A dynamic risk early warning system for chemical special operation based on STAMP and entropy increase principle

Abstract

The application discloses a dynamic risk early warning system for chemical special operation based on STAMP and entropy increase principle, and sequentially reasons risks through a dynamic Bayesian network of multi-source sensing fusion and STAMP layered control model mapping for fire and confined space operation, realizes three-level early warning by combining risk entropy and cusp catastrophe model and combining posterior probability, upgrades static approval to whole-process dynamic defense, significantly improves risk time resolution, and realizes active defense before an accident.

Description

Technical Field

[0001] This invention relates to the field of intelligent early warning technology for chemical safety production risks, specifically to a dynamic risk early warning system for special chemical operations based on the STAMP principle and the entropy increase principle. Background Technology

[0002] Hot work and confined space operations are two typical scenarios where accidents frequently occur in special operations in chemical enterprises. The current work permit system mainly completes the static access review before the operation, which is difficult to reflect the dynamic changes in gas environment fluctuations, personnel behavior deviations, and the decay of safety barrier effectiveness during the operation. The paper's statistics show that the proportion of confined space and hot work accidents that turn into major or above accidents is as high as 60.2%, and a large number of accidents occur within a specific time window after the work permit is issued, indicating that the risks of special operations have obvious characteristics of process accumulation and periodic sudden increases.

[0003] Traditional risk assessment methods have the following technical defects: (1) HAZOP and Bow-tie analysis are essentially static, the reliability of safety measures is difficult to quantify, and they are highly dependent on expert experience; (2) Existing assessment methods based on Bayesian networks are limited by the structure of directed acyclic networks and cannot effectively model the interaction, information feedback and closed-loop control relationship between system components; (3) Conventional parameter monitoring only targets single process failure indicators, lacks the ability to recognize the risk situation of multiple factors coupled together, and is difficult to capture the sudden change signal before the critical state of the accident; (4) Although the visualization management system of chemical industrial parks integrates video monitoring and gas detection, it lacks dynamic risk reasoning and early warning mechanisms based on theoretical models. Summary of the Invention

[0004] This invention aims to solve the key technical problem of the disconnect between static assessment and dynamic risk evolution in existing special operation risk management, and provides a dynamic risk early warning system based on STAMP (System-Theoretic Accident Model and Processes) control theory and the principle of entropy increase. This system can achieve real-time perception of the risk situation throughout the entire operation process, dynamic correction of posterior probabilities, and graded early warning before the critical state of an accident.

[0005] This invention provides a dynamic risk early warning system for special chemical operations based on the STAMP principle and the entropy increase principle, comprising: The dynamic risk indicator system construction module is used to construct a hierarchical dynamic risk assessment indicator system from three dimensions: inherent potential energy, dynamic disturbance, and state vulnerability. The multi-source sensing data fusion module is used to integrate gas concentration monitoring data, video AI analysis data, personnel positioning data, environmental parameters, and work permit status data at the work site. The STAMP hierarchical control model construction module is used to construct a hierarchical safety control structure for hot work and confined space operations, and to map the hierarchical safety control structure into a Bayesian network topology containing time slice sequences and feedback coupling edges. The risk entropy calculation module is used to calculate the parameter risk entropy and the total system risk entropy based on the principle of entropy increase, and to determine the risk critical state using the cusp catastrophe model; The Bayesian dynamic reasoning module is used to dynamically update the posterior probability of accident risk based on the real-time evidence provided by the multi-source perception data fusion module, using the sequential evidence update mechanism, and to reason about the accident evolution path, based on the Bayesian network topology. The graded early warning and linkage control module is used to jointly determine the early warning level based on the posterior probability of the accident risk and the total risk entropy of the system, and to trigger graded early warning and linkage control actions when a preset threshold is reached.

[0006] Furthermore, the dynamic risk indicator system construction module is based on accident case statistical analysis and STAMP control structure modeling, identifying risk factors from three dimensions: inherent potential energy, dynamic disturbances, and state vulnerability. The inherent potential energy includes the hazard of the working medium and the energy isolation status; the dynamic disturbances include real-time fluctuations in gas concentration, introduction of unauthorized ignition sources, and changes in ambient temperature and humidity; and the state vulnerability includes the attenuation of barrier effectiveness, deviations in personnel behavior, and downgrading of control measures. The hierarchical dynamic risk assessment indicator system is constructed in accordance with GB 30871-2022.

