Security fence protection method and device based on Internet of Things technology

Abstract

The invention discloses a safety fence protection method and device based on the Internet of Things technology, and belongs to the technical field of safety alarm and Internet of Things application, and the method comprises the following steps: respectively collecting the distance, heat source distribution and movement track information of a target protection area through an ultrasonic sensor, an infrared sensor and a millimeter wave radar; generating a virtual fence model with an initial alarm boundary by adopting a confidence weighted fusion algorithm based on environmental condition dynamic adjustment; environmental change parameters such as obstacle distribution and meteorological conditions are monitored in real time, when the change exceeds a preset threshold value, a dynamic boundary adjustment strategy is triggered, and the fence alarm boundary shape is optimized through a boundary node position redistribution mechanism; a distributed computing framework is adopted to decompose boundary detection tasks, and the tasks are dynamically allocated according to the real-time resource utilization rate of edge computing nodes. According to the invention, the problems of insufficient real-time performance, limited dynamic adjustment capability and low system integration degree in a complex environment in the prior art can be effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety alarm and Internet of Things application, in particular to a safety fence protection method and device based on Internet of Things technology. BACKGROUND

[0002] With the rapid development of Internet of Things technology, safety alarm technology has been widely applied in industrial facilities, substations, power plants and other scenarios. These technologies improve the intelligence and precision of safety protection through virtual fences, intelligent monitoring and early warning systems. However, the existing related technical solutions still have certain deficiencies in actual application, especially in real-time, environmental adaptability and system integration, and the safety protection requirements in complex scenarios put forward higher requirements for technology.

[0003] The invention patent with publication number CN120580774A discloses an adaptive intelligent fence system based on multi-modal perception and its early warning method, relating to the field of power facility safety protection, which includes a multi-modal sensor layer, an edge computing node, a cloud platform collaboration layer, an audible and visual alarm module and a mobile terminal. The present application integrates multi-modal sensing, edge computing and dynamic alarm strategy to solve the problems of poor environmental adaptability, serious false and missed alarms, low deployment efficiency and other problems existing in the current safety fence, and realizes the intelligence, efficiency and precision of high-risk area safety protection.

[0004] The invention patent with publication number CN120220365B discloses a substation virtual fence crossing protection early warning method and system based on Internet of Things, belonging to the technical field of virtual fence crossing protection. The method establishes a three-dimensional coordinate grid of the substation by laser radar scanning, defines the virtual fence boundary coordinate set, and deploys Internet of Things devices to collect the dynamic position of moving objects; the equal time node timing perception time is generated to include the distance and detection state of the crossing feature label; the state perception sensitivity matrix is constructed to evaluate the crossing protection fusion sensitivity and effectiveness between the perception time segments, and the time range is integrated and analyzed to lock the detection protection range. The system includes virtual fence construction, Internet of Things device management, data processing and early warning evaluation modules, realizes real-time monitoring, risk assessment and dynamic adjustment of protection range of substation crossing behavior, and improves the intelligence and precision level of substation safety protection. The patent establishes a three-dimensional coordinate grid of the substation by laser radar scanning, defines the virtual fence boundary coordinate set, and uses Internet of Things devices to collect the dynamic position of moving objects, realizes real-time monitoring and risk assessment of crossing behavior.

[0005] There are three major pain points in safety fence alarm technology:

[0006] Firstly, the environmental adaptability is weak, and it mainly depends on single sensor or fixed fusion algorithm, which cannot dynamically adjust the perception strategy according to changes in light, weather and other environments, resulting in large fluctuations in alarm accuracy;

[0007] Secondly, the dynamic adjustment capability is limited. The fence boundaries are mostly statically designed and cannot be optimized in real time according to changes in obstacles (such as temporarily stacked equipment or naturally growing vegetation), which can easily create blind spots or false alarms.

[0008] Third, the system has low integration. The detection calculation, anomaly identification and early warning feedback links lack coordinated optimization, resulting in delayed alarm response and difficulty in associating abnormal behavior with protection vulnerabilities, thus failing to improve protection capabilities from the root. Summary of the Invention

[0009] The purpose of this invention is to provide a security fence protection method and device based on Internet of Things (IoT) technology, which can effectively solve the problems of insufficient real-time performance, limited dynamic adjustment capabilities, and low system integration in existing technologies in complex environments. Existing security fence protection technologies typically rely on a single sensor or fixed algorithm, which is difficult to adapt to environmental changes and lacks the ability to analyze multi-dimensional behavioral characteristics, resulting in limited anomaly detection accuracy and response efficiency.

[0010] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0011] A security fence protection method based on Internet of Things (IoT) technology includes the following steps:

[0012] A virtual fence model based on multi-sensor data fusion is constructed. Distance information, heat source distribution information, and motion trajectory information of the target protection area are collected by ultrasonic sensors, infrared sensors, and millimeter-wave radar, respectively. A confidence-weighted fusion algorithm based on dynamic adjustment of environmental conditions is used to generate the initial fence boundary. Based on the initial fence boundary, environmental change parameters are monitored in real time. When the detected change in obstacle distribution or meteorological conditions exceeds a preset threshold, a dynamic boundary adjustment strategy is triggered to optimize the fence boundary shape through a boundary node position redistribution mechanism.

