Intelligent safety protection management method and system

Abstract

The invention relates to an intelligent safety protection management method and system. The method comprises the following steps: acquiring behavior data and position data of personnel in an industrial site; analyzing the behavior data and the position data by using a pre-trained personnel behavior model to obtain a behavior recognition result; according to the behavior recognition result and the current operation state of the target equipment, judging a risk level corresponding to the personnel behavior; based on the risk level, a corresponding safety response strategy is matched in the edge computing node, a control instruction is generated according to the safety response strategy, and the control instruction is used for driving the target device to execute a corresponding response action; and the voice interaction module is used for acquiring voice input of an on-site operator, performing semantic recognition on the voice input to obtain a voice recognition result, and executing corresponding emergency control operation to cover a current control instruction when the voice recognition result meets a preset emergency instruction condition. The method has the effect of improving the accuracy of intelligent safety management in the industrial environment.

Description

Technical Field

[0001] The present application relates to the technical field of industrial intelligent control and safety protection, and in particular to an intelligent safety protection management method and system. Background Art

[0002] As industrial automation and intelligence continue to advance, the complexity of collaborative operations between humans and high-risk equipment is increasing. Traditional security protection methods based on physical isolation and single sensor triggering are no longer able to meet the needs of dynamic risk control. Especially in application scenarios such as large factories, automated production lines, warehousing systems, and rail operations, achieving real-time perception and prediction of operator behavior, and triggering timely response mechanisms after identifying potential risks, have become key challenges in improving industrial safety.

[0003] Existing security management methods mainly rely on the following methods: first, using fixed-area facilities such as gratings, infrared, and electronic fences to determine whether people have entered restricted areas; second, setting static security boundaries through positioning technologies such as UWB; third, using simple rules to judge the status and trigger a single control response.

[0004] The above-mentioned existing technical solutions have the following defects: the behavior recognition dimension of the existing industrial safety protection methods is limited, and it is impossible to accurately judge the intention of the person's action and the degree of risk, which is prone to false alarms and missed alarms. Therefore, there is room for improvement. Summary of the Invention

[0005] In order to improve the accuracy of intelligent safety management in industrial environments, the present application provides an intelligent safety protection management method and system.

[0006] The above-mentioned invention objective of this application is achieved through the following technical solutions: An intelligent security protection management method, the intelligent security protection management method comprising: Obtain behavioral and location data of personnel at industrial sites; Analyze the behavior data and location data using a pre-trained personnel behavior model to obtain a behavior recognition result; Determine the risk level corresponding to the human behavior based on the behavior recognition result and the current operating status of the target device; Based on the risk level, a corresponding security response policy is matched in the edge computing node, and a control instruction is generated according to the security response policy, wherein the control instruction is used to drive the target device to perform a corresponding response action; The voice interaction module obtains the voice input of the on-site operator and performs semantic recognition on the voice input to obtain a voice recognition result. When the voice recognition result meets the preset emergency command condition, the corresponding emergency control operation is executed to overwrite the current control command; The behavior recognition results, risk levels, control instructions and voice interaction records are uploaded to the cloud platform, the security response strategy is optimized and updated based on the reinforcement learning model of the cloud platform, and the optimized strategy is synchronized to the edge computing node.

[0007] By adopting the above technical solution, by obtaining the behavioral data and location data of personnel in the industrial site, the spatial status and dynamic trajectory of the operating personnel can be fully perceived, thereby providing multi-dimensional data support for subsequent intelligent identification and improving the accuracy of personnel status identification; by using a pre-trained personnel behavior model to analyze the behavioral data and location data to obtain a behavior recognition result, the specific behavioral status of the personnel can be efficiently identified, and potential dangerous behaviors can be quickly classified, thereby intervening in risk prevention and control in advance; by judging the risk level corresponding to the personnel behavior based on the behavior recognition result and the current operating status of the target equipment, dynamic risk assessment can be performed based on the operating environment and equipment status, thereby avoiding fixed threshold misjudgment and improving the flexibility and accuracy of risk perception; by matching the corresponding security response strategy in the edge computing node and generating control instructions, local rapid response can be achieved, and the efficiency of issuing control instructions can be improved, thereby ensuring the real-time nature of on-site safety response.

[0008] In one example, the present application may be further configured as follows: the intelligent security protection management method is characterized in that the intelligent security protection management method includes: Obtain behavioral and location data of personnel at industrial sites; Analyze the behavior data and location data using a pre-trained personnel behavior model to obtain a behavior recognition result; Determine the risk level corresponding to the human behavior based on the behavior recognition result and the current operating status of the target device; Based on the risk level, a corresponding security response policy is matched in the edge computing node, and a control instruction is generated according to the security response policy, wherein the control instruction is used to drive the target device to perform a corresponding response action; The voice interaction module obtains the voice input of the on-site operator and performs semantic recognition on the voice input to obtain a voice recognition result. When the voice recognition result meets the preset emergency command condition, the corresponding emergency control operation is executed to overwrite the current control command; The behavior recognition results, risk levels, control instructions and voice interaction records are uploaded to the cloud platform, the security response strategy is optimized and updated based on the reinforcement learning model of the cloud platform, and the optimized strategy is synchronized to the edge computing node.

[0009] By adopting the above technical solution, by collecting image sequences through the camera and extracting posture key points and motion features, the spatial characteristics of people's postures and behavioral movements can be identified, thereby improving the system's ability to recognize complex behaviors; by collecting spatial coordinates and trajectory information through the UWB positioning module, the displacement trend of people can be grasped in real time, enhancing the system's understanding of behavioral paths; by obtaining the relative distance and approach speed information between people and equipment through millimeter wave radar, it can fill the camera blind spots and identify high-risk behaviors such as rapid approach, thereby improving the coverage and robustness of overall behavior perception.

[0010] In one example, the present application may be further configured as follows: before analyzing the behavior data and location data using the pre-trained personnel behavior model, the intelligent security protection management method further includes: Collecting a historical sample data set, wherein the historical sample data set includes image data, spatial coordinate data, and relative distance data; Manually labeling the historical sample data set, where the labeling content includes the behavior category of the person and the corresponding risk type; Based on the labeled sample data, a multimodal feature fusion model is constructed using a convolutional neural network and a time series modeling module, and the multimodal feature fusion model is trained through supervised learning to obtain the pre-trained personnel behavior model.

