Knowledge graph-based nuclear power plant risk assessment method and system, electronic device, and medium
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NEUTRON TIMES (QINGDAO) INNOVATION TECH CO LTD
- Filing Date
- 2026-02-09
- Publication Date
- 2026-06-09
Smart Images
Figure CN122175350A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of nuclear power plant safety assessment, specifically to a nuclear power plant risk assessment method and system based on knowledge graphs. Background Technology
[0002] In the field of nuclear power plant safety assessment, traditional risk assessment methods face numerous challenges due to the increasing complexity of nuclear power units and changes in the operating environment. Currently, there are two main technical approaches in this field: one is socio-technical risk analysis based on probabilistic risk assessment (PRA), and the other is nuclear power plant safety analysis based on knowledge graphs.
[0003] For the first type of technical approach, in the nuclear power plant risk assessment industry, a common method to address the issue of quantitatively analyzing the impact of organizational factors on safety risks is the Integrated Probabilistic Risk Assessment (I-PRA) method based on the SoTeRiA theoretical framework. However, this approach suffers from the following problems and drawbacks because its design principle relies on probabilistic statistics and provisional text feature extraction:
[0004] (1) Weak correlation between social and technological factors: This method only uses organizational factors as input parameters for PRA, and fails to construct a structured semantic relationship between social and technological factors, which limits the depth and breadth of risk transmission path analysis;
[0005] (2) Insufficient dynamic adaptability: The model update relies on batch-processed Bayesian updates, which makes it difficult to achieve real-time risk situation awareness and meet the needs of dynamic risk monitoring during the operation of nuclear power plants;
[0006] (3) Poor knowledge reusability: The extracted risk features are closely coupled with specific event reports, and there is a lack of a unified knowledge representation framework, which makes it difficult to share and reuse risk knowledge from different power plants and at different times.
[0007] In the second technical approach, a knowledge graph of the power plant system and assets is constructed using Model-Based Systems Engineering (MBSE) models, integrating numerical and textual data elements to assist systems engineers in analyzing equipment reliability data. However, because this approach focuses on the analysis of technical entities and asset health, it suffers from the following problems and drawbacks:
[0008] (1) Lack of social dimension: Existing knowledge graphs mainly focus on the technical status of equipment, systems and assets, and lack systematic modeling of social factors such as organizational culture, management processes and personnel behavior;
[0009] (2) Insufficient cross-domain correlation: Although the causal relationship between the technical entities was established, the social factors and technical risks were not deeply correlated and cross-reasoned, which limited the comprehensiveness of the risk root cause analysis;
[0010] (3) Weak risk assessment expertise: Existing knowledge graph designs are mainly used for safety review assistance and equipment health management, rather than specifically for comprehensive risk assessment, and lack risk reasoning mechanisms for the social-technical system characteristics of nuclear power plants.
[0011] In summary, existing technologies, whether social-technical risk analysis methods or knowledge graph technology, suffer from a disconnect between social and technical factors in nuclear power plant risk assessment. This makes it difficult to fully reflect the essential characteristics of nuclear power plants as complex social-technical systems, which in turn leads to lower accuracy in nuclear power plant risk assessment results. Summary of the Invention
[0012] The purpose of this invention is to provide a knowledge graph-based method and system for nuclear power plant risk assessment. This addresses the technical problem in existing nuclear power plant risk assessment methods that separate social and technological factors, failing to fully reflect the essential characteristics of nuclear power plants as complex socio-technical systems. Consequently, the accuracy of nuclear power plant risk assessment results is often low. The knowledge graph-based method and system proposed in this invention not only solves the problem of comprehensive nuclear power plant risk assessment but also overcomes the problems and shortcomings of existing technologies. By constructing a unified knowledge model that deeply integrates social and technological elements, it achieves a more comprehensive, accurate, and dynamic risk assessment.
[0013] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:
[0014] In a first aspect, the present invention provides a nuclear power plant risk assessment method based on knowledge graphs, comprising:
[0015] The technical entity domain, social entity domain, and risk entity domain related to nuclear power plants were obtained respectively;
[0016] Establish a cross-domain semantic relationship network model;
[0017] A knowledge graph for nuclear power plant risk assessment is constructed based on the cross-domain semantic relationship network model.
[0018] Based on the aforementioned nuclear power plant risk assessment knowledge graph, technical and social parameters of the nuclear power plant are collected in real time to obtain the nuclear power plant risk assessment results.
[0019] Furthermore, establishing the cross-domain semantic relationship network model includes:
[0020] An initial model of a cross-domain semantic relationship network was constructed based on TransR;
[0021] Construct a relational semantic space;
[0022] Each entity in the technical entity domain, the social entity domain, and the risk entity domain, as well as the corresponding cross-domain relationships, are vectorized to obtain multiple entity vectors, relation vectors corresponding to each entity vector, and projection matrices for projecting the corresponding entity vectors onto the relation semantic space.
[0023] Based on each of the projection matrices, the corresponding entity vectors are projected onto the relational semantic space, and then relational reasoning is performed to align the entities in different entity domains in the relational semantic space.
