Digital evaluation method and system for enterprise safety production and storage medium

Abstract

The invention relates to the technical field of production data evaluation, in particular to a digital evaluation method and system for enterprise safety production and a storage medium. The method comprises the following steps: acquiring enterprise safety production multi-source monitoring data including equipment operation state data, environment monitoring data, personnel behavior data and safety management system data, and performing data cleaning and multi-modal fusion processing to generate a safety production fusion data set; extracting risk features of the safety production fusion data set to construct a safety production knowledge graph; and analyzing a production operation chain of the safety production knowledge graph, and performing operation action capture based on the production operation chain to generate production action key data. According to the invention, through multi-source data fusion, dynamic risk assessment and closed-loop management, intelligentization, real-time performance and systematization of enterprise safety production are realized, and accuracy and responsiveness of digital evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of production data evaluation, and in particular to a digital evaluation method, system and storage medium for enterprise production safety. Background Art

[0002] Traditional manual inspections and paper-based record-keeping methods are no longer able to meet the efficient and accurate demands of modern safety management. To improve production safety, information technology has gradually been introduced into safety management. Initially, hardware-based approaches such as video surveillance and sensor alarms were primarily used for real-time monitoring. However, these technologies were largely reactive and lacked systematic analytical capabilities. In recent years, emerging technologies such as artificial intelligence (AI), digital twins, and edge computing have further advanced enterprise safety management towards intelligent, automated, and predictive approaches. In particular, by establishing digital assessment models, enterprises can dynamically assess and provide early warnings for safety risks, significantly improving their accident prevention and control capabilities. However, traditional methods currently rely on static indicators or historical incident analysis, making it difficult to identify dynamic risks and hidden dangers along the behavioral chain. Furthermore, risks are often limited to single-point detection and lack modeling of how risks are transmitted between the environment and behaviors. This results in low accuracy and responsiveness in digital assessments. Summary of the Invention

[0003] Based on this, it is necessary to provide a digital evaluation method, system and storage medium for enterprise safety production to solve at least one of the above technical problems.

[0004] To achieve the above objectives, a digital evaluation method for enterprise production safety is provided, the method comprising the following steps: Step S1: Acquire enterprise safety production multi-source monitoring data, including equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data, and perform data cleaning and multimodal fusion processing to generate a safety production fusion data set; Step S2: Extract risk features from the safety production fusion dataset to construct a safety production knowledge graph; analyze the production operation chain of the safety production knowledge graph, and perform operation motion capture based on the production operation chain to generate key production action data; Step S3: Perform spatial microenvironment perception based on key production action data, and conduct risk transmission on the key production action data based on the perception results to generate production risk transmission data; input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; Step S4: Dynamically divide security levels based on the security risk quantitative score to generate a hierarchical map of enterprise security situation; compare warning thresholds based on risk evolution trend prediction data to generate graded warning instructions; Step S5: Output the safety situation classification map and classification warning instructions through the visual interactive platform, and dynamically update the enterprise safety production knowledge base to complete the safety production digital evaluation closed loop.

[0005] By integrating and processing multi-source monitoring data from equipment, environment, personnel, and management systems, this invention significantly improves the comprehensiveness and accuracy of production safety data, providing a solid data foundation for subsequent risk identification and assessment. Based on a production safety knowledge graph and production operation chain analysis, it can accurately capture key production action data, perceive fine-grained spatial microenvironmental risks from the operational level, and improve the depth and accuracy of risk transmission analysis. Combining key production action data with spatial microenvironment perception results to conduct risk transmission and multi-dimensional analysis can not only quantify current safety risks but also predict risk evolution trends, achieving a shift from traditional post-event handling to pre-event prediction and in-event intervention. By dynamically dividing enterprise safety levels through safety risk quantitative scoring and implementing graded warnings based on risk trends, it is possible to achieve precise response and hierarchical management for different risk levels, improving the agility and effectiveness of safety management. Through a visual interactive platform, a safety situation classification graph and warning instructions are output in real time, and the production safety knowledge base is dynamically updated, forming a complete digital evaluation closed loop, continuously optimizing the enterprise safety management system, and improving the scientific nature and timeliness of decision-making. Therefore, the present invention realizes the intelligent, real-time and systematic production safety of enterprises through multi-source data fusion, dynamic risk assessment and closed-loop management, and improves the accuracy and responsiveness of digital evaluation.

[0006] Preferably, step S1 includes the following steps: Step S11: collecting equipment vibration spectrum, temperature and pressure data, and energy consumption curve in real time through the industrial Internet of Things terminal to generate equipment operation status data; Step S12: using a distributed sensor network to obtain environmental temperature and humidity, toxic gas concentration, and fire monitoring data to generate environmental monitoring data; Step S13: Collect personnel location trajectory, operation specification data and labor protection equipment status through smart wearable devices and video surveillance systems to generate personnel behavior data; Step S14: Perform natural language processing on the enterprise safety management system document, extract safety management elements and construct standard compliance feature vectors, thereby obtaining safety management system data; Step S15: Perform spatiotemporal fusion of equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data to generate a safety production fusion data set.

[0007] The present invention uses industrial Internet of Things terminals to collect equipment vibration spectrum, temperature, pressure and energy consumption curve data in real time, which can fully grasp the equipment operating status, identify potential failure risks in advance, and improve the level of equipment safety operation and maintenance. Based on a distributed sensor network, it obtains environmental temperature and humidity, toxic gas concentration and fire data, covering typical environmental safety indicators, realizing multi-dimensional dynamic perception of the production environment, and significantly improving the timeliness of early warning of environmental abnormal events. Through the dual-channel collection of personnel positioning trajectory, operating specifications and labor protection equipment status of smart wearable devices and video monitoring systems, high-precision monitoring of personnel behavior and immediate identification of illegal operations are achieved, strengthening personnel safety management. Natural language processing technology is applied to intelligently analyze enterprise safety management system documents, extract standard compliance feature vectors, and form a computable and comparable system data model to provide data support for system implementation monitoring and compliance assessment. By integrating multi-source data of equipment, environment, personnel and system in time and space, a unified safety production fusion data set is formed, which not only improves data integrity and consistency, but also provides a unified data foundation for subsequent risk analysis, situational awareness and decision-making.

[0008] Preferably, step S2 includes the following steps: Step S21: Graph structure modeling is performed on the safety production fusion dataset to construct an initial knowledge graph with equipment nodes, environment nodes, personnel nodes, and management nodes as entities; Step S22: Extract potential risk transmission relationships between nodes in the initial knowledge graph through a graph neural network to generate a risk propagation path graph; perform risk pattern matching on the risk propagation path graph based on a preset historical accident case library, annotate risk feature types and severity levels, and generate production safety risk features; Step S23: Optimize the initial knowledge graph according to the production safety risk characteristics to obtain a production safety knowledge graph; Step S24: Analyze the production operation chain of the safety production knowledge graph, and capture operation actions based on the production operation chain to generate key data of production actions.

