Oil extraction equipment fault monitoring system and method

Abstract

The invention provides an oil extraction equipment fault monitoring system and method, and belongs to the technical field of oil extraction equipment fault monitoring. The method comprises the following steps: acquiring operation data of oil extraction equipment, and performing feature extraction on the acquired operation data to obtain a target feature vector; fusing the obtained target feature vector with a historical fault case library and an oil extraction equipment physical constraint equation, and constructing a dynamically updated knowledge graph; based on the space-time causal adversarial network, analyzing the distribution offset of the target feature vector in the space-time dimension, detecting an abnormal event and outputting an abnormal type label; and according to the output abnormity type label, combining with a knowledge graph, tracing a propagation path of an abnormal event, and calculating a fault probability of a root cause component through a Bayesian network so as to carry out monitoring and early warning on the oil extraction equipment. According to the method, accurate fault detection and root cause positioning are realized through multi-modal data fusion and the dynamic causal knowledge graph, and the equipment shutdown risk and the operation and maintenance cost are remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of oil production equipment fault monitoring, and particularly to an oil production equipment fault monitoring system and method. Background Art

[0002] With the expansion of oilfield exploitation scale and the complexity of equipment, oil production equipment (such as pumping units, electrical submersible pumps, water injection units, etc.) operates under high-load and variable working conditions for a long time, and is prone to failures such as mechanical wear, component fatigue, and lubrication failure. Traditional equipment monitoring mainly relies on threshold alarms and regular manual inspections, which have problems such as lagging response and high missed detection rate. In recent years, predictive maintenance technologies based on vibration analysis and temperature monitoring have been gradually applied, and machine learning algorithms (such as support vector machines, random forests) are combined to classify the equipment status, realizing early warning of some faults. However, the characteristics of multi-component coupling and complex fault propagation paths in oil production equipment make it difficult for single-sensor data or static models to accurately capture the spatio-temporal evolution law of potential faults, and more intelligent monitoring means are urgently needed.

[0003] The traditional oil production equipment fault monitoring technologies generally have the following limitations: traditional methods mostly rely on single-type sensor data and static threshold judgments, and it is difficult to capture the complex fault evolution process under the coupling effect of multiple components, resulting in a high missed detection rate of early weak faults; the fault diagnosis model based on historical data lacks the dynamic adaptation ability to the real-time status and physical laws of the equipment, and the false alarm rate increases significantly when the working conditions drift or the equipment ages; at the same time, existing solutions mostly focus on anomaly detection and ignore the causal traceability of the root cause of the fault, and cannot provide accurate component-level positioning guidance for maintenance decisions. Summary of the Invention

[0004] The present invention provides an oil production equipment fault monitoring method to solve the problems of high missed detection rate and insufficient dynamic adaptation ability existing in the prior art.

[0005] To achieve the above object, on the one hand, an embodiment of the present invention provides an oil production equipment fault monitoring method, the oil production equipment fault monitoring method includes: acquiring operation data of the oil production equipment, and performing feature extraction on the acquired operation data to obtain a target feature vector; fusing the obtained target feature vector with a historical fault case library and an oil production equipment physical constraint equation to construct a dynamically updated knowledge graph; based on a spatio-temporal causal adversarial network, analyzing the distribution shift of the target feature vector in the spatio-temporal dimension, detecting abnormal events and outputting abnormal type labels; according to the output abnormal type labels, combining with the knowledge graph, tracing the propagation path of the abnormal events, and calculating the fault probability of the root cause component through a Bayesian network to monitor and give early warning to the oil production equipment.

[0006] Optionally, obtaining the operation data of the oil production equipment and performing feature extraction on the obtained operation data to obtain a target feature vector includes: according to the obtained operation data, using wavelet packet decomposition to obtain the energy entropy of the transient impact component, and extracting an initial fault feature vector; according to the extracted initial feature vector, through principal component analysis, compressing it into a target feature vector.

[0007] Optionally, fusing the obtained target feature vector with a historical fault case library and the physical constraint equations of the oil production equipment to construct a dynamically updated knowledge graph includes: according to the hierarchical relationship and physical constraint equations of each component of the oil production equipment, establishing a static knowledge skeleton of the oil production equipment structure; extracting fault events and propagation paths from the historical fault case library and loading them into the static knowledge skeleton; through a graph embedding algorithm, mapping the target feature vector to the corresponding nodes of the static knowledge skeleton to generate weighted spatio-temporal state attributes, forming the knowledge graph; based on the message passing mechanism of the graph neural network, according to the spatio-temporal similarity between the real-time data of the oil production equipment operation and the historical fault cases, updating the weight coefficients of the fault propagation edges in the knowledge graph; when detecting an unrecorded new fault mode, creating a new node and establishing its minimum causal association set with the existing nodes through transfer learning to complete the closed-loop evolution of the knowledge graph.