[0024] Based on the entities in different entity domains aligned in the relational semantic space and their corresponding cross-domain relationships, the cross-domain semantic relation network model is obtained.
[0025] Furthermore, the cross-domain relationships include: direct triggering relationships, used to describe direct causal relationships between factors; potential influence relationships, used to describe probabilistic influence relationships between factors; resource constraint relationships, used to describe the impact of resource allocation on activity execution; culture shaping relationships, used to describe the impact of organizational culture on behavioral patterns; and state-driven relationships, used to describe the impact of personnel state on operational behavior.
[0026] Furthermore, the knowledge graph-based nuclear power plant risk assessment method also includes:
[0027] The nuclear power plant risk assessment knowledge graph is dynamically updated based on real-time collected technical and social parameters of the nuclear power plant.
[0028] Based on historical and real-time data of nuclear power plant technical and social parameters, a machine learning model is constructed, and the relationship weights and risk probabilities in the nuclear power plant risk assessment knowledge graph are automatically adjusted through the machine learning model.
[0029] The risk assessment results for nuclear power plants are adjusted in real time based on the adjusted relationship weights and risk probabilities.
[0030] Furthermore, to dynamically update the nuclear power plant risk assessment knowledge graph, it is necessary to first preprocess the real-time collected technical and social parameters of the nuclear power plant:
[0031] Cross-modal feature alignment was performed on the collected technical and social parameters of nuclear power plants;
[0032] The data, after cross-modal feature alignment, is dynamically mapped to the corresponding entities and cross-domain relationships in the nuclear power plant risk assessment knowledge graph, and the nuclear power plant risk assessment knowledge graph is dynamically updated.
[0033] Furthermore, the machine learning model is constructed using the following method:
[0034] A graph neural network with a two-layer attention mechanism is used to learn the weights of semantic relations to obtain an adaptive adjustment model of relation weights. A dynamic calculation model of risk probability is obtained by modeling the probability of risk occurrence based on a Bayesian neural network.
[0035] The dual-layer attention mechanism graph neural network includes a node-level attention layer for calculating the importance weight of each entity in relation propagation, and a relation-level attention layer for calculating the dynamic weights of different types of semantic relations.
[0036] The dynamic risk probability calculation model includes an input layer for fusing entities and relationships, a hidden layer for constructing neurons, and an output layer.
[0037] Secondly, the present invention also provides a nuclear power plant risk assessment system based on a knowledge graph, comprising:
[0038] The data acquisition module is used to collect technical and social parameters of nuclear power plants in real time.
[0039] The risk assessment module is used to carry the nuclear power plant risk assessment knowledge graph as described in any of the above technical solutions, receive the data collected by the data acquisition module, input it into the nuclear power plant risk assessment knowledge graph, and output the nuclear power plant risk assessment results.
[0040] The risk assessment module also includes:
[0041] The knowledge graph dynamic update unit is used to perform cross-modal feature alignment on the collected nuclear power plant technical parameters and social parameters, and then dynamically map the cross-modal feature aligned data to the corresponding entities and cross-domain relationships in the nuclear power plant risk assessment knowledge graph, thereby dynamically updating the nuclear power plant risk assessment knowledge graph.
[0042] Furthermore, the risk assessment module also includes:
[0043] The risk assessment result dynamic update unit is used to carry the machine learning model, automatically adjust the relationship weights and risk probabilities in the nuclear power plant risk assessment knowledge graph, and then adjust the nuclear power plant risk assessment results in real time based on the adjusted relationship weights and risk probabilities.
[0044] Thirdly, the present invention also provides an electronic device, including a processor and a memory, wherein the processor is coupled to the memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the knowledge graph-based nuclear power plant risk assessment method as described above.
[0045] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer program that is executed by a processor to implement the knowledge graph-based nuclear power plant risk assessment method described above.
[0046] The present invention offers the following advantages: The knowledge graph-based nuclear power plant risk assessment method and system proposed in this invention address the technical problem of existing nuclear power plant risk assessment methods that separate social and technological factors, failing to fully reflect the essential characteristics of nuclear power plants as complex socio-technical systems. This leads to low accuracy in nuclear power plant risk assessment results. The present invention overcomes the problems and shortcomings of the aforementioned existing solutions by constructing a unified knowledge model that deeply integrates social and technological elements, achieving a more comprehensive, accurate, and dynamic risk assessment. Attached Figure Description
[0047] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will now be described in further detail with reference to the accompanying drawings, wherein:
[0048] Figure 1 This is a schematic diagram illustrating the various entity domains used to construct a knowledge graph according to an embodiment of the present invention.
[0049] Figure 2 This is a diagram illustrating a nuclear power plant maintenance embodiment of the present invention.
[0050] Figure 3 This is a cross-semantic relationship network diagram according to an embodiment of the present invention.
[0051] Figure 4 This is a diagram of a radical power generation project embodiment of the present invention.
