An oilfield perimeter intrusion early warning and emergency response linkage intelligent agent closed-loop scheduling method and system

Abstract

This invention relates to an intelligent agent closed-loop scheduling method and system for linking early warning and emergency response for oilfield perimeter intrusion, encompassing the fields of scheduling and emergency response. It addresses existing problems in oilfield perimeter security, such as insufficient multi-sensor data fusion, low early warning accuracy, lack of a tiered emergency response mechanism, and weak scheduling targeting. The method achieves accurate early warning of intrusion behavior through multi-sensor deep feature fusion of visual and millimeter-wave radar sensors. A tiered intelligent agent linkage mechanism is designed to complete differentiated emergency responses. A multi-dimensional quantitative evaluation system is constructed to achieve closed-loop feedback of response effects and dynamic optimization of scheduling strategies. The method integrates the entire process of data processing, fusion early warning, intelligent agent scheduling, emergency execution, closed-loop evaluation, strategy optimization, and visual monitoring, improving the accuracy of oilfield perimeter intrusion early warning, the targeting of emergency responses, and the adaptability of scheduling strategies. It is also applicable to oilfield perimeter security scenarios, covering industrial protection scenarios such as oilfield mining areas, well sites, and oil and gas storage areas.

Description

Technical Field

[0001] This invention relates to the field of scheduling and emergency response technology, specifically to an intelligent agent closed-loop scheduling method and system for linking early warning and emergency response of oilfield perimeter intrusion. Background Technology

[0002] Perimeter security is a core aspect of safe production in oilfields, and intrusion early warning and emergency response are key processes in perimeter protection. However, existing technologies still have many fundamental shortcomings and are difficult to meet the security needs of complex industrial environments in oilfields. Insufficient multi-sensor data fusion leads to low early warning accuracy: Traditional oilfield perimeter security relies on independent visual, radar, and other sensor data, which are not deeply fused. It judges intrusion behavior based solely on data from a single sensor, making it susceptible to environmental interference and resulting in false alarms and missed alarms. Consequently, the accuracy and reliability of intrusion early warning are poor. The emergency response lacks a tiered mechanism and the dispatching is not targeted enough: the emergency response actions of the existing security system are simplistic and there is no differentiated dispatching strategy designed according to the level of intrusion risk. There is insufficient response in high-risk scenarios and waste of resources in low-risk scenarios, resulting in low accuracy and low resource utilization in emergency response. The intelligent agent lacks a hierarchical linkage design and has a low degree of automation: the security links such as early warning, scheduling, and execution are disconnected from each other, and an integrated intelligent agent linkage system has not been built. Manual intervention is required to complete the link connection, resulting in high delays in emergency response and failing to meet the real-time requirements of oilfield security. Without a closed-loop feedback optimization system, the strategy is fixed and rigid: after the emergency response is completed, the response effect is not quantitatively evaluated, the scheduling strategy is a fixed configuration, and it is impossible to dynamically optimize according to the actual response effect. The system has poor adaptability and iteration capability. In summary, existing oilfield perimeter security systems suffer from several problems, including insufficient multi-sensor data fusion, low early warning accuracy, lack of a tiered emergency response mechanism, weak scheduling targeting, lack of hierarchical linkage design for intelligent agents, low level of automation, absence of a closed-loop feedback optimization system, and rigid and inflexible strategies. Summary of the Invention

[0003] To overcome the problems of insufficient multi-sensor data fusion, low early warning accuracy, lack of hierarchical emergency response mechanism, weak scheduling targeting, lack of hierarchical linkage design of intelligent agents, low degree of automation, lack of closed-loop feedback optimization system, and fixed and rigid strategies in existing oilfield perimeter security technologies, this invention proposes an intelligent agent closed-loop scheduling method and system for linking oilfield perimeter intrusion early warning and emergency response. The aim is to integrate the features of multiple sensors, including vision and millimeter-wave radar, to construct an intrusion early warning model, achieve accurate identification and probability prediction of intrusion behavior at the oilfield perimeter, improve the accuracy of intrusion early warning, and reduce false alarm and missed alarm rates. The invention designs a three-tiered intelligent agent linkage mechanism of early warning, scheduling, and execution, and formulates differentiated scheduling strategies based on the intrusion risk level to achieve precise scheduling of emergency response and improve resource utilization. It establishes a multi-dimensional quantitative evaluation index system based on response latency, processing success rate, and resource consumption to achieve objective evaluation of emergency response effectiveness and provide data support for strategy optimization. A closed-loop optimization system for scheduling strategies driven by evaluation results is constructed, and a strategy optimization model is trained based on the quantitative evaluation results of emergency response to achieve dynamic adaptive adjustment of the scheduling strategy. To solve the above technical problems, this invention is achieved through the following technical solutions: Option 1: This invention proposes an intelligent agent closed-loop scheduling method for linking oilfield perimeter intrusion early warning and emergency response. The method includes the following steps: Step 1: Configure data parameters in the visualization interface, and construct dual-source sensor data for synthetic data generation and real data processing based on the selected data parameters; Step 2: Based on deep learning, feature extraction is performed on the dual-source sensor data, and the depth features of the visual sensor and the radar sensor are separated. The depth features of the visual sensor and the radar sensor are fused by feature stitching to train a multi-sensor feature fusion early warning model. Step 3: Input real-time dual-source sensor data into the trained multi-sensor feature fusion early warning model and output the intrusion probability P; based on the preset threshold, perform dynamic determination of the intrusion risk level and output intrusion probability and risk level early warning information. Step 4: Construct a hierarchical intelligent agent system consisting of an early warning intelligent agent, a scheduling intelligent agent, and an execution intelligent agent. The early warning intelligent agent is used to complete multi-sensor data processing and intrusion risk assessment, and pushes early warning information to the scheduling intelligent agent through a standardized communication protocol. The scheduling intelligent agent is used to receive early warning information, match the corresponding scheduling strategy based on the intrusion risk level, and issue standardized scheduling instructions to the target execution intelligent agent. The execution intelligent agent is used to complete the corresponding emergency response actions after receiving the instructions. Step 5: Collect the scheduling and execution results from Step 4, extract the scheduling start time, execution completion time, number of successful executions, and number of schedulings; calculate the three-dimensional quantitative evaluation indicators of response latency, processing success rate, and resource consumption; calculate the comprehensive score of emergency response based on the weighted formula to form a standardized evaluation result, and store the evaluation result in the evaluation dataset. Step 6: When the sample size of the evaluation dataset reaches the preset threshold, start the scheduling strategy optimization process; use the three-dimensional quantitative indicators of the evaluation dataset as input and the optimal scheduling weight as label to train the scheduling strategy optimization model; generate the scheduling weights of the execution agents under different risk levels through the optimization model, and update the hierarchical scheduling strategy according to the weights; synchronize the optimized scheduling strategy to the scheduling agent for emergency response scheduling.

