Method and system for generating electronic fence in dangerous area of chemical plant

Abstract

The invention relates to a chemical plant dangerous area electronic fence generation method and system, and the method comprises the steps: obtaining the positioning data of a target person, the positioning data of production equipment, and the video monitoring data and environment sensing data of a target area, carrying out the recognition based on the video monitoring data, and obtaining the behavior data of the target person, the method comprises the steps of dividing a target area into virtual grids based on a dynamic grid algorithm, determining distribution of geographic features based on GIS geographic information of the target area, filtering impassable areas based on the distribution of the geographic features, aggregating and combining adjacent grids, generating a closed boundary, and establishing a risk level evaluation model to evaluate the risk level of each virtual grid. The boundary adjusting module is used for adjusting the minimum buffer area width of the closed boundary to obtain a final electronic fence boundary, and the electronic fence boundary is used for carrying out risk early warning. According to the invention, dynamic and accurate sensing and protection of the high-risk area in the chemical plant are realized.

Description

Technical Field

[0001] The present application relates to the field of safe production technology, and in particular to a method for generating an electronic fence in a dangerous area of ​​a chemical plant. Background Art

[0002] In chemical production sites, electronic fences are crucial for restricted area management and early warning systems. Chemical areas often involve high temperatures, high pressures, and toxic or flammable substances. Electronic fences monitor personnel entry and exit in real time, preventing them from entering hazardous areas and providing timely responses through audible and visual alarms or system linkage. Furthermore, electronic fences record violations for easy traceability, complying with safety regulations and significantly reducing accident risks. Compared to traditional physical fences, electronic fences offer greater flexibility and can dynamically adjust restricted areas to accommodate temporary operational needs, improving management efficiency.

[0003] Existing technologies primarily include fixed infrared beams, RFID recognition, and video surveillance linkage. Fixed electronic fences rely on pre-set geographic boundaries and cannot adapt to the expansion of the danger zone due to leak spread or weather changes. RFID requires wearing tags, which carries the risk of missed detections. Video surveillance is affected by sunlight, rain, and fog, resulting in a high false alarm rate. These technologies often operate in isolation, lack data integration capabilities, and are unable to cope with the dynamic complexity of the chemical environment. For example, gas sensors only monitor local concentrations and are not linked to the fence system, resulting in delayed warnings or missed alerts. Summary of the Invention

[0004] The embodiments of the present application provide a method and system for generating electronic fences in hazardous areas of chemical plants, aiming to solve problems such as slow response, high false alarm rate, and inability to adapt to complex environmental changes in traditional electronic fence systems.

[0005] In a first aspect, an embodiment of the present application provides a method for generating an electronic fence in a hazardous area of ​​a chemical plant, comprising: A multi-source data acquisition step, obtaining the target personnel's location data, the production equipment's location data, the target area's video surveillance data, and environmental sensor data, identifying the target personnel's behavior data based on the video surveillance data, and fusing them to obtain a multi-source data set; The electronic fence construction step divides the target area into virtual grids based on a dynamic grid algorithm, determines the distribution of geographical features based on the GIS geographic information of the target area, filters impassable areas based on the distribution of geographical features, aggregates and merges adjacent grids to generate a closed boundary, and the closed boundary is configured with a minimum buffer width; The electronic fence optimization step combines historical hidden danger data with the positioning data of production equipment and environmental sensor data in the multi-source data set for correlation analysis, evaluates the risk level of each virtual grid based on a risk level assessment model, adjusts the minimum buffer zone width based on the risk level, and uses the adjusted closed boundary as the electronic fence boundary, which is used for risk warning.

[0006] Based on the above steps, in the embodiment of the present application, during the stage of constructing the electronic fence, the target area space is divided into several virtual grids using a dynamic grid algorithm, and then the inaccessible geographical features in the area, such as buildings, pipeline areas, etc., are extracted through GIS information, and these areas are subjected to grid filtering. The remaining adjacent grids are aggregated and merged to generate a closed boundary, and a minimum buffer width with protective significance is set to restrict people from approaching the source of danger. In addition, the Alpha-Shape algorithm can be used to further extract the concave boundaries of complex areas to improve the realism of the boundaries. In the optimization stage, historical hidden danger information is introduced for correlation analysis with current environmental data, and the risk level of different grids is scored using a risk level assessment model. The higher the risk level, the larger the corresponding buffer width. The location and behavior of the target person are combined to determine whether the electronic fence is triggered, and a response is made through the emergency system after the triggering.

[0007] This method can achieve dynamic and precise perception and protection of high-risk areas in chemical plants, effectively integrate multi-source perception data to improve the reliability and accuracy of fence generation; the boundaries generated by dynamic grids and Alpha-Shape algorithms are more consistent with actual terrain and personnel flow characteristics; the buffer zone width is adjusted according to the risk level and the fence boundary is adjusted in time when the hazard source is detected, thereby improving the timeliness of hazard prevention and emergency response; the system has high visualization and automatic early warning capabilities, providing intelligent tools for enterprise safety management.

[0008] In some of the embodiments, a fence update strategy model is trained and used to output the boundary adjustment action at the next moment based on the current fence boundary, so as to improve the adaptability and intelligence of fence decisions, realize continuous online optimization and updating of strategies, and adapt to the dynamic changes of complex industrial environments.