[0052] Figure 5 This is a schematic diagram of the components of a knowledge graph-based nuclear power plant risk assessment system according to an embodiment of the present invention. Detailed Implementation
[0053] The technical solutions of some embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments disclosed in the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments disclosed in the present invention are within the scope of protection of the present invention. It should be noted that in the drawings with reference numerals, the same reference numerals and letters represent similar parts. Once a part is defined in a drawing, it will not be defined and explained again in subsequent drawings.
[0054] This invention can be applied to nuclear power plant risk assessment, solving the technical problem that existing methods for nuclear power plant risk assessment often separate social and technical factors, making it difficult to fully reflect the essential characteristics of nuclear power plants as complex socio-technical systems. This, in turn, leads to low accuracy in nuclear power plant risk assessment results.
[0055] The knowledge graph-based nuclear power plant risk assessment method and system disclosed in this invention have the following technical advantages:
[0056] (1) By constructing the knowledge graph of risk assessment for nuclear power plants, the scope of risk assessment is substantially expanded. Traditional methods mainly focus on purely technical factors such as equipment failure, while this invention systematically incorporates social factors such as organizational culture, management decisions, and personnel status into the assessment system, solving the core problem of the separation between social and technical factors, and making the risk assessment results more comprehensive and in line with the actual operating environment of nuclear power plants.
[0057] (2) Based on cross-domain semantic relationship networks, in-depth tracing of risk root causes is realized. Compared with existing technologies that can only identify surface causal relationships, this invention can trace the complete risk chain from organizational roots to technical consequences through semantic relationships such as direct triggers, potential impacts, and resource constraints, which greatly improves the depth and accuracy of risk analysis and provides technical support for formulating fundamental improvement measures.
[0058] (3) The introduction of dynamic sensing and evolution mechanism changes the traditional static assessment mode. Existing technologies mainly rely on periodic assessment and manual updates, while the present invention can sense changes in the power plant status in real time and automatically adjust the risk assessment results, realizing the transformation from post-analysis to pre-warning, and significantly improving the risk prevention and control capabilities of nuclear power plants.
[0059] (4) A unified ontology model framework was established, which solved the problem of integrating multi-source heterogeneous data. Compared with the scattered and isolated risk knowledge management methods in the existing technology, the present invention provides a standardized knowledge representation and storage framework, which enables the effective accumulation and sharing of operating experience and risk knowledge of different units and different periods, thereby improving the safety management level of the entire industry.
[0060] (5) Provides an intuitive and visual risk propagation path, greatly improving the decision-making experience. Traditional risk assessment results are often presented in the form of complex data reports, while this invention, through the visualization of knowledge graphs, can clearly present the formation mechanism and propagation path of risks, helping managers to quickly understand the nature of risks and formulate targeted measures, thereby improving the scientific nature and efficiency of decision-making.
[0061] To further illustrate the knowledge graph-based nuclear power plant risk assessment method and system provided by this invention, the following embodiments are disclosed.
[0062] In some embodiments, this embodiment provides a nuclear power plant risk assessment method based on knowledge graphs, including:
[0063] The technical entity domain, social entity domain, and risk entity domain related to nuclear power plants were obtained respectively;
[0064] Establish a cross-domain semantic relationship network model;
[0065] A knowledge graph for nuclear power plant risk assessment is constructed based on the cross-domain semantic relationship network model.
[0066] Based on the aforementioned nuclear power plant risk assessment knowledge graph, technical and social parameters of the nuclear power plant are collected in real time to obtain the nuclear power plant risk assessment results.
[0067] In some embodiments, please refer to Figure 1 The technical entity domain includes technical elements such as equipment, systems, functions, and failure modes, and is the main object of traditional risk assessment.
[0068] In some embodiments, please refer to Figure 1 The risk entity domain includes unsafe behaviors, unsafe states, events, and consequences, providing a common risk output interface for social and technical factors.
[0069] In some embodiments, please refer to Figure 1 The social entity domain includes the personnel entity domain, the organizational entity domain, and the management entity domain. Based on this, the technology entity domain, personnel entity domain, organizational entity domain, management entity domain, and risk entity domain form five core conceptual domains, creating a complete socio-technical system representation framework, which provides a basic model framework for constructing a knowledge graph for nuclear power plant risk assessment.
[0070] In some embodiments, please refer to Figure 1 The personnel entity domain includes individual operators, teams, roles, and dynamic cognitive states (such as situational awareness and workload), which for the first time incorporates the real-time cognitive state of personnel into the risk assessment system;
[0071] The organizational entity domain includes soft elements such as departmental structure, safety culture, and rules and regulations, which solves the technical problem that organizational factors are difficult to quantify and represent.
[0072] The management entity domain includes dynamic management activities such as training activities, maintenance tasks, and resource allocation, establishing a direct link between management decisions and risk status.
[0073] For example Figure 2 As shown, this is a specific application example. When a nuclear power plant is undergoing a major overhaul, the system can simultaneously characterize the process from technical, personnel, organizational, management, and risk perspectives.