[0004] Furthermore, a preferred embodiment is provided, wherein the method for dynamically determining the intrusion risk level based on a preset threshold and outputting intrusion probability and risk level warning information in step 3 is as follows:

[0005] Among them, 0.7 is the intrusion warning threshold, and 0.9 is the high-risk intrusion judgment threshold.

[0006] Furthermore, a preferred embodiment is provided in which the early warning agent, scheduling agent, and execution agent described in step 4 use the JSON standardized data format for cross-agent data transmission.

[0007] Furthermore, a preferred embodiment is provided, wherein the executing agent includes: The security notification execution intelligent agent is used to push early warning information containing intrusion location, risk level, and intrusion time to the preset security person in charge, and simultaneously push it to the oilfield security monitoring platform. The execution time is ≤200ms. The intelligent agent for executing audible and visual alarms is used to perform actions including triggering audible and visual alarm devices in the area corresponding to the intrusion point, continuously outputting warning lights and voice prompts, with an execution time of ≤100ms. The intelligent agent for closing oil pipeline valves is used to perform actions including issuing a closing command to the emergency shut-off valve of the corresponding oil pipeline, completing the valve closure and transmitting the valve position status back, with an execution time of ≤500ms.

[0008] Furthermore, a preferred embodiment is provided, wherein the method for calculating the three-dimensional quantitative evaluation indicators of response latency, processing success rate, and resource consumption in step 5 is as follows: Response latency:

[0009] Processing success rate:

[0010] Resource consumption:

[0011] in, The system timestamp for the first agent to complete a response. The system timestamp for initiating scheduling for the scheduling agent; The number of agents that successfully executed the scheduling. The total number of executing agents triggered by the scheduling; The total number of schedulable intelligent agents pre-configured for the system.

[0012] Furthermore, a preferred embodiment is provided, wherein the method for calculating the comprehensive emergency response score based on a weighted formula in step 5 to form a standardized evaluation result is as follows:

[0013]

[0014] in, The total number of schedulable intelligent agents pre-configured for the system.

[0015] Option 2: A closed-loop intelligent agent scheduling system for oilfield perimeter intrusion early warning and emergency response linkage, the system comprising: The dual-source sensor data processing module is used to configure data parameters in a visual interface and construct synthetic data generation and real data processing based on the selected data parameters. The multi-sensor feature fusion early warning module is used to extract features from the dual-source sensor data based on deep learning, separate the depth features of the visual sensor and the radar sensor, fuse the depth features of the visual sensor and the radar sensor through feature stitching, and train the multi-sensor feature fusion early warning model. The hierarchical intelligent agent linkage scheduling module is used to input real-time dual-source sensor data into the trained multi-sensor feature fusion early warning model and output the intrusion probability P; based on a preset threshold, it performs dynamic determination of the intrusion risk level and outputs intrusion probability and risk level early warning information. The emergency response execution module is used to construct a hierarchical intelligent agent system consisting of an early warning intelligent agent, a scheduling intelligent agent, and an execution intelligent agent. The early warning intelligent agent is used to complete multi-sensor data processing and intrusion risk assessment, and push early warning information to the scheduling intelligent agent through a standardized communication protocol. The scheduling intelligent agent is used to receive the early warning information, match the corresponding scheduling strategy based on the intrusion risk level, and issue standardized scheduling instructions to the target execution intelligent agent. The execution intelligent agent is used to complete the corresponding emergency response actions after receiving the instructions. The multi-dimensional closed-loop evaluation module is used to collect the scheduling and execution results from the emergency response execution module, extract scheduling start time, execution completion time, number of successful executions, and number of schedulings; and calculate three-dimensional quantitative evaluation indicators such as response delay, processing success rate, and resource consumption; calculate the comprehensive score of emergency response based on a weighted formula to form a standardized evaluation result, and store the evaluation result in the evaluation dataset. The scheduling strategy optimization module is used to initiate the scheduling strategy optimization process when the sample size of the evaluation dataset reaches a preset threshold. It trains the scheduling strategy optimization model using the three-dimensional quantitative indicators of the evaluation dataset as input and the optimal scheduling weight as the label. The optimization model generates the scheduling weights of the executing agents under different risk levels, and updates the hierarchical scheduling strategy according to the weights. The optimized scheduling strategy is then synchronized to the scheduling agents for emergency response scheduling.

