A multi-modal intelligent inspection system and device for an oilfield gathering station

Abstract

The application discloses a kind of multi-modal intelligent inspection system and device for oilfield gathering and transportation station, belong to oilfield gathering and transportation station inspection technical field, including data acquisition module, abnormality determination module, station model simulation analysis module, decision analysis and execution module, the data acquisition module, including vibration, temperature, pressure, flow, liquid level, gas, ultraviolet flame, infrared thermal imaging, visible light, depth camera sensor, to gather current oilfield gathering and transportation station process equipment, environmental state, personnel operation behavior data, record sensor acquisition data in near-field edge server and cloud server, the abnormality determination module, eliminate the time difference of acquisition data by space-time alignment algorithm, generate uniform format time sequence health vector.The application discloses a kind of multi-modal intelligent inspection system and device for oilfield gathering and transportation station, can be based on oilfield gathering and transportation station type, carries out multi-modal inspection to oilfield gathering and transportation station, identifies and handles typical exception.

Description

Technical Field

[0001] This invention belongs to the field of oilfield gathering and transportation station inspection technology, specifically a multimodal intelligent inspection system and device for oilfield gathering and transportation stations. Background Technology

[0002] Oilfield gathering and transportation stations are an important part of oilfield surface engineering. They are a general term for various stations that collect, process, store and transport mixtures of crude oil, natural gas and produced water produced from oil wells scattered throughout the oilfield.

[0003] With the continuous advancement of digitalization and intelligentization in oilfields, gathering and transportation stations face multiple challenges, including complex equipment types, harsh environmental conditions, and stringent operation and maintenance requirements. Existing inspection methods generally rely on manual labor or single-type sensing equipment, making it difficult to achieve comprehensive and real-time perception and response to complex operating conditions. This results in problems such as monitoring blind spots, delayed response, and fragmented operations, which seriously restrict the improvement of unmanned operation and intelligent maintenance levels. In particular, in typical oilfield scenarios such as deserts, remote well sites, and high-risk areas, traditional solutions are unable to cover all abnormal states and cannot adjust the execution strategy according to the actual station policies. There is an urgent need to build a multimodal intelligent inspection system that integrates perception, analysis, inference, and linkage control. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art. To this end, this invention proposes a multimodal intelligent inspection system and device for oilfield gathering and transportation stations, which solves the aforementioned technical problem by improving the detection and processing methods.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A multimodal intelligent inspection system for oilfield gathering and transportation stations includes a data acquisition module, an anomaly detection module, a station model simulation and analysis module, and a decision analysis and execution module.

[0007] The data acquisition module includes vibration, temperature, pressure, flow rate, liquid level, gas, ultraviolet flame, infrared thermal imaging, visible light, and depth camera sensors, which are used to collect data on the current oilfield gathering and transportation station's process equipment, environmental conditions, and personnel operation behavior, and record the sensor data in the near-field edge server and cloud server.

[0008] The anomaly detection module eliminates the time difference of collected data through a spatiotemporal alignment algorithm, generates a time-series health vector in a unified format, performs multidimensional analysis on the input time-series health vector through a neural network model, and outputs the risk level. At the same time, it calls a preset rule base, classifies abnormal parameters based on sensor collection parameters combined with device type and parameter thresholds, integrates rule results with AI judgment, and finally outputs the anomaly risk level and type.

[0009] The station model simulation and analysis module, based on the three-phase thermal model of oil, gas and water and the current geometric structure of the oilfield gathering and transportation station, constructs a three-layer coupled digital twin engine. It constructs a 3D model of the current station based on the on-site modeling data of the oilfield gathering and transportation station and the point cloud inverse reconstruction to reproduce the equipment and process behavior logic. It presets control strategies according to the specificity of different station types and automatically calls the corresponding process section in the digital twin model to perform strategy simulation and deduction according to the anomaly type.

[0010] The decision analysis and execution module comprehensively analyzes the equipment response speed, energy consumption performance and safety margin under each strategy, automatically performs index normalization and weighted evaluation, selects the optimal solution based on the current characteristics of the oilfield gathering and transportation station, generates adjustment logs and records the execution strategy and adjustment parameters, and manually evaluates and adjusts the strategy execution tendency.

[0011] Furthermore, the data acquisition module collects equipment parameters by installing vibration, temperature, pressure, flow, and liquid level sensors at the pump sets and pipeline locations of the oilfield gathering and transportation station. It also collects environmental data in real time by using gas, visible light, infrared thermal imaging, ultraviolet flame, and depth camera sensors to cover the work area and equipment area. The module identifies personnel outlines and records personnel work behaviors, including employee clothing and location, through the YOLO algorithm. It also identifies the location and shape of objects. By acquiring the three-dimensional spatial information of the station, the sensor data is uploaded to the near-field edge server in real time for data cleaning, filtering, and storage. The near-field edge server then uploads the processed data to the cloud server via the network.

[0012] Furthermore, the anomaly detection module eliminates the time difference in the collected data processed by the data acquisition module through a spatiotemporal alignment algorithm, and generates a time-series health vector in a unified format. The specific steps are as follows:

[0013] For the collected sensor data, the sensor timestamps are aligned using an edge clock synchronization protocol. Corresponding time-series health vectors are constructed based on the equipment types of oilfield gathering and transportation stations. Historical parameters of normal operation and typical anomalies of equipment under different types are collected according to equipment type, and a training vector set is constructed. The training vector set is divided into training set, validation set, and test set to train and obtain anomaly detection neural network models for different types of equipment. Risk levels are classified based on typical anomaly types of different equipment. The anomaly detection neural network model is used as input to the time-series health vectors to obtain typical equipment anomalies and output the corresponding risk levels, including yellow alert, orange alert, and red alert.

