Risk hidden danger checking and evaluating method and system for chemical enterprise production management

Abstract

The invention discloses a risk and hidden danger investigation and evaluation method and system for chemical enterprise production management, and relates to the technical field of enterprise risk and hidden danger investigation, and the method comprises the steps: collecting multi-source monitoring data of a chemical factory in real time through an edge sensing network, uploading the multi-source monitoring data in real time, and synchronously recording data block chain information; based on the three-dimensional model and the physical and chemical mechanism model, constructing a three-dimensional twinborn model, simulating an abnormal working condition, generating a dynamic risk thermodynamic diagram, and marking a risk hidden danger area; fusing multi-source monitoring data to construct an equipment fault knowledge graph, and associating a fault phenomenon with a fault reason; and responding to the risk hidden danger troubleshooting operation and carrying out process evidence storage. According to the method, the limitation of static monitoring and isolated evaluation is broken, an intelligent full-process risk management system with data, model and knowledge collaborative updating is formed, the real-time performance, the accuracy and the automation level of risk early warning are remarkably improved, and the scientificity and the sustainability of production safety management are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise risk and potential hazard investigation, and in particular to a risk and potential hazard investigation, assessment method and system for chemical enterprise production management. Background Art

[0002] Due to the complex factors such as high temperature, high pressure, flammability, explosiveness, and toxic and harmful substances involved in the production process of chemical enterprises, the inherent risks and potential hazards in their production management mainly include equipment failures, process out-of-control, raw material leakage, environmental pollution, and improper operation by personnel.

[0003] Firstly, equipment aging and backward technology are important sources of risks and potential hazards. After long-term high-load operation, equipment such as reactors, pipelines, and valves is prone to problems such as seal failure, corrosion and wear, and cracks. If these potential hazards are not discovered and rectified in time, they may lead to chemical leakage, fires, or even explosion accidents. Secondly, the chemical process itself is highly complex and sensitive, involving multiple-step reactions and precise process control. Any slight deviation in the parameters of any link (such as temperature, pressure, flow rate, etc.) may cause the reaction to get out of control, resulting in violent chemical reactions or the accumulation of intermediate products, forming serious safety hazards. In addition, the skill level and safety awareness of operators are directly related to the stability of the production process. Due to operational errors or insufficient emergency response, it often becomes an inducement for accidents. Moreover, the lack of informatization level makes real-time monitoring and data collection incomplete. Traditional detection means are difficult to accurately monitor complex and changing production environments, resulting in potential hazards being difficult to discover and warn in time.

[0004] Once the risks and potential hazards in chemical enterprise production management cannot be effectively identified and controlled, the consequences will have a profound and serious impact on the enterprise, the surrounding environment, and the entire society. Therefore, only through a scientific risk management system, advanced informatization monitoring means, and strict work safety systems can these potential hazards be fundamentally prevented and resolved to ensure the steady development of the enterprise in the fierce market competition.

[0005] Current methods for risk and potential hazard investigation include HAZOP (Hazard and Operability Analysis), digital technology, and Internet of Things technology, etc. HAZOP multi-disciplinary teams analyze process deviations node by node, evaluate risks and control measures, and have the advantages of strong systematicness, comprehensiveness, and high regulatory recognition. However, it relies on expert experience, takes a long time (several months for a single device), and cannot be dynamically updated. Digital Twin technology constructs a virtual factory model to map equipment status in real time (such as temperature, pressure), real-time monitoring, predictive maintenance, and supports simulation optimization. However, its initial investment is large (in the millions), the model accuracy depends on data quality, traditional methods are inefficient, digital technology costs are high, data fragmentation is serious, and dynamic risk response is insufficient. Summary of the Invention

[0006] Based on this, it is necessary to provide a risk and hidden danger investigation and assessment method and system for chemical enterprise production management in view of the above technical problems.

[0007] In the first aspect, the present invention provides a risk and hidden danger investigation and assessment method for chemical enterprise production management, and the method includes: Real-time collect multi-source monitoring data of a chemical plant through an edge perception network, and upload the multi-source monitoring data to the chain in real time to synchronously record the data blockchain information; Based on a three-dimensional model and a physical and chemical mechanism model, construct a three-dimensional twin model, simulate abnormal working conditions, generate a dynamic risk heat map, and mark the risk and hidden danger areas; Fuse multi-source monitoring data to construct an equipment failure knowledge graph, associate failure phenomena and causes, and evaluate the failure information of the risk and hidden danger areas based on the entropy method and set pair analysis technology; Respond to the risk and hidden danger investigation operation and perform process evidence storage. Dynamically calibrate the parameters of the three-dimensional twin model through knowledge base update and local training, and continuously perform the risk and hidden danger investigation of the plant.

[0008] Further, real-time collect multi-source monitoring data of a chemical plant through an edge perception network, and upload the multi-source monitoring data to the chain in real time to synchronously record the data blockchain information includes: Use explosion-proof robots to participate in intelligent inspection of chemical plants. Deploy monitoring sensors and industrial edge gateways on chemical plant equipment, and use a wireless networking method to form an edge perception network to collect multi-dimensional multi-source monitoring data through multi-source association; Adopt the consortium chain technology to package and upload the hash values of multi-source monitoring data and inspection records to the chain, set multi-level access permissions, and encrypt and store the original data in the InterPlanetary File System.

[0009] Further, based on a three-dimensional model and a physical and chemical mechanism model, construct a three-dimensional twin model, simulate abnormal working conditions, generate a dynamic risk heat map, and mark the risk and hidden danger areas includes: Use building information modeling technology to obtain the geometric information of a chemical plant, construct a three-dimensional model, and construct a physical and chemical mechanism model according to the process reaction and equipment operation process of the chemical process; Introduce multi-source monitoring data into the three-dimensional model to form a three-dimensional twin model, embed the physical and chemical mechanism model, bind the physical and chemical changes of the chemical process, and divide various operation scenarios; Use the integrated physical and chemical mechanism model to simulate abnormal working conditions in different scenarios. According to the simulation results, generate risk heat maps of each part and area, mark the risk and hidden danger areas in the risk heat map, and synchronously display them at the corresponding positions in the three-dimensional twin model.