[0016] Furthermore, a preferred embodiment is provided in which the system also includes an integrated visual monitoring module for one-click operation and real-time display of results for oilfield perimeter security.

[0017] Option 3: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Option 1.

[0018] Option 4: A computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method described in Option 1.

[0019] The advantages of this invention are: The present invention discloses an intelligent agent closed-loop scheduling method and system for linking oilfield perimeter intrusion early warning and emergency response. For the first time, it integrates the depth features of visual and millimeter-wave radar sensors to construct an intrusion probability prediction model, replacing the traditional single-sensor intrusion judgment mode. It effectively suppresses interference from complex oilfield environments such as rain, fog, strong light, and low illumination at night. Field tests show that the accuracy of intrusion early warning is improved by ≥20%, the false alarm rate is reduced by 80% in complex environments, and the missed alarm rate is reduced by 60%, solving the industry pain points of low early warning accuracy and high false alarm and missed alarm rates of traditional technologies.

[0020] The present invention discloses an intelligent agent closed-loop scheduling method and system for oilfield perimeter intrusion early warning and emergency response linkage. It proposes a hierarchical linkage mechanism of three-level intelligent agents: early warning, scheduling, and execution. That is, it designs a three-level intelligent agent system and clarifies the exclusive functional boundaries and standardized communication interaction protocols of each level to realize the automated linkage of early warning information push, scheduling strategy execution, and emergency response actions, eliminating the manual intervention link throughout the process. According to actual tests, compared with the traditional distributed multi-agent collaborative scheme, the emergency response delay is reduced by 30%, which solves the problems of low automation, disconnection of links, and high response delay of traditional technologies.

[0021] The present invention discloses an intelligent agent closed-loop scheduling method and system for linking oilfield perimeter intrusion early warning and emergency response. It proposes a differentiated scheduling strategy driven by intrusion risk level. Specifically, it divides the risk level into low, medium, and high based on the intrusion probability, designs differentiated execution agent combination scheduling rules for different risk levels, and dynamically selects the execution agent combination based on the intrusion risk level (low / medium / high) to achieve precise scheduling of emergency response. This avoids insufficient response in high-risk scenarios and prevents resource waste in low-risk scenarios. According to actual tests, the resource utilization rate is improved by more than 50%, and the overall resource utilization efficiency is improved by ≥60%, solving the problems of non-level response and low resource utilization rate of traditional technologies. The present invention discloses an intelligent agent closed-loop scheduling method and system for linking oilfield perimeter intrusion early warning and emergency response. It proposes a multi-dimensional quantitative emergency response closed-loop evaluation method, namely, establishing a three-dimensional quantitative evaluation index system of response delay, processing success rate, and resource consumption, clarifying the collection method, calculation rules and judgment criteria of each index, designing a weighted comprehensive score formula and clarifying the industry basis and experimental verification results of weight allocation, so as to achieve objective and accurate measurement of emergency response effect, provide quantifiable and traceable data basis for scheduling strategy optimization, and solve the problems of traditional technology lacking quantitative evaluation and evaluation basis. The present invention discloses an intelligent agent closed-loop scheduling method for linking oilfield perimeter intrusion early warning and emergency response, and a dynamic optimization mechanism for scheduling strategy in the system. Specifically, it constructs a deep learning strategy optimization model with evaluation indicators as input, and dynamically updates the scheduling strategy based on the actual evaluation results of emergency response. This enables the scheduling strategy to learn and optimize itself, making the strategy highly compatible with the actual security scenario of the oilfield. The overall effect of emergency response is improved by ≥15%, solving the problems of fixed scheduling strategies and poor adaptability in traditional technologies. The intelligent agent closed-loop scheduling method for oilfield perimeter intrusion early warning and emergency response linkage described in this invention, and the adaptive processing mechanism of dual-source sensor data in the system, construct a dual-source data processing mode including a synthetic data generation submodule and a real data parsing submodule. The synthetic data generation submodule can generate multi-sensor fusion data adapted to oilfield scenarios in batches based on preset parameters, effectively solving the problem of scarce real sensor data for oilfield perimeter security, providing sufficient data support for the training of fusion early warning models, greatly improving the generalization ability of the models, and solving the problems of poor generalization ability and low adaptability of traditional technology models.

[0022] The present invention describes an intelligent agent closed-loop scheduling method and system for oilfield perimeter intrusion early warning and emergency response linkage. This system employs a fully automated closed-loop architecture encompassing early warning, scheduling, execution, evaluation, and optimization. It organically combines multi-sensor fusion early warning, hierarchical intelligent agent linkage, emergency response execution, multi-dimensional quantitative evaluation, and scheduling strategy optimization to construct a fully automated closed loop. This enables data exchange and reverse optimization of results. Simultaneously, an integrated visual interface is built to display the entire process status and trace logs, solving the problems of fragmented processes and low operational efficiency in traditional technologies.

[0023] This invention is also applicable to perimeter security scenarios in oil fields, covering industrial protection scenarios such as oilfield mining areas, well sites, and oil and gas storage areas. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating an intelligent agent closed-loop scheduling method for linking oilfield perimeter intrusion early warning and emergency response, as described in Implementation Method 1.

[0025] Figure 2 This is a schematic diagram of the overall architecture of an intelligent agent closed-loop scheduling system for oilfield perimeter intrusion early warning and emergency response linkage as described in Implementation Method 2. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0027] Implementation Method 1, see Figure 1 This embodiment describes a closed-loop scheduling method for intelligent agents that links oilfield perimeter intrusion early warning and emergency response. The method specifically includes the following steps: S1: System Initialization and Parameter Configuration The visualization system is launched to complete the initialization of hardware and software modules and automatically adapt to the CPU / GPU hardware acceleration environment. Configure the system's core parameters in the visual interface, including intrusion warning threshold, risk level classification range, evaluation indicator weights, and strategy optimization training parameters; Initialize the multi-sensor fusion early warning model, hierarchical intelligent agent system, closed-loop evaluation model, and scheduling strategy optimization model to prepare for subsequent processes.

