Intelligent security and protection method and device based on dynamic environment perception and medium

Abstract

The invention provides an intelligent security and protection method and device based on dynamic environment perception and a medium, and belongs to the technical field of intelligent security and protection. The method comprises the following steps: generating a time sequence dynamic parameter group sequence based on a multi-source dynamic sensing module pre-deployed in a scene key area; generating a dynamic environment situation map according to the plurality of dynamic scene partitions and the time sequence dynamic parameter group sequence; and determining each first risk abnormal region based on each dynamic environment situation map, a historical security event database and a preset risk assessment algorithm. Inputting the dynamic environment situation map group corresponding to each first risk anomaly region into a pre-trained space-time attention coupling model to determine a risk coupling region and a second risk situation curve corresponding to the risk coupling region so as to determine a second risk anomaly region in the monitoring region; and according to the second risk abnormal region and a preset multi-dimensional intervention strategy matrix, generating graded early warning information and a self-adaptive intervention instruction, and sending the graded early warning information and the self-adaptive intervention instruction to the user terminal.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent security and protection, and in particular to an intelligent security and protection method based on dynamic environment perception, a device and a medium. BACKGROUND

[0002] With the deep development of intelligent manufacturing and industrial internet technology, the operation scale of complex scenes such as mines, ports, chemical parks and large factory areas is continuously expanding, and the degree of automation of equipment and the mobility of personnel are significantly improved, which puts unprecedented high requirements on the real-time, accuracy and dynamic adaptability of intelligent security and protection technology.

[0003] For the security and protection monitoring of the above-mentioned scenes, most of them still rely on manual background monitoring, although there are also some scenes where intelligent security and protection devices are set, but most of them rely on single perception dimension for risk judgment. The inventor found that the traditional intelligent security and protection is difficult to adapt to the above-mentioned security and protection scenes with dynamic and variable environment, multi-target collaborative operation and complex and superimposed risk types. SUMMARY

[0004] To solve the above problems, the embodiments of the present application provide an intelligent security and protection method based on dynamic environment perception, a device and a medium, which are used to adapt to dynamic environment, accurately process multiple risk couplings, and have risk prediction ability to realize intelligent security and protection.

[0005] In one aspect, the embodiments of the present application provide an intelligent security and protection method based on dynamic environment perception, which is applied to an intelligent security and protection system comprising a pre-set sensor network; the method comprises: determining multi-dimensional dynamic parameters in a monitoring area based on a multi-source dynamic perception module pre-deployed in a key area of the scene, to generate a time sequence dynamic parameter group sequence; generating a dynamic environment situation graph corresponding to each dynamic scene partition according to a plurality of dynamic scene partitions and the time sequence dynamic parameter group sequence; determining each first risk abnormal area based on each dynamic environment situation graph, a historical security and protection event database and a pre-set risk assessment algorithm; inputting a dynamic environment situation graph corresponding to each first risk abnormal area into a pre-trained space-time attention coupling model to determine a risk coupling area and a second risk situation curve corresponding thereto; determining a second risk abnormal area in the monitoring area based on the second risk situation curve and a pre-set scene positioning rule; generating graded early warning information and self-adaptive intervention instructions according to the second risk abnormal area and a pre-set multi-dimensional intervention strategy matrix, and sending them to a user terminal to continuously and dynamically monitor and actively intervene in the monitoring area.

[0006] In one implementation of this application, before generating a dynamic environmental situation map corresponding to each of the multiple dynamic scene partitions and the time-series dynamic parameter group sequence, the method further includes: Acquire basic geographic information data, operational data, and historical safety event data within a preset time period corresponding to the monitored area to construct a partitioned basic dataset; Based on the partitioned basic dataset, the preset risk-bearing capacity assessment dimensions, and the preset clustering algorithm, the monitoring area is partitioned. Based on the clustering results, the initial partition boundaries within the monitoring area are determined, and each initial partition boundary is corrected by using preset boundary topology rules and risk propagation direction rules to obtain each dynamic scene partition. The corrected dynamic scene partitions use the coverage area of ​​the corresponding monitoring sub-region along the risk propagation direction as the partition boundary, forming a non-overlapping closed-loop partition within the monitoring area.

[0007] In one implementation of this application, a dynamic environmental situation map corresponding to each of the dynamic scene partitions is generated based on multiple dynamic scene partitions and the time-series dynamic parameter group sequence, specifically including: Based on the acquisition location coordinates of each dynamic parameter in the time-series dynamic parameter group sequence, spatial matching is performed with the boundary coordinates of each dynamic scene partition to add multi-dimensional dynamic parameters within the same dynamic scene partition to the partition-related parameter set; wherein, the multi-dimensional dynamic parameters include at least environmental state parameters, target behavior parameters, and operation parameters; Based on the first preset parameter threshold matrix corresponding to each of the dynamic scene partitions, the partition-related parameter set is compared with the corresponding first preset parameter threshold matrix to determine the risk status level of the corresponding dynamic scene partition according to the first comparison result; Based on the risk situation level and historical security event data within a preset time period, the environmental situation gain coefficient corresponding to the risk situation level is fine-tuned. Based on the partition association parameter set, the risk situation level, and the fine-tuned environmental situation gain coefficient, the dynamic environmental situation map corresponding to the dynamic scene partition is constructed in chronological order.

[0008] In one implementation of this application, each first risk anomaly area is determined based on the dynamic environmental situation map, historical security event database, and preset risk assessment algorithm, specifically including: Real-time multi-dimensional dynamic parameters and environmental situation gain coefficients are extracted from each of the dynamic environmental situation maps, and the single risk type weight vector and the second preset parameter threshold matrix of the dynamic scene partition are called from the historical security event database; wherein, the second preset parameter threshold matrix matches the risk carrying capacity of the dynamic scene partition; Calculate the corresponding parameter deviation quantization value based on each of the multi-dimensional dynamic parameters and the second preset parameter threshold matrix; The basic risk value of the dynamic scenario partition is calculated based on the single risk type weight vector and the parameter deviation quantization value. The risk assessment value of the dynamic scene partition is determined based on the product of the basic risk value and the environmental situation gain coefficient. The risk assessment value of each dynamic scene partition is compared with a preset anomaly judgment threshold to determine each of the first risk anomaly areas based on the second comparison result.

