A dynamic self-adaptive oilfield electronic fence early warning area intelligent adjustment method and system

Abstract

A dynamic and adaptive intelligent adjustment method and system for early warning areas of oilfield electronic fences is disclosed, relating to the field of safety production monitoring in the petroleum industry. This invention addresses the core pain points of existing fixed electronic fences, such as poor adaptability, high false alarm rates, and inability to dynamically adjust according to personnel activity patterns and environmental changes. The method includes: constructing a multi-regional hierarchical early warning mechanism for oilfields to achieve differentiated early warning rules for different functional areas, time periods, and environments; designing an LSTM model for mining the spatiotemporal patterns of personnel activity, integrating multi-dimensional features such as personnel coordinates, time periods, and weather; proposing an intelligent fence adjustment algorithm based on behavioral prediction and dynamic adjustment of environmental factors; constructing a false alarm suppression mechanism to filter early warning results; and building a closed-loop system to support seamless switching between dual-source data. This invention is also applicable to other application areas such as oilfield field needs.

Description

Technical Field

[0001] This invention relates to the field of safety production monitoring technology in the petroleum industry, specifically to a dynamic adaptive method and system for intelligent adjustment of early warning areas in oilfield electronic fences. Background Technology

[0002] Oilfield field operations are characterized by dispersed areas, complex environments, significant differences in work periods, and high safety risks. Electronic fencing, as a core technology for ensuring the personal safety of workers, can provide early warnings of boundary crossing risks through personnel location monitoring, making it a crucial element in oilfield safety production management. However, existing electronic fencing technology exhibits a significant mismatch with the actual needs of oilfield operations. The core technological bottlenecks are as follows: Fixed fences cannot adapt to the diverse temporal and spatial patterns of oilfield operations: Oilfield sites are divided into different functional areas such as core production areas, daily operation areas, and rest and living areas. The safety control requirements for different areas change significantly throughout the day and night. Traditional fixed electronic fences use static boundaries and fixed early warning rules, which cannot adapt to the differentiated control needs of "normal daytime activities in the work area, requiring early warning at night, and normal nighttime activities in the rest area, requiring early warning during the day." This easily leads to a large number of invalid false alarms, causing workers to become desensitized to early warning information, and also poses a risk of missed alarms in critical scenarios.

[0003] The warning range cannot adaptively adjust to changes in the field environment: Oilfield field operations often face severe weather conditions such as sandstorms, heavy rain, and strong winds. Under such conditions, the safety risks for personnel approaching the core production area are significantly increased, necessitating the expansion of the warning range for advance control. However, traditional fixed fences cannot dynamically adjust the warning boundaries according to environmental factors, easily leading to safety hazards due to untimely warnings in severe weather.

[0004] The early warning mechanism lacks in-depth analysis of the patterns of human activity: the existing electronic fence only makes static judgments based on whether the current coordinates of the personnel have crossed the boundary, without considering the temporal characteristics and behavioral trends of human activities. It is impossible to predict the risk behavior of personnel through historical activity patterns, resulting in a strong lag in early warning and the inability to provide data support for the dynamic adjustment of the fence boundary.

[0005] Insufficient false alarm suppression capability and poor on-site practicality: The existing early warning mechanism does not have filtering logic for normal activity areas. It is easy to trigger false alarms for normal activities of workers in compliant areas. Indiscriminate early warning leads to a significant decrease in the trust of on-site personnel in early warning information, which seriously affects the effectiveness of safety production control. Summary of the Invention