[0028] S2: Preparation and Loading of Dual-Source Sensor Data Select the data source through the visual interface: synthetic data or real data; If synthetic data is selected: set the number of samples, the system will generate batch synthetic data of visual + radar multi-sensor fusion with intrusion annotations, and output the synthetic dataset; If real data is selected: Select the real data file path, the system will automatically load and parse the data, complete the standardization and normalization processing, and output standardized real data; The processed dataset is allocated to the training module of the fusion early warning model to provide data support for model training.

[0029] S3: Training of Multi-Sensor Fusion Early Warning Model Feature extraction is performed on dual-source sensor data to separate visual features from radar features; A multi-sensor feature fusion early warning model is trained using visual and radar features as inputs and intrusion annotations as labels. After the model training is completed, the model is optimized and saved for subsequent real-time intrusion warnings.

[0030] S4: Real-time data acquisition and fusion early warning from multiple sensors Initiate real-time data acquisition using visual and radar sensors along the oilfield perimeter to obtain continuous multi-sensor fusion data; Input real-time sensor data into the trained fusion early warning model and output the intrusion probability P; The intrusion risk level is determined based on a preset threshold. If the risk level is medium / high, an early warning message is generated and pushed to the scheduling agent; if it is low, only the data is recorded and no scheduling is triggered.

[0031] S5: Hierarchical Intelligent Agent Coordination Scheduling and Emergency Response Execution After receiving the early warning information, the scheduling agent executes the current hierarchical scheduling strategy according to the intrusion risk level and determines the execution agent that needs to be triggered. The scheduling agent sends emergency response scheduling instructions to the target execution agent and records the scheduling start time. After receiving the instruction, the intelligent agent automatically executes the corresponding emergency response action, simulating an industrial-grade low-latency execution process. After the execution is completed, the execution result and execution time are recorded.

[0032] S6: Multi-dimensional quantitative evaluation of emergency response effectiveness The closed-loop evaluation module collects scheduling and execution results and extracts core data such as scheduling start time, execution completion time, number of successful executions, and number of scheduling events. Calculate three-dimensional quantitative evaluation indicators for response latency, processing success rate, and resource consumption based on core data; The comprehensive score of emergency response is calculated based on a weighted formula to form a standardized assessment result, which is then stored in the assessment dataset.

[0033] S7: Closed-Loop Optimization and Update of Scheduling Strategy When the sample size of the evaluation dataset reaches a preset threshold, the scheduling strategy optimization process is initiated. Using the three-dimensional quantitative indicators of the evaluation dataset as input and the optimal scheduling weight as the label, a scheduling strategy optimization model is trained. The scheduling weights of the executing agents under different risk levels are generated by optimizing the model, and the hierarchical scheduling strategy is updated according to the weights. The optimized scheduling strategy will be synchronized to the scheduling agent for subsequent emergency response scheduling.

[0034] S8: Visualization and logging of the entire process results In the corresponding area of ​​the visualization interface, sensor data, intrusion probability, risk level, agent status, emergency response results, three-dimensional evaluation indicators, and current scheduling strategy are displayed in real time. The system automatically records all information of this security process, forms a standardized operation log, and stores it locally; Operation and maintenance personnel can perform operations such as parameter modification, log query, data saving, and model update in the interface, realizing full-process visual interaction. The above steps are executed sequentially, data is exchanged, and reverse optimization is performed to form a complete security closed loop of early warning, scheduling, execution, evaluation, and optimization.

[0035] Implementation Method 2, see below Figure 2 This embodiment describes an intelligent agent closed-loop scheduling system for oilfield perimeter intrusion early warning and emergency response linkage. The system architecture uses visual and radar multi-sensor feature fusion as the early warning core, three-level intelligent agent hierarchical linkage as the scheduling core, and multi-dimensional evaluation and strategy optimization as the closed-loop core. It is supported by deep learning, industrial IoT, and visual interaction technologies, constructing a six-layer closed-loop architecture. Data exchange, result feedback, and reverse optimization are achieved between each layer. The specific architecture layers are as follows: Parameter configuration layer: Enables flexible setting and updating of intrusion warning thresholds, risk level classification standards, agent scheduling strategies, evaluation index weights, and strategy optimization parameters, and outputs personalized oilfield security configuration parameters; Data processing layer: Enables adaptive generation of synthetic sensor data and standardized loading of real sensor data, providing data support for training of fusion early warning models and real-time intrusion early warning; Multi-sensor fusion early warning layer: Based on deep learning, it extracts and fuses the depth features of visual and radar sensors to achieve probability prediction and risk level determination of intrusion behavior and output early warning information; Hierarchical intelligent agent linkage layer: Based on the intrusion risk level, the corresponding early warning, scheduling and execution intelligent agents are triggered to complete hierarchical linkage scheduling and output emergency response scheduling instructions; Closed-loop feedback evaluation and optimization layer: First, the emergency response effect is quantitatively evaluated in multiple dimensions. Then, the strategy optimization model is trained based on the evaluation results to realize the dynamic optimization of the scheduling strategy and output the optimized scheduling strategy.

