Service characteristic dynamic perception intelligent agent driven nuclear power equipment thermal stratification analysis method

Abstract

The application discloses a kind of nuclear power equipment thermal stratification analysis methods of service characteristic dynamic perception intelligent agent driving, it is related to nuclear power equipment thermal state analysis technical field.The application is analyzed current operating condition by service characteristic dynamic perception intelligent agent, and is combined with historical similar working condition autonomous optimization monitoring feature and configuration state deduction task, avoid the problem that fixed monitoring feature is difficult to adapt to the change of working condition, improve the effective utilization degree of multi-source service data, by physical guidance dynamic bayesian network fusion thermal stratification formation mechanism, historical working condition knowledge and real-time monitoring data, realize the probability estimation and evolution deduction of thermal stratification state, and the uncertainty of characterization deduction result is reduced, the limitation of simply relying on data model or offline simulation is reduced, the application realizes accurate perception thermal stratification state, predicts its evolution trend and quantifies corresponding risk.

Description

Technical Field

[0001] This application relates to the field of thermal status analysis technology for nuclear power equipment, and in particular to a method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing agent for service characteristics. Background Technology

[0002] During transient operations such as start-up, shutdown, power increases / decreases, and flow rate adjustments, fluids at different temperatures in pipelines and connection areas of nuclear power equipment are prone to forming stable or alternating thermal stratification. This results in uneven local temperature fields, leading to additional thermal stress, structural deformation, and fatigue damage. Thermal stratification is characterized by strong operational condition dependence, complex evolution processes, and the difficulty in directly measuring some key features. Its continued development directly threatens the structural integrity and service reliability of the equipment. Therefore, accurately perceiving thermal stratification, predicting its evolution trend, and quantifying the corresponding risks are crucial foundations for conducting structural safety assessments and ensuring the safe operation of nuclear power equipment. Summary of the Invention

[0003] The purpose of this application is to provide a method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing agent of service characteristics, so as to accurately sense the thermal stratification state, predict its evolution trend and quantify the corresponding risks.

[0004] To achieve the above objectives, this application provides the following solution.

[0005] This application provides a method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing agent based on service characteristics, including: Construct a thermally hierarchical operating condition knowledge base for nuclear power equipment, a physically guided dynamic Bayesian network, and a service characteristic dynamic perception agent based on a large language model; Obtain the current operating condition parameters of the nuclear power equipment; the current operating condition parameters include: operation log, control commands, equipment action records and current service monitoring data; Based on the current operating parameters, the service characteristics dynamic sensing agent is used to retrieve the thermal stratification operating condition knowledge base to form a target monitoring feature combination, thermal stratification state inference task, and initial state prior distribution. Based on the target monitoring feature combination, the thermal stratification state inference task, and the initial state prior distribution, a physical-guided dynamic Bayesian network is used for analysis to obtain the thermal stratification state prediction distribution, state evolution trend, and uncertainty within the preset inference time domain.

[0006] Optionally, based on the target monitoring feature combination, the thermal stratification state inference task, and the initial state prior distribution, a physical-guided dynamic Bayesian network is used for analysis to obtain the predicted distribution of thermal stratification states, state evolution trends, and uncertainties within the preset inference time domain. This further includes: The physical-guided dynamic Bayesian network is corrected based on the service monitoring data obtained in the preset time domain, and the thermal stratification knowledge base and the service characteristic dynamic sensing agent are updated.

[0007] Optionally, based on the current operating condition parameters, the service characteristic dynamic sensing agent is used to retrieve the thermal stratification operating condition knowledge base to form a target monitoring feature combination, a thermal stratification state inference task, and an initial state prior distribution, specifically including: Transform operation logs and control commands into semantic descriptions that represent operation phases, changes in operating conditions, and adjustment events; Extract statistical features, trends, and time-series fluctuation characteristics from current service monitoring data; Based on the semantic description, the statistical features, the changing trends, and the temporal fluctuation features, a joint representation of the current operating condition is formed. Retrieve historical operating conditions that match the joint representation of the current operating condition from the thermal stratification operating condition knowledge base, and use them as the first historical matching operating condition; Based on the effective monitoring features corresponding to each first historical matching working condition and the available sensor data under the current working condition, candidate features are selected from the feature operator library. Based on the correlation between each candidate feature and the hot stratification state, the redundancy between candidate features, and the credibility of candidate features, one or more target monitoring features are selected from each candidate feature to form a target monitoring feature combination; A structured thermal stratification state deduction task is formed based on the rate of change of the current working condition, the evolution duration of each first historical matching working condition, and the data status of each target monitoring feature in the target monitoring feature combination. Based on the current joint characterization of operating conditions and the thermal stratification state inference task, retrieve historical operating conditions that match both the current joint characterization of operating conditions and the thermal stratification state inference task from the thermal stratification operating condition knowledge base, and use them as the second historical matching operating conditions. The posterior distribution of the hot-layered state of each second historical matching condition is weighted and fused according to the degree of matching between each second historical matching condition and the current condition to form the prior distribution of the initial state.