[0011] By adopting the above technical solution, by collecting historical sample data sets and manually annotating them, we can build a behavioral sample system that is actually representative, providing an accurate supervision basis for model training, thereby improving the effectiveness of model recognition; by building a multimodal feature fusion model and training it based on supervised learning, the model can have the ability to fuse and reason about multi-source data such as images, locations, and distances, thereby improving the model's behavior recognition accuracy and adaptability in complex environments.

[0012] In one example, the present application may be further configured as follows: the behavior data and location data are analyzed using a pre-trained personnel behavior model to obtain a behavior recognition result including: Performing multimodal feature fusion on the behavior data and the location data to construct a fused feature vector; The fused feature vector is input into the pre-trained personnel behavior model for inference and recognition, and the corresponding behavior recognition result is output. The behavior recognition result includes the personnel behavior category, behavior confidence, behavior occurrence area label and behavior duration.

[0013] By adopting the above technical solution, by fusing the multimodal features of behavioral data and location data and constructing a fused feature vector, it is possible to effectively integrate multi-source perception information, thereby improving the expressiveness and robustness of the model input; by inputting the fused feature vector into the personnel behavior model for inference and recognition, and outputting recognition results including behavior category, confidence, area label and duration, it is possible to finely characterize the personnel behavior status, provide a comprehensive and reliable input basis for subsequent risk assessment, and thus enhance the accuracy and granularity of the system's overall judgment.

[0014] In one example, the present application may be further configured as follows: determining the risk level corresponding to the human behavior based on the behavior recognition result and the current operating state of the target device includes: Extracting behavior category, behavior confidence, behavior occurrence area label, and behavior duration information from the behavior recognition results as a basis for determining human behavior; Perform a comprehensive analysis based on the current operating status of the target device, where the operating status includes running, paused, or standby; According to the preset risk mapping rules, the situations corresponding to the combination of the personnel behavior and the equipment status are divided into low risk, medium risk or high risk levels.

[0015] By adopting the above technical solution, by extracting behavior category, behavior confidence, behavior occurrence area label and duration information as the basis for judging human behavior, it is possible to characterize behavioral risk factors in multiple dimensions, thereby improving the flexibility and accuracy of risk judgment; by conducting comprehensive analysis in combination with the equipment operation status, it is possible to avoid making risk judgments based solely on human actions and improve situational awareness capabilities; by mapping the combination of behavior and equipment status into different risk levels according to risk mapping rules, it is possible to achieve intelligent quantification of the behavior-equipment collaborative relationship, thereby improving the rationality of the system's active response.

[0016] In one example, the present application may be further configured as follows: before classifying the situations corresponding to the combination of the human behavior and the device status into low risk, medium risk, or high risk levels according to the preset risk mapping rules, the intelligent security protection management method further includes: Based on the behavior recognition results collected in the previous operation cycle, the execution effect of the corresponding control instructions, the target device status log and the response accuracy evaluation results, the preset risk mapping rules are dynamically adjusted in real time; The dynamic adjustment includes: calculating the response effectiveness, false alarm rate and missed alarm rate indicators under different combinations of behaviors and device states, adjusting the corresponding relationship parameters between the behavior category, behavior confidence, behavior duration and device operating status, and the updated risk mapping rules are cached to the edge computing node for subsequent risk level judgment.

[0017] By adopting the above technical solution, by collecting behavior recognition results, control instruction execution effects, equipment status logs and response accuracy evaluation results in the previous operation cycle, feedback samples in real operation can be formed, thereby providing data support for rule optimization; by calculating response effectiveness, false alarm rate and missed alarm rate indicators and adjusting the mapping relationship parameters between behavior and equipment status, the original static rules can be refined and dynamically corrected, so that the risk level classification rules can adapt to different working conditions and continuously improve the judgment accuracy.

[0018] In one example, the present application may be further configured as follows: based on the risk level, matching a corresponding security response policy in the edge computing node, and generating a control instruction according to the security response policy includes: A security response policy library containing mapping relationships between multiple response levels, target device types, and control logic is preset in the edge computing node, with the risk level as the index dimension; When the risk level is determined to be high risk, an emergency response strategy is called from the strategy library to generate a power-off shutdown control instruction; When it is determined to be a medium risk or low risk level, pause action and early warning prompt control instructions are generated respectively. The control instructions are in a standardized format and are encapsulated as Modbus or CAN instruction sets according to the interface protocol of the target device to adapt to different types of devices to perform corresponding actions.

[0019] By adopting the above technical solution, by presetting a security response policy library that maps response levels, device types, and control logic in edge computing nodes and matching them with risk levels as indexes, rapid decision-making and adaptive responses can be achieved locally, thereby greatly improving the real-time and flexibility of the system; by generating different types of control instructions for different risk levels and encapsulating them into standard protocols compatible with multiple device interfaces, it can adapt to the control needs of complex equipment systems, thereby improving the versatility of system deployment and the reliability of control execution.

[0020] The second object of the present invention is achieved through the following technical solutions: An intelligent safety protection management system, comprising: Data collection module, used to obtain behavioral data and location data of personnel in industrial sites; An analysis module is used to analyze the behavior data and location data using a pre-trained personnel behavior model to obtain a behavior recognition result; A risk judgment module is used to judge the risk level corresponding to the human behavior based on the behavior recognition result and the current operating status of the target device; An instruction generation module is used to match a corresponding security response policy in the edge computing node based on the risk level, and generate a control instruction according to the security response policy, wherein the control instruction is used to drive the target device to perform a corresponding response action; A voice interaction module is used to obtain the voice input of the on-site operator through the voice interaction module, perform semantic recognition on the voice input, and obtain a voice recognition result. When the voice recognition result meets the preset emergency instruction condition, the corresponding emergency control operation is executed to overwrite the current control instruction; An update module is used to upload the behavior recognition results, risk levels, control instructions and voice interaction records to the cloud platform, optimize and update the security response strategy based on the cloud platform's reinforcement learning model, and synchronize the optimized strategy to the edge computing node.