[0036] The intelligent agent closed-loop scheduling system for oilfield perimeter intrusion early warning and emergency response linkage described in this embodiment includes a dual-source sensor data processing module, a multi-sensor feature fusion early warning module, a hierarchical intelligent agent linkage scheduling module, an emergency response execution module, a multi-dimensional closed-loop evaluation module, and a scheduling strategy optimization module.

[0037] I. The dual-source sensor data processing module includes a synthetic data generation submodule and a real data parsing submodule, constructing a dual-source data supply mechanism for synthetic and real sensor data. The synthetic data generation submodule randomly generates visual and radar features based on preset parameters, generating fused data from visual sensors and millimeter-wave radar sensors by simulating intrusion / non-intrusion scenarios at the oilfield perimeter, with intrusion annotation information. The real data parsing submodule supports loading and parsing in standardized formats, achieving seamless switching between the two data sources and providing sufficient data support for fusion early warning model training and real-time early warning. This dual-source sensor data processing module addresses the scarcity of real sensor data for oilfield perimeter security. Synthetic data can be batch-simulated using multi-sensor fusion data under different intrusion scenarios and environments. Real data achieves automated parsing and feature extraction, improving data processing efficiency by ≥50%. The synthetic data generation submodule: presets the acquisition parameters of the oilfield perimeter visual sensors and millimeter-wave radar sensors, including the frame rate and detection box output format of the visual sensors, and the detection distance, distance resolution, and speed measurement range of the millimeter-wave radar; randomly generates intrusion / non-intrusion data based on preset parameters. The system automatically labels intrusion status using visual features (detection box coordinates, target confidence, target classification type) and radar features (target distance, radial velocity, azimuth angle, echo signal strength) in non-intrusive scenarios, outputting a multi-sensor fusion dataset with complete annotations. The real data parsing submodule supports loading real visual sensor and millimeter-wave radar sensor data in mainstream formats such as CSV and JSON, automatically parsing sensor feature information, performing data cleaning, outlier removal, standardization, and normalization, and outputting standardized real data adapted to the input format of the fusion early warning model.

[0038] II. The multi-sensor feature fusion early warning module extracts depth features from visual and radar sensors based on deep learning, achieves multi-sensor feature deep fusion through feature stitching, constructs a binary classification model to output intrusion probability, and then performs dynamic judgment of intrusion risk level based on preset threshold, outputting intrusion probability and risk level early warning information to replace the traditional single-sensor intrusion judgment mode. It realizes the fusion of depth features from visual sensors and millimeter-wave radar sensors, effectively suppressing interference from complex oilfield environments such as rain, fog, strong light, and low illumination at night. According to field tests in oilfields, compared with the early warning scheme of a single visual sensor, the accuracy of intrusion early warning of this invention is improved by ≥20%, the false alarm rate is reduced by 80% in complex environments, and the missed alarm rate is reduced by 60%. Among them, the intrusion probability prediction: the fusion model processes the visual and radar depth features and outputs the intrusion probability P between 0 and 1. The larger the P value, the higher the intrusion probability. Risk Level Classification: Based on intrusion warning thresholds and probability ranges, intrusion risk levels are divided into three levels: low, medium, and high. The determination rules are as follows:

[0039] Among them, 0.7 is the core threshold for intrusion warning, and 0.9 is the threshold for high-risk intrusion judgment, which can be flexibly adjusted at the parameter configuration layer.

[0040] III. The hierarchical intelligent agent linkage scheduling module constructs a hierarchical, loosely coupled hierarchical intelligent agent system consisting of an early warning intelligent agent (Level 1), a scheduling intelligent agent (Level 2), and an execution intelligent agent (Level 3). The functional boundaries and communication interaction protocols of each level of intelligent agent are clearly defined. The early warning intelligent agent completes multi-sensor data processing and intrusion risk assessment, and pushes early warning information to the scheduling intelligent agent through a standardized communication protocol. After receiving the early warning information, the scheduling intelligent agent matches the corresponding scheduling strategy based on the intrusion risk level and issues standardized scheduling instructions to the target execution intelligent agent. After receiving the instructions, the execution intelligent agent completes the corresponding emergency response action, achieving fully automated, low-latency linkage of the three levels of intelligent agents. The hierarchical intelligent agent linkage scheduling module described in this embodiment replaces the traditional multi-agent collaborative design without hierarchy and clear interaction protocols, realizing hierarchical closed-loop linkage of early warning, scheduling, and execution, eliminating manual intervention throughout the process, and improving the automation level of emergency response scheduling by 100%. Field tests in oilfields show that compared to traditional distributed multi-agent collaborative schemes, the emergency response latency of this invention is reduced by 30%; the differentiated scheduling strategy based on risk levels improves resource utilization by more than 50%, and the overall resource utilization efficiency is improved by ≥60%.

[0041] The first-level early warning agent is used for multi-sensor data preprocessing, intrusion feature extraction, intrusion probability prediction, and intrusion risk level determination. After completing data processing, it only pushes standardized early warning information to the second-level scheduling agent, without directly interfering with the scheduling and execution stages, thus decoupling perception and decision-making. The second-level scheduling agent is used for receiving early warning information, risk level matching, scheduling strategy execution, instruction issuance, and execution status recovery. As the core layer connecting the upper and lower levels, it only receives early warning information from the early warning agent and only issues scheduling instructions to the execution agent, without participating in front-end perception and end-end execution operations, thus decoupling decision-making and execution. The third-level execution agent is used for receiving scheduling instructions, executing emergency response actions, and providing execution status feedback. It only receives standardized instructions from the scheduling agent, completes the corresponding actions, and sends the execution results back to the scheduling agent, without participating in the perception and decision-making stages, thus achieving independent control of the execution stage.