[0013] The optimized fence boundary detection algorithm is decomposed into multiple detection subtasks using a distributed computing framework. The task is dynamically allocated based on the real-time resource utilization of each edge computing node, and the boundary detection is executed in parallel on the edge computing nodes of the target protection area.

[0014] Abnormal behavior is identified based on multi-dimensional behavioral features. Features such as the number of boundary nodes, the shape of the motion trajectory, and the speed change are extracted. A pre-set machine learning classification model is used to determine whether the behavior pattern of the target object is abnormal.

[0015] When abnormal behavior is detected, feedback scheduling optimization is performed based on the frequency of occurrence of historical behavior trajectories, the monitoring sampling frequency is dynamically adjusted, and optimized monitoring data is generated. Based on the optimized monitoring data, a detailed anomaly report is generated through a multi-level judgment mechanism, the correlation between abnormal behavior and security fence protection vulnerabilities is analyzed, and vulnerability detection results are output.

[0016] In one embodiment of the present invention, the confidence-weighted fusion algorithm specifically includes:

[0017] Under sufficient lighting conditions, the detection data from infrared sensors and millimeter-wave radar are used to reduce the weighting of ultrasonic sensors.

[0018] In adverse weather conditions, the detection data from millimeter-wave radar and ultrasonic sensors are used, reducing the weighting of infrared sensors.

[0019] By dynamically adjusting the weights of each sensor, weighted fusion calculations are performed on multi-source detection data to generate a set of boundary coordinates.

[0020] In the confidence-weighted fusion algorithm, the sensor weight configuration is based on experimental data of the detection error rate of each sensor under different environmental conditions:

[0021] When the illuminance is >10000 lux, the detection error rate of the infrared sensor is <5%, the error rate of the millimeter-wave radar is <8%, and the error rate of the ultrasonic sensor due to light interference is >25%.

[0022] When visibility is less than 500m, the error rate of millimeter-wave radar is less than 6%, the error rate of ultrasonic radar is less than 12%, and the error rate of infrared radar due to fog and haze is greater than 30%. The weighting configuration is inversely proportional to the error rate.

[0023] In one embodiment of the present invention, the boundary node redistribution mechanism includes:

[0024] Define the initial set of boundary node positions and establish the correlation between node positions and environmental change parameters;

[0025] When an environmental change is detected, the position adjustment amount of each node is calculated based on the degree of environmental change. The greater the environmental change, the greater the position adjustment amount.

[0026] By iteratively calculating multiple times, the total adjustment of all nodes is minimized, achieving a smooth transition of the boundary shape.

[0027] In one embodiment of the present invention, the dynamic task allocation strategy includes:

[0028] Real-time monitoring of processor utilization, memory usage, and network communication latency of each edge computing node;

[0029] Based on the above test data, a comprehensive load score is calculated. The lower the load score, the stronger the node's processing capability.

[0030] The detection tasks are preferentially assigned to the node with the lowest overall load score to achieve load balancing.

[0031] In one embodiment of the present invention, the multi-dimensional behavioral features include:

[0032] Trajectory morphology characteristics, including total trajectory length, trajectory curvature, average velocity, and maximum acceleration;

[0033] Boundary interaction characteristics include the rate of change of the number of boundary nodes, the frequency of change of movement direction, and the dwell time in the boundary region.

[0034] In one embodiment of the present invention, the construction of the machine learning classification model includes:

[0035] Collect historical behavioral data as training samples and label them with classification tags for normal and abnormal behaviors;

[0036] The support vector machine algorithm is used to learn from the training samples and establish a mapping relationship between behavioral features and classification results;

[0037] The optimal hyperparameters (such as the penalty coefficient C and the kernel function parameter γ) were found within the preset parameter range using a grid search method, and the model parameters were finally determined to be γ=0.1 and C=1.0.

[0038] By mapping features to a high-dimensional space using kernel functions, the optimal classification decision boundary can be found.

[0039] In one embodiment of the present invention, the feedback scheduling optimization specifically includes:

[0040] Analyze the frequency of abnormal behavior in different time periods within historical behavioral patterns;

[0041] When the frequency of anomalies in a certain time period exceeds a preset threshold, the monitoring sampling frequency for that time period will be automatically increased.

[0042] The increase in sampling frequency is directly proportional to the degree to which the frequency of anomalies exceeds the threshold.

[0043] In addition, the present invention also discloses a security fence protection device based on Internet of Things (IoT) technology, used to implement the above-mentioned security fence protection method based on IoT technology, including:

[0044] The multi-sensor fusion module is used to collect environmental data through ultrasonic sensors, infrared sensors and millimeter-wave radar, and to generate the initial fence boundary using a confidence-weighted fusion algorithm that is dynamically adjusted based on environmental conditions.