[0009] In some embodiments, the training of the fence update strategy model further comprises: Define the reinforcement learning environment as a Markov decision process, expressed as: ; use The random strategy randomly selects actions and uses the reward mechanism to update the strategy; in, is the state space, the current state Including: the current electronic fence boundary, the risk level of each virtual grid, the behavior status of the target personnel, historical false alarm and missed alarm records, the current electronic fence boundary includes the boundary point coordinates, grid area or grid matrix code, is the action space, including: expanding the boundary, shrinking the boundary or keeping it unchanged. The actions of expanding the boundary and shrinking the boundary also include the corresponding action distance. is the state transfer function, modeled by environmental simulation, It is used to express the impact of the current action on the behavior of the personnel and the state of the environment. Due to the complexity of the environment, the transfer function is usually estimated by empirical sampling. is the reward function, is a discount factor used to indicate the importance of future rewards, with a value of 0.95.

[0010] The reward mechanism is set up as follows: if a violation is successfully intercepted, a value of 10 is assigned; if it is a false alarm, a value of -3 is assigned; if a response is delayed, a value of -5 is assigned; the response delay is determined based on a response threshold; if it is a missed alarm, a value of -10 is assigned. A false alarm is a situation where an alert is triggered despite not being a real risk area, and a missed alarm is a situation where an alert is not triggered despite being a real risk area.

[0011] In the initial stages, the geofence uses a random strategy to select actions, exploring a wide range of fence settings and rule combinations to collect more state-action-reward data. As the data collected continues to grow, the geofence leverages its learned experience to select the action that maximizes the expected reward based on the current state. During actual operation, the environment and the behavior patterns of the target object may change. Therefore, the geofence needs to update and adjust its strategy regularly or in real time, continuously conducting trial and error and learning.

[0012] This implementation establishes an intelligent training environment by introducing a Markov decision process, using reinforcement learning to train the fence policy model. The system updates its policy based on the current state and feedback from action selection, using a defined reward function R. This reward mechanism aims to maximize the accuracy and responsiveness of the fence system, adjusting its policy direction at each step based on environmental feedback. This strategy ensures a certain degree of exploration during the model training phase, avoiding regression into local optima. A discount factor is used to enhance the accuracy of future warnings and improve overall risk prevention and control capabilities. Ultimately, the model is able to adaptively make boundary decisions in real-world applications, improving the system's responsiveness and judgment in dynamic and complex environments. The introduction of reinforcement learning and Markov decision models imbues the electronic fence strategy with intelligent decision-making capabilities. Compared to static or rule-driven approaches, this approach can continuously self-optimize and adapt to changing risk characteristics in different environments, effectively reducing false positives and missed negatives, and improving system accuracy and robustness.

[0013] In some embodiments, the environmental risk factors are calculated based on the environmental sensor data, and the equipment risk factors are calculated based on the positioning data of the production equipment and the number of historical hidden dangers, and the risk level assessment model is established based on the environmental risk factors and the equipment risk factors.

[0014] In some embodiments, the environmental sensor data includes: wind speed, temperature, humidity, concentration of dangerous gases, Based on a safety threshold and a danger threshold, the hazard of the dangerous gas concentration is linearly scored to obtain a concentration risk score; Based on the impact of wind speed on the diffusion of hazardous gases, interval scoring is performed to obtain wind speed risk scores, which reflect the impact of diffusion capacity on the risk range; The temperature and humidity risk scores are obtained by assigning points based on the impact of temperature and humidity on the volatility, corrosivity, and sensor performance of hazardous gases. The concentration risk score, wind speed risk score, and temperature and humidity risk score are weighted and calculated to obtain the total score of environmental risk factors.

[0015] In some embodiments, historical hidden danger data of production equipment is obtained, a space-time cube model of the hidden danger data is constructed, and the space coordinates (including longitude and latitude) are used to mark the hidden danger data. i ,latitude j , floor height k ), time, risk type, and risk frequency are encoded to construct a three-dimensional hidden danger heat map. The value of the equipment hazard factor is determined based on the three-dimensional hidden danger heat map. The heat map shows the degree of concentration of hidden dangers in different areas and time periods through color intensity or numerical identification.

[0016] By introducing the space-time cube model and three-dimensional hidden danger heat map, we have achieved structured modeling and trend visualization of historical hidden danger data, making the electronic fence system more precise and dynamic in equipment risk identification and prediction. It also enables the dynamic expansion of safety distances. For example, if gas leaks frequently occur near a certain device during a certain period in winter, the system will mark the area as seasonally high-risk and automatically expand the warning area during the same period of the following year to improve the sensitivity of the fence.

[0017] In some embodiments, obtaining the target person's behavior data based on the video surveillance data further includes: Deploy a target detection model to perform real-time recognition of video surveillance data, extract the target person's location frame and posture information, and extract the target person's behavior status, which includes: standing, running, and staying.

[0018] In some embodiments, the geographic features include: buildings, obstacles, buildings, and obstacle areas configured as inaccessible areas. Buildings typically include entities such as production workshops, warehouses, and office buildings, and their physical boundaries are converted into inaccessible surface polygons; obstacles such as large equipment foundations, walls, and pools.

