Security risk cognitive training system and method based on behavior trajectory analysis

Abstract

The invention discloses a security risk cognitive training system and method based on behavior trajectory analysis. The system comprises a multi-source data acquisition module, a behavior trajectory analysis module, a dynamic training module and a real-time feedback and evaluation module. The multi-source data acquisition module comprises a sensor network unit and an environment data unit. Physiological, behavior and environmental data are integrated through the multi-source data acquisition module, multi-modal fusion analysis is realized in combination with a Transform architecture, and the one-sidedness problem of traditional single data is solved; an LSTM network and a Prophet model are used to identify an abnormal behavior mode in real time and predict a risk trend, and real-time capture and dynamic early warning of risks are realized; a customized training task is generated by means of PPO reinforcement learning, a dynamic virtual scene and self-adaptive difficulty adjustment are combined, and the problems of fixed training content and lack of personalization are solved; through a multi-modal feedback and quantitative evaluation system, the risk perception capability and the training effect are improved, and closed-loop optimization of data acquisition-analysis-training-evaluation is formed.

Description

Technical Field

[0001] The present invention relates to the technical fields of artificial intelligence, behavior analysis, and safety training, and particularly to a safety risk awareness training system and method based on behavior trajectory analysis. Background Art

[0002] In today's society, safety issues have always been of utmost importance. In industrial production, various mechanical equipment coexists with dangerous chemicals. A slight oversight may trigger serious accidents, endangering the lives of workers and causing economic losses. On campus, students study and live together. From teaching buildings to cafeterias and dormitories, once safety hazards break out, they will cause great harm to their physical and mental health. In public places, there are dense crowds and large mobility. Facility failures or emergencies are extremely likely to threaten public safety and affect social order. Therefore, whether it is industrial production, campus environment, or public places, ensuring personnel safety and preventing accidents are crucial. However, traditional safety training methods still have certain problems when dealing with the complex and changing modern environment: First, most traditional safety training methods rely on unified static teaching materials and standardized examinations. The training content fails to fully consider the differences in behavior patterns and risk perception abilities among different users, making it difficult to teach students in accordance with their aptitudes, resulting in a significant reduction in training effects and an inability to effectively improve the risk awareness ability of each user; Second, risk assessment often relies on post-event feedback information and cannot timely capture potential behavior deviations and risk hazards at the moment when the user's behavior occurs, resulting in a large delay in safety prevention and making it difficult to take effective intervention measures before an accident occurs; Third, the scenarios of virtual simulation drills are usually relatively fixed and difficult to cover the complex and diverse environments and dynamically changing risk situations in the real world, resulting in users lacking the ability to respond when facing a real and complex environment; Fourth, the behavior data collection method is limited to a single dimension, and the isolation of single-dimensional data limits the in-depth analysis and accurate judgment of safety risks; Therefore, a safety risk awareness training system and method based on behavior trajectory analysis are proposed. Summary of the Invention

[0003] The purpose of the present invention is to provide a safety risk awareness training system and method based on behavior trajectory analysis to solve the problems raised in the above background art.

[0004] To solve the above technical problems, a technical solution adopted by this application is: a safety risk awareness training system based on behavior trajectory analysis, including a multi-source data collection module, a behavior trajectory analysis module, a dynamic training module, and a real-time feedback and evaluation module; The multi-source data acquisition module includes a sensor network unit, a visual perception unit, and an environmental data unit; the sensor network unit is used to collect the physiological data of the user in real time, the visual perception unit is used to identify the behavior characteristics of the user in real time, and the environmental data unit is used to collect the environmental data in real time; The behavior trajectory analysis module includes a spatio-temporal trajectory modeling unit, a multi-modal feature fusion unit, and a risk prediction unit; the spatio-temporal trajectory modeling unit extracts the time-dependent features of the behavior sequence through an LSTM network, combines the spatial grid division algorithm to map the movement trajectory to a three-dimensional space coordinate system, and identifies abnormal behavior patterns; the multi-modal feature fusion unit uses a Transformer architecture to fuse behavior features, physiological data, and environmental data to generate a comprehensive risk index; the risk prediction unit predicts the risk trend in the next 5-10 seconds through a time series autoregressive model; The dynamic training module includes a personalized training generation unit, a virtual simulation scenario library unit, and an adaptive difficulty adjustment unit; the personalized training generation unit generates customized training tasks through a reinforcement learning model based on the user's historical behavior data and the current risk assessment results; the virtual simulation scenario library unit integrates VR / AR technology to build multi-type risk simulation environments; the adaptive difficulty adjustment unit dynamically adjusts the task difficulty according to the user's real-time performance; The real-time feedback and evaluation module includes a multi-modal feedback unit, a quantitative evaluation unit, and a training effect tracking unit; the multi-modal feedback unit is used to provide risk prompts through visual feedback, tactile feedback, and auditory feedback; the quantitative evaluation unit includes calculating the behavior compliance score and the risk perception index; the training effect tracking unit is used to establish a user behavior profile and compare the incidence of risk behaviors before and after training.

[0005] As a further preference of this technical solution: the sensor network unit includes a UWB positioning base station and a wearable device, and the wearable device is a smart bracelet; the visual perception unit uses a multi-camera array combined with an AI vision algorithm to identify the behavior characteristics of the user, and the AI vision algorithm is YOLOv8; The physiological data includes the user's heart rate, blood pressure, blood oxygen saturation, movement speed, and movement acceleration; The behavior characteristics include the user's gesture actions, body postures, facial micro-expressions, walking paths, and operation processes; The environmental data includes temperature and humidity, light intensity, noise level, harmful gas concentration, and air pressure.

[0006] As a further preferred embodiment of the present technical solution: the time series autoregressive model is a Prophet model, and the Prophet model is optimized for security risk scenarios, and the accuracy of risk trend prediction is improved by introducing spatiotemporal periodic parameters of behavior trajectory data; The reinforcement learning model is a PPO model, and the reward function of the PPO model includes multi-dimensional evaluation indicators, and the multi-dimensional evaluation indicators include behavioral compliance rewards, risk response rewards and physiological stress rewards.

[0007] As a further preferred embodiment of the present technical solution: the visual feedback provides risk warnings by displaying hot spots of behavior trajectories through a heat map, the tactile feedback provides risk warnings by vibrating a bracelet to prompt risky behaviors, and the auditory feedback provides risk warnings by voice broadcasting risk levels and improvement suggestions.

[0008] As a further preferred embodiment of the present technical solution: the behavioral compliance score is calculated in real time based on preset safety rules; the risk perception index is combined with historical training data to generate a personal risk perception radar chart including emergency response speed, risk identification accuracy, operational compliance, environmental adaptability and teamwork ability.

[0009] As a further preferred embodiment of the present technical solution: the virtual simulation scenario library unit supports various types of risk simulation environments, and dynamically generates risk factors matching real scenarios through real-time access to the environmental data interface of the environmental data unit.

[0010] As a further preferred embodiment of the present technical solution: the system also includes an interactive collaboration module and a data storage and management module; the interactive collaboration module is used for interactive collaboration training between multiple users, constructs team collaboration training tasks through virtual simulation scenarios, and synchronously collects and analyzes the collaborative compliance of multi-person behavior trajectories; the data storage and management module is used to store the data generated by the system operation, and has data backup, recovery, cleaning and access permission control functions.

