Simulation data intelligent management method and system for digital reactor

Abstract

The application discloses a simulation data intelligent management method and system for a digital reactor, relates to the technical field of data processing, and solves the technical problem that the value evaluation dimension of the prior art is single, the difference of data in physical entity correlation, time dynamic characteristics and other dimensions is ignored, the management strategies of high-value core data and low-value auxiliary data are the same, and the simulation data intelligent management method is low in efficiency; a dynamic data blood relationship diagram is constructed through physical laws; the value potential of a node is calculated through field source intensity and mixed distance to quantify the influence of physical entities; then, an improved random walk algorithm is used to simulate value flow in a flow network to obtain a value score; finally, dynamic stability statistical data is introduced to construct a decision matrix to generate a differentiated whole life cycle management strategy, the limitation of single dimension evaluation is broken through, differentiated strategy management of high-value core data and low-value auxiliary data is realized, and the efficiency and precision of simulation data intelligent management are improved.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, specifically a method and system for intelligent management of simulation data for digital reactors. Background Technology

[0002] As a core technology carrier in the field of nuclear power, digital reactors generate a large amount of simulation data under multiple operating conditions and in multiple dimensions during their simulation process. This data covers multiple physical processes such as neutron transport, thermal-hydraulic, and structural mechanics, and is characterized by large data scale, complex correlation, and strong dynamic evolution.

[0003] Current traditional simulation data management methods rely on a single dimension for node value assessment. Most methods prioritize data based solely on data size or access frequency, neglecting differences in physical entity relationships and temporal dynamics. This results in the same management strategies being applied to high-value core data and low-value auxiliary data, leading to low efficiency in intelligent simulation data management methods. Therefore, intelligent management methods for digital reactor simulation data still require further improvement. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a method and system for intelligent management of simulation data for digital reactors, which addresses the technical problem that the node value assessment dimension in the prior art is singular, ignoring the differences in data in dimensions such as physical entity association and time dynamic characteristics, resulting in the same management strategy for high-value core data and low-value auxiliary data, leading to low efficiency of intelligent management methods for simulation data.

[0005] To achieve the above objectives, the first aspect of this application provides a method for intelligent management of simulation data for digital reactors, comprising:

[0006] Acquire digital reactor simulation data and corresponding physical laws; construct a physical information loss function based on the physical laws; train a graph neural network based on the physical information loss function; and perform graph structure modeling on the digital reactor simulation data based on the graph neural network to generate a dynamic evolution data lineage diagram.

[0007] A physical entity knowledge base is acquired, analyzed, a key field source set is defined, and the corresponding field source intensities are obtained; the mixing distance from the key field source set to each node in the dynamic evolution data lineage graph is obtained; and the value potential of each node is obtained based on the field source intensities and the mixing distances.

[0008] The dynamic evolution data lineage graph is treated as a flow network, and the value potential of each node is regarded as potential energy. The value flow is simulated and iterated based on an improved random walk algorithm to obtain the value score of each node.

[0009] The dynamic stability statistics of digital reactor simulation data are obtained, and a decision matrix is ​​constructed by combining the value scores of each node. Based on the decision matrix, the simulation data management strategy of digital reactor is obtained by performing full life cycle management.

[0010] This application generates a dynamic data lineage graph that integrates physical correlation and temporal evolution by constructing a physical information loss function based on physical laws and training a graph neural network. Then, combining this with a physical entity knowledge base, it calculates the value potential of nodes using field source strength and hybrid distance to quantify the influence of physical entities. Next, an improved random walk algorithm is used to simulate value flow in a flow network, obtaining a dynamic value score. Finally, dynamic stability statistics are introduced to construct a decision matrix to generate differentiated full lifecycle management strategies. This approach breaks through the limitations of traditional single-dimensional assessment. By integrating multiple technologies such as physical correlation embedding, spatiotemporal dynamic modeling, field source potential quantification, and value flow simulation, it achieves accurate differentiation and differentiated strategy management of high-value core data and low-value auxiliary data, thereby significantly improving the efficiency and accuracy of intelligent management of simulation data.

[0011] Furthermore, the acquisition of digital reactor simulation data and corresponding physical laws; the construction of a physical information loss function based on the physical laws; the training of a graph neural network based on the physical information loss function; and the generation of a dynamic evolution data lineage graph based on the graph neural network for graph structure modeling of the digital reactor simulation data, including:

[0012] Acquire digital reactor simulation data;

[0013] The physical laws corresponding to digital reactor simulation data are based on the mass conservation equation, energy conservation equation, and momentum conservation equation of digital reactors.

[0014] The connection relationships of each node are obtained based on digital reactor simulation data as real labels, and the probability distribution of predicted node connections is output based on graph neural network.

[0015] Based on the true labels and the probability distribution, and combined with the cross-entropy loss formula, a data fitting loss term is constructed;

[0016] Calculate the residuals of the mass conservation equation, energy conservation equation, and momentum conservation equation in the physical laws respectively, and construct the physical constraint loss term through the residuals;

[0017] A physical information loss function is constructed based on the data fitting loss term and the physical constraint loss term.

[0018] A reactor correlation model is obtained by training a graph neural network based on a physical information loss function;

[0019] Digital reactor simulation data is input into the reactor correlation model to obtain a dynamic evolution data lineage diagram.

[0020] This application acquires digital reactor simulation data and uses the three conservation equations of mass, energy, and momentum as the core physical laws. In the training of the graph neural network, on the one hand, it utilizes the real labels and predicted distributions of node link relationships to construct a data fitting loss term based on cross-entropy to ensure topological accuracy. On the other hand, it calculates the residuals of the three conservation equations to construct a physical constraint loss term, thereby embedding physical laws as hard constraints into the model. By fusing the two losses to construct a physical information loss function, a reactor association model is trained, ultimately generating a dynamic evolution data lineage graph. Through a training method that deeply integrates data-driven and physical mechanisms, the generated dynamic evolution data lineage graph not only accurately reflects the association topology between data but also strictly conforms to the physical conservation laws of the reactor system, thus significantly improving the physical consistency and interpretability of the lineage graph and laying a solid foundation for subsequent multi-dimensional and highly reliable value assessment.

[0021] Furthermore, acquiring digital reactor simulation data includes:

[0022] Acquire platform interface data and reactor operation history data; the platform interface data refers to the data corresponding to the output interface of the digital reactor simulation platform.

[0023] Extract time-series and spatial distribution data of neutron transport and thermal-hydraulic processes from platform interface data;

[0024] Extract historical time-series and historical spatial data of neutron transport and thermal-hydraulic processes from reactor operation history data;

[0025] The time-series data, the spatial distribution data, the historical time-series data, and the historical spatial data are filtered and integrated under multiple operating conditions to generate digital reactor simulation data.

[0026] Furthermore, the process of acquiring a physical entity knowledge base, analyzing the physical entity knowledge base, defining a key field source set and obtaining the corresponding field source intensities; obtaining the mixing distance from the key field source set to each node in the dynamic evolution data lineage graph; and obtaining the value potential of each node based on the field source intensities and the mixing distance includes:

[0027] The physical entity knowledge base is obtained through data reactor information, reactor knowledge graph, and the physical model built into the digital reactor simulation platform.

[0028] The importance of entities in the physical entity knowledge base is evaluated to obtain a set of key field sources, and the importance of each key field source is assigned to obtain its corresponding field source intensity.

[0029] The entity refers to several physical objects related to the digital reactor, including but not limited to reactor core components;

[0030] Obtain the graph semantic distance and physical spatial distance of each node in the dynamic evolution data lineage graph of the key field source set, and calculate the mixed distance by weighted summation based on the graph semantic distance and physical spatial distance;

[0031] The value potential of each node is determined based on the source strength of each key source and the mixing distance of its corresponding node.

[0032] Furthermore, the step of obtaining the graph semantic distance and physical spatial distance of each node in the dynamic evolution data lineage graph of the key field source set, and calculating the mixed distance by weighted summation based on the graph semantic distance and physical spatial distance, includes:

[0033] Obtain the topological structure of the pedigree graph of dynamic evolutionary data;

[0034] Based on the aforementioned topology, the shortest path from each key source in the key source set to each node is obtained, and the graph semantic distance is determined based on the shortest path.

[0035] Based on the built-in physical model of the digital reactor simulation platform, spatial coordinate data of the key field source set is obtained, and spatial coordinate data of each node in the dynamic evolution data lineage diagram is obtained.

[0036] The physical spatial distance is determined based on the spatial coordinate data of each node in the lineage diagram of the key field source set and the state evolution data.

[0037] The mixed distance is calculated by weighted summation of the semantic distance and physical distance.