[0009] In one implementation of this application, a dynamic environmental situation map group corresponding to each of the first risk anomaly regions is input into a pre-trained spatiotemporal attention coupling model to determine the risk coupling region and its corresponding second risk situation curve, specifically including: The dynamic environmental situation map group is feature-encoded and dimension-mapped to construct a spatiotemporal feature matrix; wherein, the spatiotemporal feature matrix includes a time feature dimension and a spatial feature dimension; the time feature dimension includes at least an environmental state parameter vector, a target behavior parameter vector, a risk situation level encoding value, and an environmental situation gain coefficient; the spatial feature dimension includes at least the regional representation coordinate information of each of the first risk anomaly areas. The spatiotemporal feature matrix is ​​input into the pre-trained spatiotemporal attention coupling model to perform weighted fusion of the spatiotemporal feature matrix through temporal attention and spatial attention mechanisms. Density clustering is then performed on the fused spatiotemporal features to identify regions that meet a preset density connectivity criterion as risk coupling regions. The spatial attention mechanism extracts associated features of multiple risk types within the same spatiotemporal space based on the predicted target interaction range corresponding to the dynamic scene partition. The predicted target interaction range is smaller than the spatial range of the monitoring area. The preset density connectivity criterion requires at least a preset number of risk features to be density-connected within the same spatiotemporal space. The preset number of risk features originate from the same dynamic scene partition and / or different dynamic scene partitions. Based on the risk assessment value corresponding to the risk coupling region, and according to a preset time step, the corresponding risk gradient value is calculated, and the second risk situation curve is generated based on each of the risk gradient values, with time as the horizontal axis.

[0010] In one implementation of this application, determining a second risk anomaly area within the monitoring area based on the second risk situation curve and preset scenario positioning rules specifically includes: Based on preset anomaly judgment conditions, the second risk situation curve is preprocessed to extract corresponding key feature points in order to construct the first anomaly interval. Based on the basic geographic information data of the dynamic scene partition and the deployment location calibration data corresponding to the multi-source dynamic perception module, a corresponding positioning fitting equation is constructed; the positioning fitting equation includes the mapping relationship between the time dimension features of the second risk situation curve and the physical spatial coordinates of the monitoring area; According to the preset anomaly determination conditions, a preset sliding window slides along the second risk situation curve in the corresponding curve segment of the first anomaly interval, and determines in real time whether there is abnormal data in the preset sliding window; If so, the second risk anomaly region is determined based on the second anomaly interval corresponding to the abnormal data and the location fitting equation.

[0011] In one implementation of this application, determining the second risk anomaly region based on the second anomaly interval corresponding to the abnormal data and the localization fitting equation specifically includes: Determine the key time points and their corresponding risk gradient values ​​in the second abnormal interval; the key time points include at least the anomaly start time point, the risk gradient peak time point, and the anomaly termination time point. The positioning fitting equation is invoked to determine the key physical space coordinates corresponding to each of the key time points; Using the key physical spatial coordinate point corresponding to the peak time point of the risk gradient as the region center, the region expansion radius corresponding to the region center is determined based on the peak risk gradient and the duration of the anomaly. When there are no occlusion obstacles in the dynamic scene partition, a corresponding abnormal region boundary is generated based on the region center and the region expansion radius; When there are occlusions in the dynamic scene partition, the boundary is fitted according to the key physical space coordinates of the same second abnormal interval to generate the corresponding abnormal closed region boundary. The second risk anomaly region is generated based on the boundary of the anomaly region and / or the boundary of the anomaly closed region.

[0012] In one implementation of this application, based on the second risk anomaly region and a preset multi-dimensional intervention strategy matrix, hierarchical early warning information and adaptive intervention instructions are generated, specifically including: Based on the dynamic environmental situation map and the second risk situation curve corresponding to the second risk anomaly area, the corresponding risk situation level, risk gradient range and scenario type are determined; Based on the warning level corresponding to the risk gradient interval, the corresponding graded warning information is determined; The risk situation level, the risk gradient range, and the scenario type are encoded and matched with the preset multidimensional intervention strategy matrix to determine the corresponding combination of intervention methods based on the matching results. The corresponding intervention parameters are then configured according to the combination of intervention methods to generate the adaptive intervention instruction.

[0013] Secondly, embodiments of this application provide an intelligent security device based on dynamic environment perception, the device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aforementioned intelligent security method based on dynamic environment perception.

[0014] Thirdly, embodiments of this application provide a non-volatile computer storage medium storing computer-executable instructions, which are capable of executing the aforementioned intelligent security method based on dynamic environment perception.

[0015] Compared with the prior art, the significant advantages of this application are as follows: Through the above technical solution, this application utilizes a multi-source dynamic perception module at the monitoring site to collect multi-dimensional dynamic parameters and generate a dedicated situation map according to dynamic scene partitions. This allows for regional and differentiated situational quantitative assessments tailored to dynamic environments. Simultaneously, by leveraging the map and historical data, risks in individual partitions are identified, and a spatiotemporal attention coupling model is used to identify risk coupling areas that can cross partitions, enabling the prediction of implicit coupling risks. Furthermore, by combining preset scene positioning rules, the risk locations are accurately pinpointed, and tiered early warning and adaptive intervention are implemented, improving the level of intelligence in security. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating an intelligent security method based on dynamic environment perception in an embodiment of this application. Figure 2This is a schematic diagram of the structure of an intelligent security device based on dynamic environment perception in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] For security monitoring in complex scenarios such as mines, ports, chemical industrial parks, and large factories, most still rely on manual back-end monitoring. Although some scenarios have installed intelligent security equipment, they mostly rely on a single sensing dimension for risk assessment. Traditional intelligent security systems are difficult to adapt to these types of security scenarios characterized by dynamic and ever-changing environments, multi-target collaborative operations, and complex overlapping risk types.

[0019] Based on this, the embodiments of this application provide an intelligent security method, device and medium based on dynamic environment perception, to solve the problem that there is currently a lack of intelligent security technology that can adapt to dynamic environments, accurately handle multiple risk couplings and has risk prediction capabilities.

[0020] The various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0021] This application provides an intelligent security method based on dynamic environment perception. The method is applied to an intelligent security system including a pre-installed sensor network. This intelligent security system includes one or more image acquisition devices, multi-parameter sensors installed on the operating equipment, an edge computing gateway, and a user terminal. It may also include a cloud server, but this is not specifically limited. The multi-parameter sensors specifically include vibration sensors, dust sensors, and light sensors. This application uses an open-pit mine shovel and conveyor operation area as an application scenario to specifically describe the execution flow of the intelligent security method based on dynamic environment perception. Figure 1 As shown, the method may include steps S101-S106: S101, based on the multi-source dynamic perception module pre-deployed in the key areas of the scene, determines the multi-dimensional dynamic parameters within the monitoring area to generate a time-series dynamic parameter group sequence.