[0009] The present invention constructs an initial knowledge graph with equipment, environment, personnel and management as entity nodes by modeling the graph structure of the integrated production safety data set, which can break the traditional data isolation problem and fully reveal the correlation between production safety factors. The graph neural network is used to perform deep feature extraction on the initial knowledge graph, automatically discover the potential risk transmission relationship between nodes, and generate a risk propagation path graph, which greatly improves the concealment and foresight of risk identification. Based on the historical accident case library, the risk propagation path graph is pattern matched, and the risk feature type and severity are marked, which realizes the intelligent identification and quantitative description of potential risks and improves the accuracy and practicality of risk feature extraction. Through the graph optimization guided by risk features, the dynamic evolution and reasoning analysis capabilities of the production safety knowledge graph are effectively enhanced, providing a more reliable knowledge base for subsequent safety situation awareness and decision support. Based on the optimized production safety knowledge graph, the production operation chain is extracted, and combined with the operation action capture technology, the key data of the production action is accurately extracted, which can achieve the tracing and control of safety risks from the micro-operation level, laying the foundation for risk pre-intervention.

[0010] Preferably, step S24 includes the following steps: Step S241: Perform a depth-first search on the safety production knowledge graph, traverse and extract the nodes and relationships in the graph, and obtain a production operation chain; sequence the production operation chain to obtain production operation chain sequence data; Step S242: Decomposing the production operation chain based on the production operation chain sequence data to obtain operation action decomposition data; performing node action feature modeling on the operation action decomposition data to generate an operation action feature template set; Step S243: configuring motion capture points according to the operation motion feature template set, wherein the motion capture point configuration includes joints, gestures, and tool contact points, and obtaining motion capture point definition data; Step S244: Deploy and plan a field capture system based on the motion capture point definition data, and use the motion capture system to perform real-time motion capture and data synchronous collection to generate an original production motion data stream; Step S245: performing action segmentation and feature extraction on the original production action data stream to generate key production action data.

[0011] The present invention conducts a depth-first search on the safety production knowledge graph, systematically traverses nodes and relationships, accurately extracts the production operation chain, and generates serialized data to ensure that the subsequent operation action decomposition has high integrity and traceability. Action decomposition is performed based on the production operation chain sequence data, and a node action feature template set is constructed to achieve standardized description and modeling of operation actions, providing a unified and accurate action feature foundation for subsequent action capture and risk analysis. Through the definition of action capture points, including fine-grained configuration of joints, gestures, and tool contact points, the action capture area and action elements can be accurately calibrated, greatly improving the acquisition accuracy and stability of the motion capture system. The deployed motion capture system is used to achieve on-site real-time action capture and data synchronization acquisition, ensuring the timeliness and continuity of production action data, and providing reliable data support for dynamic analysis of safety risks in the production process. Action segmentation and feature extraction are performed on the original production action data stream to generate high-quality key production action data, which can be used for subsequent spatial microenvironment perception, risk transmission analysis and other links, forming an accurate deduction chain from action to risk.

[0012] Preferably, step S3 includes the following steps: Step S31: performing spatial positioning mapping based on key production action data to generate action space mapping data; performing spatial microenvironment parameter perception on the action space mapping data to generate perception results; Step S32: confirming the execution target of the key data of the production action based on the sensing result, wherein the execution target includes a dynamic target and a static target, the dynamic target is the employee, and the static target is the machine; Step S33: Modeling the spatiotemporal behavior trajectory of the dynamic target to generate dynamic behavior trajectory data; modeling the working state of the static target to generate static workflow feature data; Step S34: Using the perception results, perform an environmental modulation behavior-state linkage analysis on the dynamic behavior trajectory data and the static workflow feature data to generate production risk transmission data; Step S35: Input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data.

[0013] This invention spatially locates and maps key production action data, combined with spatial microenvironmental parameter perception, to capture dynamic environmental changes at the worksite in real time, laying the foundation for accurate perception for subsequent risk assessment. Based on the perception results, it distinguishes between dynamic targets (people) and static targets (equipment), and separately models spatiotemporal behavior trajectories and static workflows. This allows for a comprehensive description of both human-machine interactions and independent behavior patterns during the production process. Through environmentally modulated behavior-state linkage analysis, dynamic behavior trajectories are correlated with static equipment feature data to deeply explore the impact of microenvironmental changes on human and machine behavior patterns, enhancing the comprehensiveness and accuracy of production risk transmission analysis. The production risk transmission data generated through this linkage analysis accurately reveals potential risk sources, transmission chains, and affected targets, effectively enabling pre-emptive and granular risk identification. This production risk transmission data is fed into a pretrained safety risk assessment model for a multi-dimensional comprehensive analysis. This not only provides a quantitative score for current safety risks but also predicts risk evolution trends, enabling companies to proactively implement preventative measures and improve overall safety management.

[0014] Preferably, step S34 includes the following steps: The perception results are used to quantify the spatial position disturbance of the dynamic behavior trajectory data, extract the trajectory offset vector, and generate dynamic trajectory disturbance vector data; Calculate the variability of workstation nodes on the static workflow feature data to generate static workstation variability index data; Based on the dynamic trajectory disturbance vector data and the static workstation volatility index data, trajectory-node mapping analysis is performed to generate trajectory-node mapping relationship data; Perform dynamic contact frequency statistics and load superposition analysis on trajectory node mapping relationship data, extract high-frequency contact nodes and their load change rates, and generate high-risk node warning data; Perform fault pattern recognition on high-risk node warning data, extract node failure pattern characteristics, and generate node failure risk factor data; Based on the node failure risk factor data, time-series link deduction is performed to construct a risk transmission path diagram, and finally generate production risk transmission data.

[0015] This invention quantifies spatial position disturbances in dynamic behavior trajectory data and extracts trajectory offset vectors, enabling real-time monitoring of abnormal offset behaviors in production operations and improving the ability to identify risk trigger points under micro-environmental changes. Workstation node volatility is calculated based on static workflow feature data to quantitatively describe the stability and variation risk of each workstation node, providing static support data for spatial risk transmission analysis. Trajectory-node mapping analysis, based on dynamic trajectory disturbance vector data and static workstation volatility index data, clearly reveals the actual contact relationship between personnel trajectories and workstation nodes, enhancing the location of high-risk points. Dynamic contact frequency statistics and load superposition analysis extract high-frequency contact nodes and their load change rates, enabling timely identification of high-risk nodes subject to concentrated forces and prone to fatigue damage during operations, enabling dynamic early warning. Fault pattern recognition is performed on high-risk nodes, extracting node failure mode characteristics and generating node failure risk factor data. This further deepens the internal cause analysis of risk sources and supports precise prevention. Time-series link deduction based on node failure risk factor data constructs a risk transmission path diagram, clearly simulating the gradual evolution of risk from the initial node to associated nodes, providing a decision-making basis for full-chain risk control.