[0014] Based on the historical normal value ranges of sensor parameters from different devices, the parameters are statistically analyzed and combined with the 3σ principle. The mean plus or minus three standard deviations are used as the parameter judgment threshold to identify abnormal sensors. Different risk judgment criteria are then applied to the same type of equipment for further assessment.

[0015] If the parameters collected by the device exceed the proportion of the yellow alarm, it is judged as a yellow anomaly;

[0016] If the parameters collected by the device exceed the proportion of orange alarms, it is judged as an orange anomaly;

[0017] If the parameters collected by the device exceed the red alarm ratio, it is judged as a red anomaly;

[0018] The risk level output by the integrated anomaly detection neural network model is combined with the anomaly type determined by the risk assessment criteria. The model outputs the risk level or anomaly type with the highest degree of danger, where the degree of danger is red > orange > yellow.

[0019] Furthermore, the station model simulation and analysis module, based on the three-phase thermodynamic model of oil, gas and water and the current geometric structure of the oilfield gathering and transportation station, constructs a three-layer coupled digital twin engine. It recreates the current station's 3D model using on-site modeling data and point cloud reverse engineering, thereby replicating the equipment and process behavior logic. This includes the following steps:

[0020] Based on the equipment types of the current oilfield gathering and transportation station, and combined with on-site modeling data and point cloud reverse engineering, 3D models of different equipment groups of the current station are constructed as the geometric layer of the three-layer coupled digital twin engine. The construction of the geometric layer includes using laser scanners and UAVs to collect data on the oilfield gathering and transportation station, obtain the station's three-dimensional spatial information and surface texture information, filter and reduce noise on the collected point cloud data, reconstruct the station's three-dimensional model using the point cloud data, and map the collected surface texture information onto the 3D model.

[0021] The working principles and operating characteristics of different equipment in the oilfield gathering and transportation station are analyzed, and mathematical models of the equipment are established. At the same time, according to the station's technological process, a process logic model is established. The mathematical models and process logic modules constructed by different equipment groups are used as the mechanism layer input into a three-layer coupled digital twin engine. Simultaneously, the time-series health vector output by the anomaly judgment module is received and mapped to the equipment attributes in the three-dimensional model to realize the real-time update of the corresponding equipment model status and reproduce the technological behavior logic of different equipment.

[0022] Furthermore, the specific steps of pre-setting control strategies based on the specific characteristics of different station types and automatically calling the corresponding process section in the digital twin model to perform strategy simulation and deduction according to the anomaly type are as follows:

[0023] The stations under its jurisdiction are classified according to their type and... Establish a station database, create files based on type, including inter-well stations, transfer stations, and joint stations, and further classify them based on geographical environment, grouping stations of the same type into the same sub-category according to their corresponding geographical environment;

[0024] Collect typical anomaly resolution strategies from different stations, record the adjustment parameter types and control strategies of equipment associated with different typical anomalies, count the types of resolution strategies for different typical anomalies and the number of times the corresponding resolution strategies are used, sort the resolution strategies in descending order based on the number of times they are used, and select the top three resolution strategies as the basic strategies for the current typical anomaly.

[0025] Based on typical anomalies of different subcategories, the process segments corresponding to different typical anomalies are defined in the digital twin model. The specific steps are as follows:

[0026] ;

[0027] Where H is a hash function, E represents the exception type, and P represents the process segment. A hash table is used to achieve a fast mapping between exception types and process segments, mapping exception type E to the corresponding process segment P.

[0028] Select the basic strategy for typical anomalies based on the current oilfield gathering and transportation station type. According to the anomaly type and mapping relationship, the corresponding process section in the digital twin model is automatically called for simulation based on the basic strategy for typical anomalies. The preset control strategy is run on the selected process section to simulate the station's operating status under abnormal conditions. The response speed, energy consumption performance and safety margin under each strategy path are output to the decision analysis and execution module.

[0029] Furthermore, the decision analysis and execution module comprehensively analyzes the equipment response speed, energy consumption performance, and safety margin under each strategy, automatically performs index normalization and weighted evaluation, selects the optimal solution based on the current characteristics of the oilfield gathering and transportation station, generates an adjustment log and records the execution strategy and adjustment parameters, and manually evaluates and adjusts the strategy execution tendency, including the following steps:

[0030] Based on the response speed, energy consumption performance and safety margin of the current typical anomaly of the station under different strategies output by the three-layer coupled digital twin engine, the different indicators are normalized and weighted based on the weight of different indicators. Based on the comprehensive evaluation score, the basic strategy is selected as the execution strategy of the current typical anomaly.

[0031] Generate adjustment logs and record execution strategies and adjustment parameters, including actual and simulated execution strategies. Manually evaluate and adjust the strategy execution tendency. The specific steps are as follows:

[0032] Based on the actual and simulated execution strategies after each typical anomaly occurs, the optimal strategy is manually evaluated, and the weights of different parameters in all basic strategies are adjusted using the gradient descent method to maximize the comprehensive evaluation score of the manually evaluated optimal strategy. The updated weight parameters are then recorded in the decision analysis and execution module.