[0042] Each agent uses the JSON standardized data format for cross-agent data transmission, which features scalable fields and strong compatibility, and is compatible with the communication standards of the oilfield industrial Internet of Things. Warning information push protocol: The warning information pushed by the warning agent to the scheduling agent includes five core fields: timestamp, intrusion probability, risk level, intrusion target characteristics, and sensor location number. The trigger condition is that the intrusion risk level is determined to be medium or high risk; low-risk events are only recorded in local logs and do not trigger warning information push. Scheduling instruction issuance protocol: The scheduling instructions issued by the scheduling agent to the execution agent contain five core fields: unique instruction ID, timestamp, target execution agent number, emergency action type, and execution time requirement. The trigger condition is that the scheduling agent completes the matching of risk level and scheduling strategy. Execution Status Feedback Protocol: The execution results fed back by the executing agent to the scheduling agent contain five core fields: unique ID of the corresponding instruction, timestamp, execution status, execution completion time, and exception information (if any). The triggering condition is that the executing agent completes an emergency action or determines that the execution has failed.

[0043] The core logic of the hierarchical strategy is as follows: dynamically select the combination of execution agents based on the intrusion risk level (low / medium / high), preset the execution agent triggering rules corresponding to different risk levels, and the scheduling combination can be dynamically updated and adaptively adjusted through the strategy optimization module. Basic hierarchical scheduling execution rules: Low risk level (P<0.7): 0 execution agents are triggered, only data recording and log storage are completed, and no emergency response actions are triggered; Medium risk level (0.7≤P<0.9): Triggers 2 execution agents, specifically a security notification execution agent and an audible and visual alarm execution agent, to complete the synchronization of on-site warnings and security personnel information; High-risk level (P≥0.9): Triggers all 3 execution agents, specifically the security notification execution agent, the audible and visual alarm execution agent, and the oil valve closing execution agent, to simultaneously complete on-site warnings, personnel notifications, and production safety emergency response.

[0044] IV. The emergency response execution module is configured with multiple types of execution agents. Each execution agent corresponds to an emergency response action for oilfield perimeter security. After receiving instructions from the scheduling agent, it automatically executes the response action, simulating industrial-grade low-latency execution. It records response results and execution time, providing data support for closed-loop evaluation. Core improvements: It achieves automated execution of emergency response actions, simulating industrial-grade low-latency communication, with response execution latency controlled within 100-500ms, meeting the real-time requirements of oilfield security; and automatically records execution results, providing accurate raw data for closed-loop evaluation. Execution agent types and execution sequence: Three core execution agents are configured: security notification execution agent, audible and visual alarm execution agent, and oil valve closure execution agent, covering the entire emergency response scenario for oilfield perimeter security; each execution agent starts execution synchronously after receiving the dispatch command, among which: Security notification execution intelligent agent: The execution actions include pushing early warning information containing intrusion location, risk level and intrusion time to the preset security person in charge, and simultaneously pushing it to the oilfield security monitoring platform, with an execution time requirement of ≤200ms; Audible and visual alarm execution agent: The execution actions include triggering the audible and visual alarm devices in the area corresponding to the intrusion point, continuously outputting warning lights and voice prompts, and the execution time requirement is ≤100ms; Oil valve closure execution agent: The execution actions include issuing a closure command to the emergency shut-off valve of the oil pipeline in the corresponding area, completing the valve closure and returning the valve position status, with an execution time requirement of ≤500ms; Execution result recording and feedback: After each execution agent completes its execution, it automatically records the unique instruction ID, execution identifier, response action type, execution start time, execution completion time, execution status (complete / failed), and exception information (if any), forming a standardized execution result dataset. At the same time, it feeds back the execution results to the scheduling agent according to the standardized communication protocol.

[0045] V. Multi-dimensional Closed-Loop Evaluation Module: This module establishes a three-dimensional quantitative evaluation index system based on response delay, processing success rate, and resource consumption. It clarifies the data collection methods, calculation rules, and judgment criteria for each index. The module calculates the values ​​of each index based on dispatch command records and execution results, and obtains a comprehensive emergency response score through weighted summation. This achieves an objective, full-process quantitative evaluation of the response effect, providing traceable and quantifiable core input for dispatch strategy optimization. This is the first time a multi-dimensional quantitative evaluation system for emergency response at the oilfield perimeter has been established, replacing the traditional qualitative evaluation model and achieving accurate measurement of response effectiveness, providing quantifiable data for strategy optimization. The calculation methods for response latency, processing success rate, resource consumption, and the overall emergency response score are as follows: 1. Response delay ( (Unit: ms): The time difference between the start of the scheduling agent and the completion of the response action by the first executing agent, reflecting the real-time nature of the emergency response; Timestamp collection method: The scheduling start timestamp is taken from the system time when the scheduling agent issues the scheduling instruction to the first executing agent, and the execution completion timestamp is taken from the system time when the first executing agent sends the execution completion status back to the scheduling agent. The timestamp precision is in milliseconds.

[0046] in, The system timestamp for the first agent to complete a response. The system timestamp for initiating scheduling for the scheduling agent.

[0047] 2. Processing success rate ( (Unit: %): The ratio of the number of agents that successfully executed the emergency response to the total number of agents triggered in this scheduling, reflecting the effectiveness of the emergency response action; Success criteria: If the executing agent completes the preset emergency action within the time limit required by the instruction and sends back the "execution completed" status to the scheduling agent, it is considered as successful execution; if it fails to complete the action within the time limit, sends back the "execution failed" status, or sends back no status, it is considered as failed execution.