[0006] This invention addresses the core shortcomings of existing fixed electronic fences, such as poor adaptability, high false alarm rate, and inability to dynamically adjust to changes in personnel activity patterns and the environment. To this end, this invention proposes a dynamically adaptive intelligent adjustment method and system for early warning areas of oilfield electronic fences. The invention achieves these technical problems through the following technical solutions: Option 1: This invention proposes a dynamic adaptive intelligent adjustment method for early warning areas of oilfield electronic fences, the method comprising the following steps: Step 1: Configure core parameters through the visual interface. The core parameters include functional area boundary coordinates, time series length, model hyperparameters, fence scaling factor, and false alarm suppression threshold. Step 2: Select synthetic data or real data through the visual interface, convert the preprocessed synthetic data or real data into time series feature sequences, divide them into training set and test set in an 8:2 ratio, and build a standardized data loader. Step 3: Construct a multi-layer LSTM temporal feature extraction network, build a fully connected classification layer, and complete the construction and initialization of the model for mining the spatiotemporal patterns of human activities; Step 4: Input the training set into the spatiotemporal pattern mining model of personnel activities constructed in Step 3 to complete the forward propagation calculation and output the early warning prediction result; calculate the cross-entropy loss based on the prediction result and the real label, and update the model weight parameters through the backpropagation algorithm; iterate the training to the preset number of rounds and save the optimal model after training. Step 5: Load the latest personnel activity time series data, input the trained optimal model, and output the personnel behavior risk prediction results; calculate the fence center, width, and height parameters based on the current fence boundary coordinates; determine the scaling factor based on the model prediction results according to the spatiotemporal patterns of personnel activities, calculate the adjusted fence vertex coordinates, and complete the intelligent adjustment of the electronic fence warning area; Step 6: Perform batch inference on all personnel activity data to generate initial warning results; based on the preset normal activity area boundaries, execute a false alarm suppression algorithm to perform secondary filtering on the initial warning results; count the final number of warnings, warning locations, and warning time periods to generate a warning control report; Step 7: Draw the personnel activity trajectory, adjusted electronic fence boundaries, normal activity areas, and warning points on the visualization interface, and automatically record the system operation log.

[0007] Furthermore, a preferred embodiment is provided, wherein the method for converting the preprocessed synthetic data or real data into a time-series feature sequence in step 2 is as follows:

[0008]

[0009] In the formula, For the i-th time series input sequence, Let be the feature vector at time step t. For sequence length, The warning label is the one corresponding to the sequence.

[0010] Furthermore, a preferred implementation method is provided, wherein the method for constructing and initializing the multi-layer LSTM temporal feature extraction network and building a fully connected classification layer in step 3 to complete the construction and initialization of the spatiotemporal pattern mining model of personnel activities is as follows: Step 3.1: Encode the input time series sequence using a multi-layer LSTM network to capture long-range temporal dependencies of personnel activities. The calculation method is as follows:

[0011]

[0012]

[0013]

[0014]

[0015]

[0016] In the formula, These are the input gate, forget gate, and output gate, respectively. In cellular state, In hidden state, For the input features at time step t, It is the Sigmoid activation function. For Hadamah accumulation; Step 3.2: Take the hidden state of the last time step of the time series and complete the final warning classification through a fully connected layer. The calculation method is as follows:

[0017] In the formula, This represents the hidden state at the last time step of the sequence. It is a fully connected layer. This refers to the early warning prediction results output by the model.

[0018] Furthermore, a preferred embodiment is provided, wherein the method for calculating the fence center, width, and height parameters based on the current fence boundary coordinates in step 5 is as follows:

[0019]

[0020]

[0021] In the formula, Let n be the coordinates of the i-th vertex of the current fence, and n be the number of fence vertices. Here are the coordinates of the fence center, W is the fence width, and H is the fence height.

[0022] Furthermore, a preferred embodiment is provided, wherein in step 5, the scaling factor is determined based on the prediction results of the human activity spatiotemporal pattern mining model, and the adjusted fence vertex coordinates are calculated to complete the intelligent adjustment of the electronic fence early warning area.

[0023] In the formula, S is the fence scaling factor; Calculate the adjusted fence vertex coordinates:

[0024]

[0025]

[0026]

[0027] In the formula, Let be the coordinates of the i-th vertex of the fence after adjustment.

[0028] Furthermore, a preferred embodiment is provided, wherein the method for performing secondary filtering of the initial warning results in step 6 based on a preset normal activity area boundary and executing a false alarm suppression algorithm is as follows:

[0029] In the formula, The warning labels predicted by the model. This is the final warning label after false alarm suppression. The current coordinates of the personnel. This is a pre-defined set of normal activity areas.

[0030] Furthermore, a preferred embodiment is provided, in which step 6 further includes using the cross-entropy loss function as the model optimization objective, the Adam optimizer performing iterative parameter updates, and minimizing the model prediction loss through multiple rounds of iterative training, i.e.

[0031] In the formula, N is the sample size. For real labels, Predict probabilities for the model.