[0019] In some embodiments, a multi-level risk warning mechanism is configured to provide risk warnings, and the multi-level warning mechanism includes: If the target person approaches the first warning distance of the electronic fence boundary, a level 1 warning is triggered and an alert message is sent to an APP platform. For example, but not limited to, the first warning distance is 5 meters. The alert message includes the person's ID, location, behavior status and boundary information; If the target personnel gather or stay for longer than the first warning time, the second warning will be triggered, and the on-site sound and light alarm device will be activated in the target area to remind people to evacuate immediately. The on-site warning will enhance the effect of physical intervention; If the target personnel gather or stay for longer than the first warning time, and an accident occurs in the area where the target personnel are located (such as gas leakage, abnormal high temperature, pressure fluctuation, etc.), a third-level warning will be triggered, and emergency warning information will be pushed to the platform management system, emergency response system and responsible persons. Emergency response operations such as video linkage, broadcast evacuation, and closure of regional channels can be linked.

[0020] In a second aspect, an embodiment of the present application provides a system for generating an electronic fence in a hazardous area of ​​a chemical plant, comprising: Data acquisition equipment, including positioning modules, video acquisition modules, and environmental sensors deployed in the target area; The server device is used to communicate with the data acquisition device and execute the method for generating an electronic fence for a dangerous area in a chemical plant as described in the first aspect above.

[0021] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 is a flow chart of a method for generating an electronic fence in a dangerous area of ​​a chemical plant according to an embodiment of the present application; Figure 2 This is a structural block diagram of a system for generating electronic fences in dangerous areas of a chemical plant according to an embodiment of the present application; Figure 3 It is a structural diagram of a server device according to an embodiment of the present application.

[0023] In the figure: 10, data acquisition device; 20, server device; 21, processor; 22, memory. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely used to explain this application and are not intended to limit this application. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making any creative efforts are within the scope of protection of this application.

[0025] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0026] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0027] Unless otherwise defined, technical or scientific terms used herein shall have the ordinary meaning as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "an," "the," and similar expressions used herein do not denote quantitative limitations and may refer to either the singular or the plural. The terms "comprise," "include," "have," and any variations thereof, used herein, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or modules (units) is not limited to the listed steps or units but may also include steps or units not listed, or may include other steps or units inherent to the process, method, product, or apparatus. The terms "connected," "connected," "coupled," and similar expressions used herein are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used herein, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" may mean: A exists alone; A and B exist simultaneously; or B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0028] This embodiment provides a method for generating an electronic fence in a hazardous area of ​​a chemical plant. Figure 1 Flowchart of the method for generating electronic fences in dangerous areas of a chemical plant according to an embodiment of the present application. Figure 1 As shown, the process includes the following steps: Multi-source data acquisition step S101, acquiring the target person's location data, the production equipment's location data, the target area's video surveillance data, and environmental sensor data, identifying the target person's behavior data based on the video surveillance data, and fusing them to obtain a multi-source data set; The electronic fence construction step S102 divides the target area into virtual grids based on a dynamic grid algorithm, determines the distribution of geographical features based on the GIS geographic information of the target area, filters impassable areas based on the distribution of geographical features, aggregates and merges adjacent grids to generate a closed boundary, wherein the closed boundary is configured with a minimum buffer width. Optionally, the Alpha-Shape algorithm is used to extract concave boundaries on the polygons obtained by aggregation and merging. The electronic fence optimization step S103 combines historical hidden danger data with the positioning data of production equipment and environmental sensor data in the multi-source data set for correlation analysis, evaluates the risk level of each virtual grid based on a risk level assessment model, adjusts the minimum buffer zone width based on the risk level, and uses the adjusted closed boundary as the electronic fence boundary. The electronic fence boundary is used for risk warning, so that in areas with flammable and explosive materials, the areas that may be affected by danger are demarcated with irregular polygonal boundaries in combination with the surrounding building distribution, personnel traffic conditions and gas concentration distribution.

[0029] Based on the above steps, in the embodiment of the present application, during the stage of constructing the electronic fence, the target area space is divided into several virtual grids using a dynamic grid algorithm, and then the inaccessible geographical features in the area, such as buildings, pipeline areas, etc., are extracted through GIS information, and these areas are subjected to grid filtering. The remaining adjacent grids are aggregated and merged to generate a closed boundary, and a minimum buffer width with protective significance is set to restrict people from approaching the source of danger. In addition, the Alpha-Shape algorithm can be used to further extract the concave boundaries of complex areas to improve the realism of the boundaries. In the optimization stage, historical hidden danger information is introduced for correlation analysis with current environmental data, and the risk level of different grids is scored using a risk level assessment model. The higher the risk level, the larger the corresponding buffer width. The location and behavior of the target person are combined to determine whether the electronic fence is triggered, and a response is made through the emergency system after the triggering.

[0030] This method can achieve dynamic and precise perception and protection of high-risk areas in chemical plants, effectively integrate multi-source perception data to improve the reliability and accuracy of fence generation; the boundaries generated by dynamic grids and Alpha-Shape algorithms are more consistent with actual terrain and personnel flow characteristics; the buffer zone width is adjusted according to the risk level and the fence boundary is adjusted in time when the hazard source is detected, thereby improving the timeliness of hazard prevention and emergency response; the system has high visualization and automatic early warning capabilities, providing intelligent tools for enterprise safety management.

[0031] In the above embodiments, the positioning data of the target personnel and production equipment can be achieved based on Beidou technology, GPS technology, UWB technology, and Bluetooth positioning technology. Beidou technology and GPS technology are preferentially suitable for outdoor scenes, and UWB technology and Bluetooth technology collaborative positioning are suitable for indoor scenes or scenes with shielded areas.