[0074] In some embodiments, the organizational entity domain and the management entity domain are merged into an organizational management domain, and the risk entity domain is integrated into other domains as an attribute. For example, "unsafe behavior" is treated as a behavioral attribute of the personnel entity domain, rather than an independent entity. Appropriately merging the five core conceptual domains reduces model complexity while maintaining socio-technical integration. Although the model granularity becomes coarser, it still maintains the basic correlation between social and technological factors, making it suitable for scenarios with higher real-time requirements but slightly lower accuracy requirements.
[0075] In some embodiments, this embodiment also provides a nuclear power plant risk assessment method based on knowledge graphs, including:
[0076] Obtain the technical entity domain and risk entity domain related to nuclear power plants respectively, and add social dimension attributes to the technical entity domain;
[0077] Establish a cross-domain semantic relationship network model;
[0078] A knowledge graph for nuclear power plant risk assessment is constructed based on the cross-domain semantic relationship network model.
[0079] Based on the aforementioned nuclear power plant risk assessment knowledge graph, technical and social parameters of the nuclear power plant are collected in real time to obtain the nuclear power plant risk assessment results.
[0080] For example, social attributes such as maintenance team competence ratings and management system completeness can be added to pump equipment entities, and risk reasoning can be achieved through attribute relationships. This approach is compatible with existing technology maps, has low modification costs, and can still achieve basic social-technical risk analysis.
[0081] In some embodiments, please refer to Figure 3 The semantic relationships include direct triggering relationships, potential influence relationships, resource constraint relationships, culture shaping relationships, and state-driven relationships. Through carefully designed semantic relationships, various conceptual domains are closely connected to form a relationship network capable of simulating real-world risk propagation paths.
[0082] In some embodiments, establishing the cross-domain semantic relationship network includes:
[0083] Establishing the cross-domain semantic relationship network model includes:
[0084] An initial model of a cross-domain semantic relationship network was constructed based on TransR;
[0085] Construct a relational semantic space;
[0086] Each entity in the technical entity domain, the social entity domain, and the risk entity domain, as well as the corresponding cross-domain relationships, are vectorized to obtain multiple entity vectors, relation vectors corresponding to each entity vector, and projection matrices for projecting the corresponding entity vectors onto the relation semantic space.
[0087] Based on each of the projection matrices, the corresponding entity vectors are projected onto the relational semantic space, and then relational reasoning is performed to align the entities in different entity domains in the relational semantic space.
[0088] Based on the entities in different entity domains aligned in the relational semantic space and their corresponding cross-domain relationships, the cross-domain semantic relation network model is obtained.
[0089] Specifically, to achieve complex and heterogeneous semantic relationships between social and technological factors, this invention employs TransR (Translation-based Embedding for Relation) technology as the core modeling method for cross-domain semantic relationship networks. TransR can map entities in different conceptual domains to different semantic spaces and achieve accurate relational reasoning through relation-specific transformation matrices, making it particularly suitable for modeling the multi-domain and multi-type relational needs in the nuclear power plant risk assessment knowledge graph.
[0090] In some embodiments, please refer to Figure 3 The cross-domain relationships include:
[0091] Category 1: Direct triggering relationship, describing the direct causal relationship between factors, such as "operational error" directly triggering "equipment malfunction";
[0092] The second category is: potential impact relationships, which describe the probabilistic impact relationships between factors, such as the potential impact of "insufficient training" on "probability of emergency response failure";
[0093] The third category: resource constraints, which describe the impact of resource allocation on the execution of activities, such as "budget cuts" constraining "preventive maintenance quality";
[0094] Category 4: Culture-Shaping Relationships, describing the influence of organizational culture on behavioral patterns, such as "punitive culture" shaping "concealment bias";
[0095] The fifth category is state-driven relationships, which describe the impact of personnel state on operational behavior, such as "high fatigue" state driving "simplified operation process".
[0096] In some embodiments, the entities in the technical entity domain, the social entity domain, and the risk entity domain, along with their corresponding cross-domain relationships, are represented using vectorization. For example, each entity (such as "operational error," "steam generator," or "maintenance team") is represented as a uniform low-dimensional vector, denoted as [vector name missing]. Each semantic relation (such as "direct trigger" or "potential impact") is also represented as a vector and associated with a relation-specific projection matrix. This is used to project entity vectors from the entity space to the relational semantic space.
[0097] In some embodiments, based on the respective projection matrices, the corresponding entity vectors are projected into the relational semantic space, and then relational reasoning is performed to align the entities in different entity domains in the relational semantic space. For example, for any triplet... (For example, "operational error → direct trigger → device malfunction"), TransR achieves relational reasoning through the following steps:
[0098] Head entity vector Tail entity vector Through relationships Corresponding projection matrix Projection to Relationship In a specific space: ;
[0099] In the relational semantic space, relational reasoning is performed using the vector translation hypothesis: ;in, It is a relationship Vector representation in relational semantic space.
[0100] This mechanism enables entities in different conceptual domains to be aligned in the relational semantic space, thereby effectively modeling the complex relationships between "social domain entities → technological domain entities".