[0016] Preferably, step S35 includes the following steps: Step S351: Sample construction and label matching are performed on the production risk transmission data to generate risk training sample data; the risk training sample data is divided into data sets to generate model training sets and model test sets; the production risk transmission data is input into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; Step S352: Using a convolutional neural network algorithm to perform model training on the model training set, thereby generating a security risk assessment pre-model; performing model optimization iteration on the security risk assessment pre-model using the model test set, thereby generating a pre-trained security risk assessment model; Step S353: Input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data, where the multi-dimensional analysis includes safety risk quantification and time series trend prediction.

[0017] The present invention forms structured risk training sample data by constructing samples and performing label matching processing on production risk transmission data, ensuring the standardization and accuracy of model input data and providing a solid foundation for subsequent efficient modeling. Dividing the risk training sample data into a model training set and a model test set effectively prevents model overfitting and improves the generalization and adaptability of the safety risk assessment model in different scenarios. Training the model training set with a convolutional neural network (CNN) can fully tap into the potential complex feature patterns in production risk transmission data and significantly improve the accuracy of safety risk identification and quantitative assessment. Optimizing and iterating the safety risk assessment pre-model based on the model test set effectively corrects initial training deviations and ensures that the final pre-trained model has stability and high prediction accuracy. Inputting production risk transmission data into the pre-trained model for multi-dimensional analysis can not only obtain accurate safety risk quantitative scores, but also predict the direction of risk evolution based on time series trend prediction methods, assisting enterprises in formulating prevention and control strategies in advance.

[0018] Preferably, step S4 includes the following steps: Step S41: divide the security risk quantitative score data into intervals according to the set dynamic division rules to generate preliminary security level division data; Step S42: Performing a time sequence consistency check on the preliminary security level classification data, removing abnormal points of classification mutations, and generating revised security level classification data; Step S43: Perform visual mapping processing based on the revised security level classification data, draw by level nodes, and generate a hierarchical map of enterprise security situation; Step S44: Perform sliding window differential analysis on the risk evolution trend prediction data, extract short-term accelerated risk events, and generate potential risk acceleration event data; compare the potential risk acceleration event data with the set warning threshold data to generate corresponding graded warning instructions.

[0019] The present invention divides the safety risk quantitative scoring data into intervals according to the set dynamic division rules, can scientifically convert the safety risk quantitative scoring into safety levels, and generate preliminary safety level division data, laying the foundation for subsequent risk classification management. By performing a time series consistency check on the preliminary security level division data, the abnormal points of the division mutation are eliminated, and the influence of external interference and data noise on the results is effectively avoided, ensuring that the final security level division data is more stable and reliable. Based on the corrected security level division data, visual mapping processing is performed, and a graded map of the enterprise security situation is drawn according to the security level node, so that managers can intuitively and in real time grasp the overall security status of the enterprise and support more flexible decision-making. Sliding window differential analysis is performed on the risk evolution trend prediction data, which can accurately extract risk events that accelerate in the short term, timely discover potential high-risk risk events, give early warnings, and reduce the probability of sudden accidents. By comparing the potential risk acceleration event data with the set warning threshold data, the corresponding graded warning instructions can be accurately generated, so that the enterprise can respond quickly when the risk critical value appears, and prevent problems before they occur.

[0020] In this specification, a digital evaluation system for enterprise safety production is provided, which is used to execute the above-mentioned digital evaluation method for enterprise safety production. The digital evaluation system for enterprise safety production includes: The data fusion module is used to obtain multi-source monitoring data on enterprise production safety, including equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data, and perform data cleaning and multimodal fusion processing to generate a production safety fusion data set; The production monitoring module is used to extract risk features from the safety production fusion data set to build a safety production knowledge graph; analyze the production operation chain of the safety production knowledge graph, and capture operation actions based on the production operation chain to generate key production action data; The risk transmission module is used to perceive the spatial microenvironment based on key production action data, and based on the perception results, conduct risk transmission on the key production action data to generate production risk transmission data. The production risk transmission data is input into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; The production early warning module is used to dynamically divide security levels based on the quantitative safety risk scores and generate a hierarchical map of the enterprise security situation; it compares early warning thresholds based on risk evolution trend prediction data and generates graded early warning instructions; The visualization module is used to output the security situation classification map and graded warning instructions through the visual interactive platform, and dynamically update the enterprise's production safety knowledge base to complete the digital evaluation closed loop of production safety.

[0021] The present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the digital evaluation method for enterprise safety production as described above is implemented.

[0022] The beneficial effect of the present invention is that it generates a safe production fusion data set by collecting equipment operation status data, environmental monitoring data, personnel behavior data and safety management system data, and performing data cleaning and multimodal fusion processing. This module can integrate information from multiple monitoring sources to ensure the comprehensiveness of the data and provide rich basic data for subsequent analysis and decision-making. By extracting the risk characteristics of the safe production fusion data set and constructing a safe production knowledge map, it helps enterprises to fully understand the risk relationships and mutual influences of each link, and provide clear structured knowledge for the safety management of the production process. By analyzing the production operation chain and capturing the operation actions based on this, key data of production actions are generated. This module can monitor the key operation data in the production process in real time, helping enterprises to make accurate decisions and identify risks in a dynamic environment. By performing spatial microenvironment perception based on key data of production actions, the module can accurately identify the safety risks of the production site, and perform risk transmission analysis on the data through the perception results to generate production risk transmission data, which helps to predict potential safety issues in advance and ensure the safety of the production environment. By inputting the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis, it is possible to obtain safety risk quantitative scores and risk evolution trend prediction data, which can help enterprises assess the severity of safety risks and predict future risk evolution, providing a scientific basis for the enterprise's safety management. By dynamically dividing and generating safety levels based on the safety risk quantitative scores, and drawing a safety situation classification map, real-time safety situation assessment is provided to enterprise managers, facilitating quick decision-making. Based on the risk evolution trend prediction data, warning thresholds are compared and graded warning instructions are generated to ensure that enterprises can take timely measures to effectively prevent and respond to sudden safety incidents when safety risks are about to reach a dangerous critical point. By outputting the safety situation classification map and graded warning instructions through a visual interactive platform, managers can intuitively understand the current safety situation and risk situation of the enterprise, and improve the accuracy and response speed of decision-making. Therefore, the present invention realizes the intelligent, real-time and systematic production safety of enterprises through multi-source data fusion, dynamic risk assessment and closed-loop management, and improves the accuracy and responsiveness of digital evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of the steps of a digital evaluation method for enterprise safety production; Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG. Figure 3 for Figure 1Detailed implementation steps of step S3 in FIG. The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0024] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative work are within the scope of protection of the present invention.

