Safety critical system risk monitoring method and system based on multi-model analysis

Abstract

The invention discloses a multi-model analysis-based risk monitoring method and system for a safety critical system. The method comprises the following steps: acquiring physiological parameters, behavior tracks, machine operation parameters and environmental parameters of an operator in real time through intelligent risk monitoring equipment; the method comprises the following steps: constructing a database with real-time, multi-mode and multi-diversity characteristics, storing and converting historical data and virtual simulation data of a safety-critical system, preprocessing real-time risk data of people, machines and environments, and generating input parameters based on the historical data, the virtual simulation data and the preprocessed real-time risk data; based on a Bayesian network, time sequence analysis, graph calculation and machine learning, constructing a risk judgment model of the safety critical system, and calculating an overall risk value through input parameters; and according to the risk value, judging a risk grade and a risk type, and carrying out visual early warning and triggering a grading response measure. According to the invention, the risk of the safety-critical system can be dynamically and comprehensively monitored in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of risk monitoring, and in particular to a method and system for monitoring risk of safety-critical systems based on multi-model analysis. Background Art

[0002] With the rapid development of science and technology and the continuous advancement of industrial automation, the reliability and safety of safety-critical systems are becoming increasingly important. In many key industries such as chemical engineering, nuclear energy, aerospace, and transportation, safety-critical systems are responsible for ensuring the safety of personnel, the normal operation of equipment, and the protection of the environment.

[0003] Currently, existing risk monitoring methods for safety-critical systems mainly rely on a single monitoring model, which can often only conduct risk assessments on a specific aspect and lacks a comprehensive consideration of the overall system risk. For example, some methods only focus on the operating parameters of the equipment, while ignoring the impact of the operator's physiological and behavioral factors and environmental conditions on system risks; other methods focus on environmental monitoring but fail to fully consider the risks that may arise during human-computer interaction. In addition, traditional risk monitoring methods are mostly based on static analysis of historical data, which makes it difficult to capture dynamic changes and potential risk trends during system operation in real time. When faced with complex and changing actual working conditions, this single, static risk monitoring method often seems powerless and cannot accurately and promptly detect and warn of potential safety hazards, which may lead to serious accidents and losses.

[0004] Therefore, how to comprehensively consider multiple factors such as people, machines, and the environment to build a method that can monitor safety-critical system risks in real time, dynamically, and comprehensively has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the existing technology in that the risk monitoring means are single and static, and to provide a method and system that can monitor the risks of safety-critical systems in real time, dynamically and comprehensively.

[0006] To achieve the above objectives, the present invention provides a first aspect of a safety-critical system risk monitoring method based on multi-model analysis, comprising the following steps: S1. Collect risk data on people, machines, and the environment in real time through intelligent risk monitoring equipment. Human risk data includes operator physiological parameters and behavioral trajectories, machine risk data includes machine operating parameters, and environmental risk data includes environmental parameters. S2. Build a real-time, multimodal, and multivariate database to store and transform historical and virtual simulation data of safety-critical systems and pre-process real-time risk data for humans, machines, and the environment. Generate input parameters based on the historical, virtual simulation, and pre-processed real-time risk data. S3. Based on Bayesian networks, time series analysis, and machine learning, a safety-critical system risk assessment model is constructed, which is integrated with three sub-models: a human-machine interaction risk assessment model, a human-environment interaction risk assessment model, and a machine-environment interaction risk assessment model. The overall risk value of the safety-critical system is calculated using the input parameters. S4. Based on the overall risk value output by the risk assessment model, determine the risk level and risk type, and perform visual warnings and trigger graded system response measures based on the risk level.

[0007] Preferably, in step S1: Risk data on individuals is collected through wearable devices and video surveillance equipment in the intelligent risk monitoring device. The wearable devices integrate multiple types of physiological sensors to monitor changes in a person's physiological parameters, including brain waves, eye movement trajectories, electrocardiograms, heart rates, and body temperature. The video surveillance equipment integrates a video surveillance algorithm to synchronously monitor a person's behavior trajectories and generate an operation log. The risk data of the machine is collected through the machine operation sensors in the intelligent risk monitoring device, including vibration, temperature, pressure, flow, voltage and current sensors; Environmental risk data is collected by environmental monitoring sensors in the intelligent risk monitoring device.

