Distributed monitoring and early warning method for toxic and harmful gases in chemical production environment

Abstract

The invention relates to the technical field of industrial safety monitoring, and discloses a chemical production environment toxic and harmful gas distributed monitoring and early warning method comprising the following steps: S1, building a factory digital twin topological graph; s2, calculating a risk entropy value of each node; s3, dynamically scheduling the sensing nodes; s4, fusing the multi-source heterogeneous sensing data; s5, deducing an edge end risk field in real time; s6, performing graded early warning and cooperative response; and S7, self-learning optimization of the feedback driving system. According to the distributed monitoring and early warning method for the toxic and harmful gases in the chemical production environment, on the basis of digital twin topology modeling (in combination with CFD simulation stationing) and risk entropy dynamic scheduling, on-demand activation or dormancy of sensing nodes is realized, technological process, equipment and meteorological changes are adapted, monitoring blind areas are eliminated, and the monitoring efficiency is improved. And meanwhile, multi-source data are fused through temperature and humidity compensation, space-time alignment and a D-S evidence theory, so that the concentration detection error is reduced, the false alarm rate is reduced, and the data accuracy is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of industrial safety monitoring technology, specifically a distributed monitoring and early warning method for toxic and harmful gases in chemical production environments. Background Technology

[0002] Chemical production processes often produce toxic and harmful gases such as hydrogen sulfide, carbon monoxide, benzene compounds, and volatile organic compounds (VOCs). These gases are characterized by rapid diffusion, high toxicity, and high explosiveness. Once leaked, they can easily cause casualties, equipment damage, and environmental pollution accidents.

[0003] Existing gas monitoring systems generally suffer from the following problems: Static and fixed deployment of monitoring points: Traditional systems rely on fixed sensor arrays, and the deployment plan is usually determined based on the risk assessment in the design phase. It cannot be dynamically adjusted with changes in process flow, equipment aging or seasonal weather conditions, resulting in monitoring blind spots. Data silos and delayed response: Each sensor operates independently, lacking a collaborative sensing mechanism. Abnormal data identification relies on threshold alarms, which cannot distinguish between instantaneous interference and real leakage, resulting in a high false alarm rate. Furthermore, the warning information transmission chain is long, making it difficult to achieve a second-level response. The risk assessment model is static and singular: existing early warning systems mostly adopt the logic of "alarm when concentration exceeds the limit", without combining real-time wind direction, temperature and humidity, equipment status, personnel distribution and other multi-dimensional environmental factors to dynamically quantify the risk, and cannot predict the diffusion path and the scope of impact; Lack of edge intelligence capabilities: A large amount of raw data is uploaded to the central server for processing, which puts a lot of pressure on network bandwidth. The system will fail when the network is down or the server fails, and it lacks local autonomous decision-making capabilities.

[0004] Therefore, a distributed monitoring and early warning method for toxic and harmful gases in the chemical production environment is proposed to solve the problems mentioned above. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a distributed monitoring and early warning method for toxic and hazardous gases in chemical production environments. It has advantages such as risk-driven dynamic node scheduling, high-precision fusion of multi-source heterogeneous sensor data, real-time risk field simulation at the edge, adaptive early warning threshold calibration under operating conditions, hierarchical collaborative response, and continuous self-learning and evolution of the system. It solves the problems of traditional monitoring systems, such as blind spots caused by static and fixed point deployment, false alarms caused by interference of single sensors, fixed thresholds that cannot adapt to complex operating conditions, delayed early warning and lack of diffusion prediction capabilities, and the inability of the system to be continuously optimized with operating experience.

[0006] (II) Technical Solution To achieve the aforementioned objectives of "risk-driven dynamic node scheduling and high-precision fusion of multi-source heterogeneous sensor data," this invention provides the following technical solution: a distributed monitoring and early warning method for toxic and hazardous gases in chemical production environments, comprising the following steps: S1. Construct a digital twin topology map of the plant area, where nodes represent monitoring areas and edges represent gas diffusion channels; S2. Calculate the risk entropy value of each node based on historical leakage data, equipment status, process parameters, and meteorological sensitivity; S3. Dynamically schedule sensor nodes based on risk entropy values: when the risk entropy of a node is higher than the threshold, activate its surrounding redundant nodes and increase the sampling frequency; otherwise, put it into sleep mode. S4. Collect data from multiple heterogeneous sensors, unify them into a preset spatiotemporal grid through a spatiotemporal alignment module, and perform confidence fusion using DS evidence theory to output the fused gas concentration; S5. Deploy a lightweight graph neural network model on the edge gateway, using concentration, wind speed and direction, temperature and equipment status as node attributes, to predict the risk heat map in real time within the next T minutes. S6. Based on the coverage area and concentration gradient of high-risk areas in the risk heat map, trigger a graded early warning and push it to the central control room and on-site personnel terminals; S7. Record the actual handling results of the early warning event and use them as feedback signals to update the risk entropy calculation weights and graph neural network parameters online.