[0033] Repeat the above strategy execution tendency adjustment steps after each typical anomaly occurs, and update the parameter weights in real time.

[0034] Furthermore, the response speed, energy consumption, and safety margin of the current typical station anomalies under different strategies output by the three-layer coupled digital twin engine are normalized for different indicators, and a weighted evaluation is performed based on the weights of different indicators. The basic strategy is selected as the execution strategy for the current typical anomaly based on the comprehensive evaluation score. The specific steps are as follows:

[0035] For the response speed, energy consumption, and safety margin of typical anomalies in the current station under different strategies output by the three-layer coupled digital twin engine, the actual values ​​are converted into values ​​between 0 and 1 through index normalization, and weight values ​​are assigned to each value, where the weight value is 1. , where n is the total number of indicators in the response speed, energy consumption performance and safety margin of the current station's typical anomalies. By multiplying the normalized indicator values ​​by their corresponding weights and then summing them, the comprehensive evaluation score of each strategy is obtained, and the strategy with the highest comprehensive evaluation score is selected as the optimal solution.

[0036] The multimodal intelligent inspection device for oilfield gathering and transportation stations is applied to the multimodal intelligent inspection system for oilfield gathering and transportation stations. It includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the multimodal intelligent inspection system for oilfield gathering and transportation stations as described in this invention.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] 1. In this invention, by integrating multiple types of sensors such as temperature, pressure, vibration, flow, gas, flame image, infrared thermal imaging, and depth vision, a multimodal perception matrix is ​​constructed to achieve all-round, all-weather status perception of key equipment and process nodes in oilfield gathering and transportation stations. By adopting edge fusion algorithm and visual recognition technology, it can accurately identify various abnormal operating conditions such as abnormal pump vibration, water mixing imbalance, valve leakage, missing flame, and personnel crossing boundaries, providing basic data support for abnormal early warning and coordinated response.

[0039] 2. In this invention, by adopting an edge computing architecture, an AI recognition model and rule base are deployed on the local device to achieve real-time classification and dual judgment of abnormal working conditions. The edge server system supports autonomous operation when the link is disconnected. The state machine control logic and the solidified model are used to handle events in a closed loop, ensuring that remote stations still have intelligent judgment and response capabilities when communication is unstable or the network is temporarily disconnected, and ensuring safe and stable operation under unmanned or minimally staffed conditions.

[0040] 3. In this invention, a three-layer coupled digital twin model is constructed, and the basic adjustment strategies for typical anomaly generation of different oilfield gathering and transportation stations are combined. The strategy execution is simulated through the digital twin model, and the optimal strategy is selected based on the weight value. At the same time, based on the actual strategy tendency of different oilfield gathering and transportation stations, the weight values ​​of different basic strategies are adjusted in real time to ensure that the system can respond to station policies in a timely manner and that the strategy always matches the actual needs of the station. Attached Figure Description

[0041] Figure 1 This is a block diagram of a multimodal intelligent inspection system for oilfield gathering and transportation stations according to the present invention. Detailed Implementation

[0042] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example 1:

[0044] like Figure 1 As shown, a multimodal intelligent inspection system for oilfield gathering and transportation stations includes a data acquisition module, an anomaly detection module, a station model simulation and analysis module, and a decision analysis and execution module.

[0045] The data acquisition module includes vibration, temperature, pressure, flow rate, liquid level, gas, ultraviolet flame, infrared thermal imaging, visible light, and depth camera sensors, which are used to collect data on the current process equipment, environmental conditions, and personnel operation behavior of the oilfield gathering and transportation station, and record the sensor data in the near-field edge server and cloud server.

[0046] The data acquisition module collects equipment parameters by installing vibration, temperature, pressure, flow, and liquid level sensors at the pump sets and pipeline locations of oilfield gathering and transportation stations. It also collects environmental data in real time by using gas, visible light, infrared thermal imaging, ultraviolet flame, and depth camera sensors to cover the work area and equipment area. The module uses the YOLO algorithm to identify personnel silhouettes and record personnel work behaviors, including employee clothing and location. It also identifies the location and shape of objects. By acquiring the three-dimensional spatial information of the station, the sensor data is uploaded to the near-field edge server in real time for data cleaning, filtering, and storage. The near-field edge server then uploads the processed data to the cloud server via the network.

[0047] It should be noted that all sensors support standard industrial-grade plug-and-play, such as CAN, 485, and 4-20mA, facilitating maintenance, replacement, and rapid deployment during actual use. By deploying sensors at equipment pump sets and pipeline locations, real-time parameters of station equipment during oilfield gathering and transportation are collected, aiding in subsequent anomaly detection. For rotating equipment such as pumps and compressors in oilfield gathering and transportation stations, vibration sensors are installed at critical components such as bearings and motors. By monitoring parameters such as vibration frequency and amplitude, potential faults such as imbalance or loosening of the equipment can be identified. For high-temperature equipment, such as heating furnaces, temperature sensors are installed on the furnace walls and inlet / outlet pipelines. Real-time monitoring of temperature changes prevents safety accidents caused by overheating. Pressure sensors are installed at different nodes of the gathering and transportation pipeline to monitor the pressure inside the pipeline at all times, avoiding problems such as pipeline rupture due to abnormal pressure. Appropriate flow sensors, such as electromagnetic flow meters and ultrasonic flow meters, are selected according to pipeline diameter and fluid properties to accurately determine the fluid transport volume. Liquid level sensors are used to monitor the liquid level in containers such as oil tanks and water tanks. Gas sensors are installed in flammable and explosive areas and poorly ventilated areas of the station to detect the concentration of combustible and toxic gases. Visible light cameras are installed in various areas of the station to monitor personnel operation behavior, equipment appearance, etc., enabling remote monitoring and recording of on-site conditions.