[0007] Furthermore, in the multi-source sensing data fusion module, the gas concentration monitoring data is detected at a frequency of no less than once every 30 seconds; the video AI analysis data is obtained by computer vision recognition of personnel violations; and the personnel positioning data is obtained based on UWB or Bluetooth positioning tags.

[0008] Furthermore, the STAMP hierarchical control model construction module sets up a four-level control structure: enterprise decision-making level, workshop management level, field operation level, and equipment execution level. Each level forms a closed-loop control loop through sensor feedback, controller decision and actuator response, and formalizes the safety measures specified in GB 30871-2022 into safety constraints for controllers at each level.

[0009] Furthermore, when the STAMP hierarchical control model construction module maps the hierarchical safety control structure to a Bayesian network, the node types include control nodes, process nodes, and feedback nodes. By introducing time slice sequences, the static network is extended into a dynamic Bayesian network, and the feedback coupling edge represents the accumulation and propagation process of risk in the control-feedback loop.

[0010] Furthermore, in the risk entropy calculation module, the k-th type of monitoring parameter is... Risk entropy of parameters at time moment Defined as: Where K takes , For parameter fluctuation characteristic coefficients, The probability distribution of monitored values ​​deviating from the safety threshold; The total risk entropy of the system Defined as: in, The initial risk entropy, It is a positive entropy flow. It is a negative entropy flow.

[0011] Furthermore, the cusp catastrophe model employs a potential function. The state variable x represents the risk state of the system, the control variable u corresponds to the positive entropy flow, and the control variable v corresponds to the negative entropy flow. When the system state crosses the boundary of the bifurcation set, it is determined that a risk transition may occur.

[0012] Furthermore, the Bayesian dynamic reasoning module employs expert evaluation combined with DS evidence theory to fuse multi-source evidence and determine the prior probability and conditional probability tables of nodes; the sequential evidence update mechanism adopts the following formula: in, Let be the posterior probability of the accident risk at time t. The cumulative evidence sequence up to time t; the accident evolution path reasoning is based on the Bow-tie model, which embeds the evolution path of hazard release, barrier failure and accident consequence escalation into the Bayesian network causal reasoning logic.

[0013] Furthermore, the tiered early warning and linkage control module is configured with a three-level early warning mechanism: When the posterior probability of the accident risk increases by 3-5 times compared to the initial state and the total risk entropy of the system reaches the first threshold, a level 3 warning is triggered, prompting enhanced monitoring. When the posterior probability of the accident risk increases by 5-15 times compared to the initial state, a level-two warning is triggered, which triggers on-site audible and visual alarms and pushes out handling suggestions. When the posterior probability of the accident risk increases by more than 15 times compared to the initial state, a Level 1 warning is triggered, automatically triggering a work stoppage order and notifying the emergency response. The linkage control is integrated with the DCS system, and automatically executes control actions such as cutting off materials, activating inerting protection, or stopping hot work in the first-level early warning state.

[0014] On the other hand, the present invention also provides a dynamic risk early warning method for special chemical operations, applied to the system described in any of the above claims, comprising the following steps: S1. Static assessment before operation: enter work ticket information and confirm safety measures; establish initial Bayesian network; calculate initial accident risk probability and initial risk entropy. S2. Real-time data acquisition and fusion: Acquire the multi-source sensing data at a acquisition frequency of not less than 1 time / minute, and perform anomaly detection and feature extraction; S3. Dynamic risk reasoning: Input new evidence into the Bayesian network to perform sequential updates to obtain the posterior probability of accident risk, simultaneously calculate the parameter risk entropy and the total system risk entropy, and call the cusp mutation model to determine whether it is close to the critical state. S4. Early warning determination and output: Determine the early warning level based on the joint distribution of the posterior probability of the accident risk and the total risk entropy of the system, and output the early warning information; when the first-level early warning threshold is reached, automatically trigger a work stoppage command; S5. Post-operation assessment and archiving: Generate a full-process risk evolution report and archive videos, gas monitoring records, and reasoning logs.

