Intelligent management system for nuclear power plant personnel situation prediction and risk assessment

Abstract

The invention discloses an intelligent management system for nuclear power plant personnel situation prediction and risk assessment, and relates to the field of intelligent safety management systems, and the system comprises a data collection unit, a multi-dimensional situation awareness unit, a risk prediction and assessment unit, an intelligent decision intervention unit and a visual interaction unit. And multi-source data acquisition, real-time situation construction, dynamic risk prediction and evaluation, intelligent early warning intervention and information visualization are realized. According to the invention, real-time monitoring, dynamic risk prediction and intelligent management of the safety state of the operating personnel can be realized, and the defects of real-time monitoring, dynamic prediction and intelligent management of the operating personnel in a high-risk area in the prior art are overcome, so that the safety management level is improved, the life safety is guaranteed, and the accident occurrence probability is reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent safety management systems, specifically to an intelligent management system for predicting the situation and assessing risks of personnel in nuclear power plants. Background Technology

[0002] As the safety management level of nuclear power plants continues to improve, situational awareness and risk assessment for personnel working in high-risk areas have become crucial aspects of ensuring safe production. However, existing technical solutions still fall short in terms of real-time monitoring, dynamic prediction, and intelligent management of personnel working in high-risk areas of nuclear power plants, making it difficult to meet the needs of nuclear power plants for efficient and precise safety management.

[0003] Specifically, current situational awareness technologies face significant challenges in the complex and dynamic high-risk operating environment of nuclear power plants. For example, while existing solutions can achieve macroscopic disaster hazard analysis by integrating multi-source information, their application scenarios and target objects differ fundamentally from the situational awareness of personnel working in high-risk areas of nuclear power plants. These solutions fail to effectively address the dynamic monitoring of worker behavior and refined risk prediction, lacking the ability to accurately assess the safety status of individual workers. Furthermore, in high-risk areas of nuclear power plants, local risk factors such as radiation dose distribution and equipment operating status have a significant impact on worker safety, but existing solutions focus on macroscopic disaster analysis and fail to fully consider the specific risks in these complex environments. This makes them difficult to directly apply to nuclear power plant safety management and unable to provide comprehensive, real-time safety assurance. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent management system for personnel status prediction and risk assessment in nuclear power plants. It aims to address the shortcomings of existing technologies in real-time monitoring, dynamic prediction, and intelligent management of personnel working in high-risk areas of nuclear power plants, thereby achieving real-time monitoring, dynamic risk prediction, and intelligent management of personnel safety status, improving the safety management level of high-risk areas of nuclear power plants, and ensuring the life safety of personnel and the stable operation of nuclear facilities.

[0005] The technical solution of the present invention is an intelligent management system for personnel situation prediction and risk assessment in nuclear power plants, including a data acquisition unit, a multi-dimensional situation awareness unit, a risk prediction and assessment unit, an intelligent decision-making and intervention unit, and a visualization and interaction unit.

[0006] The data acquisition unit is used to acquire in real time physiological status data, environmental parameter data, precise positioning data, behavioral pattern data, nuclear facility and equipment operation status data, and operation procedure data of personnel working in high-risk areas of nuclear power plants.

[0007] The multi-dimensional situation awareness unit is electrically connected to the data acquisition unit and is used to preprocess, extract features, model context, and quantify the multi-source heterogeneous data acquired by the data acquisition unit in order to construct a real-time and comprehensive multi-dimensional situation of the operators.

[0008] The risk prediction and assessment unit is electrically connected to the multi-dimensional situational awareness unit. It is used to dynamically predict individual risks, environmental risks, and operational risks based on the multi-dimensional situational awareness of the workers output by the multi-dimensional situational awareness unit, and to comprehensively assess and generate the comprehensive risk level and confidence level of workers in high-risk areas.

[0009] The intelligent decision-making and intervention unit is electrically connected to the risk prediction and assessment unit, and is used to generate early warning information, decision suggestions or trigger automatic intervention commands based on the comprehensive risk level and confidence level output by the risk prediction and assessment unit.

[0010] The visualization and interaction unit is electrically connected to the multi-dimensional situational awareness unit and the intelligent decision-making and intervention unit, and is used to graphically display the real-time situation, risk prediction results, early warning information and decision suggestions, and provide a human-computer interaction interface.

[0011] In one embodiment of the present invention, the data acquisition unit includes:

[0012] The physiological state sensor array is used to collect physiological parameter data such as heart rate, body temperature, blood oxygen saturation, skin conductance, and fatigue level of workers;

[0013] The environmental parameter sensor group is used to collect environmental parameter data such as radiation dose rate, temperature, humidity, air pressure and air quality in the work area;

[0014] The precise positioning and behavior perception sensor group is used to acquire precise positioning data and behavior pattern data of workers in real time, such as spatial position, movement trajectory, attitude and activity intensity, through ultra-wideband positioning technology, inertial measurement unit technology and near-field communication technology.

[0015] The nuclear facility equipment operation status interface is used to interact with the distributed control system or monitoring and data acquisition system of the nuclear power plant to obtain real-time operating parameters, fault alarms and maintenance records of key equipment such as nuclear reactors, steam generators, cooling systems and electrical systems.

[0016] The work procedure and historical data interface is used to obtain work procedure data and historical data such as digital work permits, standard operating procedures, historical work records, personnel training records, and accident reports.

[0017] As one embodiment of the present invention, the multi-dimensional situational awareness unit includes:

[0018] The data preprocessing module is used to perform time synchronization, spatial calibration, data cleaning, missing value filling, and data standardization on the multi-source heterogeneous data acquired by the data acquisition unit.

[0019] The feature extraction module is used to extract multidimensional features from the preprocessed raw sensor data, including physiological feature indicators, environmental anomaly indicators, behavioral anomaly patterns, and equipment operation trend features.

[0020] The context modeling module is used to build a real-time context model of the operating environment of the operators based on work procedure data, nuclear facility and equipment operating status data and environmental parameter data, and to perform semantic modeling of the relationship between work tasks, equipment status and environmental conditions.

[0021] The situation quantification module is used to fuse the feature-extracted data with the context model to generate a high-dimensional situation vector, which quantifies the comprehensive situation of the operator in terms of current physiology, behavior, environment, equipment, and operational risks.

[0022] The context modeling module uses graph neural networks to semantically model the relationship between work tasks, equipment status, and environmental conditions.

[0023] As one embodiment of the present invention, the risk prediction and assessment unit includes:

[0024] The risk factor identification module is used to monitor abnormal patterns, trend changes, and related events in the multidimensional situation of operators in real time, so as to identify potential risk-inducing factors.

[0025] A predictive model set is used to dynamically predict identified risk factors, the predictive model set comprising:

[0026] Individual risk prediction models are used to predict workers' fatigue levels, mental stress levels, operational error probabilities, and potential health risks based on workers' physiological state data and behavioral pattern data.

[0027] Environmental risk prediction models are used to predict environmental risk events such as changes in radiation levels, leaks of hazardous gases, and the spread of hazardous areas caused by equipment failures, based on environmental parameter data and nuclear facility and equipment operating status data.

[0028] The operational risk prediction model is used to predict potential hazards such as violations of operating procedures, misoperations, and interactions with equipment by operators, based on operational procedure data, operator behavior pattern data, and nuclear facility and equipment operating status data.

[0029] The risk assessment model is used to integrate the prediction results output by the individual risk prediction model, the environmental risk prediction model, and the operational risk prediction model, and combine them with preset risk weights and risk matrices to generate the overall risk level and corresponding confidence level of the current operator.

[0030] The individual risk prediction model uses an attention-based recurrent neural network model to learn the temporal characteristics of the workers' physiological state and behavioral patterns.

[0031] The environmental risk prediction model uses a spatiotemporal graph convolutional network to model and predict the spatiotemporal correlation between environmental parameters and equipment operating status.