[0048] The anomaly detection module eliminates the time difference of collected data through a spatiotemporal alignment algorithm, generates a time-series health vector in a unified format, performs multidimensional analysis on the input time-series health vector through a neural network model, and outputs the risk level. At the same time, it calls a pre-set rule base, classifies abnormal parameters based on sensor-collected parameters, combined with device type and parameter thresholds, integrates rule results and AI judgment, and finally outputs the anomaly risk level and type.

[0049] The anomaly detection module eliminates the time difference in the acquired data processed by the data acquisition module through a spatiotemporal alignment algorithm and generates a time-series health vector in a unified format. The specific steps are as follows:

[0050] For the collected sensor data, the sensor timestamps are aligned using an edge clock synchronization protocol. Corresponding time-series health vectors are constructed based on the equipment types of oilfield gathering and transportation stations. Historical parameters of normal operation and typical anomalies of equipment under different types are collected according to equipment type, and a training vector set is constructed. The training vector set is divided into training set, validation set, and test set to train and obtain anomaly detection neural network models for different types of equipment. Risk levels are classified based on typical anomaly types of different equipment. The anomaly detection neural network model is used as input to the time-series health vectors to obtain typical equipment anomalies and output the corresponding risk levels, including yellow alert, orange alert, and red alert.

[0051] It should be noted that the specific steps for aligning sensor timestamps using the edge clock synchronization protocol are as follows:

[0052] ;

[0053] in, For sensor readings, Represents the actual sampling time. For the target alignment time, Representative sensor At the actual sampling time The measured values ​​at that time This represents the sensor alignment reading obtained after time interpolation at the target alignment time; in other words, it's the sensor value after aligning asynchronous data to the target time. Represents the sensor in The sampled values ​​at specific times show typical anomalies in oilfield gathering and transportation station equipment, including cavitation failures in pumps, bearing wear, seal leaks, compressor surge failures, valve failures, pipeline blockages, corrosion perforations, and furnace combustion failures. Historical sensor parameters under these typical anomalies are collected and a training vector set is established to specifically identify typical faults in different equipment. The time-series health vector is... ,in This represents the first time sequence after synchronization. Sensing parameters, At any given moment, the type of sensor used in different types of equipment varies, and the risk level needs to be set based on the specific typical abnormal effects of different equipment. You can consult experts in the relevant field and use experience to set the risk level, including yellow alert, orange alert, and red alert, with the degree of harm increasing in that order.

[0054] Based on the historical normal value ranges of sensor parameters from different devices, the parameters are statistically analyzed and combined with the 3σ principle. The mean plus or minus three standard deviations are used as the parameter judgment threshold to identify abnormal sensors. Different risk judgment criteria are then applied to the same type of equipment for further assessment.

[0055] If the parameters collected by the device exceed the proportion of the yellow alarm, it is judged as a yellow anomaly;

[0056] If the parameters collected by the device exceed the proportion of orange alarms, it is judged as an orange anomaly;

[0057] If the parameters collected by the device exceed the red alarm ratio, it is judged as a red anomaly;

[0058] The risk level output by the integrated anomaly detection neural network model is combined with the anomaly type determined by the risk assessment criteria. The model outputs the risk level or anomaly type with the highest degree of danger, where the degree of danger is red > orange > yellow.

[0059] It should be noted that the risk assessment criteria for different devices need to be determined based on the number of sensors and the number of abnormal sensors on the device. The yellow alarm ratio is when more than one-quarter but less than one-half of the device sensors are abnormal, the orange alarm ratio is when more than one-half but less than three-quarters of the device sensors are abnormal, and the red alarm ratio is when more than three-quarters of the device sensors are abnormal.

[0060] Example 2:

[0061] The station model simulation and analysis module, based on the three-phase thermal model of oil, gas and water and the current geometric structure of the oilfield gathering and transportation station, constructs a three-layer coupled digital twin engine. It builds a 3D model of the current station based on the on-site modeling data of the oilfield gathering and transportation station and the point cloud inverse reconstruction to reproduce the equipment and process behavior logic. It presets control strategies according to the specificity of different station types and automatically calls the corresponding process section in the digital twin model to perform strategy simulation and deduction according to the anomaly type.

[0062] The station model simulation and analysis module, based on the three-phase thermodynamic model of oil, gas and water and the current geometric structure of the oilfield gathering and transportation station, constructs a three-layer coupled digital twin engine. It builds a 3D model of the current station based on on-site modeling data and point cloud inversion to reproduce the equipment and process behavior logic. This includes the following steps:

[0063] Based on the equipment types of the current oilfield gathering and transportation station, and combined with on-site modeling data and point cloud reverse engineering, 3D models of different equipment groups of the current station are constructed as the geometric layer of the three-layer coupled digital twin engine. The construction of the geometric layer includes using laser scanners and UAVs to collect data on the oilfield gathering and transportation station, obtain the station's three-dimensional spatial information and surface texture information, filter and reduce noise on the collected point cloud data, reconstruct the station's three-dimensional model using the point cloud data, and map the collected surface texture information onto the 3D model.