[0032] In another embodiment, the historical electronic fence configuration of the target area, or the electronic fence configuration requirements input by the administrator, is obtained, and the closed boundary is adjusted based on the historical electronic fence configuration or the electronic fence configuration requirements. For example, when there is a record of frequent alarms in a certain area in the historical electronic fence configuration, the system can automatically expand the closed boundary buffer of that area, or set a larger safety buffer for a specific area during certain time periods, or set a high-sensitivity fence during specific operations. When the administrator sets a temporary pass area during a specific operation, the boundary shape can also be adjusted based on their input to temporarily relax the fence restrictions in some areas. In this embodiment, historical data and manual intervention mechanisms are introduced to optimize the boundary settings of the electronic fence, giving the system the ability to learn and dynamically adjust, further enhancing the rationality and on-site adaptability of the boundary settings, improving its controllability and compliance with actual operational needs, effectively preventing problems such as false alarms and missed alarms, and improving the overall reliability and operational convenience of the system.

[0033] In the above embodiment, historical electronic fence configurations can be derived from a variety of information systems, such as MES systems, security event databases, and GIS platforms. The administrator input interface can be implemented using a graphical visual interactive interface, voice command recognition, text input, and other methods. The spatial scale of boundary adjustments can be flexibly configured based on set policies, such as allowing the boundary to expand or contract by 1 to 10 meters. The specific value can be set based on historical trigger frequency or management level.

[0034] In another embodiment, the multi-source data acquisition step further includes: cleaning and standardizing the collected data, fusing and aligning the data through timestamps and spatial positions, for example, matching the positioning information of the personnel with the environmental parameter data at the same time, and analyzing the safety status of the environment in which the personnel are located.

[0035] In some of the embodiments, a fence update strategy model is trained and used to output the boundary adjustment action at the next moment based on the current fence boundary, so as to improve the adaptability and intelligence of fence decisions, realize continuous online optimization and updating of strategies, and adapt to the dynamic changes of complex industrial environments.

[0036] In some embodiments, the training of the fence update strategy model further comprises: Define the reinforcement learning environment as a Markov decision process, expressed as: ; use The random strategy randomly selects actions and uses the reward mechanism to update the strategy; in, is the state space, the current state Including: the current electronic fence boundary, the risk level of each virtual grid, the behavior status of the target personnel, historical false alarm and missed alarm records, the current electronic fence boundary includes the boundary point coordinates, grid area or grid matrix code, is the action space, including: expanding the boundary, shrinking the boundary or keeping it unchanged. The actions of expanding the boundary and shrinking the boundary also include the corresponding action distance. is the state transfer function, modeled by environmental simulation, It is used to express the impact of the current action on the behavior of the personnel and the state of the environment. Due to the complexity of the environment, the transfer function is usually estimated by empirical sampling. is the reward function, is a discount factor used to indicate the importance of future rewards, with a value of 0.95.

[0037] The reward mechanism is set up as follows: if a violation is successfully intercepted, a value of 10 is assigned; if it is a false alarm, a value of -3 is assigned; if a response is delayed, a value of -5 is assigned; the response delay is determined based on a response threshold; if it is a missed alarm, a value of -10 is assigned. A false alarm is a situation where an alert is triggered despite not being a real risk area, and a missed alarm is a situation where an alert is not triggered despite being a real risk area.

[0038] In the initial stages, the geofence uses a random strategy to select actions, exploring a wide range of fence settings and rule combinations to collect more state-action-reward data. As the data collected continues to grow, the geofence leverages its learned experience to select the action that maximizes the expected reward based on the current state. During actual operation, the environment and the behavior patterns of the target object may change. Therefore, the geofence needs to update and adjust its strategy regularly or in real time, continuously conducting trial and error and learning.

[0039] This implementation establishes an intelligent training environment by introducing a Markov decision process, using reinforcement learning to train the fence policy model. The system updates its policy based on the current state and feedback from action selection, using a defined reward function R. This reward mechanism aims to maximize the accuracy and responsiveness of the fence system, adjusting its policy direction at each step based on environmental feedback. This strategy ensures a certain degree of exploration during the model training phase, avoiding regression into local optima. A discount factor is used to enhance the accuracy of future warnings and improve overall risk prevention and control capabilities. Ultimately, the model is able to adaptively make boundary decisions in real-world applications, improving the system's responsiveness and judgment in dynamic and complex environments. The introduction of reinforcement learning and Markov decision models imbues the electronic fence strategy with intelligent decision-making capabilities. Compared to static or rule-driven approaches, this approach can continuously self-optimize and adapt to changing risk characteristics in different environments, effectively reducing false positives and missed negatives, and improving system accuracy and robustness.

[0040] Furthermore, the reward function can also be set as: When the electronic fence successfully prevents the target object from illegally entering the dangerous area, a higher reward will be given. The reward value can be adjusted according to the level of danger. The higher the danger level, the greater the reward. If the fence can automatically adjust its boundaries and rules based on environmental changes and changes in the behavior of the target object, and the adjustment is effective, such as reducing false alarms and improving control efficiency, a certain reward will be given; When the target object violates the rules and enters the dangerous area and the fence fails to issue a timely warning, a larger penalty will be imposed; the degree of penalty is related to the importance of the dangerous area and the severity of the violation; If the fence frequently issues false alarms, a certain penalty will be imposed; the severity of the penalty can be determined based on the number of false alarms and the scope of impact.

[0041] When the environment changes significantly or the target object exhibits abnormal behavior, if the fence fails to adjust and respond in time, corresponding penalties will be imposed.