[0022] The key areas of a scenario can be selected by the user in the actual usage scenario or based on expert experience; no specific limitations are imposed here. For example, key areas are locations prone to accidents where the development of an accident can be known immediately. The multi-source dynamic sensing module consists of industrial vision acquisition equipment and lightweight environmental sensors with built-in inertial measurement units. It is a hardware collection capable of simultaneously acquiring visual data and environmental physical parameters, achieving multi-dimensional, blind-spot-free dynamic sensing. The multi-source dynamic sensing module constructs a sensor network and collects industrial data in industrial scenarios to obtain industrial big data.

[0023] Industrial vision acquisition equipment, also known as image acquisition equipment, is deployed evenly around the periphery of high-risk areas in the scene. The coverage overlap of adjacent devices exceeds a predetermined value, such as 15%, to eliminate blind spots. High-risk areas are pre-defined by the user based on the actual usage scenario and are not specifically limited here. Lightweight environmental sensors include vibration sensors, light sensors, dust sensors, and temperature and humidity sensors. These sensors can be fixedly installed in key areas of the scene or mounted on work vehicles to acquire risk data during operations.

[0024] The multi-source dynamic sensing module can collect multi-dimensional dynamic parameters within the monitored area in real time. These multi-dimensional dynamic parameters include at least industrial data such as environmental state parameters, target behavior parameters, and operational parameters. Following a chronological order, the multi-dimensional dynamic parameters at the same timestamp are packaged to generate a time-series dynamic parameter sequence. Environmental state parameters include vibration, dust, and lighting parameters; target behavior parameters include personnel and equipment-related parameters; and operational parameters include on-site operational time-series data and scheduling plans.

[0025] It should be noted that the executing entity of this application can be the controller of the edge computing grid, the user-preset field industrial control machine, or the server. The controller, as the executing entity of the intelligent security method based on dynamic environment perception, is only an example and the executing entity is not limited to this. This application does not make any specific limitation on this.

[0026] S102, based on multiple dynamic scene partitions and time-series dynamic parameter group sequences, generate dynamic environmental situation maps corresponding to each dynamic scene partition.

[0027] In this embodiment of the application, before generating the dynamic environmental situation map corresponding to each dynamic scene partition based on multiple dynamic scene partitions and time-series dynamic parameter group sequences, it is necessary to divide the monitoring area into multiple dynamic scene partitions in order to accurately perform security monitoring. This includes: The system acquires basic geographic information data, operational data, and historical security event data for the monitored area within a preset time period to construct a partitioned basic dataset. Based on this dataset, preset risk-bearing capacity assessment dimensions, and a preset clustering algorithm, the monitored area is partitioned. Based on the clustering results, the initial partition boundaries within the monitored area are determined. These initial boundaries are then corrected using preset boundary topology rules and risk propagation direction rules to obtain dynamic scene partitions. The corrected dynamic scene partitions use the coverage area of ​​the corresponding monitored sub-area along the risk propagation direction as their partition boundaries, forming non-overlapping closed-loop partitions within the monitored area.

[0028] Specifically, this application can pre-collect basic geographic information data of the monitoring area. This data includes at least the boundary coordinates of the monitoring area, terrain slope, and the distribution location of facilities corresponding to the multi-source dynamic sensing modules. Operational data includes equipment operating range, personnel activity trajectories, and operational schedules. Historical security incident data is stored in a historical security incident database, including the location of historical risks, event type, and degree of loss. These three types of data are standardized and pre-processed to form a partitioned basic dataset. This application pre-sets risk-bearing capacity assessment dimensions, specifically including personnel and equipment density, equipment sensitivity level, operational complexity, historical risk frequency, and environmental interference intensity, and pre-sets dimension weights for different dimensions. The sum of the weights of multiple assessment dimensions is 1. Subsequently, using a pre-set K-means clustering algorithm with a pre-set number of clusters, the partitioned basic dataset is analyzed to quantify each pre-set risk-bearing capacity assessment dimension. Based on the dimension weights corresponding to the pre-set risk-bearing capacity assessment dimensions, and using the quantified values ​​of each pre-set risk-bearing capacity assessment dimension as feature vectors, cluster centers are iteratively calculated until the spatial offset between two adjacent cluster centers is less than a pre-set offset. If the clustering convergence condition is met, the cluster centers are added to the clustering results. The aforementioned quantitative values ​​for the preset risk tolerance assessment dimensions are the risk tolerance values. They represent the threshold ability of a monitored sub-area to accommodate risk factors without triggering a safety accident. For example, in the open-pit mine shovel and transport operation area: storage tank protection area: facilities are sensitive (hazardous chemicals), there are no operational activities, and there are no historical accidents. The tolerance for risks such as "personnel intrusion and equipment approach" is extremely low. As long as someone enters within 5 meters, it is close to the "overflow" threshold. Excavator turning area: equipment is dense, operations are complex, and there have been multiple historical personnel intrusion accidents. The tolerance for risks such as "personnel approach and equipment vibration exceeding the standard" is low. It is easy to trigger an accident if personnel enter the turning radius. Mine truck passage edge: only vehicles pass through, there are no sensitive facilities, and there are few historical accidents. The tolerance for risks such as "short-term vehicle speed exceeding the standard and slight dust" is moderate. Only when the vehicle speed exceeds 15 km / h and is accompanied by dust obstructing the view will the threshold be exceeded.

[0029] Subsequently, initial partition boundaries are generated based on the clustering results. Each initial partition undergoes topological verification using preset boundary topology rules. These rules include: no two partition polygons can intersect; the union of all partitions must equal the monitored area; adjacent partitions must share boundaries that completely overlap and have opposite directions; and each partition must be a simply connected region. Each partition boundary is abstracted as an edge of a graph, with boundary intersections as nodes, constructing a topological adjacency graph. Each edge is assigned a direction, and its corresponding partition ID and adjacent partition IDs are recorded. Topological rule verification is performed on the topological adjacency graph according to the above rules. If the preset boundary topology rules are not met, the boundaries are corrected to meet the rules. Simultaneously, based on the risk propagation direction obtained from historical security event data analysis, a directed graph is constructed with each partition as a node and the possible risk propagation paths as directed edges. Based on the historical security event data analysis results, each edge is assigned a weight representing the risk propagation probability / speed. Risk propagation direction rules are as follows: Rule 1: Risks prioritize propagation along the process flow / logistics direction; Rule 2: Risks prioritize propagation towards directions with favorable physical characteristics such as low-lying areas, downwind areas, and downstream areas; Rule 3: Risks prioritize propagation towards high-value areas such as densely populated areas and densely populated areas of equipment. Based on the directed graph, the risk propagation paths of each monitored sub-area are identified, and the boundaries are expanded or contracted along the risk propagation direction to form new coverage boundaries, resulting in partition boundaries. This allows for the division of the monitored area into dynamic scene partitions.