[0064] It should be noted that the geometry layer includes a tank model that recreates the structure of crude oil, water, or gas storage tanks in three dimensions and supports real-time display and positioning of liquid level changes and alarm status; a pump group model that accurately displays the appearance, model, and installation location of the delivery pump and reflects the start-up and shutdown status and operating frequency in real time based on operating data; a visualized water mixing device structure and control interface; a water mixing unit model used to simulate the linkage behavior of water mixing ratio and mixing pipeline; a heating furnace model that presents the layout of the core components of the heating furnace and the structure of the combustion chamber and supports real-time thermal field coverage and linkage demonstration of operating status; and a pipeline segment model that reproduces the main and branch pipeline network based on the actual process diagram and has the functions of flow direction indication, logical connectivity, and dynamic pressure drop simulation.

[0065] It can also support quick highlighting of target equipment in a 3D scene by equipment code or name, which is convenient for fault location and maintenance guidance. It provides virtual 3D inspection function for stations, supports rotation, zoom, flight view and cross-section view, which is convenient for understanding the overall structure. It can overlay thermal attribute layers such as temperature, flow rate, and pressure onto the solid model to realize intuitive visualization of abnormal hot areas and tracking of temperature rise trends.

[0066] The working principles and operating characteristics of different equipment in the oilfield gathering and transportation station are analyzed, and mathematical models of the equipment are established. At the same time, according to the station's technological process, a process logic model is established. The mathematical models and process logic modules constructed by different equipment groups are used as the mechanism layer input into a three-layer coupled digital twin engine. Simultaneously, the time-series health vector output by the anomaly judgment module is received and mapped to the equipment attributes in the three-dimensional model to realize the real-time update of the corresponding equipment model status and reproduce the technological behavior logic of different equipment.

[0067] It should be noted that the mechanism layer includes a heat exchange model that simulates the heat transfer process in the heating furnace, predicts in real time the temperature difference between inlet and outlet, heat exchange efficiency and their dynamic impact on energy consumption, simulates the separation and transport state of three-phase fluids (oil, water and gas) in equipment and pipelines, analyzes the three-phase flow model of flow velocity, proportion and turbulence distribution, dynamically simulates the coupling relationship between water mixing ratio, inlet temperature, outlet temperature and mixing point pressure, a water mixing balance model to assist in optimizing water mixing strategy, a pressure difference calculation model that calculates the pressure difference before and after key nodes based on pressure sensor data to determine whether there is blockage, air resistance or abnormal energy consumption, and an interface that connects real-time operating data to thermodynamic and fluid dynamic state equations to drive the model to calculate the state equation of equipment response and process behavior.

[0068] Based on the specific control strategies for different station types, and according to the anomaly type, the corresponding process section in the digital twin model is automatically invoked for strategy simulation and deduction. The specific steps are as follows:

[0069] The stations under its jurisdiction are classified according to their type and... Establish a station database, create files based on type, including inter-well stations, transfer stations, and joint stations, and further classify them based on geographical environment, grouping stations of the same type into the same sub-category according to their corresponding geographical environment;

[0070] It should be noted that the geographical environment includes typical oilfield scenarios such as deserts, remote well sites, and high-risk areas.

[0071] Collect typical anomaly resolution strategies from different stations, record the adjustment parameter types and control strategies of equipment associated with different typical anomalies, count the types of resolution strategies for different typical anomalies and the number of times the corresponding resolution strategies are used, sort the resolution strategies in descending order based on the number of times they are used, and select the top three resolution strategies as the basic strategies for the current typical anomaly.

[0072] Based on typical anomalies of different subcategories, the process segments corresponding to different typical anomalies are defined in the digital twin model. The specific steps are as follows:

[0073] ;

[0074] Where H is a hash function, E represents the exception type, and P represents the process segment. A hash table is used to achieve a fast mapping between exception types and process segments, mapping exception type E to the corresponding process segment P.

[0075] It should be noted that after mapping different typical anomalies to process segments using hash functions, the digital twin model needs to be verified to ensure the accuracy and reliability of the process segments in the model. By comparing and calibrating with actual station data, the accuracy of the digital twin model can be improved.

[0076] Select the basic strategy for typical anomalies based on the current oilfield gathering and transportation station type. Based on the anomaly type and mapping relationship, the corresponding process section in the digital twin model is automatically called for simulation based on the basic strategy for typical anomalies. The preset control strategy is run on the selected process section to simulate the station's operating status under abnormal conditions. The response speed, energy consumption performance and safety margin under each strategy path are output to the decision analysis and execution module.

[0077] It should be noted that the data layer in the three-layer coupled digital twin engine includes receiving the health vector output by the edge analysis module, which is mapped to the device attributes of the 3D model to achieve real-time status linkage. It binds the device's operating status with visual elements such as alarm colors and animation effects to achieve visual reflection of the status. Through real-time data flow, it drives the behavior changes of the twin model, simulates the device's operating trend and predicts the evolution of the working condition. For devices or areas in abnormal states, it automatically highlights and flashes alarm colors and adds risk labels and text descriptions to achieve visual alarm prompts.