[0042] This implementation enhances the reinforcement learning model's adaptability to complex industrial scenarios by further refining the reward function. During fence policy training, the system not only applies simple rewards and penalties based on successful interception of the target object, but also incorporates multi-dimensional evaluation metrics closely related to hazard level, behavioral response efficiency, and environmental adaptability. For successful interception of intrusions in high-risk areas, a reward value is dynamically calculated based on the area's preset hazard level, reflecting its risk control value. Furthermore, if the fence system proactively adapts to environmental changes and changes in human behavior patterns, significantly reducing false alarms and improving control effectiveness, it is considered an excellent strategy update and rewarded to encourage continued evolution. Conversely, if missed alerts occur, particularly if timely warnings are not issued in high-risk areas, penalties are assigned based on the severity of the violation and the importance of the area. Frequent false alarms and failure to respond promptly to environmental changes also constitute negative feedback. Through the dynamic feedback of this refined reward mechanism, the value of various response behaviors can be more accurately evaluated and learned during policy optimization, resulting in a more stable and reliable fence policy.

[0043] In this embodiment, the reward function can be designed in a hierarchical manner: sub-functions are set for different types of rewards and penalties to perform sub-item evaluations, such as a successful interception reward function and a false alarm penalty function, and finally a weighted total reward is synthesized. The danger level can be generated by the risk level assessment model of this application, the expert knowledge base, historical accident data derivation, or automatic classification by a deep learning model. The false alarm judgment criteria can also be flexibly adjusted according to the tolerance threshold set by the user, and its severity can be assessed based on the duration, coverage area, and interference impact of the false alarm.

[0044] In some embodiments, the environmental risk factors are calculated based on the environmental sensor data, and the equipment risk factors are calculated based on the positioning data of the production equipment and the number of historical hidden dangers. The risk level assessment model is established based on the environmental risk factors and the equipment risk factors. The environmental sensor data includes: wind speed, temperature, humidity, and hazardous gas concentration; calculating the environmental risk factors based on the environmental sensor data further includes: Step S201: Linearly assigning a score to the hazardous gas concentration based on a safety threshold and a danger threshold to obtain a concentration risk score. For example, but not limited to, if the safety threshold concentration is set to TLV (safe exposure threshold, Threshold Limit Value) 10% LEL (Lower Explosive Limit of flammable gas / vapor), the danger threshold is IDLH or 50% LEL. Optionally, corresponding safety thresholds and danger thresholds can be flexibly configured for different gases, and a threshold adaptation module based on the gas database is set, which is not specifically limited here. Step S202: Based on the wind speed, the impact of the diffusion of hazardous gases is scored in intervals to obtain a wind speed risk score, which reflects the impact of diffusion capacity on the risk range. For example, but not limited to, if the wind speed is 1-3 m / s, it is considered to achieve ideal diffusion, and the wind speed risk score is 0 points, achieving the best diffusion effect. If the wind speed is 0.5-1 m / s, it is considered to be at low risk and is assigned 30 points. If the wind speed is less than 0.5 m / s, it is considered that there is gas accumulation and is at medium risk, and is assigned 70 points. If there is a fire point or a polluted gas leak, if the wind speed is 3-5 m / s, it is considered to be at low risk and is also assigned 30 points. If the wind speed is 5-10 m / s, it is considered that the pollution spreads too quickly and is at medium risk, and is also assigned 70 points. If the wind speed is greater than 10 m / s, it is considered to expand the pollution / explosion risk and is at high risk and is assigned 100 points.

[0045] Step S203: Based on the impact of temperature and humidity on the volatility, corrosivity, and sensor performance of hazardous gases, scores are assigned to obtain a temperature and humidity risk score. For example, but not limited to, when the temperature is 10°C to 30°C or the humidity is 40% to 70%, a score of 0 is assigned; when the temperature is 30°C to 35°C or the humidity is 70% to 80%, a score of 40 is assigned; when the temperature is greater than 35°C (accelerated reaction), the humidity is less than 30% (static risk), or the humidity is greater than 80% (corrosion / false alarm), a score of 80 is assigned; Step S204: performing weighted calculation on the concentration risk score, wind speed risk score, and temperature and humidity risk score to obtain a total score of the environmental risk factors.

[0046] Based on the above steps, the calculation of environmental risk factors relies on multi-dimensional sensor data analysis. Through refined interval division and assignment strategies, the impact of environmental parameters on safety risks is fully explored, generating more sensitive and controllable hazard level assessment results. In particular, the responsiveness to scenarios such as gas leakage, diffusion, and accumulation is significantly enhanced, making the electronic fence system more adaptable to environmental conditions and proactively predictive. Furthermore, the modular design of the assignment rules provides flexible configuration space for actual deployment, improving the system's versatility and deployment efficiency.

[0047] In another embodiment, the wind speed assignment strategy can be regionalized based on factors such as plant building distribution and ventilation structure. Specifically, based on a 2D CAD or 3D GIS model of the chemical plant, the plant is divided into several functional areas (such as storage tank area, production area, office area, and unloading area). Real-time wind speed data is collected at different locations and combined with historical wind speed distribution to form a wind field distribution map. The impact of the plant building layout (such as density, height, and orientation) and obstacles on natural ventilation is analyzed to establish a local ventilation resistance model. CFD (computational fluid dynamics) simulation or empirical formulas are used to calculate the "effective wind speed factor" for each area. For example, in a high-density equipment area, the actual wind speed impact can be reduced to 60% of the original value. Therefore, if the original wind speed is 1.5 m / s, K = 1.0 in open areas, and the assignment is calculated according to the standard process; however, in poorly ventilated areas, K = 0.6, the adjusted wind speed is 0.9 m / s, and the corresponding assignment is 30 points instead of 0 points.