[0045] The dynamic boundary adjustment module is used to monitor environmental change parameters and trigger a boundary node redistribution mechanism to optimize the fence boundary.

[0046] The distributed computing module is used to decompose the detection algorithm into multiple subtasks and dynamically allocate and execute tasks in parallel based on the resource utilization of each edge computing node.

[0047] The abnormal behavior recognition module is used to extract multi-dimensional behavioral features and judge abnormal behavior through machine learning classification models.

[0048] The feedback optimization module is used to dynamically adjust the monitoring sampling frequency based on the frequency of occurrence of historical behavior trajectories;

[0049] The vulnerability detection module is used to generate anomaly reports through a multi-level judgment mechanism and analyze the correlation of vulnerabilities.

[0050] In one embodiment of the present invention, the multi-sensor fusion module includes a weight configuration unit, a data fusion unit, and a quality control unit.

[0051] The weighting configuration unit is used to dynamically adjust the confidence weight ratio of each sensor according to environmental conditions;

[0052] The data fusion unit is used to generate a set of boundary coordinates through weighted fusion calculation;

[0053] The quality control unit is used to verify the validity and consistency of the fused data.

[0054] In one embodiment of the present invention, the dynamic boundary adjustment module includes an environmental monitoring unit, a node distribution optimization unit, and a boundary reconstruction unit; the environmental monitoring unit is used to collect obstacle distribution and meteorological condition parameters in real time; the node distribution optimization unit is used to perform position redistribution calculation of boundary nodes; and the boundary reconstruction unit is used to regenerate the fence boundary based on the optimized node distribution.

[0055] In one embodiment of the present invention, the distributed computing module includes a task decomposition unit, a load monitoring unit, and a task scheduling unit; the task decomposition unit is used to divide the detection algorithm into multiple sub-tasks according to regions; the load monitoring unit is used to monitor the resource utilization status of each computing node in real time; and the task scheduling unit is used to allocate tasks based on a comprehensive load score.

[0056] In one embodiment of the present invention, the abnormal behavior recognition module includes:

[0057] The feature extraction unit is used to calculate trajectory morphology features and boundary interaction features;

[0058] The model computation unit is used to execute machine learning classification algorithms;

[0059] The result determination unit is used to output the result of behavior anomaly determination.

[0060] In one embodiment of the present invention, the feedback optimization module includes:

[0061] Frequency statistics unit, used to record the frequency of occurrence of historical behavior trajectories;

[0062] The parameter adjustment unit is used to dynamically adjust the monitoring sampling frequency;

[0063] The optimization verification unit is used to verify the optimization effect and provide feedback for adjustments.

[0064] Furthermore, in some preferred embodiments, the multi-sensor data fusion uses the IEEE 1588 precise time protocol to achieve time synchronization with a synchronization accuracy of <1ms, and weighted fusion of timestamp-aligned data is performed through a sliding window caching mechanism.

[0065] Furthermore, the position redistribution mechanism employs a gradient descent algorithm with a momentum term, a momentum coefficient μ=0.9, and a learning rate α=0.01 that is adaptively adjusted. Convergence is determined when the objective function descent rate is <0.1% for 10 consecutive iterations.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] This invention generates an initial fence boundary by collecting data from three types of sensors: ultrasonic, infrared, and millimeter-wave radar, combined with a confidence-weighted fusion algorithm that dynamically adjusts to environmental conditions. The multi-source data fusion approach effectively overcomes the limitations of single sensors, improving the accuracy and robustness of the fence boundary. Simultaneously, a boundary node position redistribution mechanism is triggered based on environmental change parameters (obstacle distribution, weather conditions) to optimize the fence shape, breaking through the limitations of traditional static fence design and enhancing the system's adaptability. This invention decomposes the detection algorithm into multiple sub-tasks and dynamically allocates tasks based on the real-time resource utilization of each node, significantly improving computational efficiency and resource utilization. More importantly, by extracting multi-dimensional behavioral features (trajectory morphology, boundary interactions, etc.) and combining them with a machine learning classification model, accurate identification and judgment of abnormal behavior are achieved. Attached Figure Description

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0069] Figure 1 This is a schematic diagram illustrating the working principle of the dynamic boundary adjustment module of the present invention. Detailed Implementation

[0070] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0071] The following is in conjunction with the appendix Figure 1 The embodiments of the present invention will be described in detail below.

[0072] Example 1:

[0073] This embodiment discloses a security fence protection method based on Internet of Things (IoT) technology. The method and device achieve efficient security protection of the target protection area through the collaborative work of a multi-sensor fusion module, a dynamic boundary adjustment module, a distributed computing module, an abnormal behavior recognition module, and a feedback optimization module.

[0074] When constructing a virtual fence model based on multi-sensor data fusion, the multi-sensor fusion module serves as the core component, including a weight configuration unit, a data fusion unit, and a quality control unit. The weight configuration unit dynamically adjusts the confidence weight ratios of the ultrasonic sensor, infrared sensor, and millimeter-wave radar according to environmental conditions.