[0078] The decision analysis and execution module comprehensively analyzes the equipment response speed, energy consumption, and safety margin under each strategy, automatically performs index normalization and weighted evaluation, selects the optimal solution based on the current characteristics of the oilfield gathering and transportation station, generates an adjustment log and records the execution strategy and adjustment parameters, and manually evaluates and adjusts the strategy execution tendency, including the following steps:

[0079] Based on the response speed, energy consumption, and safety margin of typical anomalies in the current station under different strategies output by the three-layer coupled digital twin engine, the different indicators are normalized and weighted based on their respective weights. A basic strategy is selected as the execution strategy for the current typical anomaly based on the comprehensive evaluation score. The specific steps are as follows:

[0080] For the response speed, energy consumption, and safety margin of typical anomalies in the current station under different strategies output by the three-layer coupled digital twin engine, the actual values ​​are converted into values ​​between 0 and 1 through index normalization, and weight values ​​are assigned to each value, where the weight value is 1. , where n is the total number of indicators in the response speed, energy consumption performance and safety margin of the current station's typical anomalies. By multiplying the normalized indicator values ​​by their corresponding weights and then summing them, the comprehensive evaluation score of each strategy is obtained, and the strategy with the highest comprehensive evaluation score is selected as the optimal solution.

[0081] It should be noted that the response speed data is the time from receiving the instruction to reaching a stable operating state for the corresponding abnormal device, the energy consumption data is the total power and gas consumption during the operation of the devices involved in the current strategy, the safety margin data is the difference between the device operating parameters and the safety threshold, and the normalization is selected as Z-Score normalization.

[0082] It should be noted that the system supports multiple channel modes such as cellular communication, LoRa, Wi-Fi, and industrial short-range wireless, and has the ability to determine link status. When the main link is interrupted, the system automatically switches to the backup communication channel. If all external communication links fail, the system will switch to the local state machine control mode, trigger local alarms, log archiving, equipment protection, and other operations according to preset levels, and maintain local inspection closed-loop operation to ensure that it still has basic decision-making and handling capabilities in environments such as no network, remote areas, and desert sites to the greatest extent.

[0083] Generate adjustment logs and record execution strategies and adjustment parameters, including actual and simulated execution strategies. Manually evaluate and adjust the strategy execution tendency. The specific steps are as follows:

[0084] Based on the actual and simulated execution strategies after each typical anomaly occurs, the optimal strategy is manually evaluated, and the weights of different parameters in all basic strategies are adjusted using the gradient descent method to maximize the comprehensive evaluation score of the manually evaluated optimal strategy. The updated weight parameters are then recorded in the decision analysis and execution module.

[0085] Repeat the above strategy execution tendency adjustment steps after each typical anomaly occurs, and update the parameter weights in real time.

[0086] It should be noted that by repeating the above strategy execution tendency adjustment steps after each typical anomaly occurs and updating the parameter weights in real time, it is possible to adapt to the real-time strategy tendency changes of different oilfield gathering and transportation stations. The optimal solution for typical anomaly resolution is to continuously adjust the parameters to execute the basic strategy that best suits the current situation of the oilfield gathering and transportation station. During the weight parameter adjustment process, the goal is to achieve the highest comprehensive evaluation score of the optimal strategy selected by manual evaluation. The weights of different parameters in all basic strategies are adjusted, and the gradient of the score with respect to the weight is calculated based on the gradient descent method. The weights are adjusted along the gradient direction until the state with the highest target score is reached, and the corresponding weight adjustment value is obtained.

[0087] It should be noted that during the execution of the optimal strategy, the selected strategy path is converted into a sequence of control commands that the device can recognize. It supports industrial protocols such as Modbus, OPC UA, and 4-20mA. The control command transmitter is linked with the device, and the commands are sent to the target device, such as the frequency converter, actuator valve, and pump control module, through the edge controller to start the actual control operation. The device execution results, such as frequency changes, current response, and valve position, are collected for closed-loop control confirmation.

[0088] A multimodal intelligent inspection device for oilfield gathering and transportation stations, applied to a multimodal intelligent inspection system for oilfield gathering and transportation stations, includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the multimodal intelligent inspection system for oilfield gathering and transportation stations as described in this invention.

[0089] This invention discloses a multimodal intelligent inspection system and device for oilfield gathering and transportation stations. During operation, it integrates multiple types of sensors, including temperature, pressure, vibration, flow rate, gas, flame images, infrared thermal imaging, and depth vision, to construct a multimodal perception matrix. This enables comprehensive, all-weather status perception of key equipment and process nodes in oilfield gathering and transportation stations. Employing edge fusion algorithms and visual recognition technology, it can accurately identify various abnormal operating conditions such as abnormal pump vibration, water mixing imbalance, valve leakage, missing flames, and personnel crossing boundaries, providing basic data support for abnormal early warning and coordinated response. By adopting an edge computing architecture, it deploys AI recognition models and rule bases on local devices to achieve real-time classification and dual judgment of abnormal operating conditions. The edge server system supports autonomous operation even when the link is disconnected. Through state machine control logic and a fixed model, it performs closed-loop processing of events, ensuring that remote stations still have intelligent judgment and response capabilities when communication is unstable or temporarily disconnected, ensuring safe and stable operation under unmanned or minimally staffed conditions.