[0010] Furthermore, construct an equipment failure knowledge graph by integrating multi-source monitoring data, associate failure phenomena and causes, and based on the entropy value method and set pair analysis technology, evaluate the failure information of risk areas, including: Perform noise removal, missing value processing, and standardization processing on the collected multi-source monitoring data in sequence, and fuse the processed multi-source monitoring data into a multi-dimensional data set; Use a bidirectional transducer model for failure annotation. By constructing an equipment failure knowledge graph, automatically associate the failure causes corresponding to the failure phenomena in the chemical plant; According to the association results of the equipment failure knowledge graph, set failure evaluation indicators. Based on the evaluation indicators, set a troubleshooting plan and a standard plan, and calculate the weights of each indicator using entropy theory; Calculate the similarity between the troubleshooting plan corresponding to the current risk area and the standard plan, determine the failure phenomenon and the probability of failure according to the ranking of advantages and disadvantages, and output the failure information.

[0011] Furthermore, use a bidirectional transducer model for failure annotation. By constructing an equipment failure knowledge graph, automatically associate the failure causes corresponding to the failure phenomena in the chemical plant, including: Extract the maintenance history and failure logs from the historical knowledge base, use the bidirectional transducer model to perform natural language processing on the historical maintenance history and failure logs, and identify historical failure phenomena and causes; Use a graph database to store and manage failure phenomena, potential causes, and their association relationships, set nodes and edges, and use a graph neural network to reason about the equipment failure knowledge graph; Use a bidirectional transducer model to annotate equipment failures. By comparing failure phenomena and their potential failure causes, automatically associate the failure causes corresponding to various types of failure phenomena.

[0012] Furthermore, based on the evaluation indicators, set a troubleshooting plan and a standard plan, and calculate the weights of each indicator using entropy theory, including: Obtain the standard values of each evaluation indicator, combine them to form a standard plan, form a troubleshooting plan from the evaluation indicators extracted from the current multi-source monitoring data, calculate the identity degree of each evaluation indicator in each troubleshooting plan and the corresponding evaluation indicator in the standard plan, and form an identity degree matrix; Obtain the optimal value and the worst value of each evaluation indicator, perform normalization processing on the identity degree matrix, set the information entropy weight calculation formula, and calculate the information entropy weights of each evaluation indicator.

[0013] Furthermore, the information entropy weight calculation formula includes: ; ; where wk Denote the information entropy weight value of the k-th evaluation index; H k Denote the information entropy of the k-th evaluation index; n denotes the number of evaluation indexes in the troubleshooting plan; d ik Denote the identity degree between the k-th evaluation index in the troubleshooting plan and the i-th evaluation index in the standard plan; m denotes the number of evaluation indexes in the standard plan; x ik Denote the corresponding evaluation index value; d min Denote the worst value of each evaluation index; d max Denote the optimal value of each evaluation index; b ik Denote the value after normalizing the identity degree matrix.

[0014] Furthermore, calculate the similarity between the troubleshooting plan corresponding to the current risk and hidden danger area and the target plan, determine the fault phenomenon and the probability of fault occurrence according to the ranking of advantages and disadvantages, and output the fault information including: Obtain the multi-source monitoring data of the current risk and hidden danger area, extract the fault evaluation indexes to form a troubleshooting plan, calculate the similarity between the current troubleshooting plan and the standard plan according to the weight matrix composed of information entropy weight values, and form a similarity matrix; Screen the number of advantages and disadvantages of the current troubleshooting plan according to the order of similarity in the similarity matrix. If the similarity difference is lower than the preset threshold, mark that there are risks and hidden dangers in the current troubleshooting plan, and divide the risk level according to the similarity difference as the fault information.

[0015] Furthermore, respond to the risk and hidden danger troubleshooting operation and perform process archiving. Through knowledge base update and local training, dynamically calibrate the parameters of the three-dimensional twin model, and continuously execute the factory risk and hidden danger troubleshooting including: Based on the risk and hidden danger assessment results, trigger the risk and hidden danger troubleshooting task, and automatically update the equipment fault knowledge base according to the troubleshooting feedback, and extract new fault modes and associated causes; Collect the new data generated during the troubleshooting process by the edge computing node, and perform local processing. Through online learning and incremental training methods, optimize the risk and hidden danger fault assessment technology; Combine the real-time monitoring data with the three-dimensional twin model, update the factory equipment status in the model, and based on the equipment entity status and physical and chemical mechanism model, simulate the equipment operation in real time. Based on the fault data and on-site troubleshooting results, dynamically adjust the model parameters of the three-dimensional twin model.

[0016] On the second aspect, the present invention also provides a risk and hidden danger troubleshooting and assessment system for chemical enterprise production management, and the system includes: A data acquisition module, which is used to collect multi-source monitoring data of a chemical plant in real time through an edge perception network, and upload the multi-source monitoring data to the chain in real time, and synchronously record the data blockchain information; The mapping simulation module is used to construct a three-dimensional twin model based on a three-dimensional model and a physical and chemical mechanism model, simulate abnormal working conditions, generate a dynamic risk heat map, and mark risk hidden areas; The analysis and evaluation module is used to construct an equipment failure knowledge graph by integrating multi-source monitoring data, associate failure phenomena and potential causes, and evaluate the failure information of risk hidden areas based on the entropy method and set pair analysis technology; The troubleshooting and update module is used to respond to risk hidden area troubleshooting operations and perform process archiving. Through knowledge base updates and local training, it dynamically calibrates the parameters of the three-dimensional twin model and continuously performs factory risk hidden area troubleshooting.