[0007] Preferably, the construction of the digital twin topology map of the plant area in step S1 also includes using computational fluid dynamics (CFD) to simulate the gas diffusion in different areas and determine the initial monitoring node placement locations.

[0008] Preferably, the risk entropy calculation formula in step S2 is: ; in The normalized weight of the Kth type of risk factor (such as equipment aging, process hazard, historical leakage frequency, and meteorological sensitivity coefficient) in region i.

[0009] Preferably, the sensing node in step S3 is equipped with a power module, a data processing unit, and a wireless communication module, supports the LoRa-WiFi dual-mode networking protocol, realizes the self-organizing network function, and when a node fails, the adjacent node automatically takes over.

[0010] Preferably, in step S4, before collecting data from the multi-source heterogeneous sensors, the temperature and humidity compensation formula is first applied. The original gas concentration data was corrected, among which... The value ranges from 0.003 to 0.005 per ℃. The value is 0.001~0.002 / %RH.

[0011] Preferably, the lightweight graph neural network model in step S5 adopts a two-layer GraphSAGE structure, where each layer aggregates neighbor node information and processes it through the ReLU activation function, ultimately outputting a node risk score.

[0012] Preferably, in step S6, the tiered early warning rule is as follows: A yellow alert is triggered when any area on the risk heatmap has a risk score exceeding 0.6 for at least 30 seconds. An orange alert is triggered if three or more adjacent areas have a risk score exceeding 0.75. A red alert is triggered if the risk score within 50 meters of the predicted leak source exceeds 0.9 or a lethal concentration is detected.

[0013] Preferably, the early warning information in step S6 is visualized using Geographic Information System (GIS) technology and simultaneously pushed to the central control room and on-site personnel terminals.

[0014] Preferably, the feedback signals in step S7 include a true leak confirmation flag, emergency response time, and the number of affected personnel. These feedback signals are used for: (1) Adjust the weights of each risk factor and the parameters of the graph neural network model in the risk entropy calculation; (2) Optimize the normal fluctuation model of gas concentration based on historical data, and update the dynamic warning threshold based on the optimized model.

[0015] (III) Beneficial Effects Compared with existing technologies, this invention provides a distributed monitoring and early warning method for toxic and harmful gases in chemical production environments, which has the following beneficial effects: 1. This distributed monitoring and early warning method for toxic and harmful gases in chemical production environments utilizes digital twin topology modeling (combined with CFD simulation and point placement) and dynamic scheduling of risk entropy to enable sensor nodes to be activated or dormant as needed, adapting to changes in process flow, equipment, and weather conditions, eliminating monitoring blind spots. Simultaneously, by integrating multi-source data through temperature and humidity compensation, spatiotemporal alignment, and DS evidence theory, concentration detection errors are reduced, false alarm rates are decreased, and data accuracy is ensured.

[0016] 2. This distributed monitoring and early warning method for toxic and hazardous gases in chemical production environments can use a lightweight GraphSAGE model at the edge to generate a risk heat map in real time for the next T minutes. Combined with GIS visualization-based hierarchical early warning and second-level cross-terminal push notifications, it can predict diffusion paths in advance and shorten emergency response time. Furthermore, based on the feedback of disposal results, it optimizes risk entropy weights and model parameters to achieve system self-learning and evolution, maintaining the accuracy of risk assessment and the adaptability of early warning in the long term, providing dynamic and efficient safety assurance for chemical production. Attached Figure Description

[0017] Figure 1 This is a flowchart of the distributed monitoring and early warning method for toxic and harmful gases in the chemical production environment according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 Please see Figure 1 The distributed monitoring and early warning method for toxic and hazardous gases in chemical production environments is based on the core logic of "risk-driven—intelligent sensing—dynamic simulation—closed-loop optimization," constructing an integrated technical system covering the sensing layer, edge computing layer, and decision feedback layer. Specifically, it includes the following steps: S1. Construct a digital twin topology map of the factory area. Based on the physical layout, equipment distribution, and ventilation structure of the chemical plant area, a digital twin topology map is constructed, in which the nodes in the map represent the preset monitoring areas, and the edges represent the channels through which gas may diffuse. The initial node placement is determined in conjunction with the results of computational fluid dynamics (CFD) simulations, focusing on covering areas with historically high leakage risk, gas diffusion paths, and densely populated areas to ensure the scientific nature and coverage integrity of the monitoring network.