[0090] By constructing a three-layer coupled digital twin model and combining it with the basic adjustment strategies for typical anomaly generation at different oilfield gathering and transportation stations, the system simulates strategy execution through the digital twin model and selects the optimal strategy based on weight values. At the same time, based on the actual strategy tendencies of different oilfield gathering and transportation stations, the system adjusts the weight values ​​of different basic strategies in real time to ensure that the system can respond to station policies in a timely manner and that the strategies always match the actual needs of the stations.

[0091] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may exist in actual implementation. Modules described as separate components may or may not be physically separated, and components shown as modules may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the method in this embodiment according to actual needs.

[0092] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A multimodal intelligent inspection system for oilfield gathering and transportation stations, characterized in that: It includes a data acquisition module, an anomaly detection module, a station model simulation and analysis module, and a decision analysis and execution module; The data acquisition module includes vibration, temperature, pressure, flow rate, liquid level, gas, ultraviolet flame, infrared thermal imaging, visible light, and depth camera sensors, which are used to collect data on the current oilfield gathering and transportation station's process equipment, environmental conditions, and personnel operation behavior, and record the sensor data in the near-field edge server and cloud server. The anomaly detection module eliminates the time difference of collected data through a spatiotemporal alignment algorithm, generates a time-series health vector in a unified format, performs multidimensional analysis on the input time-series health vector through a neural network model, and outputs the risk level. At the same time, it calls a preset rule base, classifies abnormal parameters based on sensor collection parameters combined with device type and parameter thresholds, integrates rule results with AI judgment, and finally outputs the anomaly risk level and type. The station model simulation and analysis module, based on the three-phase thermal model of oil, gas and water and the current geometric structure of the oilfield gathering and transportation station, constructs a three-layer coupled digital twin engine. It constructs a 3D model of the current station based on the on-site modeling data of the oilfield gathering and transportation station and the point cloud inverse reconstruction to reproduce the equipment and process behavior logic. It presets control strategies according to the specificity of different station types and automatically calls the corresponding process section in the digital twin model to perform strategy simulation and deduction according to the anomaly type. The decision analysis and execution module comprehensively analyzes the equipment response speed, energy consumption performance and safety margin under each strategy, automatically performs index normalization and weighted evaluation, selects the optimal solution based on the current characteristics of the oilfield gathering and transportation station, generates adjustment logs and records the execution strategy and adjustment parameters, and manually evaluates and adjusts the strategy execution tendency.

2. The multimodal intelligent inspection system for oilfield gathering and transportation stations according to claim 1, characterized in that: The data acquisition module collects equipment parameters by installing vibration, temperature, pressure, flow, and liquid level sensors at the pump sets and pipeline locations of the oilfield gathering and transportation station. It also collects environmental data in real time by using gas, visible light, infrared thermal imaging, ultraviolet flame, and depth camera sensors to cover the work area and equipment area. The module uses the YOLO algorithm to identify personnel silhouettes and record personnel work behaviors, including employee clothing and location. It also identifies the location and shape of objects. By acquiring the three-dimensional spatial information of the station, the sensor data is uploaded to the near-field edge server in real time for data cleaning, filtering, and storage. The near-field edge server then uploads the processed data to the cloud server via the network.

3. The multimodal intelligent inspection system for oilfield gathering and transportation stations according to claim 2, characterized in that: The anomaly detection module eliminates the time difference in the collected data after processing by the data acquisition module through a spatiotemporal alignment algorithm, and generates a time-series health vector in a unified format. The specific steps are as follows: For the collected sensor data, the sensor timestamps are aligned using an edge clock synchronization protocol. Corresponding time-series health vectors are constructed based on the equipment types of oilfield gathering and transportation stations. Historical parameters of normal operation and typical anomalies of equipment under different types are collected according to equipment type, and a training vector set is constructed. The training vector set is divided into training set, validation set, and test set to train and obtain anomaly detection neural network models for different types of equipment. Risk levels are classified based on typical anomaly types of different equipment. The anomaly detection neural network model is used as input to the time-series health vectors to obtain typical equipment anomalies and output the corresponding risk levels, including yellow alert, orange alert, and red alert. Based on the historical normal value ranges of sensor parameters from different devices, the parameters are statistically analyzed and combined with the 3σ principle. The mean plus or minus three standard deviations are used as the parameter judgment threshold to identify abnormal sensors. Different risk judgment criteria are then applied to the same type of equipment for further assessment. If the parameters collected by the device exceed the proportion of the yellow alarm, it is judged as a yellow anomaly; If the parameters collected by the device exceed the proportion of orange alarms, it is judged as an orange anomaly; If the parameters collected by the device exceed the red alarm ratio, it is judged as a red anomaly; The risk level output by the integrated anomaly detection neural network model is combined with the anomaly type determined by the risk assessment criteria. The model outputs the risk level or anomaly type with the highest degree of danger, where the degree of danger is red > orange > yellow.