[0015] Compared with the prior art, the present invention has the following beneficial effects: To overcome the shortcomings of static access control: the traditional work permit system’s static management model of “approval equals safety” is upgraded to a “full-process dynamic perception-reasoning-early warning” model, which reflects dynamic risk factors such as gas environment fluctuations and personnel behavior deviations in real time during the operation. Improve the time resolution of risk monitoring: Compared with the traditional manual detection of 2 hours, the time resolution is improved by about 120 times with an update frequency of 1 minute. The system response delay is less than 5 seconds. It can automatically trigger graded early warning and shutdown instructions 1 to 2 stages before the critical state of accidents such as flash explosions. Achieve proactive early warning: By analyzing the cumulative trend of risk entropy and identifying cusp mutations, identify early warning signals of risk transitions that are difficult to detect by conventional parameter monitoring, and transform accident management from "post-event response" to "proactive defense"; The model is highly interpretable: the STAMP control structure provides a systematic modeling framework for risk causation, and Bayesian posterior probability inference can trace the key causal paths of risk escalation, assisting on-site decision-makers in implementing precise measures. Taking into account technical, human, and organizational factors: The STAMP method can model multi-dimensional risk factors and their interactive coupling relationships, including equipment failure, human error, and management deficiencies, breaking through the technical limitations of traditional methods that only focus on single failures. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall system architecture, showing the data flow and control flow relationships between various functional modules. Detailed Implementation

[0017] 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.

[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, not all, of the embodiments of the present invention. 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] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0021] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0022] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0024] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0025] Example 1: This example uses a Class 1 hot work operation in a chemical plant's tank area as an application scenario to describe the deployment and operation process of the system of this invention in detail. The tanks store naphtha with a flash point below -20°C, classifying it as a Class B flammable medium. The hot work operation is located at the flange connection of the inlet and outlet pipelines of the tanks, requiring gas welding and cutting. The working environment presents a risk of residual flammable gas release, and other normally operating tanks and transfer pump sets are present nearby, making energy isolation and operation control quite challenging.

[0026] (a) System deployment: such as Figure 1 As shown, the system includes: The dynamic risk indicator system construction module is used to construct a hierarchical dynamic risk assessment indicator system from three dimensions: inherent potential energy, dynamic disturbance, and state vulnerability. The multi-source sensing data fusion module is used to integrate gas concentration monitoring data, video AI analysis data, personnel positioning data, environmental parameters, and work permit status data at the work site. The STAMP hierarchical control model construction module is used to construct a hierarchical safety control structure for hot work and confined space operations, and to map the hierarchical safety control structure into a Bayesian network topology containing time slice sequences and feedback coupling edges. The risk entropy calculation module is used to calculate the parameter risk entropy and the total system risk entropy based on the principle of entropy increase, and to determine the risk critical state using the cusp catastrophe model; The Bayesian dynamic reasoning module is used to dynamically update the posterior probability of accident risk based on the real-time evidence provided by the multi-source perception data fusion module, using the sequential evidence update mechanism, and to reason about the accident evolution path, based on the Bayesian network topology. The graded early warning and linkage control module is used to jointly determine the early warning level based on the posterior probability of the accident risk and the total risk entropy of the system, and to trigger graded early warning and linkage control actions when a preset threshold is reached.

[0027] The following multi-source sensing equipment will be deployed at the hot work site: Gas concentration monitoring unit: 4 explosion-proof composite gas detectors, placed 1 meter and 3 meters downwind of the hot work point, next to the tank drain valve, and at an upwind reference point on site. Detection parameters include the concentrations of combustible gas (LEL%), hydrogen sulfide (H2S), carbon monoxide (CO), and oxygen (O2). The sampling interval is set to 30 seconds, and the detection data is connected to the on-site explosion-proof data acquisition unit via an RS-485 bus.

[0028] Video AI Analysis Unit: Two explosion-proof network cameras, covering the hot work area and the personnel passageway respectively. Video signals are transmitted to an edge AI analysis server via gigabit industrial Ethernet. The server deploys YOLOv7-based models for pedestrian detection, workwear / safety helmet recognition, detection of crossing boundaries in hot work spark areas, and detection of supervisors leaving their posts. The inference frame rate is no less than 15fps, and a violation tag is output every 10 seconds.

