A regional alarm system based on infrared electronic fence array

Abstract

The application discloses a regional alarm system based on an infrared electronic fence array, relates to the technical field of intelligent monitoring, and collects multidimensional data through an infrared electronic fence array and multiple sensors of a multi-source sensing module; a learning feature extraction module adopts a dynamic weighted aggregation strategy, generates a global space-time correlation matrix through a multi-model parallel architecture; a dynamic risk prediction module inputs the global space-time correlation matrix and a construction safety knowledge graph into a space-time graph neural network, and outputs an accurate risk probability value in combination with a cellular automaton simulator; and an intelligent execution module divides three alert zones and triggers a ladder response based on a risk probability value and a dynamic safety distance. The application solves the problems of poor adaptability and lagging risk prediction of traditional protection, realizes real-time control of construction risks, and improves the accuracy of safety protection and work efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and more specifically to a regional alarm system based on an infrared electronic fence array. Background Technology

[0002] In construction scenarios (especially construction on railway lines near operational levels), there are many irregular terrains such as deep foundation pits and tower crane operation areas, and frequent cross-operations, which pose multiple safety risks such as unauthorized entry, equipment failure, and personnel fatigue.

[0003] Traditional safety protection relies on physical fences and dedicated personnel for monitoring, but these traditional methods have obvious limitations. First, physical fences and warning lines have limited interception effects and relatively insufficient warning function, failing to effectively prevent personnel from accidentally entering dangerous areas. Second, on-site monitoring requires a large investment of manpower, and the monitoring personnel themselves may become fatigued due to long working hours, affecting the monitoring effect. In addition, traditional warning zones often need to be set much larger than the actual safe distance, which not only wastes resources but may also cause unnecessary interference with normal operations. Existing alarm systems often lack multi-source data fusion capabilities, resulting in insufficient risk prediction accuracy and failing to achieve early warning and precise control of risks.

[0004] Therefore, how to achieve precise control of construction risks is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a regional alarm system based on an infrared electronic fence array. By collecting multi-source data and combining dynamic safety distance, it realizes the division of warning areas and achieves real-time control of construction risks.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An area alarm system based on an infrared electronic fence array includes: The multi-source sensing module collects multi-source heterogeneous data from the construction scene, including an infrared electronic fence array, a personnel physiological sensing unit, an equipment status sensing unit, and an environmental sensing unit. The feature extraction module employs a dynamic weighted aggregation strategy to perform distributed feature extraction on the multi-source heterogeneous data, generating a global spatiotemporal correlation matrix. The dynamic risk prediction module inputs the global spatiotemporal correlation matrix and the construction safety knowledge graph into a preset spatiotemporal graph neural network to obtain a fused feature matrix; the cellular automaton simulator uses the fused features as input and outputs the spatiotemporal location and the risk probability value corresponding to the location. The intelligent execution module divides the warning zone based on the risk probability value and the preset dynamic safety distance, and triggers a graded response operation.

[0007] Preferably, multi-view 3D point cloud data is obtained by scanning the construction area, and the multi-view 3D point cloud data is fused and registered using the iterative nearest point algorithm. A triangular mesh terrain model is generated based on the registered data, and the key areas in the triangular mesh terrain are segmented using the region growing algorithm to form a terrain topology map. The 3D deployment coordinates of the infrared electronic fence array are generated by using the terrain topology map as input and the PSO algorithm.

[0008] Preferably, the distributed feature extraction adopts a multi-model parallel architecture: the environmental sensing unit collects environmental perception data and extracts temporal dependency features through an LSTM network; the device status sensing unit collects device status data and extracts abnormal pattern features through an autoencoder; the personnel physiological sensing unit collects physiological data of employees and extracts group interaction features through a graph attention network, with feature dimensions including heart rate, blood oxygen, fatigue level, and location coordinates; all features are aligned to spatiotemporal resolution through deformable convolution and a cross-modal attention mechanism is used to establish dynamic associations.

[0009] Preferably, the dynamic weighted aggregation strategy specifically includes: ; in, The aggregate weight of the h-th data source; This is a time decay factor that is dynamically adjusted according to the data acquisition duration. The spatiotemporal coverage completeness of the h-th data source; Let h be the noise level of the h-th data source; Let h be the local feature matrix of the h-th data source; The global spatiotemporal correlation matrix has dimensions of . N is the number of monitoring points, T is the time step, D is the feature dimension, and m is the total number of data sources.