[0030] By dividing the scene into dynamic zones as described above, non-overlapping closed-loop zones can be formed in the space of the monitoring area. By correcting the risk propagation direction, the zone boundaries can be made to have a risk prevention and control guidance function. Based on the relationship between the zone boundaries and risk propagation, the risk spread path can be predicted in advance, and early warning can be achieved. The combination of the two forms a dual guarantee, providing a solid foundation for intelligent security in complex scenarios.

[0031] In this embodiment of the application, the above-mentioned generation of dynamic environmental situation maps corresponding to each dynamic scene partition based on multiple dynamic scene partitions and time-series dynamic parameter group sequences specifically includes: Based on the acquisition location coordinates of each dynamic parameter in the time-series dynamic parameter group sequence, spatial matching is performed with the boundary coordinates of each dynamic scene partition to add multi-dimensional dynamic parameters within the same dynamic scene partition to the partition-related parameter set. These multi-dimensional dynamic parameters include at least environmental state parameters, target behavior parameters, and operational parameters. Based on the first preset parameter threshold matrix corresponding to each dynamic scene partition, the partition-related parameter set is compared with the corresponding first preset parameter threshold matrix to determine the risk status level of the corresponding dynamic scene partition based on the comparison result. Based on the risk status level and historical security event data within a preset time period, the environmental status gain coefficient corresponding to the risk status level is fine-tuned. Based on the partition-related parameter set, risk status level, and fine-tuned environmental status gain coefficient, a dynamic environmental status map corresponding to the dynamic scene partition is constructed in chronological order.

[0032] In other words, this application can establish a spatial correlation mapping between dynamic parameters and dynamic scene partitions by collecting location coordinates, and simultaneously store the corresponding partition correlation parameter set. It calls the first preset parameter threshold matrix for each dynamic scene partition and compares it with the real-time parameters in the partition correlation parameter set. This first preset parameter threshold matrix can be set based on the pre-defined risk-bearing capacity of the dynamic scene partition. The matrix contains different threshold ranges for each dimension of parameters, such as safe threshold ranges and dangerous threshold ranges. The parameters in the parameter set are compared with the threshold ranges. When all parameters are within the safe threshold range, the risk situation level is determined to be normal; when 1-2 types of parameters exceed the safe threshold but do not reach the dangerous threshold, it is determined to be alert; when more than or equal to 3 types of parameters exceed the safe threshold or any parameter reaches the dangerous threshold, it is determined to be extreme. The switching timestamp and trigger parameters of the risk situation level are recorded synchronously. Subsequently, for different risk situation levels, a basic environmental situation gain coefficient is pre-set, such as 1.0 for normal, 1.2 for alert, and 1.5 for extreme. Based on historical security event data within a preset time period, the basic environmental situation gain coefficient is fine-tuned. The preset time period can be set by the user according to actual usage, and is not specifically limited here, for example, it can be set to 3 months. For example, if the incidence rate of risk events under extreme historical conditions in a region is ≥15%, the extreme situation gain coefficient is increased by 0.1; if there are no risk event records under normal conditions for more than 6 months, the normal situation gain coefficient is decreased by 0.05. After fine-tuning, the gain coefficient range is limited to 0.95-1.6 to ensure the rationality of the coefficient. Integrating the above-mentioned region-related parameter set, risk situation level, and environmental situation gain coefficient, a four-dimensional dynamic environmental situation map is constructed in chronological order, including a time axis, multi-dimensional dynamic parameters, risk situation level, and environmental situation gain coefficient.

[0033] By constructing a dynamic environmental situation map, the dynamic environmental state can be quantified to facilitate subsequent risk and anomaly assessment.

[0034] S103, based on dynamic environmental situation maps, historical security event databases and preset risk assessment algorithms, determines each first-risk abnormal area.

[0035] In this embodiment of the application, based on various dynamic environmental situation maps, historical security event databases, and preset risk assessment algorithms, each first risk anomaly area is determined, specifically including: Real-time multi-dimensional dynamic parameters and environmental situation gain coefficients are extracted from various dynamic environmental situation maps. Single-risk type weight vectors and second preset parameter threshold matrices for dynamic scene partitions are retrieved from the historical security event database. The second preset parameter threshold matrix is ​​matched to the risk-bearing capacity of the dynamic scene partitions. Based on the multi-dimensional dynamic parameters and the second preset parameter threshold matrix, corresponding parameter deviation quantification values ​​are calculated. Based on the single-risk type weight vectors and parameter deviation quantification values, the basic risk value of the dynamic scene partition is calculated. The risk assessment value of the dynamic scene partition is determined based on the product of the basic risk value and the environmental situation gain coefficient. The risk assessment value of each dynamic scene partition is compared with a preset anomaly judgment threshold to determine each first-risk anomaly area based on the second comparison result.

[0036] In other words, parameters are extracted from the dynamic environmental situation map, and a pre-set single-risk type weight vector for the dynamic scene partition is retrieved from the historical security event database. This single-risk type weight vector is obtained by the user setting the weights for each individual risk that can exist in the dynamic scene partition based on expert experience. For example, for dynamic scene partition A, its single-risk type weight vector may contain elements such as personnel intrusion (0.7), equipment collision (0.1), and severe environment (0.2). Simultaneously, a second pre-set parameter threshold matrix is ​​retrieved. This second pre-set parameter threshold matrix is ​​trained based on nearly 6 months of valid data from the historical security event database and matches the risk-bearing capacity of the dynamic scene partition. It includes partition safety thresholds under different parameter dimensions, such as a personnel safety distance threshold of 2 meters in densely equipped areas and 1.5 meters in densely populated areas. The parameter deviation quantification value is calculated by using the formula: Parameter Deviation Quantification Value = |Real-time Multi-dimensional Dynamic Parameters - Partition Safety Threshold| ÷ Partition Safety Threshold. Then, each parameter deviation quantification value is multiplied by its corresponding single-risk type weight, and the results are summed to obtain the basic risk value: Basic Risk Value = ∑(Single Risk Type Weight × Corresponding Parameter Deviation Quantification Value). The corresponding parameter deviation quantification value refers to the parameter deviation quantification value that matches a single risk type. For example, if the single risk type is personnel intrusion, then the dynamic parameter corresponding to the parameter deviation quantification value should be the personnel intrusion related parameter.