[0029] Personnel positioning unit: Operators, supervisors, and approval personnel wear UWB positioning tags. Four UWB positioning base stations are installed on site to achieve centimeter-level real-time positioning with a positioning refresh frequency of 1Hz. After the positioning data is processed by the UWB engine, the coordinates are output and the electronic fence boundary crossing event is judged.

[0030] Environmental parameter monitoring: one explosion-proof temperature and humidity sensor and one anemometer, with monitoring data updated once per minute.

[0031] Electronic work permit system interface: It connects with the company's existing electronic work permit management system through the OPC-UA protocol to read the work permit approval status, safety measures confirmation records, gas analysis results within 30 minutes before the operation, and the qualification information of the operators in real time.

[0032] The system server is deployed in the enterprise security management center, configured with two Intel Xeon Gold processors, 128GB of memory, and 4TB of SSD storage, running the CentOS operating system and deploying the dynamic risk inference engine and early warning service of this invention. The server is interconnected with the enterprise's DCS system, fire control system, and emergency broadcast system via industrial Ethernet, configured with Modbus TCP / IP and OPC-UA communication protocols. On-site audible and visual alarms are controlled via relay output modules, and an industrial explosion-proof flat panel terminal serves as the information receiving terminal for supervisors.

[0033] (II) Static assessment before operation: Before the work begins, safety management personnel enter the hot work permit information into the system, including: the work unit, the workers, the supervisor, the start and end times of the work, the location of the hot work and surrounding process conditions, the energy isolation method (pipeline purging, cleaning and replacement, blind flange isolation), and the characteristics of the hazardous media involved. The system automatically reads the safety measures list from the electronic work permit system and checks each item to confirm, including: covering the trench and manhole within 15 meters of the hot work point, ensuring the combustible gas detection result is less than 10% LEL, and ensuring that fire-fighting equipment is in place.

[0034] Subsequently, the system performs the following initial modeling steps: 1. Construct a STAMP hierarchical control structure. For this hot work operation in the tank area, a four-level control structure is established: the enterprise decision-making level (Vice President of Safety, Vice President of Production) is responsible for approving the special-level hot work operation and supervising company-level safety measures; the workshop management level (workshop director, safety officer) is responsible for implementing the work plan, arranging supervisors and energy isolation; the on-site operation level (hot work operator, supervisor, process operator) is responsible for executing the hot work operation, continuous gas monitoring, and on-site emergency response; the equipment execution level (combustible gas detector, DCS emergency shut-off valve, fire sprinkler system) is responsible for sensing physical parameters and executing safety controls. Based on the safety requirements for hot work operations in GB 30871-2022, the system generates formalized safety constraints for each controller, such as "the supervisor must continuously monitor the hot work area and must not leave," and "the concentration of combustible gas within 10 meters of the hot work point must not reach 10% of the lower explosive limit," etc. 2. Generate the initial Bayesian network. Map the above STAMP control structure to a Bayesian network, where control nodes include "effectiveness of enterprise approval decisions," "implementation level of workshop safety measures," "monitor's performance status," "tendency of hot work operators to violate regulations," and "operating status of gas detectors," etc.; process nodes include "intensity of combustible gas release," "possibility of ignition source existence," "ventilation dilution effect," and "barrier effectiveness," etc.; feedback nodes correspond to sensor observations (such as "combustible gas detection readings," "personnel location data," and "video violation recognition results"). Construct the initial network using time slices $t_0$ (at the start of the operation) and set the state transition relationships between time slices to establish a dynamic Bayesian network framework; 3. Calculate the initial risk probability and risk entropy. Based on the characteristics of the hazardous medium in this operation (naphtha flash point below -20℃, lower explosive limit 1.2%), energy isolation status (blind flange installed and nitrogen purged), and the experience of the operators, the prior probability of each root node is determined by combining expert prior scoring with DS evidence theory. For example, the prior probability of the node "accidental release of flammable gas" is set to 0.05, and the prior probability of "supervisor absent from duty" is set to 0.02. The initial posterior probability of the accident risk is obtained through forward inference using a Bayesian network. Simultaneously, the initial total system risk entropy is calculated based on the formula. (Normalized value). This benchmark value serves as a reference for subsequent dynamic changes in risk.