[0038] This application constructs a physical entity knowledge base by integrating multi-source information such as data reactor data, knowledge graphs, and built-in physical models. From this base, key physical entities are evaluated and defined as field sources, and their importance is assessed as field source strength. Furthermore, to comprehensively measure the association between nodes and field sources, a hybrid distance concept is proposed. This hybrid distance is calculated by weighted fusion of two distances: first, a graph semantic distance measured by the shortest path length based on the dynamic evolution data lineage graph topology, used to capture the logical and evolutionary associations between data; and second, a physical spatial distance calculated based on simulation platform coordinate data, used to reflect the real spatial relationships between entities. Finally, the value potential of each node is calculated based on the field source strength and the hybrid distance. By defining key field sources through multi-source knowledge fusion and utilizing the hybrid distance that fuses topological and spatial associations, a rich and multi-dimensional quantitative assessment of the physical connotation of node value potential is achieved. This effectively overcomes the shortcomings of traditional methods that ignore physical entities and spatial characteristics, making the value assessment results more consistent with the physical reality of digital reactors and the logic of data associations.

[0039] Furthermore, the dynamic evolution data lineage graph is treated as a flow network, and the value potential of each node is considered as potential energy; the value flow is simulated and iterated based on an improved random walk algorithm to obtain the value score of each node, including:

[0040] Obtain the potential energy of several nodes;

[0041] Calculate the potential energy difference between two adjacent nodes, and determine the improved transition probability based on the potential energy difference;

[0042] An improved random walk algorithm is obtained by introducing improved transition probabilities into the random walk algorithm;

[0043] The improved random walk algorithm is used to simulate and iterate the value flow in the flow network, obtain the flow data of each node in several iterations, and calculate the flow change rate of each node between two adjacent iterations.

[0044] When the rate of change of traffic at each node is less than the threshold for the rate of change of traffic, the simulation iteration stops, the traffic data of each node is output, and the value score of each node is determined based on the traffic data of each node.

[0045] Otherwise, continue with the simulation iteration.

[0046] Furthermore, the calculation of the potential energy difference between two adjacent nodes and the determination of the improved transition probability based on the potential energy difference include:

[0047] Set a target node and obtain the potential energy difference between all adjacent nodes of the target node;

[0048] The total potential energy difference of the target node is obtained by summing the potential energy differences of all its adjacent nodes.

[0049] The transition probability from the target node to each of its neighboring nodes is obtained by dividing the potential energy difference of each neighboring node by the total potential energy difference.

[0050] The transition probability is defined as the improved transition probability.

[0051] Furthermore, the process of acquiring dynamic stability statistics of digital reactor simulation data, analyzing and constructing a decision matrix in conjunction with the value scores of each node, and then executing full lifecycle management based on the decision matrix to obtain a digital reactor simulation data management strategy includes:

[0052] Obtain the iteration records of digital reactor simulation data from dynamic stability statistics, and statistically analyze the update frequency, version consistency, and historical retention duration of digital reactor simulation data;

[0053] The dynamic stability statistics and the value scores of each node are standardized.

[0054] The decision matrix is ​​determined based on standardized dynamic stability statistics and the value scores of each node;

[0055] Set value scoring thresholds and dynamic stability statistics thresholds, construct a decision coordinate system with value scoring as the horizontal axis, dynamic stability statistics as the vertical axis, and the coordinate combination of value scoring thresholds and dynamic stability statistics thresholds as the origin;

[0056] By mapping the decision matrix to the decision coordinate system, formulating full lifecycle management rules, and dividing the decision matrix into quadrants in the decision coordinate system, and combining the full lifecycle management rules for management, a simulation data management strategy for digital reactors is obtained.

[0057] Furthermore, the process of mapping the decision matrix to a decision coordinate system, formulating full lifecycle management rules, and managing the data based on the quadrant division of the decision matrix in the decision coordinate system, combined with the full lifecycle management rules, yields a simulation data management strategy for the digital reactor, including:

[0058] When the decision matrix is ​​mapped to the first quadrant of the decision coordinate system, the decision matrix category is set to high-value, high-stability data.

[0059] When the decision matrix is ​​mapped to the second quadrant of the decision coordinate system, the decision matrix category is set to high-value, low-stability data.

[0060] When the decision matrix is ​​mapped to the third quadrant of the decision coordinate system, the decision matrix category is set to low-value, high-stability data.

[0061] When the decision matrix is ​​mapped to the fourth quadrant of the decision coordinate system, the decision matrix category is set to low-value, low-stability data.

[0062] Points on the positive half-axis of the horizontal coordinate, points on the positive half-axis of the vertical coordinate, and the origin of the decision coordinate system are assigned to the first quadrant; points on the negative half-axis of the horizontal coordinate are assigned to the second quadrant; and points on the negative half-axis of the vertical coordinate are assigned to the third quadrant.

[0063] The decision matrix categories are used to obtain the simulation data management strategy for digital reactors through full lifecycle management rules;

[0064] The full lifecycle management rules include: configuring high-performance storage for high-value, high-stability data and setting priority access policies; backing up and encrypting high-value, low-stability data and deploying dynamic monitoring policies; allocating low-value, high-stability data to low-cost storage and archiving it regularly; and cleaning up low-value, low-stability data as needed and setting access restriction policies.

[0065] This application constructs a decision matrix by acquiring and standardizing dynamic stability statistics data and value scores of each node from digital reactor simulation data. Then, a decision coordinate system is constructed with set value and stability thresholds as the origin, and the decision matrix is ​​mapped into it. Based on the quadrant in which the data points are located, they are precisely classified into four categories: high value and high stability, high value and low stability, low value and high stability, and low value and low stability. Finally, for each category of data, clear full lifecycle management rules are formulated and applied to achieve refined and differentiated strategy management for different categories of data. This changes the one-size-fits-all management approach from a single dimension, enabling the automatic generation and execution of optimal and differentiated resource allocation and control strategies based on the intrinsic value attributes and external state characteristics of the data. This significantly improves the security and utilization efficiency of high-value core data while effectively reducing the storage and management costs of low-value data, ultimately achieving a leapfrog improvement in the overall efficiency and refined management level of digital reactor simulation data management.

[0066] A second aspect of the present invention provides an intelligent management system for simulation data of a digital reactor, comprising: a data acquisition module, a data construction module, a data analysis module, a simulation iteration module, and an intelligent management module; wherein the data acquisition module is connected to the data construction module, the data construction module is connected to the data analysis module, the data analysis module is connected to the simulation iteration module, and the simulation iteration module is connected to the intelligent management module;

[0067] The data acquisition module acquires digital reactor simulation data, as well as the corresponding physical laws, physical entity knowledge base, and dynamic stability statistics of digital reactor simulation data through data acquisition equipment.

[0068] The data construction module: constructs a physical information loss function based on the physical laws, trains a graph neural network based on the physical information loss function, and performs graph structure modeling on digital reactor simulation data based on the graph neural network to generate a dynamic evolution data lineage map;

[0069] The data analysis module analyzes the physical entity knowledge base, defines a set of key field sources and obtains the corresponding field source intensities; obtains the mixing distance from the set of key field sources to each node in the dynamic evolution data lineage graph; and obtains the value potential of each node based on the field source intensities and the mixing distance.

[0070] The simulation iteration module treats the dynamic evolution data lineage graph as a flow network and the value potential of each node as potential energy; it simulates and iterates the value flow based on an improved random walk algorithm to obtain the value score of each node.

[0071] The intelligent management module analyzes and constructs a decision matrix by combining dynamic stability statistics with the value scores of each node, and then executes full lifecycle management based on the decision matrix to obtain the simulation data management strategy for the digital reactor.

[0072] Compared with the prior art, the beneficial effects of this application are:

[0073] 1. This application generates a dynamic data lineage graph that integrates physical correlation and temporal evolution by constructing a physical information loss function based on physical laws and training a graph neural network. Then, combining this with a physical entity knowledge base, it calculates the value potential of nodes using field source strength and hybrid distance to quantify the influence of physical entities. Next, an improved random walk algorithm is used to simulate value flow in a flow network to obtain dynamic value scores. Finally, dynamic stability statistics are introduced to construct a decision matrix to generate differentiated full lifecycle management strategies. This approach breaks through the limitations of traditional single-dimensional assessment. By integrating multiple technologies such as physical correlation embedding, spatiotemporal dynamic modeling, field source potential quantification, and value flow simulation, it achieves accurate differentiation and differentiated strategy management of high-value core data and low-value auxiliary data, thereby significantly improving the efficiency and accuracy of intelligent management of simulation data.

[0074] 2. This application constructs a physical entity knowledge base by integrating multi-source information such as data reactor data, knowledge graphs, and built-in physical models. From this base, key physical entities are evaluated and defined as field sources, and their importance is assessed as field source strength. Furthermore, to comprehensively measure the association between nodes and field sources, a hybrid distance concept is proposed. This hybrid distance is calculated by weighted fusion of two distances: first, a graph semantic distance measured by the shortest path length based on the dynamic evolution data lineage graph topology, used to capture the logical and evolutionary associations between data; second, a physical spatial distance calculated based on simulation platform coordinate data, used to reflect the real spatial relationships between entities. Finally, the value potential of each node is calculated based on the field source strength and the hybrid distance. By defining key field sources through multi-source knowledge fusion and utilizing the hybrid distance that fuses topological and spatial associations, a rich and multi-dimensional quantitative assessment of the physical connotation of node value potential is achieved. This effectively overcomes the shortcomings of traditional methods that ignore physical entities and spatial characteristics, making the value assessment results more consistent with the physical reality and data association logic of digital reactors.