[0010] Preferably, the dynamic safety distance formula is: ; in, β This is a correction factor for the machine type; R m The rated safe operating radius for construction machinery; r Personnel density in the work area; i To monitor the slope of the terrain in the area; c This is the equipment risk amplification factor; H For equipment health; L This refers to the cumulative operating time since the equipment's most recent maintenance. d For safety measure compensation coefficient; S To ensure the implementation of safety measures, when personnel are detected entering a deep foundation pit or within the slewing radius of a tower crane, a 15% safety distance expansion is triggered.

[0011] Preferably, the output spatiotemporal location and the corresponding risk probability value specifically include:

[0012] in, For the graph embedding vector of the construction safety knowledge graph; Let be the risk probability value at the spatiotemporal location (t,s). For spatial deviation, For time step deviation, The gated convolution feature weight matrix is... GatedConv ( X ) represents the gated spatiotemporal convolution operation. Conv3D ( X () represents a 3D convolution operation. W t Here is the Transformer feature weight matrix. For a cellular automaton simulator, the state transition formula is: ; Where b is the bias term. N ( s ) represents the neighborhood of cell s; P This is a multi-granularity risk perception module. The output results are, in order, device-level, regional-level, and global-level risk results. For cells s The risk probability value at time t+1 Let be the weight coefficient of the neighboring cell s'. Let be the risk probability value of the neighboring cell s' at time t. Let be the cosine similarity between the knowledge graph entities corresponding to cell s and their neighboring cells s'.

[0013] Preferably, the delineation of the warning zone specifically includes: Warning zone ,scope This triggers an audio-visual warning and simultaneously pushes a voice prompt to the construction worker's smart helmet; Quarantine area ,scope Initiate directional acoustic wave dispersal; Restricted area ,scope The power supply to the equipment is cut off, and the rescue information is pushed to the cloud-based collaborative platform.

[0014] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a regional alarm system based on an infrared electronic fence array. It optimizes the deployment of the infrared electronic fence array through visual SLAM and PSO algorithms, accurately adapts to irregular terrains such as deep foundation pits and tower crane operation areas, and overcomes the limitations of traditional fence coverage. It integrates multi-source heterogeneous data and construction safety knowledge graphs, and uses spatiotemporal graph neural networks to achieve accurate risk prediction, avoiding the misjudgment problem of single data monitoring. It divides three-level warning zones based on dynamic safety distance and risk probability values ​​to achieve tiered intelligent response, balances safety protection and construction efficiency, and has good scene adaptability and flexibility. It can be widely used in various construction and high-risk operation areas. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the structure provided by the present invention; Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention discloses a regional alarm system based on an infrared electronic fence array, such as... Figure 1 As shown, it includes: The multi-source sensing module collects multi-source heterogeneous data from the construction scene, including an infrared electronic fence array, a personnel physiological sensing unit, an equipment status sensing unit, and an environmental sensing unit. The feature extraction module employs a dynamic weighted aggregation strategy to perform distributed feature extraction on multi-source heterogeneous data and generate a global spatiotemporal correlation matrix. The dynamic risk prediction module inputs the global spatiotemporal correlation matrix and the construction safety knowledge graph into a preset spatiotemporal graph neural network to obtain a fused feature matrix; the cellular automaton simulator takes the fused features as input and outputs the spatiotemporal location and the risk probability value corresponding to the location; among them, the construction safety knowledge graph includes construction element entities, safety specification constraints and accident case correlations; The intelligent execution module divides warning zones and triggers tiered response operations based on risk probability values ​​and preset dynamic safety distances.

[0019] In one specific embodiment, the spatiotemporal graph neural network includes: Spatial graph convolutional branch (capturing spatial correlations of monitoring points) Graph structure construction: Based on the topology map of the construction area, construct a spatial association graph G=(V,E), where node V corresponds to a monitoring point, and the weight of edge E is determined by the distance between monitoring points and the occlusion situation; Operation module: A 2-layer stacked graph attention network (GAT) is used, with 4 attention heads per layer and LeakyReLU activation function (negative slope 0.2). Level 1 GAT: Input H norm Spatial slices (dimension N×D) are output with dimension N×128, and the feature weights of monitoring points in high-risk areas are enhanced through an attention mechanism. Layer 2 GAT: Integrates knowledge graph node embeddings, concatenating the features of each monitoring point with the knowledge graph entity embeddings of the corresponding region (dimension N×(128+128)), outputting a spatial feature matrix of dimension N×128. F space .