[0017] The beneficial effects of the present invention are as follows: 1. Real-time collect multi-source monitoring data through an edge perception network and upload it to the chain in real time to ensure the authenticity, reliability, and traceability of the data; use three-dimensional twin technology combined with a physical and chemical mechanism model to simulate abnormal working conditions in the factory in real time, generate a dynamic risk heat map, and intuitively identify risk hidden areas; then integrate multi-source data to construct an equipment failure knowledge graph and combine the entropy method and set pair analysis to quantitatively evaluate the failure information. Finally, through responsive hidden area troubleshooting, process archiving, knowledge base updates, and local training, realize the dynamic calibration of the parameters of the three-dimensional twin model and the continuous closed-loop of risk management. The present invention breaks through the limitations of static monitoring and isolated evaluation, forms an intelligent full-process risk management system with collaborative updates of data, models, and knowledge, significantly improves the real-time performance, accuracy, and automation level of risk early warning, reduces the cost of manual intervention, and enhances the scientific nature and sustainability of production safety management.

[0018] 2. By using a bidirectional transducer model for failure annotation, constructing an equipment failure knowledge graph, and automatically associating failure phenomena and potential failure causes, the intelligent and automated level of failure diagnosis is improved; calculating weights based on the entropy method, combining the equipment failure knowledge graph and evaluation indicators, systematically evaluating risk hidden areas, and providing a standardized troubleshooting plan, improving the accuracy and effectiveness of risk assessment, optimizing the failure identification and troubleshooting process, realizing intelligent and automated failure diagnosis and risk management, significantly improving the efficiency and accuracy of production safety management, reducing the need for manual intervention, and enhancing the real-time performance and reliability of failure prediction and early warning. Description of the Drawings

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings: Figure 1 is a flowchart of a risk hidden area troubleshooting and evaluation method for chemical enterprise production management according to an embodiment of the present invention; Figure 2It is a system principle block diagram of a risk and hidden danger investigation and assessment system for chemical enterprise production management according to an embodiment of the present invention.

[0020] Reference numerals in the attached drawings: 1. Data acquisition module; 2. Mapping simulation module; 3. Analysis and evaluation module; 4. Investigation and update module. Detailed implementation manners

[0021] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0022] Please refer to Figure 1 , a risk and hidden danger investigation and assessment method for chemical enterprise production management is provided, and the method includes: S1. Real-time collect multi-source monitoring data of a chemical plant through an edge perception network, and upload the multi-source monitoring data to the chain in real time, and synchronously record the data blockchain information.

[0023] In the description of the present invention, real-time collecting multi-source monitoring data of a chemical plant through an edge perception network, and uploading the multi-source monitoring data to the chain in real time, and synchronously recording the data blockchain information includes: S11. Use explosion-proof robots to participate in intelligent inspection of chemical plants. Deploy monitoring sensors and industrial edge gateways in chemical plant equipment, and form an edge perception network by using a wireless networking method to collect multi-dimensional multi-source monitoring data through multi-source association.

[0024] Specifically, for the selection and installation of monitoring sensors, it can be summarized as the following aspects: 1. Key equipment monitoring Fiber optic strain sensor: Installed in stress concentration areas such as reaction kettles and pipe elbows to monitor deformation and pressure fluctuations (range 0-50 MPa, accuracy ±0.1%).

[0025] MEMS gas detection chip: Deployed in storage tank areas and ventilation openings to detect dangerous gases such as VOCs and H2S (detection limit in ppm level, anti-hydrogen sulfide corrosion coating).

[0026] High-temperature resistant vibration sensor: Fixed on the bearing seats of pumps and compressors to collect vibration spectra (frequency range 0-10 kHz, temperature resistance 200 °C).

[0027] 2. Wireless networking Adopt ZigBee+LoRa hybrid networking to cover complex plant structures (ZigBee for high-density area coverage, LoRa for long-distance transmission).

[0028] Explosion-proof certification: The sensor has passed ATEX / IECEx certification and meets the requirements of Zone 1 explosive environment.

[0029] For various types of sensors, the sensor reference value needs to be adjusted under no-load conditions (such as calibrating the pressure sensor to atmospheric pressure). Compare the sensor data through a handheld high-precision detector (such as Fluke 754), and an automatic alarm will be triggered when the error exceeds ±2%.

[0030] For the intelligent inspection of explosion-proof robots, the technology and operations include the following aspects: 1. Robot configuration and path planning Hardware configuration: Explosion-proof chassis (Ex d IIB T4 certified), equipped with a dual-light cloud platform (visible light + thermal imaging, temperature measurement range -20°C to 600°C). Multi-gas detection module (PID + electrochemical sensors, detecting 8 types of gases such as benzene and ammonia).

[0031] Autonomous navigation: Based on laser SLAM + UWB positioning to build a high-precision map of the factory (accuracy ±5 cm), dynamic obstacle avoidance (using 3D ToF sensors). Preset inspection routes (circling once per hour) and emergency response trigger mechanisms (such as automatically going for re-inspection when the gas concentration exceeds the limit).

[0032] 2. Data collection and transmission Real-time video stream: Transmit 4K video to the control center through a 5G network (SA networking, UPF sinking), with a delay <50 ms.

[0033] Abnormality marking: The AI edge computing box (NVIDIA Jetson AGX Orin) analyzes the thermal imaging data in real time and marks the temperature abnormality points (such as local overheating of the pipeline >150°C).

[0034] S12. Adopt the consortium blockchain technology to package the hash values of multi-source monitoring data and inspection records onto the chain, set multi-level access permissions, and encrypt the original data and store it in the InterPlanetary File System.

[0035] Specifically, the blockchain network is built using the Hyperledger Fabric consortium blockchain. The nodes include enterprise servers, regulatory agencies, and third-party auditing units. The data deposit contract automatically packages the hash value (SHA-256) of the sensor data and the inspection record onto the chain (block generation interval 2 s); set multi-level access permissions (such as the environmental protection bureau can query the emission data, and the supplier can only view the device status).

[0036] Trusted storage and auditing can achieve on-chain and off-chain collaborative storage, that is, encrypt the original data and store it in IPFS (InterPlanetary File System), upload the hash fingerprint to the chain, and back up the full data in the cloud to support quick retrieval during judicial forensics. In addition, a standardized API is provided for regulatory authorities to check the data integrity (such as verifying whether there is tampering through the hash value).