[0037] After obtaining the basic risk value, the product of the basic risk value and the fine-tuned environmental situation gain coefficient is further calculated to obtain the final risk assessment value. The final risk assessment value is compared with a preset anomaly detection threshold used to determine abnormal areas. If the risk assessment value is greater than the preset anomaly detection threshold, the corresponding dynamic scene partition is determined to be the first risk anomaly area. The preset anomaly detection threshold is based on expert experience and is not specifically limited here.

[0038] By utilizing relevant data from the historical security incident database, the risks of dynamic scene partitions are quantified and initially identified, providing a data foundation for subsequent identification of risk-coupled areas.

[0039] S104. Input the dynamic environmental situation map group corresponding to each first risk anomaly area into the pre-trained spatiotemporal attention coupling model to determine the risk coupling area and its corresponding second risk situation curve.

[0040] In this embodiment of the application, the dynamic environmental situation map group corresponding to each first risk anomaly region is input into a pre-trained spatiotemporal attention coupling model to determine the risk coupling region and its corresponding second risk situation curve, specifically including: The dynamic environmental situation map group is feature-encoded and dimension-mapped to construct a spatiotemporal feature matrix. This matrix includes both temporal and spatial feature dimensions. The temporal feature dimension includes at least an environmental state parameter vector, a target behavior parameter vector, a risk situation level encoding value, and an environmental situation gain coefficient. The spatial feature dimension includes at least the regional representation coordinate information of each first-risk anomaly region. The spatiotemporal feature matrix is ​​input into a pre-trained spatiotemporal attention coupling model. Through temporal and spatial attention mechanisms, the spatiotemporal feature matrix is ​​weighted and fused, and density clustering is performed on the fused spatiotemporal features to identify regions that meet preset density connectivity criteria as risk coupling regions. The spatial attention mechanism extracts associated features of multiple risk types within the same spatiotemporal space based on the predicted target interaction range corresponding to the dynamic scene partition. The predicted target interaction range is smaller than the spatial range of the monitored area. The preset density connectivity criteria require at least a preset number of risk features to be density-connected within the same spatiotemporal space. These preset number of risk features originate from the same dynamic scene partition and / or different dynamic scene partitions. Based on the risk assessment value corresponding to the risk coupling area, and according to the preset time step, the corresponding risk gradient value is calculated, and a second risk situation curve is generated based on each risk gradient value with time as the horizontal axis.

[0041] Specifically, the aforementioned spatiotemporal attention coupling model can be a deep learning-based neural network model. Its network architecture includes an input layer, a feature extraction layer, a spatiotemporal attention coupling layer, a feature fusion layer, and an output layer. The input layer maps the raw sensor data to a unified feature space, including timestamp embedding and spatial location encoding, and standardizes the input features to accelerate model convergence and improve training stability. At the same time, it connects temporal features (such as time difference and temporal encoding) with spatial features (such as coordinates and region ID) to form an initial spatiotemporal representation. The feature extraction layer uses a convolutional neural network (CNN) or a graph neural network (GNN) to capture local spatial dependencies, uses one-dimensional convolution or temporal attention to capture the changing trends in the time series, and extracts spatiotemporal features of different granularities through pyramid pooling. The spatiotemporal attention coupling layer is the core component of the model, which includes a temporal attention mechanism and a spatial attention mechanism. The temporal attention mechanism can treat the sequence data as multiple tokens in the time dimension, calculate the correlation between each time point and other time points, and use a multi-head self-attention mechanism to capture the dependencies of different time scales in parallel. The spatial attention mechanism calculates the importance weights between spatial locations within a single preset time step, and generates a spatial attention map through global average pooling and max pooling to highlight key areas in space. Furthermore, the information corresponding to the temporal attention layer and the spatial attention layer is coupled through a spatiotemporal coupling strategy. The coupling method can be, for example, dividing the time series into blocks and performing spatiotemporal joint attention calculation in each block. Other coupling methods can also be used, which are not specifically limited here. The feature fusion layer and output layer fuse the spatiotemporal attention output with the original features through residual concatenation or splicing. Then, using density-based clustering algorithms such as Density-Based Spatial Clustering of Applications with Noise (DBSCAN), regions meeting preset density connectivity criteria are marked as risk-coupled regions. The output layer outputs the risk assessment value and corresponding region boundaries for these risk-coupled regions. These risk-coupled regions can be understood as high-risk areas formed by the superposition of multiple risk types in the same spatiotemporal space, requiring identification and targeted intervention through the coupling model.

[0042] The aforementioned spatiotemporal attention coupling model can be trained using several historical multi-risk overlay scenario training samples. Each training sample set includes at least a dynamic environment situation map group, risk type annotations, and risk coupling relationship labels. The samples are divided into a training set and a validation set in a 7:3 ratio. The training set samples are clustered using the model to be trained, and the clustering results are compared with the risk coupling relationship labels of the samples to adjust the model's temporal attention weights (memory duration 0-3 seconds) and spatial attention weights (prediction range 0-5 meters). Training stops when the risk coupling identification accuracy of the validation set is ≥92%, resulting in a pre-trained spatiotemporal attention coupling model. The time step of the second risk situation curve output by the model can be 100ms, and the risk gradient calculation accuracy is ±0.01. In the preset density connection judgment condition, there must be at least a preset number (e.g., 2) of risk feature density vectors in the same spatiotemporal space. This same spatiotemporal space can be the same dynamic scene partition or a region spliced ​​together from adjacent dynamic scene partitions; no specific limitation is made here.

[0043] When generating the second risk situation curve, the risk gradient value corresponding to each preset time step of 100ms can be extracted. The comprehensive risk gradient value = (current risk assessment value - risk assessment value of the previous time step) / time step. The second risk situation curve is plotted in chronological order with time as the horizontal axis and the risk gradient value as the vertical axis. During the curve plotting process, algorithms such as Gaussian filtering can also be used for curve smoothing and noise removal.

[0044] The above scheme can couple and associate the risks of one or more dynamic scene zones within the monitoring area. Instead of performing security monitoring on a single perception dimension, it couples the various risks within the area to facilitate subsequent anomaly identification, thereby enabling efficient dynamic security monitoring.

[0045] S105, based on the second risk situation curve and preset scenario positioning rules, determines the second risk anomaly area within the monitoring area.

[0046] In this embodiment of the application, based on the second risk situation curve and preset scene positioning rules, the second risk anomaly area within the monitoring area is determined, specifically including: Based on preset anomaly detection criteria, the second risk situation curve is preprocessed to extract corresponding key feature points to construct the first anomaly interval. According to the basic geographic information data of the dynamic scene partition and the deployment location calibration data corresponding to the multi-source dynamic sensing modules, a corresponding positioning fitting equation is constructed. The positioning fitting equation includes the mapping relationship between the time dimension features of the second risk situation curve and the physical spatial coordinates of the monitored area. Based on the preset anomaly detection criteria, a preset sliding window slides along the second risk situation curve within the curve segment corresponding to the first anomaly interval, and the presence of abnormal data within the preset sliding window is determined in real time. If so, the second risk anomaly area is determined based on the second anomaly interval corresponding to the abnormal data and the positioning fitting equation.