[0020] Temporal Transformer branch (capturing temporal dependencies)

[0021] Module structure: It adopts a 3-layer spatiotemporal Transformer encoder, with each layer containing a spatiotemporal multi-head attention mechanism and a feedforward neural network (FFN). Spatiotemporal attention mechanism: Simultaneously calculates temporal and spatial attention, with a temporal attention window size of 5 (i.e., each time step is associated with two adjacent time steps), and spatial attention is based on the monitoring point association graph G constraint; Feedforward neural networks (FFNs) enhance the nonlinear representation of temporal features; Output: After 3 layers of encoding, a time feature matrix of dimension T×N×128 is obtained. F time The transposed value is N×T×128, which is aligned with the output dimension of the spatial branch.

[0022] 3. Cross-modal attention alignment layer

[0023] Core functionality: Aligning spatial branches F space With time branch F time The feature semantics are enhanced to strengthen the feature expression of key spatiotemporal combinations; Operating procedure:F space (N×128) is expanded to N×T×128, and... F time The concatenated features are N×T×256; the spatiotemporal alignment weights are calculated using a single-head attention mechanism, and the aligned spatiotemporal features are finally output. F st = W align F time (Dimensions N×T×128).

[0024] Knowledge Graph Gating Fusion Layer

[0025] Gating mechanism design: Gated convolution is introduced to integrate spatiotemporal features and knowledge graph constraints. The formula is as follows: ; Integration process: Aligned spatiotemporal features F st Embedded with the mapped knowledge graph KG′ embed (Dimensions N×T×128) are concatenated to obtain a mixed feature of N×T×256; Input a gated convolutional layer (kernel size 3×3×3, stride 1, padding=1), output a gated fusion feature Fgate with dimension N×T×256; A linear transformation is performed using the Transformer feature weight matrix Wt (256×256 dimensions) to strengthen the constraints of security rules on features.

[0026] Deformable convolution enhancement layer adapts to spatiotemporal distortions caused by irregular terrain in the construction area and dynamically adjusts the convolution receptive field; Operational parameters: Two layers of deformable convolution (DeformableConv3D) are used, with a kernel size of 3×3×3, an offset range of [-2,2], and the activation function is ReLU; the output dimension of each layer is kept at N×T×256, and feature bias caused by terrain occlusion is corrected by dynamic sampling.

[0027] Global average pooling is performed on the enhanced features to compress the time dimension T, resulting in an N×256 fused feature matrix. F final ; In one specific embodiment, multi-view 3D point cloud data is obtained by scanning the construction area. The multi-view 3D point cloud data is fused and registered using the iterative nearest point algorithm. A triangular mesh terrain model is generated based on the registered data. The key areas in the triangular mesh terrain are segmented using the region growing algorithm to form a terrain topology map. The 3D deployment coordinates of the infrared electronic fence array are generated using the PSO algorithm with the terrain topology map as input.

[0028] In one specific embodiment, distributed feature extraction employs a multi-model parallel architecture: the environmental sensing unit collects environmental perception data and extracts temporal dependency features through an LSTM network; the device status sensing unit collects device status data and extracts abnormal pattern features through an autoencoder; the personnel physiological sensing unit collects physiological data of employees and extracts group interaction features through a graph attention network, with feature dimensions including heart rate, blood oxygen, fatigue level, and location coordinates; all features are aligned to spatiotemporal resolution through deformable convolutions, and a cross-modal attention mechanism is used to establish dynamic associations.

[0029] Data Acquisition Principle: A mobile platform equipped with a LiDAR and an RGB camera performs a comprehensive scan of the construction area to acquire multi-view 3D point cloud data. ,in For the first i The three-dimensional coordinates of each point, where N is the total number of points in the point cloud.

[0030] The Iterative Closest Point (ICP) algorithm is used to achieve multi-view point cloud fusion, with the objective function being to minimize the Euclidean distance between corresponding points. ; in: For target point cloud data, The source is point cloud data, and R is the rotation matrix. Let M be the translation vector and M be the number of corresponding point pairs; the iteration stopping condition is... .

[0031] 3D terrain modeling: Based on the registered point cloud data, a triangular mesh terrain model is generated. The terrain surface is then fitted using a Poisson surface reconstruction algorithm, with the following formula: ; in: It is an implicit function. This is the weighted density function for the point cloud normal vectors.

[0032] Construction area topology extraction: Key regions in the terrain are segmented using a region growing algorithm, and topology nodes are extracted. Where K is the number of topological nodes; topological edges This forms the initial deployment topology graph G=(V,E).