[0037] S2. Based on the three-dimensional model and the physical and chemical mechanism model, construct a three-dimensional twin model, simulate abnormal working conditions, generate a dynamic risk heat map, and mark the risk hidden areas.

[0038] In the description of the present invention, based on the three-dimensional model and the physical and chemical mechanism model, constructing a three-dimensional twin model, simulating abnormal working conditions, generating a dynamic risk heat map, and marking the risk hidden areas include: S21. Use building information modeling technology to obtain the geometric information of a chemical plant, construct a three-dimensional model, and construct a physical and chemical mechanism model according to the process reactions and equipment operation processes of the chemical process.

[0039] Specifically, use BIM software (such as Revit, Tekla, or ARCHICAD) to obtain geometric information such as the overall building layout, equipment installation location, and pipeline routing of the chemical plant; combine on-site laser scanning (LiDAR) and two-dimensional CAD drawings to generate accurate point cloud data, and then convert it into a three-dimensional model. Integrate all key facilities (reactors, storage tanks, pipelines, valves, etc.) into a unified three-dimensional model according to the actual layout to ensure that the model has high precision and visualization effects, facilitating subsequent data binding and simulation analysis.

[0040] According to the chemical production process, sort out the process flow, reaction path, and heat and mass transfer processes of each equipment, and establish a physical and chemical mechanism model, including a reaction kinetics model, a heat and mass transfer model, and a fluid mechanics model. Among them, the reaction kinetics model establishes reaction rate equations, catalyst activity decay models, etc., to describe the rate and equilibrium of chemical reactions. The heat and mass transfer model establishes heat and mass transfer equations (such as the NTU method, LMTD method) to describe the variation laws of temperature, fluid, concentration, etc. inside the equipment. The fluid mechanics model uses a CFD (Computational Fluid Dynamics) model to model the flow conditions inside pipelines and reactors, and predicts flow velocity, pressure distribution, etc.

[0041] S22. Introduce multi-source monitoring data into the three-dimensional model to form a three-dimensional twin model, embed the physical and chemical mechanism model, bind the physical and chemical changes of the chemical process, and divide various operation scenarios.

[0042] Specifically, the operating data of the factory are collected in real time through sensors (temperature, pressure, flow rate, vibration, gas concentration, etc.) deployed on key equipment; at the same time, environmental and equipment status information is obtained using devices such as video monitoring and infrared thermal imaging. The multi-source data collected are transmitted to the data center in real time through edge computing nodes or industrial Internet of Things platforms and associated with the corresponding equipment or areas in the 3D model. Data interfaces (APIs) are set for each component such as equipment and pipelines in the 3D model to achieve dynamic binding of data and the model, thus forming a digitally-twinned model that is updated in real time.

[0043] Couple the physical and chemical mechanism model established above with the 3D geometric model. Use simulation software to take the real-time data as boundary conditions and initial conditions and dynamically solve the physical and chemical state changes of each component. According to the actual operating conditions of the factory, divide the entire process flow into multiple operating scenarios (such as normal operation, process parameter fluctuations, equipment start-up / shutdown, abnormal conditions, etc.) and distinguish them in the 3D twin model. Finally, use the real-time data to drive the physical mechanism model so that the digitally-twinned model can dynamically reflect the changes in key parameters such as temperature, pressure, and flow rate in the chemical process, ensuring a high degree of consistency between the virtual model and the physical field state.

[0044] S23. Use the integrated physical and chemical mechanism model to simulate abnormal operating conditions in different scenarios. According to the simulation results, generate a risk heat map for each part and area, mark the risk hazard areas in the risk heat map, and synchronously display them at the corresponding positions in the 3D twin model.

[0045] Specifically, in the coupled 3D twin model, set different abnormal operating condition scenarios by modifying key parameters (such as reducing the cooling water flow rate, increasing the reactor temperature, simulating pipeline leakage, etc.). Use the simulation module integrated with physical and chemical mechanisms to dynamically simulate each abnormal scenario and predict the changes in key indicators (temperature, pressure, stress, concentration, etc.) under abnormal conditions. Set the safety thresholds for each indicator according to process safety standards and historical data to facilitate the judgment of the severity of abnormal operating conditions.

[0046] Extract the fault evaluation indicators from the simulation results and use data visualization tools (such as Power BI, Tableau, or a custom visualization platform) to display the risk scores of each equipment and area in the form of the depth of color. The red area indicates high risk (such as the equipment temperature far exceeding the safety upper limit, abnormal pressure, etc.), the yellow is medium risk, and the green is low risk. Link the generated risk heat map with the 3D twin model in real time and synchronously display it at the corresponding positions in the model with graphical identifiers (such as color overlay, heat circle, risk icon, etc.) to facilitate the operator to intuitively locate the risk hazard areas.

[0047] For example, the integrated physical and chemical mechanism model can be used to simulate different abnormal operating conditions, including: 1. Cooling system failure: Simulate the cooling system failure by changing the cooling water flow rate or temperature.

[0048] 2. Catalyst failure: Simulate the reduction in reaction rate caused by catalyst aging or blockage.

[0049] 3. Pipeline leakage: Simulate the pipeline pressure or leakage and observe its impact on the process flow.

[0050] Perform time-varying analysis on the system through numerical solutions (such as CFD, finite element analysis, etc.), observe the system response, and identify possible faults and risks.

[0051] S3. Integrate multi-source monitoring data to construct a device fault knowledge graph, associate fault phenomena with fault causes, and evaluate the fault information in the risk hidden area based on the entropy method and set pair analysis technology.

[0052] It should be noted that the purpose of the knowledge graph is to organize and store the knowledge and experience of device faults by establishing an association network, so as to comprehensively manage and reason about the device status, fault causes, and their solutions. The knowledge graph can help structure the knowledge of device faults, associate different types of fault phenomena with potential causes, and ensure the comprehensiveness of fault diagnosis; it can also achieve automatic reasoning and knowledge query through a graph database and an inference engine. For example, if the device shows abnormal phenomena (such as too high temperature), the system can infer the possible fault causes (such as cooling system failure) and solutions based on the knowledge graph. With the accumulation of device operation data and historical fault data, the knowledge graph can be continuously updated and expanded, improving the accuracy and reliability of fault diagnosis.