[0008] Preferably, the database further includes an integration module, which performs timestamp alignment and format conversion on the risk data of the person, machine, and environment, eliminates redundant information, applies a filtering algorithm to reduce noise on the data of physiological parameters, and normalizes the data of environmental parameters.

[0009] Preferably, the real-time nature in step S2 means that all data are uploaded to the database in real time for storage and processing via wireless network or wired transmission; Multimodality means that the database supports a variety of data types and can meet the unified storage, management and query of structured and unstructured multimodal data; The diversity means that the database is configured with a historical data interface and a virtual simulation data interface, which can import the system's historical data, maintenance records, life cycle, and virtual simulation data of people, machines, and environments in special operating scenarios into the database.

[0010] Preferably, the storage of the multimodal data is implemented through a relational database and a distributed file system. The structured data includes operator physiological parameter tables, machine operation parameter tables, maintenance record tables, and environmental parameter tables. The unstructured data includes video surveillance files, operation log texts, and virtual simulation 3D scene files. The database supports cross-modal joint queries through extended syntax of the structured query language. If the historical data comes from the intelligent risk monitoring device, the historical data interface is connected to the intelligent risk monitoring device through the API to capture the historical data regularly or on demand. If the historical data is stored in an external file, the batch import tool is used to import the historical data into the database; The virtual simulation interface is connected to the data output interface of the simulation platform, and stores the physiological parameters, behavioral trajectories, machine operation parameters and environmental parameters of the virtual operator under various working conditions into a database.

[0011] Preferably, the human-machine interaction risk judgment model defines human nodes and machine nodes based on the Bayesian network according to the input parameters of the human and the machine, determines the conditional probability distribution of the two nodes through training through the input parameters, predicts the short-term changes in the machine operating parameters and the human fatigue trend through time series analysis, combines machine learning to identify the abnormal operation mode in the operator's behavior trajectory, quantifies the risk of human error, identifies the impact of human operation standardization on machine stability, predicts the risk of machine failure caused by human factors, and constructs and trains a classification model through machine learning to identify risk patterns.

[0012] Preferably, the human-environment interaction risk judgment model defines human nodes and environmental nodes based on the Bayesian network according to the input parameters of human and environment, and establishes a causal chain of various factors in the risk data of human and environment. Combined with historical data, it quantifies the joint probability of human operation errors under specific environmental parameter conditions, uses time series analysis to predict the changing trend of environmental parameters and the changes in physiological parameters of people in specific environments, and constructs and trains classification models through machine learning to identify risk patterns.

[0013] Preferably, the machine-environment interaction risk judgment model defines machine nodes and environment nodes based on the Bayesian network according to the input parameters of the machine and the environment, determines the conditional probability distribution of the two nodes through training through the input parameters, predicts the trend of machine operating parameters under changes in environmental parameters through time series analysis, and constructs and trains a classification model through machine learning to identify risk patterns.

[0014] Preferably, the safety-critical system risk determination model constructs a human-machine-environment relationship graph based on a graph computing unit, and analyzes the dependency relationships and risk propagation paths between nodes.