[0033] The PSO algorithm takes the terrain topology map generated by visual SLAM as input and optimizes the three-dimensional deployment coordinates of the infrared electronic fence array to adapt to the protection needs of irregular terrain.

[0034] Each particle The three-dimensional deployment coordinates of m infrared sensors are given. Let m be the coordinates of the i-th sensor in the k-th particle (i=1,2,...,m). The number of sensors m is determined by the area of ​​the monitoring region (1 sensor is deployed for every 100m²).

[0035] Fitness function: Combining coverage integrity, blind spot area, and deployment cost, the formula is:

[0036] in: To ensure complete coverage, The area covered by the sensor's laser grid. The total area of ​​the construction area; The percentage of blind spots, This refers to the area of ​​the uncovered region. To reduce deployment density costs and avoid excessive sensor density; The weights are determined through construction scenario tests.

[0037] PSO Iterative Optimization Formula

[0038] The velocity update formula considers the dynamic decay of inertia weights to balance global search and local convergence.

[0039] in: Let the velocity of the t-th generation particle k in the i-th dimension be a velocity constraint. ; For dynamic inertia weights, This represents the maximum number of iterations. As a learning factor, It is a random number; Let i be the historical best position of particle k in the i-th dimension. This is the globally optimal position.

[0040] The location update formula adjusts the sensor coordinates based on terrain constraints: ; Position constraints: Horizontal direction (construction area boundary): ; Vertical direction: ,in This represents the elevation value of the terrain elevation model at (x,y), ensuring that the vertical layer spacing is adjustable.

[0041] Output: Global optimal particle The corresponding sensor 3D deployment coordinates are used to deploy an infrared electronic fence array based on these coordinates. The laser grid spatial positioning accuracy is ≤10cm, and each layer of laser independently detects the location of intruders.

[0042] In one specific embodiment, the dynamic weighted aggregation strategy includes: ; in, The aggregate weight of the h-th data source; This is a time decay factor that is dynamically adjusted according to the data acquisition duration. The spatiotemporal coverage completeness of the h-th data source; Let h be the noise level of the h-th data source; Let h be the local feature matrix of the h-th data source; The global spatiotemporal correlation matrix has dimensions of . N is the number of monitoring points, T is the time step, D is the feature dimension, and m is the total number of data sources.

[0043] In one specific embodiment, the dynamic safety distance formula is: ; Where β∈[1.0,1.5] is the machine type correction coefficient; R m The rated safe operating radius of the construction machinery is taken from the preset parameters of the equipment status sensing unit. r The personnel density in the work area is calculated jointly by an infrared electronic fence array and a visual sensor. i To monitor the terrain slope of the area, a visual SLAM algorithm is used to obtain the slope. c H represents the equipment risk amplification factor; H is the equipment health status, calculated from equipment vibration, current, and other data through a self-encoder. L This refers to the cumulative operating time since the equipment's most recent maintenance. d For safety measure compensation coefficient; S To ensure the implementation of safety measures, when personnel are detected entering a deep foundation pit or within the slewing radius of a tower crane, a 15% safety distance expansion is triggered.

[0044] In a specific embodiment, the output of the spatiotemporal location and the corresponding risk probability value specifically includes:

[0045] in, For the graph embedding vector of the construction safety knowledge graph; Let be the risk probability value at the spatiotemporal location (t,s). For spatial deviation, For time step deviation, The gated convolution feature weight matrix is... GatedConv ( X ) represents the gated spatiotemporal convolution operation. Conv3D ( X () represents a 3D convolution operation. W t Here is the Transformer feature weight matrix. For a cellular automaton simulator, the state transition formula is: ; Where b is the bias term. N ( s ) represents the neighborhood of cell s; P This is a multi-granularity risk perception module. The output results are, in order, device-level, regional-level, and global-level risk results. For cells s The risk probability value at time t+1 Let be the weight coefficient of the neighboring cell s'. Let be the risk probability value of the neighboring cell s' at time t. Let be the cosine similarity between the knowledge graph entities corresponding to cell s and their neighboring cells s'.

[0046] In one specific embodiment, the delineation of the warning zone includes: Warning zone ,scope This triggers an audio-visual warning and simultaneously pushes a voice prompt to the construction worker's smart helmet; Quarantine area ,scope Initiate directional acoustic wave dispersal; Restricted area ,scope The power supply to the equipment is cut off, and the rescue information is pushed to the cloud-based collaborative platform.

[0047] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the methods disclosed in the embodiments; relevant parts can be found in the method section.