[0021] By adopting the above technical solution, by obtaining the behavioral data and location data of personnel in the industrial site, the spatial status and dynamic trajectory of the operating personnel can be fully perceived, thereby providing multi-dimensional data support for subsequent intelligent identification and improving the accuracy of personnel status identification; by using a pre-trained personnel behavior model to analyze the behavioral data and location data to obtain a behavior recognition result, the specific behavioral status of the personnel can be efficiently identified, and potential dangerous behaviors can be quickly classified, thereby intervening in risk prevention and control in advance; by judging the risk level corresponding to the personnel behavior based on the behavior recognition result and the current operating status of the target equipment, dynamic risk assessment can be performed based on the operating environment and equipment status, thereby avoiding fixed threshold misjudgment and improving the flexibility and accuracy of risk perception; by matching the corresponding security response strategy in the edge computing node and generating control instructions, local rapid response can be achieved, and the efficiency of issuing control instructions can be improved, thereby ensuring the real-time nature of on-site safety response.

[0022] In summary, this application has the following beneficial technical effects: 1. By acquiring the behavioral and location data of personnel at industrial sites, the spatial status and dynamic trajectory of the operators can be fully perceived, thereby providing multi-dimensional data support for subsequent intelligent identification and improving the accuracy of personnel status identification. By analyzing the behavioral and location data using pre-trained personnel behavior models, the behavior recognition results can be obtained, which can efficiently identify the specific behavioral status of personnel and quickly classify potentially dangerous behaviors, thereby intervening in risk prevention and control in advance. 2. By judging the risk level corresponding to the personnel behavior based on the behavior recognition results and the current operating status of the target equipment, dynamic risk assessment can be performed based on the operating environment and equipment status, thereby avoiding misjudgment of fixed thresholds and improving the flexibility and accuracy of risk perception; by matching the corresponding security response strategy in the edge computing node and generating control instructions, local rapid response can be achieved, improving the efficiency of issuing control instructions, thereby ensuring the real-time nature of on-site safety response. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of an intelligent security protection management method in one embodiment of the present application; Figure 2 This is a flowchart for implementing step S10 in an intelligent security protection management method in one embodiment of the present application; Figure 3 This is a flowchart for implementing step S20 in an intelligent security protection management method in one embodiment of the present application; Figure 4 This is another implementation flowchart of step S20 in an intelligent security protection management method in one embodiment of the present application; Figure 5 This is a flowchart for implementing step S30 in an intelligent security protection management method in one embodiment of the present application; Figure 6 This is a flowchart for implementing step S33 in an intelligent security protection management method in one embodiment of the present application; Figure 7 This is a flowchart for implementing step S40 in an intelligent security protection management method in one embodiment of the present application; Figure 8 This is a principle block diagram of an intelligent safety protection management system in one embodiment of the present application. DETAILED DESCRIPTION

[0024] The present application is further described in detail below with reference to the accompanying drawings.

[0025] In one embodiment, if Figure 1 As shown, this application discloses an intelligent security protection management method, which specifically includes the following steps: S10: Obtain behavioral data and location data of personnel in the industrial site.

[0026] Specifically, the behavioral data and location data of personnel in the industrial site are obtained, including reading real-time image frame sequences from the image acquisition equipment deployed on site and caching them in the local data buffer, and receiving synchronized spatial location information from the positioning module. The image frames may include RGB images or infrared images for subsequent image feature extraction. The spatial location information may include two-dimensional or three-dimensional coordinate data, reflecting the real-time coordinate change trajectory of personnel in a specific area, providing support for the subsequent construction of the dynamic movement trend of personnel by the system. In addition, in the initial stage, the basic data of the electronic fence used to set the virtual safety boundary is also loaded, and the spatial relationship between personnel and the boundary is continuously tracked.

[0027] S20: Analyze the behavior data and location data using the pre-trained personnel behavior model to obtain behavior recognition results.

[0028] Specifically, the behavior data and location data are analyzed using a pre-trained personnel behavior model to obtain behavior recognition results, including inputting the cached image frames into the image preprocessing module for feature enhancement and noise reduction, and constructing it together with the location information into a unified multimodal input vector. The input vector is sent to the personnel behavior recognition model for forward reasoning. The recognition model makes behavior judgments based on the fused convolutional visual features and sequence trajectory features, and outputs structured recognition results, which include not only the behavior type label and its confidence value, but also whether the inferred current behavior is abnormal and whether there is a high-risk intention tendency. At the same time, it dynamically analyzes the spatial behavior patterns of personnel at multiple time points and judges the stability of the behavior characteristics to support the dynamic boundary contraction mechanism. S30: Determine the risk level corresponding to the human behavior based on the behavior recognition result and the current operating status of the target device.

[0029] Specifically, based on the behavior recognition results and the current operating status of the target device, the risk level corresponding to the human behavior is judged, including obtaining real-time operating data from the target device operating status monitoring interface. The data includes parameters such as motor status, load changes, and working condition switching stages. At the same time, the device status is jointly matched with the identified human behavior, and the danger correlation between the two is evaluated through a weight model or a judgment matrix, and a risk level label is output. On this basis, if it is detected that the abnormal trend of human behavior continues and is close to the established boundary area, the dynamic safety boundary adjustment mechanism is further triggered, the expansion range of the danger zone of the electronic fence is fine-tuned, and the expansion information is fed back to the local safety response strategy module to enhance the system's perception sensitivity to risk spread.

[0030] S40: Based on the risk level, the corresponding security response strategy is matched in the edge computing node, and a control instruction is generated according to the security response strategy. The control instruction is used to drive the target device to perform the corresponding response action.

[0031] Specifically, based on the risk level, the corresponding security response strategy is matched in the edge computing node, and control instructions are generated according to the security response strategy. The control instructions are used to drive the target device to perform corresponding response actions, including loading the policy rules bound to the corresponding risk level in the policy matching engine, and extracting the response items that match the current working conditions from the local policy library. The response policy items include control action codes, execution logic structures, associated device lists, expected execution effects and adjustable delay parameters. The strategy also adjusts the response area according to the current dynamic security boundary position, such as adding a monitoring warning point after the boundary is expanded or triggering an early warning action when approaching the new boundary. Finally, the matching results are encapsulated into a standardized control instruction structure, and the communication interface module of the edge node is called to transcode it into the protocol format supported by the target device (such as Modbus, CANopen, etc.) and send it for execution.

[0032] S50: The voice input of the on-site operator is obtained through the voice interaction module, and semantic recognition is performed on the voice input to obtain a voice recognition result. When the voice recognition result meets the preset emergency instruction condition, the corresponding emergency control operation is executed to cover the current control instruction.