[0032] Option 2: A dynamically adaptive intelligent adjustment system for early warning areas of oilfield electronic fences, the system comprising: The dual-source data processing module is used to configure core parameters through a visual interface. These core parameters include functional area boundary coordinates, time series length, model hyperparameters, fence scaling factor, and false alarm suppression threshold. The module for mining spatiotemporal patterns of human activities is used to select synthetic or real data through a visual interface, convert the preprocessed synthetic or real data into a time-series feature sequence, divide it into training and test sets in an 8:2 ratio, and build a standardized data loader. The model building module is used to construct a multi-layer LSTM temporal feature extraction network, build a fully connected classification layer, and complete the construction and initialization of the model for mining the spatiotemporal patterns of human activities. The graded early warning rules and false alarm suppression module is used to input the training set into the spatiotemporal pattern mining model of personnel activities built by the model building module to complete the forward propagation calculation and output the early warning prediction result; calculate the cross-entropy loss based on the prediction result and the true label, update the model weight parameters through the backpropagation algorithm; iterate the training to the preset number of rounds and save the optimal model after training. The dynamic adaptive fence area intelligent adjustment module is used to load the latest human activity time series data, input the trained optimal model, and output the human behavior risk prediction results; based on the current fence boundary coordinates, it calculates the fence center, width, and height parameters; based on the spatiotemporal patterns of human activity, it mines the model prediction results to determine the scaling factor, calculates the adjusted fence vertex coordinates, and completes the intelligent adjustment of the electronic fence warning area; The model inference and false alarm suppression module is used to perform batch inference on all personnel activity data and generate initial warning results; based on the preset normal activity area boundaries, it executes a false alarm suppression algorithm to perform secondary filtering on the initial warning results; and it counts the final number of warnings, warning locations, and warning time periods to generate a warning control report. The visual interactive management module is used to draw personnel activity trajectories, adjusted electronic fence boundaries, normal activity areas, and early warning points on a visual interface, and automatically record system operation logs.

[0033] 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.

[0034] 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.

[0035] The advantages of this invention are: The present invention discloses a dynamic adaptive method and system for intelligent adjustment of early warning areas for oilfield electronic fences. It proposes a multi-regional hierarchical differentiated early warning mechanism adapted to the spatiotemporal patterns of oilfield operations. Targeting the different functional attributes of the core production area, work area, and rest area of ​​the oilfield, it constructs hierarchical early warning rules that link time period, area, and environment. This achieves refined early warning for "all-time control of the core area, differentiated day and night control of the work area, differentiated day and night control of the rest area, and expanded control in severe weather." It fundamentally solves the problem of false alarms and missed alarms caused by the inability of traditional fixed electronic fences to adapt to the day and night activity patterns of oilfield operations, and achieves accurate early warning in different scenarios.

[0036] The present invention discloses a dynamic adaptive method and system for intelligent adjustment of early warning areas of oilfield electronic fences. It designs a spatiotemporal sequence mining model of oilfield personnel activities based on LSTM. Targeting the temporal dependence characteristics of oilfield personnel activities, it constructs an LSTM deep network that integrates multi-dimensional features such as personnel coordinates, time periods, and weather environment. This accurately mines the spatiotemporal patterns and behavioral trends of personnel activities, enabling early prediction of risky behaviors in the future. This breaks through the limitation of traditional fences that rely solely on static coordinate judgment and provides a reliable decision-making basis for the dynamic adjustment of electronic fences.

[0037] The present invention discloses a dynamic adaptive intelligent adjustment method and system for early warning areas of oilfield electronic fences. It proposes a dynamic adaptive intelligent adjustment algorithm for electronic fence areas. Based on the prediction results of personnel behavior risks, it constructs a proportional scaling adjustment mechanism with the fence center as the reference. When an early warning risk is predicted, the early warning range is expanded by 20% to achieve early control. When there is no risk, the control range is reduced by 10% to reduce interference with normal operations. This realizes the dynamic adaptive adjustment of the early warning area according to personnel activity patterns and environmental factors, taking into account both oilfield safety production control and on-site operation convenience.

[0038] The present invention describes a dynamic adaptive method and system for intelligent adjustment of early warning areas in oilfield electronic fences. It constructs an early warning false alarm suppression mechanism based on normal activity areas. By presetting the boundary range of normal operation and rest areas in the oilfield, the early warning results predicted by the model are filtered twice to eliminate invalid early warnings in normal activity areas. Under the premise of ensuring that the missed detection rate does not increase, the false alarm rate of early warnings is significantly reduced, effectively solving the problem of personnel complacency caused by traditional indiscriminate early warnings, and significantly improving the effectiveness and field applicability of early warning information.