[0048] In another embodiment, the impact of temperature and humidity can also be considered by combining meteorological forecast data and real-time multi-point monitoring data to construct a multi-source perception fusion model. Specifically, spatial interpolation (e.g., Kriging method) is performed on the collected sensor data to form a continuous distribution map to fill in perception blind spots. At the same time, real-time meteorological forecast data (e.g., obtained through an API interface) is accessed to predict temperature and humidity trends in advance and establish the impact function of temperature and humidity conditions on the volatility, corrosiveness, and static electricity risk of hazardous gases. For example, a 10°C temperature increase doubles the volatility rate; when humidity is <30%, the static electricity accumulation rate increases by 50%. Such empirical functions can be obtained from standard databases or literature. A perception fusion model based on Bayesian networks or fuzzy logic is constructed, combining real-time perception data with meteorological forecasts to output a comprehensive temperature and humidity risk score.

[0049] In some embodiments, historical hidden danger data of production equipment is obtained, a space-time cube model of the hidden danger data is constructed, and the space coordinates (including longitude and latitude) are used to mark the hidden danger data. i ,latitude j , floor height k), time, risk type, and risk frequency are encoded to construct a three-dimensional hidden danger heat map. The value of the equipment hazard factor is determined based on the three-dimensional hidden danger heat map. The heat map shows the degree of concentration of hidden dangers in different areas and time periods through color intensity or numerical identification.

[0050] The three-dimensional hidden danger heat map has multi-dimensional analysis capabilities, which can screen out high-frequency hidden danger areas at different spatial levels (such as 1 meter, 5 meters, and 10 meters around the equipment) and time segments (such as spring, winter, and night) and identify voxels. Cumulative density of medium-risk events Areas exceeding the set threshold are identified as "high-risk spaces" and their corresponding equipment risk factor values ​​are further calculated. Specifically, each production equipment is assumed to have a set radius (such as 5 meters, 10 meters) extending outward from its coordinate point to form a risk domain around the equipment, and all voxels within the domain are extracted. The risk factor of the equipment is obtained by weighted average of the risk density value set. Expressed as: , in, is the distance from the voxel to the production equipment The associated decay weight, .

[0051] By introducing the space-time cube model and three-dimensional hidden danger heat map, we have achieved structured modeling and trend visualization of historical hidden danger data, making the electronic fence system more precise and dynamic in equipment risk identification and prediction. It also enables the dynamic expansion of safety distances. For example, if gas leaks frequently occur near a certain device during a certain period in winter, the system will mark the area as seasonally high-risk and automatically expand the warning area during the same period of the following year to improve the sensitivity of the fence.

[0052] In this embodiment, when determining a dangerous area for a certain sub-area, past accident records at that location, time period, and weather conditions can be quickly queried.

[0053] In another embodiment, equipment risk factors may include not only location data and historical hazards but also dimensions such as equipment type, operating hours, and maintenance cycles. Risk assessment models can utilize rule systems based on expert experience or employ neural networks to implement complex nonlinear mapping capabilities to adapt to the dynamic characteristics of diverse work environments.

[0054] Based on the above steps, in this embodiment of the present application, at the environmental level, sensor data is collected to extract safety-related indicator variables, and then an environmental hazard factor representing the degree of environmental risk is calculated. At the equipment level, the equipment hazard factor is generated by combining information such as the spatial location of various types of production equipment within the factory and the number and type of historical hidden dangers. Subsequently, these two factors are introduced as input variables into the risk level assessment model, which can use linear weighting, the Analytic Hierarchy Process (AHP), or machine learning methods (such as decision trees and Support Vector Machines) to calculate the virtual grid risk level. This mechanism effectively integrates spatial distribution information with historical behavioral characteristics, making the risk assessment process more targeted and practical.

[0055] By integrating data from both the environment and the equipment, the ability to perceive regional risk conditions is effectively enhanced. Compared to methods that rely solely on static geographic or personnel distribution information, this approach incorporates historical hidden dangers and real-time sensor signals, giving the electronic fence system the ability to respond dynamically and identify risks. This allows for more scientific guidance on the generation and optimization of fence boundaries, improving the accuracy of early warning mechanisms.

[0056] In some embodiments, obtaining the target person's behavior data based on the video surveillance data further includes: Deploy a target detection model to perform real-time recognition of video surveillance data, extract the target person's location frame and posture information, and extract the target person's behavior status, which includes: standing, running, and staying.

[0057] In the above embodiment, the object detection model is a YOLOv5 model, which uses a convolutional neural network to extract features from the input image, generate location boxes and category labels for all human targets in the frame, and output the spatial coordinates of the target people in real time. Furthermore, a posture recognition module (such as OpenPose or HRNet) is integrated to extract key points (such as the head, shoulders, elbows, knees, and feet) of the detected human targets and generate a pose sequence. In another embodiment, the object detection model can be replaced with other deep object detection frameworks such as YOLOv5, Faster R-CNN, and EfficientDet, depending on the deployment environment and hardware resources.