[0033] Specifically, the voice input of the on-site operator is obtained through the voice interaction module, and the voice input is semantically recognized to obtain a voice recognition result. When the voice recognition result meets the preset emergency instruction conditions, the corresponding emergency control operation is executed to cover the current control instruction, including starting the voice collection channel to monitor personnel instructions in real time, and parsing the instruction content through the lightweight semantic understanding module built into the local edge node. The parsing process combines keyword matching, sentence structure recognition and context association analysis to extract execution intentions. When keywords such as "emergency stop", "evacuation" and "release confirmation" are identified, the instruction coverage processor is called to suspend the current safety response process, triggering an emergency policy item with a higher priority and immediately sending it to the device. At the same time, the voice trigger and its corresponding boundary status and personnel behavior are recorded as corpus samples for subsequent strategy optimization.

[0034] S60: Upload behavior recognition results, risk levels, control instructions, and voice interaction records to the cloud platform, optimize and update the security response strategy based on the cloud platform's reinforcement learning model, and synchronize the optimized strategy to the edge computing node.

[0035] Specifically, the behavior recognition results, risk levels, control instructions and voice interaction records are uploaded to the cloud platform, and the security response strategy is optimized and updated based on the reinforcement learning model of the cloud platform, and the optimized strategy is synchronized to the edge computing node, including the timed or triggered collection of all event records, execution responses and final feedback information in the current cycle, which are packaged into a structured log data stream and sent to the cloud. After receiving the data, the cloud platform calls the policy optimization model in training and uses the behavior-response-result samples as input for multiple rounds of policy value updates. The reinforcement learning model adjusts the policy score or penalty value according to the actual feedback signal, and dynamically adjusts the policy priority and execution conditions in the policy library. At the same time, based on the continuous approach to the boundary behavior, the virtual boundary scaling coefficient is automatically optimized and a boundary fine-tuning strategy is generated. Finally, the policy set and boundary configuration file are compressed and transmitted back to the edge node and hot-loaded into the policy matching module in real time for response decision and control instruction generation in the next cycle.

[0036] By adopting the above technical solution, by obtaining the behavioral data and location data of personnel in the industrial site, the spatial status and dynamic trajectory of the operating personnel can be fully perceived, thereby providing multi-dimensional data support for subsequent intelligent identification and improving the accuracy of personnel status identification; by using a pre-trained personnel behavior model to analyze the behavioral data and location data to obtain a behavior recognition result, the specific behavioral status of the personnel can be efficiently identified, and potential dangerous behaviors can be quickly classified, thereby intervening in risk prevention and control in advance; by judging the risk level corresponding to the personnel behavior based on the behavior recognition result and the current operating status of the target equipment, dynamic risk assessment can be performed based on the operating environment and equipment status, thereby avoiding fixed threshold misjudgment and improving the flexibility and accuracy of risk perception; by matching the corresponding security response strategy in the edge computing node and generating control instructions, local rapid response can be achieved, and the efficiency of issuing control instructions can be improved, thereby ensuring the real-time nature of on-site safety response.

[0037] In one embodiment, if Figure 2 As shown, in step S10, the behavior data and location data of personnel in the industrial site are obtained, which specifically includes: S11: Collect an image sequence of the person through a camera, and extract posture key points and motion features from the image sequence.

[0038] Specifically, a camera is used to collect image sequences of people, and posture key points and motion features are extracted from the image sequences, including deploying fixed wide-angle cameras or hemispherical smart cameras on site to collect video frames. The collected video sequences are cached at a set frame rate and input into the image processing module. The image processing module uses a human key point extraction algorithm to identify the skeleton nodes of the people appearing in the image frames, and outputs a two-dimensional or three-dimensional coordinate set containing multiple joint points such as shoulders, elbows, and knees. At the same time, the inter-frame skeleton displacement is combined to extract the motion speed and rhythm information to assist in judging whether the person is in a standing, running, bending or leaning over state, and such motion features are used as behavioral clues on the image side to participate in subsequent multimodal fusion analysis.

[0039] S12: The spatial coordinates and movement trajectory information of the person in the target area are collected through the UWB positioning module.

[0040] Specifically, the UWB positioning module is used to collect the spatial coordinates and movement trajectory information of personnel in the target area, including deploying multiple UWB base stations in the target work area and requiring on-site workers to wear UWB tags or embedded work cards. The positioning module reads the two-way time difference of arrival (TDoA) data between the tag and the base station at a fixed frequency, and uses the positioning algorithm to solve the spatial coordinate information of the personnel in real time and convert it into a positioning reference value relative to the virtual safety boundary. At the same time, the spatial movement path of the personnel in continuous time slices is cached and its trajectory change curve is constructed. The trajectory information can be used to analyze the behavioral tendency characteristics such as the way, speed, and degree of path curvature of the personnel entering the area, thereby providing input basis for the system to determine whether there are potential abnormal behaviors such as wandering, straying or rapid intrusion.

[0041] S13: Obtaining relative distance information and approach speed information between the person and the target device through millimeter wave radar.

[0042] Specifically, the relative distance information and approach speed information between the person and the target device are obtained through the millimeter-wave radar, including using the millimeter-wave radar array arranged around the device to periodically scan the target area and send back the reflected signal data. The radar processing module solves the Doppler frequency shift and delay parameters to obtain the distance and speed vector information of the reflected target, and identifies the person's vital sign echo based on the target contour feature matching. The dynamic approach vector is constructed in combination with the relative orientation between the person and the target device. When the approach speed exceeds the preset threshold or the approach distance is less than the set boundary, it is marked as a high-concern behavior point, and this information is used as an enhanced channel in conjunction with image recognition and positioning information to determine the current risk level or boundary intrusion behavior trend.

[0043] In one embodiment, if Figure 3As shown, before step S20, that is, before analyzing the behavior data and location data using the pre-trained personnel behavior model, the intelligent security protection management method further includes: S201: Collect a historical sample data set, where the historical sample data set includes image data, spatial coordinate data, and relative distance data.

