of the initial data stream collected by the sensors, and calculate the protocol matching score. ,in represents the weight of the h-th feature, Represents the matching degree of the hth feature. When it matches, the value is 1 and when it does not match, the value is 0. When it is used, it means it belongs to the corresponding agreement; Add a timestamp to the data collected by each sensor, align the sensor data in time through interpolation, and perform semantic mapping based on the type of sensor and the meaning of the collected data.

3. The intelligent inspection fault diagnosis system using six-dimensional judgment according to claim 2 is characterized in that: The method encapsulates the data collected by single-point sensors in the oilfield gathering and transportation process into a unified measurement point object UPO, and constructs a standard data structure including design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters and upstream and downstream pointers, specifically including the following steps: Define the structure of the unified measurement point object UPO, including device ID, design attributes, real-time value, historical sequence, alarm threshold, mechanism parameters and upstream and downstream pointers; Encapsulate the data after protocol parsing, time alignment and semantic mapping into UPO to build a standard data structure ,in It is a combination code of device number and measuring point sensor type. Represents the real-time value collected, Store the historical sequence formed by the latest N real-time values ​​for the time window, Represents the rated value in the design properties of the current measurement point object. are the high-alarm threshold and the low-alarm threshold, It is the ID of the upstream and downstream pointers, i.e. the predecessor node and the successor node. Represents the parameters of the mechanism equation.

4. The intelligent inspection fault diagnosis system using six-dimensional judgment according to claim 3 is characterized in that: The edge analysis module dynamically constructs a single-point model for each UPO object, forming a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table. Specifically, the following steps are included: Based on the unified measurement point object UPO encapsulated by the data access module, a single-point model is dynamically constructed for the standard data structure, and the static parameters in the UPO are constructed as static attribute sub-tables, including design attributes and alarm thresholds; By using the real-time values ​​and historical sequence data in the UPO object, a dynamic state table is constructed, and by calculating the statistical characteristics of the data, the trend slope of the measurement point is calculated in real time using the linear regression equation; According to the upstream and downstream association pointers in the UPO object, an upstream and downstream linked list is constructed so that each node in the upstream and downstream linked list corresponds to a measuring point, and the upstream and downstream relationships are represented by pointer connections, that is, other measuring points associated with the current measuring point; The mechanism parameters are extracted from the UPO object to construct the mechanism parameter sub-table, and the static attribute sub-table, dynamic state sub-table, mechanism parameter table and upstream and downstream link tables are integrated to form a decidable multi-dimensional model structure.

5. The intelligent inspection fault diagnosis system using six-dimensional judgment according to claim 4 is characterized in that: The six-dimensional judgment score calculation is performed on the constructed single-point model to output a six-dimensional anomaly score vector, and the six-dimensional score vector is fused with Bayesian posterior update and logical voting to output an anomaly level label, including the following steps: Based on the single-point models constructed at different measuring points, six-dimensional judgment scores are calculated for different measuring point models and a six-dimensional abnormality score vector is output. The six dimensions include design boundaries, operating boundaries, upstream and downstream coupling, mechanism residuals, historical evolution, high and low reporting thresholds, to form a normalized six-dimensional score vector. ,in , specifically including the following steps: Design boundary scoring , determine the normal range of design parameters through the static property table , based on the real-time value of the current measuring point , calculate the design boundary deviation : ； Deviation from the design boundary Map to Get design boundary score within the interval ; Determine the operation boundary by the statistical characteristics of historical data in the dynamic state table, based on the mean value in the historical data and standard deviation Get the operating range ,in is a constant to calculate the operating boundary deviation , get the running boundary score ; According to the current measuring point and the upstream and downstream related measuring points, calculate the Pearson correlation coefficient between the real-time value of the current measuring point and the real-time value of the upstream and downstream related measuring points ,in is the current measuring point, For other measurement points in the upstream and downstream associated measurement point set, based on the normal correlation range , calculate the upstream and downstream coupling deviation : ； in, The average value of the real-time value of the current measuring point and the correlation coefficient of the upstream and downstream measuring points is used to obtain the upstream and downstream coupling score. , according to the parameters in the mechanism equation parameter subtable, use the mechanism equation to calculate the theoretical value of the measuring point And calculate the residual , according to the upper limit of the normal range of residuals , and obtain the mechanism residual score ,when hour, ; The trend slope obtained from the historical data series , according to the normal trend slope range , the calculation can obtain the historical evolution deviation , get the historical evolution score : ； Based on high and low alarm thresholds, when If it falls within the high and low reporting threshold range, the high and low reporting threshold score ,when If it does not fall within the high or low reporting threshold range, the high or low reporting threshold score ; Based on the abnormality level normal, attention, abnormal, severe, respectively , according to historical data statistics, the prior probability of each abnormal level is obtained ,and , for each abnormal level , respectively calculate the six-dimensional scoring vectors under different abnormal levels The likelihood probability , calculate the posterior probability according to the Bayesian formula ,in Represents the observed six-dimensional rating vector After that, the current measurement point belongs to the abnormal level At the same time, when the score of any of the four dimensions is greater than 0.7, a response is triggered synchronously, the abnormality level is judged to be severe, and the abnormality level label is output to enter the subsequent module.