[0015] To achieve the objectives of the present invention, a second aspect provides a safety-critical system risk monitoring system based on multi-model analysis, applying the safety-critical system risk monitoring method based on multi-model analysis described in the above technical solution, the system comprising: The risk perception layer is used to collect risk data on humans, machines, and the environment in real time through intelligent risk monitoring equipment. Human risk data includes the physiological parameters and behavioral trajectories of operators, machine risk data includes machine operating parameters, and environmental risk data includes environmental parameters. The information storage and conversion layer is used to build a database with real-time, multimodal and multivariate characteristics, store and convert historical data and virtual simulation data of safety-critical systems, pre-process real-time risk data of humans, machines and the environment, and generate input parameters based on historical data, virtual simulation data and pre-processed real-time risk data; a risk decision layer, configured to construct a safety-critical system risk assessment model based on Bayesian networks, time series analysis, and machine learning, which is an integration of three sub-models: a human-machine interaction risk assessment model, a human-environment interaction risk assessment model, and a machine-environment interaction risk assessment model, and calculate the overall risk value of the safety-critical system using the input parameters; The risk warning and response layer is used to determine the risk level and risk type based on the overall risk value output by the risk judgment model, and to provide visual warnings and trigger graded system response measures based on the risk level.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention collects risk data on humans, machines, and the environment through intelligent risk monitoring equipment, develops a database with real-time, multimodal, and multivariate characteristics for data storage and conversion, and uses system science theory combined with Bayesian networks, time series analysis, and machine learning to construct a comprehensive and integrated safety-critical system risk determination model. This model comprehensively considers the interaction factors between humans, machines, and the environment, and implements intelligent risk warning and response based on the model calculation results, which can monitor safety-critical system risks in real time, dynamically, and comprehensively. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a method for monitoring safety-critical system risks based on multi-model analysis provided in an embodiment of the present application; Figure 2 Schematic diagram of the intelligent risk monitoring device provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a safety-critical system risk monitoring system based on multi-model analysis provided in an embodiment of the present application; Figure 4A schematic diagram of some steps of the safety-critical system risk determination model provided in an embodiment of the present application for performing risk determination and trend prediction based on risk data.

[0018] Description of reference numerals: 201—Environmental parameters, 202—Environmental monitoring sensors, 203—Machine operation parameters or operator behavior trajectories, 204—Machine operation sensors, 205—Video monitoring equipment, 206—Wearable devices, 207—Human physiological parameters, 208—Wearable device processor, 209—Database. DETAILED DESCRIPTION

[0019] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0020] Example 1 See also Figure 1 This embodiment provides a safety-critical system risk monitoring method based on multi-model analysis, including the following steps: S1. Collect risk data on people, machines, and the environment in real time through intelligent risk monitoring equipment. Human risk data includes operator physiological parameters and behavioral trajectories, machine risk data includes machine operating parameters, and environmental risk data includes environmental parameters. S2. Build a real-time, multimodal, and multivariate database to store and transform historical and virtual simulation data of safety-critical systems and pre-process real-time risk data for humans, machines, and the environment. Generate input parameters based on the historical, virtual simulation, and pre-processed real-time risk data. S3. Based on Bayesian networks, time series analysis, graph computing, and machine learning, a safety-critical system risk assessment model is constructed, which is integrated with three sub-models: a human-machine interaction risk assessment model, a human-environment interaction risk assessment model, and a machine-environment interaction risk assessment model. The overall risk value of the safety-critical system is calculated using the input parameters. S4. Based on the overall risk value output by the risk assessment model, determine the risk level and risk type, and perform visual warnings and trigger graded system response measures based on the risk level.

[0021] like Figure 2As shown, in step S1, human risk data is collected by the wearable device 206 and video surveillance device 205 within the intelligent risk monitoring system. The wearable device integrates multiple physiological sensors, enabling real-time monitoring of physiological and emotional changes in the operator, including brain waves, eye movements, electrocardiogram (ECG), heart rate, and body temperature. Combined with the video surveillance system, this allows for comprehensive monitoring of the operator's physiological parameters and behavioral trajectories, as well as the entire process of machine operation. The video surveillance system integrates a video monitoring algorithm to simultaneously monitor human behavior, particularly machine operation, and generate an operation log. Machine risk data is collected by the machine operation sensors 204 within the intelligent risk monitoring system, including vibration, temperature, pressure, flow, voltage, and current sensors. Environmental risk data is collected by the environmental monitoring sensors 202 within the intelligent risk monitoring system. The intelligent risk monitoring system enables more comprehensive monitoring of system risk data from the three perspectives of human, machine, and environment, thereby improving system reliability.