[0075] 3. This application treats the value potential of each node in the dynamic evolution data lineage graph as potential energy, and drives the direction and probability of value flow based on the potential energy difference between adjacent nodes. The sum of the potential energy differences between the target node and all adjacent nodes is calculated as the total driving force, and the ratio of the potential energy difference of each adjacent node to this total driving force is defined as the improved transfer probability between nodes. This physically simulates the natural diffusion of value from high potential energy areas to low potential energy areas. Furthermore, this improved transfer probability is introduced into a random walk algorithm to iteratively simulate value flow in the flow network. The simulation reaches a stable convergence state by monitoring whether the flow change rate of all nodes is lower than a threshold. Finally, the stable flow data of each node at the convergence point is used as its value score, realizing the dynamic and adaptive transmission and redistribution of node value in the network. This makes the final value score not only reflect the static value potential of the node itself, but also capture its dynamic influence and pivotal role in the overall data flow network, thereby avoiding the one-sidedness of static weighted summation and making the identification of core high-value nodes more accurate. Attached Figure Description

[0076] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0077] Figure 1 This is a flowchart of the intelligent management method for simulation data of digital reactors according to this application;

[0078] Figure 2This is a schematic diagram of the intelligent management system for simulation data of digital reactors according to this application. Detailed Implementation

[0079] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0080] Please see Figure 1 The first aspect of this application provides a method for intelligent management of simulation data for digital reactors, including:

[0081] Acquire digital reactor simulation data and corresponding physical laws; construct a physical information loss function based on the physical laws; train a graph neural network based on the physical information loss function; and perform graph structure modeling on the digital reactor simulation data based on the graph neural network to generate a dynamic evolution data lineage diagram.

[0082] The graph neural network in this embodiment includes an input layer, a core layer, and an output layer; wherein the input layer includes a graph structure and the corresponding features of the graph structure, and the graph structure includes nodes and edges; the core layer consists of graph convolutional layers, used for message passing and neighbor aggregation operations; the output layer includes operations such as node embedding, edge prediction, and graph representation; the training process of the graph neural network model includes forward propagation, loss calculation, backpropagation to update parameters, and iteration to convergence steps;

[0083] The generation process of the dynamic evolution data lineage graph in this embodiment includes: representing digital reactor simulation data as a time series graph; the time series graph includes nodes and edges, nodes include tables and fields, and edges include dependencies and lineages; learning the time series evolution law through a graph neural network model; the training objective is to predict the lineage edges at the next moment; the inference is to predict the lineage edges time by time to generate a lineage graph sequence that evolves over time.

[0084] The system acquires a physical entity knowledge base, analyzes the physical entity knowledge base, defines a set of key field sources and obtains the corresponding field source intensities; obtains the mixing distance of each node from the set of key field sources to the lineage graph of dynamic evolution data; and obtains the value potential of each node based on the field source intensities and mixing distances.

[0085] The dynamic evolution data lineage graph is treated as a flow network, and the value potential of each node is regarded as potential energy. The value flow is simulated and iterated based on an improved random walk algorithm to obtain the value score of each node.

[0086] The dynamic stability statistics of digital reactor simulation data are obtained, and a decision matrix is ​​constructed by combining the value scores of each node. Based on the decision matrix, the simulation data management strategy of digital reactor is obtained by performing full life cycle management.

[0087] This embodiment involves acquiring digital reactor simulation data and corresponding physical laws; constructing a physical information loss function based on the physical laws; training a graph neural network based on the physical information loss function; and performing graph structure modeling on the digital reactor simulation data using the graph neural network to generate a dynamic evolution data lineage graph, including:

[0088] Acquire digital reactor simulation data;

[0089] The physical laws corresponding to digital reactor simulation data are based on the mass conservation equation, energy conservation equation, and momentum conservation equation of digital reactors. It can be understood that by obtaining relevant data of digital reactor simulation through multiple channels, the core physical laws that digital reactors follow during operation can be obtained, such as the mass conservation equation for the mass flow rate of coolant in the reactor core, the energy conservation equation for the balance between the heat release of combustion rods and the heat absorption of coolant, and the momentum conservation equation for the change of momentum of coolant flow in pipes.

[0090] The connection relationships of each node are obtained based on digital reactor simulation data as real labels, and the probability distribution of predicted node connections is output based on graph neural network.

[0091] Based on the true labels and probability distributions, and combined with the cross-entropy loss formula, a data fitting loss term is constructed;

[0092] In this embodiment, the data fitting loss term is calculated by obtaining the connection relationships of each node from digital reactor simulation data as the true labels, predicting the probability distribution of node connections based on the graph neural network output, and combining the cross-entropy loss formula. The data fitting loss term measures the deviation between the model's predicted node connection relationships and the actual situation. It should be noted that this step constructs the data fitting loss term through cross-entropy loss, which improves the sensitivity to differences in probability distribution and effectively amplifies the deviation between the predicted probability and the true label. Especially in the node connection relationships, it can improve the prediction accuracy of discrete relationships such as the existence or non-existence of links. At the same time, the gradient characteristics of cross-entropy loss improve the convergence stability of the graph neural network.

[0093] In this embodiment, three sets of true link relationships between nodes are extracted from the digital reactor simulation data as labels. These include a 1 if there is a link between the core neutron field node and the combustion rod data node; a 1 if there is a link between the main coolant temperature field node and the main coolant pipe node; and a 0 if there is no link between the core neutron field node and the pressure field node. The digital reactor simulation data is input into a graph neural network, which outputs the predicted probability distributions of the three sets of nodes: a probability distribution of 0.92 between the core neutron field node and the combustion rod data node; a probability distribution of 0.88 between the main coolant temperature field node and the main coolant pipe node; and a probability distribution of 0.15 between the core neutron field node and the pressure field node. The deviation is then calculated using the cross-entropy loss formula, resulting in a value of 0.162, which is the data fitting loss term.

[0094] The residuals of the mass conservation equation, energy conservation equation, and momentum conservation equation in the physical laws are calculated respectively, and the physical constraint loss term is constructed through the residuals. The residual is represented as the difference between the theoretical value of the equation and the predicted value of the model output. In this embodiment, the physical constraint loss term is constructed by quantifying the residual by taking the sum of squares or L1 norm.

[0095] In this embodiment, the residual corresponding to the mass conservation equation refers to the error between the rate of change of mass in the numerical solution and the actual mass non-conservation; the residual corresponding to the energy conservation equation refers to the difference between the rate of change of energy in the numerical solution and the thermal diffusion-viscous coupling source phase, reflecting the energy non-conservation error; the residual corresponding to the momentum conservation equation is expressed as the difference between the rate of change of momentum in the numerical solution and the pressure-viscous force source term, reflecting the momentum non-conservation error.

[0096] It should be noted that the mass conservation equation can be constructed by the mass flow rate of coolant flowing into the reactor core being equal to the mass flow rate of coolant flowing out of the reactor core, the energy conservation equation can be obtained based on the theoretical value of the heat released by the combustion rods being equal to the theoretical value of the heat absorbed by the coolant, and the momentum conservation equation can be obtained by the change in momentum being equal to the impulse of the force.

[0097] In this embodiment, for the mass conservation equation, the theoretical value of the coolant flow rate into the reactor core is 2000 kg / s, and the coolant flow rate predicted by the graph neural network is 2015 kg / s, with a residual of 15 kg / s. For the energy conservation equation, the theoretical value of the heat released by the combustion rods is 1500 MW, and the model predicts that the coolant absorbs 1488 MW, with a residual of 12 MW. For the momentum conservation equation, the theoretical value of the momentum change of the coolant in the pipe is 5000 kg·m / s², and the model predicts that it is 5020 kg·m / s², with a residual of 20 kg·m / s². After removing the units from the three residuals, the values ​​are squared and then summed to obtain a physical constraint loss term of 769. In this embodiment, removing the units means dividing each residual by its corresponding characteristic scale to obtain dimensionless data.

[0098] A physical information loss function is constructed based on the data fitting loss term and the physical constraint loss term.

[0099] In this embodiment, the physical information loss function is constructed by first preprocessing the data fitting loss term and the physical constraint loss term, setting the weight parameters of the data fitting loss term and the physical constraint loss term respectively, and then combining the data fitting loss term and the physical constraint loss term. It should be noted that the weight parameters of the data fitting loss term and the physical constraint loss term are set according to the core requirements of digital reactor simulation data management. The core requirements are specifically expressed as improving the accuracy of data association modeling while ensuring strict constraints of physical laws. Therefore, in this embodiment, the weight parameter of the data fitting loss term is set to 0.6, and the weight parameter of the physical constraint loss term is set to 0.4.