[0032] As one embodiment of the present invention, the intelligent decision-making and intervention unit includes:

[0033] The early warning generation module is used to trigger different levels of early warning information when the risk level exceeds a preset threshold, based on the comprehensive risk level and confidence level output by the risk assessment model. These include vibration alerts from the local wearable devices of the workers, alarms from the mobile terminals of the work team leaders, and audible and visual alarms from the central control room.

[0034] The decision recommendation module is used to generate and push the optimal risk avoidance strategy, emergency response plan or work process adjustment suggestions to operators or managers based on a preset expert rule base and reinforcement learning model for specific risk scenarios and early warning information.

[0035] The automatic intervention module is used to trigger an automated safety mechanism in extreme emergency risk situations when the comprehensive risk level reaches the highest preset threshold and the confidence level is high. The safety mechanism includes area isolation, emergency equipment shutdown, broadcast of forced evacuation instructions, and automatic activation of fire protection or cooling systems.

[0036] The feedback and optimization module is used to record the execution status and actual effects of all early warning information, decision suggestions, and automatic intervention instructions, and transmit these feedback data to the risk prediction and assessment unit for continuous training and optimization of the model.

[0037] As one embodiment of the present invention, the visualization and interaction unit includes:

[0038] The real-time situation display module is used to display the precise location, movement trajectory, physiological state trend, environmental parameter distribution, and nuclear facility and equipment operation status of workers on a high-risk area map in real time through two-dimensional or three-dimensional visualization technology, and to identify risk areas with color coding or dynamic icons.

[0039] The risk dashboard module is used to centrally display the comprehensive risk level of operators, various risk prediction curves, historical risk event records, and current warning status, and supports data filtering and analysis by time, region, or personnel.

[0040] The work management interface module provides functions such as digital work ticket management, work task allocation, safety procedure query, and emergency plan invocation, and allows managers to confirm or modify risk warnings and decision suggestions;

[0041] The reporting and retrospective analysis module is used to generate periodic safety reports, incident investigation reports, and performance evaluation reports. It also supports retrospective analysis of historical work processes to identify safety management vulnerabilities and summarize lessons learned.

[0042] Compared with the prior art, the advantages of the present invention are as follows:

[0043] This invention establishes a multi-dimensional data acquisition unit, integrating heterogeneous data such as physiological state, environmental parameters, precise positioning, behavioral patterns, equipment operating status, and work procedures to construct a comprehensive multi-dimensional situational awareness capability for operators. This overcomes the limitations of existing technologies that rely on single data sources or macro-level situational awareness, enabling refined and real-time monitoring of the status of operators in high-risk areas of nuclear power plants. This provides a solid data foundation for subsequent risk prediction and assessment, significantly improving the comprehensiveness and accuracy of situational awareness.

[0044] This invention introduces a risk prediction and assessment unit, utilizing individual risk prediction models, environmental risk prediction models, and operational risk prediction models to dynamically predict various potential risks in the future, and comprehensively assess and generate a comprehensive risk level and confidence level. This innovative design elevates risk management from passive response to proactive prediction, enabling the identification of potential hazards before risk events occur, providing managers with valuable early warning and intervention time, and significantly reducing the probability of accidents and their potential impact.

[0045] The intelligent decision-making and intervention unit provided by this invention can intelligently generate multi-level early warning information, decision suggestions, and even automatic intervention commands based on the predicted risk level and confidence level. This realizes the transformation from traditional manual experience-based decision-making to intelligent, data-driven decision-making, avoiding the subjectivity and lag of human judgment, ensuring the efficiency and accuracy of decision-making in emergency situations, and thus significantly improving the level of intelligent safety management and response efficiency in high-risk areas of nuclear power plants.

[0046] This invention incorporates visualization and interaction units, along with a continuous feedback and optimization mechanism. This enables the system to present complex situational information, risk prediction results, and decision recommendations to users in an intuitive and easy-to-understand manner. Furthermore, it allows for iterative model training and strategy optimization based on actual operational results and expert feedback. This closed-loop self-learning and adaptive capability ensures the system can continuously adapt to the complex and ever-changing safety management needs of nuclear power plants, maintaining its advanced and effective safety management capabilities in high-risk scenarios.

[0047] Combining the above advantages, this invention constructs a closed-loop management system covering the entire chain, from data acquisition and multi-dimensional situational awareness to risk prediction and assessment and intelligent decision-making intervention. This provides comprehensive intelligent protection for the safety of personnel working in high-risk areas of nuclear power plants. This system not only effectively prevents accidents and reduces casualties and property damage, but also optimizes operational processes, enhances the overall safety resilience and operational efficiency of nuclear power plants, and provides strong technical support for the safe development of the nuclear power industry. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the technical architecture of the intelligent management system proposed in this invention;

[0049] Figure 2 This is a schematic diagram of the core principle framework of the risk prediction and assessment unit in this invention;

[0050] Figure 3 This is a logical flow diagram of the multi-dimensional situational awareness unit in this invention;

[0051] Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the intelligent decision-making and intervention unit and external entities in this invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0053] Please refer to the attached document. Figure 1This invention provides an intelligent management system for personnel situation prediction and risk assessment in nuclear power plants. It aims to address the shortcomings of existing technologies in real-time monitoring, dynamic prediction, and intelligent management of personnel working in high-risk areas of nuclear power plants. This system achieves real-time monitoring of personnel safety status, dynamic risk prediction, and intelligent management, thereby improving the safety management level of high-risk areas in nuclear power plants and ensuring the safety of personnel and the stable operation of nuclear facilities. The system includes a data acquisition unit 101, a multi-dimensional situation awareness unit 102, a risk prediction and assessment unit 103, an intelligent decision-making and intervention unit 104, and a visualization and interaction unit 105. In this embodiment, the system is deployed in a multi-level distributed computing environment consisting of an internal data center of the nuclear power plant, edge computing nodes, and cloud servers to ensure real-time data processing, security, and large-scale computing capabilities.

[0054] The overall technical process of the system begins with the data acquisition unit 101 acquiring real-time heterogeneous data from multiple sources, including the physiological state, environmental parameters, precise location, behavioral patterns, nuclear facility and equipment operating status, and operating procedures of personnel working in high-risk areas of the nuclear power plant. This raw data is then transmitted to the multi-dimensional situational awareness unit 102 for preprocessing, feature extraction, contextual modeling, and situational quantification to construct a real-time, comprehensive multi-dimensional situational awareness of the personnel. The multi-dimensional situational awareness of the personnel output by the multi-dimensional situational awareness unit 102 serves as input, and the risk prediction and assessment unit 103 uses a prediction model to dynamically predict individual risk, environmental risk, and operational risk, and comprehensively assesses and generates a comprehensive risk level and confidence level for personnel working in high-risk areas. The intelligent decision-making and intervention unit 104 receives the comprehensive risk level and confidence level, and generates early warning information, decision suggestions, or triggers automatic intervention commands based on preset safety strategies and expert rules. Finally, the visualization and interaction unit 105 graphically displays all real-time situations, risk prediction results, early warning information, and decision suggestions, and provides a human-computer interaction interface to support monitoring and intervention by management personnel. In addition, the system achieves continuous model iteration and performance improvement through feedback and optimization mechanisms.

[0055] The data acquisition unit 101 aims to provide comprehensive, multi-dimensional, and highly real-time data input for subsequent intelligent decision-making. Please refer to the appendix. Figure 1 The data acquisition unit 101 is an integrated data interface platform that collects data from various sources through a variety of physical sensor networks and software interfaces. The data acquisition unit 101 includes a physiological state sensor group, an environmental parameter sensor group, a precise positioning and behavior perception sensor group, a nuclear facility equipment operation status interface, and an operating procedure and historical data interface.