[0008] Optionally, the spatio-temporal causal adversarial network includes a generator and a discriminator. Based on the spatio-temporal causal adversarial network, analyzing the distribution shift of the target feature vector in the spatio-temporal dimension, detecting abnormal events and outputting abnormal type labels includes: the generator extracts the spatial pattern of the target feature vector through a 3D convolutional network and combines it with an LSTM network to learn the time evolution law to generate a normal state feature distribution; the discriminator calculates the spatio-temporal distribution difference according to the target feature vector and the generated normal state feature distribution through a multi-head attention mechanism, and combines the physical constraints of the knowledge graph to identify abnormal events and calculate an abnormal confidence score; based on the calculated abnormal confidence score, retrieving the fault mode with the highest similarity from the historical fault feature library and outputting an abnormal type label with an attached probability.

[0009] Optionally, according to the output abnormal type label, combining it with the knowledge graph to trace the propagation path of the abnormal event and calculating the fault probability of the root cause component through a Bayesian network includes: based on the fault propagation edges in the knowledge graph, traversing backward along the starting node corresponding to the abnormal type label to trace the potential propagation path of the abnormal event; based on the Bayesian network and the real-time operation data of the oil production equipment, calculating the fault probability of the root cause component in each propagation path.

[0010] Optionally, after tracing the propagation path of the abnormal event according to the knowledge graph and calculating the failure probability of the root cause component through the Bayesian network, the oil production equipment failure monitoring method further includes: generating a maintenance work order for the component to be inspected according to the failure probability; predicting the remaining service life of the component based on the component degradation trend, and generating an optimal maintenance plan in combination with the maintenance cost; dynamically correcting the node weights of the knowledge graph according to the maintenance feedback data, and triggering incremental training of the spatio-temporal causal adversarial network to achieve closed-loop optimization.

[0011] Optionally, the predicting the remaining service life of the component based on the component degradation trend and generating an optimal maintenance plan in combination with the maintenance cost includes: analyzing the component degradation trend according to the historical failure data and physical wear rules of the oil production equipment, and predicting the remaining service life of each component; calculating the total cost at different maintenance time points according to the predicted remaining life; generating an optimal maintenance plan through a genetic algorithm according to the calculated total cost and the potential failure risk of the oil production equipment.

[0012] On the other hand, the present invention also provides an oil production equipment failure monitoring system for implementing the above-mentioned oil production equipment failure monitoring method. The oil production equipment failure monitoring system includes: a data acquisition module, including a sensor array, for real-time acquisition of the operation data of the oil production equipment; an intelligent analysis module, for analyzing the acquired operation data, predicting the failure probability of each component in the oil production equipment, generating a maintenance work order for the component to be inspected, and formulating an optimal maintenance plan; generating a failure warning message according to the failure probability; a terminal server, configured with a display unit, for displaying the real-time operation data of the oil production equipment and dispatching maintenance work orders.

[0013] Optionally, the intelligent analysis module is further configured to output a failure warning signal according to the failure probability, and the display unit is further configured to display a failure warning message according to the failure warning signal.

[0014] Optionally, the oil production equipment failure monitoring system further includes: a communication module, for realizing data transmission between the data acquisition module, the intelligent analysis module and the terminal server, and receiving remote control instructions.

[0015] The oil production equipment failure monitoring system and method provided by the present invention fuse multi-source sensor data to construct a dynamically updated knowledge graph, combine a spatio-temporal causal adversarial network and a Bayesian network to achieve accurate fault detection and root cause tracing, effectively solving the problems of high missed detection rate, high false alarm rate and fuzzy root cause positioning in traditional methods; at the same time, an optimal maintenance plan is generated based on a multi-objective optimization algorithm of equipment degradation trend and maintenance cost, significantly improving the maintenance efficiency and reducing the operation and maintenance cost. Its closed-loop self-optimization mechanism can dynamically adapt to equipment aging and working conditions changes, providing an efficient and reliable solution for intelligent operation and maintenance of oilfield equipment. Brief Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings: Figure 1 is a flowchart of a method for monitoring faults in oil production equipment provided by an embodiment of the present invention; Figure 2 is another flowchart of a method for monitoring faults in oil production equipment provided by an embodiment of the present invention; Figure 3 is a flowchart of the construction and update of the knowledge graph provided by an embodiment of the present invention; Figure 4 is a flowchart of anomaly detection and root cause location provided by an embodiment of the present invention; Figure 5 is a flowchart of maintenance decision-making and closed-loop optimization provided by an embodiment of the present invention; Figure 6 is a structural diagram of a system for monitoring faults in oil production equipment provided by an embodiment of the present invention. Detailed Embodiments

[0017] The following will describe in detail the specific embodiments of the embodiments of the present invention with reference to the drawings. It should be understood that the specific embodiments described here are only used to illustrate and explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0018] It should be noted that the acquisition, transmission, storage, use, processing, etc. of data in the technical solution of this application all comply with the relevant regulations of national laws and regulations. In the embodiments of this application, some industry-existing solutions such as certain software, components, models, etc. may be mentioned. They should be regarded as exemplary, and their purpose is only to illustrate the feasibility in the implementation of the technical solution of this application, but it does not mean that the applicant has already or necessarily used this solution.