[0048] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A regional alarm system based on an infrared electronic fence array, characterized in that, include: Multi-source sensing module to collect multi-source heterogeneous data from the construction scene; This includes an infrared electronic fence array, a personnel physiological sensing unit, an equipment status sensing unit, and an environmental sensing unit; The feature extraction module employs a dynamic weighted aggregation strategy to perform distributed feature extraction on the multi-source heterogeneous data, generating a global spatiotemporal correlation matrix. The dynamic risk prediction module inputs the global spatiotemporal correlation matrix and the construction safety knowledge graph into a preset spatiotemporal graph neural network to obtain a fusion feature matrix; The cellular automaton simulator takes the fused features as input and outputs the spatiotemporal location and the risk probability value corresponding to the location. The intelligent execution module divides the warning zone based on the risk probability value and the preset dynamic safety distance, and triggers a graded response operation.

2. The area alarm system based on an infrared electronic fence array according to claim 1, characterized in that, By scanning the construction area, multi-view 3D point cloud data is obtained. The iterative nearest point algorithm is used to fuse and register the multi-view 3D point cloud data. Based on the registration data, a triangular mesh terrain model is generated. The region growing algorithm is used to segment the key areas in the triangular mesh terrain to form a terrain topology map. Using the terrain topology map as input, the PSO algorithm is used to generate the three-dimensional deployment coordinates of the infrared electronic fence array.

3. A regional alarm system based on an infrared electronic fence array according to claim 1, characterized in that, The distributed feature extraction adopts a multi-model parallel architecture: the environmental sensing unit collects environmental perception data and extracts time-dependent features through an LSTM network; The device status sensing unit collects device status data and extracts abnormal pattern features through an autoencoder. The personnel physiological sensing unit collects physiological data of employees and extracts group interaction features through graph attention network. The feature dimensions include heart rate, blood oxygen, fatigue level, and location coordinates. All features are aligned to spatiotemporal resolution through deformable convolution, and dynamic associations are established using a cross-modal attention mechanism.

4. A regional alarm system based on an infrared electronic fence array according to claim 1, characterized in that, The adoption of the dynamic weighted aggregation strategy specifically includes: ； in, The aggregate weight of the h-th data source; This is a time decay factor that is dynamically adjusted according to the data acquisition duration. The spatiotemporal coverage completeness of the h-th data source; Let h be the noise level of the h-th data source; Let h be the local feature matrix of the h-th data source; The global spatiotemporal correlation matrix has dimensions of . N is the number of monitoring points, T is the time step, D is the feature dimension, and m is the total number of data sources.

5. A regional alarm system based on an infrared electronic fence array according to claim 1, characterized in that, The dynamic safety distance formula is: ； in, β This is a correction factor for the machine type; R m The rated safe operating radius for construction machinery; ρ Personnel density in the work area; θ To monitor the slope of the terrain in the area; γ This is the equipment risk amplification factor; H For equipment health; L This refers to the cumulative operating time since the equipment's most recent maintenance. δ For safety measure compensation coefficient; S To ensure the implementation of safety measures, when personnel are detected entering a deep foundation pit or within the slewing radius of a tower crane, a 15% safety distance expansion is triggered.

6. A regional alarm system based on an infrared electronic fence array according to claim 4, characterized in that, The output spatiotemporal location and the corresponding risk probability value specifically include: in, For the graph embedding vector of the construction safety knowledge graph; Let be the risk probability value at the spatiotemporal location (t,s). For spatial deviation, For time step deviation, The gated convolution feature weight matrix is... GatedConv ( X ) represents the gated spatiotemporal convolution operation. Conv3D ( X () represents a 3D convolution operation. W t Here is the Transformer feature weight matrix. For a cellular automaton simulator, the state transition formula is: ； Where b is the bias term. N ( s ) represents the neighborhood of cell s; Ψ This is a multi-granularity risk perception module. The output results are, in order, device-level, regional-level, and global-level risk results. For cells s The risk probability value at time t+1 Let be the weight coefficient of the neighboring cell s'. Let be the risk probability value of the neighboring cell s' at time t. Let be the cosine similarity between the knowledge graph entities corresponding to cell s and their neighboring cells s'.

7. A regional alarm system based on an infrared electronic fence array according to claim 6, characterized in that, The delineation of warning zones specifically includes: Warning zone ,scope This triggers an audio-visual warning and simultaneously pushes a voice prompt to the construction worker's smart helmet; Quarantine area ,scope Initiate directional acoustic wave dispersal; Restricted area ,scope The power supply to the equipment is cut off, and the rescue information is pushed to the cloud-based collaborative platform.

Images