[0100] In this embodiment, the physical information loss function satisfies:

[0101] ;in, and These represent the weight parameters for the data fitting loss term and the physical constraint loss term, respectively. and These are represented as the data fitting loss term and the physical constraint loss term, respectively.

[0102] The data fitting loss term satisfies:

[0103] ;in, Let be the spatial and temporal coordinates corresponding to data point i; x represents the spatial coordinates, and t represents the temporal coordinates. Represented as the total number of data points; Indicated as in The predicted value; Indicated as in The true value;

[0104] The physical constraint loss term satisfies:

[0105] ;in, , and These are represented as the residuals of the mass conservation equation, the energy conservation equation, and the momentum conservation equation, respectively.

[0106] The residuals corresponding to the mass conservation equation satisfy:

[0107] ;in, This represents the number of sampling points in the PDE residual; j represents the number corresponding to the sampling point. It represents the rate of generation / consumption of a physical quantity per unit time and per unit volume, and is the external / internal factor driving the change of the physical quantity;

[0108] The residuals corresponding to the energy conservation equation satisfy:

[0109] ; This represents the number of sampling points for the boundary conditions; m represents the number corresponding to the sampling points for the boundary conditions. Represented as the true value of the boundary conditions;

[0110] The residuals corresponding to the momentum conservation equation satisfy:

[0111] ; This represents the number of sampling points for the initial conditions; n represents the number corresponding to the sampling points for the initial conditions. Represented as the actual value corresponding to the initial time;

[0112] In this embodiment, the data fitting loss term is 0.162, and the physical constraint loss term is 769. The data fitting loss term and the physical constraint loss term are preprocessed using a max-min normalization method. During training, the maximum value of the data fitting loss term is 1, and the minimum value is 0; the maximum value of the physical constraint loss term is 1000, and the minimum value is 0. Therefore, after normalization, the data fitting loss term is 0.162, and the physical constraint loss term is 0.769. The weight parameter of the data fitting loss term is set to 0.6, so the weight parameter of the physical constraint loss term is 1 - 0.6 = 0.4. Substituting the data, the physical information loss is calculated to be 0.6 × 0.162 + 0.4 × 0.769 = 0.405.

[0113] A reactor correlation model is obtained by training a graph neural network based on the physical information loss function. Digital reactor simulation data is used as the training set, the physical information loss function is used as the optimization objective, and the graph neural network is trained using the gradient descent algorithm. During training, the network parameters are continuously adjusted to minimize the physical information loss function value, thus obtaining a reactor correlation model that accurately reflects the correlation of digital reactor data. It should be noted that the gradient descent algorithm iteratively updates the network parameters according to the gradient direction of the loss function, thereby obtaining the parameter combination that minimizes the loss function, i.e., the digital reactor data correlation relationship.

[0114] Digital reactor simulation data is input into the reactor association model to obtain a dynamic evolution data lineage diagram. In this embodiment, the latest multi-condition digital reactor simulation data is input into the trained reactor association model. The model analysis yields the association relationships between the core neutron field nodes and the combustion rod data nodes, the main coolant temperature field nodes and the pressure field nodes, etc., and constructs and outputs a dynamic evolution data lineage diagram of a pressurized water reactor.

[0115] This embodiment proposes a modeling method that integrates data-driven and physical mechanism approaches to construct a lineage graph of reactor dynamic evolution data. First, simulation data of a digital reactor is acquired, with the three conservation equations of mass, energy, and momentum serving as core physical constraints. During the training of the graph neural network, on the one hand, the actual connections between nodes and their predicted distributions are utilized to construct data fitting terms through cross-entropy loss, ensuring the accuracy of the topology. On the other hand, the residuals of the three conservation equations on the graph structure are calculated to construct physical constraint loss terms, embedding conservation laws into the model as hard constraints. These two losses together constitute a joint loss function with enhanced physical information, used to train the reactor association model, ultimately generating a lineage graph of dynamic evolution data. By deeply coupling data-driven learning with prior knowledge of physical laws, this method not only ensures high-fidelity reproduction of the data association topology but also strictly satisfies the physical conservation characteristics of the reactor system, significantly improving its physical consistency and interpretability. This lays a solid foundation for subsequent multi-dimensional, high-reliability value assessment.

[0116] In this embodiment, acquiring digital reactor simulation data includes:

[0117] Acquire platform interface data and reactor operation history data; platform interface data refers to the data corresponding to the output interface of the digital reactor simulation platform.

[0118] The time-series data and spatial distribution data of neutron transport and thermal-hydraulic processes are extracted from the platform interface data. In this embodiment, neutron transport is important data that determines the intensity of the reactor chain reaction and the nuclear power distribution, while thermal-hydraulic processes are important processes related to coolant heat transfer efficiency and equipment safety. The two are directly related to core objectives such as reactor safety and power control. Therefore, by acquiring the time-series data and spatial distribution data of the two, a subsequent dynamic evolution data lineage map is constructed, thereby realizing intelligent management of the reactor.

[0119] In this embodiment, the pressurized water reactor digital reactor is managed, and the time-series data for the neutron transport process includes: at t1=0s, the neutron flux at the core center is 1.2×10¹. 5 When n / (cm²·s) and t²=10s, it is 1.22×10¹ 5 n / (cm²·s), …, when t100=990s, is 1.35×10¹ 5 n / (cm²·s), spatial distribution data includes: at t50=490s, the neutron flux at the core radial radius of 0cm is 1.5×10¹ 5 n / (cm²·s), the neutron flux at 5 cm is 1.4 × 10¹ 5 n / (cm²·s), the neutron flux at 10 cm is 1.2 × 10¹ 5n / (cm²·s); The time series data of the thermal-hydraulic process includes: the main coolant outlet temperature is 300℃ at t1=0s, 302℃ at t2=10s, ..., the main coolant outlet temperature is 320℃ at t100=990s. The spatial distribution data includes: at t50=490s, the main coolant outlet temperature is 280℃ at 0m, 350℃ at 1m, and 420℃ at 2m.

[0120] Extract historical time-series and historical spatial data of neutron transport and thermal-hydraulic processes from reactor operation history data;

[0121] In this embodiment, historical time-series data of neutron transport processes over the past 5 years are extracted from the reactor operation history data of the aforementioned reactor, including the core neutron flux of 1.18 × 10¹ at 00:00 on January 1st of last year. 5 The core neutron flux at n / (cm²·s) is 1.19 × 10¹. 5 n / (cm²·s), …, the core neutron flux at 23:00 on January 31 was 1.21 × 10¹ 5 Hourly variation data of n / (cm²·s), historical spatial data including the neutron flux of 1.45 × 10¹ at the core radial radius of 0 cm during maintenance on January 15, 2023. 5 The neutron flux at 5 cm is 1.38 × 10¹ (n / (cm²·s)). 5 The neutron flux at 10 cm is 1.15 × 10¹ (n / (cm²·s)). 5 n / (cm²·s); Historical time-series data of thermal-hydraulic processes, including the main coolant temperature of 298℃ at 0:00 on February 1, 2023, 299℃ at 1:00, ..., and the main coolant temperature of 305℃ at 23:00 on February 28; Historical spatial data includes the temperature of 275℃ at 0m, 345℃ at 1m, and 410℃ at 2m when measured on February 20 last year;

[0122] Multi-condition screening and integration of time-series data, spatially distributed data, historical time-series data, and historical spatial data are used to generate digital reactor simulation data. In this embodiment, multi-condition screening and integration refers to removing redundant and abnormal data to obtain complete, accurate, and representative digital reactor simulation data. It should be noted that if the data obtained directly from the digital reactor simulation platform may be biased, this step verifies the accuracy of the data extracted by the simulation platform by combining the data obtained from the digital reactor simulation platform with historical data, thereby correcting the bias.

[0123] In other embodiments, when performing intelligent management of a pressurized water reactor digital reactor, the extracted time-series data, spatial distribution data, historical time-series data, and historical spatial data are filtered to retain data under steady-state conditions such as rated power operation, or data under transient conditions such as rated power operation, power increase, and load reduction. Abnormal data caused by sensor failures, such as abnormal values ​​of sudden changes in neutron flux, are eliminated. Similar data are then integrated according to time and spatial dimensions to obtain pressurized water reactor digital reactor simulation data.

[0124] In this embodiment, the process of acquiring a physical entity knowledge base involves analyzing the knowledge base, defining a set of key field sources and obtaining the corresponding field source intensities, acquiring the mixing distance from the key field source set to each node in the dynamic evolution data lineage graph, and obtaining the value potential of each node based on the field source intensities and mixing distances.