[0039] The present invention describes a dynamic adaptive intelligent adjustment method and system for early warning areas of oilfield electronic fences. It establishes a closed-loop system integrating dual-source data adaptation, model training, dynamic adjustment, and visualized control. It supports seamless dual-source access of synthetic simulation data and real personnel positioning data, and is equipped with a full-process visualized human-computer interaction interface. It realizes intelligent control of the entire process of oilfield electronic fences from data processing, model training, area adjustment to early warning display. It does not require complicated code operations, which greatly reduces the deployment and use threshold in oilfields and is fully adapted to the actual control needs of oilfield safety production.

[0040] This invention is also applicable to fields such as safety early warning and control of workers in complex field oilfield scenarios. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating a dynamic adaptive intelligent adjustment method for early warning areas of oilfield electronic fences, as described in Implementation Method 1.

[0042] Figure 2 This is a schematic diagram of the hierarchical early warning rules and false alarm suppression logic described in Implementation Method 1. Detailed Implementation

[0043] 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.

[0044] Implementation Method 1, see Figures 1 to 2 This embodiment describes a dynamic and adaptive intelligent adjustment method for early warning areas of oilfield electronic fences. The method specifically includes the following steps: Step 1: System Initialization and Parameter Configuration Configure the system hardware environment and deep learning framework, and determine the computing devices (CPU / GPU) for model training and inference. Set the core parameters for oilfield electronic fence control: functional area boundary coordinates, time series length, model hyperparameters, fence scaling factor, false alarm suppression threshold, etc. Initialize the hierarchical early warning rules, model network structure, and weight parameters.

[0045] Step 2: Generation / Loading and Preprocessing of Dual-Source Data Select the data source through the visual interface: synthetic data or real CSV data. If you choose to use synthetic data: set the number of samples and the time span to generate a batch of synthetic datasets of personnel activity trajectories that conform to the operation patterns of oilfields; If real data is selected: Load real personnel location data in CSV format and complete preprocessing operations such as timestamp conversion, feature standardization, and outlier filtering; The preprocessed dataset is converted into a time-series feature sequence, divided into training and test sets in an 8:2 ratio, and a standardized data loader is constructed.

[0046] This step establishes a dual-source supply mechanism that combines synthetic data generation with standardized processing of real-world location data. On one hand, it generates high-fidelity synthetic data by simulating the spatiotemporal patterns and environmental characteristics of oilfield operations, addressing the challenges of collecting real-world personnel location data and the incomplete scene coverage in oilfields. On the other hand, it completes standardized preprocessing of real-world CSV-format location data, enabling seamless switching and unified input between the two data sources, and providing a standardized time-series dataset for model training.

[0047] Oilfield personnel activity composite data generation The boundary coordinates of three core functional areas of the oilfield are preset: core production area, daily work area, and rest and living area. Based on the day and night pattern of oilfield operations, the distribution of personnel activities at different times is simulated: during the day, there is a 70% probability of personnel being distributed in the work area, and at night, there is a 70% probability of personnel being distributed in the rest area. At the same time, risk behaviors such as crossing the core area boundary and non-compliant activities are randomly generated.

[0048] By simultaneously overlaying environmental and temporal features, a full-dimensional dataset is generated, which includes timestamps, personnel coordinates (x, y), time period labels (daytime / nighttime), weather labels (normal / severe), and warning labels. The dataset covers full-scene samples of different time periods, environments, and activity types.

[0049] The generation of early warning labels follows the oilfield classification and control rules, and the core formula is:

[0050] In the formula, label=1 is the warning label, and label=0 is the normal label.

[0051] Automatically load real personnel location data in CSV format, complete timestamp format conversion and feature standardization processing; based on the preset sequence length, convert continuous personnel activity data into time-series feature sequences, each sequence containing feature data of multiple consecutive time steps, corresponding to the warning label of the next time step, and construct a standardized dataset that can be directly used for model training.

[0052] The formula for constructing a time series is:

[0053]

[0054] In the formula, For the i-th time series input sequence, The feature vector at time step t contains four dimensions: x-coordinate, y-coordinate, time period label, and weather label. For sequence length, The warning label is the one corresponding to the sequence.

[0055] Step 3: Construction and Initialization of the Spatiotemporal Pattern Mining Model for Human Activities Construct a multi-layer LSTM temporal feature extraction network and set core parameters such as input dimension, hidden layer dimension, number of network layers, and number of output categories; Build a fully connected classification layer to complete the construction of the overall model architecture; Initialize the model weight parameters, set the loss function and optimizer, and complete the model initialization configuration.