[0020] S2. Calculate the risk entropy value of each node. By comprehensively considering historical leakage data, equipment operating status (such as aging degree and sealing performance), real-time process parameters (such as pressure, temperature, and material type), and meteorological sensitivity (such as the impact of wind speed, wind direction, and humidity on diffusion), multidimensional risk factors are quantified for each monitoring area and normalized into a probability distribution. Then, the risk entropy value is calculated: ; This entropy value reflects the uncertainty and complexity of regional risks and serves as the basis for subsequent resource allocation.

[0021] S3. Dynamic scheduling of sensor nodes Implement adaptive node management based on risk entropy values: when a certain area... When the preset threshold is exceeded, the surrounding redundant sensor nodes are automatically activated and the sampling frequency is increased (e.g., from 1 time / minute to 5 times / second); conversely, the low-risk area nodes are put into a low-power sleep state. The sensor node integrates a power module, an embedded data processing unit and a LoRa-WiFi dual-mode wireless communication module, supports self-organizing networking, and when any node fails, the adjacent node automatically takes over the communication and monitoring tasks to ensure network robustness.

[0022] S4. Multi-source heterogeneous sensor data fusion Each node synchronously acquires raw data from multiple types of sensors, including electrochemical, infrared spectroscopy, and metal oxide sensors, and first applies the temperature and humidity compensation formula: (in The value ranges from 0.003 to 0.005 per ℃. The concentration values ​​(ranging from 0.001 to 0.002 / %RH) are adjusted for environmental conditions. Subsequently, a Kalman filter with an attention mechanism is used to achieve spatiotemporal alignment of multi-sensor data, and an improved DS evidence theory is used for confidence fusion to output high-precision, interference-resistant fused gas concentration and component identification results.

[0023] S5. Real-time simulation of risk fields at the edge A lightweight graph neural network (GNN) model (preferably a two-layer GraphSAGE structure, with each layer aggregating neighbor node information and then activated by ReLU) is deployed on the edge gateway. The model uses fused concentration, wind speed and direction, ambient temperature and humidity, and equipment start-up and shutdown status as node attributes. Based on the plant topology, the model can extrapolate the risk heat map in real time within the next T minutes (e.g., 5-15 minutes) to predict gas diffusion trends and the evolution of high-risk areas.

[0024] S6. Tiered Early Warning and Coordinated Response Based on the risk heatmap, a three-level early warning rule is set: Yellow alert: Risk score > 0.6 in any area for ≥ 30 seconds; Orange alert: There are ≥3 adjacent areas with a risk score >0.75; Red alert: Risk score >0.9 or lethal concentration detected within 50 meters of the leak source.

[0025] The early warning information is visualized through a geographic information system (GIS) and simultaneously pushed to the central control room's large screen, mobile terminals, and on-site audible and visual alarm devices, achieving a response time of ≤3 seconds.

[0026] S7. Feedback-driven system self-learning optimization The system records the actual handling results of each early warning event, including the actual leak confirmation signs, emergency response time, and number of affected personnel, forming a structured feedback signal. This signal is used to adjust the weights of each factor and the parameters of the graph neural network in the online risk entropy calculation, and to optimize the normal fluctuation model of gas concentration built based on historical data, thereby dynamically updating the early warning threshold and enabling the system to continuously evolve.

[0027] Through the above steps, this technical solution achieves a fundamental transformation from static monitoring to dynamic sensing, from threshold alarms to risk prediction, and from isolated responses to closed-loop optimization, significantly improving the accuracy of toxic and harmful gas monitoring in chemical production environments, the foresight of early warnings, and the synergy of emergency response.

[0028] Example 2 Scenario Example Background: A large chlor-alkali chemical plant has a liquid chlorine storage tank area, an electrolysis workshop, a chlorine compressor room, and surrounding pipe corridors. The liquid chlorine storage tank area is a high-risk area, where a small amount of chlorine gas has been leaked in the past. The electrolysis workshop has complex processes and dense equipment, posing a risk of cross-contamination of multiple toxic gases such as ammonia and chlorine. The prevailing wind direction in the plant area is southeasterly year-round, with high temperature and humidity in summer and dry and windy conditions in winter. Meteorological conditions have a significant impact on gas diffusion.