[0053] The entropy method and set pair analysis technology are used to evaluate and quantify the risks of devices from different perspectives, especially in an environment with high dynamics and uncertainty. Among them, the entropy method is an evaluation method based on information theory. It evaluates the stability and potential risks of devices by calculating the uncertainty of device monitoring data. Areas with larger entropy values indicate that the information in these areas is more complex and the system is more unstable, so the possibility of faults is higher. Through the entropy method, the risk changes at different time points or under different operating conditions during the device operation process can be reflected, helping to warn of possible fault areas.

[0054] Set pair analysis evaluates the operating status and risks of devices by defining the relationships between device states (such as normal, abnormal, faulty), helping to quantify the probability and severity of fault occurrence. By performing set pair analysis on multiple device states (such as the inputs of multiple sensors), the risks of mutual influence between devices can be evaluated, especially in complex fault situations under the combined action of multiple factors.

[0055] In the description of the present invention, constructing a device fault knowledge graph by fusing multi-source monitoring data, associating fault phenomena and fault causes, and evaluating the fault information of risk hidden areas based on the entropy method and set pair analysis technology includes: S31. Perform noise removal, missing value processing, and normalization processing on the collected multi-source monitoring data in sequence, and fuse the processed multi-source monitoring data into a multi-dimensional data set.

[0056] S32. Use a bidirectional transducer model (BERT model, full name Bidirectional Encoder Representations from Transformers) for fault annotation. By constructing a device fault knowledge graph, automatically associate the fault causes corresponding to the fault phenomena in the chemical plant.

[0057] In the description of the present invention, using a bidirectional transducer model for fault annotation, and automatically associating the fault causes corresponding to the fault phenomena in the chemical plant by constructing a device fault knowledge graph includes: S321. Extract the maintenance history and fault logs from the historical knowledge base, use the bidirectional transducer model to perform natural language processing on the historical maintenance history and fault logs, and identify historical fault phenomena and fault causes.

[0058] Specifically, extract structured and unstructured text data such as maintenance records and fault logs from the historical knowledge base, and perform preprocessing on the text data: including operations such as word segmentation, stop word removal, noise cleaning, and sentence segmentation to construct a standardized text input format.

[0059] Use the pre-trained BERT model to fine-tune the fault text using transfer learning. Use the labeled fault text data (including labels of fault phenomena and corresponding fault causes) to train the model so that it can perform named entity recognition (NER) or sequence annotation tasks, and automatically extract fault-related keywords, phrases, and their semantic information. When implementing, an open-source library (such as Hugging Face Transformers) can be used to load the model, encode the input text, and output the labels and confidence levels of fault phenomena (such as "temperature anomaly", "severe vibration") and fault causes (such as "cooling system failure", "bearing damage").

[0060] S322. Use a graph database to store and manage fault phenomena, potential causes, and their association relationships, set nodes and edges, and use a graph neural network to reason about the device fault knowledge graph.

[0061] Specifically, for graph construction, it includes the following aspects: 1. Node design: In a graph database (such as Neo4j), create multiple node types, including "equipment", "fault phenomenon", "fault cause", "maintenance measures", etc.

[0062] 2. Edge relationship definition: Define the edges between nodes, such as "manifests as", "leads to", "associated with", etc., to reflect the causal or co-occurrence relationship between the fault phenomenon and the fault cause.

[0063] The fault phenomena and fault causes extracted in the S321 stage and their relationships (such as "temperature abnormality" → "cooling system failure") are stored in the graph database to form a preliminary knowledge graph.

[0064] Use graph neural network (GNN) models (such as GraphSAGE, GCN, or GAT) to learn embedded representations of nodes in the graph and capture the structural and semantic information between nodes. By training or fine-tuning the GNN model, perform node classification or edge prediction tasks to further explore potential associations (for example, through similarity judgment, it is found that certain fault phenomena may be associated with multiple fault causes). The reasoning results can be used to enhance the knowledge graph, such as automatically updating newly discovered fault modes or supplementing existing relationships to provide decision support for subsequent automatic associations.

[0065] S323. Use the bidirectional converter model to mark the equipment faults, and automatically associate the fault causes corresponding to each type of fault phenomenon by comparing the fault phenomenon and its potential fault causes.

[0066] Specifically, for newly input equipment fault reports or real-time fault logs, a fine-tuned BERT model is used to encode and annotate the text, extracting the fault phenomenon keywords and related semantic information. The annotation results are semantically vectorized, and the fault phenomenon is represented as a vector of fixed dimension, which is convenient for comparison with the fault cause description stored in the knowledge graph.

[0067] Use semantic similarity calculation (such as cosine similarity) to compare the current fault phenomenon vector with the vector representation of each fault cause in the knowledge graph, and automatically match the fault cause with the highest similarity. According to the similarity threshold and historical matching rules, automatically assign the most likely fault cause label to the fault phenomenon and output the labeling result.

[0068] S33. According to the association results of the equipment fault knowledge graph, set the fault evaluation index, set the troubleshooting plan and standard plan based on the evaluation index, and use the entropy theory to calculate the weight of each index.

[0069] In the description of the present invention, based on the evaluation indicators, the screening scheme and the standard scheme are set, and the entropy theory is used to calculate the weight of each indicator, including: S331. Obtain the standard values of various evaluation indicators, combine them to form a standard solution, form a troubleshooting solution by extracting the evaluation indicators from the current multi-source monitoring data, calculate the identity degree of each evaluation indicator in each troubleshooting solution and the corresponding evaluation indicator in the standard solution, and form an identity degree matrix.