[0056] A physiological state sensor array is used to collect core physiological parameter data from workers. This sensor array is integrated into the worker's smart wearable device, including smartwatches, smart bracelets, and smart work clothes, and specifically includes the following sensors:

[0057] Heart rate sensor: The real-time heart rate of the operator is measured using photoplethysmography, with a measurement range of 30 to 200 beats per minute and a sampling frequency of 1 Hz. The data is transmitted to the local edge computing node via Bluetooth Low Energy or Wi-Fi protocol using standard medical-grade protocols.

[0058] Body temperature sensor: Employs a high-precision thermistor or infrared sensor to measure the skin temperature of workers, with a measurement accuracy of ±0.1 degrees Celsius and a measurement range of 20 to 45 degrees Celsius. The sampling frequency is 0.5 Hz. When the body temperature exceeds the preset safety threshold, i.e., below 35 degrees Celsius or above 38.5 degrees Celsius, high-frequency sampling and local warning are automatically triggered.

[0059] Blood oxygen saturation sensor: Peripheral blood oxygen saturation is measured using pulse oximetry technology, with a measurement range of 70% to 100% and a measurement accuracy of ±2%. Data is collected every 5 seconds.

[0060] Skin conductivity sensor: This sensor reflects the activity of the sympathetic nervous system of workers by measuring changes in skin resistance, indirectly assessing their mental stress level. The sensor uses two silver / silver chloride electrodes, with a measurement range of 0.1 microsiemens to 20 microsiemens and a sampling frequency of 1 Hz.

[0061] Fatigue Assessment Module: This module comprehensively analyzes data from the worker's heart rate variability, skin conductance, accelerometer, and a pre-set physiological model to assess the worker's real-time fatigue status. The output is a fatigue index from 0 to 100, where 0 represents complete wakefulness and 100 represents extreme fatigue. This module outputs the fatigue index every minute and packages the results together with raw data such as heart rate and body temperature before sending them.

[0062] An environmental parameter sensor array is used to collect key environmental parameter data in the work area. This sensor array is distributed at various key locations in the high-risk area of ​​the nuclear power plant, forming a dense environmental monitoring network. Specifically, it includes the following sensors:

[0063] Radiation dose rate sensor: Employing a Geiger counter or semiconductor detector, it monitors the dose rate of ionizing radiation such as gamma rays and X-rays in real time, with a measurement range of 0.01 microsieverts / hour to 100 millisieverts / hour and a measurement accuracy of ±15%. Data acquisition frequency is once per second, and data is transmitted via industrial Ethernet or a dedicated wireless network.

[0064] Temperature sensor: Uses platinum resistance or thermocouple to measure ambient air temperature, with a measurement accuracy of ±0.5 degrees Celsius and a measurement range of 0 to 60 degrees Celsius.

[0065] Humidity sensor: measures the relative humidity of ambient air, with a measurement accuracy of ±3% relative humidity and a measurement range of 10% to 95% relative humidity.

[0066] Barometric pressure sensor: measures ambient atmospheric pressure with an accuracy of ±0.1 kPa and a measurement range of 80 kPa to 110 kPa.

[0067] Air quality sensor: Integrates multiple gas detectors to monitor the concentration of harmful gases, including hydrogen, carbon monoxide, nitrogen oxides, and inert gas leaks, with a detection accuracy on the order of one part per million; when the concentration of a specific gas exceeds the safety threshold, a local emergency alarm is triggered.

[0068] All environmental sensor data is transmitted to the regional data aggregation node via wired or wireless Industrial IoT protocols, and then sent to the data center. Data transmission uses encryption protocols to ensure data integrity and security.

[0069] The precise positioning and behavior perception sensor group is used to acquire precise positioning data and behavior pattern data of workers in real time, such as spatial location, movement trajectory, posture and activity intensity.

[0070] Ultra-wideband (UWB) positioning technology: A network of UWB base stations is deployed in high-risk areas of nuclear power plants. Workers wear UWB tags, and by measuring the time difference of arrival (TET) or round-trip time (RTT) of the UWB signal, sub-meter level high-precision real-time positioning is achieved. The positioning refresh rate is 10 Hz, and the 3D spatial coordinates (X, Y, Z) of the workers are output in real time. The positioning data is smoothed using algorithms such as Kalman filtering to eliminate noise.

[0071] Inertial Measurement Unit (IMU) technology: The IMU, integrated into the wearable device of the operator, includes a 3-axis accelerometer, a 3-axis gyroscope, and a 3-axis magnetometer; the accelerometer measures the linear acceleration of the operator, the gyroscope measures the angular velocity, and the magnetometer provides attitude reference, with a data sampling frequency of 100 Hz; the real-time attitude, direction of movement, and speed of the operator are calculated through sensor fusion algorithms and extended Kalman filtering.

[0072] Near Field Communication (NFC) technology: NFC tags worn by workers interact with NFC readers installed on equipment, tools, and control panels. When workers approach specific equipment or operating points, the system uses the NFC reader to sense close-range interaction events between the worker and the equipment, identify the type of equipment being operated, the operation time, and the operation sequence, thereby inferring the worker's specific behavioral patterns, including equipment inspection, valve switching, and instrument readings.

[0073] Precise positioning data, movement trajectory, and attitude data are all transmitted in real time via wireless local area network, while behavioral pattern data is initially identified locally and then packaged and sent to the back-end processing system of data acquisition unit 101.

[0074] The nuclear facility equipment operation status interface is used for data exchange with the nuclear power plant's distributed control system (DCS) or monitoring and data acquisition system (SCADA). This interface is a secure and highly reliable industrial data gateway that supports industrial communication protocols such as ModbusTCP, OPC UA, and IEC 61850. Through this interface, the system can obtain the operating parameters of the following key equipment in real time:

[0075] Nuclear reactor: Acquire core operating parameters such as core temperature, pressure, neutron flux, control rod position, coolant flow rate, and reactor power.

[0076] Steam generator: acquires parameters such as steam pressure, temperature, water level, and steam flow rate.

[0077] Cooling system: Acquires parameters such as the operating status of the main coolant pump, cooling water temperature, flow rate, and valve opening.

[0078] Electrical system: Acquire parameters such as generator output power, voltage, current, and circuit breaker status.

[0079] Fault Alarm: Receives information such as equipment faults, abnormal alarms, and over-limit alarms generated by the DCS / SCADA system, including alarm time, type, level, and corresponding equipment identifier.

[0080] Maintenance records: Obtain historical maintenance records, overhaul plans, and operational life assessment data for the equipment.

[0081] All device operating status data is updated every second in a structured data format via an encrypted channel to ensure data synchronization and security.

[0082] The work procedure and historical data interface is used to acquire work procedure data and historical data such as digital work permits, standard operating procedures (SOPs), historical work records, personnel training records, and incident reports. This interface is typically integrated with the nuclear power plant's asset management system (AMS), training management system, and safety management database, exchanging data through database connections or customized APIs.

[0083] Digital Work Permit: Obtain detailed information on currently valid work permits, including work name, work area, list of workers, start and end times of work, required tools, safety measures, and risk assessment conclusions.

[0084] Standard Operating Procedures: Obtain the text, flowcharts, and safety checklists of standard operating procedures related to operations in high-risk areas to guide behavioral pattern analysis and risk prediction.

[0085] Historical work records: Obtain all work data of personnel in high-risk areas in the past, including work duration, operation sequence, encountered anomalies, and handling results. This data is crucial for training behavior pattern recognition and prediction models.

[0086] Personnel training records: Obtain professional skills certifications, safety training records, simulated operation performance, and health records of operators to assess individual capabilities and potential risks.

[0087] Incident and Accident Reporting: Obtain detailed reports on historical incidents, events, and near misses, including causes, impacts, handling procedures, and lessons learned. These reports are used to build a risk knowledge base and optimize risk assessment models.