[0022] In step S2, the real-time, multimodal, and multivariate characteristics of the database specifically include: The real-time feature means that all data is uploaded to the database in real time for storage and processing via wireless network or wired transmission. The database has the function of real-time data processing and analysis, supports the capture, processing and analysis of real-time data, and provides timely and accurate data input for the risk decision-making model. The multimodal feature means that the database supports a variety of data types and can meet the unified management and query of structured, unstructured and other multimodal data to achieve seamless integration and efficient use of data. The diversity feature represents the configuration of historical data and virtual simulation data interfaces in the database, which can import the system's historical data, maintenance records, life cycle and virtual simulation data of people, machines and environments in special operating scenarios into the database, making the source of risk data richer and providing more comprehensive data support for the risk judgment model of safety-critical systems.

[0023] The database also includes an integration module, which aligns timestamps and converts the format of risk data of people, machines, and the environment, eliminates redundant information, applies filtering algorithms to reduce noise on physiological parameter data, and normalizes environmental parameter data.

[0024] The human-machine interaction risk determination model in step S3 defines human nodes and machine nodes based on the Bayesian network according to the input parameters of humans and machines, determines the conditional probability distribution of the two nodes through training through the input parameters, predicts the short-term changes in machine operating parameters and human fatigue trends through time series analysis, combines machine learning to identify abnormal operating modes in the operator's behavior trajectory, quantifies the risk of human error, identifies the impact of human operation standardization on machine stability, predicts the risk of machine failure caused by human factors, and constructs and trains a classification model through machine learning to identify risk patterns.

[0025] The human-environment interaction risk assessment model uses Bayesian networks to define human and environmental nodes based on input parameters of both individuals and the environment, and establishes a causal chain between various factors in the risk data for both individuals and the environment. Integrating historical data, it quantifies the joint probability of human operational errors under specific environmental parameters. It uses time series analysis to predict trends in environmental parameters and changes in human physiological parameters under specific circumstances. It then constructs and trains a classification model through machine learning to identify risk patterns. Its function is to assess the direct impact of the environment on human health and decision-making, provide early warning of physiological abnormalities or behavioral disturbances caused by environmental deterioration, and help optimize the design of the work environment.

[0026] The machine-environment interaction risk assessment model uses a Bayesian network to define machine and environment nodes based on machine and environment input parameters. Training with the input parameters determines the conditional probability distribution of the two nodes. Time series analysis is used to predict trends in machine operating parameters under changing environmental parameters. Machine learning is then used to construct and train a classification model to identify risk patterns. Its function is to assess the impact of environmental factors on machine operational stability.

[0027] The safety-critical system risk assessment model also constructs a human-machine-environment relationship map based on graph computing units to analyze the dependencies and risk propagation paths between nodes.

[0028] In the past, environmental factors were considered in isolation in risk monitoring and assessment work, and the contribution of "human-environment interaction" and "machine-environment interaction" to the overall risk of the system was not fully recognized, which easily led to errors in the quantification of risks.

[0029] In step S4, the visual risk warning interface includes real-time data charts, risk types and levels, and warning information. This interface is interactive, allowing users to interact with the data by clicking and dragging, gaining a deeper understanding of detailed information. It also features alarm and notification functions. When the overall risk value exceeds a threshold or reaches an alarm condition, it automatically sends an alert to relevant personnel, helping them to quickly take risk mitigation measures.