[0035] (III) Dynamic Risk Reasoning and Early Warning in the Operation Process: The operation officially began at 9:00 AM, and the system entered dynamic monitoring mode, executing steps S2-S4 at a fusion frequency of once per minute. The following three typical moments during the operation illustrate the dynamic inference and early warning process.

[0036] Time 9:15—Normal Operation Phase: Monitoring data shows that the concentration of combustible gas is stable at around 3% LEL, the oxygen concentration is 20.9%, and no toxic gases are detected; video analysis did not identify any violations, and all personnel are located within the compliant area; the ambient temperature is 26℃, the humidity is 55%, and the wind speed is 1.2 m / s. The evidence vector $E_t$ at this time indicates that each observation node is in a normal state.

[0037] The Bayesian inference engine performs sequential updates: it uses the posterior probability distribution of the previous time slice as the prior of the current time slice, and calculates the current evidence as input. The result is The risk probability increased by only about 1.09 times compared to the initial value, which is lower than the level 3 warning threshold.

[0038] Meanwhile, the parameter risk entropy calculation module calculates the parameter risk entropy of the combustible gas concentration sequence based on a sliding window of nearly 10 minutes. The monitored values ​​remained stable around 3% LEL, with a standard deviation of 0.4% LEL. The probability distribution of deviations from the safety threshold (10% LEL) was concentrated. The calculated values... The risk entropy of environmental parameters and personnel status parameters is also low, with positive entropy flow. Minimal. Negative entropy flow. It persists as long as the guardian confirms that safety measures are in place and gas detection is functioning normally. Total system risk entropy. The value increased only slightly from the initial value. The cusp mutation detection module calculates the system state. Located in the stable region, no warning output is issued. The system only records data in the background and displays a "normal operation" status.

[0039] Time 10:02—Abnormal gas fluctuations trigger Level II warning: Starting from 9:58, the readings from combustible gas detector A (1 meter downwind of the hot work point) showed an abnormal increase: 5% LEL at 9:58, jumping to 12% LEL at 9:59, reaching 18% LEL at 10:00, and dropping back to 15% LEL at 10:01. Oxygen concentration remained at 20.8%, and other detection points showed no significant changes. At 10:00:30, the video AI detected that the hot work operator briefly crossed the fire blanket's shielding area while adjusting the cutting torch. Personnel location data showed the supervisor was 8 meters from the hot work point and had not yet left the monitored area.

[0040] The system's fusion processing at 10:02 was as follows: Evidence extraction: Combustible gas concentration extraction "seriously deviated from the baseline, up to 18% LEL, exceeding the 10% LEL safety threshold", generating evidence "combustible gas barrier partially failed"; video generation "operation position offset, splashing sparks may have entered the unshielded area"; positioning data was normal.

[0041] Sequential evidence update: The above evidence is input into the Bayesian network, affecting the nodes "combustible gas release intensity," "probability of abnormal ignition source introduction," and "effectiveness of physical isolation barriers." Through probability propagation, the posterior probability of the 10:02 accident risk is... Updated to , is the initial value It is approximately 9.1 times that of the previous level, falling into the level II warning range (5-15 times).

[0042] Risk entropy calculation: The fluctuation range of the combustible gas concentration sequence has increased sharply in the past 5 minutes, and the risk entropy parameter $H^{comb}$ has jumped to 0.62. Due to the increased gas release and barrier failure, the positive entropy flow... Significantly increased, while negative entropy flow There was no significant increase in the total risk entropy of the system. Accumulated to 0.48. The cusp mutation model was invoked. It is approaching the boundary of the bifurcation set, and the system judges that the risk is in the early warning stage of the critical domain, but has not yet entered the mutation domain.