[0075] In specific implementation: The multi-sensor data fusion adopts the IEEE 1588 precise time protocol for time synchronization, with a synchronization accuracy of <1ms. Each sensor data is accompanied by a timestamp. A sliding window caching mechanism (window size 100ms) is used to weight and fuse the timestamp-aligned data to ensure the time consistency of multi-source data. The sliding window size is dynamically adjusted according to the sensor sampling frequency, with a minimum window size of 50ms and a maximum window size of 200ms, to balance real-time performance and data integrity.

[0076] In low-light or rainy / foggy weather, the weights of infrared sensors and millimeter-wave radar are increased to compensate for the decreased accuracy of ultrasonic sensors. The data fusion unit receives data from each sensor, generates distance distribution maps, heat source distribution maps, and trajectory distribution maps, and integrates them into a set of coordinates for the initial fence boundary using a weighted fusion algorithm. The quality control unit is responsible for verifying the validity and consistency of the fused data, ensuring that the generated fence boundary geometry meets actual requirements and protection specifications.

[0077] The dynamic boundary adjustment module collects real-time obstacle distribution information and meteorological condition parameters within the target protection area through the environmental monitoring unit, including temperature, humidity, wind speed, and obstacle density. When an environmental change exceeds a preset threshold, the node distribution optimization unit triggers a boundary node redistribution mechanism.

[0078] The node distribution optimization unit optimizes the position distribution of boundary nodes through multiple iterations, ensuring a smooth transition in boundary shape and minimizing the total adjustment. The boundary reconstruction unit regenerates the fence boundary based on the optimized node distribution and updates the fence model. The environmental monitoring unit, node distribution optimization unit, and boundary reconstruction unit are connected sequentially to form a closed-loop feedback system, ensuring that the fence boundary can adapt to complex environmental changes in real time.

[0079] The task decomposition unit of the distributed computing module divides the target protection area into multiple sub-regions, each corresponding to a detection sub-task. The load monitoring unit monitors the processor utilization, memory usage, and network communication latency of each distributed computing node in real time, and calculates a comprehensive load score based on these parameters. The task scheduling unit prioritizes the allocation of detection tasks to nodes with lower loads based on the comprehensive load score, thereby achieving load balancing. After the detection sub-tasks are executed in parallel on the edge computing nodes, the detection results of each sub-task are aggregated to generate a complete fence boundary detection report.

[0080] The feature extraction unit of the abnormal behavior recognition module is responsible for extracting trajectory morphology features and boundary interaction features from the behavioral data of the target object. Trajectory morphology features include total trajectory length, curvature, average speed, and maximum acceleration, while boundary interaction features cover the rate of change of the number of boundary nodes, the frequency of change of movement direction, and the dwell time in the boundary region. The model calculation unit uses the support vector machine algorithm to learn from historical behavioral data, establishing a mapping relationship between behavioral features and classification results, and uses a kernel function to map the features to a high-dimensional space to find the optimal classification decision boundary. The result judgment unit inputs the extracted behavioral features into the classification model to determine whether the behavioral pattern of the target object is abnormal. This process relies on a comprehensive analysis of multi-dimensional behavioral features to ensure the accuracy of abnormal behavior recognition.

[0081] The frequency statistics unit of the feedback optimization module records the frequency of abnormal behavior in each time period of the historical behavior trajectory and defines a threshold range for the frequency of abnormal behavior. When the frequency of abnormal behavior in a certain time period exceeds the preset threshold, the parameter adjustment unit automatically increases the monitoring sampling frequency for that time period. The increase in sampling frequency is proportional to the degree to which the abnormal frequency exceeds the threshold. The optimization verification unit verifies the optimization effect and provides feedback for adjustment, generating optimized monitoring data to update the monitoring plan. By dynamically adjusting the monitoring frequency, the system's response efficiency and resource utilization are improved.

[0082] The vulnerability detection module generates detailed anomaly reports through a multi-level judgment mechanism, analyzing the correlation between anomalous behavior and vulnerabilities protected by the security fence. After extracting the temporal distribution characteristics, spatial distribution characteristics, and behavioral pattern characteristics of the anomalous behavior, the specific location and type of the vulnerability are determined. The generated anomaly report includes a description of the anomalous behavior, the vulnerability detection results, and improvement suggestions, and records the handling process of the anomalous behavior.

[0083] To facilitate a better understanding of the present invention by those skilled in the art, the present invention will be further elaborated below with reference to specific algorithm models.

[0084] A specific area of ​​an industrial plant needs to be protected.

[0085] First, a virtual fence model is constructed using a multi-sensor fusion module to define the boundary of the target protection area and divide it into grid cells. Ultrasonic sensors, infrared sensors, and millimeter-wave radar collect distance information, heat source distribution information, and motion trajectory information, respectively, to generate the initial fence boundary.