[0056] To address the temporal dependence of personnel activities in oilfields, a deep LSTM network with multiple feature inputs is constructed to uncover the spatiotemporal patterns and behavioral trends of these activities, enabling risk warning and prediction for future moments. The model captures long-term behavioral dependencies through LSTM layers and outputs the final warning classification results through fully connected layers, overcoming the limitations of traditional static coordinate judgments and providing accurate decision-making basis for dynamic fence adjustments.

[0057] It adopts a two-level architecture of "LSTM feature extraction - fully connected layer classification", with the input being the dimension. The time-series feature sequence is used to output a 2-classification early warning prediction result (normal / early warning).

[0058] LSTM temporal feature extraction: This method encodes the input temporal sequence using a multi-layer LSTM network to capture long-range temporal dependencies in human activities. The core formula is:

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] In the formula, These are the input gate, forget gate, and output gate, respectively. In cellular state, In hidden state, For the input features at time step t, It is the Sigmoid activation function. For Hadama accumulation.

[0065] Classification Output: Take the hidden state of the last time step of the time series and complete the final warning classification through a fully connected layer. The core formula is:

[0066] In the formula, This represents the hidden state at the last time step of the sequence. It is a fully connected layer. This refers to the early warning prediction results output by the model.

[0067] Step 4: Model Training and Parameter Optimization Input the training set data into the model, complete the forward propagation calculation, and output the early warning prediction result; The cross-entropy loss is calculated based on the prediction results and the true labels, and the model weight parameters are updated through the backpropagation algorithm. The model is trained iteratively to a preset number of rounds. The prediction accuracy and generalization ability of the model are verified through a test set, and the optimal model after training is saved.

[0068] Step 5: Dynamic adaptive adjustment of the electronic fence area Load the latest time-series data of personnel activities, input the trained optimal model, and output the personnel behavior risk prediction results; Based on the current fence boundary coordinates, calculate the core parameters of the fence center, width, and height; Based on the model prediction results, the scaling factor is determined, the adjusted fence vertex coordinates are calculated, and the intelligent adjustment of the electronic fence warning area is completed.

[0069] This step, based on the prediction results of personnel behavior risks, constructs a proportional scaling adjustment mechanism with the fence center as the reference, enabling dynamic adaptive adjustment of the electronic fence warning area. When the model predicts a warning risk, the warning range is expanded to achieve early control; when there is no warning risk, the warning range is appropriately reduced to minimize interference with normal operations.

[0070] 1. Calculation of core parameters of the current fence: Based on the vertex coordinates of the current fence, calculate the center coordinates, width, and height of the fence. The core formula is:

[0071]

[0072]

[0073] In the formula, Let n be the coordinates of the i-th vertex of the current fence, and n be the number of fence vertices. Here are the coordinates of the fence center, W is the fence width, and H is the fence height.

[0074] 2. Scaling Factor Determination: The fence scaling factor is determined based on the model's prediction results, using the following formula:

[0075] In the formula, S is the fence scaling factor.

[0076] 3. Calculation of adjusted fence vertex coordinates: Based on the fence center, width, height, and scaling factor, calculate the adjusted fence vertex coordinates. For a quadrilateral fence, the core formula is:

[0077]

[0078]

[0079]

[0080] In the formula, Let be the coordinates of the i-th vertex of the fence after adjustment.

[0081] Step 6: Early Warning Reasoning and False Alarm Suppression Perform batch inference on all personnel activity data to generate initial early warning results; Based on the preset boundaries of the normal activity area, a false alarm suppression algorithm is executed to perform secondary filtering on the initial warning results; The system collects key information such as the final number of warnings, warning locations, and warning periods, and generates a warning and control report.

[0082] This step involves constructing tiered early warning rules adapted to oilfield scenarios, clarifying the control requirements for different regions and time periods; at the same time, it designs a false alarm suppression mechanism based on normal activity areas to perform secondary filtering on the early warning results predicted by the model, eliminating invalid early warnings in normal activity areas, significantly reducing the false alarm rate, and improving the effectiveness of early warning information.

[0083] For the three core functional areas of the oilfield, a hierarchical control rule is established that links time period, region, and environment: Core production area: High-level control around the clock, with an alarm triggered immediately upon personnel entry; Daily work area: During the daytime, which is the normal working period, personnel entering will not trigger an alarm; at night, which is the non-working period, personnel entering will trigger an alarm. Rest and living area: Nighttime is the normal rest period, and personnel entering will not trigger an alarm; Daytime is the non-rest period, and personnel entering will trigger an alarm. Severe weather scenario: Expand the warning range of the core production area. The warning will be triggered when personnel enter the expanded control area.