[0019] With the expansion of the scale of oilfield exploitation and the improvement of equipment complexity, oil production equipment operates under harsh working conditions of high load and strong corrosion for a long time, and faults such as mechanical wear and component fatigue occur frequently. Traditional monitoring means rely on manual inspections and static threshold alarms, and there are problems such as response lag and high missed detection rate. Therefore, it is particularly important to develop a more accurate and effective system and method for monitoring faults in oil production equipment.

[0020] To address this problem, the present invention provides an oil production equipment fault monitoring system and method. The present invention accurately locates the root cause of faults through multi-modal data fusion and dynamic causal knowledge graphs, continuously improves the monitoring accuracy by combining a closed-loop self-optimization mechanism, and generates an optimal maintenance strategy based on the equipment degradation state and cost model, significantly reducing the missed detection rate, false alarm rate, and operation and maintenance costs, and realizing intelligent and efficient operation and maintenance of oil production equipment.

[0021] The following will Figures 1 - 6 describe the present invention in detail.

[0022] As Figure 1 shown, an embodiment of the present invention provides an oil production equipment fault monitoring method, and the oil production equipment fault monitoring method includes the following steps: S101: Obtain the operation data of the oil production equipment, and perform feature extraction on the obtained operation data to obtain a target feature vector; S102: Integrate the obtained target feature vector with the historical fault case library and the physical constraint equations of the oil production equipment to construct a dynamically updated knowledge graph; S103: Based on the spatio-temporal causal adversarial network, analyze the distribution shift of the target feature vector in the spatio-temporal dimension, detect abnormal events, and output abnormal type labels; S104: According to the output abnormal type labels, combine with the knowledge graph, trace the propagation path of the abnormal events, and calculate the fault probability of the root cause components through a Bayesian network to monitor and warn the oil production equipment.

[0023] As Figure 2 shown, the oil production equipment fault monitoring method provided by the present invention first collects the vibration, temperature, pressure, and flow data of the oil production equipment in real time through multi-source sensors, performs feature extraction on the original data, and generates a low-dimensional feature vector after eliminating noise interference; then integrates the feature vector with the equipment physical constraint equations (such as mechanical balance, heat conduction law) and the historical fault case library to construct a dynamic knowledge graph reflecting the association relationship between equipment components and the fault propagation logic; then analyzes the distribution shift of the feature vector in the spatio-temporal dimension based on the spatio-temporal causal adversarial network, detects abnormal events, and outputs classification labels; finally, traces the abnormal propagation path in reverse according to the knowledge graph, combines with the Bayesian network to calculate the fault probability of the root cause components, and realizes accurate fault location and warning. This method solves the problems of one-sidedness of single-sensor data and poor adaptability of static models in traditional methods through multi-source data fusion and dynamic knowledge graph modeling; the spatio-temporal causal adversarial network combined with physical constraints significantly improves the sensitivity of abnormal detection, and the Bayesian root cause tracing mechanism ensures the accuracy of location. The overall solution can reduce the missed detection rate and false alarm rate, providing technical support for the efficient operation and maintenance of oil production equipment.

[0024] Preferably, in step S101, the obtaining of the operation data of the oil production equipment and the feature extraction of the obtained operation data to obtain the target feature vector include: according to the obtained operation data, using wavelet packet decomposition to obtain the energy entropy of the transient shock component, and extracting the initial fault feature vector; according to the extracted initial feature vector, through principal component analysis, compressing it into the target feature vector.