[0086] Subsequently, the dynamic boundary adjustment module monitors environmental change parameters in real time. When changes in obstacle distribution or weather conditions exceed preset thresholds, it triggers a dynamic boundary adjustment strategy to optimize the fence boundary shape. The distributed computing module decomposes the detection task into multiple sub-tasks and dynamically allocates tasks based on node resource utilization, executing boundary detection in parallel on edge computing nodes. The abnormal behavior recognition module extracts multi-dimensional behavioral features and uses a machine learning classification model to determine abnormal behavior. When abnormal behavior is detected, the feedback optimization module dynamically adjusts the monitoring sampling frequency based on the frequency of occurrence of historical behavioral trajectories. Finally, the vulnerability detection module generates a detailed anomaly report through a multi-level judgment mechanism, analyzes vulnerability correlations, and outputs detection results.

[0087] The various modules and units are tightly interconnected through data and control flows, forming a highly integrated system. The multi-sensor fusion module provides fundamental data support for the entire system, the dynamic boundary adjustment module ensures the flexibility of the fence boundaries, the distributed computing module improves detection efficiency, the abnormal behavior recognition module achieves accurate anomaly detection, and the feedback optimization module and vulnerability detection module further optimize system performance and provide improvement suggestions. This modular design not only enhances the system's scalability but also significantly improves security protection capabilities in complex environments.

[0088] In practical implementation, the confidence-weighted fusion algorithm includes the following steps:

[0089] Sensor reliability coefficient mapping: Establishing a quantitative mapping relationship between environmental conditions and sensor reliability coefficients.

[0090] Under sufficient lighting conditions: illuminance > 10000 lux, the reliability coefficient of the infrared sensor is 0.85, the reliability coefficient of the millimeter-wave radar is 0.75, and the reliability coefficient of the ultrasonic sensor is 0.40;

[0091] Under adverse weather conditions: visibility <500m, the reliability coefficient of the infrared sensor is 0.25, the reliability coefficient of the millimeter-wave radar is 0.90, and the reliability coefficient of the ultrasonic sensor is 0.75.

[0092] Dynamic weight calculation model:

[0093] The real-time weights of each sensor are calculated using the following formula:

[0094] ;

[0095] in:

[0096] : Fusion weights of sensor i;

[0097] : Sensor reliability coefficient under current conditions;

[0098] Inherent accuracy coefficient of the sensor; infrared sensor Millimeter-wave radar Ultrasonic sensor ;

[0099] : Real-time confidence level of the sensor, based on signal quality assessment, with a value range of 0.6-1.0.

[0100] Among them, real-time confidence Based on signal-to-noise ratio evaluation:

[0101] ;

[0102] in, Current signal-to-noise ratio; Minimum acceptable signal-to-noise ratio; Ideal signal-to-noise ratio.

[0103] Boundary coordinate fusion calculation employs a weighted fusion method for the coordinates of each boundary point:

[0104] ;

[0105] in, The coordinates of the boundary points detected by each sensor must be fused with a coordinate error of ≤0.5 meters.

[0106] Furthermore, the location redistribution mechanism is implemented using an optimization algorithm, with the following objective function:

[0107] ;

[0108] in,

[0109] : The total number of boundary nodes; : This is the neighbor node relationship matrix, generated using the Delaunay triangulation algorithm: when node i and node j are adjacent in the triangulation, =1, otherwise 0. Let be the Euclidean distance between node i and node j. The ideal node spacing is calculated based on a total boundary length of 1600m and a number of 200 nodes, and λ=0.3 is the smoothing coefficient.

[0110] Total number of boundary nodes : Adjacent node relationship matrix, when node i is adjacent to node j, Otherwise, it is 0; Let be the Euclidean distance between node i and its neighboring node j. =8 meters is the ideal node spacing, calculated based on a total boundary length of 1600m and a number of 200 nodes; The smoothing coefficient has a default value of 0.3; iterative optimization is performed using gradient descent, with a maximum number of iterations. Convergence threshold .

[0111] In one embodiment, the location redistribution mechanism employs a gradient descent algorithm with a momentum term, where the momentum coefficient μ = 0.9 and the learning rate α = 0.01 is adaptively adjusted. Convergence is determined when the objective function descent rate is <0.1% for 10 consecutive iterations, ensuring optimization stability under complex terrain conditions.

[0112] Each iteration is based on: Update node position.

[0113] Environmental change assessment, calculating the environmental change impact factors for each node:

[0114] ;

[0115] in: ; ; .

[0116] Position adjustment calculation: , This is a proportionality coefficient, with a value range of 0.1-0.5.

[0117] in, Based on environmental change impact factors Adaptive adjustment:

[0118] when When environmental changes are small; Fine adjustment point;

[0119] when When the environment changes, ;

[0120] when When the environment changes drastically, Significantly adjust nodes to enhance the rationality of boundary adjustments.

[0121] Convergence criterion: When the total adjustment change between two consecutive iterations is less than 1 / 3... Stop iterating when the time comes.