[0030] Example 2 This embodiment 2 provides a safety-critical system risk monitoring system based on multi-model analysis, such as Figure 3 As shown, the method for monitoring risk of a safety-critical system based on multi-model analysis described in Example 1 is applied, and the system includes: The risk perception layer is used to collect risk data on humans, machines, and the environment in real time through intelligent risk monitoring equipment. Human risk data includes the physiological parameters and behavioral trajectories of operators, machine risk data includes machine operating parameters, and environmental risk data includes environmental parameters. The information storage and conversion layer is used to build a database with real-time, multimodal and multivariate characteristics, store and convert historical data and virtual simulation data of safety-critical systems, pre-process real-time risk data of humans, machines and the environment, and generate input parameters based on historical data, virtual simulation data and pre-processed real-time risk data; a risk decision layer, configured to construct a safety-critical system risk assessment model based on Bayesian networks, time series analysis, and machine learning, which is an integration of three sub-models: a human-machine interaction risk assessment model, a human-environment interaction risk assessment model, and a machine-environment interaction risk assessment model, and calculate the overall risk value of the safety-critical system using the input parameters; The risk warning and response layer is used to determine the risk level and risk type based on the overall risk value output by the risk judgment model, and to provide visual warnings and trigger graded system response measures based on the risk level.

[0031] like Figure 3 As shown in the figure, the risk perception layer collects and aggregates data from various sources, including human risk data, machine risk data, and environmental risk data (environmental factors). Human risk data can be collected through wearable devices such as biosensors and motion sensors, which monitor a person's physiological state and behavioral trajectory in real time. Video surveillance equipment can also be used to monitor a person's behavioral trajectory and identify illegal operations or abnormal behavior through visual algorithms. Machine risk data collection includes vibration signals, temperature, pressure, flow rate, current, voltage, and other information. Environmental factor data is collected by deploying environmental sensors or monitoring devices to obtain information on changes in temperature, humidity, air pressure, hazardous gas concentrations, noise levels, and light intensity.

[0032] The information storage and conversion layer integrates, cleans, and models the various data obtained in step S1 (data storage, data conversion, and data pre-processing), providing accurate and effective input for subsequent risk assessment. Historical operational data includes previous operating records, accident data, maintenance and repair records, and historical environmental data curves. Real-time monitoring data obtained in S1 is aggregated to form a current state parameter set. For additional system assessments in data-deficient or extreme scenarios, supplementary virtual simulation data is generated through virtual simulation. These three types of data are aggregated in a database to generate input parameters for the safety-critical system risk assessment model.

[0033] The data analysis modules in the risk decision layer include machine learning units, graph computing units, time series analysis units, and Bayesian network units. The machine learning unit uses classification models such as random forests and LSTM neural networks to train and identify risk patterns. The graph computing unit constructs a "human-machine-environment" relationship map to analyze inter-node dependencies and risk propagation paths. The time series analysis unit uses algorithms such as ARIMA to predict equipment failure cycles or environmental parameter trends. The Bayesian network unit calculates the joint probability of multiple factors and infers the probability of risk occurrence in complex scenarios. These three models are combined to construct a comprehensive safety-critical system risk assessment model consisting of a "human-machine interaction risk assessment model," a "human-environment interaction risk assessment model," and a "machine-environment interaction risk assessment model."

[0034] The risk warning and response layer divides the risk level into multiple levels (such as low risk, medium risk, and high risk) based on the calculation results of the risk judgment model in S3, and can automatically identify and manage them in a hierarchical manner according to pre-set thresholds or evaluation standards; when the judgment result reaches or exceeds the pre-set risk threshold, the system automatically issues a risk warning signal and can implement corresponding disposal measures according to the emergency plan, including but not limited to personnel evacuation, equipment shutdown, alarm notification, safety isolation, etc.

[0035] The safety-critical system risk determination model is pre-trained by using historical data and virtual simulation data as a training set. The pre-trained safety-critical system risk determination model determines the risk level, and the processing results are stored in a database.

[0036] Example 3 This Example 3 is based on Example 1 and Example 2 to further illustrate the specific implementation of the present invention.