[0025] In a preferred embodiment of the present invention, wavelet packet decomposition is a time-frequency analysis method that is more refined than wavelet decomposition. It can decompose a signal at multiple levels in different scales and frequencies, and decompose the signal into sub-signals in different frequency bands. For the operation data of oil production equipment, when the equipment fails, transient shock signals are often generated, and these signals contain important information about the fault. Energy entropy reflects the energy distribution of the signal in different frequency bands. In the fault detection of oil production equipment, different fault types may cause changes in the energy distribution of the operation data in different frequency bands. Calculating the energy entropy of the transient shock component can quantify this energy distribution feature, thereby extracting the initial fault feature vector. In the process of feature extraction, first use wavelet packet decomposition to accurately capture the transient shock component from the operation data, and obtain the initial fault feature vector by calculating its energy entropy. This method can sensitively capture the subtle changes in the energy distribution at the initial stage of equipment failure, providing key clues for fault judgment. Then, use principal component analysis to compress the initial feature vector into the target feature vector, while maximizing the retention of key information, greatly reducing the data dimension, effectively reducing the computational complexity of subsequent analysis, and avoiding problems such as overfitting caused by high data dimension, thus laying a solid and reliable data foundation for the entire oil production equipment fault monitoring process.

[0026] As Figure 3 shown, preferably, in step S102, the fusing of the obtained target feature vector with the historical fault case library and the physical constraint equation of the oil production equipment to construct a dynamically updated knowledge graph includes: establishing a static knowledge framework of the oil production equipment structure according to the hierarchical relationship and physical constraint equation of each component of the oil production equipment; extracting the fault events and propagation paths from the historical fault case library and loading them into the static knowledge framework; through the graph embedding algorithm, mapping the target feature vector to the corresponding nodes of the static knowledge framework to generate weighted spatio-temporal state attributes, forming the knowledge graph; based on the message passing mechanism of the graph neural network, updating the weight coefficient of the fault propagation edge in the knowledge graph according to the spatio-temporal similarity between the real-time data of the oil production equipment operation and the historical fault cases; when detecting an unrecorded new fault mode, creating a new node and establishing its minimum causal association set with the existing nodes through transfer learning to complete the closed-loop evolution of the knowledge graph.

[0027] In a preferred embodiment of the present invention, when constructing the knowledge graph, first, a static knowledge skeleton is built based on the hierarchical relationship of oil production equipment components and physical constraint equations, which provides a stable structural foundation for the knowledge graph and ensures that it can reflect the physical essence of the equipment. Then, fault events and propagation paths are loaded from the historical fault case library, enabling the knowledge graph to incorporate actual fault information and be more practical. The target feature vectors are mapped to the corresponding nodes using a graph embedding algorithm, and spatio-temporal state attributes with weights are generated, realizing the organic combination of real-time operation data and the knowledge graph and dynamically reflecting the operating state of the equipment. In terms of the update mechanism, based on the message passing mechanism of graph neural networks, the weight coefficients of fault propagation edges are updated by combining the spatio-temporal similarity of real-time data and historical cases, allowing the knowledge graph to be dynamically adjusted according to the equipment operating conditions and improving the accuracy of fault analysis. When encountering a new fault mode, a new node is created and a minimum causal association set is established using transfer learning to achieve the closed-loop evolution of the knowledge graph, ensuring that the knowledge graph can continuously adapt to new situations and improve its own knowledge system. Generally speaking, this construction and update process comprehensively applies multidisciplinary methods and integrates various data, making the knowledge graph both stable and flexible, and capable of providing comprehensive and dynamic knowledge support for the fault monitoring and diagnosis of oil production equipment.

[0028] For example, taking the gearbox of an oilfield pumping unit as an example, the specific implementation process of constructing a dynamic knowledge graph is as follows: First, based on the component hierarchical relationship of the gearbox (such as the assembly topology of gears, bearings, shafts, and seals) and physical constraint equations (such as gear meshing dynamics equations and bearing wear models), a static knowledge skeleton containing node attributes (material hardness, design life) and edge relationships (force transmission, heat conduction) is established; Subsequently, typical fault cases (such as gear fracture caused by bearing wear and lubricating oil leakage caused by seal failure) are extracted from the historical fault library, and they are transformed into (fault component, fault type, affected component) triples and loaded into the skeleton to form an initial knowledge graph; After real-time acquisition of the vibration and temperature signals of the gearbox, the frequency band energy entropy features are extracted through wavelet packet decomposition, and low-dimensional feature vectors are generated through PCA dimensionality reduction. The graph embedding algorithm is used to map the vibration feature vectors to the bearing nodes and the temperature features to the gear nodes, and dynamic weights are assigned to the nodes (such as bearing vibration energy entropy = 0.85, gear temperature gradient = 0.72); When it is detected that the current vibration energy entropy (for example, 0.91) exceeds the historical case threshold, the graph neural network updates the fault propagation edge weights through the message passing mechanism (such as the fault propagation weight from bearing to gear increases from 0.6 to 0.8); If a new fault mode not recorded is found (such as high-frequency noise caused by seal aging), a new node is created and associated with the lubricating oil temperature rise node through transfer learning to form the minimum causal chain of "seal aging → lubricating oil temperature rise → gear wear", completing the dynamic expansion of the graph. The preferred embodiment of the present invention enables the knowledge graph to continuously adapt to equipment degradation and working condition changes through a static skeleton + dynamic attribute update + incremental expansion mechanism, greatly shortening the diagnostic response time for new fault modes and improving the positioning accuracy rate.