[0122] The quantitative implementation of the dynamic task allocation strategy, and the comprehensive load scoring model are as follows:

[0123] ;

[0124] in: Real-time monitoring of processor utilization of each edge computing node; Memory usage (%) ; Average network communication latency (ms) of the last 10 ping tests ; The unit is milliseconds. It is measured by the ping command as the average of the last 10 times. The score range is 0-1, and the smaller the value, the lighter the load.

[0125] Load status determination:

[0126] At that time, the node can accept new tasks;

[0127] when The node load is moderate;

[0128] when When a node triggers a task migration mechanism, it migrates 50% of its computational tasks to [a new node]. The lowest adjacent node.

[0129] Task allocation principle: Prioritize assigning detection tasks to the node with the lowest overall load score to achieve load balancing.

[0130] The technical parameters of the machine learning classification model are as follows:

[0131] The support vector machine model uses a radial basis function (RBF) kernel function with parameters... Penalty coefficient Perform z-score normalization on each feature dimension:

[0132] ;

[0133] in, The characteristic mean, The standard deviation is given. The model performance requirements are training accuracy ≥ 95% and test set recall ≥ 90%.

[0134] In some preferred embodiments, the training of the support vector machine model uses five-fold cross-validation to determine the optimal hyperparameters, specifically by grid search within the following parameter ranges: the penalty coefficient C ranges from [0.1, 1, 10], and the kernel function parameter γ ranges from [0.01, 0.1, 1].

[0135] Feedback scheduling optimization, the formula for dynamically adjusting the sampling frequency is as follows:

[0136] ;

[0137] in, : Basic sampling frequency (1Hz); Current anomaly frequency; Abnormal frequency threshold (0.1 times / minute); The adjustment factor (default 2.0) has a maximum sampling frequency of no more than 10Hz and a minimum adjustment interval of 5 minutes.

[0138] The protected area has an area of ​​500m × 300m, a temperature of 25℃, a visibility of 1000m, and no obvious obstacles; according to the perimeter calculation formula:

[0139] ,

[0140] The number of boundary nodes is set to 200, and the ideal spacing between adjacent nodes is... Environmental monitoring parameter thresholds are set as follows: temperature change threshold ±5℃, wind speed change threshold ±3m / s, obstacle density change threshold ±15%. Distributed computing node configuration requirements: 4-core CPU, 8GB memory, network latency <100ms. Anomaly detection performance requirements: training accuracy ≥95%, response time ≤2 seconds.

[0141] In actual testing, on an edge computing node with a 4-core CPU and 8GB of memory, the SVM model inference time averaged 0.8 seconds, meeting the ≤2-second response requirement. The model employs feature dimension compression technology, selecting the top 15 most important dimensions from the original 50 features for inference, with an accuracy loss of <3%. Feature importance evaluation uses the information gain method, selecting the top 15 feature dimensions with the highest information gain values ​​to ensure that key behavioral features are preserved.

[0142] This invention decomposes the complex problem of environmental adaptability into a collaborative reconstruction of three fundamental principles: sensor physical characteristics, the mathematical essence of optimization algorithms, and computational resource allocation. Traditional solutions treat multi-sensor fusion as a static configuration, while this invention starts from the physical characteristics of sensors: ultrasonic waves scattering under strong light, infrared radiation attenuating in haze, and the all-weather stability of millimeter-wave radar. By establishing a quantitative mapping between environmental parameters and sensor error rates (where the ultrasonic error rate is >25% when illuminance >10000 lux), the weight allocation is transformed into a calculation based on an inverse proportional function of signal-to-noise ratio, achieving a fundamental shift in data fusion from empirical judgment to a physical model-driven approach.

[0143] Simultaneously, the boundary optimization problem is reduced to a mathematical optimization objective of the node position distribution. Traditional gradient descent is prone to getting trapped in local optima in complex terrain. This invention introduces a momentum term and an adaptive learning rate mechanism, and ensures the mathematical determinism of a smooth transition of the boundary shape through a convergence condition of a descent rate of <0.1% for 10 consecutive iterations.

[0144] The computational task is decomposed into a load scoring model based on processor, memory, and network status (0.4 + 0.3 + 0.3 = 1), transforming resource allocation from simple round-robin to optimal solution search based on real-time load vectors. Anomaly detection uses information gain to compress feature dimensions (50-dimensional → 15-dimensional), reducing inference time to 0.8 seconds while maintaining 95% accuracy, achieving a physical balance between detection accuracy and real-time performance.

[0145] This invention fundamentally solves the three major technical contradictions faced in dynamic environments: fluctuations in sensor reliability, poor convergence of boundary optimization, and low utilization of computing resources, and realizes a paradigm shift from experience-based adaptation to model-driven approaches.