[0011] To solve the above technical problems, another technical solution adopted by the present application is: a safety risk cognition training method based on behavior trajectory analysis, comprising the following steps: Step 1: Use a multi-source data collection module to collect the user's physiological data, behavioral characteristics and environmental data in real time; Step 2: The collected physiological data, behavioral characteristics and environmental data are transmitted to the behavioral trajectory analysis module for analysis and processing, and the current risk assessment results are generated; Step 3: Based on the user's historical behavior data and the current risk assessment results, the dynamic training module generates customized training tasks through a reinforcement learning model and executes the tasks in the VR / AR risk simulation environment integrated in the virtual simulation scenario library unit. The adaptive difficulty adjustment unit dynamically adjusts the task difficulty according to the user's real-time performance; Step 4: The real-time feedback and evaluation module provides feedback through the multi-modal feedback unit, calculates the behavior compliance score and risk awareness index using the quantitative evaluation unit, and the training effect tracking unit establishes a user behavior profile and compares the incidence of risk behaviors before and after training.

[0012] As a further optimization of this technical solution: In Step 2, the method for generating the current risk assessment result includes the following steps: Step 201: The spatio-temporal trajectory modeling unit extracts the time-dependent features of the behavior sequence through an LSTM network, maps the motion trajectory to a three-dimensional space coordinate system in combination with a spatial grid division algorithm, and identifies abnormal behavior patterns by setting dynamic thresholds; Step 202: The multi-modal feature fusion unit uses the Transformer architecture to perform cross-modal feature fusion on behavior features, physiological data, and environmental data, and assigns dynamic weights to different modal data through an attention mechanism to generate a comprehensive risk index; Step 203: The risk prediction unit uses the Prophet model, introduces spatio-temporal periodic parameters, and predicts the risk trend in the next 5-10 seconds; Step 204: Combine the abnormal behavior pattern, the comprehensive risk index, and the risk trend in the next 5-10 seconds to comprehensively evaluate the current safety risk status and generate the current risk assessment result.

[0013] As a further optimization of this technical solution: When executing tasks in the VR / AR risk simulation environment integrated in the virtual simulation scenario library unit in Step 3, if an error occurs, the system immediately prompts an operation error and generates a targeted training plan.

[0014] Advantages of the present invention: 1. The present invention integrates a sensor network unit, a visual perception unit, and an environmental data unit through a multi-source data acquisition module to collect physiological data, behavior features, and environmental data in real time, and performs multi-modal feature fusion in combination with the Transformer architecture, solving the isolation problem of single-dimensional data acquisition in traditional solutions and realizing a comprehensive and in-depth analysis of safety risks; 2. Through the spatio-temporal trajectory modeling unit of the behavior trajectory analysis module, the present invention uses an LSTM network to extract the time-dependent features of the behavior sequence, combines a spatial grid division algorithm to map the three-dimensional motion trajectory, and real-time identifies abnormal behavior patterns. At the same time, by introducing spatio-temporal periodic parameters through the Prophet model to predict the risk trend in the next 5-10 seconds, the problem of lagging risk assessment is solved, and the real-time capture and dynamic warning of potential risks are realized; 3. Through the personalized training generation unit of the dynamic training module, the base PPO reinforcement learning model and the multi-dimensional reward function generate customized training tasks. Combining the virtual simulation scenario library unit to access environmental data in real-time to dynamically generate risk factors, and the adaptive difficulty adjustment unit to dynamically adjust the task difficulty according to the user's performance, the problems of fixed traditional training content and lack of personalization are solved, and "teaching students in accordance with their aptitude" and the real simulation of complex scenarios are realized; 4. Through the multi-modal feedback mechanism (visual, tactile, auditory) of the real-time feedback and evaluation module to provide real-time risk prompts, the quantitative evaluation system generates a radar chart of personal risk perception ability, and the interactive collaboration module supports multi-user team training and collaborative compliance analysis, solving the problems of single feedback form and difficult-to-quantify training effect, and improving the user's risk perception ability and the team's collaborative emergency handling ability. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0016] Figure 1 It is a schematic diagram of the functional modules of a safety risk perception training system based on behavior trajectory analysis of the present invention; Figure 2 It is a schematic diagram of the flow of a safety risk perception training method based on behavior trajectory analysis of the present invention; Figure 3 It is a schematic diagram of the flow of a method for generating the current risk assessment result of the present invention. Detailed Embodiments

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment Figure 1 is a schematic diagram of the functional modules of a safety risk perception training system based on behavior trajectory analysis according to an embodiment of the present application. As Figure 1 shown, a safety risk perception training system based on behavior trajectory analysis includes a multi-source data acquisition module, a behavior trajectory analysis module, a dynamic training module, and a real-time feedback and evaluation module; The multi-source data acquisition module includes a sensor network unit, a visual perception unit, and an environmental data unit; the sensor network unit is used to collect the physiological data of the user in real time, the visual perception unit is used to identify the behavior characteristics of the user in real time, and the environmental data unit is used to collect the environmental data in real time; Specifically, the sensor network unit selects UWB positioning base stations and smart bracelets integrated with multiple sensors; the UWB positioning base stations are deployed in the training area, and by sending and receiving ultra-wideband signals, high-precision positioning of the smart bracelets is achieved, and the positioning accuracy can reach the centimeter level; the smart bracelets are built-in with heart rate sensors, blood pressure sensors, blood oxygen sensors, accelerometers, gyroscopes, etc.; the heart rate sensors use photoplethysmography (PPG) technology to obtain heart rate data by detecting changes in blood vessel volume, and the sampling frequency is set to 1 time per second to monitor the user's heart rate changes in real time; the blood pressure sensors use the oscillometric method to measure blood pressure and automatically measure it once every 5 minutes to ensure the acquisition of relatively stable blood pressure data; the blood oxygen sensors are also based on PPG technology to monitor the blood oxygen saturation in real time; the accelerometers and gyroscopes collect motion speed and motion acceleration data at a frequency of 50 Hz, which can accurately capture changes in the user's motion state; the smart bracelets transmit the collected physiological data to the data aggregation node through Bluetooth, and then the aggregation node transmits the data to the behavior trajectory analysis module through a wired network (such as Ethernet) to ensure the stability and real-time of data transmission; The visual perception unit uses high-definition cameras. By installing multiple high-definition cameras in the training area, a multi-camera array is formed to ensure visual dead angle-free coverage; the high-definition cameras are selected to have a high frame rate (such as 60 fps) and a wide dynamic range to meet the shooting requirements under different lighting conditions. The cameras are connected to an edge computing device, which is equipped with a YOLOv8 algorithm model. The YOLOv8 algorithm has been pre-trained and fine-tuned for behavior characteristics such as gesture actions, body postures, facial micro-expressions, walking paths, and operation processes, and can quickly and accurately identify these behavior characteristics. The edge computing device performs real-time analysis on the video data collected by the cameras, extracts the behavior characteristic data, and transmits the processed characteristic data to the behavior trajectory analysis module through a network (such as Wi-Fi) to reduce the data transmission volume and improve the system response speed; The environmental data unit adopts an environmental monitoring device integrated with temperature and humidity sensors, light sensors, noise sensors, harmful gas sensors, and barometric pressure sensors; the temperature and humidity sensors are high-precision digital sensors with measurement accuracies of ±0.5°C and ±2%RH respectively, and data is collected once every 1 minute; the light sensor uses the principle of a photodiode and can accurately measure the light intensity, with a sampling frequency set to 1 time per minute; the noise sensor obtains the noise level by measuring the sound pressure level in the environment and collects data once per second; the harmful gas sensor detects common harmful gases (such as carbon monoxide, carbon dioxide, formaldehyde, etc.) and selects corresponding detection principles (such as electrochemistry, semiconductor, etc.) according to the characteristics of different gases, and collects data once every 5 minutes; the barometric pressure sensor measures the atmospheric pressure and collects data once every 2 minutes; the environmental monitoring device sends the collected environmental data to the data gateway through a wireless transmission module (such as ZigBee), and then the data gateway transmits the data to the behavior trajectory analysis module to achieve real-time collection and transmission of environmental data.