[0025] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0026] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0027] To achieve this, please refer to Figures 1 to 3 A digital evaluation method for enterprise safety production, comprising the following steps: Step S1: Acquire enterprise safety production multi-source monitoring data, including equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data, and perform data cleaning and multimodal fusion processing to generate a safety production fusion data set; Step S2: Extract risk features from the safety production fusion dataset to construct a safety production knowledge graph; analyze the production operation chain of the safety production knowledge graph, and perform operation motion capture based on the production operation chain to generate key production action data; Step S3: Perform spatial microenvironment perception based on key production action data, and conduct risk transmission on the key production action data based on the perception results to generate production risk transmission data; input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; Step S4: Dynamically divide security levels based on the security risk quantitative score to generate a hierarchical map of enterprise security situation; compare warning thresholds based on risk evolution trend prediction data to generate graded warning instructions; Step S5: Output the safety situation classification map and classification warning instructions through the visual interactive platform, and dynamically update the enterprise safety production knowledge base to complete the safety production digital evaluation closed loop.

[0028] By integrating and processing multi-source monitoring data from equipment, environment, personnel, and management systems, this invention significantly improves the comprehensiveness and accuracy of production safety data, providing a solid data foundation for subsequent risk identification and assessment. Based on a production safety knowledge graph and production operation chain analysis, it can accurately capture key production action data, perceive fine-grained spatial microenvironmental risks from the operational level, and improve the depth and accuracy of risk transmission analysis. Combining key production action data with spatial microenvironment perception results to conduct risk transmission and multi-dimensional analysis can not only quantify current safety risks but also predict risk evolution trends, achieving a shift from traditional post-event handling to pre-event prediction and in-event intervention. By dynamically dividing enterprise safety levels through safety risk quantitative scoring and implementing graded warnings based on risk trends, it is possible to achieve precise response and hierarchical management for different risk levels, improving the agility and effectiveness of safety management. Through a visual interactive platform, a safety situation classification graph and warning instructions are output in real time, and the production safety knowledge base is dynamically updated, forming a complete digital evaluation closed loop, continuously optimizing the enterprise safety management system, and improving the scientific nature and timeliness of decision-making.

[0029] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a digital evaluation method for enterprise safety production according to the present invention. In this example, the digital evaluation method for enterprise safety production includes the following steps: Step S1: Acquire enterprise safety production multi-source monitoring data, including equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data, and perform data cleaning and multimodal fusion processing to generate a safety production fusion data set; Step S2: Extract risk features from the safety production fusion dataset to construct a safety production knowledge graph; analyze the production operation chain of the safety production knowledge graph, and perform operation motion capture based on the production operation chain to generate key production action data; Step S3: Perform spatial microenvironment perception based on key production action data, and conduct risk transmission on the key production action data based on the perception results to generate production risk transmission data; input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; Step S4: Dynamically divide security levels based on the security risk quantitative score to generate a hierarchical map of enterprise security situation; compare warning thresholds based on risk evolution trend prediction data to generate graded warning instructions; Step S5: Output the safety situation classification map and classification warning instructions through the visual interactive platform, and dynamically update the enterprise safety production knowledge base to complete the safety production digital evaluation closed loop.

[0030] In an embodiment of the present invention, multi-source monitoring data from an enterprise's production safety process is acquired, specifically including equipment operating status data (such as real-time parameters such as temperature, current, voltage, and vibration), environmental monitoring data (such as hazardous gas concentrations, humidity, dust, and noise), personnel behavior data (such as personnel location trajectories, work postures, and violation identification), and safety management system data (such as operating procedures, emergency plans, and historical violation records). High-frequency sampling is achieved through data acquisition terminals, sensor networks, and edge computing devices, and the collected data is cleaned, including missing value filling, outlier removal, time alignment, and unified format conversion. Subsequently, based on a multimodal fusion algorithm (such as a cross-modal embedding model driven by an attention mechanism), structured and unstructured data (images, text, and sensor signals) are semantically aligned and jointly modeled, ultimately generating a unified format of a fused production safety dataset with high-dimensional expression capabilities and cross-domain connectivity. In step S2, risk features are extracted from the fused dataset, including those related to equipment failures, environmental anomalies, behavioral violations, and deficiencies in management systems. High-value information is extracted using clustering, principal component analysis (PCA), and deep feature embedding networks. A safety production knowledge graph is constructed based on the extracted results. The knowledge graph nodes include four core elements: equipment, personnel, environment, and systems, while edges represent causal, temporal, and correlational relationships. The production operation chain within the knowledge graph is further analyzed to identify typical operational sequences. Action capture is then performed using video surveillance, sensor tracking, or AR-assisted capture systems. Core operational behaviors such as startup, inspection, assembly, and cleaning are identified, and key spatiotemporal nodes and posture features are extracted to generate structured production action key data. Next, in step S3, this extracted production action key data is combined with on-site environmental microsensors (such as UWB positioning, ambient light sensors, and temperature and humidity probes) to perform spatial microenvironmental perception, generating coupled action-environment data pairs. Based on this coupling relationship, a risk transmission model is constructed, taking into account action execution errors, environmental interference levels, and historical accident paths. This model models the step-by-step transmission of risk within the operational chain, generating production risk transmission data with a chain structure. This data is then fed into a pre-trained safety risk assessment model based on a graph neural network (GNN) and time series modeling for multi-dimensional risk analysis. The output includes static safety risk quantitative scores (such as risk level values ​​and confidence intervals) and dynamic risk evolution trend prediction data (such as risk growth rate within the next 24 hours and inflection point identification). In step S4, based on the generated risk quantitative score data, a multi-level risk level standard (such as red, orange, yellow, and blue) is set to dynamically divide security levels, generating a hierarchical map of the enterprise security situation. Simultaneously, a real-time comparison is performed based on the trend prediction data and a historical warning threshold library. When the predicted risk growth exceeds the set threshold or a level jump occurs, a prioritized hierarchical warning instruction is automatically generated, including the warning level, responsible department, and recommended response measures.Finally, in step S5, the security situation classification map and the corresponding graded warning instructions are intuitively presented to management personnel in the form of charts, heat maps, dynamic curves, etc. through the enterprise-side visual interactive platform. The platform also supports multi-terminal linkage operations and remote response mechanisms. In addition, the platform supports backflow processing of feedback results, automatically stores data such as risk detection, response execution, and warning effects at each stage, and uses them to update the enterprise's production safety knowledge base, realizing a fully closed-loop digital dynamic evaluation system for production safety of "perception-identification-assessment-warning-response-learning", and promoting the continuous evolution and closed-loop optimization of the enterprise's intelligent risk governance capabilities.