[0146] Example 2:

[0147] In substation outdoor protection scenarios, edge nodes are mostly powered by lithium batteries. The dynamic task allocation strategy in Example 1 only considers load balancing and does not address node energy consumption. Edge nodes have short battery life, averaging only 24 hours, requiring frequent battery replacements and resulting in high maintenance costs. Sensors continue to operate even in stable environments, causing unnecessary energy waste and shortening sensor lifespan, averaging only 12 months. There is a trade-off between energy consumption and protection performance; simply reducing energy consumption may lead to delayed detection response, failing to meet the real-time safety protection requirements of substations, which require a response time of ≤2 seconds. Based on this, this embodiment is a further optimization of Example 1. In this embodiment, the dynamic task allocation strategy also includes coordinated control of energy consumption monitoring and sensor sleep mode. The specific steps are as follows:

[0148] (1) Monitor the energy consumption parameters of each distributed computing node in real time, including CPU energy consumption (unit: W), memory energy consumption (unit: W) and network communication energy consumption (unit: W / transmission), and the real-time energy consumption value of the computing node. ;

[0149] (2) Construct a dual-objective comprehensive scoring model of energy consumption and load:

[0150] ;

[0151] in, The comprehensive load score described in the embodiment. The rated energy consumption of the node is preset to 2.5W.

[0152] (3) When And environmental change influencing factors When the sensor sleep mechanism is triggered: the ultrasonic sensor and infrared sensor enter sleep mode, and the sleep cycle is [number missing]. The millimeter-wave radar maintains 1 / 3 of its operating frequency; when it detects... or This will wake up all sensors and restore them to normal operation.

[0153] Furthermore, the real-time energy consumption value of the node The calculation method is as follows:

[0154] ;

[0155] Among them: CPU energy consumption coefficient ; Network communication energy consumption coefficient ; CPU real-time utilization (in %) Real-time memory utilization (in %) This represents the number of times data is transmitted between nodes within 5 minutes. , and The values ​​were obtained through hardware power consumption testing and calibration. The specific values ​​were determined based on power consumption model experiments of a typical edge node (4-core CPU, 8GB memory). In actual applications, recalibration can be performed according to the specific hardware configuration.

[0156] In practice, the parameters are set as follows:

[0157] The edge nodes are powered by lithium batteries (rated capacity 10Ah, rated voltage 3.7V), and the node's rated power consumption is [not specified]. The energy consumption coefficient was calibrated through experiments: , , Environmental change influencing factors The calculation uses the temperature, wind speed, and obstacle density parameters from the original patent, with the threshold set to 0.2.

[0158] Real-time energy consumption calculation for a node: When the energy consumption of a certain edge node... (If there is no data transmission within 5 minutes), according to the formula:

[0159] ;

[0160] Substituting into the formula, we get:

[0161] ;

[0162] Based on the dual-objective comprehensive score calculation, the comprehensive load score of this node is as follows:

[0163] ;

[0164] According to the formula: ;

[0165] Substituting into the formula, we get:

[0166] 0.306, close to the 0.3 threshold;

[0167] when When, substitute into the formula:

[0168] ;

[0169] ;

[0170] ;

[0171] and This triggers the sensor's sleep mechanism.

[0172] After triggering sleep mode, the ultrasonic sensor and infrared sensor enter sleep mode. The millimeter-wave radar's operating frequency was reduced from 2Hz to 0.67Hz. When a person is detected approaching, it causes... Immediately, all sensors are activated, and the millimeter-wave radar resumes its 2Hz frequency. The value was recalculated to 0.69 (≥0.5), and the detection tasks were allocated according to the original load balancing strategy.

[0173] Through testing, the method described in this embodiment has achieved the following technical effects in outdoor substation protection scenarios: the edge node's battery life has been extended from 24 hours to 72 hours, a 200% improvement, significantly reducing battery replacement frequency; the average node energy consumption has decreased from 1.2W to 0.4W, a 66.7% reduction, with sensor power consumption in sleep mode only 8% of that during operation, avoiding unnecessary energy consumption; in actual deployment tests, with a scale of 200 edge nodes, the average response time from the occurrence of abnormal behavior to the generation of an early warning report is 1.8 ± 0.3 seconds, meeting the real-time requirement of ≤2 seconds; thus meeting the real-time requirements for substation safety protection; and the theoretical lifespan of the sensor has been extended from 12 months to 36 months. This embodiment, through a dual-objective optimization mechanism of energy consumption and performance, effectively solves the industry pain points of high energy consumption and short battery life of edge nodes without compromising protection performance.

[0174] 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.