[0044] Specifically, historical sample data sets are collected, which include image data, spatial coordinate data and relative distance data. Specifically, the collection process includes extracting representative video segments from video frame sequences automatically archived in multiple past operation cycles, and at the same time associating historical positioning records with radar echo data to construct complete behavior context information. Image data is used to characterize personnel movement postures and behavior profiles, spatial coordinate data is used to restore the trajectory distribution pattern of personnel in the work area, and relative distance data is used to capture the dynamic distance change trend of personnel approaching the equipment. Various types of data are aligned according to timestamps to form multi-modal behavior samples with consistent time sequence. Each sample also contains the time of behavior occurrence, personnel identity number and area identification information, which are used for subsequent analysis of behavior background conditions and typical working conditions with high risks.

[0045] S202: Manually label the historical sample data set, where the labeling content includes the behavior category of the personnel and the corresponding risk type.

[0046] Specifically, the historical sample data set is manually annotated, and the annotated content includes the behavior categories of personnel and their corresponding risk types. Specifically, manual annotation is performed by security analysts with field experience. The annotation interface displays image frames, trajectory line graphs and distance change curves, and provides a timeline adjustment function for locating peak behavior periods. Behavior categories include but are not limited to normal inspections, standing observations, rapid approaches, bending operations, wandering and sprinting, etc. Risk types are divided into low, medium and high levels according to the enterprise safety level standards. Annotators assign risk labels to samples based on the environment in which the behavior occurs, the state of proximity to the equipment and the duration. At the same time, a two-person review mechanism is used to ensure the consistency and accuracy of the sample labels, and ultimately form a high-credibility supervision sample set for training.

[0047] S203: Based on the labeled sample data, a multimodal feature fusion model is constructed using a convolutional neural network and a time series modeling module, and the multimodal feature fusion model is trained through supervised learning to obtain a pre-trained personnel behavior model.

[0048] Specifically, based on the labeled sample data, a multimodal feature fusion model is constructed using convolutional neural networks and time series modeling modules, and the multimodal feature fusion model is trained through supervised learning to obtain a pre-trained personnel behavior model. Specifically, the image sequence of each sample is first input into the convolutional neural network model to extract the spatial image feature tensor. At the same time, the corresponding positioning coordinates and relative distance sequences are input into the time series modeling module such as the bidirectional long short-term memory network (Bi-LSTM) to capture the spatial movement rules and approach trend changes. The feature fusion module performs vector splicing and weighted fusion on the above two types of features to form a unified feature expression vector and input it into the fully connected layer for classification learning. During the training process, the cross-entropy loss function is used for dual-task supervised learning of behavior categories and risk levels. After the model training is completed, it can output structured behavior recognition results and has a certain generalization ability to adapt to different personnel and working conditions.

[0049] In one embodiment, if Figure 4 As shown, in step S20, the behavior data and location data are analyzed using the pre-trained personnel behavior model to obtain the behavior recognition result, which specifically includes: S21: Perform multimodal feature fusion on the behavior data and location data to construct a fused feature vector.

[0050] Specifically, the behavioral data and position data are fused with multimodal features to construct a fused feature vector. Specifically, this includes extracting skeleton posture key points from each frame in the image sequence to construct a static action feature map, and standardizing and dimensionally expanding the spatial position coordinates and the relative speed collected by the radar at the same time point. The missing frames are aligned through an interpolation algorithm, and the image features, trajectory coordinate sequences, and distance information sequences are synchronously spliced ​​into a unified temporal feature matrix through the time axis. The attention mechanism is introduced in the fusion layer to weight the importance of each modal information, generating a fusion vector with behavioral action patterns, position trends, and proximity dynamic features. The fused feature vector has cross-modal feature consistency and strong context association capabilities, providing more sufficient input expression for the reasoning and recognition of subsequent models.

[0051] S22: Input the fused feature vector into a pre-trained personnel behavior model for inference and recognition, and output the corresponding behavior recognition result. The behavior recognition result includes the personnel's behavior category, behavior confidence, behavior occurrence area label and behavior duration.

[0052] Specifically, the fused feature vector is input into a pre-trained personnel behavior model for inference and identification, and the corresponding behavior recognition result is output. The behavior recognition result includes the personnel behavior category, behavior confidence, behavior occurrence area label and behavior duration. Specifically, it includes calling the personnel behavior model inference interface deployed in the edge computing node, and inputting the fused feature vector into the neural network inference module. The module performs forward calculation on the input features based on the weight parameters obtained during training. The model output is a multi-dimensional vector, which is subjected to Softmax normalization and maximum value selection operations by the post-processing module to determine the final behavior category, and at the same time calculates the confidence probability of the category as a behavior confidence indicator. The recognition result also includes the current personnel area label determined based on the mapping relationship between trajectory data and the on-site area, such as a safe area, buffer zone or restricted area. In addition, the current recognition behavior duration is accumulated on the timeline as a reference dimension for risk assessment, and the result is structured and output to the subsequent risk judgment module for use.

[0053] In one embodiment, if Figure 5 As shown, in step S30, the risk level corresponding to the human behavior is determined based on the behavior recognition result and the current operating status of the target device, specifically including: S31: Extract the behavior category, behavior confidence, behavior occurrence area label and behavior duration information from the behavior recognition results as the basis for determining the human behavior.

[0054] Specifically, the behavior category, behavior confidence, behavior occurrence area label and behavior duration information in the behavior recognition results are extracted as the basis for determining the human behavior. Specifically, it includes reading the behavior category label from the structured recognition result output in the previous stage and mapping it to the system preset behavior type set, determining the reliability level of the current recognition result through the confidence score threshold, and parsing the spatial area label carried in the recognition result to determine whether the person is in a restricted area, warning area or safe area. Then, the cumulative time that the behavior category has lasted in the current scenario is synchronously extracted from the system records. If the time exceeds the behavior maintenance threshold defined in the policy, the behavior is regarded as a continuous behavior and is marked with a high risk. This judgment basis serves as one of the key input dimensions for the subsequent behavior-device joint risk analysis.

[0055] S32: Perform comprehensive analysis based on the current operating status of the target device, where the operating status includes running, paused, or standby.

[0056] Specifically, a comprehensive analysis is performed based on the current operating status of the target device. The operating status includes running, paused or standby status. Specifically, it includes obtaining the latest operating status flag of the device through the standard communication interface between the edge node and the device management system. The status information includes the working condition code of the device main control module, current load status, operating cycle stage and other core fields. The system decodes the operating mode of the device according to the status field and maps it into three status labels: running, paused or standby. In the data fusion engine, the personnel behavior status and the equipment operating status are jointly calculated, and the risk sensitivity of the current human-machine collaborative relationship is matched through the state combination matrix. For example, it is determined whether the personnel approaching the rotating mechanism during equipment operation or moving repeatedly in the pause state constitutes a potential dangerous behavior trend.