[0031] Preferably, step S1 includes the following steps: Step S11: collecting equipment vibration spectrum, temperature and pressure data, and energy consumption curve in real time through the industrial Internet of Things terminal to generate equipment operation status data; Step S12: using a distributed sensor network to obtain environmental temperature and humidity, toxic gas concentration, and fire monitoring data to generate environmental monitoring data; Step S13: Collect personnel location trajectory, operation specification data and labor protection equipment status through smart wearable devices and video surveillance systems to generate personnel behavior data; Step S14: Perform natural language processing on the enterprise safety management system document, extract safety management elements and construct standard compliance feature vectors, thereby obtaining safety management system data; Step S15: Perform spatiotemporal fusion of equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data to generate a safety production fusion data set.

[0032] In an embodiment of the present invention, in step S11, multi-dimensional signals in the equipment operation process are collected in real time through the industrial Internet of Things terminal deployed on the enterprise's production equipment, including but not limited to the equipment vibration spectrum (for identifying structural abnormalities or mechanical failure warnings), temperature and pressure sensor data (for monitoring whether the equipment working condition is stable), and time series curve data of power consumption (for evaluating the equipment load and energy efficiency status), to comprehensively generate high-precision and time-series complete equipment operation status data; in step S12, distributed environmental sensor nodes are deployed in key areas to obtain fire monitoring data including the trend of environmental temperature and humidity changes, concentration values ​​of toxic and harmful gases (such as CO, H2S, VOC, etc.), and smoke, flames, etc., and regional mapping processing of environmental data is realized in combination with spatial positioning tags, thereby generating environmental monitoring data covering the entire area; in step S13, with the help of smart wearable devices (such as smart helmets, positioning badges, physiological status monitoring bracelets) and video surveillance systems, the spatial positioning trajectory of front-line operators and the standardization of their work movements (through posture monitoring) are collected. The system uses natural language processing technology to extract information and perform semantic analysis on the internal safety management system documents (such as operating procedures, safety training records, and accident emergency plans), extract key safety management elements (such as job responsibilities, operating permissions, and safety levels), and construct standardized compliance feature vectors to evaluate the degree of system implementation coverage and weak links, ultimately forming machine-readable safety management system data. In step S15, the four core data sources mentioned above—equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data—are integrated. A fusion framework based on timestamps and spatial coordinates is used, combined with multimodal data embedding technology to achieve data synchronization and unification in the time and space dimensions, generating a safe production fusion dataset with a clear structure and complete data suitable for subsequent modeling and analysis.

[0033] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: Graph structure modeling is performed on the safety production fusion dataset to construct an initial knowledge graph with equipment nodes, environment nodes, personnel nodes, and management nodes as entities; Step S22: Extract potential risk transmission relationships between nodes in the initial knowledge graph through a graph neural network to generate a risk propagation path graph; perform risk pattern matching on the risk propagation path graph based on a preset historical accident case library, annotate risk feature types and severity levels, and generate production safety risk features; Step S23: Optimize the initial knowledge graph according to the production safety risk characteristics to obtain a production safety knowledge graph; Step S24: Analyze the production operation chain of the safety production knowledge graph, and capture operation actions based on the production operation chain to generate key data of production actions.

[0034] In an embodiment of the present invention, in step S21, graph structure modeling is performed on the safety production fusion dataset generated in step S1, and an initial knowledge graph is constructed in the form of an entity-relationship graph (Entity-Relation Graph), wherein the equipment operating status, environmental monitoring, personnel behavior, and management system are mapped to equipment nodes, environment nodes, personnel nodes, and management nodes, respectively. The association between the nodes is established through temporal co-occurrence, spatial coupling, and management dependency. For example, relationships such as "operator-control-equipment", "equipment-operating-environment", and "personnel-controlled-system" are established to construct an initial safety production knowledge graph with clear entity definitions and semantic annotations. In step S22, a graph neural network (GNN), such as GraphSAGE or GAT (graph attention network), is introduced to perform deep embedding feature extraction on the initial graph, identify implicit risk transmission paths between nodes, and construct a risk propagation path graph. The generated path graph is semantically matched and structurally aligned with a preset historical accident case library to identify potential risk scenario patterns (such as equipment overload-temperature rise-personnel proximity-poisoning accident), and combined with indicators such as loss and casualty levels recorded in the case, Each path node and its combination is labeled with the risk feature type (such as mechanical injury, electrical fire, chemical poisoning, etc.) and severity level (such as Level I-extremely high, Level II-high, Level III-medium, etc.) to generate production safety risk features; in step S23, based on the extracted production safety risk features, the initial knowledge graph is structurally optimized and semantically enhanced, including adding node label attributes (such as risk labels, historical occurrence probabilities), strengthening the relationship weights of high-risk nodes, merging homogeneous node clusters, and other operations, ultimately obtaining a production safety knowledge graph with clearer risk relationships and more accurate transmission chains; in step S24, the chain collaborative relationship between entities such as equipment, personnel, and environment in various types of operation tasks in the production safety knowledge graph is further analyzed to construct a production operation chain model, and the operation actions are dynamically captured and time series encoded based on video surveillance data, wearable device motion capture data, etc., to extract key production action nodes (such as illegal operations, accidental touch of equipment, high-frequency abnormal behaviors, etc.) to generate key production action data.

[0035] Preferably, step S24 includes the following steps: Step S241: Perform a depth-first search on the safety production knowledge graph, traverse and extract the nodes and relationships in the graph, and obtain a production operation chain; sequence the production operation chain to obtain production operation chain sequence data; Step S242: Decomposing the production operation chain based on the production operation chain sequence data to obtain operation action decomposition data; performing node action feature modeling on the operation action decomposition data to generate an operation action feature template set; Step S243: configuring motion capture points according to the operation motion feature template set, wherein the motion capture point configuration includes joints, gestures, and tool contact points, and obtaining motion capture point definition data; Step S244: Deploy and plan a field capture system based on the motion capture point definition data, and use the motion capture system to perform real-time motion capture and data synchronous collection to generate an original production motion data stream; Step S245: performing action segmentation and feature extraction on the original production action data stream to generate key production action data.