[0037] In a specific embodiment, monitoring data sources include sensors, monitoring equipment, and wearable devices worn by workers. All data is uploaded to a database in real time via wireless or wired transmission for storage and processing, achieving real-time performance. Multimodal storage can be implemented through a relational database + distributed file system, where structured data storage includes equipment operating parameter tables and maintenance record tables, and unstructured data storage includes video surveillance clips, operation log text, and virtual simulation 3D scene files. Cross-modal joint queries are supported through the extended syntax of the Structured Query Language. Historical data interface configuration: If historical data comes from a real-time monitoring system, it can be connected to the monitoring system (i.e., intelligent risk monitoring equipment) through an API to capture data regularly or on demand. For example, a REST API can be used to obtain real-time operating data of the equipment and store it in the database. If historical data is stored in external files such as Excel, CSV, or XML, batch import tools such as MySQL's LOAD DATA INFILE and PostgreSQL's COPY command can be used to import the data into the database. The virtual simulation interface can access Unity / ANSYS simulation platform data to store virtual human physiological parameters, behavioral trajectories, machine operating parameters, and environmental parameters under various working conditions in the database, achieving database diversity.

[0038] The schematic diagram of some steps of the risk assessment model for safety-critical systems based on risk data for risk assessment and trend prediction is as follows: Figure 4 shown.

[0039] Risk variables include human-machine interaction risk variables, human-environment interaction risk variables, and machine-environment interaction risk variables. These risk variables are transmitted to the data processing module via local terminals or wireless networks, and then stored and preprocessed in a database to prepare for subsequent Bayesian network analysis. A Bayesian network is constructed and graph-based calculations are performed to describe the dependencies between variables. The overall risk level is calculated by integrating the output nodes of the Bayesian network with the causal relationships between nodes and the conditional probability distribution. To identify the dynamic changes in risk over time, a time series analysis model is introduced to predict risk values in the short or medium to long term. A classification model is constructed and trained through machine learning to identify risk patterns. Figure 4 In the time series analysis diagram, the solid line represents the real-time monitored risk value, and the dashed line represents the risk trend predicted by the model. The system can set one or more thresholds and compare the measured risk value with the predicted risk value. If the measured value is significantly higher than the predicted value, it may indicate a sudden failure or abnormal event. If the measured value is lower than the predicted value, the system security status may be better than expected, but continued monitoring is still required to prevent hidden risks.

[0040] In an application embodiment of a chemical production scenario, a risk variable node is defined, where: Human nodes include : Worker fatigue index; : Operational standardization; : Training level; the above nodes are affected by the physiological parameter analysis of wearable devices, video recognition of illegal operations, and the operation error rate in historical data; Machine nodes include : Reactor temperature; : Vibration amplitude of the stirring motor; the above nodes are monitored and collected in real time by temperature sensors and vibration sensors; The environment nodes include : Toxic gas concentration; : Ambient temperature; the above nodes are determined by gas sensors and temperature and humidity sensors; The dependency relationship between nodes in the Bayesian network has human-computer interaction: When fatigue occurs, The probability of reactor temperature control errors also increases; Human-environment interaction: If workers do not wear protective equipment, they will inhale The risk of toxic gases in the air increases accordingly; Machine-environment interaction: The reaction temperature rises, The ambient temperature also rises, exacerbating the Worker fatigue index. By inputting real-time and historical data, the probability of equipment overheating leading to gas leakage can be obtained.

[0041] Next, we will conduct time series analysis and prediction. Real-time prediction requires updating data every 5 minutes. and PPM is the high-risk threshold; trend forecast is based on ARIMA model analysis Based on historical data, the temperature trend for the next two hours is predicted and compared with the high risk threshold. The corresponding response is made according to the following risk determination logic: Low risk: ppm and , only log; medium risk: Slightly exceeding the limit, the ventilation volume will be automatically increased; High risk: ppm and , triggering an alarm and shutting down; Emergency risks: ppm and , activate the sprinkler system and evacuate personnel.

[0042] In addition, the safety-critical system risk assessment model can be adapted to different industry sectors by adjusting model variables and structures according to different application scenarios, and has good flexibility.