[0019] The behavior trajectory analysis module includes a spatio-temporal trajectory modeling unit, a multi-modal feature fusion unit, and a risk prediction unit; the spatio-temporal trajectory modeling unit extracts the time-dependent features of the behavior sequence through an LSTM network, combines the spatial grid division algorithm to map the movement trajectory to a three-dimensional space coordinate system, and identifies abnormal behavior patterns; the multi-modal feature fusion unit uses a Transformer architecture to fuse behavior features, physiological data, and environmental data to generate a comprehensive risk index; the risk prediction unit predicts the risk trend in the next 5 - 10 seconds through a time series autoregressive model. Specifically, the specific method for identifying abnormal behavior patterns is as follows: First, obtain user behavior sequence data (such as walking paths, operation procedures), physiological data (movement speed, acceleration to assist in judging the movement state), and environmental data (spatial layout affects the movement trajectory) from the multi-source data collection module, clean the behavior sequence data, and remove outliers and duplicate data; align the physiological data and environmental data with the behavior sequence data in time to ensure the relevance of the data. Then, construct an LSTM network with 3 hidden layers, with 128 neurons set in each hidden layer. The input layer receives the preprocessed behavior sequence data, and the data dimension is determined according to the number of behavior characteristics. For example, if there are 5 behavior characteristics, the input dimension is 5; the time step of the LSTM network is set to 10, that is, the behavior sequence data of 10 time points is processed each time to fully capture the time-dependent features; the activation function selects the ReLU function, the optimizer uses the Adam optimizer, and the learning rate is set to 0.001. Next, according to the actual spatial range of the training scenario, it is divided into three-dimensional spatial grids of equal size. For example, for a training space that is 50 meters long, 30 meters wide, and 5 meters high, it can be divided into grids with a side length of 1 meter. Each grid has a unique coordinate identifier for positioning the user's movement trajectory; Finally, the LSTM network outputs the time-dependent feature vectors of the behavior sequence. By combining it with the result of the spatial grid division, and by calculating indicators such as the residence time, movement speed, and movement direction of the user in different grids, and comparing them with the preset normal behavior pattern threshold. For example, if the user stays in a dangerous area grid (such as a grid near dangerous equipment) for more than 10 seconds, or the movement speed is abnormally fast (more than twice the normal walking speed), it is determined as an abnormal behavior pattern, and the time, location, and type of the abnormal behavior are recorded; The specific method for generating the comprehensive risk index is as follows: First, collect the behavior feature data (gesture actions, body postures, etc.) from the visual perception unit, the physiological data (heart rate, blood pressure, etc.) from the sensor network unit, and the environmental data (temperature and humidity, harmful gas concentration, etc.) from the environmental data unit. Normalize these data and map the data to the [0, 1] interval for easy model processing; Then, construct a Transformer architecture containing 6 encoder layers. Each encoder layer consists of a multi-head attention mechanism, a feed-forward neural network, and layer normalization; the multi-head attention mechanism is set with 8 heads to capture data features from different perspectives; the feed-forward neural network contains two fully connected layers, and the activation function is selected as the GELU function; Finally, use the normalized behavior features, physiological data, and environmental data as the input of the Transformer architecture. The Transformer architecture automatically learns the associations between different modality data through the multi-head attention mechanism and fuses the data; at the output layer, the fused features are mapped to a scalar value, that is, the comprehensive risk index, and the value range of the comprehensive risk index is set to [0, 100]. The larger the value, the higher the risk; for example, when the behavior features show that the user has dangerous operations, the physiological data indicates that the user is in a tense state (fast heart rate), and the environmental data shows the presence of harmful gas leakage, the comprehensive risk index will increase accordingly; The specific method for predicting the risk trend in the next 5 - 10 seconds is as follows: First, obtain the comprehensive risk index sequence generated by the multi-modal feature fusion unit and the abnormal behavior pattern data identified by the spatio-temporal trajectory modeling unit, and arrange these data in chronological order to form time series data; Then, the Prophet model is used for risk trend prediction. When initializing the model, appropriate seasonal cycle parameters are set according to the characteristics of the security risk scenarios. For example, for industrial production scenarios, considering the work shifts of workers (such as three-shift work), daily and weekly cycles are set; for campus scenarios, semester and weekday / weekend cycles are set. At the same time, the smoothing parameter of the model is adjusted through cross-validation to optimize the prediction performance of the model; Finally, the sorted time series data is input into the optimized Prophet model. The model predicts the risk trend in the next 5 - 10 seconds based on the trends and seasonal variations in the historical data. The prediction results are presented in the form of risk indices, including information such as predicted values and confidence intervals. For example, the prediction results show that there is an 80% probability that the risk index will be between 70 - 80 in the next 5 seconds, indicating an upward trend in risk and the need to take timely measures for intervention.

[0020] The dynamic training module includes a personalized training generation unit, a virtual simulation scenario library unit, and an adaptive difficulty adjustment unit. The personalized training generation unit generates customized training tasks through a reinforcement learning model based on the user's historical behavior data and the current risk assessment results. The virtual simulation scenario library unit integrates VR / AR technologies to construct various types of risk simulation environments. The adaptive difficulty adjustment unit dynamically adjusts the task difficulty according to the user's real-time performance. Specifically, the specific implementation method of the personalized training generation unit is as follows: First, obtain the user's historical behavior data from the data storage and management module, including information such as operation processes, risk response situations, and behavior compliance in previous training. At the same time, receive the current risk assessment results output by the behavior trajectory analysis module, such as comprehensive risk indices and abnormal behavior patterns. Then, perform feature extraction on these data. For example, calculate the average reaction time and risk response success rate of the user in different types of risk scenarios, and normalize the extracted features to make them comparable. Then, select the Proximal Policy Optimization (PPO) reinforcement learning model, which has high sample efficiency and stability and is suitable for generating customized training tasks. Define the state space of the PPO model, including user historical behavior features, current risk assessment indicators, etc. The action space is the set of training tasks, such as operation tasks and emergency handling tasks at different difficulty levels. At the same time, design a reward function, comprehensively considering behavior compliance rewards, risk response rewards, and physiological stress rewards. For example, if the user strictly adheres to safety rules during training, give a behavior compliance reward; if the user successfully responds to a risk event, give a risk response reward according to the response effect; if the physiological stress indicators (such as heart rate and blood pressure) are within a reasonable range, give a physiological stress reward. Finally, the normalized feature data is input into the PPO model, and the model selects the optimal action according to the current state, that is, generates customized training tasks; the training tasks include information such as task objectives, task scenario descriptions, operation step requirements, etc.; for example, for users with relatively weak risk perception ability, generate simple operation process training tasks, with the goal of enabling users to familiarize themselves with basic safety operations; for users with relatively strong risk response ability, generate complex emergency handling tasks to test the decision-making and operation ability of users in high-risk scenarios. The specific implementation of the virtual simulation scenario library unit is as follows: First, adopt a mature VR / AR development engine, such as Unity or Unreal Engine, integrate a head-mounted display device (such as Oculus Rift, HTC Vive) and a gesture recognition device (such as Leap Motion) to achieve an immersive virtual simulation experience; and use 3D modeling software (such as Blender, 3ds Max) to create three-dimensional models of various training scenarios, including industrial production workshops, campus environments, public places, etc., and perform texture mapping, lighting settings, etc. on the models to make them realistic. Next, construct multiple risk simulation environments according to different application scenarios and risk types; for example, in the industrial production scenario, simulate risk events such as mechanical equipment failures and chemical leaks; in the campus scenario, simulate natural disasters such as fires and earthquakes; in the public place scenario, simulate emergencies such as stampedes and electrical fires; and design detailed scenario scripts for each risk simulation environment, including event trigger conditions, development processes, influencing factors, etc.; for example, in the chemical leak scenario, set parameters such as the location of the leak source, the leak rate, and the toxicity of the leaked substance, and simulate the diffusion process of the leak and its impact on the environment and personnel according to these parameters. Finally, connect to the environmental data interface of the environmental data unit in real time to obtain the current environmental data, such as temperature and humidity, light intensity, noise level, etc., and dynamically adjust the parameters of the risk simulation environment according to these environmental data to make the virtual scene match the real environment; for example, in a high-temperature environment, increase the likelihood of a fire and the speed of fire spread; in a low-light environment, reduce the user's visual perception ability and increase the risk of operation errors. The specific implementation of the adaptive difficulty adjustment unit is as follows: First, during the process of the user performing training tasks, collect the user's operation data in real time, including operation time, operation steps, operation results, etc. At the same time, monitor the user's physiological data, such as heart rate, blood pressure, blood oxygen saturation, etc., to evaluate the user's stress state; and transmit the collected data to the analysis module to calculate the user's real-time performance indicators, such as task completion time, error rate, risk response score, etc. Then, according to the user's real-time performance metrics, a difficulty adjustment strategy is formulated; for example, if the user has a high error rate and a long task completion time in the current task, it indicates that the task difficulty is too high and the difficulty needs to be reduced; if the user can complete the task quickly and accurately and the physiological stress metrics are stable, it indicates that the task difficulty is too low and the difficulty needs to be increased; among them, the difficulty adjustment can be achieved by adjusting the task complexity, the occurrence frequency of risk events, the operation time limit, etc.; for example, in the easy difficulty level, the occurrence frequency of risk events is low and the number of operation steps is small; in the difficult difficulty level, risk events occur frequently, the operation steps are complex, and there are strict restrictions on the operation time. Finally, according to the difficulty adjustment strategy, the task difficulty in the virtual simulation scenario is adjusted in real time. During the task execution process, the task parameters are dynamically modified, such as increasing or decreasing the number of risk events, changing the order of operation steps, etc.; at the same time, prompt information about the difficulty adjustment is provided to the user so that the user can understand the change of the task difficulty; for example, inform the user "The task difficulty has been increased, please get ready" through voice prompts or screen displays.