[0036] In this embodiment of the present invention, in step S241, a depth-first search (DFS) algorithm is used to traverse the constructed safety production knowledge graph, identifying and extracting entity nodes and the relationship edges between them in the graph, thereby constructing a complete production operation chain path from the starting operation to the target result. Based on this, each operation chain is process-encoded to generate serially numbered production operation chain sequence data to ensure consistency between the operation sequence and the node timing. In step S242, based on the constructed production operation chain sequence data, the operation action is decomposed based on the entity categories (such as equipment operation, personnel behavior, and environmental monitoring) to which different nodes belong. The complex continuous operation process is broken down into several basic action units, and characteristic parameters such as spatiotemporal posture, mechanical contact, and action duration of each action unit are extracted to generate standardized operation action decomposition data. Node action feature modeling is then performed on the decomposed action units, extracting their key physical features and semantic attributes (such as joint range of motion, posture change trend, and action intention recognition), thereby constructing a set of high-precision, multimodal fusion operation action feature templates. In step S243, based on the constructed set of action feature templates, motion capture points are assigned for various key actions, including human joint positions (such as elbows, knees, and wrists), key hand points (such as the tip of the index finger and the palm), and points of interaction with equipment or tools (such as buttons, valves, and joysticks). This generates motion capture point definition data, providing deployment parameter basis for subsequent action data collection. Next, in step S244, based on the motion capture point definition data, a high-precision motion capture system is deployed at the industrial site. The system utilizes various technologies, such as inertial sensors, optical tracking cameras, and RFID positioning devices, to achieve multi-channel, multi-node synchronous data collection, capturing the operator's dynamic behavior during real-world operations and generating a real-time raw production action data stream containing spatial trajectory, action posture, and interaction events. In the final step S245, the sliding window algorithm and dynamic time warping (DTW) method are used to segment the original production action data stream into time periods and identify action boundaries to achieve accurate segmentation of continuous actions. At the same time, deep feature extraction technology (such as convolutional neural network (CNN) or temporal convolution (TCN)) is used to extract representative key action features from the segmented action segments, ultimately generating refined key production action data that can be used for subsequent risk assessment and modeling.

[0037] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes: Step S31: performing spatial positioning mapping based on key production action data to generate action space mapping data; performing spatial microenvironment parameter perception on the action space mapping data to generate perception results; Step S32: confirming the execution target of the key data of the production action based on the sensing result, wherein the execution target includes a dynamic target and a static target, the dynamic target is the employee, and the static target is the machine; Step S33: Modeling the spatiotemporal behavior trajectory of the dynamic target to generate dynamic behavior trajectory data; modeling the working state of the static target to generate static workflow feature data; Step S34: Using the perception results, perform an environmental modulation behavior-state linkage analysis on the dynamic behavior trajectory data and the static workflow feature data to generate production risk transmission data; Step S35: Input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data.

[0038] In an embodiment of the present invention, based on the aforementioned key production action data, industrial scene modeling and spatial coordinate mapping technology are used to perform three-dimensional spatial positioning and trajectory reconstruction, thereby restoring the position and posture of the work action in the actual site, thereby generating action space mapping data; then, the action space mapping data is subjected to micro-environment parameter perception, including temperature and humidity, noise, light, harmful gas concentration, electromagnetic interference, etc., and the real-time environmental status is obtained using a distributed environmental sensor network, and the action position data is integrated to construct a high-precision spatial micro-environment perception result. In step S32, target attribution analysis is performed on the key production action data based on the perception results to identify the target entity affected by the action, where dynamic targets are on-site workers with behavioral timing and displacement characteristics; static targets are fixed equipment or workbenches with working status or operation response characteristics; the action target type is established through the target recognition mechanism and classified as dynamic targets or static targets, and the action execution object and its attributes are clarified. In step S33, dynamic and static targets are modeled separately. For dynamic targets (e.g., employees), a spatiotemporal behavior trajectory model is constructed to extract information such as their position, posture, speed, and behavior labels during the work process. Dynamic behavior trajectory data is then generated using algorithms such as sliding windows and long-short-term memory (LSTM) networks. For static targets (e.g., equipment), a working state model is constructed to extract operational state indicators (e.g., vibration spectrum, current variation, task execution phase, etc.). Workflow features are then constructed based on working condition rules to generate static workflow feature data. In step S34, based on the aforementioned perception results, an environmentally modulated behavior-state linkage analysis model is used to fuse and analyze the dynamic behavior trajectory data with the static workflow feature data. For example, this model analyzes the impact of elevated temperature and humidity on the accuracy of human behavior, or the probability of misoperation caused by equipment response delays in high-noise environments. This identifies potential collaborative risk mechanisms caused by environmental parameters and generates production risk transmission data reflecting the linkage between behavior and state. In step S35, the above-mentioned production risk transmission data is input into a pre-trained safety risk assessment model. The model is constructed based on graph neural networks, multi-dimensional feature fusion and Bayesian reasoning mechanism. It can perform cross-analysis and risk deduction from four dimensions: personnel, equipment, environment and system. It outputs a quantitative safety risk score under the current operation scenario, and combines the risk evolution model to predict its trend changes. Finally, the safety risk quantitative score and risk evolution trend prediction data are obtained for subsequent safety level classification and early warning strategy formulation.

[0039] Preferably, step S34 includes the following steps: The perception results are used to quantify the spatial position disturbance of the dynamic behavior trajectory data, extract the trajectory offset vector, and generate dynamic trajectory disturbance vector data; Calculate the variability of workstation nodes on the static workflow feature data to generate static workstation variability index data; Based on the dynamic trajectory disturbance vector data and the static workstation volatility index data, trajectory-node mapping analysis is performed to generate trajectory-node mapping relationship data; Perform dynamic contact frequency statistics and load superposition analysis on trajectory node mapping relationship data, extract high-frequency contact nodes and their load change rates, and generate high-risk node warning data; Perform fault pattern recognition on high-risk node warning data, extract node failure pattern characteristics, and generate node failure risk factor data; Based on the node failure risk factor data, time-series link deduction is performed to construct a risk transmission path diagram, and finally generate production risk transmission data.

[0040] In an embodiment of the present invention, a sensor fusion positioning algorithm is used to quantify spatial position disturbances in dynamic behavior trajectory data based on the aforementioned spatial microenvironment perception results. This process primarily analyzes the spatial offset between the operator's ideal trajectory and the actual execution trajectory. Key offset parameters are extracted by constructing a trajectory residual model, generating dynamic trajectory disturbance vector data representing the deviation in action execution. Each vector element describes the direction and magnitude of the offset at a specific moment or node. Subsequently, static workflow feature data is analyzed. Using a workstation node disturbance sensitivity calculation method, combined with equipment response delay, error tolerance, and historical fault records, the stability threshold and response volatility of each workstation node under external disturbances are calculated. This generates static workstation volatility index data, which quantifies the volatility of each workstation node to external action disturbances. Based on this, trajectory-node mapping analysis is performed. By projecting the dynamic trajectory disturbance vector onto the static workstation spatial layout, the static nodes affected by the disturbance behavior in the spatiotemporal dimensions are identified, generating trajectory-node mapping relationship data. This mapping not only considers geometric distance but also incorporates historical intervention data and the coupling strength of node functions, improving mapping accuracy. Dynamic contact frequency statistics and load superposition analysis are performed on the mapping relationship data. The high-frequency impact points and superimposed load trends are analyzed, and then high-frequency contact nodes and their load change rates are extracted. Warning data for high-risk nodes is output. These nodes are frequently disturbed and experience abnormal load changes, posing potential failure risks. Furthermore, fault pattern recognition is performed on the high-risk node warning data. Combining existing node function models and failure case libraries, machine learning methods (such as support vector machines or fault tree analysis) are used to identify typical failure modes faced by nodes. Failure trends, causal paths, and response characteristics that characterize their failure behaviors are extracted, ultimately forming node failure risk factor data. Finally, based on these risk factors, temporal link deduction is performed to construct causal dependencies and time evolution paths between nodes. This comprehensive visualization of the risk transmission path diagram is formed, clarifying the transmission relationship and triggering pattern between high-risk nodes, and ultimately generating production risk transmission data that reflects the overall operation risk evolution mechanism.