[0029] To improve intrinsic safety, the plant deployed the distributed monitoring and early warning system described in this invention, including the following implementation steps: S1. Constructing a digital twin topology graph Based on the plant's CAD drawings and 3D laser scanning data, a digital twin topology map containing 42 monitoring nodes was constructed. The nodes cover the liquid chlorine storage tank area (one node every 6 meters), the area around key equipment in the electrolysis workshop (one node every 8 meters), pipe gallery intersections, and personnel inspection passages. The initial node layout plan referenced CFD simulation results: the nodes were densely deployed within 50 meters downwind of the prevailing wind direction, while avoiding "dead zones" formed by building blind spots.

[0030] S2. Calculate the risk entropy value The system automatically updates the risk entropy of each node every day at midnight; taking node #07 of the liquid chlorine storage tank area as an example, its risk factor normalized weights are as follows: Equipment aging (12 years of service): 0.35 Process hazard level (liquid chlorine stored at ambient pressure): 0.40 Historical leakage frequency (2 times in the last 3 years): 0.15 Weather sensitivity coefficient (current wind speed 3.2 m / s, humidity 70%): 0.10 Substituting into the formula, we get: =-(0.35log0.35+0.40log0.40+0.15log0.15+0.10log0.10)≈1.28; If the value exceeds the preset threshold of 1.2, it is identified as a high-risk node.

[0031] S3. Dynamic scheduling of sensor nodes The system automatically activates node #07 and its three surrounding redundant nodes (#06, #08, and #15), increasing the sampling frequency from the default 1 time / minute to 5 times / second. Meanwhile, edge nodes #38-#42, which are far from the tank area, enter a dormant state, only waking up once every 10 minutes to report their status. All nodes use LoRa-WiFi dual-mode communication. When node #07 experiences a brief power outage due to a lightning strike, node #06 automatically takes over its data forwarding task, ensuring uninterrupted network connectivity.

[0032] S4. Data Acquisition and Fusion Chlorine concentration data are collected synchronously at each node: Raw reading from the electrochemical sensor: 3.8 ppm Raw infrared sensor reading: 4.1 ppm Raw reading of metal oxide sensor: 4.5 ppm The current ambient temperature is 32℃ and the humidity is 68%RH. Substituting these values ​​into the temperature and humidity compensation formula (taking a=0.004 / ℃, b=0.0015 / %RH): = 1.055; After compensation, the three data points were 4.01ppm, 4.33ppm, and 4.75ppm, respectively. After spatiotemporal alignment and fusion with the improved DS evidence theory, the final fused concentration was 4.36ppm, with a component identification confidence level of 98.7%.

[0033] S5. Risk Simulation at the Edge The edge gateway (Huawei Atlas 500) runs a lightweight GraphSAGE model (2 layers, 64 hidden dimensions). The inputs include fused concentration, real-time wind direction (southeast, 2.8 m / s), and equipment status (tank valve is open). The model extrapolates the risk heat map for the next 10 minutes, predicting that chlorine will spread to the pipeline area in a southeast direction. The risk scores of nodes #12 and #13 will rise to 0.78 after 5 minutes.

[0034] S6. Tiered Early Warning and Response The system detected a risk score of 0.82 for node #07 for 35 seconds, triggering a yellow alert. Two minutes later, the risk scores for nodes #07, #12, and #13 all exceeded 0.75, and the system upgraded to an orange alert. The alert information was displayed as a heat map on the central control room screen via the GIS platform, and SMS messages were automatically sent to the safety officer's mobile phone. At the same time, the on-site audible and visual alarms were activated; the entire alert chain had a delay of 2.4 seconds.

[0035] S7. Feedback and Self-Learning On-site inspection confirmed a minor leak in the valve seal. The emergency response took 8 minutes, and no personnel were affected. The system recorded the following feedback signals: Actual Leakage Indicator = 1, Response Time = 8 minutes, Number of Affected Personnel = 0. Therefore: 1. Adjust the weight of the "equipment aging" factor from 0.35 to 0.38; 2. Fine-tune the parameters of the last layer of the GraphSAGE model; 3. Update the chlorine normal fluctuation model and lower the baseline threshold for this operating condition from 5 ppm to 4.5 ppm.