[0125] The physical entity knowledge base is obtained through data reactor information, reactor knowledge graphs, and the physical models built into the digital reactor simulation platform. In this embodiment, the data reactor information includes core structure design drawings and combustion rod parameter manuals; the reactor knowledge graph includes a graph containing entities such as the core, combustion rods, and coolant pipes, as well as their relationships; the physical models built into the digital reactor simulation platform include neutron transport models and thermal-hydraulic calculation models. By integrating the above data and removing duplicates, the physical entity knowledge base of the pressurized water reactor is constructed.

[0126] By constructing a physical entity knowledge base to organize entity information, a basis is provided for subsequent screening of key field sources; at the same time, the entity attributes and relationships in the physical entity knowledge base provide data support for physical spatial coordinates and graph semantic association analysis in subsequent hybrid distance calculations.

[0127] The importance of entities in the physical entity knowledge base is evaluated to obtain a set of key field sources, and each key field source is assigned an importance value to obtain its corresponding field source strength. In this embodiment, the importance includes security priority, simulation accuracy, and operation and maintenance criticality. Entities with high evaluation scores are defined as the set of key field sources, and each key field source is assigned an importance value according to the evaluation score to obtain the corresponding field source strength.

[0128] In this embodiment, the evaluation score is obtained by evaluating the entity's security priority, simulation accuracy, and operational criticality, etc., obtaining a separate score for each dimension, and setting a weight parameter for each dimension. The weight parameter is set according to the importance of different dimensions. In this embodiment, the weight parameter of security priority is set to 0.5, the weight parameter of simulation accuracy is set to 0.3, and the weight parameter of operational criticality is set to 0.2, thereby obtaining the evaluation score.

[0129] An evaluation score greater than or equal to the evaluation score threshold is considered a high evaluation score. The evaluation score threshold can be set by the lowest impact score of key entities in historical accident cases. In this embodiment, it is set to 8.0. The field source strength is quantified based on the evaluation score. In this embodiment, the evaluation score is directly regarded as the quantified value of the field source strength.

[0130] In this embodiment, entities such as the reactor core, combustion rods, main coolant pipes, neutron field, and temperature field in the physical entity knowledge base of the pressurized water reactor digital reactor are evaluated. The neutron field has a safety priority of 9.5, a simulation accuracy impact of 8.5, and an operation and maintenance criticality of 9.0, resulting in an evaluation score of 9.5 × 0.5 + 8.5 × 0.3 + 9.0 × 0.2 = 9.1. The temperature field has a safety priority of 9.0, a simulation accuracy impact of 8.0, and an operation and maintenance criticality of 8.5, resulting in an evaluation score of 9.0 × 0.5 + 8.0 × 0.2 = 9.1. 0.3 + 8.5 × 0.2 = 8.6; Core safety priority 8.8, simulation accuracy impact 8.2, operation and maintenance criticality 8.6, evaluation score is 8.8 × 0.5 + 8.2 × 0.3 + 8.6 × 0.2 = 8.58; The evaluation scores of the above entities are all greater than or equal to 8.0, so these three entities are defined as the set of critical field sources; and the evaluation scores are used as the quantitative values ​​of the field source strength, that is, the field source strength of the neutron field is 9.1, the field source strength of the temperature field is 8.6, and the field source strength of the core is 8.58;

[0131] Here, an entity is represented as several physical objects related to a digital reactor, including but not limited to core reactor components;

[0132] Obtain the graph semantic distance and physical spatial distance of each node in the dynamic evolution data lineage graph by set up key field sources, and calculate the mixed distance by weighted summation based on the graph semantic distance and physical spatial distance;

[0133] In this embodiment, the graph semantic distance is obtained through the topology of the dynamically evolving data lineage graph. The shortest path from each key source in the key source set to each node is obtained using the topology, and this shortest path is used as the graph semantic distance. In this embodiment, the topology of the dynamically evolving data lineage graph includes node N1 corresponding to the neutron field, node N2 corresponding to the temperature field, node N3 corresponding to the reactor core, and node N4 corresponding to the combustion rod data node within the key source. The paths from N1 to node N4 obtained through the topology of the dynamically evolving data lineage graph are N1 to N3 to N4 and N1 to N2 to N3 to N4. The shortest path is N1 to N3 to N4, with a length of 2. Therefore, the graph semantic distance from N1 to N4 is 2. The shortest path length from other key sources to each node is calculated using the same method and used as the graph semantic distance.

[0134] In this embodiment, the physical spatial distance is obtained by combining the spatial coordinate data of the key field source set with the spatial coordinate data of each node in the dynamic evolution data lineage diagram. The Euclidean distance between the spatial coordinate data of the key field source set and the spatial coordinate data of each node in the dynamic evolution data lineage diagram is used as the physical spatial distance. In this embodiment, the spatial coordinates of the neutron field within the key field source are (10m, 10m, 10m), the spatial coordinates of the temperature field are (12m, 10m, 10m), and the spatial coordinates of the combustion rod data node are (11m, 10m, 10m). The Euclidean distance from the neutron field to the combustion rod data node is calculated to be 1m, and the Euclidean distance from the temperature field to the combustion rod data node is also calculated to be 1m. The Euclidean distance is then used as the physical spatial distance.

[0135] In this embodiment, a hybrid distance is calculated by weighted summation of graph semantic distance and physical spatial distance. Graph semantic distance reflects the tightness of association between data nodes, while physical spatial distance reflects the physical location relationship of entities. Using only one distance may be one-sided. Therefore, by setting weight parameters for each distance separately, the hybrid distance is calculated by weighting the two distances. This hybrid distance reflects both the tightness of association between data nodes and the physical location relationship of entities, providing a basis for value potential calculation. The weight parameters are determined according to the management objectives of the digital reactor. If the emphasis is on the logicality of data association, the weight parameter of graph semantic distance is increased; if the emphasis is on the safety of physical entities, the weight parameter of graph semantic distance is increased. To ensure security, the weighting parameter of physical spatial distance is increased. In this embodiment, it is necessary to balance the logicality of data association and the security of physical entities, so the weighting parameter of both is set to 0.5. In this embodiment, the graph semantic distances of node N4 are 2 from the neutron field to N4, 3 from the temperature field to N4, and 1 from the core to N4; the physical spatial distances are 1m from the neutron field to N4, 1m from the temperature field to N4, and 0.5m from the core to N4; therefore, the mixing distance r from the neutron field to N4 is 0.5×2+0.5×1=1.5, the mixing distance r from the temperature field to N4 is 0.5×3+0.5×1=2, and the mixing distance r from the core to N4 is 0.5×1+0.5×0.5=0.75.

[0136] The value potential of each node is determined based on the source intensity of each key source and the mixing distance of its corresponding node.

[0137] In this embodiment, the value potential formula satisfies:

[0138] Where d represents the node number, Let be the value potential of node d, Q be the source strength of the key field source, and r be the mixing distance of the corresponding node. In this embodiment, the source strength and mixing distance of each key field source corresponding to the combustion rod are as follows: the source strength of the neutron field is 9.5, and the mixing distance of the corresponding node is 1.5; the source strength of the temperature field is 9.0, and the mixing distance of the corresponding node is 2; the source strength of the core is 8.8, and the mixing distance of the corresponding node is 0.75. Substituting into the value potential formula, the value potential of the combustion rod is obtained as 22.27.

[0139] In this embodiment, the key field source set is obtained into the graph semantic distance and physical spatial distance of each node in the dynamic evolution data lineage graph. A weighted summation of the graph semantic distance and physical spatial distance is then performed to calculate the mixed distance, including:

[0140] Obtain the topological structure of the pedigree graph of dynamic evolutionary data;

[0141] Based on the topology, the shortest path from each key source in the key source set to each node is obtained, and the graph semantic distance is determined based on the shortest path.

[0142] Based on the built-in physical model of the digital reactor simulation platform, spatial coordinate data of the key field source set is obtained, and spatial coordinate data of each node in the dynamic evolution data lineage diagram is obtained.

[0143] The physical spatial distance is determined based on the spatial coordinate data of each node in the lineage diagram of the key field source set and the state evolution data.

[0144] The mixed distance is calculated by weighted summation of graph semantic distance and physical spatial distance.

[0145] In this embodiment, the dynamic evolution data lineage graph is treated as a flow network, and the value potential of each node is considered as potential energy. The value flow is simulated and iterated based on an improved random walk algorithm to obtain the value score of each node, including:

[0146] Obtain the potential energy of several nodes;

[0147] Calculate the potential energy difference between two adjacent nodes and determine the improved transition probability based on the potential energy difference;

[0148] An improved random walk algorithm is obtained by introducing improved transition probabilities into the random walk algorithm. It should be noted that the traditional random walk algorithm uses equal transition probabilities, which cannot reflect the value differences between nodes. However, the value of data nodes in a digital reactor is affected by key field sources differently. For example, there is a value difference between core data nodes and ordinary pipeline data nodes. Equal probabilities will cause the value flow simulation to deviate from the actual value distribution. By using the improved transition probabilities of potential energy difference to quantify the value flow dynamics between nodes, and introducing them can replace the traditional equal probabilities to form an improved algorithm.