[0043] Early warning triggering and linkage: Based on the joint judgment criteria (posterior probability reaches the second-level threshold and risk entropy) The system triggers a level 2 warning (alarm). The warning information is output in the following ways: (1) The explosion-proof sound and light alarm at the hot work site emits a flashing yellow light and intermittent buzzing sound; (2) The supervisor receives the alarm information on the explosion-proof tablet and displays: "The concentration of combustible gas at the hot work point exceeds the standard (peak value 18% LEL). The operation spark may enter the weak protection area. Stop the operation immediately and check the gas pipe isolation and fire blanket arrangement!" and displays the risk cause path diagram - "micro-leakage of the storage tank breathing valve → accumulation of combustible gas → insufficient physical barrier shielding → increased risk of ignition source contact with combustible gas"; (3) The monitoring screen of the operation point and the risk parameter trend diagram automatically pop up on the large screen of the central control room. The person in charge of the operation immediately stops the hot work according to the alarm, re-checks the energy isolation measures, adds fire blankets and increases the nitrogen purging volume. After 10 minutes of treatment, the gas concentration drops back to below 4% LEL, the system posterior probability gradually recovers to the baseline level, and the warning is automatically lifted.

[0044] At 11:20—A Level 1 Warning was triggered by a combination of risks: After work resumed at 10:50, the supervisor briefly left the monitoring area (15 meters away from the hot work point) to verify the blind flange isolation status, and remained off-duty for 2 minutes between 11:18 and 11:20. Simultaneously, as the cutting operation entered the pipe circumferential weld stage, the flame intensity needed to be increased. Oxygen and acetylene cylinder hoses were bent by being stepped on on the ground, causing a brief backfire that produced a small flash explosion. Although it did not ignite the surrounding environment, the video AI detected an abnormal flash, and the combustible gas detection suddenly rose to 30% LEL at 11:19 (due to the vaporization of trace amounts of residual liquid in the pipe), while the oxygen concentration showed a short-term consumption decrease to 20.4%.

[0045] At 11:20, the system performed data fusion and reasoning: evidence extraction: "the guardian left the supervision post for more than 2 minutes", "flash explosion sound and abnormal flame appeared at the work site", "the concentration of combustible gas exceeded the limit (30% LEL)", and "the oxygen concentration fluctuated".

[0046] Sequential Update: The aforementioned evidence strongly influences key indicators such as "guardian's immediate intervention capability," "ignition source control," and "intensity of combustible gas leakage," and the post-accident probability. leap to , and initial value In comparison, the risk has increased by approximately 20.9 times, far exceeding the first-level warning threshold (15 times).

[0047] Risk entropy synchronous calculation: Combustible gas concentration sequences and video events enable the calculation of parameter risk entropy. , All increased sharply, positive entropy flow It presents a pulsed influx, It surged to 0.85. Cusp mutation model assessment. Having crossed the bifurcation set and entered the mutation domain, the system has an extremely high probability of mutating from a "risk accumulation state" to an "accident state".

[0048] Level 1 Warning Triggered: The system immediately triggers a Level 1 warning (emergency) and executes the following linked actions: The on-site audible and visual alarms emitted red flashing lights and a high-intensity siren. An emergency shutdown command is sent to the DCS system. The DCS automatically closes the emergency shut-off valves of the upstream and downstream pipelines of the hot work point and starts the fixed fire sprinkler system of the storage tank to cool and protect the adjacent storage tanks. Send a start signal to the fire control system to grant remote control access to the on-site fire monitors; The emergency broadcast system was used to issue a voice prompt to the entire plant: "Hot work in the tank area must be stopped immediately, and personnel must evacuate immediately." Send alarm information containing risk details, cause chain, and coordinates to the enterprise's security officer, the 119 command center's preset interface, and the company's emergency command center.

[0049] Thanks to early warning and automatic shut-off and inerting protection, the flash explosion energy was effectively suppressed, preventing fire and personal injury. After the operation was terminated, on-site inspection confirmed the presence of residual liquid in the pipeline that had not been completely drained. Combined with the supervisor's absence from their post and improper operation, this constituted a major near miss.

[0050] (iv) Post-work evaluation and archiving: After the operation is terminated, the system automatically generates a full-process risk evolution report for this hot work operation, which includes: The curve showing the change of risk posterior probability over time is marked with three warning trigger times and their corresponding posterior probability values ​​and changes in risk entropy. The ranking of key contributing factors by their impact intensity is as follows: accidental release of combustible gas contributed the most (41%), guardian leaving their post (27%), insufficient physical barriers (18%), and operational violations (14%). System response performance statistics: The average delay from evidence fusion to the issuance of the early warning signal is 3.8 seconds, and the data fusion cycle is stable at 1 minute; Based on the early warning lead time of the cusp mutation model: the first-level warning is triggered about 30 seconds before the flash explosion energy is released, which provides a critical time window for automatic protection actions.