[0021] The real-time feedback and evaluation module includes a multi-modal feedback unit, a quantitative evaluation unit, and a training effect tracking unit; the multi-modal feedback unit is used to provide risk prompts through visual feedback, tactile feedback, and auditory feedback; the quantitative evaluation unit includes behavior compliance scoring and risk perception index calculation; the training effect tracking unit is used to establish a user behavior profile and compare the incidence of risk behaviors before and after training. Specifically, for the multi-modal feedback unit, it is used to provide risk prompts through visual feedback, tactile feedback, and auditory feedback; among them, the specific implementation method of visual feedback is as follows: On the display interface of the training scenario, the hot spots of the behavior trajectory are displayed in the form of a heat map. By analyzing the user behavior data, the areas where risk behaviors or abnormal behaviors frequently occur are determined and presented on the heat map with different colors and transparencies; for example, use red to represent high-risk areas and orange to represent medium-risk areas. When the user approaches or enters these areas, the heat map will flash to prompt; at the same time, eye-catching warning icons, such as a red exclamation mark, are set on the interface. When the risk level increases, the icon will become larger and flash; in addition, using AR technology, virtual risk prompt signs are superimposed on the real scenario to help users more intuitively understand the location of the risk. The specific implementation method of tactile feedback is as follows: The smart bracelet worn by the user integrates a vibration function. When the system detects a risk behavior, the smart bracelet vibrates at different frequencies and intensities according to the risk level; in the case of low risk, the vibration frequency is low and the intensity is weak; in the case of high risk, the vibration frequency speeds up and the intensity increases; for example, when the user stays in a dangerous area for too long, the bracelet will vibrate quickly and strongly to attract the user's attention. The specific implementation method of auditory feedback is as follows: Adopt speech synthesis technology to broadcast different contents to the user according to the risk level and specific situation; when the risk level is low, the voice prompt is "Please note that there is a certain risk in the current area. Please operate carefully"; when the risk level is medium, the prompt is "The risk level has risen. Please adjust your behavior immediately"; when the risk level is high, an alarm sound is issued, and the prompt is "High risk! Please stop the operation immediately and take corresponding measures"; at the same time, specific improvement suggestions are given for different types of risks, such as "There is a problem with your operation process. Please operate according to the standard steps".

[0022] The quantitative evaluation unit includes the calculation of behavior compliance score and risk perception index; Among them, the calculation method of the behavior compliance score is as follows: According to the preset safety rule library, compare the user's real-time behavior data with it. The safety rule library covers content such as operation process specifications and behavior guidelines. For example, in an industrial scenario, the operation step sequence of specific equipment is specified; check each step of the user's operation. If it conforms to the rules, a certain score is given, and if it does not conform, the corresponding score is deducted; set different score weights according to the importance and risk degree of the behavior. For example, more points are deducted for violations of key operation steps; calculate the behavior compliance score in real time and display it to the user in the form of a progress bar or a number on the training interface, so that the user can understand their behavior compliance situation at any time; The calculation method of the risk perception index is as follows: Collect multiple data of the user during the training process, including the reaction time to risk events, the risk identification accuracy rate, the operation compliance score, the environmental adaptability (performance under different environmental conditions), and the teamwork performance (if team training is involved); assign weights to each data. For example, the weight of the risk identification accuracy rate is relatively high, and the weight of the environmental adaptability is relatively low; calculate the risk perception index by weighted average. The formula is: Risk perception index = reaction time score × weight 1 + risk identification accuracy rate score × weight 2 + operation compliance score × weight 3 + environmental adaptability score × weight 4 + teamwork ability score × weight 5; present the risk perception index in the form of a radar chart to clearly show the user's risk perception ability in different dimensions; The implementation method of using the training effect tracking unit to establish a user behavior profile and compare the incidence of risk behaviors before and after training is as follows: When the user first uses the system, create a dedicated behavior profile, which includes the user's basic information, training history records (training time, training scenario, training task completion situation, etc.), behavior data (behavior trajectory, operation records), evaluation results (behavior compliance score and risk perception index for each training), etc. As the number of user training times increases, continuously update the profile content; After each training session, count the number of risk behaviors occurred by the user during the training process, calculate the incidence rate of risk behaviors, and compare the incidence rate of the current training with that of previous trainings; if the incidence rate of risk behaviors decreases, it indicates that the training has achieved certain effects; if the incidence rate increases, analyze the reasons, such as unreasonable training difficulty settings or the user's unfamiliarity with certain risk scenarios; by tracking the changes in the incidence rate of risk behaviors in the long term, evaluate the training effect of the system and provide a basis for subsequent optimization.