[0047] The process involves executing preset scene positioning rules through pre-defined anomaly conditions and positioning fitting equations. The preset anomaly judgment conditions include at least a risk gradient threshold for preliminary judgment of whether data is anomalous. Within a preset sliding window, a strict judgment criterion is applied to determine the presence of anomalous data; this strict criterion requires that the risk gradient value be greater than the gradient threshold for n consecutive time steps within the window. In the preprocessing stage of the second risk situation curve, only the risk gradient threshold from the preset anomaly judgment conditions is used. The process iterates along the second risk situation curve, identifying the anomaly start time point that first satisfies the preset anomaly judgment conditions and the anomaly termination time point that first deviates from the preset anomaly judgment conditions after the anomaly start time point. Between the anomaly start time point and the anomaly termination time point, the risk gradient peak time point corresponding to the curve's peak point is determined. Key feature points are then extracted, and the anomaly start time point and the anomaly termination time point are used as endpoints to construct the first anomaly interval.

[0048] This application also pre-establishes a positioning fitting equation for the time step and physical spatial coordinates of the second risk situation curve based on the basic geographic information data of the dynamic scene partition (such as partition boundary coordinates, center coordinates and partition perimeter) and the deployment location calibration data of the multi-source dynamic perception module, such as: P(x,y)=a×x+b×y+c, where P(x,y) represents the physical spatial coordinate point (x,y) in the second risk situation curve, x represents the time step, y represents the risk gradient value of the corresponding time step, and a, b, and c are the fitting coefficients obtained by training based on the deployment location of the multi-source dynamic perception module and the geographic coordinates of the partition.

[0049] Subsequently, a preset sliding window, with a size of 5 time steps and a movement step of 1 time step, slides along the curve segment of the first abnormal interval in the second risk situation curve. Using the aforementioned strict judgment criteria, it is determined whether the preset sliding window detects abnormal data that meets the corresponding criteria. This further filters out the first abnormal interval containing anomalies, resulting in the second abnormal interval. Next, the physical spatial coordinates corresponding to the first abnormal interval are identified using the aforementioned positioning fitting equation, thus obtaining the second risk anomaly area.

[0050] If no abnormal data is found in any of the first abnormal intervals, then the determination of the second risk abnormal region will not proceed.

[0051] This enables security monitoring that transcends single perception dimensions and single zones, ensuring the identification of the actual geographical location of risk-coupled areas across multiple dynamic scene zones. Furthermore, by extracting key feature points through curve preprocessing, the range of the preset sliding window is narrowed, reducing computational costs and improving the efficiency of anomaly area location.

[0052] More specifically, the determination of the second risk anomaly region based on the second anomaly interval corresponding to the abnormal data and the localization fitting equation includes: Identify key time points and their corresponding risk gradient values ​​within the second anomaly interval. Key time points must include at least the anomaly start time, the risk gradient peak time, and the anomaly termination time. Use the localization fitting equation to determine the key physical spatial coordinates corresponding to each key time point. Using the key physical spatial coordinates corresponding to the risk gradient peak time as the region center, determine the region expansion radius based on the risk gradient peak and the anomaly duration. When there are no occlusion obstacles in the dynamic scene partition, generate the corresponding anomaly region boundary based on the region center and the region expansion radius. When there are occlusion obstacles in the dynamic scene partition, perform boundary fitting based on the key physical spatial coordinates corresponding to the same second anomaly interval to generate the corresponding anomaly closed region boundary. Generate the second risk anomaly region based on the anomaly region boundary and / or the anomaly closed region boundary.

[0053] In other words, key time points and their corresponding risk gradient values ​​are input into the location fitting equation to obtain the key physical spatial coordinates corresponding to each key time point. Then, the peak coordinates corresponding to the peak time point of the risk gradient are used as the center of the region, and the region expansion radius is calculated based on the peak risk gradient and the duration of the anomaly, as shown in the formula below. ,in, For the radius of the region expansion, This is a preset proportionality constant, set based on expert experience. The peak value of the risk gradient. This indicates the duration of the anomaly; 1000 is used to convert milliseconds to seconds. A circular area can be defined with the region center as the midpoint and the region's expansion radius. The boundary of this circular area is the boundary of the anomaly region. This type of anomaly region boundary is suitable when there are no obstructions within the monitored area. If obstructions exist within the monitored area, polygon fitting is required based on the coordinates of each key physical space point, while simultaneously removing the coordinate intervals corresponding to the obstruction areas to construct a closed anomaly region boundary. Further, based on the constructed region boundary, a second risk anomaly region is generated.

[0054] The above technical solution transforms the anomalies in the risk situation curve into precise risk areas in the physical space, enabling accurate security monitoring and improving the efficiency of early warning for risk events.

[0055] S106, based on the second risk anomaly area and the preset multi-dimensional intervention strategy matrix, generates graded early warning information and adaptive intervention instructions, and sends them to the user terminal to conduct continuous dynamic security monitoring and proactive intervention in the monitored area.

[0056] User terminals can be handheld terminals used by security personnel in the monitored area, such as mobile phones, smartwatches, smart bracelets, etc., or backend monitoring computers, on-site alarm edge devices, etc., without specific limitations.

[0057] In this embodiment, based on the second risk anomaly region and a preset multi-dimensional intervention strategy matrix, hierarchical early warning information and adaptive intervention instructions are generated, specifically including: Based on the dynamic environmental situation map and the second risk situation curve corresponding to the second risk anomaly area, the corresponding risk situation level, risk gradient range, and scenario type are determined. Based on the warning level corresponding to the risk gradient range, the corresponding graded warning information is determined. The risk situation level, risk gradient range, and scenario type are encoded and matched with a preset multi-dimensional intervention strategy matrix to determine the corresponding combination of intervention methods based on the matching results. The corresponding intervention parameters are then configured according to the combination of intervention methods to generate adaptive intervention instructions.