[0101] In some embodiments, an application example in risk reasoning is provided, where when the system detects an "operational error" event, its entity vector e_"error" is extracted. All potentially associated relationships are then queried within the knowledge graph. Tail-end entity The TransR model is used to calculate the distance between e_"error" and each candidate tail entity in the relational semantic space, and the relation-entity pair with the smallest distance is selected as the reasoning result. For example, if the distance between the "direct trigger" relation and the "equipment malfunction" entity is the smallest, the system determines that the operational error may directly trigger the equipment malfunction and activates the corresponding risk assessment and early warning process.
[0102] like Figure 4The diagram illustrates a specific implementation. In a typical risk attribution analysis, the system can deduce the complete socio-technical risk chain through a relational network. This reasoning process changes the traditional isolation between social and technological factors in risk assessment, enabling true root cause analysis.
[0103] In some embodiments, the knowledge graph-based nuclear power plant risk assessment method further includes:
[0104] The nuclear power plant risk assessment knowledge graph is dynamically updated based on real-time collected technical and social parameters of the nuclear power plant.
[0105] Based on historical and real-time data of nuclear power plant technical and social parameters, a machine learning model is constructed, and the relationship weights and risk probabilities in the nuclear power plant risk assessment knowledge graph are automatically adjusted through the machine learning model.
[0106] The risk assessment results for nuclear power plants are adjusted in real time based on the adjusted relationship weights and risk probabilities.
[0107] Establish a dynamic update mechanism based on multi-source data fusion to enable the knowledge graph to continuously evolve. This will allow for an adaptable risk assessment process to address the different situations at different nuclear power plants, resulting in risk assessment outcomes tailored to the specific circumstances.
[0108] In some embodiments, the real-time collected technical and social parameters of the nuclear power plant include: real-time data on equipment status, personnel operation, and management activities collected from systems such as DCS, SIS, and EAM.
[0109] In some embodiments, dynamically updating the nuclear power plant risk assessment knowledge graph requires preprocessing the real-time collected technical and social parameters of the nuclear power plant:
[0110] Cross-modal feature alignment was performed on the collected technical and social parameters of nuclear power plants;
[0111] The data, after cross-modal feature alignment, is dynamically mapped to the corresponding entities and cross-domain relationships in the nuclear power plant risk assessment knowledge graph, and the nuclear power plant risk assessment knowledge graph is dynamically updated.
[0112] In this embodiment, through state perception and mapping, real-time data is mapped to corresponding entities and relationships in the knowledge graph, realizing the transformation from a "static knowledge base" to a "dynamic perception system".
[0113] In some embodiments, cross-modal feature alignment of collected nuclear power plant technical and social parameters includes: constructing multimodal data feature representations and cross-modal semantic alignment;
[0114] The multimodal data feature representation includes establishing a unified feature representation framework for each type of data modality.
[0115] For example, for unstructured text data, given a text sequence Semantic representation vectors are obtained through domain-pre-trained language models. :
[0116] ;
[0117] in, It can adopt architectures such as BERT and RoBERTa, and make fine adjustments on the corpus of nuclear power plants. The dimension is The real vector space.
[0118] For structured time-series data, suppose the sensor sampling sequence of a certain device within a time window is as follows: Extract its state feature vector through a time encoder : ;in, This is the semantic vector obtained by encoding the input sequence S using a Long Short-Term Memory (LSTM) network. This is the semantic vector obtained by the Transformer encoder encoding the input sequence S. The dimension is The real vector space.
[0119] For discrete state data, such as work order status and alarm category, an embedding layer is used to map it into a dense vector. :
[0120] ;in, For a trainable embedding matrix, It is a one-hot vector. For dimension The real vector space.
[0121] Cross-modal feature alignment of collected nuclear power plant technical and social parameters is a unified processing of structured data (sensor readings) and unstructured data (work logs, meeting minutes). Through cross-modal feature alignment and knowledge consistency fusion mechanism, it provides high-quality and traceable data input for dynamic knowledge graphs.
[0122] To achieve alignment of data from different modalities within a unified semantic space, the cross-modal semantic alignment employs a contrastive learning objective function for joint training. Let the text description vectors from the same event instance be... The corresponding sensor feature vector is The alignment loss function is then defined as:
[0123] ;
[0124] in, For cosine similarity, For temperature coefficient, denoted as the number of negative samples in the batch. This loss function encourages different modal vectors describing the same event to be closer to each other in the semantic space.
[0125] Here is an example of cross-modal feature alignment of collected technical and social parameters of nuclear power plants:
[0126] Taking the handling of "abnormal pump vibration" incidents as an example,
[0127] enter:
[0128] Text log: "Inspector reports abnormal vibration of P-101 pump, with a muffled sound."
[0129] Sensor data: Peak vibration acceleration sequence (Exceeding the threshold).
[0130] Maintenance work order: Equipment ID=P-101, Status=“Abnormal”, Priority=“High”.
[0131] Processing procedure:
[0132] Text encoding: ;
[0133] Sequence encoding: ;
[0134] Cross-modal alignment: Using a trained alignment model, ... and Mapped to the same semantic space.
[0135] Entity Link: Link “P-101 Pump” to the corresponding device node in the diagram.