[0023] In this embodiment, specifically: the sensor network unit includes UWB positioning base stations and wearable devices, and the wearable device is a smart bracelet; the visual perception unit uses a multi-camera array combined with an AI vision algorithm for user behavior feature recognition, and the AI vision algorithm is YOLOv8; Among them, UWB positioning base stations are generally reasonably deployed in the training area. For example, in an industrial scenario, they are installed at key positions in the workshop to ensure that the signal covers the entire working area, thereby achieving precise positioning of users wearing smart bracelets. The positioning accuracy can reach a high level, usually accurate to dozens of centimeters; as a wearable device, the smart bracelet integrates multiple sensors. Its built-in heart rate sensor uses photoplethysmography technology to measure the heart rate by irradiating the skin with light of a specific wavelength and according to the absorption and reflection changes of light by blood, and can obtain the user's heart rate data in real time and accurately. The data acquisition frequency is generally set to once per second; the blood pressure sensor uses the oscillometric method to measure blood pressure by using the changes in oscillatory waves and can automatically measure it every few minutes; the blood oxygen saturation sensor is also based on photoplethysmography technology and can continuously monitor the blood oxygen condition; while the accelerometer and gyroscope are responsible for collecting motion speed and motion acceleration data. The accelerometer can detect the magnitude and direction of the user's motion acceleration, and the gyroscope can sense the rotational changes of the user's body. They collect data at a high frequency (such as 50Hz) to provide detailed information about the user's motion state for the system; Among them, the layout of the multi-camera array fully considers the range and perspective of the training scenario to ensure that the user's behaviors can be comprehensively captured. These cameras are connected to a device with a certain computing power to run the YOLOv8 algorithm; the YOLOv8 algorithm is an advanced object detection algorithm. After being trained according to the requirements of this system, it can quickly identify various behavior features of users; during the training process, a large number of image data containing different gesture actions, body postures, facial micro-expressions, walking paths, and operation processes of users are used to optimize the algorithm, so that it can accurately identify targets in complex scenarios; during actual operation, the cameras collect images at a certain frame rate (such as 30fps), and the YOLOv8 algorithm analyzes these images in real time, identifies the user's behavior features, and transmits the results to subsequent modules in a timely manner; Physiological data includes the user's heart rate, blood pressure, blood oxygen saturation, movement speed, and movement acceleration. After collecting these physiological data through various sensors on the smart bracelet, preliminary processing and integration will be carried out. For example, the heart rate data will be filtered to remove noise interference and ensure the accuracy of the data. The processed data will be stored in chronological order and the collection time points will be marked to facilitate subsequent correlation analysis with other data. These physiological data can not only reflect the user's current physical state but also assist in judging the user's psychological stress response in the face of different situations, providing important basis for risk assessment. Behavioral characteristics include the user's gesture actions, body postures, facial micro-expressions, walking paths, and operation procedures. Among them, for gesture actions, the system will judge whether they comply with safety operation specifications, such as whether the specific gestures during device operation are correct. Body postures will focus on whether there are dangerous postures, such as excessive body tilt during high-altitude work. Facial micro-expressions can reflect the user's emotional state, such as emotions like tension and anxiety affecting the accuracy of operation. The walking path will be compared with the preset safe path to determine whether the user enters a dangerous area. The operation procedure will be compared with the preset standard procedure to check for any illegal operation steps. Through a comprehensive analysis of these behavioral characteristics, the system can more accurately assess the safety risks faced by the user.

[0024] Environmental data includes temperature and humidity, light intensity, noise level, harmful gas concentration, and air pressure. By installing temperature and humidity sensors, light sensors, noise sensors, harmful gas sensors, and air pressure sensors at different positions in the training scenario, comprehensive and accurate environmental data can be obtained. The temperature and humidity sensors will continuously monitor the changes in environmental temperature and humidity, providing a basis for evaluating the impact of the environment on the user's body and equipment. The light sensors are used to detect the light intensity to ensure that the lighting conditions in the training environment meet safety standards, as both too strong and too weak light will affect the user's operation and visual judgment. The noise sensors can monitor the environmental noise level, and excessive noise will interfere with the user's attention. The harmful gas sensors will detect specific harmful gases, and once the harmful gas concentration exceeds the standard, an alarm will be issued in a timely manner. The air pressure sensors are used to measure the environmental air pressure, and in some special scenarios, changes in air pressure will have an impact on equipment operation or personnel health. These environmental data will be collected and transmitted in real time and combined with the user's physiological data and behavioral characteristic data to provide more comprehensive risk assessment information for the system.

[0025] In this embodiment, specifically, the time series autoregressive model is the Prophet model. The Prophet model is optimized for safety risk scenarios. By introducing the spatio-temporal periodic parameters of behavioral trajectory data, the accuracy of risk trend prediction is improved. Specifically, in the security risk scenario, the behavioral trajectory data has spatio-temporal periodicity. Taking the industrial production scenario as an example, the operation behaviors of workers show certain patterns on weekdays. The working hours and operation processes are similar every day, and there are also similarities between different weekdays. At the same time, in different areas of the workshop, the probability and pattern of risk occurrence also show periodic changes over time. By introducing these spatio-temporal periodicity parameters, such as the daily cycle in days, the weekly cycle in weeks, and the risk distribution cycle in different areas in space, the Prophet model can use these parameters to better fit the periodic patterns in historical data, thus making more accurate predictions about the risk trends in the next 5 - 10 seconds. The training and tuning of the Prophet model are as follows: Collect a large amount of historical behavioral trajectory data, risk event data, and corresponding time and space information to form a training dataset. During the training process, adjust the hyperparameters of the Prophet model according to the characteristics of the security risk scenario. For example, adjust the smoothing parameter of the model to balance the fitting degree of the model to trends and seasonality, making it more adaptable to the changing characteristics of security risk data. Select the optimal combination of hyperparameters through cross-validation to improve the prediction accuracy of the model. After being trained and optimized, the Prophet model can predict the risk trends in the next 5 - 10 seconds after receiving new behavioral trajectory data, time, and space information. The prediction results not only show the change trend of the risk index but also provide the prediction confidence interval to help the system judge the reliability of the prediction. For example, the model predicts that there is a 70% probability that the risk index will be between 60 - 70 in the next 5 seconds, indicating an upward trend in risk. The system can issue an early warning and take corresponding measures accordingly.

[0026] The reinforcement learning model is the PPO model. The reward function of the PPO model includes multi-dimensional evaluation indicators, and the multi-dimensional evaluation indicators include behavioral compliance rewards, risk response rewards, and physiological stress rewards. Among them, the application of the multi-dimensional reward function is as follows: Behavioral compliance rewards: Monitor the user's behavior during the training process in real time according to the preset security rules. When the user's operation process and behavioral actions comply with the security rules, give certain behavioral compliance rewards. For example, in the simulated chemical production operation training, when the user starts the equipment according to the standard process, the system will give corresponding reward scores according to the accuracy and completion time of the operation. The setting of the reward scores is related to the importance and difficulty of the operation, and higher rewards will be obtained for the compliance completion of key operation steps. Risk response rewards: various risk events are set up in virtual simulation scenarios. When users successfully respond to risks, they will be rewarded based on the response effect. The evaluation of response effect includes response time, correctness and effectiveness of handling methods, etc. For example, in a fire simulation scenario, users can quickly discover a fire and take correct fire-fighting measures to control the fire in time. Based on these performances, corresponding rewards will be given. The amount of rewards will be adjusted according to the severity of the risk. Successful response to high-risk events will receive more generous rewards. Physiological stress reward: Monitor the user's physiological data during training, such as heart rate, blood pressure, blood oxygen saturation, etc., through smart bracelets and other devices. When the user faces risks, the physiological indicators are within the normal range, indicating that the user can remain calm and respond, and a physiological stress reward is given. For example, in a high-pressure risk scenario, the user's heart rate does not rise excessively and the blood pressure is stable, indicating that the user's psychological and physiological state is good, and the system will give a certain reward. The calculation of the reward will comprehensively consider the changes in multiple physiological indicators and the reasonable range of these indicators in different risk scenarios.

[0027] Among them, the training and task generation methods of the PPO model are as follows: Integrate behavioral compliance rewards, risk response rewards, and physiological stress rewards into the reward function of the PPO model to build a complete reward system; use this reward system to train the PPO model so that the model learns how to generate more reasonable training tasks based on the user's behavior and physiological state; during the training process, the model continuously adjusts its strategy to maximize the cumulative reward; the trained PPO model can generate personalized training tasks based on the user's historical behavior data and current risk assessment results, thereby improving the pertinence and effectiveness of the training.