[0058] In other words, this application can extract risk status levels from dynamic environmental situation maps and risk gradient intervals from second risk status curves. It can also analyze and identify scenario types from data in the dynamic environmental situation maps, and users can specify scenario types for monitoring areas and / or dynamic scenario partitions based on actual usage scenarios; no specific limitations are made here. Scenario types include, for example, equipment-intensive, personnel-intensive, and sensitive facility-intensive scenarios. This application pre-sets different tiered warning information for different warning levels. For example, a level 1 warning includes risk location, risk type, and basic avoidance suggestions; a level 2 warning supplements the risk gradient change trend and intervention measures; a level 3 warning adds a link to a real-time risk heat map and emergency contact information. The warning information format is adapted to different user terminals (pop-up + voice broadcast for mobile apps, vibration + text prompts for wristbands, and pop-up + sound and light alarms for central control screens), with concise and clear text content (≤100 characters) and voice broadcast duration ≤5 seconds. The warning level is determined by the risk gradient range. For example, the risk gradient range (a, b) is a Level 1 warning, and the risk gradient range [b, c) is a Level 2 warning, etc. The specific level can be set by the user according to the actual usage scenario, and is not specifically limited here. The preset multi-dimensional intervention strategy matrix contains the correspondence between several intervention method combinations and multi-dimensional parameter codes. The multi-dimensional parameter codes are obtained by encoding the risk situation level, risk gradient range, and scenario type, and are used to index the intervention method combinations in the matrix. An example of an intervention method combination is information prompts + low-intensity equipment control. Intervention parameters are configured, such as display duration and volume / brightness for information prompts, and deceleration acceleration and braking response time for equipment control. Then, based on the intervention method combination and intervention parameters, adaptive intervention commands are generated to provide prompts and / or control the equipment.

[0059] Through the above technical solution, this application utilizes a multi-source dynamic perception module at the monitoring site to collect multi-dimensional dynamic parameters and generate a dedicated situation map according to dynamic scene partitions. This allows for regional and differentiated situational quantitative assessments tailored to dynamic environments. Simultaneously, by leveraging the map and historical data, risks in individual partitions are identified, and a spatiotemporal attention coupling model is used to identify risk coupling areas that can cross partitions, enabling the prediction of implicit coupling risks. Furthermore, by combining preset scene positioning rules, the risk locations are accurately pinpointed, and tiered early warning and adaptive intervention are implemented, improving the level of intelligence in security.

[0060] Figure 2 A schematic diagram of the structure of an intelligent security device based on dynamic environment perception is provided for an embodiment of this application, as shown below. Figure 2 As shown, the device includes: At least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Based on multi-source dynamic perception modules pre-deployed in key areas of the scene, multi-dimensional dynamic parameters within the monitoring area are determined to generate a time-series dynamic parameter set sequence. According to multiple dynamic scene partitions and the time-series dynamic parameter set sequence, dynamic environmental situation maps corresponding to each dynamic scene partition are generated. Based on each dynamic environmental situation map, a historical security event database, and a preset risk assessment algorithm, each first-risk anomaly area is identified. The dynamic environmental situation map set corresponding to each first-risk anomaly area is input into a pre-trained spatiotemporal attention coupling model to determine the risk coupling area and its corresponding second-risk situation curve. Based on the second-risk situation curve and preset scene positioning rules, the second-risk anomaly area within the monitoring area is determined. Based on the second-risk anomaly area and a preset multi-dimensional intervention strategy matrix, tiered early warning information and adaptive intervention instructions are generated and sent to the user terminal for continuous dynamic security monitoring and proactive intervention in the monitoring area.

[0061] This application embodiment also provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows: Based on multi-source dynamic perception modules pre-deployed in key areas of the scene, multi-dimensional dynamic parameters within the monitoring area are determined to generate a time-series dynamic parameter set sequence. According to multiple dynamic scene partitions and the time-series dynamic parameter set sequence, dynamic environmental situation maps corresponding to each dynamic scene partition are generated. Based on each dynamic environmental situation map, a historical security event database, and a preset risk assessment algorithm, each first-risk anomaly area is identified. The dynamic environmental situation map set corresponding to each first-risk anomaly area is input into a pre-trained spatiotemporal attention coupling model to determine the risk coupling area and its corresponding second-risk situation curve. Based on the second-risk situation curve and preset scene positioning rules, the second-risk anomaly area within the monitoring area is determined. Based on the second-risk anomaly area and a preset multi-dimensional intervention strategy matrix, tiered early warning information and adaptive intervention instructions are generated and sent to the user terminal for continuous dynamic security monitoring and proactive intervention in the monitoring area.

[0062] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0063] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0064] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0065] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An intelligent security method based on dynamic environmental perception, characterized in that, The method is applied to an intelligent security system that includes a pre-installed sensor network; the method includes: Based on multi-source dynamic perception modules pre-deployed in key areas of the scene, multi-dimensional dynamic parameters within the monitoring area are determined to generate a time-series dynamic parameter set sequence. Based on multiple dynamic scene partitions and the time-series dynamic parameter group sequence, a dynamic environmental situation map corresponding to each dynamic scene partition is generated. Based on the dynamic environmental situation map, historical security event database and preset risk assessment algorithm, each first risk anomaly area is determined; The dynamic environmental situation map group corresponding to each of the first risk anomaly regions is input into the pre-trained spatiotemporal attention coupling model to determine the risk coupling region and its corresponding second risk situation curve. Based on the second risk situation curve and the preset scenario positioning rules, the second risk anomaly area within the monitoring area is determined; Based on the second risk anomaly area and the preset multi-dimensional intervention strategy matrix, hierarchical early warning information and adaptive intervention instructions are generated and sent to the user terminal to conduct continuous dynamic security monitoring and proactive intervention in the monitored area.

2. The intelligent security method based on dynamic environment perception according to claim 1, characterized in that, Before generating dynamic environmental situation maps corresponding to each of the multiple dynamic scene partitions and the time-series dynamic parameter group sequences, the method further includes: Acquire basic geographic information data, operational data, and historical safety event data within a preset time period corresponding to the monitored area to construct a partitioned basic dataset; Based on the partitioned basic dataset, the preset risk-bearing capacity assessment dimensions, and the preset clustering algorithm, the monitoring area is partitioned. Based on the clustering results, the initial partition boundaries within the monitoring area are determined, and each initial partition boundary is corrected by using preset boundary topology rules and risk propagation direction rules to obtain each dynamic scene partition. The corrected dynamic scene partitions use the coverage area of ​​the corresponding monitoring sub-region along the risk propagation direction as the partition boundary, forming a non-overlapping closed-loop partition within the monitoring area.