[0088] This data typically exists in the form of text, documents, or structured database records, and is periodically synchronized to the system's data storage layer via secure file transfer protocols or database interfaces.

[0089] Combined with appendix Figure 3 The multi-dimensional situation awareness unit 102 is electrically connected to the data acquisition unit 101, and its purpose is to perform in-depth processing on the multi-source heterogeneous data acquired by the data acquisition unit 101, thereby constructing a real-time and comprehensive multi-dimensional situation awareness of the personnel. The multi-dimensional situation awareness unit 102 is one of the core intelligent processing layers of this system, ensuring the effective transformation from raw data to high-level situation information. The multi-dimensional situation awareness unit 102 includes a data preprocessing module, a feature extraction module, a context modeling module, and a situation quantization module.

[0090] The data preprocessing module is used to perform time synchronization, spatial calibration, data cleaning, missing value filling, and data standardization on the multi-source heterogeneous data acquired by the data acquisition unit 101.

[0091] Time synchronization: To address the issue of inconsistent data acquisition frequencies from various sensors and interfaces, a high-precision timestamp server and network time protocol are used for time alignment to ensure precise consistency of all data in the time dimension; a sliding time window method is used to aggregate or interpolate data with different sampling rates to a unified time granularity.

[0092] Spatial calibration: Based on precise positioning data and environmental parameter data, all geographic coordinates and equipment locations are mapped to a unified 3D spatial coordinate system of the nuclear power plant, eliminating sensor installation errors and inconsistencies in reference systems. A sensor network-based self-calibration or preset base station coordinate calibration algorithm is employed to ensure high-precision correspondence of spatial information.

[0093] Data cleaning: Identify and remove noise, outliers and redundant information from the raw data. Use statistical methods and machine learning anomaly detection algorithms to detect and remove outliers in physiological parameters, environmental parameters and equipment operating parameters. For data drift or abrupt changes caused by sensor failures, apply digital filtering for smoothing.

[0094] Missing value imputation: For some sensor data missing due to transmission interruption or transient failure, interpolation, nearest neighbor imputation, or prediction imputation based on machine learning model are used to maintain the continuity and integrity of the data stream.

[0095] Data standardization: After cleaning and filling, the data is normalized or standardized, scaling the data to between 0 and 1, or converting it into a standard normal distribution with a mean of 0 and a variance of 1, in order to eliminate the dimensional differences between different data sources.

[0096] The feature extraction module is used to extract multi-dimensional features from preprocessed sensor data, transforming low-level raw data into high-level semantic features, revealing the potential risk patterns in the status of workers, environmental conditions, and equipment operation. The extracted features include the following four categories: physiological characteristic indicators, environmental anomaly indicators, behavioral anomaly patterns, and equipment operation trend features.

[0097] Physiological indicators include: Heart rate variability: reflecting the activity state of the autonomic nervous system by calculating the root mean square difference and low-frequency / high-frequency power ratio statistical characteristics; Body temperature fluctuation rate: the rate of change of body temperature per unit time, indicating physiological stress or abnormal response; Blood oxygen saturation trend: the long-term trend of blood oxygen saturation change, assessing the health status of the respiratory and circulatory systems; Fatigue index change rate: the dynamic change of the fatigue index, used to predict the degree of fatigue accumulation; Mental stress index: a real-time psychological stress assessment value that integrates indicators such as skin conductance and heart rate variability.

[0098] Anomalies in the environment include: sudden changes in radiation dose rate: significant changes in radiation dose rate within a short period of time; abnormal concentrations of harmful gases: gas concentrations exceeding safety thresholds or showing a rapid upward trend; and abnormal fluctuations in environmental parameters: temperature, humidity, and air pressure exceeding normal ranges or experiencing drastic changes.

[0099] Abnormal behavior patterns include: prolonged stillness: prolonged inactivity in a high-risk area, which may indicate coma or loss of mobility; abnormally rapid movement: unexpected high-speed movement, which may indicate an emergency escape or panic; fall / collision events: falls or collisions identified through inertial measurement unit data; boundary crossing behavior: entering unauthorized or dangerous areas; and non-compliant operation sequences: the sequence or duration of equipment operation that does not conform to standard operating procedures.

[0100] Equipment operation trend characteristics include: key parameter drift: parameters such as reactor power and coolant flow rate deviate from the normal operating range; fault alarm frequency: the number of equipment alarms increases abnormally per unit time; maintenance status change: the equipment changes from normal operating status to maintenance status.

[0101] The aforementioned features are extracted through statistical analysis, frequency domain analysis, wavelet transform, and deep learning time-series feature learning models, and output in the form of structured feature vectors to provide data support for subsequent risk assessment.

[0102] The context modeling module is used to construct a real-time context model of the operating environment of operators based on work procedure data, nuclear facility and equipment operating status data, and environmental parameter data, and to perform semantic modeling of the correlation between work tasks, equipment status, and environmental conditions. This module integrates discrete and heterogeneous context information into a semantically rich knowledge graph.

[0103] This module uses graph neural networks (GNNs) to semantically model the relationships between work tasks, equipment status, and environmental conditions. It abstracts entities such as work tasks, nuclear facilities and equipment, environmental conditions, and workers as nodes in a graph, and the relationships between nodes are abstracted as edges in the graph.

[0104] Graph neural networks use multi-layer graph convolution operations to aggregate feature information from neighboring nodes to update the representation of the current node, thereby capturing the complex relationships between nodes. Specifically, if a worker is operating a valve in a nearby high-radiation area, the graph neural network can fuse information such as the worker, the valve, and the high-radiation area through edge connections to form a semantic context that includes "the worker is operating a specific valve in a specific high-radiation area".

[0105] The specific processing flow of this module includes:

[0106] Entity recognition and relationship extraction: Automatically identify entities and the relationships between them from work procedure texts, equipment operation logs, and environmental monitoring reports.

[0107] Knowledge graph construction: The identified entities and relationships are constructed into a dynamically changing knowledge graph. The nodes and edges in the graph are accompanied by attribute information, including the real-time operating parameters of the equipment and the current location of the personnel.

[0108] Graph Embedding and Semantic Representation: Using a graph neural network model, the Graph Convolutional Network (GCN) learns from the knowledge graph to generate low-dimensional, dense vector representations of nodes and edges, capturing their semantic and structural information.

[0109] By learning contextual information, this module can provide a deeper semantic understanding for subsequent risk prediction and assessment, and identify potential risks that may be triggered by specific tasks under specific environmental conditions.

[0110] The situation quantification module is used to fuse the feature-extracted data with the context model to generate a high-dimensional situation vector, which quantifies the comprehensive situation of the operator in terms of current physiology, behavior, environment, equipment, and operational risks.

[0111] The core of this module is the effective fusion of numerical and temporal features from the feature extraction module with the semantic graph embedding vector generated by the context modeling module. The fusion process employs a multimodal fusion technique—weighted fusion using an attention mechanism. Specifically, the construction of the high-dimensional situation vector includes:

[0112] Feature vector concatenation: The physiological feature index vector, environmental anomaly index vector, behavioral anomaly pattern vector, and equipment operation trend feature vector are concatenated along the feature dimension to form a comprehensive numerical feature vector.

[0113] Context Embedding Fusion: The knowledge graph embedding vector output by the context modeling module is fused with the concatenated numerical feature vector. The fusion method can be through deep learning models such as Gated Recurrent Unit (GRU) and Transformer encoder to learn the interaction between different modal information and generate a more expressive fused situation vector.

[0114] Situation Indicator Calculation: Based on the high-dimensional situation vector, a series of quantified situation indicators are calculated, including:

[0115] Real-time physiological load index: a combination of heart rate, body temperature, fatigue level, etc.; environmental exposure index: based on radiation dose rate, harmful gas concentration, exposure duration, etc.; behavioral compliance index: based on the matching degree between behavioral patterns and operating procedures; equipment-related risk index: based on the closeness of interaction between personnel and risky equipment.