[0149] The improved random walk algorithm can more accurately reflect the tendency of value flow compared with the traditional algorithm. Specifically, the probability of transfer between nodes with large potential energy differences is higher. At the same time, it improves the differentiation of node value scores, avoids the problem of average value scores in the traditional algorithm, makes the scores of high-value nodes more prominent, and provides a basis for the formulation of subsequent management strategies.

[0150] The improved random walk algorithm is used to simulate and iterate the value flow in the flow network. The flow data of each node is obtained in several iterations, and the flow change rate of each node between two adjacent iterations is calculated. The flow change rate reflects the dynamic trend of value flow.

[0151] In this embodiment, the dynamic evolution data lineage graph is used as a flow network. Initially, each node is assigned the same initial flow, which is set to 10 units in this embodiment. The first iteration is performed using an improved random walk algorithm, resulting in flow rates of N3=12.5, N4=9.8, N5=11.2, and N6=8.5 for each node. After the second iteration, the flow rates are N3=13.1, N4=9.5, N5=10.8, and N6=8.9. The flow rate change rate for each iteration is calculated simultaneously: the change rate for N3 = |13.1-12.5| / 12.5 = 4.8%, and the change rate for N4 = |9.5-9.8| / 9.8 ≈ 3.06%. This method is used to obtain the flow rate change rate for all nodes.

[0152] When the rate of change of traffic at each node is less than the threshold for the rate of change of traffic, the simulation iteration stops, the traffic data of each node is output, and the value score of each node is determined based on the traffic data of each node. Specifically, a threshold for the rate of change of traffic is set, and the rate of change of traffic at each node is compared with the threshold for the rate of change of traffic. When the rate of change of traffic at all nodes is less than the threshold for the rate of change of traffic, it indicates that the value flow has reached a balanced state, and the iteration stops. At this time, the steady-state traffic of each node is the value score of the node.

[0153] In this embodiment, the threshold for the rate of change of flow is set according to the scoring requirements. If high-precision value scoring is required, the threshold can be set to 1%-2% to ensure that the value flow converges fully. If the focus is on calculation efficiency, the threshold can be set to 3%-5% to avoid excessive iteration and resource consumption. In this embodiment, 2% is taken as the threshold for balancing accuracy and efficiency.

[0154] It is understandable that the steady-state flow of a high-value potential node must be higher. Therefore, the steady-state flow can directly quantify the relative value of the node, and thus it can be regarded as the value score of the node.

[0155] Otherwise, continue the simulation iteration;

[0156] In this embodiment, the threshold for the rate of change of traffic is 2%. After multiple iterations, when it reaches the 10th iteration, the rates of change of traffic for each node are: N3=1.5%, N4=1.2%, N5=1.8%, and N6=1.6%, all of which are less than 2%, so the iteration stops. At this time, the steady-state traffic for each node is: N3=14.2, N4=10.1, N5=11.5, and N6=9.3. These steady-state traffic values ​​are used as the value scores for each node.

[0157] In this embodiment, calculating the potential energy difference between two adjacent nodes and determining the improved transition probability based on the potential energy difference includes:

[0158] A target node is defined, and the potential energy difference between all its neighboring nodes is obtained. The potential energy difference between the target node and its neighboring nodes reflects the driving force of value flow between nodes. In this embodiment, node N4 is selected as the target node, where the value potential of node N4 is 22.56, the value potential of its neighboring node N3 is 25.8, the value potential of its neighboring node N5 is 18.2, and the value potential of its neighboring node N6 is 20.1. N6 is a pressure data node. The potential energy difference between node N4 and node N3 is calculated to be 3.24, the potential energy difference between node N4 and node N5 is 4.36, and the potential energy difference between node N4 and node N6 is 2.46.

[0159] The total potential energy difference of the target node is obtained by summing the potential energy differences of all its neighboring nodes.

[0160] The transition probability from the target node to each neighboring node is obtained by dividing the potential energy difference of each neighboring node of the target node by the total potential energy difference of the target node. Specifically, the transition probability from the target node to each neighboring node is obtained by dividing the potential energy difference of each neighboring node of the target node by the total potential energy difference of the target node. The larger the value of the transition probability, the stronger the tendency for value to flow from the target node to that neighboring node.

[0161] In this embodiment, with node N4 as the target node, the total potential difference between its adjacent nodes is 3.24 + 4.36 + 2.46 = 10.06; the improved transfer probability from node N4 to node N3 is 3.24 / 10.06 ≈ 0.32, the improved transfer probability from node N4 to node N5 is 4.36 / 10.06 ≈ 0.43, and the improved transfer probability from node N4 to node N6 is 2.46 / 10.06 ≈ 0.25; therefore, the probability of value flowing from N4 to N3 is 0.32, to N5 is 0.43, and to N6 is 0.25, indicating that node N4 has a stronger tendency to flow to node N5;

[0162] The transition probability is defined as the improved transition probability.

[0163] This embodiment treats the value potential of each node in the dynamic evolution data lineage graph as potential energy in a physical sense, and drives the direction and probability of value flow based on the potential energy difference between adjacent nodes: by calculating the sum of the potential energy differences between the target node and all its adjacent nodes as the total driving force, the ratio of the potential energy difference of each adjacent node to this total driving force is defined as the improved transfer probability, thereby naturally simulating the diffusion process of value from high potential energy regions to low potential energy regions through a physical mechanism; subsequently, this improved transfer probability is embedded into a random walk algorithm to iteratively simulate the dynamic propagation of value in the flow network; by monitoring whether the rate of change of traffic of all nodes is lower than a preset threshold, it is determined whether the system has reached a stable convergence state; finally, the stable traffic of each node at the time of convergence is used as its value score, realizing the adaptive transmission and redistribution of value in the network. This not only comprehensively considers the static value potential of the node itself, but also more effectively captures its dynamic influence and pivotal role in the overall data flow network, overcoming the one-sidedness of the traditional static weighted summation method, and significantly improving the accuracy and physical rationality of high-value core node identification.

[0164] In this embodiment, dynamic stability statistics of digital reactor simulation data are acquired and analyzed in conjunction with the value scores of each node to construct a decision matrix. Based on the decision matrix, full lifecycle management is performed to obtain the digital reactor simulation data management strategy, including:

[0165] This involves acquiring iteration records of digital reactor simulation data from dynamic stability statistics, as well as statistically analyzing the update frequency, version consistency, and historical retention duration of the digital reactor simulation data. It's important to note that the value of digital reactor simulation data depends not only on its inherent importance (node ​​value score) but also on its dynamic characteristics over time. Therefore, joint analysis with dynamic stability statistics is necessary. Update frequency reflects the timeliness requirements of the data, version consistency relates to data reliability, and historical retention duration reflects the data's reusability value. By combining dynamic stability statistics with node value scores, we can avoid biased management strategies and provide a basis for constructing a decision matrix and implementing differentiated management.

[0166] In this embodiment, the version iteration records of the digital reactor simulation data over the past year are obtained. The update frequency of node N3 is twice a month, the version consistency is 98%, and the historical retention period is 5 years; the update frequency of node N4 is three times a month, the version consistency is 95%, and the historical retention period is 4 years. According to the method in this step, the dynamic stability statistics of all nodes can be obtained.

[0167] The dynamic stability statistics and the value scores of each node are standardized; in this embodiment, the standardization process uses the min-max standardization method and the average value operation.

[0168] In this embodiment, the maximum value score is 14.2 and the minimum value is 9.3. Therefore, the standardized value score of node N3 is (14.2-9.3) / (14.2-9.3)=1; the standardized value score of node N4 is (10.1-9.3) / (14.2-9.3)≈0.16. The dynamic stability statistics include three indicators: update frequency, version consistency, and historical retention time. These need to be standardized and their average values ​​taken as the standardized dynamic stability statistics. Taking update frequency as an example, the unit is times / month, with a maximum of 5 times / month and a minimum of 1 time / month. The update frequency of node N3 is 2 times / month. The standardization update frequency for each month is (2-1) / (5-1)=0.25; the maximum version consistency is 99% and the minimum is 90%, the N3 version consistency is 98%, and the standardization version consistency is (98-90) / (99-90)≈0.89; the maximum historical retention period is 6 years and the minimum is 2 years, the N3 retention period is 5 years, and the standardization retention period is (5-2) / (6-2)=0.75; then the dynamic stability statistics of N3 after standardization are (0.25+0.89+0.75) / 3≈0.63, and the dynamic stability statistics of all nodes after standardization are obtained using the same method in this step;

[0169] The decision matrix is ​​determined based on the standardized dynamic stability statistics and the value scores of each node. In this embodiment, four key nodes are selected to construct the decision matrix. The standardized value scores and dynamic stability data of the four key nodes are N3(1,0.63), N4(0.16,0.7), N5(0.45,0.6), and N6(0,0.8), respectively. The decision matrix is ​​[[1,0.63],[0.16,0.7],[0.45,0.6],[0,0.8]].