[0043] Visual risk warning can intuitively display monitoring data and risk warning results through graphs, tables, signal lights, flashing icons and text. The risk warning visualization interface of this embodiment includes real-time data charts, risk types and levels, warning information, etc. The visualization interface has interactive functions. Users can interact with the data by clicking, dragging, etc. to gain an in-depth understanding of detailed information. At the same time, the visualization interface has alarm and notification functions. When the risk exceeds the threshold or reaches the alarm condition, it can automatically send an alarm message to the relevant personnel to help them take quick measures to avoid the risk. The use of risk visualization technology can improve the intuitiveness of monitoring data and help improve the accuracy and timeliness of decision-making.

[0044] In summary, the present invention utilizes systems science theory, Bayesian networks, graph computing, machine learning, and time series analysis to construct a comprehensive safety-critical system risk assessment model that integrates a "human-machine interaction risk assessment model," a "human-environment interaction risk assessment model," and a "machine-environment interaction risk assessment model." This model combines historical and virtual simulation data with real-time monitoring data to generate an immediate analysis of system risk trends. This safety-critical system risk assessment model transcends the limitations of single models, no longer assessing "human-machine," "human-environment," and "machine-environment" risks in isolation, but instead captures the chain reaction between these three. Using a Bayesian network to calculate the joint probability of multiple factors, it addresses the quantification error of environmental factor contributions in traditional methods. By integrating real-time monitoring data, historical data, and virtual simulation data, and using time series analysis and machine learning, dynamic risk trend prediction and risk pattern classification are achieved. Leveraging the real-time, multimodal, and multivariate nature of the database, this model supports the unified analysis of structured (device parameters) and unstructured (video surveillance, operation logs), enhancing the comprehensiveness of risk identification. In addition, the model can be adapted to different industry fields by adjusting model variables and structures according to different application scenarios, and has good flexibility.

[0045] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A safety-critical system risk monitoring method based on multi-model analysis, characterized in that: The following steps are involved: S1. Collect risk data on people, machines, and the environment in real time through intelligent risk monitoring equipment. Human risk data includes operator physiological parameters and behavioral trajectories, machine risk data includes machine operating parameters, and environmental risk data includes environmental parameters. S2. Build a real-time, multimodal, and multivariate database to store and transform historical and virtual simulation data of safety-critical systems and pre-process real-time risk data for humans, machines, and the environment. Generate input parameters based on the historical, virtual simulation, and pre-processed real-time risk data. S3. Based on Bayesian networks, time series analysis, graph computing, and machine learning, a safety-critical system risk assessment model is constructed, which is integrated with three sub-models: a human-machine interaction risk assessment model, a human-environment interaction risk assessment model, and a machine-environment interaction risk assessment model. The overall risk value of the safety-critical system is calculated using the input parameters. S4. Based on the overall risk value output by the risk assessment model, determine the risk level and risk type, and perform visual warnings and trigger graded system response measures based on the risk level.

2. The method according to claim 1, characterized in that In the step S1: Risk data on individuals is collected through wearable devices and video surveillance equipment in the intelligent risk monitoring device. The wearable devices integrate multiple types of physiological sensors to monitor changes in a person's physiological parameters, including brain waves, eye movement trajectories, electrocardiograms, heart rates, and body temperature. The video surveillance equipment integrates a video surveillance algorithm to synchronously monitor a person's behavior trajectories and generate an operation log. The risk data of the machine is collected through the machine operation sensors in the intelligent risk monitoring device, including vibration, temperature, pressure, flow, voltage and current sensors; Environmental risk data is collected by environmental monitoring sensors in the intelligent risk monitoring device.

3. The method according to claim 1, characterized in that The database also includes an integration module, which aligns the timestamps and converts the formats of the risk data of people, machines, and the environment, eliminates redundant information, applies a filtering algorithm to reduce noise on the data of physiological parameters, and normalizes the data of environmental parameters.