[0084] Based on the preset boundaries of normal activity areas, the early warning results predicted by the model are filtered. The core formula is:

[0085] In the formula, The warning labels predicted by the model. This is the final warning label after false alarm suppression. The current coordinates of the personnel. This is a pre-defined set of normal activity areas.

[0086] Step 7: Visualization and Data Recording The system displays personnel activity trajectories, adjusted electronic fence boundaries, normal activity areas, and early warning points in a visual interface, providing an intuitive view of the system's operational performance. The system automatically records operation logs, including core data such as data loading information, model training process, fence adjustment records, and early warning results, enabling full traceability of the control process. It supports local storage of early warning reports, trajectory data, and fence boundary data, meeting the archiving requirements for oilfield safety production management.

[0087] This step involves constructing a complete model training and daily inference process, optimizing model parameters through iterative training to improve the accuracy of human behavior prediction, and designing an automated daily inference process to complete the entire process of dynamic fence adjustment, batch early warning prediction, and false alarm suppression, thereby achieving automated operation of the system.

[0088] Model training process: The cross-entropy loss function is used as the model optimization objective. The Adam optimizer iteratively updates the parameters and minimizes the model's prediction loss through multiple rounds of iterative training to improve the accuracy of risk prediction. The core loss formula is:

[0089] In the formula, N is the sample size. For real labels, Predict probabilities for the model.

[0090] Daily automated inference process: Step 1: Load the time series data of personnel activities for the day, input it into the trained model, and complete the adaptive adjustment of the fenced area; Step 2: Perform batch early warning prediction on all personnel activity data for the day and generate initial early warning results; Step 3: Execute the false alarm suppression algorithm to filter the initial warning results and output the final accurate warning information.

[0091] This implementation method establishes an integrated human-computer interaction GUI interface, integrates the full-process functions of the system, realizes visualized operation and result display, reduces the usage threshold for oilfield personnel, and achieves full-process visualized control of electronic fences.

[0092] Data source selection: Supports switching between synthetic data and real CSV data, and provides CSV file selection functionality; Core operation buttons: Integrates three core functions: data loading, model training, and daily fence adjustment; Visualized display area: Enables the visualization of personnel activity trajectories, electronic fence boundaries, normal activity areas, and warning points, clearly showing the effects of fence adjustments and warning distribution.

[0093] Implementation Method 2: This implementation method proposes a dynamically adaptive intelligent adjustment system for oilfield electronic fence early warning areas. The system architecture is based on a model for mining the spatiotemporal patterns of human activities, and is designed with oilfield scenario adaptability and field practicality as the guiding principles. It constructs a closed-loop architecture, enabling data exchange and result feedback between different levels, from bottom to top as follows: The dual-source data processing module is used to configure core parameters through a visual interface. These core parameters include functional area boundary coordinates, time series length, model hyperparameters, fence scaling factor, and false alarm suppression threshold. The module for mining spatiotemporal patterns of human activities is used to select synthetic or real data through a visual interface, convert the preprocessed synthetic or real data into a time-series feature sequence, divide it into training and test sets in an 8:2 ratio, and build a standardized data loader. The model building module is used to construct a multi-layer LSTM temporal feature extraction network, build a fully connected classification layer, and complete the construction and initialization of the model for mining the spatiotemporal patterns of human activities. The graded early warning rules and false alarm suppression module is used to input the training set into the spatiotemporal pattern mining model of personnel activities built by the model building module to complete the forward propagation calculation and output the early warning prediction result; calculate the cross-entropy loss based on the prediction result and the true label, update the model weight parameters through the backpropagation algorithm; iterate the training to the preset number of rounds and save the optimal model after training. The dynamic adaptive fence area intelligent adjustment module is used to load the latest human activity time series data, input the trained optimal model, and output the human behavior risk prediction results; based on the current fence boundary coordinates, it calculates the fence center, width, and height parameters; based on the spatiotemporal patterns of human activity, it mines the model prediction results to determine the scaling factor, calculates the adjusted fence vertex coordinates, and completes the intelligent adjustment of the electronic fence warning area; The model inference and false alarm suppression module is used to perform batch inference on all personnel activity data and generate initial warning results; based on the preset normal activity area boundaries, it executes a false alarm suppression algorithm to perform secondary filtering on the initial warning results; and it counts the final number of warnings, warning locations, and warning time periods to generate a warning control report. The visual interactive management module is used to draw personnel activity trajectories, adjusted electronic fence boundaries, normal activity areas, and early warning points on a visual interface, and automatically record system operation logs.