[0116] The final generated situation quantification vector and various indicators are output in a standardized format as input to the risk prediction and assessment unit 103.

[0117] Combined with appendix Figure 2The risk prediction and assessment unit 103 is electrically connected to the multi-dimensional situational awareness unit 102. Based on the multi-dimensional situational awareness of workers output by the multi-dimensional situational awareness unit 102, it uses a prediction model to dynamically predict individual risks, environmental risks, and operational risks, and comprehensively assesses and generates the overall risk level and confidence level of workers in high-risk areas. The risk prediction and assessment unit 103 is the core intelligent analysis layer of this system, transforming situational awareness information into actionable risk insights. The risk prediction and assessment unit 103 includes a risk factor identification module, a prediction model group, and a risk assessment model.

[0118] The risk factor identification module is used to monitor abnormal patterns, trend changes, and related events in the multidimensional situation of operators in real time, in order to identify potential risk-inducing factors.

[0119] This module continuously analyzes the high-dimensional situation vector and various situation indicators output by the situation quantification module, and adopts a method that combines rule-based expert system with machine learning anomaly detection.

[0120] Threshold rule recognition: Set static and dynamic thresholds for various physiological, environmental and behavioral indicators. When the heart rate exceeds 150 beats / minute for 2 minutes, the radiation dose rate increases more than 10 times within 1 minute, or the worker stays in an unauthorized area for more than 30 seconds, the system will immediately identify them as potential risk factors.

[0121] Temporal pattern recognition: Using anomaly detection models based on statistical process control (SPC) or recurrent neural networks (RNN), non-periodic fluctuations, long-term trend drifts, or abnormal linkages with other indicators of situational indicators are identified. Specifically, the correlation between the continuous upward trend of individual fatigue index and the probability of operational errors, or the correlation between abnormal increase in ambient temperature and increase in cooling system load.

[0122] Correlation analysis: By constructing a knowledge graph through the context modeling module and combining it with event flow processing (CEP) technology, the causal or correlation relationships between multi-source events can be identified, including concurrent events such as "equipment malfunction alarm" and "operator approaching the equipment", as well as correlation events such as "hazardous gas leak" and "ventilation system malfunction", thereby more accurately identifying the starting point of the risk chain.

[0123] The risk factor identification module will identify all potential risk triggers, including their type, timing, intensity, and scope of impact, and output them to the prediction model group for more refined risk prediction.

[0124] The prediction model group is used to dynamically predict the identified risk factors. The prediction model group includes individual risk prediction models, environmental risk prediction models, and operational risk prediction models.

[0125] Individual risk prediction models are used to predict workers' fatigue levels, mental stress levels, operational error probabilities, and potential health risks based on workers' physiological state data and behavioral pattern data.

[0126] The model uses an attention-based recurrent neural network to learn the temporal characteristics of workers' physiological states and behavioral patterns.

[0127] Model structure description:

[0128] The individual risk prediction model is a multi-layer neural network structure, with a recurrent neural network (RNN) at its core, capable of effectively processing time-series data. An attention mechanism is introduced above the RNN layers. This attention mechanism allows the model to dynamically assign different weights to different time steps in the input sequence when processing long-sequence data, thus focusing more on the physiological or behavioral features most relevant to the current prediction task.

[0129] Input data: The model's input is time-series data, including physiological state data from data acquisition unit 101, namely heart rate, body temperature, blood oxygen saturation, skin conductance, and fatigue level, as well as behavioral pattern data from the precise positioning and behavior perception sensor group, namely movement trajectory, posture, activity intensity, and interaction events. This data is processed by the data preprocessing module and the feature extraction module to form a unified, multi-dimensional temporal feature vector sequence.

[0130] Model training process:

[0131] Dataset construction: Collect a large amount of historical data on nuclear power plant workers under different physiological and behavioral conditions, and label their corresponding fatigue level, stress level, operational errors and health risk events.

[0132] Feature engineering: Converting raw time-series data into feature vectors that are acceptable to the model.

[0133] Model Architecture: Construct an encoder-decoder structure containing multiple layers of LSTM or GRU units, where the encoder learns the contextual representation of the input sequence, and the decoder generates the prediction results. Add attention layers between the encoder and decoder or within the encoder to enable it to "pay attention" to the most important parts of the sequence when processing the input at each time step.

[0134] Loss Function and Optimizer: The mean squared error (MSE) is used as the loss function for regression prediction, or cross-entropy is used as the loss function for classification prediction. The optimizer employs optimization algorithms such as Adaptive Moment Estimation (Adam).

[0135] Model output: The prediction results include the following for workers in the next 15 minutes:

[0136] Fatigue state prediction: quantified fatigue index and its confidence interval.

[0137] Mental stress level prediction: quantified stress index and its confidence interval.

[0138] Operational error probability prediction: The percentage probability of an operational error occurring in a specific operation.

[0139] Potential health risk prediction: probability of heatstroke and cardiovascular events.

[0140] Environmental risk prediction models are used to predict environmental risk events such as changes in radiation levels, leaks of hazardous gases, and the spread of hazardous areas caused by equipment failures, based on environmental parameter data and nuclear facility and equipment operating status data.

[0141] The model uses the Spatiotemporal Graph Convolutional Network (STGCN) to model and predict the spatiotemporal correlation between environmental parameters and equipment operating status.

[0142] Model structure description:

[0143] Spatiotemporal Graph Convolutional Network (STGCN) is a model that combines Graph Convolutional Network (GCN) and Temporal Convolutional Network (TCN), and is particularly suitable for processing data with complex spatiotemporal dependencies. In this model, sensor nodes and critical equipment in high-risk areas of nuclear power plants are abstracted as nodes of a graph, and the spatial connections between them constitute the edges of the graph. The temporal convolutional network is responsible for capturing the temporal series features of each node, while the graph convolutional network is responsible for capturing the spatial dependencies between nodes.

[0144] Input data: The model input consists of environmental parameter data and nuclear facility equipment operating status data. This data is constructed into a series of graph signals over time steps, where the feature vector of each node contains its environmental parameters and equipment operating status at the current time step.

[0145] Model training process:

[0146] Graph structure definition: Based on the physical layout and equipment connection relationships of a nuclear power plant, construct a topology graph that includes sensors and equipment.

[0147] Dataset construction: Collect historical environmental parameters and equipment operating status data, and label the occurrence time and scope of environmental risk events.

[0148] Model Architecture: A network consisting of multiple stacked STGCN layers is constructed. Each STGCN layer contains a graph convolutional module and a temporal convolutional module, which handle feature learning in the spatial and temporal dimensions, respectively.

[0149] Loss function and optimizer: The loss function used for time series prediction tasks is mean squared error (MSE), and the optimizer used is Adam.

[0150] Model output: Predictions include the next 30 minutes:

[0151] Radiation level change prediction: Predict the trend of radiation dose rate changes at different locations in high-risk areas and the formation of potential high-radiation areas.

[0152] Hazardous gas leak prediction: Predicting the likelihood of hazardous gas leaks, the type of gas, the range of concentration diffusion, and the speed.

[0153] Prediction of hazardous area spread due to equipment failure: potential spread of steam leakage or high-temperature areas due to cooling pump failure.

[0154] The operational risk prediction model is used to predict potential hazards such as violations of operating procedures, misoperations, and interactions with equipment by operators, based on operational procedure data, operator behavior pattern data, and nuclear facility equipment operation status data.

[0155] The model employs a hybrid model based on the fusion of Transformer encoder and expert knowledge.