[0029] As Figure 4 shown, preferably, the spatio-temporal causal adversarial network includes a generator and a discriminator. Based on the spatio-temporal causal adversarial network, the distribution shift of the target feature vector in the spatio-temporal dimension is analyzed to detect abnormal events and output abnormal type labels, including: The generator extracts the spatial pattern of the target feature vector through a 3D convolutional network and combines with an LSTM network to learn the time evolution law to generate a normal state feature distribution; The discriminator calculates the spatio-temporal distribution difference according to the target feature vector and the generated normal state feature distribution through a multi-head attention mechanism, and combines with the physical constraints of the knowledge graph to identify abnormal events and calculate an abnormal confidence score; Based on the calculated abnormal confidence score, the fault mode with the highest similarity is retrieved from the historical fault feature library, and an abnormal type label with an attached probability is output.

[0030] In a preferred embodiment of the present invention, the spatio-temporal causal adversarial network realizes anomaly detection through the collaborative architecture of a generator and a discriminator: the generator uses a 3D convolutional network to capture the spatial coupling features of multi-sensor data (such as the spatial distribution pattern of vibration signals on bearing seats and gearboxes), combines with an LSTM network to learn the temporal evolution law of the equipment state (such as the cumulative trend of temperature gradient over operating time), and reconstructs the feature distribution under normal operating conditions; the discriminator calculates the distribution difference between the real-time feature vector and the output of the generator through a multi-head attention mechanism, and at the same time introduces physical constraints in the knowledge graph (such as "when the bearing is abnormal, the vibration energy entropy and temperature should rise synchronously") as a regularization term to suppress false alarm signals that do not conform to the equipment mechanism, and finally outputs an anomaly type label with an attached probability. Its advantage lies in the dual verification of spatio-temporal feature joint modeling and physical rules, effectively improving the detection sensitivity to weak early faults (such as bearing pitting) and reducing the false alarm rate.

[0031] For example, taking the overheating fault monitoring of the stator winding of an oilfield ESP motor as an example, the implementation process of the spatio-temporal causal adversarial network is as follows: the generator receives three-dimensional spatio-temporal data of vibration, temperature, and current (the spatial dimension is the sensor layout position, and the temporal dimension is a continuous 10-minute window), extracts the vibration spatial pattern caused by winding looseness through a 3D convolutional layer, and combines with LSTM to predict the normal temperature rise curve; the discriminator performs a multi-head attention comparison on the real-time data and the normal distribution output by the generator, and finds that the temperature of a certain winding node deviates from the predicted value by 2.5°C and the vibration energy entropy is abnormal. At the same time, it verifies the physical rule of "winding looseness → vibration enhancement → temperature rise acceleration" in the knowledge graph. After confirming that it conforms to the fault propagation logic, it outputs the label "stator winding looseness" (probability 87%).

[0032] Preferably, according to the output anomaly type label, in combination with the knowledge graph, tracing the propagation path of the abnormal event, and calculating the fault probability of the root cause component through a Bayesian network, including: based on the fault propagation edges in the knowledge graph, performing a reverse traversal along the starting node corresponding to the anomaly type label to trace the potential propagation path of the abnormal event; based on the Bayesian network and the real-time operation data of the oil production equipment, calculating the fault probability of the root cause component in each propagation path.

[0033] In a preferred embodiment of the present invention, the knowledge graph stores the relationships between various components of the oil production equipment and the historical information of fault propagation. The root cause location process is divided into two technical stages: 1. Reverse traversal of the fault propagation path: Based on the predefined fault propagation edges in the dynamic knowledge graph (such as "bearing wear → increased gear vibration → increased oil temperature"), starting from the starting node corresponding to the abnormal type label (such as the "abnormal oil temperature" node), traverse all possible upstream nodes along the reverse path (abnormal oil temperature ← abnormal gear vibration ← bearing wear); The path search adopts a depth-first search combined with a causal constraint pruning strategy, only retaining the propagation chains that conform to the physical equations, and eliminating invalid paths (such as skipping the seal failure path that has nothing to do with the oil temperature) to ensure the engineering rationality of the candidate path set. 2. Bayesian network probability calculation: Construct a Bayesian network topology, where the nodes correspond to the states of the equipment components, and the edge weights are jointly determined by historical fault statistics and equipment physical constraints (such as the initial conditional probability of bearing wear to gear damage = 0.7), and the real-time operation data dynamically updates the fault probabilities of the nodes. By continuously collecting and analyzing fault data, the model of the Bayesian network can be continuously optimized, so that the calculation of the fault probability is more accurate, further improving the performance of the entire oil production equipment fault monitoring system.