[0051] All video clips, raw gas monitoring data, inference process logs, and early warning records are automatically archived into the safety event database, serving as the basis for subsequent model parameter optimization and historical accident analysis.

[0052] Example 2: This example applies the invention to a confined space operation scenario in a fertilizer plant, demonstrating the system's adaptability to different special operation types. The confined space is an underground fire water tank dredging operation, with an operating depth of 6 meters, which may pose risks of oxygen deficiency and hydrogen sulfide accumulation.

[0053] The system deploys gas detectors, UWB personnel positioning (including personnel entry and exit statistics), wellhead video surveillance, and Bluetooth relay for portable gas detectors within the confined space. In the STAMP modeling, "ventilation effectiveness," "continuous gas detection," and "rescue equipment positioning" are designated as critical control loops. The Bayesian network structure is adjusted based on the risk characteristics of the confined space, adding "poisoning / asphyxiation" as a key consequence node.

[0054] During the operation, at a certain moment, the system detected an unauthorized auxiliary personnel entering the confined space (electronic fence breach). Simultaneously, a portable gas detector showed that the oxygen concentration slowly decreased from 20.9% to 19.7%. Although this was not yet at the alarm limit of 19.5%, the system detected the change in information entropy due to the continuous monotonic downward trend of the oxygen concentration through risk entropy calculation, and the unauthorized entry was confirmed by video AI. After the Bayesian posterior probability was updated, it increased to 6.8 times the initial value, triggering a level-two warning. The monitoring terminal received the instruction "Unauthorized personnel have appeared in the confined space, and the oxygen concentration is showing a downward trend. Immediately evacuate personnel and increase ventilation," thus promptly eliminating the risk and preventing a potential oxygen deprivation and asphyxiation accident.

[0055] This embodiment demonstrates that the present invention, through a unified STAMP control structure mapping and entropy increase analysis framework, can adapt to the dynamic risk evolution characteristics of different scenarios such as hot work and confined space operations, and can achieve early warning in all of them.

[0056] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic risk early warning system for special chemical operations based on the STAMP principle and the entropy increase principle, characterized in that, include: The dynamic risk indicator system construction module is used to construct a hierarchical dynamic risk assessment indicator system from three dimensions: inherent potential energy, dynamic disturbance, and state vulnerability. The multi-source sensing data fusion module is used to integrate gas concentration monitoring data, video AI analysis data, personnel positioning data, environmental parameters, and work permit status data at the work site. The STAMP hierarchical control model construction module is used to construct a hierarchical safety control structure for hot work and confined space operations, and to map the hierarchical safety control structure into a Bayesian network topology containing time slice sequences and feedback coupling edges. The risk entropy calculation module is used to calculate the parameter risk entropy and the total system risk entropy based on the principle of entropy increase, and to determine the risk critical state using the cusp catastrophe model; The Bayesian dynamic reasoning module is used to dynamically update the posterior probability of accident risk based on the real-time evidence provided by the multi-source perception data fusion module, using the sequential evidence update mechanism, and to reason about the accident evolution path, based on the Bayesian network topology. The graded early warning and linkage control module is used to jointly determine the early warning level based on the posterior probability of the accident risk and the total risk entropy of the system, and to trigger graded early warning and linkage control actions when a preset threshold is reached.

2. The system according to claim 1, characterized in that, The dynamic risk indicator system construction module is based on accident case statistical analysis and STAMP control structure modeling, and identifies risk factors from three dimensions: inherent potential energy, dynamic disturbances, and state vulnerability. The inherent potential energy includes the hazard of the working medium and the energy isolation status; the dynamic disturbances include real-time fluctuations in gas concentration, introduction of unauthorized ignition sources, and changes in ambient temperature and humidity; and the state vulnerability includes the decay of barrier effectiveness, deviations in personnel behavior, and downgrading of control measures. The hierarchical dynamic risk assessment indicator system is constructed in accordance with GB 30871-2022.