[0041] Preferably, step S35 includes the following steps: Step S351: Sample construction and label matching are performed on the production risk transmission data to generate risk training sample data; the risk training sample data is divided into data sets to generate model training sets and model test sets; the production risk transmission data is input into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; Step S352: Using a convolutional neural network algorithm to perform model training on the model training set, thereby generating a security risk assessment pre-model; performing model optimization iteration on the security risk assessment pre-model using the model test set, thereby generating a pre-trained security risk assessment model; Step S353: Input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data, where the multi-dimensional analysis includes safety risk quantification and time series trend prediction.

[0042] In this embodiment of the present invention, risk training sample data with input features and target labels is generated by encoding the nodes, edges, and their attribute features in the risk transmission path and annotating them with labels based on known risk levels and evolution processes from a historical accident case database. Next, the constructed risk training sample data is partitioned into a dataset according to a set ratio (e.g., 8:2 or 7:3) to generate a model training set and a model test set, respectively, to meet the needs of model training and validation. In step S352, a convolutional neural network (CNN) algorithm is used to train the model training set. By setting appropriate convolution kernel sizes, network depths, and activation functions, spatial dependencies and feature maps in the production risk transmission data are extracted to generate an initial safety risk assessment pre-model. During the training process, a loss function such as cross-entropy or mean squared error is used to measure the deviation between the predicted output and the true label, and the network weights are continuously optimized through backpropagation and gradient descent algorithms. Subsequently, the model test set is used to perform performance evaluation and error feedback on the safety risk assessment pre-model. The network structure and hyperparameters are fine-tuned based on the test results, and model optimization iterations are performed. Ultimately, a pre-trained safety risk assessment model with enhanced generalization and higher prediction accuracy is obtained. In step S353, the actual production risk transmission data generated is input into the trained safety risk assessment model for multi-dimensional analysis. This analysis consists of two core components: first, safety risk quantification. Based on the input risk data characteristics, the model outputs a risk level score for the current operation scenario or operation node (e.g., a continuous value between 0 and 1, or a five-level segmented rating). Second, time series trend prediction. By incorporating a time-dimensional modeling mechanism (e.g., a temporal convolutional layer or a CNN model nested within an LSTM layer), combined with historical risk evolution data, the model predicts future risk trends and outputs risk evolution trend prediction data, such as a curve showing the likelihood of risk increasing, remaining flat, or decreasing. The resulting safety risk quantification score and risk evolution trend prediction data will serve as a crucial basis for the company's subsequent safety situation analysis and early warning decisions.

[0043] Preferably, step S4 includes the following steps: Step S41: divide the security risk quantitative score data into intervals according to the set dynamic division rules to generate preliminary security level division data; Step S42: Performing a time sequence consistency check on the preliminary security level classification data, removing abnormal points of classification mutations, and generating revised security level classification data; Step S43: Perform visual mapping processing based on the revised security level classification data, draw by level nodes, and generate a hierarchical map of enterprise security situation; Step S44: Perform sliding window differential analysis on the risk evolution trend prediction data, extract short-term accelerated risk events, and generate potential risk acceleration event data; compare the potential risk acceleration event data with the set warning threshold data to generate corresponding graded warning instructions.

[0044] In an embodiment of the present invention, quantitative safety risk score data is segmented according to pre-defined dynamic segmentation rules (e.g., based on quantile segmentation, K-means clustering, or empirical risk level intervals). The score data is divided into multiple risk level intervals (e.g., Level I - safe, Level II - low risk, Level III - medium risk, Level IV - high risk, and Level V - severe risk), thereby generating corresponding preliminary safety level classification data. The segmentation rules can be adaptively adjusted, such as dynamically setting interval boundaries to accommodate shifts in the risk score distribution. The generated preliminary safety level classification data is then subjected to time series validation. A sliding window detection mechanism and local fitting methods (e.g., LOESS smoothing, moving average, or local volatility detection) are used to identify and remove short-term abnormal jump points or discrete sudden abnormal level nodes, thereby enhancing the stability and interpretability of the classification. This process can also be combined with risk evolution trend data for trend consistency analysis to ensure that the level transition conforms to the actual evolution logic. The revised safety level classification data is graphically represented and multi-dimensional node mapping is performed using graph theory visualization tools (e.g., force-directed layout, heat map grid layout, etc.). The map uses the "enterprise spatial structure" or "process flow path" as its framework, coloring nodes by security level (e.g., green for safe, red for high risk) to create a hierarchical map of enterprise security posture. It supports dynamic updates, hierarchical aggregation, and risk heat projection, enhancing visualization and decision support. Sliding window differential analysis is performed on risk evolution trend forecast data to identify sharp increases in risk scores over short periods of time, generating data on potential risk acceleration events. This analysis employs methods such as linear fitting within the window and acceleration calculations (e.g., first-order and second-order differences) to identify nodes with sudden increases. The potential risk acceleration event data is then compared with pre-set tiered warning thresholds (e.g., each tier corresponds to a different rate of increase threshold) to determine whether an alert is triggered. If the triggering conditions are met, specific tiered warning instructions are generated based on the risk level, event location, and degree of acceleration, indicating the response level and recommended actions.

[0045] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0046] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A digital evaluation method for enterprise safety production, characterized in that: The following steps are involved: Step S1: Acquire enterprise safety production multi-source monitoring data, including equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data, and perform data cleaning and multimodal fusion processing to generate a safety production fusion data set; Step S2: Extract risk features of the safety production fusion dataset to construct a safety production knowledge graph; Analyze the production operation chain of the safety production knowledge graph, and capture operation actions based on the production operation chain to generate key production action data; Step S3: Perform spatial microenvironment perception based on key production action data, and conduct risk transmission on the key production action data based on the perception results to generate production risk transmission data; Input production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; Step S4: Dynamically divide security levels based on the security risk quantitative score to generate a hierarchical map of enterprise security situation; compare warning thresholds based on risk evolution trend prediction data to generate graded warning instructions; Step S5: Output the safety situation classification map and classification warning instructions through the visual interactive platform, and dynamically update the enterprise safety production knowledge base to complete the safety production digital evaluation closed loop.