[0070] Specifically, according to the design standards, process requirements, and safety specifications of chemical equipment, pre-define the ideal standard values of various evaluation indicators. For example, assume the common evaluation indicators are: Temperature deviation: The standard value is 0 °C (i.e., no deviation); Pressure deviation: The standard value is 0 kPa; Vibration amplitude: The standard value is 5 mm / s; Flow deviation: The standard value is 0%.

[0071] Standard solution: Combine the standard values of the above indicators into a standard solution vector.

[0072] Extract the real-time measurement values of various evaluation indicators from the current multi-source monitoring data to form a troubleshooting solution vector. For example: The current temperature deviation is 5 °C; the pressure deviation is 10 kPa; the vibration amplitude is 8 mm / s; the flow deviation is 3%.

[0073] For each indicator, calculate the "identity degree" between the current troubleshooting solution and the standard solution, that is, measure the similarity between the actual value and the standard value, then form an identity degree vector from the identity degrees of all indicators, and further form an identity degree matrix.

[0074] S332. Obtain the optimal value and the worst value of each evaluation indicator, perform normalization processing on the identity degree matrix, set the information entropy weight value calculation formula, and calculate the information entropy weight value of each evaluation indicator.

[0075] In the description of the present invention, the information entropy weight value calculation formula includes: .

[0076] .

[0077] In the formula, w k represents the information entropy weight value of the kth evaluation indicator; H k represents the information entropy of the kth evaluation indicator; n represents the number of evaluation indicators in the troubleshooting solution; d ik represents the identity degree of the kth evaluation indicator in the troubleshooting solution and the ith evaluation indicator in the standard solution; m represents the number of evaluation indicators in the standard solution; x ik represents the corresponding evaluation indicator value; d min represents the worst value of each evaluation indicator; d max represents the optimal value of each evaluation indicator; b ikIt represents the normalized value of the same degree matrix.

[0078] S34. Calculate the similarity between the inspection plan corresponding to the current risk and hidden danger area and the standard plan, determine the fault phenomenon and the probability of fault occurrence according to the ranking of advantages and disadvantages, and output the fault information.

[0079] In the description of the present invention, calculating the similarity between the inspection plan corresponding to the current risk and hidden danger area and the target plan, determining the fault phenomenon and the probability of fault occurrence according to the ranking of advantages and disadvantages, and outputting the fault information includes: S341. Obtain the multi-source monitoring data of the current risk and hidden danger area, extract the fault evaluation indicators to form an inspection plan, and calculate the similarity between the current inspection plan and the standard plan according to the weight matrix composed of information entropy weights to form a similarity matrix.

[0080] S342. Screen the number of advantages and disadvantages of the current inspection plan according to the order of similarity in the similarity matrix. If the similarity difference is lower than the preset threshold, mark that there are risks and hidden dangers in the current inspection plan, and divide the risk level according to the similarity difference as the fault information.

[0081] Specifically, sort according to the similarity of each area in the similarity matrix. If the similarity difference is lower than the preset threshold (for example, 0.75), it means that the similarity between the inspection plan and the standard plan is insufficient and there are potential risks. Compare the similarity differences of different areas or devices, and screen out the areas with relatively lower number of advantages and disadvantages, that is, the areas with higher risks.

[0082] Divide the risk level according to the similarity difference: for example, when the difference is lower than 0.25, it is classified as "high risk", 0.25 - 0.5 as "medium risk", and higher than 0.5 as "low risk". When outputting the fault information, combine the similarity results to describe the abnormal conditions of each index in the current inspection plan and the probability of fault occurrence. For example: "The temperature and vibration indicators of this device deviate greatly from the standard, and the similarity is only 0.6, and the fault risk is medium to high." S4. Respond to the risk and hidden danger inspection operation and perform process archiving. Dynamically calibrate the parameters of the three-dimensional twin model through knowledge base update and local training, and continuously perform the factory risk and hidden danger inspection.

[0083] In the description of the present invention, responding to the risk and hidden danger inspection operation and performing process archiving, dynamically calibrating the parameters of the three-dimensional twin model through knowledge base update and local training, and continuously performing the factory risk and hidden danger inspection includes: S41. Based on the risk and hidden danger assessment results, trigger the risk and hidden danger inspection task, and automatically update the device fault knowledge base according to the inspection feedback, and extract new fault modes and associated causes.

[0084] S42. The edge computing node collects the new data generated during the troubleshooting process and performs local processing. Through online learning and incremental training methods, the risk and hidden danger fault assessment technology is optimized.

[0085] Specifically, during the execution of the troubleshooting task, the edge computing node (such as an industrial PC or edge server at the factory site) continuously collects new data generated on-site, including sensor data, video images, fault feedback records, etc. Utilizing the advantages of edge computing, real-time preprocessing (such as data cleaning, feature extraction, data fusion, etc.) is performed before the data is transmitted to the central platform to reduce the data volume and improve the data quality.

[0086] Based on the newly collected data, the risk and hidden danger fault assessment model is optimized using online learning and incremental training methods. The deep learning framework (such as TensorFlow or PyTorch) is used to fine-tune the original model. By combining the new data with historical data, the model's recognition ability for new fault modes is improved through continuous training. This enables the model to timely capture the changes in the on-site equipment status, improve the prediction accuracy, and continuously adapt to the dynamic changes in the equipment operating environment.

[0087] S43. Combine the real-time monitoring data with the three-dimensional twin model, update the factory equipment status in the model, and based on the equipment entity status and physical and chemical mechanism model, simulate the equipment operation in real time. Based on the fault data and on-site troubleshooting results, dynamically adjust the model parameters of the three-dimensional twin model.

[0088] Specifically, closely combine the real-time monitoring data (from sensors, inspection equipment, video surveillance, etc.) with the three-dimensional twin model to dynamically update the operating status of the factory equipment. Each key equipment is set with a data interface in the three-dimensional model, enabling parameters such as temperature, pressure, and flow rate to be reflected on the model in real time, ensuring that the virtual model is highly consistent with the physical equipment status.