[0034] As Figure 5 shown, preferably, after tracing the propagation path of the abnormal event according to the knowledge graph and calculating the fault probability of the root cause component through the Bayesian network, the oil production equipment fault monitoring method further includes: generating a maintenance work order for the component to be inspected according to the fault probability; predicting the remaining service life of the component based on the component degradation trend, and generating an optimal maintenance plan in combination with the maintenance cost; dynamically correcting the node weights of the knowledge graph according to the maintenance feedback data, and triggering the incremental training of the spatio-temporal causal adversarial network to achieve closed-loop optimization.

[0035] Further preferably, the predicting the remaining service life of the component based on the component degradation trend and generating an optimal maintenance plan in combination with the maintenance cost includes: analyzing the component degradation trend according to the historical fault data and physical wear law of the oil production equipment, and predicting the remaining service life of each component; calculating the total cost at different maintenance time points according to the predicted remaining life; generating an optimal maintenance plan through a genetic algorithm according to the calculated total cost and the potential fault risk of the oil production equipment.

[0036] In a preferred embodiment of the present invention, after root cause localization, the system predicts the remaining service life based on the component degradation trend, constructs a multi-objective optimization problem by combining a maintenance cost model (spare part cost, man-hour, downtime loss, and failure risk penalty term), and uses a genetic algorithm to solve for the Pareto optimal maintenance time window to generate an optimal maintenance plan that takes into account both economy and reliability. At the same time, the system dynamically corrects the node weights of the knowledge graph through maintenance feedback data and triggers incremental training of the spatio-temporal causal adversarial network to achieve system self-evolution. Its collaborative advantage lies in the deep integration of physics-data fusion life prediction and multi-objective dynamic optimization, which improves the economy of the maintenance plan by 35%, increases the work order execution rate from 70% to 95%, and reduces the unplanned downtime by 30%.

[0037] For example, taking the wear failure of the motor bearing of an oilfield electric submersible pump as an example, the process of generating maintenance decisions and implementing closed-loop optimization is as follows: It is detected that the vibration energy entropy of the bearing exceeds the standard (abnormal confidence level of 92%), and the root cause is located as bearing wear (probability of 85%). Based on historical wear data and the Archard wear model, the remaining life of the bearing is predicted to be T RUL =T0±ΔT (where T0 is the reference life and ΔT is the prediction error); a maintenance cost model is established, including the bearing replacement cost C b 、labor cost C l 、downtime loss C d and the failure risk penalty term C r ×P f (P f is the probability of motor burnout caused by delayed maintenance, and C r is the estimated loss). Calculate the total cost at different maintenance time points: The total cost of immediate maintenance = C b +C l +C d ×T1 (T1 is the downtime), and the failure risk is 0%; the total cost of delayed maintenance = C b +C l +C d ×T1+C r ×P f , and the failure risk is P f ; through iterative optimization of the genetic algorithm, the Pareto optimal solution is selected as "maintenance within T1 hours", with a relatively low predicted cost and a risk approaching zero; after maintenance execution, the actual bearing wear degree is feedback to be consistent with the prediction, the edge weight of "bearing wear → vibration anomaly" in the knowledge graph is dynamically corrected from w1 to w2, and the wear coefficient of the RUL prediction model is adjusted, so that the subsequent prediction error is reduced from ΔT1 to ΔT2. At the same time, incremental training of the spatio-temporal causal adversarial network is triggered, the physical constraint threshold of the discriminator is updated, and the detection response time of new failure modes is shortened. The preferred embodiment of the present invention reduces the maintenance cost through multi-objective optimization and a closed-loop self-correction mechanism.

[0038] As Figure 6 shown, an oil production equipment fault monitoring system provided by an embodiment of the present invention is used to implement the above-mentioned oil production equipment fault monitoring method. The oil production equipment fault monitoring system includes: a data acquisition module, including a sensor array, for real-time acquisition of the operation data of the oil production equipment; an intelligent analysis module, for analyzing the acquired operation data, predicting the fault probability of each component in the oil production equipment, generating a maintenance work order for the components to be inspected, and formulating an optimal maintenance plan; generating a fault warning message according to the fault probability; a terminal server, configured with a display unit, for displaying the real-time operation data of the oil production equipment and dispatching maintenance work orders.

[0039] Preferably, the intelligent analysis module is further used to output a fault warning signal according to the fault probability, and the display unit is further used to display a fault warning message according to the fault warning signal.