[0008] Optionally, based on the target monitoring feature combination, the thermal stratification state inference task, and the initial state prior distribution, a physical-guided dynamic Bayesian network is used for analysis to obtain the predicted distribution of thermal stratification states, state evolution trends, and uncertainties within the preset inference time domain, specifically including: The task parameters of the physical-guided dynamic Bayesian network are set based on the thermal stratification state inference task, and the composition of the thermal stratification hidden state variables of the physical-guided dynamic Bayesian network is determined based on the combination of target monitoring features. Starting from the initial state prior distribution, the thermal stratification state prediction distribution within the preset extrapolation time domain is obtained by recursively extrapolating along the time slices of the physically guided dynamic Bayesian network. The expected prediction is calculated based on the predicted distribution of thermal stratification state within the preset simulation time domain. The state evolution trend is determined based on the predicted expectation; Based on the degree of dispersion and confidence interval of the predicted distribution of thermal stratification state within the preset extrapolation time domain, uncertainty is quantified using the probability of exceeding the risk threshold.

[0009] Optionally, the probability relationship of a single time slice in the physically guided dynamic Bayesian network is expressed as follows: ; in, For the thermal stratification hidden state variables of time slice t, Let be the observed variable for time slice t. For the execution boundary variables of time slice t, The posterior distribution of the state after fusing the observed variables at time slice t, Let represent the observation probability relationship between the observed variables and the latent state variables of the thermal stratification at time slice t. Let be the prior distribution of the state in time slice t. The marginal probabilities of observed variables are determined under the constraints of the boundary variables at time slice t, and are used to normalize the posterior distribution of the state.

[0010] Optionally, the joint probability distribution of time-slice recursion in the physically guided dynamic Bayesian network is represented as: ; in, For all observed variables up to time slice t, Let these be the hidden state variables of the hot stratification at future time t+k. Let these be the hidden state variables of the hot stratification at future time t+k-1. Let be the boundary variables for operation at a future time t+k. Predict the distribution of thermal stratification states for future time t+k. Hidden state variables for thermal stratification are determined by Transfer to The probability, Predict the distribution of thermal stratification states for future time t+k-1.

[0011] Optionally, the state transition relationship between adjacent time slices in a physically guided dynamic Bayesian network is as follows: ; ; in, , These are the thermally layered hidden state variables for time slices t and t-1, respectively. Let be the observed variable for time slice t. For the execution boundary variables of time slice t, , These represent state evolution error and observation noise, respectively. The state transition function is established based on the causal dependency between the thermal stratification state and the operating boundary, and its parameters are... Calibration was performed using a combination of multiphysics simulation samples and historical service data. The parameter is an observation mapping function established based on the observation relationship between thermal stratification stealth state and service monitoring information. It is calibrated based on historical service monitoring data.

[0012] Optionally, the thermal stratification operating condition knowledge base of the nuclear power equipment includes: Thermal stratification knowledge ternary set under different historical operating conditions; Each thermal stratification condition knowledge triplet corresponds to an attribute feature; the thermal stratification condition knowledge triplet is associated with and stored with its corresponding attribute feature, which includes historical monitoring sequences, state estimation results, and evolutionary features; Each thermal stratified operating condition knowledge triple corresponds to a semantic index for operating condition description and a numerical index for operating parameters.

[0013] Optionally, the service characteristic dynamic perception intelligent agent includes a working condition semantic parsing module, a knowledge retrieval module, a candidate feature generation module, a numerical feature evaluation module, and a deduction task configuration module.