[0058] The temporal changes in posture sequences are fed into a time series model (such as LSTM, Transformer, or 3D CNN) for behavior recognition. The model can identify behavior categories such as "normal passage," "running," "falling," "wandering," "entering a dangerous area," and "forced valve opening." The behavioral state recognition results are integrated with the target person's location data as input for behavioral risk assessment, providing accurate triggering for the geo-fence warning mechanism.

[0059] By integrating YOLOv5 with posture analysis, the system achieves rapid and accurate identification of the target person's location and behavior from video surveillance data. With its real-time and high accuracy, it can proactively identify abnormal behavior (such as falls, trespassing, and lingering) and trigger early warnings, significantly enhancing the automation and intelligence of personnel safety management in chemical plants. This approach provides critical human dynamic input to the electronic fence system, enhancing its dynamic response capabilities and risk prevention and control accuracy.

[0060] In some embodiments, the geographic features include: buildings, obstacles, buildings, and obstacle areas configured as inaccessible areas. Buildings typically include entities such as production workshops, warehouses, and office buildings, and their physical boundaries are converted into inaccessible surface polygons; obstacles such as large equipment foundations, walls, and pools.

[0061] Furthermore, the geographic features also include motor vehicle lanes, which are typically designated as off-limits to the target personnel. Specifically, these are extracted based on factory traffic planning or road layer data, and logical restriction rules are set to define these areas as off-limits to the target personnel, allowing vehicles to enter but prohibiting people from crossing. When a person's location data indicates they have entered this area, a violation flag is triggered for subsequent risk assessment and early warning processing.

[0062] Based on the above configuration, the embodiments of this application improve the feasibility and decision-making rationality of the electronic fence model by introducing detailed geographic information. Setting buildings and obstacles as impassable areas effectively prevents fence boundaries from mistakenly crossing physical structures, ensuring that the generated results are consistent with the actual terrain. Identifying motor vehicle lanes as restricted areas can improve the accuracy of personnel management, prevent traffic conflicts or safety accidents caused by incorrect paths or management negligence, and enhance the compliance and rigor of the overall safety strategy.

[0063] In another embodiment, the geographic features can be expanded to include "flammable material storage area," "high-voltage equipment area," "automated robot work area," and other areas based on different factory area types. Each area can be marked as "restricted personnel access" or "equipment access only" based on its functional attributes. Building and obstacle boundary extraction can be generated through drone-based scene modeling or automatic conversion of laser scanning point cloud data. Lane recognition can also be updated through methods such as video image recognition and road image segmentation.

[0064] In some embodiments, a multi-level risk warning mechanism is configured to provide risk warnings, and the multi-level warning mechanism includes: If the target person approaches the first warning distance of the electronic fence boundary, a level 1 warning is triggered and an alert message is sent to an APP platform. For example, but not limited to, the first warning distance is 5 meters. The alert message includes the person's ID, location, behavior status and boundary information; If the target personnel gather or stay for longer than the first warning time, the second warning will be triggered, and the on-site sound and light alarm device will be activated in the target area to remind people to evacuate immediately. The on-site warning will enhance the effect of physical intervention; If the target personnel gather or stay for longer than the first warning time, and an accident occurs in the area where the target personnel are located (such as gas leakage, abnormal high temperature, pressure fluctuation, etc.), a third-level warning will be triggered, and emergency warning information will be pushed to the platform management system, emergency response system and responsible persons. Emergency response operations such as video linkage, broadcast evacuation, and closure of regional channels can be linked.

[0065] In the above embodiment, the first warning distance and warning duration parameters can be dynamically configured based on regional risk level, historical behavior patterns, or personnel role permissions. Level 1 warning messages can be published through various platforms, such as WeChat mini-programs, internal enterprise OA, and SMS platforms. The audio and visual alarm system can be configured to be controlled by a programmable controller (PLC), supporting various sound level and frequency combinations.

[0066] The early warning mechanism in this embodiment implements integrated safety management throughout the entire process, from prompting to intervention and emergency response, by setting progressive, multi-level response thresholds. In the early stages, it accurately identifies risk trends and provides remote notifications, improving management efficiency. In the mid-term, it strengthens behavioral control through physical, acoustic, and visual interventions. In the high-risk phase, it links incidents to rapid decision-making and response, significantly reducing risk resolution latency. This mechanism improves the electronic fence's behavioral recognition sensitivity and incident response accuracy, enhancing the intelligent management of complex behaviors such as gatherings, wandering, and stray entry.

[0067] This embodiment also provides a system for generating electronic fences in dangerous areas of chemical plants. Figure 2 As shown, the system is used to implement the above-mentioned embodiments and preferred implementations, and details that have been described will not be repeated. As used below, the terms "module," "unit," "subunit," etc. may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0068] Data acquisition equipment 10, including a positioning module, a video acquisition module, and an environmental sensor deployed in the target area; The server device 20 is used to communicate with the data acquisition device 10 and execute the method for generating an electronic fence for a dangerous area in a chemical plant as described in the above embodiment.