[0040] More preferably, the oil production equipment fault monitoring system further includes: a communication module, for realizing data transmission between the data acquisition module, the intelligent analysis module and the terminal server, and receiving remote control instructions.

[0041] The oil production equipment fault monitoring system provided by the embodiment of the present invention uses a multi-modal sensor array (vibration, temperature, pressure) to collect equipment operation data in real time, and encrypts and transmits it to the intelligent analysis module through a communication module (supporting 4G / 5G and Modbus protocol conversion). This module uses wavelet packet decomposition and principal component analysis (PCA) for feature dimension reduction, and constructs a dynamic knowledge graph by integrating equipment physical constraints and historical fault cases. It uses a spatio-temporal causal adversarial network (ST-CAN) to achieve anomaly detection and classification, and then calculates the root cause probability by traversing the fault propagation path in the graph in reverse through a Bayesian network. Finally, the terminal server generates a visual maintenance work order (health heat map, fault tracing tree, multi-scheme decision-making dashboard); after the maintenance is executed, the system dynamically corrects the node weights of the knowledge graph through feedback data and triggers incremental training of ST-CAN (for example, adjusting the generator LSTM time window). At the same time, the communication module issues a parameter update instruction to the sensor, forming a self-enhancing cycle of "data acquisition → intelligent diagnosis → decision optimization → closed-loop feedback". Through the collaborative design of multi-modal perception, causal knowledge-driven and dynamic self-optimization, this system architecture realizes end-to-end low latency, reduces the comprehensive operation and maintenance cost, improves the recognition rate of new fault modes, and provides a highly reliable and adaptive integrated solution for the intelligent operation and maintenance of oilfield equipment.

[0042] In summary, the oil production equipment fault monitoring system and method provided by the present invention significantly improve the sensitivity of fault detection and the accuracy of root cause location through multi-modal data fusion and dynamic knowledge graph construction. Combining the physical constraint verification mechanism of the spatio-temporal causal adversarial network significantly reduces the false alarm rate. Its closed-loop self-optimization system corrects the knowledge graph weights and model parameters in real time by maintaining feedback data, enabling the system to adapt to equipment aging and working condition drift, shortening the recognition response time of new fault modes. At the same time, the maintenance plan is optimized based on the multi-objective genetic algorithm, effectively reducing the comprehensive operation and maintenance cost, providing a full-chain intelligent solution for oilfield equipment from abnormal perception to precise maintenance, effectively solving industry problems such as high missed detection rate and insufficient dynamic adaptation ability in traditional methods, and having significant economic benefits and engineering practical value.

[0043] It should be understood that in various embodiments of the present invention, the magnitudes of the serial numbers of the above processes do not mean the order of execution is prior or subsequent. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0044] In addition, the terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article merely describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0045] It should be understood that in the embodiments of the present invention, "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0046] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0047] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0048] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Additionally, the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can also be electrical, mechanical, or other forms of connection.

[0049] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments of the present invention.

[0050] Furthermore, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0051] Through the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or firmware, or a combination thereof. When implemented in software, the above functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. The computer-readable medium includes computer storage media and communication media, where the communication media includes any medium that facilitates the transmission of a computer program from one place to another. The storage media can be any available medium that can be accessed by a computer. By way of example but not limitation: the computer-readable medium can include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer. In addition, any connection can suitably be a computer-readable medium. For example, if the software is transmitted using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL) or wireless technologies such as infrared, radio and microwave from a website, server or other remote source, then the coaxial cable, fiber optic cable, twisted pair, DSL or wireless technologies such as infrared, wireless and microwave are included in the definition of the medium. As used in the present invention, disk and disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, where disks generally reproduce data magnetically, while discs reproduce data optically with a laser. The above combinations should also be included within the scope of protection of the computer-readable medium.

[0052] In summary, the above description is only a preferred embodiment of the technical solution of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for monitoring faults of oil production equipment, characterized in that, The oil production equipment fault monitoring method includes: Obtain the operation data of the oil production equipment, and perform feature extraction on the obtained operation data to obtain a target feature vector; Fuse the obtained target feature vector with the historical fault case library and the physical constraint equations of the oil production equipment to construct a dynamically updated knowledge graph; Based on the spatio-temporal causal adversarial network, analyze the distribution shift of the target feature vector in the spatio-temporal dimension, detect abnormal events and output abnormal type labels; According to the output abnormal type labels, combined with the knowledge graph, trace the propagation path of the abnormal events, and calculate the fault probability of the root cause components through the Bayesian network to monitor and warn the oil production equipment.

2. The oil production equipment fault monitoring method according to claim 1, wherein The obtaining the operation data of the oil production equipment, and performing feature extraction on the obtained operation data to obtain a target feature vector includes: According to the obtained operation data, use wavelet packet decomposition to obtain the energy entropy of the transient impact component, and extract the initial fault feature vector; According to the extracted initial feature vector, compress it into a target feature vector through principal component analysis.