[0057] S33: Classify situations corresponding to combinations of human behavior and equipment status into low risk, medium risk, or high risk levels based on preset risk mapping rules.

[0058] Specifically, based on the preset risk mapping rules, the situations corresponding to the combination of personnel behavior and equipment status are divided into low risk, medium risk or high risk levels. Specifically, it includes loading a risk level rule table for the joint mapping of behavior categories and equipment status from the risk rule storage module. The rule table defines a combination of multiple behavior status and equipment operation status. Each set of correspondences is accompanied by a corresponding risk level label and explanation. The system retrieves matching items based on the behavior category, area label and current status of the equipment. If there is a clear combination item, the corresponding risk level is directly obtained. If the combined behavior is in the gray area or the confidence level is critical, an approximate level identifier is assigned after fuzzy reasoning scoring through dynamic compensation rules. The risk level result is used as the core index key of the subsequent response strategy matching module, and at the same time triggers the risk level update counter to support the self-optimization learning of the cloud strategy model.

[0059] Furthermore, the system dynamically adjusts the security boundary based on the evolving trends of human behavior and risk level results to enhance the timeliness and adaptability of protection strategies. Specifically, this involves performing a time-series analysis of high-frequency, high-risk behaviors identified over multiple consecutive judgment cycles. The system then calculates the frequency of spatial overlap between the behavior occurrence area and the device boundary and the duration of the behavior. Combined with UWB trajectory trends, the system determines whether individuals frequently wander near the boundary, stray into the boundary, or rapidly approach the boundary. Furthermore, the system references the current operating conditions of the device and the risk level results. If the intensity of these behaviors gradually increases or the risk level remains high over multiple cycles, a dynamic security boundary adjustment mechanism is activated. The current boundary shape and coordinate parameters are loaded into the edge computing node. The boundary expansion ratio, buffer zone thickness, or stretch coefficient in a specific direction are used as dynamic variables to calculate a new boundary range and update the electronic fence control module. The new boundary is then written to the edge control cache to guide policy matching range changes. The dynamically expanded boundary allows the system to trigger the buffer control strategy earlier, improving the lead time of the protection response and the accuracy of the warning. When human behavior gradually returns to normal, the system can shrink the boundary according to a set time threshold, restoring it to its initial state or a more optimized configuration, achieving periodic adaptive adjustment of the boundary.

[0060] In one embodiment, if Figure 6 As shown, before step S33, that is, before classifying the situations corresponding to the combination of personnel behavior and equipment status into low risk, medium risk or high risk levels according to the preset risk mapping rules, the intelligent safety protection management method further includes: S3301: Based on the behavior recognition results collected in the previous operation cycle, the execution effect of the corresponding control instructions, the target device status log and the response accuracy evaluation results, the preset risk mapping rules are adjusted in real time and dynamically.

[0061] Specifically, based on the behavior recognition results collected in the previous operation cycle, the execution effect of the corresponding control instructions, the target equipment status log and the response accuracy evaluation results, the preset risk mapping rules are adjusted in real time and dynamically. Specifically, it includes retrieving the behavior recognition log, control instruction execution feedback and equipment operating condition change records stored in the previous cycle from the edge node. The behavior log contains the behavior type, confidence, trigger time and area label corresponding to each recognition event. The control feedback data records whether the response action is successfully executed, the execution delay time and whether it is interrupted by the voice command. The equipment status log provides the load curve and status change mark of the equipment during the risk judgment period. At the same time, the false alarm events and missed events that actually occurred in this cycle are automatically marked as optimized samples. Feature hierarchical analysis is performed through these historical data, and the rule correction engine is called to perform weighted adjustment on the level boundary conditions in the original risk mapping rules to complete the microstructure update of the risk decision-making mechanism to adapt to the new trend of current personnel behavior characteristics and equipment response deviations.

[0062] S3302: Dynamic adjustment includes: calculating the response effectiveness, false alarm rate and missed alarm rate indicators under different combinations of behaviors and device states, adjusting the corresponding relationship parameters between behavior category, behavior confidence, behavior duration and device operating status, and the updated risk mapping rules are cached to the edge computing node for subsequent risk level judgment.

[0063] Specifically, dynamic adjustment includes: calculating the response effectiveness, false alarm rate and missed alarm rate indicators under different combinations of behaviors and device states, adjusting the corresponding relationship parameters between behavior category, behavior confidence, behavior duration and device operating state, and the updated risk mapping rules are cached to the edge computing node for subsequent risk level judgment. Specifically, it includes clustering the behavior-device state combination samples according to the behavior label dimension, and statistically analyzing whether the corresponding response action of each combination is effectively executed, whether there is excessive response or missing response, and calculating its response accuracy and risk perception deviation coefficient based on the statistics. If a combination shows a high false alarm or missed alarm tendency in two consecutive rounds of evaluation, the weight adjustment operation of its corresponding parameters is triggered, such as increasing the tolerance threshold for low-confidence short-term behavior or reducing the response sensitivity to medium-risk behavior in the non-operating state. After the adjustment is completed, a new version of the rule mapping table is generated and loaded into the local mapping engine of the edge node in cache form to replace the original rule to ensure the optimal adaptability and response robustness of the rules in the next judgment cycle.

[0064] In one embodiment, if Figure 7 As shown, in step S40, based on the risk level, the corresponding security response strategy is matched in the edge computing node, and a control instruction is generated according to the security response strategy, specifically including: S41: A security response policy library containing mapping relationships between multiple response levels, target device types, and control logic is preset in the edge computing node, and the policy library uses risk level as an index dimension.

[0065] Specifically, a security response policy library containing multiple response levels, target device types and control logic mapping relationships is preset in the edge computing node. The policy library uses risk level as the index dimension. Specifically, it includes establishing a response policy library index structure in the local database of the edge node. The structure adopts a multi-level hash table design. The first-level index is the risk level label, the second-level index is the device type identifier, and the third level is the specific control logic entry. Each policy record contains the execution type of the response action, the response priority, the applicable scenario label, the expected execution time and the fault-tolerant recovery plan. The policy library is loaded during the system initialization phase, and the updated version can also be pushed regularly or on demand through the cloud-based policy update module, and the current version is automatically compared for hot replacement to ensure that the policy is always consistent with the latest operating data and optimization logic. When the system is called, the policy entry is quickly found by matching the current risk level and locating the corresponding device category, and it is used as the basis structure for generating control instructions.