[0014] Optionally, the physically guided dynamic Bayesian network is corrected based on the service monitoring data obtained subsequently within the preset time domain, and the thermally stratified operating condition knowledge base and the service characteristic dynamic sensing agent are updated, specifically including: In subsequent monitoring, service monitoring data within the preset simulation time domain will be acquired; Thermal stratification state estimation is performed on the service monitoring data obtained within the preset simulation time domain to obtain the estimated distribution of thermal stratification state within the preset simulation time domain. Compare the estimated distribution of thermal stratification states with the predicted distribution of thermal stratification states in the preset extrapolation time domain, and calculate the state prediction deviation; When the absolute value of the state prediction deviation exceeds a preset threshold, the state transition parameters and observation parameters of the physically guided dynamic Bayesian network are corrected. The state transition parameters are the parameters of the thermally hierarchical state transition function, and the observation parameters are the parameters in the observation mapping function. The system associates the current operating conditions, target monitoring characteristics, thermal stratification state estimation results, evolution prediction results, and comparative verification results of the current analysis cycle, and updates them to the thermal stratification operating condition knowledge base and adjusts the optimal configuration of the operating condition retrieval and monitoring characteristics of the dynamic sensing agent for service characteristics.

[0015] According to the specific embodiments provided in this application, this application has the following technical effects.

[0016] This application provides a method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing agent for service characteristics. This application analyzes the current operating conditions through a dynamic sensing agent for service characteristics, and autonomously selects monitoring features and configuration status inference tasks by combining historical similar operating conditions. This avoids the problem that fixed monitoring features are difficult to adapt to changes in operating conditions, improves the effective utilization of multi-source service data, and achieves probability estimation and evolution inference of thermal stratification state by fusing thermal stratification formation mechanism, historical operating condition knowledge and real-time monitoring data through a physically guided dynamic Bayesian network. It also characterizes the uncertainty of the inference results and reduces the limitations of simply relying on data models or offline simulation. This application achieves accurate perception of thermal stratification state, prediction of its evolution trend and quantification of corresponding risks. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments 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.

[0018] Figure 1 This is a flowchart illustrating a method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing agent for service characteristics, provided as an embodiment of this application.

[0019] Figure 2 This is a schematic diagram illustrating the specific implementation process of a nuclear power equipment thermal stratification analysis method driven by a dynamic sensing intelligent agent for service characteristics, provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of this application.

[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Most methods for analyzing thermal stratification in nuclear power equipment rely on fixed measurement point layouts and preset temperature threshold criteria for identification, followed by offline numerical simulation for condition assessment. These methods separate monitoring configuration from condition projection, making it difficult to adjust the monitoring focus autonomously based on real-time operating condition changes, and also hindering the effective reuse of analytical experience accumulated from similar historical operating conditions. More importantly, they generally lack quantitative characterization of uncertainties caused by measurement noise, operating condition deviations, and model errors, making it difficult to accurately estimate the unmeasurable thermal stratification state and its evolutionary risks under complex operating conditions. Furthermore, they struggle to continuously revise the condition projection model, operating condition knowledge, and monitoring feature configuration using subsequent monitoring results, resulting in insufficient adaptability and analytical reliability under long-term operating conditions.

[0023] In one exemplary embodiment, a method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing agent for service characteristics is provided, such as... Figure 1 As shown, it includes the following steps 101-104.

[0024] Step 101: Construct a thermally hierarchical operating condition knowledge base, a physically guided dynamic Bayesian network, and a service characteristic dynamic perception agent based on a large language model for nuclear power equipment. Step 102: Obtain the current operating parameters of the nuclear power equipment; the current operating parameters include: operation log, control commands, equipment action records, and current service monitoring data; Step 103: Based on the current operating condition parameters, use the service characteristic dynamic sensing agent to search the thermal stratification operating condition knowledge base to form a target monitoring feature combination, thermal stratification state inference task, and initial state prior distribution. Step 104: Based on the target monitoring feature combination, the thermal stratification state inference task, and the initial state prior distribution, a physical-guided dynamic Bayesian network is used for analysis to obtain the thermal stratification state prediction distribution, state evolution trend, and uncertainty within the preset inference time domain.

[0025] Implementing steps 101-104 above enables accurate perception of thermal stratification, prediction of its evolution trend, and quantification of corresponding risks.

[0026] In another exemplary embodiment, step 104 is followed by step 105.

[0027] Step 105: Correct the physically guided dynamic Bayesian network based on the subsequently obtained service monitoring data within the preset time domain, and update the thermally stratified operating condition knowledge base and the service characteristic dynamic sensing agent. At this time, as... Figure 2 As shown, the specific implementation process of the above method can be referred to as steps S1-S5 below.