[0136] Status Update: Update the "health status" attribute of the device node according to the fusion vector, and establish a semantic chain of "inspector → report → abnormal pump vibration → potential fault".
[0137] For the machine learning model portion of the adaptive evolution, to achieve dynamic adjustment of relation weights and risk probabilities in the knowledge graph, this invention employs an adaptive evolution framework based on machine learning and online learning. This framework not only supports real-time updates of the knowledge graph but also continuously optimizes the accuracy and timeliness of risk assessment based on power plant operation data.
[0138] In some embodiments, the machine learning model is constructed using the following methods:
[0139] A two-layer attention-based graph neural network (GNN) is used to learn the weights of semantic relationships, resulting in an adaptive adjustment model for relationship weights. A dynamic calculation model for risk probability is obtained by modeling the probability of risk occurrence using a Bayesian neural network.
[0140] The dual-layer attention mechanism graph neural network includes a node-level attention layer for calculating the importance weight of each entity in relation propagation, and a relation-level attention layer for calculating the dynamic weights of different types of semantic relations.
[0141] Specifically, at the node-level attention layer, the importance weight of each entity node in relation propagation is calculated:
[0142] ;
[0143] in, , , For entities eigenvectors, The weight matrix is a learnable matrix. For attention vectors, For nodes The set of neighbors.
[0144] Specifically, the relation-level attention layer assigns dynamic weights to different types of semantic relations. :
[0145] ;
[0146] in, For relation vectors, , , For trainable parameters, This is the Sigmoid function.
[0147] The model's weight update mechanism is relation weights. The adjustment adopts the online gradient descent method, and each time a new batch of event data is received... This means performing a weight update once:
[0148] ;
[0149] in For learning rate, For entity vectors and For the tail entity vector, It is a relationship Vector representation in relational semantic space, loss function Using marginal ranking loss:
[0150] ;
[0151] in, For marginal parameters, For the TransR scoring function, For negative sampling triples.
[0152] The dynamic risk probability calculation model uses a Bayesian Neural Network (BNN) to model the probability of risk occurrence. The network structure is a three-layer fully connected network, including an input layer, a hidden layer, and an output layer.
[0153] in,
[0154] Input layer: Feature vector fused from entity and relation data, with the optimal dimension selected using... ;
[0155] Hidden layers: 2 layers, 128 neurons per layer, preferably using ReLU activation;
[0156] Output layer: 1 neuron, using the sigmoid function to output probability. .
[0157] The model's probability update formula is given new evidence. (e.g., "an increase in the number of operational errors"), risk events The posterior probability is calculated using Bayesian update:
[0158] ;
[0159] in, This is the prior probability, obtained from historical data statistics; The likelihood function is learned by a BNN; the posterior probability is... The probability attributes of corresponding risk nodes in the knowledge graph will be updated in real time. The model's online learning strategy uses variational inference to update the BNN parameters online, performing parameter retraining every 24 hours (or when the accumulated data reaches 1000 records), with 1000 iterations per iteration and a batch size of... .
[0160] Here is a specific example of real-time adjustment of nuclear power plant risk assessment results, which demonstrates the system evolution process:
[0161] The system detected that "the main pump vibration value exceeded the threshold three times consecutively":
[0162] Data input: Vibration sensor sequence Maintenance log text ;
[0163] Feature extraction: , ;
[0164] Relationship weight adjustment: Update the relationship weights of "vibration anomaly → potential fault" through the GNN attention mechanism;
[0165] Probability update: The probability of "main pump failure" calculated by BNN has been increased from 0.15 to 0.48;
[0166] Anomaly detection: The autoencoder reconstruction error exceeds the threshold, triggering an anomaly flag;
[0167] Knowledge Accumulation: Generate a new rule, “High-frequency vibration + lack of timely maintenance → increased probability of failure,” and store it in the nuclear power plant risk assessment knowledge graph.
[0168] In some embodiments, please refer to Figure 5 This embodiment also provides a nuclear power plant risk assessment system based on knowledge graphs, including:
[0169] The data acquisition module is used to collect technical and social parameters of nuclear power plants in real time.
[0170] The risk assessment module is used to carry the nuclear power plant risk assessment knowledge graph as described in any of the above technical solutions, receive the data collected by the data acquisition module, input it into the nuclear power plant risk assessment knowledge graph, and output the nuclear power plant risk assessment results.
[0171] The risk assessment module also includes:
[0172] The knowledge graph dynamic update unit is used to perform cross-modal feature alignment on the collected nuclear power plant technical parameters and social parameters, and then dynamically map the cross-modal feature aligned data to the corresponding entities and cross-domain relationships in the nuclear power plant risk assessment knowledge graph, thereby dynamically updating the nuclear power plant risk assessment knowledge graph.
[0173] In some embodiments, the risk assessment module further includes:
[0174] The risk assessment result dynamic update unit is used to carry the machine learning model, automatically adjust the relationship weights and risk probabilities in the nuclear power plant risk assessment knowledge graph, and then adjust the nuclear power plant risk assessment results in real time based on the adjusted relationship weights and risk probabilities.