2. The digital evaluation method for enterprise safety production according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: collecting equipment vibration spectrum, temperature and pressure data, and energy consumption curve in real time through the industrial Internet of Things terminal to generate equipment operation status data; Step S12: using a distributed sensor network to obtain environmental temperature and humidity, toxic gas concentration, and fire monitoring data to generate environmental monitoring data; Step S13: Collect personnel location trajectory, operation specification data and labor protection equipment status through smart wearable devices and video surveillance systems to generate personnel behavior data; Step S14: Perform natural language processing on the enterprise safety management system document, extract safety management elements and construct standard compliance feature vectors, thereby obtaining safety management system data; Step S15: Perform spatiotemporal fusion of equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data to generate a safety production fusion data set.

3. The digital evaluation method for enterprise safety production according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: Graph structure modeling is performed on the safety production fusion dataset to construct an initial knowledge graph with equipment nodes, environment nodes, personnel nodes, and management nodes as entities; Step S22: Extract potential risk transmission relationships between nodes in the initial knowledge graph through a graph neural network to generate a risk propagation path graph; perform risk pattern matching on the risk propagation path graph based on a preset historical accident case library, annotate risk feature types and severity levels, and generate production safety risk features; Step S23: Optimize the initial knowledge graph according to the production safety risk characteristics to obtain a production safety knowledge graph; Step S24: Analyze the production operation chain of the safety production knowledge graph, and capture operation actions based on the production operation chain to generate key data of production actions.

4. The digital evaluation method for enterprise safety production according to claim 3 is characterized in that: Step S24 includes the following steps: Step S241: Perform a depth-first search on the safety production knowledge graph, traverse and extract the nodes and relationships in the graph, and obtain a production operation chain; sequence the production operation chain to obtain production operation chain sequence data; Step S242: Decomposing the production operation chain based on the production operation chain sequence data to obtain operation action decomposition data; performing node action feature modeling on the operation action decomposition data to generate an operation action feature template set; Step S243: configuring motion capture points according to the operation motion feature template set, wherein the motion capture point configuration includes joints, gestures, and tool contact points, and obtaining motion capture point definition data; Step S244: Deploy and plan a field capture system based on the motion capture point definition data, and use the motion capture system to perform real-time motion capture and data synchronous collection to generate an original production motion data stream; Step S245: performing action segmentation and feature extraction on the original production action data stream to generate key production action data.

5. The digital evaluation method for enterprise safety production according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: performing spatial positioning mapping based on key production action data to generate action space mapping data; performing spatial microenvironment parameter perception on the action space mapping data to generate perception results; Step S32: confirming the execution target of the key data of the production action based on the sensing result, wherein the execution target includes a dynamic target and a static target, the dynamic target is the employee, and the static target is the machine; Step S33: Modeling the spatiotemporal behavior trajectory of the dynamic target to generate dynamic behavior trajectory data; modeling the working state of the static target to generate static workflow feature data; Step S34: Using the perception results, perform an environmental modulation behavior-state linkage analysis on the dynamic behavior trajectory data and the static workflow feature data to generate production risk transmission data; Step S35: Input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data.

6. The digital evaluation method for enterprise safety production according to claim 5, characterized in that: Step S34 includes the following steps: The perception results are used to quantify the spatial position disturbance of the dynamic behavior trajectory data, extract the trajectory offset vector, and generate dynamic trajectory disturbance vector data; Calculate the variability of workstation nodes on the static workflow feature data to generate static workstation variability index data; Based on the dynamic trajectory disturbance vector data and the static workstation volatility index data, trajectory-node mapping analysis is performed to generate trajectory-node mapping relationship data; Perform dynamic contact frequency statistics and load superposition analysis on trajectory node mapping relationship data, extract high-frequency contact nodes and their load change rates, and generate high-risk node warning data; Perform fault pattern recognition on high-risk node warning data, extract node failure pattern characteristics, and generate node failure risk factor data; Based on the node failure risk factor data, time-series link deduction is performed to construct a risk transmission path diagram, and finally generate production risk transmission data.

7. The digital evaluation method for enterprise safety production according to claim 5 is characterized in that: Step S35 includes the following steps: Step S351: Sample construction and label matching are performed on the production risk transmission data to generate risk training sample data; the risk training sample data is divided into data sets to generate model training sets and model test sets; the production risk transmission data is input into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; Step S352: Using a convolutional neural network algorithm to perform model training on the model training set, thereby generating a security risk assessment pre-model; performing model optimization iteration on the security risk assessment pre-model using the model test set, thereby generating a pre-trained security risk assessment model; Step S353: Input the production risk transmission data into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data, where the multi-dimensional analysis includes safety risk quantification and time series trend prediction.

8. The digital evaluation method for enterprise safety production according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: divide the security risk quantitative score data into intervals according to the set dynamic division rules to generate preliminary security level division data; Step S42: Performing a time sequence consistency check on the preliminary security level classification data, removing abnormal points of classification mutations, and generating revised security level classification data; Step S43: Perform visual mapping processing based on the revised security level classification data, draw by level nodes, and generate a hierarchical map of enterprise security situation; Step S44: Perform sliding window differential analysis on the risk evolution trend prediction data, extract short-term accelerated risk events, and generate potential risk acceleration event data; compare the potential risk acceleration event data with the set warning threshold data to generate corresponding graded warning instructions.

9. A digital evaluation system for enterprise safety production, characterized in that: For executing the digital evaluation method for enterprise safety production according to claim 1, the digital evaluation system for enterprise safety production comprises: The data fusion module is used to obtain multi-source monitoring data on enterprise production safety, including equipment operation status data, environmental monitoring data, personnel behavior data, and safety management system data, and perform data cleaning and multimodal fusion processing to generate a production safety fusion data set; The production monitoring module is used to extract risk features from the safety production fusion data set to build a safety production knowledge graph; analyze the production operation chain of the safety production knowledge graph, and capture operation actions based on the production operation chain to generate key production action data; The risk transmission module is used to perceive the spatial microenvironment based on key production action data, and based on the perception results, conduct risk transmission on the key production action data to generate production risk transmission data. The production risk transmission data is input into the pre-trained safety risk assessment model for multi-dimensional analysis to obtain safety risk quantitative scores and risk evolution trend prediction data; The production early warning module is used to dynamically divide security levels based on the quantitative safety risk scores and generate a hierarchical map of the enterprise security situation; it compares early warning thresholds based on risk evolution trend prediction data and generates graded early warning instructions; The visualization module is used to output the security situation classification map and graded warning instructions through the visual interactive platform, and dynamically update the enterprise's production safety knowledge base to complete the digital evaluation closed loop of production safety.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the digital evaluation method for enterprise safety production as described in any one of claims 1 to 8 is implemented.

Images