3. The oil production equipment fault monitoring method according to claim 1, wherein The fusing the obtained target feature vector with the historical fault case library and the physical constraint equations of the oil production equipment to construct a dynamically updated knowledge graph includes: Establish a static knowledge skeleton of the oil production equipment structure according to the hierarchical relationship and physical constraint equations of each component of the oil production equipment; Extract fault events and propagation paths from the historical fault case library and load them into the static knowledge skeleton; Through the graph embedding algorithm, map the target feature vector to the corresponding nodes of the static knowledge skeleton, generate spatio-temporal state attributes with weights, and form the knowledge graph; Based on the message passing mechanism of the graph neural network, update the weight coefficients of the fault propagation edges in the knowledge graph according to the spatio-temporal similarity between the real-time data of the oil production equipment operation and the historical fault cases; When a new unrecorded fault mode is detected, create a new node and establish its minimum causal association set with the existing nodes through transfer learning to complete the closed-loop evolution of the knowledge graph.

4. The oil production equipment fault monitoring method according to claim 1, characterized in that The spatio-temporal causal adversarial network includes a generator and a discriminator. The analyzing the distribution shift of the target feature vector in the spatio-temporal dimension based on the spatio-temporal causal adversarial network, detecting abnormal events and outputting abnormal type labels includes: The generator extracts the spatial pattern of the target feature vector through a 3D convolutional network, and combines with the LSTM network to learn the time evolution law to generate a normal state feature distribution; The discriminator calculates the spatio-temporal distribution difference through a multi-head attention mechanism according to the target feature vector and the generated normal state feature distribution, and combines with the physical constraints of the knowledge graph to identify abnormal events and calculate the abnormal confidence score; Based on the calculated abnormal confidence score, retrieve the fault mode with the highest similarity from the historical fault feature library and output the abnormal type label with probability.

5. The oil production equipment fault monitoring method according to claim 4, characterized in that, The tracing the propagation path of the abnormal events according to the output abnormal type labels, combined with the knowledge graph, and calculating the fault probability of the root cause components through the Bayesian network includes: Based on the fault propagation edges in the knowledge graph, perform backward traversal along the starting node corresponding to the abnormal type label to trace the potential propagation path of the abnormal event; Based on the Bayesian network and the real-time operation data of the oil production equipment, calculate the fault probabilities of the root cause components in each propagation path.

6. The method for monitoring faults of oil production equipment according to claim 5, characterized in that, After tracing the propagation path of the abnormal event according to the knowledge graph and calculating the fault probabilities of the root cause components through the Bayesian network, the oil production equipment fault monitoring method further includes: Generate a maintenance work order for the components to be inspected according to the fault probabilities; Predict the remaining service life of the components based on the component degradation trend, and generate an optimal maintenance plan in combination with the maintenance cost; Dynamically correct the node weights of the knowledge graph according to the maintenance feedback data, and trigger the incremental training of the spatio-temporal causal adversarial network to achieve closed-loop optimization.

7. The oil production equipment fault monitoring method according to claim 6, wherein, The predicting the remaining service life of the components based on the component degradation trend and generating an optimal maintenance plan in combination with the maintenance cost includes: Analyze the component degradation trend according to the historical fault data of the oil production equipment and the physical wear law, and predict the remaining service life of each component; Calculate the total cost at different maintenance time points according to the predicted remaining life; Generate an optimal maintenance plan through a genetic algorithm according to the calculated total cost and the potential fault risk of the oil production equipment.

8. An oil production equipment fault monitoring system, characterized in that, To implement the oil production equipment fault monitoring method described in any one of claims 1-7, the oil production equipment fault monitoring system includes: A data acquisition module, including a sensor array, for real-time acquisition of the operation data of the oil production equipment; An intelligent analysis module, for analyzing the acquired operation data, predicting the fault probabilities of the components in the oil production equipment, generating a maintenance work order for the components to be inspected, and formulating an optimal maintenance plan; A terminal server, configured with a display unit, for displaying the real-time operation data of the oil production equipment and dispatching maintenance work orders.

9. The oil production equipment fault monitoring system according to claim 8, wherein The intelligent analysis module is further configured to output a fault warning signal according to the fault probabilities, and the display unit is further configured to display fault warning information according to the fault warning signal.

10. The oil production equipment fault monitoring system according to claim 8, wherein, The oil production equipment fault monitoring system further includes: A communication module, for realizing data transmission between the data acquisition module, the intelligent analysis module, and the terminal server, and receiving remote control instructions.

Images