[0156] Model structure description:

[0157] The operational risk prediction model combines the temporal modeling capabilities of deep learning with the logical reasoning abilities of expert knowledge. At its core is a Transformer encoder, used to capture the temporal dependencies and contextual semantics of worker behavior patterns. With its self-attention mechanism, the Transformer encoder effectively handles long-distance dependencies and assigns different importance to each behavioral event in the sequence. The expert knowledge fusion module embeds standardized operating procedures (SOPs), safety rules, and historical accident experience from nuclear power plants into the model in the form of rule sets or knowledge graphs to assist in judging and correcting the model output.

[0158] Input data: The model's input includes work procedure data, operator behavior pattern data, and equipment operating status data. Work procedure data includes digitized Standard Operating Procedures (SOPs) and work permit information; operator behavior pattern data includes operation sequences, dwell time, movement trajectory, and speed; and equipment operating status data includes equipment alarms and mode switching. This data is encoded into a series of discrete or continuous sequences, which serve as the input to the Transformer encoder.

[0159] Model training process:

[0160] Dataset construction: Collect a large number of historical operation records, including normal operation and violation operation events, and label the corresponding operation risk level.

[0161] Behavior sequence coding: Encoding the behavior sequence of operators and the status of related equipment.

[0162] Model Architecture: A multi-layer Transformer encoder is constructed, taking the encoded behavior sequence and device state as input. The encoder's output sequence representation is then passed through a fully connected layer for risk prediction.

[0163] Expert knowledge fusion: During the model training or inference phase, the expert rule base is used as a regularization term or post-processing module to verify and adjust the model output. If the model predicts that a certain operation sequence is safe, but the expert rules clearly indicate that the sequence is illegal under a specific device state, the risk prediction result is adjusted.

[0164] Model output: Predictions include the next 11 minutes:

[0165] Probability of violation of operating procedures: The probability of violating the SOP in a specific work step.

[0166] Probability of misoperation: The probability of misoperation due to negligence or error during human-computer interaction.

[0167] Potential hazards when interacting with equipment: the probability of jamming or leakage when operating critical valves.

[0168] The risk assessment model is used to integrate the prediction results output by the individual risk prediction model, the environmental risk prediction model, and the operational risk prediction model, and combine them with preset risk weights and risk matrices to generate the overall risk level and corresponding confidence level of the current operator.

[0169] This module is the final output of the risk prediction and assessment unit 103, and aims to provide a unified and quantitative view of risk.

[0170] Comprehensive approach:

[0171] Risk index normalization: Normalize the output results of individual risk prediction, environmental risk prediction and operational risk prediction models to a unified risk score range.

[0172] Risk weighting: Based on expert experience and safety regulations for nuclear power plant safety management, different weights are assigned to individual risks, environmental risks, and operational risks; in high-radiation areas, the weight of environmental risks may be higher than that of individual fatigue risks. These weights are configurable and dynamically adjustable.

[0173] Risk matrix analysis: Using a standard or customized risk matrix from the nuclear power industry, different risk categories and their prediction results are mapped to preset risk levels. The risk matrix typically categorizes risks into low, medium, high, and extremely high levels based on the probability of occurrence and the severity of consequences.

[0174] Multi-source information fusion: By using methods such as weighted summation, fuzzy logic reasoning, or Bayesian networks, the prediction results and weights of different categories of risks are fused together to calculate a comprehensive risk value.

[0175] The comprehensive risk value is transformed through a nonlinear mapping function to generate a comprehensive risk level: Level 1: low risk, Level 2: medium risk, Level 3: high risk, Level 4: extremely high risk.

[0176] Confidence score calculation: The confidence score of the comprehensive risk level is calculated using methods such as Monte Carlo simulation, Bayesian inference, or the variance of the probability distribution output by a deep learning model. Confidence score reflects the uncertainty of the model regarding its own prediction results. When the confidence score reaches 95%, it means that the model believes that the current risk level has a 95% probability of being accurate. The higher the confidence score, the more likely the decision-making unit is to adopt its output.

[0177] The risk assessment model outputs the comprehensive risk level and its confidence level to the intelligent decision-making and intervention unit 104.

[0178] Combined with appendix Figure 4 The intelligent decision-making and intervention unit 104 is electrically connected to the risk prediction and assessment unit 103. It generates early warning information, decision suggestions, or triggers automatic intervention commands based on the comprehensive risk level and confidence level output by the risk prediction and assessment unit 103. This unit is the key execution layer for the system to achieve intelligent risk management and security assurance. The intelligent decision-making and intervention unit 104 includes an early warning generation module, a decision recommendation module, an automatic intervention module, and a feedback and optimization module.

[0179] The early warning generation module is used to trigger early warning information of different levels when the risk level exceeds a preset threshold, based on the comprehensive risk level and confidence level output by the risk assessment model.

[0180] The core of this module is a tiered early warning mechanism to ensure that early warning information can be delivered to relevant personnel in a timely, accurate and appropriate manner.

[0181] Early warning level definition: Based on the comprehensive risk level and confidence level, a multi-level early warning response is defined:

[0182] Level 1 Warning (Low Risk): No warning issued, continuous monitoring.

[0183] Level 2 Warning (Medium Risk): When the risk level reaches Level 2 and the confidence level is higher than 70%, a slight vibration is triggered on the worker's local wearable device to alert them. At the same time, the work team leader's mobile terminal receives a low-priority alarm message.

[0184] Level 3 Warning (High Risk): When the risk level reaches Level 3 and the confidence level is higher than 80%, a strong vibration and audible / visual alert is triggered on the worker's wearable device, and the specific risk details are displayed. The work team leader's mobile terminal receives a medium-priority alarm with a detailed risk description. A flashing warning appears on the monitoring screen in the central control room.

[0185] Level 4 Warning (Extremely High Risk): When the risk level reaches Level 4 and the confidence level is higher than 90%, an emergency audible and visual alarm is triggered on the equipment worn by the workers, and an evacuation order is forcibly displayed. The work team leader's mobile terminal receives the highest priority alarm and a pop-up window is forcibly displayed to show the emergency response plan. The audible and visual alarm system in the central control room is activated and automatically notifies the emergency response team.

[0186] Warning content generation: Warning information includes not only the risk level, but also the specific risk type, affected area, recommended physiological indicators, environmental parameters or equipment status; for "radiation exceeding the standard risk", the warning content may include "the current radiation dose rate in the area exceeds the safety threshold by 2 times, it is recommended to immediately check personal protective equipment and evacuate to the designated safe area".

[0187] Information distribution: Warning information is distributed through multiple communication channels, including wireless LAN, dedicated digital trunking communication (DMR), SMS, and voice broadcasting systems, to ensure reliable delivery of information.

[0188] The decision recommendation module is used to generate and push optimal risk avoidance strategies, emergency response plans, or work process adjustment suggestions to operators or managers based on a preset expert rule base and reinforcement learning model for specific risk scenarios and early warning information.

[0189] This module aims to provide intelligent action guidance, transforming risk assessments into actionable decisions.

[0190] Expert rule base: Construct a comprehensive expert knowledge base, which includes various regulations, emergency plans, accident handling procedures and expert experience for the safe operation of nuclear power plants; the rule base adopts condition-action rules in the form of IF-THEN, and supports dynamic updates and maintenance.

[0191] Reinforcement learning model: When the risk scenario is complex and the rule base is difficult to fully cover, a reinforcement learning (RL) model is used to find the optimal decision.

[0192] Model training process:

[0193] Environment definition: The dynamic situation, equipment status, and personnel behavior of high-risk areas in nuclear power plants are defined as the state space of the reinforcement learning environment.

[0194] Action space: Defines the actions the model can take, including "suggest checking equipment", "suggest adjusting work posture", "suggest pausing work", "suggest evacuation", and "starting the ventilation system", etc.

[0195] Reward function: Design a reward function to incentivize the model to select actions that reduce risk, ensure safety, and minimize operational disruption; successful risk avoidance earns high scores, while accidents or ineffective interventions result in penalties.