3. The system according to claim 1, characterized in that, In the multi-source sensing data fusion module, the gas concentration monitoring data is detected at a frequency of no less than once every 30 seconds; the video AI analysis data is obtained by computer vision recognition of personnel violations; and the personnel positioning data is obtained based on UWB or Bluetooth positioning tags.

4. The system according to claim 1, characterized in that, The STAMP hierarchical control model construction module sets up a four-level control structure: enterprise decision-making level, workshop management level, field operation level, and equipment execution level. Each level forms a closed-loop control loop through sensor feedback, controller decision and actuator response, and formalizes the safety measures specified in GB 30871-2022 into safety constraints for controllers at each level.

5. The system according to claim 1 or 4, characterized in that, When the STAMP hierarchical control model construction module maps the hierarchical safety control structure to a Bayesian network, the node types include control nodes, process nodes, and feedback nodes. By introducing time slice sequences, the static network is extended into a dynamic Bayesian network, and the feedback coupling edge represents the accumulation and propagation process of risk in the control-feedback loop.

6. The system according to claim 1, characterized in that, In the risk entropy calculation module, the k-th type of monitoring parameter is... Risk entropy of parameters at time moment Defined as: Where K is taken as, For parameter fluctuation characteristic coefficients, The probability distribution of monitored values ​​deviating from the safety threshold; The total risk entropy of the system Defined as: Where is the initial risk entropy. It is a positive entropy flow. It is a negative entropy flow.

7. The system according to claim 1 or 6, characterized in that, The cusp mutation model uses a potential function. The state variable x represents the risk state of the system, the control variable u corresponds to the positive entropy flow, and the control variable v corresponds to the negative entropy flow. When the system state crosses the boundary of the bifurcation set, it is determined that a risk transition may occur.

8. The system according to claim 1, characterized in that, The Bayesian dynamic inference module uses expert evaluation combined with DS evidence theory to fuse multi-source evidence to determine the prior probability and conditional probability tables of nodes; the sequential evidence update mechanism adopts the following formula: in, Let be the posterior probability of the accident risk at time t. The cumulative evidence sequence up to time t; the accident evolution path reasoning is based on the Bow-tie model, which embeds the evolution path of hazard release, barrier failure and accident consequence escalation into the Bayesian network causal reasoning logic.

9. The system according to claim 1, characterized in that, The hierarchical early warning and linkage control module is configured with a three-level early warning mechanism: When the posterior probability of the accident risk increases by 3-5 times compared to the initial state and the total risk entropy of the system reaches the first threshold, a level 3 warning is triggered, prompting enhanced monitoring. When the posterior probability of the accident risk increases by 5-15 times compared to the initial state, a level-two warning is triggered, which triggers on-site audible and visual alarms and pushes out handling suggestions. When the posterior probability of the accident risk increases by more than 15 times compared to the initial state, a Level 1 warning is triggered, automatically triggering a work stoppage order and notifying the emergency response. The linkage control is integrated with the DCS system, and automatically executes control actions such as cutting off materials, activating inerting protection, or stopping hot work in the first-level early warning state.

10. A dynamic risk early warning method for special chemical operations, applied to the system described in any one of claims 1 to 9, characterized in that, Includes the following steps: S1. Static assessment before operation: enter work ticket information and confirm safety measures; establish initial Bayesian network; calculate initial accident risk probability and initial risk entropy. S2. Real-time data acquisition and fusion: Acquire the multi-source sensing data at a acquisition frequency of not less than 1 time / minute, and perform anomaly detection and feature extraction; S3. Dynamic risk reasoning: Input new evidence into the Bayesian network to perform sequential updates to obtain the posterior probability of accident risk, simultaneously calculate the parameter risk entropy and the total system risk entropy, and call the cusp mutation model to determine whether it is close to the critical state. S4. Early warning determination and output: Determine the early warning level based on the joint distribution of the posterior probability of the accident risk and the total risk entropy of the system, and output the early warning information; When the Level 1 warning threshold is reached, a work stoppage command will be automatically triggered. S5. Post-operation assessment and archiving: Generate a full-process risk evolution report and archive videos, gas monitoring records, and reasoning logs.

Images