[0066] S42: When the risk level is determined to be high risk, an emergency response strategy is called from the strategy library to generate a power-off shutdown control instruction.

[0067] Specifically, when the determined risk level is high, the emergency response strategy is called from the strategy library to generate power-off and shutdown control instructions. Specifically, it includes locating the strategy list under the first-level index with a high risk level, and filtering out the strategy items marked as emergency stop, emergency stop power off, and safety isolation. The control instruction constructor is called to generate a standard control instruction data packet containing the control target, action type, device channel, and execution delay of zero. This instruction type is a non-buffered forced execution instruction with an execution priority higher than any existing task flow. Before packaging, the device identification code and command number are attached through the security signature module to prevent instruction hijacking, and the device communication port of the edge node is immediately called to send the power-off instruction directly to the target device controller end, forming a physical-level risk-cutting action, which is suitable for situations such as personnel rushing into high-voltage areas, approaching the blind spot of robotic arm rotation, or approaching high-speed transmission devices.

[0068] S43: When it is determined to be a medium risk or low risk level, pause action and early warning prompt control instructions are generated respectively. The control instructions are in a standardized format and are encapsulated as Modbus or CAN instruction sets according to the interface protocol of the target device to adapt to different types of devices to perform corresponding actions.

[0069] Specifically, when it is determined to be a medium risk or low risk level, pause action and early warning prompt control instructions are generated respectively. The control instructions are in a standardized format and are encapsulated as Modbus or CAN instruction sets according to the interface protocol of the target device, which are used to adapt to different types of devices to perform corresponding actions. Specifically, it includes retrieving policy entries marked as "delayed processing" or "warning response" under the corresponding risk level from the policy library, and generating a standard control structure with action types of "terminate current task", "temporary standby", "sound and light alarm activation" or "screen prompt" according to the policy definition. The control structure contains the execution type, target device number, effective duration, whether repeated triggering and overwrite mark are supported, and then the instruction encapsulation template is selected according to the communication protocol type recorded in the device registration information. If the device communication interface is the serial Modbus protocol, the package is in the Modbus RTU frame format. If the device uses CAN communication, it is assembled into a CAN data frame format with an identifier. Finally, it is uniformly sent to the target device for execution through the protocol middleware to ensure the compatibility, stability and effectiveness of the control process.

[0070] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0071] In one embodiment, an intelligent security protection management system is provided, which corresponds one-to-one to an intelligent security protection management method in the above embodiment. Figure 8 As shown, the intelligent security protection management system includes a data collection module, an analysis module, a risk judgment module, an instruction generation module, a voice interaction module and an update module. The functional modules are described in detail as follows: Data collection module, used to obtain behavioral data and location data of personnel in industrial sites; The analysis module is used to analyze the behavior data and location data using the pre-trained personnel behavior model to obtain behavior recognition results; The risk assessment module is used to determine the risk level of human behavior based on the behavior recognition results and the current operating status of the target device; The instruction generation module is used to match the corresponding security response strategy in the edge computing node based on the risk level and generate control instructions based on the security response strategy. The control instructions are used to drive the target device to perform the corresponding response action; The voice interaction module is used to obtain the voice input of the on-site operator through the voice interaction module, perform semantic recognition on the voice input, and obtain a voice recognition result. When the voice recognition result meets the preset emergency command condition, the corresponding emergency control operation is executed to overwrite the current control command; The update module is used to upload behavior recognition results, risk levels, control instructions and voice interaction records to the cloud platform, optimize and update the security response strategy based on the cloud platform's reinforcement learning model, and synchronize the optimized strategy to the edge computing node.

[0072] Optionally, the data collection module includes: The image acquisition submodule is used to collect image sequences of people through cameras and extract posture key points and motion features from the image sequences; The coordinate acquisition submodule is used to collect the spatial coordinates and movement trajectory information of personnel in the target area through the UWB positioning module; The distance acquisition submodule is used to obtain the relative distance information and approach speed information between the person and the target device through the millimeter wave radar.

[0073] Optionally, the intelligent safety protection management system further includes: The historical sample collection module is used to collect historical sample data sets, which include image data, spatial coordinate data and relative distance data; The annotation module is used to manually annotate historical sample data sets, including the behavior categories of personnel and their corresponding risk types; The model building module is used to build a multimodal feature fusion model based on the labeled sample data using the convolutional neural network and the time series modeling module, and train the multimodal feature fusion model through supervised learning to obtain a pre-trained personnel behavior model.

[0074] Optional analysis modules include: The data fusion submodule is used to fuse the behavior data and location data into multimodal features and construct the fused feature vector; The inference and recognition submodule is used to input the fused feature vector into a pre-trained personnel behavior model for inference and recognition, and output the corresponding behavior recognition results. The behavior recognition results include the personnel's behavior category, behavior confidence, behavior occurrence area label and behavior duration.

[0075] Optional risk assessment modules include: The feature extraction submodule is used to extract the behavior category, behavior confidence, behavior occurrence area label and behavior duration information from the behavior recognition results as the basis for judging human behavior; The combined analysis submodule is used to conduct a comprehensive analysis based on the current operating status of the target device, which includes running, paused, or standby; The judgment submodule is used to classify the situations corresponding to the combination of personnel behavior and equipment status into low risk, medium risk or high risk levels according to the preset risk mapping rules.

[0076] Optionally, the intelligent safety protection management system further includes: An evaluation module is used to dynamically adjust the preset risk mapping rules in real time based on the behavior recognition results collected in the previous operation cycle, the execution effect of the corresponding control instructions, the target device status log, and the response accuracy evaluation results; The adjustment module is used for dynamic adjustment, including: calculating the response effectiveness, false alarm rate and missed alarm rate indicators under different combinations of behaviors and device states, adjusting the corresponding relationship parameters between behavior category, behavior confidence, behavior duration and device operating status, and the updated risk mapping rules are cached to the edge computing node for subsequent risk level judgment.