[0094] Example 1: To verify the technical advantages and application effects of the present invention, a special test dataset was constructed based on an oilfield operation scenario. A traditional fixed electronic fence solution was selected as a control group. A comparative experiment was conducted under the same hardware environment and dataset, as detailed below: Experimental Environment and Dataset Hardware environment: CPU Intel Core i5-12400, GPU NVIDIA RTX 5070Ti, 16GB memory; Software environment: Python 3.9, PyTorch 2.0, Windows 10 operating system; Test dataset: Synthesized 30-day activity trajectory data of oilfield personnel, with a total of 3,000 valid records, covering daytime / nighttime periods and normal / severe weather scenarios, including various risk behaviors such as crossing the core area boundary, nighttime intrusion into the work area, and daytime intrusion into the rest area, and marked with accurate warning labels.

[0095] Table 1 Core Performance Comparison Experiment

[0096] Experimental results show that, compared with traditional fixed fences, the present invention significantly improves the accuracy of early warning, while significantly reducing the false alarm rate and the missed alarm rate, thus solving the core pain points of the traditional solution.

[0097] Table 2 Performance Verification by Scenario

[0098] Experimental results show that the solution of the present invention has excellent performance in various typical operating scenarios in oil fields, especially in day-night difference scenarios and severe weather scenarios, which fully demonstrates the strong adaptability of the solution to complex oil field scenarios.

[0099] System performance verification: Model training time: 3000 samples trained for 10 rounds, total time ≤90 seconds; Daily data fence adjustment and early warning inference time: ≤50ms; Visual interface operation response latency: ≤100ms; Fully meets the performance requirements of real-time control in oilfields.

[0100] In summary, this implementation method constructs a full-process technical solution that integrates multi-feature time-series mining, hierarchical early warning rules, dynamic regional adjustment, and false alarm suppression. This enables adaptive intelligent adjustment of the early warning area of ​​the oilfield electronic fence, significantly reducing the false alarm rate while ensuring operational safety, and fully adapting to the on-site management needs of oilfield field operations.

[0101] 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.

[0102] 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 method for intelligent adjustment of early warning areas in oilfield electronic fences with dynamic and adaptive characteristics, characterized in that... The method includes the following steps: Step 1: Configure core parameters through the visual interface. The core parameters include functional area boundary coordinates, time series length, model hyperparameters, fence scaling factor, and false alarm suppression threshold. Step 2: Select synthetic data or real data through the visual interface, convert the preprocessed synthetic data or real data into time series feature sequences, divide them into training set and test set in an 8:2 ratio, and build a standardized data loader. Step 3: Construct a multi-layer LSTM temporal feature extraction network, build a fully connected classification layer, and complete the construction and initialization of the model for mining the spatiotemporal patterns of human activities; Step 4: Input the training set into the spatiotemporal pattern mining model of personnel activities constructed in Step 3 to complete the forward propagation calculation and output the early warning prediction result; calculate the cross-entropy loss based on the prediction result and the real label, and update the model weight parameters through the backpropagation algorithm; iterate the training to the preset number of rounds and save the optimal model after training. Step 5: Load the latest personnel activity time series data, input the trained optimal model, and output the personnel behavior risk prediction results; calculate the fence center, width, and height parameters based on the current fence boundary coordinates; determine the scaling factor based on the model prediction results according to the spatiotemporal patterns of personnel activities, calculate the adjusted fence vertex coordinates, and complete the intelligent adjustment of the electronic fence warning area; Step 6: Perform batch inference on all personnel activity data to generate initial warning results; based on the preset normal activity area boundaries, execute a false alarm suppression algorithm to perform secondary filtering on the initial warning results; count the final number of warnings, warning locations, and warning time periods to generate a warning control report; Step 7: Draw the personnel activity trajectory, adjusted electronic fence boundaries, normal activity areas, and warning points on the visualization interface, and automatically record the system operation log.

2. The method for intelligent adjustment of early warning areas of oilfield electronic fences with dynamic adaptation according to claim 1, characterized in that, The method for converting the preprocessed synthetic data or real data into a time-series feature sequence in step 2 is as follows: In the formula, For the i-th time series input sequence, Let be the feature vector at time step t. For sequence length, The warning label is the one corresponding to the sequence.