[0170] A decision coordinate system is constructed by setting a value scoring threshold and a dynamic stability statistics threshold, with the value score as the horizontal axis, the dynamic stability statistics as the vertical axis, and the coordinate combination of the value scoring threshold and the dynamic stability statistics threshold as the origin. In this embodiment, both the value scoring threshold and the dynamic stability statistics threshold are set to 0.5. The standardized data are mapped to the [0,1] interval. The median value of 0.5 can divide all nodes into two levels, high and low. This avoids the problem of too few high-value and high-stability nodes due to excessively high thresholds, and also prevents low-value and low-stability nodes from being mixed into the high-priority quadrant due to excessively low thresholds, thereby ensuring the rationality of the coordinate system division.

[0171] By mapping the decision matrix to the decision coordinate system, formulating full lifecycle management rules, and dividing the decision matrix into quadrants in the decision coordinate system, and combining the full lifecycle management rules for management, a simulation data management strategy for digital reactors is obtained.

[0172] In this embodiment, the decision matrix is ​​mapped to a decision coordinate system, full lifecycle management rules are formulated, and management is carried out based on the quadrant division of the decision matrix in the decision coordinate system and the full lifecycle management rules. This yields a simulation data management strategy for the digital reactor, including:

[0173] When the decision matrix is ​​mapped to the first quadrant of the decision coordinate system, the decision matrix category is set to high-value, high-stability data.

[0174] When the decision matrix is ​​mapped to the second quadrant of the decision coordinate system, the decision matrix category is set to high-value, low-stability data.

[0175] When the decision matrix is ​​mapped to the third quadrant of the decision coordinate system, the decision matrix category is set to low-value, high-stability data.

[0176] When the decision matrix is ​​mapped to the fourth quadrant of the decision coordinate system, the decision matrix category is set to low-value, low-stability data.

[0177] Points on the positive half-axis of the horizontal coordinate, points on the positive half-axis of the vertical coordinate, and the origin of the decision coordinate system are assigned to the first quadrant; points on the negative half-axis of the horizontal coordinate are assigned to the second quadrant; and points on the negative half-axis of the vertical coordinate are assigned to the third quadrant.

[0178] In this embodiment, node N3 (1, 0.9) is mapped to the first quadrant of the decision coordinate system; node N4 (0.16, 0.7) is mapped to the third quadrant of the decision coordinate system; node N5 (0.45, 0.6) is mapped to the third quadrant of the decision coordinate system; and node N6 (0, 0.8) is mapped to the third quadrant of the decision coordinate system.

[0179] The decision matrix categories are used to obtain the simulation data management strategy for digital reactors through full lifecycle management rules;

[0180] The full lifecycle management rules include: configuring high-performance storage for high-value, high-stability data and setting priority access policies; backing up and encrypting high-value, low-stability data and deploying dynamic monitoring policies; allocating low-value, high-stability data to low-cost storage and archiving it regularly; and cleaning up low-value, low-stability data as needed and setting access restriction policies.

[0181] In this embodiment, high-performance storage, such as an SSD storage array, is configured for node N3 in the first quadrant, and priority access is set, specifically, it is given priority access by the core simulation team; low-cost storage, such as tape library storage, is allocated for nodes N4, N5, and N6 in the third quadrant, and a regular archiving plan is formulated, specifically, archiving once per quarter; by integrating the full lifecycle management rules, a simulation data management strategy for a pressurized water reactor digital reactor is obtained.

[0182] In other embodiments, if a second quadrant node exists, it is backed up and encrypted, such as by using the AES-256 encryption algorithm, and dynamic monitoring is deployed, specifically by real-time monitoring for abnormal data updates; if a fourth quadrant node exists, a cleanup rule is set for it, such as cleaning it up if it has not been accessed for more than 6 months, and access permissions are restricted, specifically by allowing only maintenance personnel to view it.

[0183] This embodiment constructs a multi-dimensional decision matrix by collecting and standardizing dynamic stability statistics and value scores of each node in digital reactor simulation data. Based on this, a two-dimensional decision coordinate system is established with preset value and stability thresholds as the origin, mapping the data points in the decision matrix to this space and precisely classifying them into four categories according to their quadrants. For each category, a full lifecycle management strategy matching its characteristics is formulated and implemented, achieving refined and differentiated control of simulation data. This breaks through the traditional single-dimensional, one-size-fits-all data management paradigm, automatically generating and executing optimal resource allocation and control strategies based on the intrinsic value attributes and external state characteristics of the data. While significantly improving the security and utilization efficiency of high-value core data, it effectively reduces the storage and maintenance costs of low-value data, thereby driving a leap forward in the overall efficiency and refined governance capabilities of digital reactor simulation data management.

[0184] Please see Figure 2 A second aspect of this application provides an intelligent management system for simulation data of a digital reactor, comprising: a data acquisition module, a data construction module, a data analysis module, a simulation iteration module, and an intelligent management module; the data acquisition module is connected to the data construction module, the data construction module is connected to the data analysis module, the data analysis module is connected to the simulation iteration module, and the simulation iteration module is connected to the intelligent management module.

[0185] Data acquisition module: Acquires digital reactor simulation data, as well as corresponding physical laws, physical entity knowledge base, and dynamic stability statistics of digital reactor simulation data through data acquisition equipment; the data acquisition equipment includes several sensors, etc.

[0186] Data construction module: Constructs a physical information loss function based on physical laws, trains a graph neural network based on the physical information loss function, and performs graph structure modeling on digital reactor simulation data based on the graph neural network to generate a dynamic evolution data lineage map;

[0187] Data Analysis Module: Analyzes the physical entity knowledge base, defines the key field source set and obtains the corresponding field source intensity; obtains the mixing distance of the key field source set to each node in the dynamic evolution data lineage graph; obtains the value potential of each node based on the field source intensity and mixing distance;

[0188] Simulation Iteration Module: Treats the dynamic evolution data lineage graph as a flow network and the value potential of each node as potential energy; simulates and iterates the value flow based on an improved random walk algorithm to obtain the value score of each node;

[0189] Intelligent Management Module: By analyzing dynamic stability statistics and the value scores of each node, a decision matrix is ​​constructed. Based on the decision matrix, full lifecycle management is performed to obtain the simulation data management strategy for the digital reactor.

[0190] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0191] The working principle of this application is as follows: By constructing a physical information loss function based on physical laws and training a graph neural network, a dynamic data lineage graph integrating physical correlation and temporal evolution is generated. Then, combined with a physical entity knowledge base, the value potential of nodes is calculated through field source strength and mixing distance to quantify the influence of physical entities. Next, an improved random walk algorithm is used to simulate value flow in a flow network to obtain dynamic value scores. Finally, dynamic stability statistics are introduced to construct a decision matrix to generate differentiated full life cycle management strategies. This approach breaks through the limitations of traditional single-dimensional assessment. By integrating multiple dimensions such as physical correlation embedding, spatiotemporal dynamic modeling, field source potential quantification, and value flow simulation, it achieves accurate differentiation and differentiated strategy management of high-value core data and low-value auxiliary data. This significantly improves the efficiency and accuracy of intelligent simulation data management, avoiding the problem of low efficiency in intelligent simulation data management methods caused by the single-dimensional node value assessment in existing technologies, which ignores the differences in data in dimensions such as physical entity correlation and temporal dynamic characteristics, resulting in the same management strategies for high-value core data and low-value auxiliary data.

[0192] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. A method for intelligent management of simulation data for digital reactors, characterized in that, include: To acquire digital reactor simulation data and the corresponding physical laws; Based on the aforementioned physical laws, a physical information loss function is constructed. Based on the physical information loss function, a graph neural network is trained. Based on the graph neural network, a graph structure model is performed on the digital reactor simulation data to generate a dynamic evolution data lineage diagram. Obtain a physical entity knowledge base, analyze the physical entity knowledge base, define a set of key field sources and obtain the corresponding field source intensities; obtain the mixing distance of the key field source set to each node in the dynamic evolution data lineage graph; The value potential of each node is obtained based on the source strength and the mixing distance; The dynamic evolutionary data lineage graph is viewed as a flow network, and the value potential of each node is viewed as potential energy. The value flow is simulated and iterated based on an improved random walk algorithm to obtain the value score of each node. The dynamic stability statistics of digital reactor simulation data are obtained, and a decision matrix is ​​constructed by combining the value scores of each node. Based on the decision matrix, the simulation data management strategy of digital reactor is obtained by performing full life cycle management.