[0028] In this embodiment, specifically: visual feedback provides risk warnings by displaying hot spots of behavior trajectories in a heat map, tactile feedback provides risk warnings by vibrating a wristband to prompt risk behaviors, and auditory feedback provides risk warnings by voice broadcasting risk levels and improvement suggestions; The specific implementation of visual feedback is as follows: First, the user's behavior trajectory data is obtained from the behavior trajectory analysis module. This data records the user's movement path and stop location in the training scenario. The behavior trajectory data is cleaned and preprocessed to remove abnormal data points to ensure the accuracy and reliability of the data. Then, based on the behavior trajectory data, the behavior hotspots of different areas are calculated. Methods such as kernel density estimation can be used to calculate the heat value of each area using the user's stay time, passing frequency, etc. as weights; Next, use visualization tools (such as Python's Matplotlib, Seaborn libraries, etc.) to map the heat values to the color space to generate a heatmap. Generally speaking, the darker the color (such as red), the higher the degree of behavior hotspots in that area and the relatively greater the risk; the lighter the color (such as blue), the lower the degree of hotspots and the relatively smaller the risk. Finally, display the generated heatmap on the monitoring interface of the training scenario or the screen of the user's wearable device. When the user approaches or enters a high-risk hotspot area, the area can be further highlighted by changing the color, flashing, etc. of the heatmap to remind the user of potential risks. The specific implementation method of tactile feedback is as follows: First, the behavior trajectory analysis module monitors the user's behavior in real time and determines whether a risk behavior occurs according to preset risk rules. For example, situations such as the user entering a dangerous area or violating the operation process can be determined as risk behaviors. Then, different vibration modes are set according to the severity and type of the risk behavior. For example, for mild risk behaviors, it can be set to short and weak vibrations; for severe risk behaviors, it is set to long and strong continuous vibrations. When a risk behavior is detected, the system sends a vibration instruction to the smart bracelet worn by the user. After receiving the instruction, the bracelet vibrates according to the preset vibration mode, thus giving the user a tactile risk reminder. The specific implementation method of auditory feedback is as follows: First, based on multi-source data (such as physiological data, behavior characteristics, environmental data, etc.), the behavior trajectory analysis module evaluates the current risk level. The risk level can be divided into three levels: low, medium, and high, or can be more finely divided according to specific needs. Then, according to the evaluated risk level, corresponding voice broadcast content is generated. In addition to informing the user of the current risk level, specific improvement suggestions are also provided. For example, when the risk level is low, the voice can prompt "The current risk level is relatively low, but still pay attention to the operation specifications"; when the risk level is high, the voice can warn "High risk! Immediately stop the current operation and handle it according to the safety process." Finally, the generated voice content is broadcast to the user through an audio device such as a speaker or the user's headphones, ensuring that the voice is clear and easy to understand, and can convey risk information and improvement suggestions in a timely manner.

[0029] In this embodiment, specifically: The behavior compliance score is calculated in real time based on preset safety rules. Among them, the specific calculation method of the behavior compliance score is as follows: First, according to different application scenarios, such as industrial production, campus security, public place emergencies, etc., detailed and comprehensive safety rules are formulated. Taking the industrial production scenario as an example, the rules cover equipment operation procedures, requirements for wearing protective equipment, norms for entering dangerous areas, etc. These rules exist in the form of clear and definite clauses and can be transformed into logical judgment conditions recognizable by computers. And a corresponding score weight is set for each rule to reflect its importance in overall safety. For example, the rules related to the operation of key equipment have a higher weight, while some relatively minor operation specifications have a lower weight. Then, the system collects the behavior data of users in training or actual operations in real time. This data comes from a multi-source data acquisition module, including behavior characteristics obtained by the visual perception unit (such as gesture actions, operation procedures), physiological data obtained by the sensor network unit (to assist in judging the behavior state), etc. And the collected real-time behavior data is compared one by one with the preset safety rules. For each rule, it is judged whether the user's behavior meets the requirements. For example, it is checked whether the user performs the operation in accordance with the specified step sequence when operating a specific device and whether the protective equipment is worn correctly, etc. Finally, according to the comparison results, corresponding scores are given to the behaviors that comply with the rules, and corresponding scores are deducted from the behaviors that do not comply with the rules. The initial value of the behavior compliance score can be set to full marks. As the rules are checked and compared, addition or subtraction operations are performed according to the results. And the behavior compliance score is updated in real time and displayed on the user's operation interface or monitoring terminal, so that users and managers can understand the behavior compliance situation at any time. For example, in a chemical production operation training, if the user correctly completes the first few steps of starting the equipment, the system will increase the corresponding score according to the score of the corresponding rule. If the user fails to wear protective gloves as required in subsequent operations, the corresponding score will be deducted.

[0030] The risk perception index generates a personal risk perception ability radar chart including emergency response speed, risk identification accuracy rate, operation compliance, environmental adaptability, and teamwork ability in combination with historical training data. Emergency response speed: Calculate the average response time of the user when facing a risk event, that is, the time interval from the occurrence of the risk event to the user starting to take countermeasures. Statistical analysis is performed on the response time data in multiple trainings to obtain a relatively accurate emergency response speed index. Risk identification accuracy rate: Statistically calculate the ratio of the number of times the user correctly identifies risk events to the total number of risk events in training. For example, in multiple simulated fire scene trainings, the ratio of the number of times the user correctly identifies a fire to the total number of simulations is the risk identification accuracy rate. Operation compliance: Directly use the historical average value of the behavior compliance score as the value of this index to reflect the operation compliance degree of the user in long-term training. Environmental adaptability: Analyze the user's performance under different environmental conditions (such as different temperature and humidity, light intensity, noise level, etc.), and comprehensively consider factors such as the accuracy, efficiency of operations, and physiological stress responses to calculate an environmental adaptability index. For example, evaluate environmental adaptability by comparing data such as the operation error rate and heart rate changes of the user in a high-temperature environment and a normal-temperature environment. Team collaboration ability: If the training involves team collaboration scenarios, count data such as the completion of team tasks and the degree of cooperation among members. For example, calculate indicators such as the completion time of team tasks and the communication efficiency among members during the task to comprehensively evaluate the team collaboration ability.

[0031] In this embodiment, specifically: The virtual simulation scenario library unit supports various types of risk simulation environments and dynamically generates risk elements that match the real scenario through the environmental data interface of the environmental data unit in real time. Specifically, the virtual simulation scenario library unit has pre-constructed a rich and diverse set of risk simulation scenarios, covering different application fields and potential risk scenarios. In industrial scenarios, it can simulate chemical leaks and explosions during the chemical production process, equipment failures and human operation errors in mechanical manufacturing, etc. In the field of public safety, there are disaster scenarios such as fires, earthquakes, and stampedes. In campus scenarios, it can simulate risk situations such as laboratory accidents and campus fire accidents. By constructing these different types of scenarios, it meets the safety risk awareness training needs of different user groups (such as industrial practitioners, students, and public place managers), enabling them to become familiar with various risk scenarios in a virtual environment and enhance their response capabilities. The virtual simulation scenario library unit is connected to the environmental data interface of the environmental data unit in real time and continuously obtains environmental data in the real scenario, such as information on temperature and humidity, light intensity, noise level, harmful gas concentration, and air pressure. When the harmful gas sensor detects an increase in the harmful gas concentration in the real environment, the virtual simulation scenario library unit will, based on this data, generate risk elements of harmful gas leakage in the corresponding chemical production simulation scenario, including the leakage location, leakage volume, diffusion range, etc., all of which match the real situation. If the temperature and humidity data change, in the fire simulation scenario, the risk elements such as the fire spread speed and smoke diffusion range will be adjusted according to these changes, making the risk situation in the virtual scenario closely related to the real environment and providing users with a highly realistic risk simulation experience to enhance the training effect.