[0069] For example, but not limited to, the positioning module is deployed at UWB positioning base stations at the main entrances and exits of the target factory and in the work areas. Target personnel wear safety helmets or ID cards with Bluetooth tags, and use Beidou or GPS equipment to obtain precise locations. The video acquisition module is composed of video surveillance cameras deployed in various sub-areas of the target factory and around hazardous sources. After acquiring video surveillance data, it uses target detection algorithms to identify personnel locations and behaviors. Wind speed detection sensors, temperature and humidity detection sensors, and hazardous gas concentration detection sensors are deployed in key areas. Data is collected and uploaded once per minute. The hazardous gas concentration detection sensors in this embodiment include, but are not limited to, hydrocarbon gas detection sensors, hydrogen detection sensors, carbon monoxide detection sensors, hydrogen sulfide detection sensors, nitrogen oxide detection sensors, chlorine / ammonia detection sensors, or any combination thereof, thereby enabling the detection of explosive or toxic gases.

[0070] The server device may include a processor 21 and a memory 22 storing computer program instructions.

[0071] Specifically, the processor 21 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0072] Memory 22 may include a large-capacity memory for data or instructions. By way of example, and not limitation, memory 22 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 22 may include removable or non-removable (or fixed) media. Where appropriate, memory 22 may be internal or external to the data processing device. In certain embodiments, memory 22 is non-volatile memory. In certain embodiments, memory 22 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data out dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0073] The memory 22 may be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 21 .

[0074] The processor 21 reads and executes computer program instructions stored in the memory 22 to implement any one of the methods for generating an electronic fence for a hazardous area in a chemical plant in the above embodiments.

[0075] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for generating an electronic fence in a dangerous area of ​​a chemical plant, characterized in that: include: A multi-source data acquisition step, obtaining the target personnel's location data, the production equipment's location data, the target area's video surveillance data, and environmental sensor data, identifying the target personnel's behavior data based on the video surveillance data, and fusing them to obtain a multi-source data set; The electronic fence construction step divides the target area into virtual grids based on a dynamic grid algorithm, determines the distribution of geographical features based on the GIS geographic information of the target area, filters impassable areas based on the distribution of geographical features, aggregates and merges adjacent grids to generate a closed boundary, and the closed boundary is configured with a minimum buffer width; The electronic fence optimization step establishes a risk level assessment model based on the multi-source data set to assess the risk level of each virtual grid, adjusts the minimum buffer width based on the risk level, and uses the adjusted closed boundary as the electronic fence boundary, which is used for risk warning.

2. The method for generating an electronic fence in a dangerous area of ​​a chemical plant according to claim 1, characterized in that: Train the fence update strategy model and use it to output the boundary adjustment action at the next moment based on the current fence boundary.

3. The method for generating an electronic fence in a dangerous area of ​​a chemical plant according to claim 2, characterized in that: The training fence update strategy model further includes: Define the reinforcement learning environment as a Markov decision process, expressed as: ; use The random strategy randomly selects actions and uses the reward mechanism to update the strategy; in, is the state space, is the action space, is the state transition function, is the reward function, is the discount factor.

4. The method for generating an electronic fence for a dangerous area in a chemical plant according to any one of claims 1 to 3, characterized in that: The environmental risk factors are calculated based on the environmental sensor data, and the equipment risk factors are calculated based on the positioning data of the production equipment and the number of historical hidden dangers. The risk level assessment model is established based on the environmental risk factors and the equipment risk factors.

5. The method for generating an electronic fence in a dangerous area of ​​a chemical plant according to claim 4, characterized in that: The environmental sensing data include: wind speed, temperature, humidity, dangerous gas concentration, Based on a safety threshold and a danger threshold, the hazard of the dangerous gas concentration is linearly scored to obtain a concentration risk score; Based on the impact of wind speed on the diffusion of hazardous gases, the wind speed risk score is obtained; The temperature and humidity risk scores are obtained by assigning points based on the impact of temperature and humidity on the volatility, corrosivity, and sensor performance of hazardous gases. The concentration risk score, wind speed risk score, and temperature and humidity risk score are weighted and calculated to obtain the total score of environmental risk factors.

6. The method for generating an electronic fence in a dangerous area of ​​a chemical plant according to claim 5, characterized in that: Obtain historical hidden danger data of production equipment, build a space-time cube model to mark the hidden danger data, encode it by spatial coordinates, time, risk type, and risk frequency, build a three-dimensional hidden danger heat map, and determine the value of the equipment hazard factor based on the three-dimensional hidden danger heat map.

7. The method for generating an electronic fence in a dangerous area of ​​a chemical plant according to claim 6, characterized in that: Obtaining behavioral data of a target person based on the video surveillance data further includes: Deploy the target detection model to perform real-time recognition of video surveillance data, extract the location frame and posture information of the target person, and extract the target person's behavioral status.

8. The method for generating an electronic fence in a dangerous area of ​​a chemical plant according to claim 6, characterized in that: The geographical features include: buildings, obstacles, and areas of buildings and obstacles are configured as inaccessible areas.

9. The method for generating an electronic fence in a dangerous area of ​​a chemical plant according to claim 6, characterized in that: Configure a multi-level risk warning mechanism for risk warning, the multi-level warning mechanism includes: If the target person approaches the first warning distance of the electronic fence boundary, a first-level warning will be triggered; If the target personnel gather or stay for longer than the first warning time, the second warning will be triggered; If the target personnel gather or stay for longer than the first warning time, and an accident occurs in the area where the target personnel are located, a third-level warning will be triggered.

10. A system for generating electronic fences in dangerous areas of chemical plants, characterized by: include: Data acquisition equipment, including positioning modules, video acquisition modules, and environmental sensors deployed in the target area; The server device is used to communicate with the data acquisition device and execute the method for generating an electronic fence for a dangerous area in a chemical plant according to any one of claims 1 to 9.

Images