[0175] In some embodiments, an electronic device is also provided, including a processor and a memory coupled to the memory; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the knowledge graph-based nuclear power plant risk assessment method as described above.
[0176] In some embodiments, a computer-readable storage medium is also provided for storing a computer program that is executed by a processor to implement the knowledge graph-based nuclear power plant risk assessment method as described above.
[0177] The core innovation of this invention lies in constructing a social-technological integrated knowledge graph ontology model. This model deeply integrates social factors such as organization, management, and personnel with technical factors such as equipment, systems, and faults through a cross-domain semantic relationship network, and establishes a dynamic perception and evolution mechanism to achieve comprehensive, accurate, and real-time risk assessment of nuclear power plants. Specifically, the innovation of this invention is reflected in the following three aspects:
[0178] (1) A unified ontology model framework was established, defining five core concept domains: technology domain, personnel domain, organization domain, management domain, and risk domain, thus solving the problem of the separation between social and technological factors in existing technologies.
[0179] (2) A cross-domain semantic relationship network was designed, which realizes the deep association and cross-reasoning of social factors and technological factors through semantic relationships such as "direct trigger", "potential influence" and "resource constraint".
[0180] (3) A dynamic perception and evolution mechanism was introduced. Through real-time data access and machine learning technology, the knowledge graph can be dynamically updated with the operating status of the nuclear power plant, overcoming the limitations of the traditional static model.
[0181] Here is a specific implementation example: In the daily operation of a nuclear power plant, this system can monitor the workload indicators of the main control room operators in real time (such as alarm handling frequency and number of operation steps). When the workload is detected to be continuously higher than a threshold, it automatically strengthens the weight of the potential impact relationship between high workload and the probability of cognitive error; at the same time, it issues a warning to management: "The current workload level will lead to an increase of X% in the probability of human error"; based on actual event data, it continuously adjusts the relationship weights to improve prediction accuracy. This mechanism overcomes the shortcomings of traditional static models that are difficult to adapt to changes in the operating environment, making risk assessment results more accurate and reliable.
[0182] In the description disclosed in this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to has a specific orientation, or is constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the disclosure of this invention.
[0183] Unless the context otherwise requires, throughout the specification and claims, the term "comprising" is interpreted as open-ended and encompassing, meaning "including, but not limited to." In the description, terms such as "one embodiment," "some embodiments," "exemplary embodiment," "exemplary," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example disclosed in the invention. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics mentioned may be included in any suitable manner in any one or more embodiments or examples.
[0184] The terms "first" and "second" are used merely to distinguish different descriptive objects and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated; that is, they do not limit the position, order, priority, quantity, or content of the described objects. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments disclosed in this invention, unless otherwise stated, "a plurality of" means two or more.
[0185] In describing some embodiments, the term "connection" and its derivative expressions may be used. For example, the term "connection" may be used in describing some embodiments to indicate that two or more components have direct physical or electrical contact with each other. The embodiments disclosed herein are not necessarily limited to the content of this document.
[0186] "At least one of A, B, and C" has the same meaning as "at least one of A, B, or C," both including the following combinations of A, B, and C: only A, only B, only C, a combination of A and B, a combination of A and C, a combination of B and C, and a combination of A, B, and C. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B; "A and / or B" includes the following three combinations: only A, only B, and a combination of A and B.
[0187] The use of “configured as” in this article implies an open and inclusive language that does not exclude the applicability to or configuration of devices to perform additional tasks or steps.
[0188] In addition, the use of “based on” implies openness and inclusivity, because processes, steps, calculations or other actions “based on” one or more of the stated conditions or values may in practice be based on additional conditions or values beyond those stated.
[0189] As used herein, “about” and “approximately” include the values stated and the average values within an acceptable range of deviation from a particular value, wherein the acceptable range of deviation is determined by a person skilled in the art taking into account the measurement under discussion and the error associated with the measurement of the particular quantity (i.e., the limitations of the measurement system).
[0190] This document describes exemplary embodiments with reference to cross-sectional views and / or plan views, which are idealized exemplary drawings. In the drawings, the thickness of layers and regions is enlarged for clarity. Therefore, variations in shape relative to the drawings are contemplated due to, for example, manufacturing techniques and / or tolerances. Thus, exemplary embodiments should not be construed as limited to the shapes of the regions shown herein, but rather include shape deviations due to, for example, manufacturing processes. For example, etched regions shown as rectangular would typically have curved features. Therefore, the regions shown in the drawings are schematic in nature, and their shapes are not intended to show the actual shapes of the regions of the device, nor are they intended to limit the scope of the exemplary embodiments.
[0191] The above description is merely a specific embodiment of the present invention, but the scope of protection disclosed in the present invention is not limited thereto. Any variations or substitutions conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection disclosed in the present invention. Therefore, the scope of protection disclosed in the present invention should be determined by the scope of the claims.