[0089] The three-dimensional twin model embeds physical and chemical mechanism models (such as heat and mass transfer, fluid mechanics, and reaction kinetics models). Using the current equipment status and fault data, it simulates the equipment behavior under different operating conditions. By comparing the simulation results with the actual detection data, the deviation between the model prediction and the on-site reality is identified. Based on the on-site troubleshooting feedback and fault data, adaptive control and parameter optimization algorithms are used to dynamically calibrate the three-dimensional twin model. For example, key parameters such as the pipeline corrosion rate and the reactor heat transfer coefficient are adjusted, enabling the model to more accurately predict equipment anomalies. The calibrated model parameters will be fed back to the risk assessment system to achieve a closed-loop optimization of the entire process, ensuring that the system can continuously and efficiently perform the factory risk and hidden danger troubleshooting.

[0090] Please refer to Figure 2, a risk and hidden danger investigation and assessment system for chemical enterprise production management is also provided. The system includes: A data collection module 1, configured to collect multi-source monitoring data of a chemical plant in real time through an edge perception network, and upload the multi-source monitoring data to the blockchain in real time, and synchronously record the data blockchain information.

[0091] A mapping and simulation module 2, configured to construct a three-dimensional twin model based on a three-dimensional model and a physical and chemical mechanism model, simulate abnormal working conditions, generate a dynamic risk heat map, and mark the risk and hidden danger areas.

[0092] An analysis and assessment module 3, configured to fuse multi-source monitoring data to construct an equipment failure knowledge graph, associate failure phenomena and potential causes, and evaluate the failure information of the risk and hidden danger areas based on the entropy method and the set pair analysis technology.

[0093] An investigation and update module 4, configured to respond to the risk and hidden danger investigation operation and perform process evidence preservation, dynamically calibrate the parameters of the three-dimensional twin model through knowledge base update and local training, and continuously perform the risk and hidden danger investigation of the plant.

[0094] In summary, by means of the above technical solutions of the present invention, multi-source monitoring data is collected in real time through an edge perception network and uploaded to the blockchain in real time, ensuring that the data is true, reliable and traceable; the three-dimensional twin technology is used to combine with the physical and chemical mechanism model to simulate the abnormal working conditions of the plant in real time, generate a dynamic risk heat map, and intuitively identify the risk and hidden danger areas; then fuse multi-source data to construct an equipment failure knowledge graph and combine the entropy method and the set pair analysis to quantitatively evaluate the failure information, and finally through responsive hidden danger investigation, process evidence preservation, knowledge base update and local training, realize the dynamic calibration of the parameters of the three-dimensional twin model and the continuous closed-loop of risk management; the present invention breaks the limitations of static monitoring and isolated assessment, forms an intelligent full-process risk management system with collaborative update of data, models and knowledge, significantly improves the real-time, accuracy and automation level of risk early warning, reduces the manual intervention cost at the same time, and enhances the scientificity and sustainability of production safety management. By using a bidirectional converter model for fault annotation, constructing an equipment failure knowledge graph, automatically associating failure phenomena and potential failure causes, the intelligent and automation level of fault diagnosis is improved; based on the entropy method to calculate weights, combined with the equipment failure knowledge graph and evaluation indicators, systematically evaluate the risk and hidden danger areas, and provide a standardized investigation plan, improve the accuracy and effectiveness of risk assessment, optimize the fault identification and investigation process, realize intelligent and automated fault diagnosis and risk management, significantly improve the efficiency and accuracy of production safety management, reduce the need for manual intervention, and enhance the real-time and reliability of fault prediction and early warning.

[0095] It should be understood that although the steps in the flowchart of the accompanying drawings are shown sequentially according to the indication of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear indication in this document, there is no strict order restriction for the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

Claims

1. A method for investigating and assessing risk hazards in production management of a chemical enterprise, characterized in that: The method includes: The multi-source monitoring data of chemical plants is collected in real time through the edge sensing network, and the multi-source monitoring data is uploaded to the blockchain in real time, and the data blockchain information is recorded synchronously; Based on the 3D model and the physical and chemical mechanism model, a 3D twin model is constructed to simulate abnormal working conditions, generate a dynamic risk heat map, and mark the risk potential areas; Integrate multi-source monitoring data to build a knowledge graph of equipment failures, associate failure phenomena and causes, and evaluate failure information in risk areas based on entropy method and set pair analysis technology; Respond to risk and hidden danger investigation operations and conduct process evidence recording. Through knowledge base updates and local training, dynamically calibrate 3D twin model parameters and continuously perform factory risk and hidden danger investigation.

2. The method for identifying and assessing risk hazards in production management of a chemical enterprise according to claim 1, characterized in that: The multi-source monitoring data of the chemical plant is collected in real time through the edge sensing network, and the multi-source monitoring data is uploaded to the chain in real time, and the data blockchain information is recorded synchronously, including: Use explosion-proof robots to participate in intelligent inspections of chemical plants, deploy monitoring sensors and industrial edge gateways on chemical plant equipment, use wireless networking to form an edge perception network, and collect multi-dimensional multi-source monitoring data through multi-source association; Using consortium chain technology, hash values ​​of multi-source monitoring data and inspection records are packaged and uploaded to the chain, multi-level access rights are set, and the original data is encrypted and stored in the Interstellar File System.

3. The method for identifying and assessing risk hazards in production management of a chemical enterprise according to claim 1, characterized in that: The three-dimensional model and the physical and chemical mechanism model are used to construct a three-dimensional twin model, simulate abnormal working conditions, generate a dynamic risk heat map, and mark the risk potential areas, including: Use building information modeling technology to obtain geometric information of chemical plants, build three-dimensional models, and build physical and chemical mechanism models based on the process reactions and equipment operation processes of chemical processes; Introduce multi-source monitoring data into the three-dimensional model to form a three-dimensional twin model, and embed the physical and chemical mechanism model to bind the physical and chemical changes of the chemical process and divide various operation scenarios; The integrated physical and chemical mechanism model is used to simulate abnormal operating conditions in different scenarios. Based on the simulation results, risk heat maps of various parts and regions are generated. Risk potential areas are marked in the risk heat map and displayed simultaneously at corresponding positions in the three-dimensional twin model.