[0175] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A safety fence protection method based on Internet of Things technology, characterized in that, The method comprises the following steps: Construct a virtual fence model based on multi-sensor data fusion, collect distance information, heat source distribution information and motion trajectory information of the target protection area through ultrasonic sensors, infrared sensors and millimeter wave radars respectively, and generate an initial fence boundary by using a confidence weighted fusion algorithm based on dynamic adjustment of environmental conditions; Based on the initial fence boundary, real-time monitoring of environmental change parameters is performed, and when the detection of obstacle distribution changes or meteorological condition changes exceeds a preset threshold, a dynamic boundary adjustment strategy is triggered, and the fence boundary shape is optimized through a position redistribution mechanism of the boundary nodes; The optimized fence boundary detection algorithm is decomposed into multiple detection sub-tasks using a distributed computing framework, dynamic task allocation is performed according to the real-time resource utilization of each edge computing node, and boundary detection is performed in parallel on the edge computing nodes in the target protection area; Based on multi-dimensional behavior characteristics, abnormal behavior recognition is performed, boundary node quantity change characteristics, motion trajectory shape characteristics and speed change characteristics are extracted, and a preset machine learning classification model is used to determine whether the behavior mode of the target measured object is abnormal; When abnormal behavior is detected, feedback scheduling optimization is performed according to the frequency of occurrence of historical behavior trajectories, the monitoring sampling frequency is dynamically adjusted, and optimized monitoring data is generated; based on the optimized monitoring data, a detailed abnormality report is generated through a multi-level judgment mechanism, the relevance of abnormal behavior and security fence protection vulnerabilities is analyzed, and the vulnerability detection result is output. 2.The safety fence protection method based on the Internet of Things technology according to claim 1, characterized in that, The confidence weighted fusion algorithm specifically includes: Under sufficient light conditions: when the illuminance is >10000 lux, the detection data of the infrared sensor and the millimeter wave radar are used, and the weight proportion of the ultrasonic sensor is reduced; Under adverse weather conditions: when the visibility is <500 m, the detection data of the millimeter wave radar and the ultrasonic sensor is used, and the weight proportion of the infrared sensor is reduced; Through dynamic adjustment of the weight of each sensor, weighted fusion calculation is performed on the multi-source detection data to generate a boundary coordinate set. 3.The security fence protection method based on the Internet of Things technology according to claim 1, characterized in that, The position redistribution mechanism of the boundary nodes includes: Defining an initial position set of the boundary nodes, establishing an association between the node positions and the environmental change parameters; When environmental changes are detected, the position adjustment amount of each node is calculated according to the degree of environmental change, and the greater the environmental change, the greater the position adjustment amount; Through multiple iterations, the total adjustment amount of all nodes is minimized to achieve smooth transition of the boundary shape.

4. The security fence protection method based on the Internet of Things technology according to claim 2, characterized in that, The dynamic task allocation strategy includes: Real-time monitoring of the processor utilization, memory occupancy and network communication delay of each edge computing node; According to the above detection data, a comprehensive load score is calculated, and the lower the load score, the stronger the node processing capability; The detection task is preferentially assigned to the node with the lowest comprehensive load score for execution to achieve load balancing.

5. The security fence protection method based on the Internet of Things technology according to claim 1, characterized in that, The multi-dimensional behavior characteristics include: Trajectory shape characteristics, including total trajectory length, trajectory curvature, average motion speed and maximum acceleration; Boundary interaction characteristics, including boundary node quantity change rate, motion direction change frequency and residence time in the boundary area. 6.The security fence protection method based on the Internet of Things technology according to claim 1, characterized in that, The construction of the machine learning classification model includes: Collect historical behavior data as training samples, and label the classification tags of normal behavior and abnormal behavior; Adopt support vector machine algorithm to learn the training samples, and establish the mapping relationship between behavior characteristics and classification results; Map the characteristics to high-dimensional space through kernel function, and find the optimal classification decision boundary.

7. A safety fence protection device based on Internet of Things technology, used for implementing the safety fence protection method based on Internet of Things technology in any one of claims 1-6, characterized in that, It includes: A multi-sensor fusion module for collecting environmental data through ultrasonic sensors, infrared sensors and millimeter wave radars, and generating initial fence boundaries using a confidence weighted fusion algorithm based on dynamic adjustment of environmental conditions; A dynamic boundary adjustment module for monitoring environmental change parameters and triggering a position redistribution mechanism of boundary nodes to optimize the fence boundaries; A distributed computing module for decomposing the detection algorithm into multiple sub-tasks, and dynamically allocating and executing the tasks in parallel according to the resource utilization of each edge computing node; An abnormal behavior recognition module for extracting multi-dimensional behavior characteristics and determining behavior abnormalities through a machine learning classification model; A feedback optimization module for dynamically adjusting the monitoring sampling frequency according to the frequency of historical behavior trajectories; A vulnerability detection module for generating abnormal reports through a multi-level judgment mechanism and analyzing vulnerability correlation.

8. The security fence protection device based on Internet of Things technology according to claim 7, characterized in that, The multi-sensor fusion module includes a weight configuration unit, a data fusion unit and a quality control unit, The weight configuration unit is used to dynamically adjust the confidence weight proportion of each sensor according to the environmental conditions; The data fusion unit is used to generate a boundary coordinate set through weighted fusion calculation; The quality control unit is used to verify the validity and consistency of the fused data. 9.The safety fence protection device based on the Internet of Things technology of claim 7, wherein, The dynamic boundary adjustment module includes an environment monitoring unit, a node distribution optimization unit and a boundary reconstruction unit; The environment monitoring unit is used to collect obstacle distribution and meteorological condition parameters in real time; The node distribution optimization unit is used to perform position redistribution calculation of boundary nodes; The boundary reconstruction unit is used to regenerate the fence boundaries according to the optimized node distribution.

Images