[0032] In this embodiment, specifically: The system also includes an interaction and collaboration module and a data storage and management module. The interaction and collaboration module is used for interactive collaboration training among multiple users, constructs team collaboration training tasks through virtual simulation scenarios, and synchronously collects and analyzes the collaborative compliance of the behavior trajectories of multiple people. The data storage and management module is used to store the data generated during the operation of the system and has functions such as data backup, recovery, cleaning, and access permission control. Among them, the interactive collaboration module mainly focuses on the interactive collaboration training among multiple users. It creates team collaboration training tasks with the help of virtual simulation scenarios, and synchronously collects and analyzes the collaborative compliance of the behavior trajectories of multiple people, so as to improve the collaboration ability and response level of team members when facing risks. The specific implementation methods are as follows: Based on various risk simulation environments in the virtual simulation scenario library unit, different types of team collaboration training tasks are designed. For example, in an industrial scenario, an emergency treatment task for chemical leakage is set up, and team members need to assume different roles, such as leakage source plugging personnel, evacuation guidance personnel, medical rescue personnel, etc. The goals, processes, and rules of each task are clarified to ensure that team members are clear about their respective responsibilities and collaboration requirements. At the same time, according to the difficulty and complexity of the task, risk elements in the virtual scenario are adjusted, such as the leakage speed, diffusion range, etc., to meet the training needs of teams at different levels; Provide multi-user access function, allowing multiple users to enter the same virtual simulation scenario for training at the same time. Users can log in to the system through their respective terminal devices (such as VR devices, computers, etc.) and join the specified training task to achieve real-time interaction function, enabling team members to communicate by voice and cooperate in actions, etc. For example, in a fire rescue training, team members can determine the rescue plan through voice communication and synchronously execute rescue actions in the virtual scenario; Synchronously collect the behavior trajectory data of each team member, including their movement paths, operation behaviors, etc. in the virtual scenario. These data can be obtained through visual perception units, sensor network units, etc. Analyze the collaborative compliance of the behavior trajectories of multiple people according to the preset collaborative rules and safety standards. For example, check whether team members operate according to the specified processes and sequences when performing tasks, and whether they cooperate tacitly, etc. For behaviors that do not meet the collaborative requirements, give feedback and prompts in time to help team members improve their collaboration methods.

[0033] Among them, the data storage and management module is responsible for storing various types of data generated during the operation of the system, and at the same time has functions such as data backup, recovery, cleaning, and access permission control to ensure the security, integrity, and availability of the data. The specific implementation methods are as follows: Adopt a suitable database management system (such as relational database MySQL, non-relational database MongoDB, etc.) to store the data generated during the operation of the system. These data include user basic information, training records, behavior trajectory data, evaluation results, environmental data, etc. Classify and store different types of data and establish corresponding indexes for quick query and retrieval. For example, establish a time index for user training records to facilitate querying and statistics according to the training time; Develop a regular data backup strategy to back up the data in the database to external storage devices (such as tapes, disk arrays, etc.) or cloud storage services. The backup period can be set according to the importance and update frequency of the data. For example, a full backup can be performed daily, weekly, or monthly. At the same time, provide a data recovery function so that when the database fails or data is lost, the data can be quickly restored from the backup to ensure the normal operation of the system. The recovery process should be simple, reliable, and able to verify the integrity of the restored data. Regularly clean up expired or useless data to free up database storage space and improve system performance. For example, delete training records, temporary data, etc. that have exceeded the retention period. Before cleaning up the data, archive and back up the data to ensure that important data is not accidentally deleted. At the same time, record the operation logs of data cleaning for auditing and traceability. Establish a user role and permission management system to assign different access permissions to different users or user groups. For example, system administrators can have the highest permissions and be able to perform operations such as reading, writing, backing up, and restoring data. Ordinary users can only view their own training records and evaluation results. Through the authentication and authorization mechanism, ensure that only authorized users can access and operate the data. When a user logs in to the system, verify their identity information and restrict the access scope to the data according to their permission level. At the same time, record the access operation logs of users for security auditing and monitoring.

[0034] In summary, a security risk awareness training system based on behavior trajectory analysis provided by an embodiment of the present invention comprehensively collects user physiological, behavioral, and environmental data through a multi-source data collection module; the behavior trajectory analysis module conducts in-depth analysis to generate risk assessment results; the dynamic training module provides customized training; the real-time feedback and evaluation module realizes effective feedback and evaluation; in addition, the system also includes an interactive collaboration module and a data storage and management module to support team training and ensure data security; each module of the system complements each other, effectively solves the problems existing in traditional security training, and comprehensively improves the user's security risk awareness and response ability.

[0035] Figure 2 It is a flow schematic diagram of a security risk awareness training method based on behavior trajectory analysis according to an embodiment of the present invention. It should be noted that if there are substantially the same results, the method of the present application is not limited to Figure 2 the flow sequence shown. As Figure 2 - Figure 3 shown: A security risk awareness training method based on behavior trajectory analysis includes the following steps: Step 1: Use the multi-source data collection module to collect the user's physiological data, behavioral characteristics, and environmental data in real time; these data provide the basis for subsequent analysis and training. Step 2: Transmit the collected physiological data, behavioral characteristics, and environmental data to the behavioral trajectory analysis module for analysis and processing, and generate the current risk assessment result; Step 3: Based on the user's historical behavior data and the current risk assessment result, the dynamic training module generates customized training tasks through a reinforcement learning model and executes the tasks in the VR / AR risk simulation environment integrated in the virtual simulation scenario library unit. The adaptive difficulty adjustment unit dynamically adjusts the task difficulty according to the user's real-time performance; Step 4: The real-time feedback and evaluation module provides feedback through the multi-modal feedback unit, calculates the behavior compliance score and the risk perception index using the quantitative evaluation unit, and the training effect tracking unit establishes a user behavior profile and compares the incidence of risk behaviors before and after training.

[0036] In this embodiment, specifically, as Figure 3 shown: The method for generating the current risk assessment result includes the following steps: Step 201: The spatio-temporal trajectory modeling unit extracts the time-dependent features of the behavior sequence through an LSTM network, maps the movement trajectory to a three-dimensional space coordinate system in combination with the spatial grid division algorithm, and identifies abnormal behavior patterns by setting a dynamic threshold; Step 202: The multi-modal feature fusion unit uses a Transformer architecture to perform cross-modal feature fusion on the behavioral characteristics, physiological data, and environmental data, and assigns dynamic weights to different modal data through an attention mechanism to generate a comprehensive risk index; Step 203: The risk prediction unit uses the Prophet model, introduces spatio-temporal periodic parameters, and predicts the risk trend in the next 5-10 seconds; Step 204: Combine the abnormal behavior pattern, the comprehensive risk index, and the risk trend in the next 5-10 seconds to comprehensively evaluate the current safety risk status and generate the current risk assessment result.

[0037] In this embodiment, specifically: When executing the task in the VR / AR risk simulation environment integrated in the virtual simulation scenario library unit in Step 3, if an error occurs, the system immediately prompts the operation error and generates a targeted training plan; at the same time, the system uses the reinforcement learning model to generate a targeted training plan according to the error type and the user's historical data to help the user make up for the lack of knowledge and skills.