6. The intelligent inspection fault diagnosis system using six-dimensional judgment according to claim 1 is characterized in that: The intelligent reasoning module comprises the following steps: By deploying an embedded Neo4j graph database, we constructed a five-element graph of "equipment-measurement point-process-fault-measurement." Based on the anomaly level labels of the single-point model, we mapped the abnormal measurement points to knowledge graph entities. Using causal chain weights, upstream and downstream impact paths, and mechanism residual information, we calculated the fault path score and used a reinforcement learning ranking model to output the top-N root cause sequence, forming a prioritized list of issues to be addressed. The specific steps are as follows: Collect historical data related to equipment, measurement points, processes, faults, and measures, including basic equipment information, real-time data from measurement points, process flow and parameters, historical records of faults, and corresponding control measures. Create nodes and relationships in the embedded Neo4j graph database to construct a five-element graph of "equipment-measurement-process-fault-measurement". Determine the causal relationship between measurement points, processes, and faults based on historical data, and use the analytic hierarchy process to evaluate the importance of the causal relationship to obtain the weight of each causal chain. Based on the physical principles of the system and process knowledge, establish the mechanism equation between measurement points and process parameters. ,in represents the theoretical value of the process parameter, Represents the measured value of the measuring point, the actual measured value of the process parameter Compared with the theoretical calculated value Perform difference calculation to obtain residuals, distribute the residuals to each measuring point, and obtain the residual contribution value of each measuring point; Match the identified abnormal measurement points with the measurement point nodes in the knowledge graph, map the abnormal information to the knowledge graph, use breadth-first search to search all possible paths from the abnormal measurement point to the fault node in the knowledge graph, and calculate the score of each fault path by combining the causal chain weight, upstream and downstream impact paths and mechanism residual information. : ； in, Represents the fault path contains nodes and edges, Representative Nodes to The causal chain weight of each node, represent The residual contribution value of each node is calculated, and an offline Q-learning architecture reinforcement learning ranking model is used to output the Top-N root cause sequence to form a priority list of problems to be handled.

7. The intelligent inspection fault diagnosis system using six-dimensional judgment according to claim 6, characterized in that: The intelligent reasoning module outputs adjustment results based on the priority list of issues to be addressed, combined with the current equipment site conditions and operation goals, including the following steps: Each root cause in the root cause list and the fault context information of the historical case are converted into a semantic vector. The text information is encoded into a vector using the BERT model. The cosine similarity is used to calculate the similarity between the root cause list and the historical case. The semantic vector of , historical cases The fault context semantic vector is , cosine similarity for: ； Based on the similarity threshold The root cause similarity in the root cause list is greater than The historical cases of similar fault scenarios are combined to form a case set of similar fault scenarios. The corresponding historical disposal paths are extracted from the case set of similar fault scenarios. The on-site working condition information of the current equipment is recorded as a state vector. The operation goals of the equipment are clarified, including safety goals, energy consumption goals, and response efficiency goals. The historical disposal paths are used as reinforcement learning actions. The reward function is used to evaluate the pros and cons of each action. The optimal strategy is obtained through Q-learning to output the adjustment results.

8. The intelligent inspection fault diagnosis system using six-dimensional judgment according to claim 1 is characterized in that: The execution control module sends control instructions to the DCS, PLC or intelligent actuator through the standard industrial protocol based on the output adjustment results, adjusts and controls the abnormal equipment, and completes the closed loop response to the abnormal state, including the following steps: For measurement points with severe abnormality levels, control instructions are sent to the DCS, PLC, or intelligent actuator based on the upstream and downstream indicators of the measurement point. Safety interlocks are performed on the measurement point and its upstream and downstream measurement points, and the equipment is stopped and started. Based on the adjustment results output by the intelligent reasoning module, the abnormal equipment is adjusted and controlled according to the control instructions in the adjustment results. At the same time, the status feedback data of the measuring points are collected in real time and transmitted to the edge analysis module and the intelligent reasoning module.

9. An intelligent inspection fault diagnosis method using six-dimensional judgment, characterized in that: The method adopts an intelligent inspection fault diagnosis system using six-dimensional judgment as described in any one of claims 1 to 8, comprising the following steps: S1. Collect information from field sensors deployed in equipment at different sites during oilfield gathering and transportation, perform protocol parsing, time alignment, and semantic mapping on the collected data, encapsulate the data collected by single-point sensors during oilfield gathering and transportation into a unified measurement point object (UPO), and construct a standard data structure containing device ID, design attributes, real-time values, historical sequences, alarm thresholds, mechanism parameters, and upstream and downstream pointers. S2. Dynamically construct a single-point model for each UPO object, forming a static attribute sub-table, a dynamic state sub-table, a mechanism parameter table, and an upstream and downstream link table. Perform a six-dimensional judgment score calculation on the constructed single-point model, output a six-dimensional anomaly score vector, and perform Bayesian posterior update and logical voting fusion on the six-dimensional score vector to output an anomaly level label. S3. Based on the abnormality level label of the single-point model, the abnormal measurement point is mapped to the knowledge graph entity. Through the causal chain weight, upstream and downstream impact path and mechanism residual information, the fault path score is calculated and the reinforcement learning sorting model is used to output the Top-N root cause sequence to form a priority list of problems to be handled. Combined with the current equipment site conditions and operating goals, the adjustment results are output. Based on the output adjustment results, the control instructions are sent to the DCS, PLC or intelligent actuator through the standard industrial protocol to adjust and control the abnormal equipment, completing the response closed loop to the abnormal state.

Images