[0196] Training algorithms: Deep reinforcement learning algorithms such as Deep Q-Network (DQN), Asynchronous Advantageous Actor Commenting (A3C), or Proximal Policy Optimization (PPO) are used to train the model through interaction with the simulated environment and learn the strategy of taking the optimal action under different risk situations.

[0197] Model Output: Based on the current risk level, risk type, confidence level, and contextual information, the decision recommendation module comprehensively utilizes expert rules and reinforcement learning models to output priority-ranked decision suggestions, specifically: "Emergency evacuation to a safe area," "Pause current equipment operation and check coolant pressure," "Wear higher-level protective equipment," and "Adjust work plan to avoid the area for a period of time." These suggestions are displayed through the visualization and interaction unit 105 and pushed to the terminals of operators or managers.

[0198] The automatic intervention module is used to trigger an automated safety protection mechanism in extreme emergency risk situations when the comprehensive risk level reaches the highest preset threshold and the confidence level is high.

[0199] This module is the last line of defense for this system in dealing with sudden and high-risk situations, and is designed to minimize casualties and damage to nuclear facilities.

[0200] Intervention triggering conditions: Multiple high-risk indicators usually need to be triggered simultaneously, and the risk assessment confidence level must be extremely high to avoid false triggering; when an extremely high radiation dose rate is detected, a worker falls or falls into a coma, and their physiological indicators are severely abnormal, the system may trigger automatic intervention.

[0201] Automatic intervention instructions:

[0202] Area isolation: Instructs the automatic access control system to close the passage to a specific area to prevent more people from entering the dangerous area, while locking stranded personnel in a safer sub-area.

[0203] Emergency equipment shutdown: By linking the nuclear facility equipment operation status interface with the DCS / SCADA system, high-risk equipment is instructed to perform emergency shutdown or load reduction operations to prevent the fault from escalating or the damage from worsening.

[0204] Mandatory evacuation order broadcast: The mandatory evacuation order and evacuation route are broadcast to all personnel in the hazardous area through the nuclear power plant's broadcast system. At the same time, the maximum intensity alarm is issued via remote control equipment worn by the personnel.

[0205] Automatic activation of fire suppression or cooling systems: Based on feedback from environmental parameter sensors, if a fire risk or equipment overheating is detected, the preset fire suppression system or emergency cooling system will be automatically activated.

[0206] The execution of automated intervention commands has the highest priority and requires strict security logic verification and multiple confirmation mechanisms to prevent secondary harm caused by misoperation. All automated intervention actions and their results are recorded and incorporated into the feedback and optimization module for analysis.

[0207] The feedback and optimization module is used to record the execution status and actual effects of all early warning information, decision suggestions, and automatic intervention instructions, and transmit these feedback data to the risk prediction and assessment unit 103 for continuous training and optimization of the model.

[0208] This module constructs a closed-loop learning mechanism to ensure that the system can continuously improve its performance and adaptability over time.

[0209] Data Recording: Detailed records of the trigger time, type, level, recommended decision, actual intervention measures taken, and changes in the risk situation and final results after the implementation of these measures for each risk warning.

[0210] Effectiveness evaluation: The effectiveness of early warning, decision-making, and intervention measures is assessed by comparing the predicted results with the actual situation and the changes in risk indicators before and after the intervention.

[0211] Model retraining: Using labeled feedback data as new training samples, the individual risk prediction model, environmental risk prediction model, occupational risk prediction model, and reinforcement learning model are retrained periodically or when a decline in model performance is detected. Retraining can be performed using incremental learning or full retraining.

[0212] Parameter optimization: Based on feedback data and effect evaluation results, dynamically adjust the risk weights and early warning thresholds in the risk assessment model, as well as the expert rules and reward functions of the reinforcement learning model in the decision recommendation module.

[0213] The feedback and optimization module is key to the system’s self-adaptation and self-learning capabilities, ensuring that it can continuously adapt to the complex and ever-changing safety management needs of nuclear power plants.

[0214] The visualization and interaction unit 105 is electrically connected to the multi-dimensional situational awareness unit 102 and the intelligent decision-making and intervention unit 104. It is used to graphically display real-time situation, risk prediction results, early warning information, and decision suggestions, and to provide a human-computer interaction interface. This unit is the core interface for effective information exchange and control between the user and the system. The visualization and interaction unit 105 includes a real-time situation display module, a risk dashboard module, an operation management interface module, and a reporting and retrospective analysis module.

[0215] The real-time situation display module is used to display the precise location, movement trajectory, physiological state trend, environmental parameter distribution, and nuclear facility equipment operation status of workers on a high-risk area map in real time through two-dimensional or three-dimensional visualization technology, and to identify risk areas with color coding or dynamic icons.

[0216] This module provides an intuitive, real-time panoramic view of the work site.

[0217] Map rendering: Using the building information model (BIM) data or high-precision 2D floor plan of the nuclear power plant as the base map, the map is displayed in real time through a graphics rendering engine.

[0218] Personnel situation visualization:

[0219] Precise location and trajectory: The real-time location of the operator is displayed on the map as a dynamic icon, and their recent movement trajectory is shown through the dotted line or trajectory color change.

[0220] Physiological status: The floating label or color change next to the personnel icon indicates key physiological parameters such as heart rate, body temperature, and fatigue level in real time, and provides historical trend charts.

[0221] Behavioral patterns: Visualize the current behavioral state of personnel through animation effects or icon changes.

[0222] Visualization of environmental parameter distribution:

[0223] Radiation dose rate: The radiation dose rate distribution in high-risk areas is displayed in real time on the map through heat maps, contour maps, or color gradient areas. The darker the color, the higher the dose rate.

[0224] Hazardous gases: The source of the leak and the range of concentration diffusion are identified by diffusion simulation animations or specific gas icons.

[0225] Temperature / humidity / air pressure: displayed by color-changing areas or numerical labels.

[0226] Equipment operation status visualization: Key nuclear facilities and equipment are displayed on the map as icons, and their real-time operation status is shown through icon color, flashing frequency or animation.

[0227] Risk area identification: Based on the comprehensive risk level output by the risk prediction and assessment unit 103, the current or predicted risk areas are dynamically identified on the map using overlays of different colors or transparency: red areas indicate extremely high risk, and orange areas indicate high risk.

[0228] The risk dashboard module is used to centrally display the comprehensive risk level of operators, various risk prediction curves, historical risk event records, and current warning status, and supports data filtering and analysis by time, region, or personnel.

[0229] This module serves as the primary interface for managers to conduct macro-level risk monitoring and in-depth analysis.

[0230] Comprehensive Risk Overview: Visually displays the average comprehensive risk level, current highest risk level, and risk distribution for all operators in the form of a dashboard, progress bar, or pie chart.

[0231] Risk Prediction Curves: Displays the predicted curves of individual risk, environmental risk, operational risk, and overall risk over a future period, including confidence intervals, to help managers predict risk trends.

[0232] Historical Risk Event Log: Provides a searchable and filterable list of events, detailing every warning, decision, intervention, and its outcome. It supports filtering by risk type, time frame, personnel, and region, and allows users to click to view event details.

[0233] Current alert status: Displays a real-time list of alert information for all activities, including alert level, type, trigger time, and affected personnel and areas.

[0234] Data filtering and analysis: Provides flexible interactive charts and tables, supporting managers to perform in-depth drill-down and cross-analysis of risk data by time granularity, geographic region, worker ID, and other dimensions, identifying risk accumulation patterns of specific personnel in specific regions.

[0235] The job management interface module provides functions such as digital job ticket management, job task allocation, safety procedure query, and emergency plan invocation, and allows managers to confirm or modify risk warnings and decision suggestions.

[0236] This module enables deep integration of the system with nuclear power plant operating procedures.

[0237] Digital work permit management: Supports administrators in creating, approving, assigning, and revoking digital work permits online. The system automatically associates work permits with workers, work areas, and relevant safety procedures.