4. The multimodal intelligent inspection system for oilfield gathering and transportation stations according to claim 2, characterized in that: The station model simulation and analysis module, based on the three-phase thermodynamic model of oil, gas and water and the current geometric structure of the oilfield gathering and transportation station, constructs a three-layer coupled digital twin engine. It builds a 3D model of the current station based on on-site modeling data and point cloud inversion to reproduce the equipment and process behavior logic. This includes the following steps: Based on the equipment types of the current oilfield gathering and transportation station, and combined with on-site modeling data and point cloud reverse engineering, 3D models of different equipment groups of the current station are constructed as the geometric layer of the three-layer coupled digital twin engine. The construction of the geometric layer includes using laser scanners and UAVs to collect data on the oilfield gathering and transportation station, obtain the station's three-dimensional spatial information and surface texture information, filter and reduce noise on the collected point cloud data, reconstruct the station's three-dimensional model using the point cloud data, and map the collected surface texture information onto the 3D model. The working principles and operating characteristics of different equipment in the oilfield gathering and transportation station are analyzed, and mathematical models of the equipment are established. At the same time, according to the station's technological process, a process logic model is established. The mathematical models and process logic modules constructed by different equipment groups are used as the mechanism layer input into a three-layer coupled digital twin engine. Simultaneously, the time-series health vector output by the anomaly judgment module is received and mapped to the equipment attributes in the three-dimensional model to realize the real-time update of the corresponding equipment model status and reproduce the technological behavior logic of different equipment.

5. A multimodal intelligent inspection system for oilfield gathering and transportation stations according to claim 4, characterized in that: The specific steps of the method of pre-setting control strategies based on the specific characteristics of different station types and automatically calling the corresponding process section in the digital twin model to perform strategy simulation and deduction according to the anomaly type are as follows: The stations under its jurisdiction are classified according to their type and... Establish a station database, create files based on type, including inter-well stations, transfer stations, and joint stations, and further classify them based on geographical environment, grouping stations of the same type into the same sub-category according to their corresponding geographical environment; Collect typical anomaly resolution strategies from different stations, record the adjustment parameter types and control strategies of equipment associated with different typical anomalies, count the types of resolution strategies for different typical anomalies and the number of times the corresponding resolution strategies are used, sort the resolution strategies in descending order based on the number of times they are used, and select the top three resolution strategies as the basic strategies for the current typical anomaly. Based on typical anomalies of different subcategories, the process segments corresponding to different typical anomalies are defined in the digital twin model. The specific steps are as follows: ； Where H is a hash function, E represents the exception type, and P represents the process segment. A hash table is used to achieve a fast mapping between exception types and process segments, mapping exception type E to the corresponding process segment P. Select the basic strategy for typical anomalies based on the current oilfield gathering and transportation station type. Based on the anomaly type and mapping relationship, the corresponding process section in the digital twin model is automatically called for simulation based on the basic strategy for typical anomalies. The preset control strategy is run on the selected process section to simulate the station's operating status under abnormal conditions. The response speed, energy consumption performance and safety margin under each strategy path are output to the decision analysis and execution module.

6. A multimodal intelligent inspection system for oilfield gathering and transportation stations according to claim 5, characterized in that: The decision analysis and execution module comprehensively analyzes the equipment response speed, energy consumption, and safety margin under each strategy, automatically performs index normalization and weighted evaluation, selects the optimal solution based on the current characteristics of the oilfield gathering and transportation station, generates an adjustment log and records the execution strategy and adjustment parameters, and manually evaluates and adjusts the strategy execution tendency, including the following steps: Based on the response speed, energy consumption performance and safety margin of the current typical anomaly of the station under different strategies output by the three-layer coupled digital twin engine, the different indicators are normalized and weighted based on the weight of different indicators. Based on the comprehensive evaluation score, the basic strategy is selected as the execution strategy of the current typical anomaly. Generate adjustment logs and record execution strategies and adjustment parameters, including actual and simulated execution strategies. Manually evaluate and adjust the strategy execution tendency. The specific steps are as follows: Based on the actual and simulated execution strategies after each typical anomaly occurs, the optimal strategy is manually evaluated, and the weights of different parameters in all basic strategies are adjusted using the gradient descent method to maximize the comprehensive evaluation score of the manually evaluated optimal strategy. The updated weight parameters are then recorded in the decision analysis and execution module. Repeat the above strategy execution tendency adjustment steps after each typical anomaly occurs, and update the parameter weights in real time.

7. A multimodal intelligent inspection system for oilfield gathering and transportation stations according to claim 6, characterized in that: The response speed, energy consumption, and safety margin of typical anomalies at the current station under different strategies output by the three-layer coupled digital twin engine are normalized for different indicators, and a weighted evaluation is performed based on the weights of different indicators. A basic strategy is selected as the execution strategy for the current typical anomaly based on the comprehensive evaluation score. The specific steps are as follows: For the response speed, energy consumption, and safety margin of typical anomalies in the current station under different strategies output by the three-layer coupled digital twin engine, the actual values ​​are converted into values ​​between 0 and 1 through index normalization, and weight values ​​are assigned to each value, where the weight value is 1. , where n is the total number of indicators in the response speed, energy consumption performance and safety margin of the current station's typical anomalies. By multiplying the normalized indicator values ​​by their corresponding weights and then summing them, the comprehensive evaluation score of each strategy is obtained, and the strategy with the highest comprehensive evaluation score is selected as the optimal solution.

8. A multimodal intelligent inspection device for oilfield gathering and transportation stations, characterized in that, The device is applied to a multimodal intelligent inspection system for oilfield gathering and transportation stations, and includes a memory and a processor: Memory is used to store computer-readable instructions in a non-transitory manner. Processor, for executing the computer-readable instructions; When the computer-readable instructions are executed by the processor, they perform the steps in the system according to any one of claims 1-7.

Images