4. The method for identifying and assessing risk hazards in production management of a chemical enterprise according to claim 1, characterized in that: The fusion of multi-source monitoring data to build a knowledge graph of equipment failures, associate failure phenomena and failure causes, and evaluate the failure information of risk potential areas based on the entropy method and set pair analysis technology, including: The collected multi-source monitoring data are subjected to noise removal, missing value processing and standardization processing in turn, and the processed multi-source monitoring data are fused into a multi-dimensional data set; A bidirectional converter model is used for fault annotation. By building a knowledge graph of equipment faults, the fault causes corresponding to the fault phenomena in the chemical plant are automatically associated. According to the results of the equipment fault knowledge graph association, set the fault evaluation index, set the troubleshooting plan and standard plan based on the evaluation index, and use the entropy theory to calculate the weight of each index; Calculate the similarity between the troubleshooting plan corresponding to the current risk hidden danger area and the standard plan, determine the fault phenomenon and fault probability according to the order of merit, and output the fault information.

5. A method for identifying and assessing risks and hidden dangers for production management in a chemical enterprise according to claim 4, characterized in that: The bidirectional converter model is used for fault labeling, and the fault causes corresponding to the fault phenomena in the chemical plant are automatically associated by constructing a knowledge graph of equipment faults, including: Extract maintenance history and fault logs from the historical knowledge base, use the bidirectional transformer model to perform natural language processing on the historical maintenance history and fault logs, and identify historical fault phenomena and fault causes; Use graph database to store and manage fault phenomena, potential causes and their correlations, set nodes and edges, and use graph neural network to reason about equipment fault knowledge graph; The bidirectional converter model is used to mark equipment faults, and the fault causes corresponding to each type of fault phenomenon are automatically associated by comparing the fault phenomena and their potential fault causes.

6. A method for identifying and assessing risk hazards in production management of a chemical enterprise according to claim 4, characterized in that: The method of setting the screening plan and standard plan based on the evaluation indicators and using the entropy theory to calculate the weight of each indicator includes: Obtain the standard values ​​of each evaluation indicator, combine them into a standard plan, combine the evaluation indicators extracted from the current multi-source monitoring data into a screening plan, calculate the identity of each evaluation indicator in each screening plan with the corresponding evaluation indicator in the standard plan, and form an identity matrix; Obtain the optimal and worst values ​​of each evaluation indicator, normalize the identity matrix, set the information entropy weight calculation formula, and calculate the information entropy weight of each evaluation indicator.

7. A method for identifying and assessing risk hazards in production management of a chemical enterprise according to claim 6, characterized in that: The information entropy weight calculation formula includes: ； ； In the formula, w k represents the information entropy weight of the kth evaluation index; H k represents the information entropy of the kth evaluation indicator; n represents the number of evaluation indicators in the investigation plan; d ik represents the degree of identity between the kth evaluation indicator in the screening solution and the ith evaluation indicator in the standard solution; m represents the number of evaluation indicators in the standard solution; x ik Indicates the corresponding evaluation index value; d min Indicates the worst value of each evaluation index; d max represents the optimal value of each evaluation index; b ik Represents the normalized value of the identity matrix.

8. A method for identifying and assessing risks and hidden dangers for production management in a chemical enterprise according to claim 6, characterized in that: The calculation of the similarity between the troubleshooting scheme corresponding to the current risk potential area and the target scheme, the determination of the fault phenomenon and the probability of fault occurrence according to the order of merit, and the output of the fault information include: Obtain multi-source monitoring data of the current risk hidden danger area, extract fault assessment indicators to form a troubleshooting plan, calculate the similarity between the current troubleshooting plan and the standard plan based on the weight matrix composed of information entropy weights, and form a similarity matrix; The number of advantages and disadvantages of the current troubleshooting plan is filtered according to the order of similarity in the similarity matrix. If the similarity difference is lower than the preset threshold, the current troubleshooting plan is marked as having risk hazards, and the risk level is divided according to the similarity difference as fault information.

9. A method for identifying and assessing risks and hidden dangers for production management in a chemical enterprise according to claim 1, characterized in that: The response risk hazard investigation operation and process evidence storage, through knowledge base update and local training, dynamically calibrate the 3D twin model parameters, and continuously perform factory risk hazard investigation, including: Based on the risk assessment results, the risk investigation task is triggered, and according to the investigation feedback, the equipment failure knowledge base is automatically updated to extract new failure modes and related causes; The edge computing nodes collect new data generated during the troubleshooting process and process it locally, optimizing the risk and hidden danger fault assessment technology through online learning and incremental training. Combine real-time monitoring data with the three-dimensional twin model, update the status of factory equipment in the model, and simulate equipment operation in real time based on the equipment's physical status and physical and chemical mechanism model. Dynamically adjust the model parameters of the three-dimensional twin model based on fault data and on-site troubleshooting results.

10. A system for checking and evaluating risks and hidden dangers for production management of chemical enterprises, used to implement the method for checking and evaluating risks and hidden dangers for production management of chemical enterprises according to any one of claims 1 to 9, characterized in that: The system includes: The data collection module is used to collect multi-source monitoring data of chemical plants in real time through the edge sensing network, upload the multi-source monitoring data to the blockchain in real time, and synchronously record the data blockchain information; The mapping simulation module is used to build a three-dimensional twin model based on the three-dimensional model and the physical and chemical mechanism model, simulate abnormal working conditions, generate dynamic risk heat maps, and mark risk potential areas; The analysis and evaluation module is used to integrate multi-source monitoring data to build a knowledge graph of equipment failures, associate failure phenomena with potential causes, and evaluate failure information in risk areas based on entropy method and set pair analysis technology; The investigation and update module is used to respond to risk and hidden danger investigation operations and to store process evidence. Through knowledge base updates and local training, it dynamically calibrates the 3D twin model parameters and continuously performs factory risk and hidden danger investigations.

Images