[0048] in, This represents the number of agents that successfully executed the task in this scheduling process. This represents the total number of executing agents triggered by this scheduling. 3. Resource consumption The ratio of the number of executing agents triggered in this scheduling to the total number of executing agents configured in the system reflects the resource consumption and utilization rate of the emergency response.

[0049] in, The total number of schedulable intelligent agents pre-configured for the system; 4. Emergency Response Overall Score (S, value 0~1): Based on the weighted calculation of each indicator, it comprehensively reflects the overall effectiveness of the emergency response. The weights can be flexibly adjusted at the parameter configuration level. Basic weight allocation: Success rate weight 0.6, response latency weight 0.2, resource consumption weight 0.2; Weighting criteria: Based on the core industry needs of oilfield perimeter security, the successful execution of emergency response actions is crucial to ensuring oilfield production safety; therefore, the success rate is assigned the highest weight. Oilfield industrial scenarios have high real-time requirements for responding to intrusion events; therefore, response delay is assigned a secondary weight. Resource consumption is an auxiliary optimization indicator; while ensuring safe handling, resource utilization efficiency is also considered, thus it is assigned a secondary weight equal to response delay. This weighting allocation has been verified through multi-scenario field testing in oilfields and can achieve an optimal balance between security and resource utilization. .

[0050] VI. The scheduling strategy optimization module uses the three-dimensional quantitative indicators of closed-loop evaluation as input to construct a deep learning strategy optimization model. Through model training, the scheduling weights of the executing agents under different intrusion risk levels are obtained. Executing agents are selected from high to low weights to form an optimized hierarchical scheduling strategy, achieving dynamic adaptive adjustment of the scheduling strategy. This replaces the traditional fixed scheduling strategy design, constructing an evaluation result-driven closed-loop optimization system to achieve self-learning and self-optimization of the scheduling strategy, making the scheduling strategy highly compatible with the actual security scenario of the oilfield, and improving the overall emergency response effect by ≥15%. Using evaluation indicators (response latency / 1000, processing success rate, resource consumption) as input features and the optimal scheduling weights under different risk levels as labels, a fully connected neural network model is trained, outputting the scheduling weights of the executing agents (values ​​from 0 to 1). For low, medium, and high intrusion risk levels, the optimization model generates scheduling weights for the executing agents respectively. The corresponding number of executing agents are selected from high to low weights to form a new hierarchical scheduling strategy, replacing the original strategy.

[0051] VII. Integrated Visual Monitoring Module: By building a visual interactive interface, it integrates the entire process functions such as parameter configuration, data management, model training, fusion early warning, agent status, emergency response, closed-loop evaluation, and log recording. It realizes one-click operation and real-time display of results for oilfield perimeter security, while automatically recording all operation and logs, and supporting local data storage and traceability. Core improvements: Addressing the fragmented processes of traditional security systems by achieving integrated functionality across the entire workflow; providing a visual display of the system's full status, with intuitive operation and reduced operational costs for maintenance personnel by ≥60%; automating log recording to resolve the issue of untraceable early warning and scheduling data. Interactive configuration supports visual modification and updates of early warning thresholds, risk level ranges, evaluation indicator weights, and strategy optimization parameters; real-time display synchronously shows sensor data, intrusion probability, risk level, agent status, emergency response results, 3D evaluation indicators, and scheduling strategies; automatic logging of timestamps, sensor data, intrusion probability, risk level, scheduling results, execution results, evaluation indicators, and comprehensive scores, supporting log querying, clearing, and local saving; and integrated visualization of synthetic data generation, real data loading, and fusion early warning model training, enabling model training and updates without requiring specialized coding.

[0052] In summary, the intelligent agent closed-loop scheduling method and system for oilfield perimeter intrusion early warning and emergency response linkage described in this embodiment is the first to fuse visual and millimeter-wave radar sensor depth features to construct an intrusion probability prediction model, replacing the traditional single-sensor intrusion judgment mode. It effectively suppresses interference from complex oilfield environments such as rain, fog, strong light, and low illumination at night. Field tests show that the accuracy of intrusion early warning is improved by ≥20%, the false alarm rate is reduced by 80% in complex environments, and the missed alarm rate is reduced by 60%, solving the industry pain points of low early warning accuracy and high false alarm and missed alarm rates of traditional technologies.

[0053] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. This is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0054] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A closed-loop scheduling method for intelligent agents linking early warning and emergency response for oilfield perimeter intrusion, characterized in that, The method includes the following steps: Step 1: Configure data parameters in the visualization interface, and construct dual-source sensor data for synthetic data generation and real data processing based on the selected data parameters; Step 2: Based on deep learning, feature extraction is performed on the dual-source sensor data, and the depth features of the visual sensor and the radar sensor are separated. The depth features of the visual sensor and the radar sensor are fused by feature stitching to train a multi-sensor feature fusion early warning model. Step 3: Input real-time dual-source sensor data into the trained multi-sensor feature fusion early warning model and output the intrusion probability P; based on the preset threshold, perform dynamic determination of the intrusion risk level and output intrusion probability and risk level early warning information. Step 4: Construct a hierarchical intelligent agent system consisting of an early warning intelligent agent, a scheduling intelligent agent, and an execution intelligent agent. The early warning intelligent agent is used to complete multi-sensor data processing and intrusion risk assessment, and pushes early warning information to the scheduling intelligent agent through a standardized communication protocol. The scheduling intelligent agent is used to receive early warning information, match the corresponding scheduling strategy based on the intrusion risk level, and issue standardized scheduling instructions to the target execution intelligent agent. The execution intelligent agent is used to complete the corresponding emergency response actions after receiving the instructions. Step 5: Collect the scheduling and execution results from Step 4, extract the scheduling start time, execution completion time, number of successful executions, and number of schedulings; calculate the three-dimensional quantitative evaluation indicators of response latency, processing success rate, and resource consumption; calculate the comprehensive score of emergency response based on the weighted formula to form a standardized evaluation result, and store the evaluation result in the evaluation dataset. Step 6: When the sample size of the evaluation dataset reaches the preset threshold, start the scheduling strategy optimization process; use the three-dimensional quantitative indicators of the evaluation dataset as input and the optimal scheduling weight as label to train the scheduling strategy optimization model; generate the scheduling weights of the execution agents under different risk levels through the optimization model, and update the hierarchical scheduling strategy according to the weights; synchronize the optimized scheduling strategy to the scheduling agent for emergency response scheduling.