Claims
1. A nuclear power plant risk assessment method based on knowledge graphs, characterized in that, include: The technical entity domain, social entity domain, and risk entity domain related to nuclear power plants were obtained respectively; Establish a cross-domain semantic relationship network model; A knowledge graph for nuclear power plant risk assessment is constructed based on the cross-domain semantic relationship network model. Based on the aforementioned nuclear power plant risk assessment knowledge graph, technical and social parameters of the nuclear power plant are collected in real time to obtain the nuclear power plant risk assessment results.
2. The nuclear power plant risk assessment method based on knowledge graphs according to claim 1, characterized in that, Establishing the cross-domain semantic relationship network model includes: An initial model of a cross-domain semantic relationship network was constructed based on TransR; Construct a relational semantic space; Each entity in the technical entity domain, the social entity domain, and the risk entity domain, as well as the corresponding cross-domain relationships, are vectorized to obtain multiple entity vectors, relation vectors corresponding to each entity vector, and projection matrices for projecting the corresponding entity vectors onto the relation semantic space. Based on each of the projection matrices, the corresponding entity vectors are projected onto the relational semantic space, and then relational reasoning is performed to align the entities in different entity domains in the relational semantic space. Based on the entities in different entity domains aligned in the relational semantic space and their corresponding cross-domain relationships, the cross-domain semantic relation network model is obtained.
3. The nuclear power plant risk assessment method based on knowledge graphs according to claim 1, characterized in that, The cross-domain relationships include: direct triggering relationships, used to describe direct causal relationships between factors; potential influence relationships, used to describe probabilistic influence relationships between factors; resource constraint relationships, used to describe the impact of resource allocation on activity execution; culture shaping relationships, used to describe the impact of organizational culture on behavioral patterns; and state-driven relationships, used to describe the impact of personnel state on operational behavior.
4. The nuclear power plant risk assessment method based on knowledge graphs according to claim 1, characterized in that, Also includes: The nuclear power plant risk assessment knowledge graph is dynamically updated based on real-time collected technical and social parameters of the nuclear power plant. Based on historical and real-time data of nuclear power plant technical and social parameters, a machine learning model is constructed, and the relationship weights and risk probabilities in the nuclear power plant risk assessment knowledge graph are automatically adjusted through the machine learning model. The risk assessment results for nuclear power plants are adjusted in real time based on the adjusted relationship weights and risk probabilities.
5. The nuclear power plant risk assessment method based on knowledge graphs according to claim 1, characterized in that, To dynamically update the nuclear power plant risk assessment knowledge graph, it is necessary to first preprocess the real-time collected technical and social parameters of the nuclear power plant. Cross-modal feature alignment was performed on the collected technical and social parameters of nuclear power plants; The data, after cross-modal feature alignment, is dynamically mapped to the corresponding entities and cross-domain relationships in the nuclear power plant risk assessment knowledge graph, and the nuclear power plant risk assessment knowledge graph is dynamically updated.
6. The nuclear power plant risk assessment method based on knowledge graphs according to claim 1, characterized in that, The machine learning model is constructed using the following method: A graph neural network with a two-layer attention mechanism is used to learn the weights of semantic relations to obtain an adaptive adjustment model of relation weights. A dynamic calculation model of risk probability is obtained by modeling the probability of risk occurrence based on a Bayesian neural network. The dual-layer attention mechanism graph neural network includes a node-level attention layer for calculating the importance weight of each entity in relation propagation, and a relation-level attention layer for calculating the dynamic weights of different types of semantic relations. The dynamic risk probability calculation model includes an input layer for fusing entities and relationships, a hidden layer for constructing neurons, and an output layer.
7. A nuclear power plant risk assessment system based on knowledge graphs, characterized in that, include: The data acquisition module is used to collect technical and social parameters of nuclear power plants in real time. The risk assessment module is used to carry the nuclear power plant risk assessment knowledge graph as described in any one of claims 1 to 6, receive the data collected by the data acquisition module, input it into the nuclear power plant risk assessment knowledge graph, and output the nuclear power plant risk assessment result. The risk assessment module also includes: The knowledge graph dynamic update unit is used to perform cross-modal feature alignment on the collected nuclear power plant technical parameters and social parameters, and then dynamically map the cross-modal feature aligned data to the corresponding entities and cross-domain relationships in the nuclear power plant risk assessment knowledge graph, thereby dynamically updating the nuclear power plant risk assessment knowledge graph.
8. The nuclear power plant risk assessment system based on knowledge graphs according to claim 1, characterized in that, The risk assessment module also includes: The risk assessment result dynamic update unit is used to carry the machine learning model, automatically adjust the relationship weights and risk probabilities in the nuclear power plant risk assessment knowledge graph, and then adjust the nuclear power plant risk assessment results in real time based on the adjusted relationship weights and risk probabilities.
9. An electronic device, characterized in that, Includes a processor and a memory, wherein the processor is coupled to the memory; The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the knowledge graph-based nuclear power plant risk assessment method as described in any one of claims 1 to 6.
10. A computer-readable storage medium for storing computer programs, characterized in that, The computer program is executed by a processor to implement the knowledge graph-based nuclear power plant risk assessment method as described in any one of claims 1 to 6.