3. The method for intelligent adjustment of early warning areas of oilfield electronic fences with dynamic adaptation according to claim 1, characterized in that, The method for constructing and initializing the multi-layer LSTM temporal feature extraction network and building a fully connected classification layer in step 3 to complete the construction of the spatiotemporal pattern mining model for personnel activities is as follows: Step 3.1: Encode the input time series sequence using a multi-layer LSTM network to capture long-range temporal dependencies of personnel activities. The calculation method is as follows: In the formula, These are the input gate, forget gate, and output gate, respectively. In cellular state, In hidden state, For the input features at time step t, It is the Sigmoid activation function. For Hadamah accumulation; Step 3.2: Take the hidden state of the last time step of the time series and complete the final warning classification through a fully connected layer. The calculation method is as follows: In the formula, This represents the hidden state at the last time step of the sequence. It is a fully connected layer. This refers to the early warning prediction results output by the model.

4. The method for intelligent adjustment of early warning areas of oilfield electronic fences with dynamic adaptation according to claim 1, characterized in that, The method for calculating the fence center, width, and height parameters based on the current fence boundary coordinates in step 5 is as follows: In the formula, Let n be the coordinates of the i-th vertex of the current fence, and n be the number of fence vertices. Here are the coordinates of the fence center, W is the fence width, and H is the fence height.

5. The method for intelligent adjustment of early warning areas of oilfield electronic fences with dynamic adaptation according to claim 1, characterized in that, Step 5 involves determining the scaling factor based on the prediction results of the model based on the spatiotemporal patterns of human activities, calculating the adjusted fence vertex coordinates, and completing the intelligent adjustment of the electronic fence warning area. In the formula, S is the fence scaling factor; Calculate the adjusted fence vertex coordinates: In the formula, Let be the coordinates of the i-th vertex of the fence after adjustment.

6. The method for intelligent adjustment of early warning areas of oilfield electronic fences with dynamic adaptation according to claim 1, characterized in that, In step 6, based on the preset normal activity area boundary, the false alarm suppression algorithm is executed to perform secondary filtering on the initial warning results. The method is as follows: In the formula, The warning labels predicted by the model. This is the final warning label after false alarm suppression. The current coordinates of the personnel. This is a pre-defined set of normal activity areas.

7. The method for intelligent adjustment of early warning areas of oilfield electronic fences with dynamic adaptation according to claim 1, characterized in that, Step 6 also includes using the cross-entropy loss function as the model optimization objective, iteratively updating the parameters of the Adam optimizer, and minimizing the model's prediction loss through multiple rounds of iterative training. In the formula, N is the sample size. For real labels, Predict probabilities for the model.

8. A dynamically adaptive intelligent adjustment system for early warning areas of oilfield electronic fences, characterized in that, The system includes: The dual-source data processing module is used to configure core parameters through a visual interface. These core parameters include functional area boundary coordinates, time series length, model hyperparameters, fence scaling factor, and false alarm suppression threshold. The module for mining spatiotemporal patterns of human activities is used to select synthetic or real data through a visual interface, convert the preprocessed synthetic or real data into a time-series feature sequence, divide it into training and test sets in an 8:2 ratio, and build a standardized data loader. The model building module is used to construct a multi-layer LSTM temporal feature extraction network, build a fully connected classification layer, and complete the construction and initialization of the model for mining the spatiotemporal patterns of human activities. The graded early warning rules and false alarm suppression module is used to input the training set into the spatiotemporal pattern mining model of personnel activities built by the model building module to complete the forward propagation calculation and output the early warning prediction result; calculate the cross-entropy loss based on the prediction result and the true label, update the model weight parameters through the backpropagation algorithm; iterate the training to the preset number of rounds and save the optimal model after training. The dynamic adaptive fence area intelligent adjustment module is used to load the latest human activity time series data, input the trained optimal model, and output the human behavior risk prediction results; based on the current fence boundary coordinates, it calculates the fence center, width, and height parameters; based on the spatiotemporal patterns of human activity, it mines the model prediction results to determine the scaling factor, calculates the adjusted fence vertex coordinates, and completes the intelligent adjustment of the electronic fence warning area; The model inference and false alarm suppression module is used to perform batch inference on all personnel activity data and generate initial warning results; based on the preset normal activity area boundaries, it executes a false alarm suppression algorithm to perform secondary filtering on the initial warning results; and it counts the final number of warnings, warning locations, and warning time periods to generate a warning control report. The visual interactive management module is used to draw personnel activity trajectories, adjusted electronic fence boundaries, normal activity areas, and early warning points on a visual interface, and automatically record system operation logs.

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 1-7.

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 1-7.

Images