[0036] Implementation effect Three months after its launch, the system effectively alerted four leak incidents, reducing the false alarm rate from 23% to 8% and the average warning lead time to 9.2 minutes. It meets the standards for toxic gas detection and significantly improves the inherent safety level of the factory.

[0037] This embodiment fully demonstrates the technical advantages of the present invention in dynamic perception, precise fusion, intelligent inference and closed-loop evolution, and has good engineering applicability and promotion value.

[0038] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A distributed monitoring and early warning method for toxic and harmful gases in chemical production environments, characterized in that, Includes the following steps: S1. Construct a digital twin topology map of the plant area, where nodes represent monitoring areas and edges represent gas diffusion channels; S2. Calculate the risk entropy value of each node based on historical leakage data, equipment status, process parameters, and meteorological sensitivity; S3. Dynamically schedule sensor nodes based on risk entropy values: when the risk entropy of a node is higher than the threshold, activate its surrounding redundant nodes and increase the sampling frequency; otherwise, put it into sleep mode. S4. Collect data from multiple heterogeneous sensors, unify them into a preset spatiotemporal grid through a spatiotemporal alignment module, and perform confidence fusion using DS evidence theory to output the fused gas concentration; S5. Deploy a lightweight graph neural network model on the edge gateway, using concentration, wind speed and direction, temperature and equipment status as node attributes, to predict the risk heat map in real time within the next T minutes. S6. Based on the coverage area and concentration gradient of high-risk areas in the risk heat map, trigger a graded early warning and push it to the central control room and on-site personnel terminals; S7. Record the actual handling results of the early warning event and use them as feedback signals to update the risk entropy calculation weights and graph neural network parameters online.

2. The distributed monitoring and early warning method for toxic and harmful gases in chemical production environments according to claim 1, characterized in that: The construction of the digital twin topology map of the plant area in step S1 also includes using computational fluid dynamics (CFD) to simulate the gas diffusion in different areas and determine the initial monitoring node placement locations.

3. The distributed monitoring and early warning method for toxic and harmful gases in chemical production environments according to claim 1, characterized in that: The formula for calculating risk entropy in step S2 is as follows: ； in The normalized weight of the Kth type of risk factor (such as equipment aging, process hazard, historical leakage frequency, and meteorological sensitivity coefficient) in region i.

4. The distributed monitoring and early warning method for toxic and harmful gases in chemical production environments according to claim 1, characterized in that: The sensor node in step S3 is equipped with a power module, a data processing unit and a wireless communication module, supports the LoRa-WiFi dual-mode networking protocol, and realizes the self-organizing network function. When a node fails, the adjacent node automatically takes over.

5. The distributed monitoring and early warning method for toxic and harmful gases in chemical production environments according to claim 1, characterized in that: In step S4, before collecting data from the multi-source heterogeneous sensors, the temperature and humidity compensation formula is first applied. The original gas concentration data was corrected, among which... The value ranges from 0.003 to 0.005 per ℃. The value is 0.001~0.002 / %RH.

6. The distributed monitoring and early warning method for toxic and harmful gases in chemical production environments according to claim 1, characterized in that: The lightweight graph neural network model in step S5 adopts a two-layer GraphSAGE structure. Each layer aggregates neighbor node information and processes it through the ReLU activation function, and finally outputs the node risk score.

7. The distributed monitoring and early warning method for toxic and harmful gases in chemical production environments according to claim 1, characterized in that: In step S6, the tiered early warning rules are as follows: A yellow alert is triggered when any area on the risk heatmap has a risk score exceeding 0.6 for at least 30 seconds. An orange alert is triggered if three or more adjacent areas have a risk score exceeding 0.

75. A red alert is triggered if the risk score within 50 meters of the predicted leak source exceeds 0.9 or a lethal concentration is detected.

8. The method for distributed monitoring and early warning of toxic and harmful gases in chemical production environments according to claim 1, characterized in that: The early warning information in step S6 is visualized using Geographic Information System (GIS) technology and simultaneously pushed to the central control room and on-site personnel terminals.

9. The distributed monitoring and early warning method for toxic and harmful gases in chemical production environments according to claim 1, characterized in that: The feedback signals in step S7 include a true leak confirmation indicator, emergency response time, and the number of affected personnel. These feedback signals are used for: (1) Adjust the weights of each risk factor and the parameters of the graph neural network model in the risk entropy calculation; (2) Optimize the normal fluctuation model of gas concentration based on historical data, and update the dynamic warning threshold based on the optimized model.

Images