[0077] Optionally, the instruction generation module includes: A preset submodule is used to preset a security response policy library containing multiple response levels, target device types, and control logic mapping relationships in the edge computing node. The policy library uses risk level as the index dimension; The high-risk call submodule is used to call the emergency response strategy from the strategy library and generate power-off and shutdown control instructions when the risk level is determined to be high risk; The medium and low risk call submodule is used to generate pause action and early warning prompt control instructions when the risk level is determined to be medium or low. The control instructions are in a standardized format and are encapsulated as Modbus or CAN instruction sets according to the interface protocol of the target device, which are used to adapt to different types of devices to perform corresponding actions.

[0078] The specific definition of an intelligent security protection management system can be found in the definition of an intelligent security protection management method above and will not be repeated here. Each module in the aforementioned intelligent security protection management system may be implemented in whole or in part through software, hardware, or a combination thereof. Each of these modules may be embedded in or independent of a processor in a computer device in hardware form, or may be stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0079] Those skilled in the art will clearly understand that for the sake of convenience and brevity in description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0080] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. An intelligent security protection management method, characterized in that: The intelligent security protection management method includes: Obtain behavioral and location data of personnel at industrial sites; Analyze the behavior data and location data using a pre-trained personnel behavior model to obtain a behavior recognition result; Determine the risk level corresponding to the human behavior based on the behavior recognition result and the current operating status of the target device; Based on the risk level, a corresponding security response policy is matched in the edge computing node, and a control instruction is generated according to the security response policy, wherein the control instruction is used to drive the target device to perform a corresponding response action; The voice interaction module obtains the voice input of the on-site operator and performs semantic recognition on the voice input to obtain a voice recognition result. When the voice recognition result meets the preset emergency command condition, the corresponding emergency control operation is executed to overwrite the current control command; The behavior recognition results, risk levels, control instructions and voice interaction records are uploaded to the cloud platform, the security response strategy is optimized and updated based on the reinforcement learning model of the cloud platform, and the optimized strategy is synchronized to the edge computing node.

2. The intelligent security protection management method according to claim 1, characterized in that: The acquisition of behavior data and location data of personnel in the industrial site includes: Collecting a sequence of images of a person through a camera, and extracting posture key points and motion features from the sequence of images; The UWB positioning module is used to collect the spatial coordinates and movement trajectory information of people in the target area; The relative distance information and approach speed information between the person and the target device are obtained through millimeter wave radar.

3. The intelligent security protection management method according to claim 1, characterized in that: Before analyzing the behavior data and location data using the pre-trained personnel behavior model, the intelligent security protection management method further includes: Collecting a historical sample data set, wherein the historical sample data set includes image data, spatial coordinate data, and relative distance data; Manually labeling the historical sample data set, where the labeling content includes the behavior category of the person and the corresponding risk type; Based on the labeled sample data, a multimodal feature fusion model is constructed using a convolutional neural network and a time series modeling module, and the multimodal feature fusion model is trained through supervised learning to obtain the pre-trained personnel behavior model.

4. The intelligent security protection management method according to claim 1, characterized in that: The behavior data and location data are analyzed using a pre-trained personnel behavior model to obtain a behavior recognition result including: Performing multimodal feature fusion on the behavior data and the location data to construct a fused feature vector; The fused feature vector is input into the pre-trained personnel behavior model for inference and recognition, and the corresponding behavior recognition result is output. The behavior recognition result includes the personnel behavior category, behavior confidence, behavior occurrence area label and behavior duration.

5. The intelligent security protection management method according to claim 4, characterized in that: Determining the risk level corresponding to the human behavior based on the behavior recognition result and the current operating status of the target device includes: Extracting behavior category, behavior confidence, behavior occurrence area label, and behavior duration information from the behavior recognition results as a basis for determining human behavior; Perform a comprehensive analysis based on the current operating status of the target device, where the operating status includes running, paused, or standby; According to the preset risk mapping rules, the situations corresponding to the combination of the personnel behavior and the equipment status are divided into low risk, medium risk or high risk levels.

6. The intelligent security protection management method according to claim 1, characterized in that: Before classifying the situations corresponding to the combination of personnel behavior and equipment status into low risk, medium risk, or high risk levels according to the preset risk mapping rules, the intelligent safety protection management method further includes: Based on the behavior recognition results collected in the previous operation cycle, the execution effect of the corresponding control instructions, the target device status log and the response accuracy evaluation results, the preset risk mapping rules are dynamically adjusted in real time; The dynamic adjustment includes: calculating the response effectiveness, false alarm rate and missed alarm rate indicators under different combinations of behaviors and device states, adjusting the corresponding relationship parameters between the behavior category, behavior confidence, behavior duration and device operating status, and the updated risk mapping rules are cached to the edge computing node for subsequent risk level judgment.

7. The intelligent security protection management method according to claim 1, characterized in that: The matching of a corresponding security response strategy in the edge computing node based on the risk level and generating a control instruction according to the security response strategy includes: A security response policy library containing mapping relationships between multiple response levels, target device types, and control logic is preset in the edge computing node, with the risk level as the index dimension; When the risk level is determined to be high risk, an emergency response strategy is called from the strategy library to generate a power-off shutdown control instruction; When it is determined to be a medium risk or low risk level, pause action and early warning prompt control instructions are generated respectively. The control instructions are in a standardized format and are encapsulated as Modbus or CAN instruction sets according to the interface protocol of the target device to adapt to different types of devices to perform corresponding actions.

8. An intelligent safety protection management system, characterized in that: The intelligent safety protection management system includes: Data collection module, used to obtain behavioral data and location data of personnel in industrial sites; An analysis module is used to analyze the behavior data and location data using a pre-trained personnel behavior model to obtain a behavior recognition result; A risk judgment module is used to judge the risk level corresponding to the human behavior based on the behavior recognition result and the current operating status of the target device; An instruction generation module is used to match a corresponding security response policy in the edge computing node based on the risk level, and generate a control instruction according to the security response policy, wherein the control instruction is used to drive the target device to perform a corresponding response action; A voice interaction module is used to obtain the voice input of the on-site operator through the voice interaction module, perform semantic recognition on the voice input, and obtain a voice recognition result. When the voice recognition result meets the preset emergency instruction condition, the corresponding emergency control operation is executed to overwrite the current control instruction; An update module is used to upload the behavior recognition results, risk levels, control instructions and voice interaction records to the cloud platform, optimize and update the security response strategy based on the cloud platform's reinforcement learning model, and synchronize the optimized strategy to the edge computing node.

Images