[0038] In summary, a safety risk perception training method based on behavioral trajectory analysis provided by an embodiment of the present invention first uses a multi-source data acquisition module to collect the physiological data, behavioral characteristics, and environmental data of a user in real time; then transmits the data to a behavioral trajectory analysis module, and generates a current risk assessment result through spatio-temporal trajectory modeling, multi-modal feature fusion, and risk prediction; then, according to the user's historical behavior data and the current assessment result, a dynamic training module generates a customized training task by means of a reinforcement learning model, executes it in a VR / AR simulation environment, adjusts the difficulty according to the user's real-time performance, and generates a targeted training plan when an error occurs; finally, a real-time feedback and evaluation module enables the user to understand the training situation in a timely manner through multi-modal feedback, quantitative evaluation, and training effect tracking; this method breaks through the limitations of traditional safety training, solves problems such as one-sided and lagging data, lack of personalization in training, and single feedback, improves the user's risk perception and response ability, enhances the practicality of training, and effectively reduces safety risks in various scenarios.

[0039] It should be noted that the embodiments in this specification are all described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and reference can be made to the relevant parts of the method embodiments for the relevant parts.

[0040] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A security risk perception training system based on behavioral trajectory analysis, characterized in that, It includes a multi-source data acquisition module, a behavior trajectory analysis module, a dynamic training module, and a real-time feedback and evaluation module; The multi-source data acquisition module includes a sensor network unit, a visual perception unit, and an environmental data unit; the sensor network unit is used to collect the user's physiological data in real time, the visual perception unit is used to identify the user's behavior characteristics in real time, and the environmental data unit is used to collect environmental data in real time; The behavior trajectory analysis module includes a spatio-temporal trajectory modeling unit, a multi-modal feature fusion unit, and a risk prediction unit; the spatio-temporal trajectory modeling unit extracts the time-dependent features of the behavior sequence through an LSTM network, combines the spatial grid division algorithm to map the movement trajectory to a three-dimensional space coordinate system, and identifies abnormal behavior patterns; the multi-modal feature fusion unit uses the Transformer architecture to fuse behavior features, physiological data, and environmental data to generate a comprehensive risk index; the risk prediction unit predicts the risk trend in the next 5-10 seconds through a time series autoregressive model; The dynamic training module includes a personalized training generation unit, a virtual simulation scenario library unit, and an adaptive difficulty adjustment unit; The personalized training generation unit generates customized training tasks through a reinforcement learning model based on the user's historical behavior data and the current risk assessment results; The virtual simulation scenario library unit integrates VR / AR technology to build multi-type risk simulation environments; the adaptive difficulty adjustment unit dynamically adjusts the task difficulty according to the user's real-time performance; The real-time feedback and evaluation module includes a multi-modal feedback unit, a quantitative evaluation unit, and a training effect tracking unit; The multi-modal feedback unit is used to provide risk prompts through visual feedback, tactile feedback, and auditory feedback; the quantitative evaluation unit includes behavior compliance scoring and risk awareness index calculation; the training effect tracking unit is used to establish a user behavior profile and compare the incidence of risk behaviors before and after training.

2. The safety risk perception training system based on behavioral trajectory analysis according to claim 1, characterized in that The sensor network unit includes a UWB positioning base station and a wearable device, and the wearable device is a smart bracelet; the visual perception unit uses a multi-camera array combined with an AI vision algorithm to identify the user's behavior characteristics, and the AI vision algorithm is YOLOv8; The physiological data includes the user's heart rate, blood pressure, blood oxygen saturation, movement speed, and movement acceleration; The behavior characteristics include the user's gesture actions, body postures, facial micro-expressions, walking paths, and operation processes; The environmental data includes temperature and humidity, light intensity, noise level, harmful gas concentration, and air pressure.

3. The safety risk perception training system based on behavioral trajectory analysis according to claim 1, characterized in that, The time series autoregressive model is the Prophet model, and the Prophet model is optimized for safety risk scenarios. By introducing spatio-temporal periodic parameters of behavior trajectory data, the accuracy of risk trend prediction is improved; The reinforcement learning model is the PPO model, and the reward function of the PPO model includes multi-dimensional evaluation indicators, and the multi-dimensional evaluation indicators include behavior compliance rewards, risk response rewards, and physiological stress rewards.

4. A safety risk perception training system based on behavioral trajectory analysis according to claim 1, characterized in that, The visual feedback provides risk warnings by displaying hot spots of behavior trajectories through a heat map, the tactile feedback provides risk warnings by vibrating a bracelet to prompt risky behaviors, and the auditory feedback provides risk warnings by voice broadcasting risk levels and improvement suggestions.

5. The safety risk perception training system based on behavior trajectory analysis according to claim 1, characterized in that, The behavioral compliance score is calculated in real time based on preset safety rules; the risk perception index is combined with historical training data to generate a personal risk perception radar chart including emergency response speed, risk identification accuracy, operational compliance, environmental adaptability and teamwork ability.

6. The safety risk perception training system based on behavioral trajectory analysis according to claim 1, wherein The virtual simulation scenario library unit supports various types of risk simulation environments, and dynamically generates risk factors matching real scenarios through real-time access to the environmental data interface of the environmental data unit.

7. A safety risk perception training system based on behavioral trajectory analysis according to claim 1, characterized in that, The system also includes an interactive collaboration module and a data storage and management module; The interactive collaboration module is used for interactive collaboration training among multiple users. It builds team collaboration training tasks through virtual simulation scenarios, and simultaneously collects and analyzes the collaborative compliance of multi-person behavior trajectories. The data storage and management module is used to store data generated by system operation, and has data backup, recovery, cleaning and access permission control functions.

8. A safety risk perception training method based on behavior trajectory analysis, applied to a safety risk perception training system based on behavior trajectory analysis according to any one of claims 1-7, characterized in that, The following steps are involved: Use multi-source data acquisition modules to collect users' physiological data, behavioral characteristics and environmental data in real time; The collected physiological data, behavioral characteristics and environmental data are transmitted to the behavioral trajectory analysis module for analysis and processing, and the current risk assessment results are generated; Based on the user's historical behavior data and current risk assessment results, the dynamic training module generates customized training tasks through the reinforcement learning model and executes the tasks in the VR / AR risk simulation environment integrated by the virtual simulation scene library unit. The adaptive difficulty adjustment unit dynamically adjusts the task difficulty according to the user's real-time performance; The real-time feedback and evaluation module provides feedback through a multimodal feedback unit, uses a quantitative evaluation unit to calculate behavioral compliance scores and risk perception indexes, and the training effect tracking unit establishes user behavior profiles and compares the incidence of risky behaviors before and after training.

9. The safety risk perception training method based on behavior trajectory analysis according to claim 8, wherein The method for generating the current risk assessment result comprises the following steps: The spatiotemporal trajectory modeling unit extracts the time-dependent features of the behavior sequence through the LSTM network, maps the motion trajectory to a three-dimensional space coordinate system in combination with the spatial grid partitioning algorithm, and identifies abnormal behavior patterns by setting dynamic thresholds; The multimodal feature fusion unit adopts the Transformer architecture to perform cross-modal feature fusion on behavioral features, physiological data, and environmental data, and assigns dynamic weights to different modal data through the attention mechanism to generate a comprehensive risk index; The risk prediction unit uses the Prophet model and introduces spatiotemporal periodic parameters to predict the risk trend in the next 5-10 seconds; Combined with abnormal behavior patterns, comprehensive risk index and risk trends in the next 5-10 seconds, a comprehensive assessment of the current security risk situation is conducted to generate the current risk assessment results.

10. A safety risk perception training method based on behavioral trajectory analysis according to claim 8, characterized in that, When performing tasks in the VR / AR risk simulation environment integrated with the virtual simulation scene library unit, if an error occurs, the system immediately prompts the operation error and generates a targeted training plan.

Images