[0238] Task assignment: Managers can assign new tasks to specific workers or work groups, and the system will automatically update the task status.

[0239] Safety Procedures Inquiry: Provides a searchable knowledge base that allows managers and operators to quickly access relevant standard operating procedures (SOPs), safety checklists, and equipment operation manuals.

[0240] Emergency Response Plan Activation: When an alert is triggered, the system automatically recommends the corresponding emergency response plan, which can be activated with a single click or selected manually by administrators. The plan content is presented in the form of a step-by-step checklist, illustrations, or videos.

[0241] Warning and Decision Confirmation / Modification: Managers can view the warning information and decision suggestions generated by the system and confirm or modify them; after receiving the system's evacuation suggestion, managers can confirm execution, postpone execution, or issue different instructions based on the on-site situation; all confirmation and modification operations are recorded as input to the feedback and optimization module.

[0242] The reporting and retrospective analysis module is used to generate periodic safety reports, incident investigation reports, and performance evaluation reports. It also supports retrospective analysis of historical work processes to identify safety management vulnerabilities and summarize lessons learned.

[0243] This module provides data support and decision-making reference for the safety management of nuclear power plants.

[0244] Periodic security reports: Automatically generate daily, weekly, monthly or quarterly security reports, summarizing key indicators such as the overall security status of high-risk areas, risk trends, number of warnings, and effectiveness of intervention measures, presented in charts and text.

[0245] Accident and Incident Investigation Report: When an accident or major incident occurs, the system can automatically collect all relevant historical data, including situational data before and after the incident, prediction results, intervention records, personnel behavior trajectories, and equipment operating parameters, and generate a detailed incident retrospective report to assist investigators in analyzing the cause of the accident.

[0246] Performance evaluation report: Evaluates the safety performance of the system itself and the operators, assesses the accuracy of different prediction models and the timeliness of early warnings, or evaluates the response efficiency and compliance of specific operators in different risk scenarios.

[0247] Retrospective analysis: Provides powerful historical data query and playback functions. Managers can select any historical time period to "replay" the operation process, risk evolution, and system response in the real-time situation display module, thereby identifying potential safety management vulnerabilities, optimizing work processes, and summarizing experience for training.

[0248] This embodiment of the intelligent management system for situation prediction and risk assessment of personnel working in high-risk areas of nuclear power plants, through the close collaboration of the aforementioned units, constructs a closed-loop intelligent management platform covering the entire chain, from data acquisition, multi-dimensional perception, intelligent prediction and assessment to decision intervention and continuous optimization. This not only significantly improves the safety management level and response efficiency of personnel working in high-risk areas of nuclear power plants, but also effectively prevents accidents, reduces casualties and property losses, and provides strong technical support for safe production in the nuclear power industry.

Claims

1. An intelligent management system for personnel situation prediction and risk assessment in nuclear power plants, characterized in that, include: The data acquisition unit is used to acquire in real time physiological status data, environmental parameter data, precise location data, behavioral pattern data, nuclear facility and equipment operation status data, and operation procedure data of personnel working in high-risk areas of nuclear power plants; A multi-dimensional situation awareness unit, electrically connected to the data acquisition unit, is used to preprocess, extract features, model context, and quantify the multi-source heterogeneous data acquired by the data acquisition unit to construct a real-time and comprehensive multi-dimensional situation of the operators. The context modeling uses a graph neural network to semantically model the correlation between work tasks, equipment status, and environmental conditions. The risk prediction and assessment unit is electrically connected to the multi-dimensional situational awareness unit. It is used to dynamically predict individual risks, environmental risks, and operational risks based on the multi-dimensional situational awareness of the workers output by the multi-dimensional situational awareness unit and using a prediction model. The individual risk prediction model uses a recurrent neural network model based on an attention mechanism to learn the temporal characteristics of the workers' physiological state and behavioral patterns. The environmental risk prediction model uses a spatiotemporal graph convolutional network to model and predict the spatiotemporal correlation between environmental parameters and equipment operating status, and comprehensively evaluates and generates the comprehensive risk level and confidence level of workers in high-risk areas. The intelligent decision-making and intervention unit is electrically connected to the risk prediction and assessment unit and is used to generate early warning information, decision suggestions or trigger automatic intervention commands based on the comprehensive risk level and confidence level output by the risk prediction and assessment unit. The visualization and interaction unit is electrically connected to the multi-dimensional situational awareness unit and the intelligent decision-making and intervention unit, and is used to graphically display the real-time situation, risk prediction results, early warning information and decision suggestions, and provide a human-computer interaction interface.

2. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The data acquisition unit includes a physiological state sensor group, an environmental parameter sensor group, a precise positioning and behavior perception sensor group, a nuclear facility equipment operation status interface, and an operation procedure and historical data interface. The physiological state sensor group is used to collect physiological parameter data such as heart rate, body temperature, and blood oxygen saturation. The environmental parameter sensor group is used to collect environmental parameter data such as radiation dose rate, temperature, and humidity. The precise positioning and behavior perception sensor group is used to acquire the spatial location, movement trajectory, and behavior pattern data of the operators.

3. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 2, characterized in that, The physiological state sensor group is used to collect data on the worker's heart rate, body temperature, blood oxygen saturation, skin conductance, and fatigue level.

4. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The data preprocessing module of the multi-dimensional situational awareness unit is used to perform time synchronization on the multi-source heterogeneous data to ensure the data's precise consistency in the time dimension, perform spatial calibration to eliminate sensor installation errors and inconsistencies in the reference system, perform data cleaning to identify and remove noise, outliers, and redundant information, perform missing value filling to maintain the continuity and integrity of the data stream, and perform data standardization to eliminate dimensional differences between different data sources.

5. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The feature extraction module of the multidimensional situational awareness unit is used to extract multidimensional features from the preprocessed raw sensor data. The multidimensional features include physiological feature indicators based on heart rate variability, fatigue index change rate, and mental stress index; environmental anomaly indicators based on radiation dose rate mutation, abnormal concentration of harmful gases, and abnormal temperature fluctuation; behavioral anomaly patterns based on long-term stillness, abnormal rapid movement, fall events, boundary crossing behavior, and violation operation sequences; and equipment operation trend features based on equipment key parameter drift, fault alarm frequency, and maintenance status changes.

6. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The risk prediction and assessment unit also includes a risk factor identification module, which is used to monitor abnormal patterns, trend changes and related events in the multidimensional situation of operators in real time, so as to identify potential risk-inducing factors.

7. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The risk prediction and assessment unit also includes a risk assessment model, which integrates the prediction results output by the individual risk prediction model, the environmental risk prediction model, and the operational risk prediction model, and combines them with preset risk weights and risk matrices to generate the current operator's comprehensive risk level and corresponding confidence level.

8. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The intelligent decision-making and intervention unit also includes a decision recommendation module, which is used to generate and push the optimal risk avoidance strategy, emergency response plan or work process adjustment suggestions to operators or managers based on a preset expert rule base and reinforcement learning model for specific risk scenarios and early warning information.

9. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The intelligent decision-making and intervention unit also includes an automatic intervention module, which is used to trigger an automated safety protection mechanism in extreme emergency risk situations when the comprehensive risk level reaches the highest preset threshold and the confidence level is high. The safety protection mechanism includes area isolation, emergency equipment shutdown, broadcasting of forced evacuation instructions, and automatic activation of fire protection or cooling systems.

10. The intelligent management system for personnel situation prediction and risk assessment in nuclear power plants according to claim 1, characterized in that, The visualization and interaction unit also includes a real-time situation display module, which uses 2D or 3D visualization technology to display the precise location, movement trajectory, physiological state trend, environmental parameter distribution, and nuclear facility equipment operation status of workers on a high-risk area map in real time, and identifies risk areas with color coding or dynamic icons.

Images