3. The intelligent security method based on dynamic environment perception according to claim 1, characterized in that, Based on multiple dynamic scene partitions and the time-series dynamic parameter group sequence, a dynamic environment situation map corresponding to each of the dynamic scene partitions is generated, specifically including: Based on the acquisition location coordinates of each dynamic parameter in the time-series dynamic parameter group sequence, spatial matching is performed with the boundary coordinates of each dynamic scene partition to add multi-dimensional dynamic parameters within the same dynamic scene partition to the partition-related parameter set; wherein, the multi-dimensional dynamic parameters include at least environmental state parameters, target behavior parameters, and operation parameters; Based on the first preset parameter threshold matrix corresponding to each of the dynamic scene partitions, the partition-related parameter set is compared with the corresponding first preset parameter threshold matrix to determine the risk status level of the corresponding dynamic scene partition according to the first comparison result; Based on the risk situation level and historical security event data within a preset time period, the environmental situation gain coefficient corresponding to the risk situation level is fine-tuned. Based on the partition association parameter set, the risk situation level, and the fine-tuned environmental situation gain coefficient, the dynamic environmental situation map corresponding to the dynamic scene partition is constructed in chronological order.

4. The intelligent security method based on dynamic environment perception according to claim 3, characterized in that, Based on the dynamic environmental situation maps, historical security event databases, and preset risk assessment algorithms, each first-risk anomaly area is determined, specifically including: Real-time multi-dimensional dynamic parameters and environmental situation gain coefficients are extracted from each of the dynamic environmental situation maps, and the single risk type weight vector and the second preset parameter threshold matrix of the dynamic scene partition are called from the historical security event database; wherein, the second preset parameter threshold matrix matches the risk carrying capacity of the dynamic scene partition; Calculate the corresponding parameter deviation quantization value based on each of the multi-dimensional dynamic parameters and the second preset parameter threshold matrix; The basic risk value of the dynamic scenario partition is calculated based on the single risk type weight vector and the parameter deviation quantization value. The risk assessment value of the dynamic scene partition is determined based on the product of the basic risk value and the environmental situation gain coefficient. The risk assessment value of each dynamic scene partition is compared with a preset anomaly judgment threshold to determine each of the first risk anomaly areas based on the second comparison result.

5. The intelligent security method based on dynamic environment perception according to claim 1, characterized in that, The dynamic environmental situation map group corresponding to each of the first risk anomaly regions is input into a pre-trained spatiotemporal attention coupling model to determine the risk coupling region and its corresponding second risk situation curve, specifically including: The dynamic environmental situation map group is feature-encoded and dimension-mapped to construct a spatiotemporal feature matrix; wherein, the spatiotemporal feature matrix includes a time feature dimension and a spatial feature dimension; the time feature dimension includes at least an environmental state parameter vector, a target behavior parameter vector, a risk situation level encoding value, and an environmental situation gain coefficient; the spatial feature dimension includes at least the regional representation coordinate information of each of the first risk anomaly areas. The spatiotemporal feature matrix is ​​input into the pre-trained spatiotemporal attention coupling model to perform weighted fusion of the spatiotemporal feature matrix through temporal attention and spatial attention mechanisms. Density clustering is then performed on the fused spatiotemporal features to identify regions that meet a preset density connectivity criterion as risk coupling regions. The spatial attention mechanism extracts associated features of multiple risk types within the same spatiotemporal space based on the predicted target interaction range corresponding to the dynamic scene partition. The predicted target interaction range is smaller than the spatial range of the monitoring area. The preset density connectivity criterion requires at least a preset number of risk features to be density-connected within the same spatiotemporal space. The preset number of risk features originate from the same dynamic scene partition and / or different dynamic scene partitions. Based on the risk assessment value corresponding to the risk coupling region, and according to a preset time step, the corresponding risk gradient value is calculated, and the second risk situation curve is generated with time as the horizontal axis based on each of the risk gradient values.

6. The intelligent security method based on dynamic environment perception according to claim 1, characterized in that, Based on the second risk situation curve and the preset scenario positioning rules, a second risk anomaly area is determined within the monitoring area, specifically including: Based on preset anomaly judgment conditions, the second risk situation curve is preprocessed to extract corresponding key feature points in order to construct the first anomaly interval. Based on the basic geographic information data of the dynamic scene partition and the deployment location calibration data corresponding to the multi-source dynamic perception module, a corresponding positioning fitting equation is constructed; the positioning fitting equation includes the mapping relationship between the time dimension features of the second risk situation curve and the physical spatial coordinates of the monitoring area; According to the preset anomaly determination conditions, a preset sliding window slides along the second risk situation curve in the corresponding curve segment of the first anomaly interval, and determines in real time whether there is abnormal data in the preset sliding window; If so, the second risk anomaly region is determined based on the second anomaly interval corresponding to the abnormal data and the location fitting equation.

7. The intelligent security method based on dynamic environment perception according to claim 6, characterized in that, Based on the second abnormal interval corresponding to the abnormal data and the localization fitting equation, the second risk abnormal region is determined, specifically including: Determine the key time points and their corresponding risk gradient values ​​in the second abnormal interval; the key time points include at least the anomaly start time point, the risk gradient peak time point, and the anomaly termination time point. The positioning fitting equation is invoked to determine the key physical space coordinates corresponding to each of the key time points; Using the key physical spatial coordinate point corresponding to the peak time point of the risk gradient as the region center, the region expansion radius corresponding to the region center is determined based on the peak risk gradient and the duration of the anomaly. When there are no occlusion obstacles in the dynamic scene partition, a corresponding abnormal region boundary is generated based on the region center and the region expansion radius; When there are occlusions in the dynamic scene partition, the boundary is fitted according to the key physical space coordinates of the same second abnormal interval to generate the corresponding abnormal closed region boundary. The second risk anomaly region is generated based on the boundary of the anomaly region and / or the boundary of the anomaly closed region.

8. The intelligent security method based on dynamic environment perception according to claim 1, characterized in that, Based on the second risk anomaly area and the preset multidimensional intervention strategy matrix, hierarchical early warning information and adaptive intervention instructions are generated, specifically including: Based on the dynamic environmental situation map and the second risk situation curve corresponding to the second risk anomaly area, the corresponding risk situation level, risk gradient range and scenario type are determined; Based on the warning level corresponding to the risk gradient interval, the corresponding graded warning information is determined; The risk situation level, the risk gradient range, and the scenario type are encoded and matched with the preset multidimensional intervention strategy matrix to determine the corresponding combination of intervention methods based on the matching results. The corresponding intervention parameters are then configured according to the combination of intervention methods to generate the adaptive intervention instruction.

9. An intelligent security device based on dynamic environmental perception, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the intelligent security method based on dynamic environment perception as described in any one of claims 1-8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of executing the intelligent security method based on dynamic environment perception as described in any one of claims 1-8.

Images