2. The intelligent management method for simulation data of digital reactors according to claim 1, characterized in that, The process includes acquiring digital reactor simulation data and corresponding physical laws; constructing a physical information loss function based on the physical laws; training a graph neural network based on the physical information loss function; and performing graph structure modeling on the digital reactor simulation data based on the graph neural network to generate a dynamic evolution data lineage graph, including: Acquire digital reactor simulation data; The physical laws corresponding to digital reactor simulation data are based on the mass conservation equation, energy conservation equation, and momentum conservation equation of digital reactors. The connection relationships of each node are obtained based on digital reactor simulation data as real labels, and the probability distribution of predicted node connections is output based on graph neural network. Based on the true labels and the probability distribution, and combined with the cross-entropy loss formula, a data fitting loss term is constructed; Calculate the residuals of the mass conservation equation, energy conservation equation, and momentum conservation equation in the physical laws respectively, and construct the physical constraint loss term through the residuals; A physical information loss function is constructed based on the data fitting loss term and the physical constraint loss term. A reactor correlation model is obtained by training a graph neural network based on a physical information loss function; Digital reactor simulation data is input into the reactor correlation model to obtain a dynamic evolution data lineage diagram.

3. The intelligent management method for simulation data of digital reactors according to claim 1, characterized in that, The process involves acquiring a physical entity knowledge base, analyzing the physical entity knowledge base, defining a set of key field sources and acquiring the corresponding field source intensities, and acquiring the mixing distance from the set of key field sources to each node in the dynamic evolution data lineage graph. The value potential of each node is obtained based on the source strength and the mixing distance, including: The physical entity knowledge base is obtained through data reactor information, reactor knowledge graph, and the physical model built into the digital reactor simulation platform. The importance of entities in the physical entity knowledge base is evaluated to obtain a set of key field sources, and the importance of each key field source is assigned to obtain its corresponding field source intensity. The entity refers to several physical objects related to a digital reactor; Obtain the graph semantic distance and physical spatial distance of each node in the dynamic evolution data lineage graph of the key field source set, and calculate the mixed distance by weighted summation based on the graph semantic distance and physical spatial distance; The value potential of each node is determined based on the source strength of each key source and the mixing distance of its corresponding node.

4. The intelligent management method for simulation data of digital reactors according to claim 3, characterized in that, The process of acquiring the key field source set into the dynamic evolution data lineage graph, calculating the graph semantic distance and physical spatial distance of each node, and then performing a weighted summation of the graph semantic distance and physical spatial distance to obtain the mixed distance includes: Obtain the topological structure of the pedigree graph of dynamic evolutionary data; Based on the aforementioned topology, the shortest path from each key source in the key source set to each node is obtained, and the graph semantic distance is determined based on the shortest path. Based on the built-in physical model of the digital reactor simulation platform, spatial coordinate data of the key field source set is obtained, and spatial coordinate data of each node in the dynamic evolution data lineage diagram is obtained. The physical spatial distance is determined based on the spatial coordinate data of each node in the lineage diagram of the key field source set and the state evolution data. The mixed distance is calculated by weighted summation of the semantic distance and physical distance.

5. The intelligent management method for simulation data of digital reactors according to claim 1, characterized in that, The dynamic evolution data lineage graph is viewed as a flow network, and the value potential of each node is viewed as potential energy. The value flow is simulated iteratively based on an improved random walk algorithm to obtain the value score of each node, including: Obtain the potential energy of several nodes; Calculate the potential energy difference between two adjacent nodes, and determine the improved transition probability based on the potential energy difference; An improved random walk algorithm is obtained by introducing improved transition probabilities into the random walk algorithm; The improved random walk algorithm is used to simulate and iterate the value flow in the flow network, obtain the flow data of each node in several iterations, and calculate the flow change rate of each node between two adjacent iterations. When the rate of change of traffic at each node is less than the threshold for the rate of change of traffic, the simulation iteration stops, the traffic data of each node is output, and the value score of each node is determined based on the traffic data of each node. Otherwise, continue with the simulation iteration.

6. The intelligent management method for simulation data of a digital reactor according to claim 5, characterized in that, The calculation of the potential energy difference between two adjacent nodes, and the determination of the improved transition probability based on the potential energy difference, includes: Set a target node and obtain the potential energy difference between all adjacent nodes of the target node; The total potential energy difference of the target node is obtained by summing the potential energy differences of all its adjacent nodes. The transition probability from the target node to each of its neighboring nodes is obtained by dividing the potential energy difference of each neighboring node by the total potential energy difference. The transition probability is defined as the improved transition probability.

7. The intelligent management method for simulation data of a digital reactor according to claim 1, characterized in that, The process involves acquiring dynamic stability statistics of digital reactor simulation data, analyzing them in conjunction with the value scores of each node to construct a decision matrix, and then implementing full lifecycle management based on the decision matrix to obtain a digital reactor simulation data management strategy, including: Obtain the iteration records of digital reactor simulation data from dynamic stability statistics, and statistically analyze the update frequency, version consistency, and historical retention duration of digital reactor simulation data; The dynamic stability statistics and the value scores of each node are standardized. The decision matrix is ​​determined based on standardized dynamic stability statistics and the value scores of each node; Set value scoring thresholds and dynamic stability statistics thresholds, construct a decision coordinate system with value scoring as the horizontal axis, dynamic stability statistics as the vertical axis, and the coordinate combination of value scoring thresholds and dynamic stability statistics thresholds as the origin; By mapping the decision matrix to the decision coordinate system, formulating full lifecycle management rules, and dividing the decision matrix into quadrants in the decision coordinate system, and combining the full lifecycle management rules for management, a simulation data management strategy for digital reactors is obtained.

8. The intelligent management method for simulation data of a digital reactor according to claim 7, characterized in that, The process of mapping the decision matrix to a decision coordinate system, formulating full lifecycle management rules, and managing the simulation data of a digital reactor based on the quadrant division of the decision matrix in the decision coordinate system and the full lifecycle management rules, includes: When the decision matrix is ​​mapped to the first quadrant of the decision coordinate system, the decision matrix category is set to high-value, high-stability data. When the decision matrix is ​​mapped to the second quadrant of the decision coordinate system, the decision matrix category is set to high-value, low-stability data. When the decision matrix is ​​mapped to the third quadrant of the decision coordinate system, the decision matrix category is set to low-value, high-stability data. When the decision matrix is ​​mapped to the fourth quadrant of the decision coordinate system, the decision matrix category is set to low-value, low-stability data. Points on the positive half-axis of the horizontal coordinate, points on the positive half-axis of the vertical coordinate, and the origin of the decision coordinate system are assigned to the first quadrant; points on the negative half-axis of the horizontal coordinate are assigned to the second quadrant; and points on the negative half-axis of the vertical coordinate are assigned to the third quadrant. The decision matrix categories are used to obtain the simulation data management strategy for digital reactors through full lifecycle management rules; The full lifecycle management rules include: configuring high-performance storage for high-value, high-stability data and setting priority access policies; backing up and encrypting high-value, low-stability data and deploying dynamic monitoring policies; allocating low-value, high-stability data to low-cost storage and archiving it regularly; and cleaning up low-value, low-stability data as needed and setting access restriction policies.

9. The intelligent management method for simulation data of a digital reactor according to claim 1, characterized in that, The acquisition of digital reactor simulation data includes: Acquire platform interface data and reactor operation history data; the platform interface data refers to the data corresponding to the output interface of the digital reactor simulation platform. Extract time-series and spatial distribution data of neutron transport and thermal-hydraulic processes from platform interface data; Extract historical time-series data and historical spatial data of neutron transport and thermal-hydraulic processes from reactor operation history data; The time-series data, the spatial distribution data, the historical time-series data, and the historical spatial data are filtered and integrated under multiple operating conditions to generate digital reactor simulation data.

10. A simulation data intelligent management system for digital reactors, characterized in that, include: The system comprises a data acquisition module, a data construction module, a data analysis module, a simulation iteration module, and an intelligent management module; the data acquisition module is connected to the data construction module, the data construction module is connected to the data analysis module, the data analysis module is connected to the simulation iteration module, and the simulation iteration module is connected to the intelligent management module. The data acquisition module acquires digital reactor simulation data, as well as the corresponding physical laws, physical entity knowledge base, and dynamic stability statistics of digital reactor simulation data through data acquisition equipment. The data construction module: constructs a physical information loss function based on the physical laws, trains a graph neural network based on the physical information loss function, and performs graph structure modeling on digital reactor simulation data based on the graph neural network to generate a dynamic evolution data lineage map; The data analysis module analyzes the physical entity knowledge base, defines a set of key field sources and obtains the corresponding field source intensities; and obtains the mixing distance from the set of key field sources to each node in the dynamic evolution data lineage graph. The value potential of each node is obtained based on the source strength and the mixing distance; The simulation iteration module treats the dynamic evolution data lineage graph as a flow network and the value potential of each node as potential energy; it simulates and iterates the value flow based on an improved random walk algorithm to obtain the value score of each node. The intelligent management module analyzes and constructs a decision matrix by combining dynamic stability statistics with the value scores of each node, and then executes full lifecycle management based on the decision matrix to obtain the simulation data management strategy for the digital reactor.

Images