2. The intelligent agent closed-loop scheduling method for oilfield perimeter intrusion early warning and emergency response linkage according to claim 1, characterized in that, Step 3 involves dynamically determining the intrusion risk level based on a preset threshold and outputting intrusion probability and risk level warning information as follows: Among them, 0.7 is the intrusion warning threshold, and 0.9 is the high-risk intrusion judgment threshold.

3. The intelligent agent closed-loop scheduling method for oilfield perimeter intrusion early warning and emergency response linkage according to claim 1, characterized in that, In step 4, the early warning agent, scheduling agent, and execution agent use the JSON standardized data format for cross-agent data transmission.

4. The intelligent agent closed-loop scheduling method for oilfield perimeter intrusion early warning and emergency response linkage according to claim 1, characterized in that, The executing intelligent agent includes: The security notification execution intelligent agent is used to push early warning information containing intrusion location, risk level, and intrusion time to the preset security person in charge, and simultaneously push it to the oilfield security monitoring platform. The execution time is ≤200ms. The intelligent agent for executing audible and visual alarms is used to perform actions including triggering audible and visual alarm devices in the area corresponding to the intrusion point, continuously outputting warning lights and voice prompts, with an execution time of ≤100ms. The intelligent agent for closing oil pipeline valves is used to perform actions including issuing a closing command to the emergency shut-off valve of the corresponding oil pipeline, completing the valve closure and transmitting the valve position status back, with an execution time of ≤500ms.

5. The intelligent agent closed-loop scheduling method for oilfield perimeter intrusion early warning and emergency response linkage according to claim 1, characterized in that, The method for calculating the three-dimensional quantitative evaluation indicators of response latency, processing success rate, and resource consumption in step 5 is as follows: Response latency: Processing success rate: Resource consumption: in, The system timestamp for the first agent to complete a response. The system timestamp for initiating scheduling for the scheduling agent; The number of agents that successfully executed the scheduling. The total number of executing agents triggered by the scheduling; The total number of schedulable intelligent agents pre-configured for the system.

6. The intelligent agent closed-loop scheduling method for oilfield perimeter intrusion early warning and emergency response linkage according to claim 5, characterized in that, The method for calculating the comprehensive emergency response score based on the weighted formula in step 5 to form a standardized evaluation result is as follows: in, The total number of schedulable intelligent agents pre-configured for the system.

7. A closed-loop intelligent agent scheduling system for oilfield perimeter intrusion early warning and emergency response linkage, characterized in that, The system includes: The dual-source sensor data processing module is used to configure data parameters in a visual interface and construct synthetic data generation and real data processing based on the selected data parameters. The multi-sensor feature fusion early warning module is used to extract features from the dual-source sensor data based on deep learning, separate the depth features of the visual sensor and the radar sensor, fuse the depth features of the visual sensor and the radar sensor through feature stitching, and train the multi-sensor feature fusion early warning model. The hierarchical intelligent agent linkage scheduling module is used to input real-time dual-source sensor data into the trained multi-sensor feature fusion early warning model and output the intrusion probability P; based on a preset threshold, it performs dynamic determination of the intrusion risk level and outputs intrusion probability and risk level early warning information. The emergency response execution module is used to construct a hierarchical intelligent agent system consisting of an early warning intelligent agent, a scheduling intelligent agent, and an execution intelligent agent. The early warning intelligent agent is used to complete multi-sensor data processing and intrusion risk assessment, and push early warning information to the scheduling intelligent agent through a standardized communication protocol. The scheduling intelligent agent is used to receive the early warning information, match the corresponding scheduling strategy based on the intrusion risk level, and issue standardized scheduling instructions to the target execution intelligent agent. The execution intelligent agent is used to complete the corresponding emergency response actions after receiving the instructions. The multi-dimensional closed-loop evaluation module is used to collect the scheduling and execution results from the emergency response execution module, extract scheduling start time, execution completion time, number of successful executions, and number of schedulings; and calculate three-dimensional quantitative evaluation indicators such as response delay, processing success rate, and resource consumption; calculate the comprehensive score of emergency response based on a weighted formula to form a standardized evaluation result, and store the evaluation result in the evaluation dataset. The scheduling strategy optimization module is used to initiate the scheduling strategy optimization process when the sample size of the evaluation dataset reaches a preset threshold. It trains the scheduling strategy optimization model using the three-dimensional quantitative indicators of the evaluation dataset as input and the optimal scheduling weight as the label. The optimization model generates the scheduling weights of the executing agents under different risk levels, and updates the hierarchical scheduling strategy according to the weights. The optimized scheduling strategy is then synchronized to the scheduling agents for emergency response scheduling.

8. The intelligent agent closed-loop scheduling system for oilfield perimeter intrusion early warning and emergency response linkage according to claim 7, characterized in that, The system also includes an integrated visual monitoring module for one-click operation and real-time display of results for oilfield perimeter security.

9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-6.

10. A computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method of any one of claims 1-6.

Images