4. The method according to claim 1, wherein The real-time feature in step S2 means that all data are uploaded to the database in real time for storage and processing via wireless network or wired transmission; The multimodal feature indicates that the database supports a variety of data types and can meet the unified storage, management and query of structured and unstructured multimodal data; The diversity feature indicates that the database is configured with a historical data interface and a virtual simulation data interface, which can import the system's historical data, maintenance records, life cycle, and virtual simulation data of people, machines, and environments in special operating scenarios into the database.

5. The method according to claim 4, characterized in that The storage of the multimodal data is achieved through a relational database and a distributed file system. The structured data includes operator physiological parameter tables, machine operation parameter tables, maintenance record tables, and environmental parameter tables. The unstructured data includes video surveillance files, operation log text, and virtual simulation 3D scene files. The database supports cross-modal joint queries through extended syntax of the structured query language. If the historical data comes from the intelligent risk monitoring device, the historical data interface is connected to the intelligent risk monitoring device through the API to capture the historical data regularly or on demand. If the historical data is stored in an external file, the batch import tool is used to import the historical data into the database; The virtual simulation interface is connected to the data output interface of the simulation platform, and stores the physiological parameters, behavioral trajectories, machine operation parameters and environmental parameters of the virtual operator under various working conditions into a database.

6. The method according to claim 1, characterized in that The human-machine interaction risk assessment model defines human nodes and machine nodes based on a Bayesian network according to the input parameters of humans and machines, determines the conditional probability distribution of the two nodes through training through the input parameters, predicts the short-term changes in machine operating parameters and human fatigue trends through time series analysis, combines machine learning to identify abnormal operating modes in the operator's behavior trajectory, quantifies the risk of human error, identifies the impact of human operating norms on machine stability, predicts the risk of machine failure caused by human factors, and constructs and trains a classification model through machine learning to identify risk patterns.

7. The method according to claim 1, characterized in that The human-environment interaction risk assessment model defines human nodes and environmental nodes based on the Bayesian network according to the input parameters of humans and the environment, and establishes a causal chain of various factors in the risk data of humans and the environment. In combination with historical data, it quantifies the joint probability of human operational errors under specific environmental parameter conditions, uses time series analysis to predict the changing trends of environmental parameters and the changes in physiological parameters of humans in specific environments, and constructs and trains a classification model through machine learning to identify risk patterns.

8. The method according to claim 1, characterized in that The machine-environment interaction risk determination model defines machine nodes and environment nodes based on the Bayesian network according to the input parameters of the machine and the environment, determines the conditional probability distribution of the two nodes through training through the input parameters, predicts the trend of machine operating parameters under changes in environmental parameters through time series analysis, and constructs and trains a classification model through machine learning to identify risk patterns.

9. The method according to any one of claims 6 to 8, characterized in that: The safety-critical system risk determination model constructs a human-machine-environment relationship graph based on a graph computing unit and analyzes the dependency relationships and risk propagation paths between nodes.

10. A safety-critical system risk monitoring system based on multi-model analysis, applying the safety-critical system risk monitoring method based on multi-model analysis according to any one of claims 1 to 9, characterized in that: The system comprises: The risk perception layer is used to collect risk data on humans, machines, and the environment in real time through intelligent risk monitoring equipment. Human risk data includes the physiological parameters and behavioral trajectories of operators, machine risk data includes machine operating parameters, and environmental risk data includes environmental parameters. The information storage and conversion layer is used to build a database with real-time, multimodal and multivariate characteristics, store and convert historical data and virtual simulation data of safety-critical systems, pre-process real-time risk data of humans, machines and the environment, and generate input parameters based on historical data, virtual simulation data and pre-processed real-time risk data; a risk decision layer, configured to construct a safety-critical system risk assessment model based on Bayesian networks, time series analysis, and machine learning, which is an integration of three sub-models: a human-machine interaction risk assessment model, a human-environment interaction risk assessment model, and a machine-environment interaction risk assessment model, and calculate the overall risk value of the safety-critical system using the input parameters; The risk warning and response layer is used to determine the risk level and risk type based on the overall risk value output by the risk judgment model, and to provide visual warnings and trigger graded system response measures based on the risk level.

Images