[0028] S1: Obtain multi-source service information and thermal stratification analysis knowledge of nuclear power equipment, establish the correlation between operating conditions, monitoring characteristics and thermal stratification status, and construct a knowledge base of thermal stratification conditions for nuclear power equipment; S2: Based on the formation and evolution mechanism of thermal stratification in nuclear power equipment, a physical-guided dynamic Bayesian network is constructed to realize thermal stratification state estimation, evolution prediction and uncertainty quantification. S3: Construct a dynamic perception agent for service characteristics based on a large language model, analyze the current operating conditions and retrieve the hot stratification operating condition knowledge base in each analysis cycle, autonomously select monitoring features and configure hot stratification state inference tasks. S4: Extract historical similar operating conditions from the thermal stratification operating condition knowledge base to construct the prior distribution of the current operating condition, and use the physical-guided dynamic Bayesian network to fuse the current service monitoring data to estimate the thermal stratification state and predict its evolution and uncertainty. S5: Output the current thermal stratification state, the evolution trend within the preset time domain, the uncertainty and risk level, and correct the physical-guided dynamic Bayesian network based on subsequent service monitoring data, update the thermal stratification operating condition knowledge base and the monitoring feature configuration of the agent, for state inference in the next analysis cycle.

[0029] In another exemplary embodiment, during the construction of the aforementioned thermal stratification knowledge base, the following steps are taken: First, the structural design data, operating procedures, historical monitoring data, thermal stratification simulation analysis results, and historical analysis records of the surge tube are acquired. The historical monitoring data includes wall temperature at different locations on the surge tube, system pressure, pressurizer pressure and water level, reactor power, main pump operating status, pressurizer spray valve opening and spray flow rate, etc. Based on unit start-up and shutdown, power changes, and equipment adjustment events, the continuous monitoring data is segmented into operating conditions, and data from different sources is time-aligned, anomaly corrected, and missing data filled in. Combining the upper and lower wall temperature differences, axial temperature gradients, and thermal stress analysis results, the thermal stratification state and evolution stage of each operating condition segment are labeled.

[0030] Based on the processed multi-source information, the correlations between operating conditions and monitoring features, between monitoring features and thermal stratification states, and between thermal stratification states and evolution results are extracted and represented as thermal stratification operating condition knowledge triples. The corresponding historical monitoring sequences, state estimation results, and evolution features are stored as attribute associations. This knowledge is organized according to the structural location of the waveguide and the unit's operating stage, establishing a semantic index for operating condition description and a numerical index for operating parameters to construct a thermal stratification operating condition knowledge base. This knowledge base can retrieve historically similar operating conditions based on the current operating condition representation, returning effective monitoring features, thermal stratification state information, and historical analysis experience, providing support for subsequent monitoring feature optimization and state prior construction.

[0031] In another exemplary embodiment, the above-described process for constructing the thermal stratification knowledge base is as follows: S11: Obtain design and operation data, historical monitoring data, thermal stratification simulation analysis results and analysis cases of nuclear power equipment, and perform operating condition segmentation, time sequence alignment and status labeling on the obtained information; S12: Extract the correlation between operating conditions, monitoring characteristics, state representation and state evolution to form a thermal stratification operating condition knowledge triplet; S13: The knowledge triples are associated and stored according to the nuclear power equipment location and operation stage, and a joint index of semantic information and numerical attributes is established to construct a hot-layered operating condition knowledge base for the dynamic sensing agent of service characteristics to retrieve and call.

[0032] In another exemplary embodiment, the construction process of the physical-guided dynamic Bayesian network described above is as follows: S21: Using the operating boundary of nuclear power equipment as the conditional variable, service monitoring information as the observation variable, and thermal stratification state as the latent variable, the causal dependencies between the variables are determined based on fluid transport, buoyancy stratification and wall heat transfer mechanisms, and a single-time-slice Bayesian network is constructed.

[0033] For example, consider a waveguide.

[0034] Using the operating boundary of the waveguide as the conditional variable, service monitoring information such as wall temperature, pressure, flow rate, and equipment status as the observed variables, and the temperature difference between the upper and lower fluids, the location of the stratification interface, the axial temperature gradient, and the thermal stress level as the latent state variables of thermal stratification, time slices are used. The operational boundary variables, observation variables, and thermal stratification hidden state variables are denoted as follows: , and ,in: In the formula, The temperature difference between the upper and lower fluids inside the pipe. For the location of the layered interface, For the axial temperature gradient, The thermal stress level is considered. Based on the mechanisms of fluid transport, density difference buoyancy, hot-cold mixing, and heat transfer through the pipe wall, the causal dependencies between variables are determined, and a single-time-slice Bayesian network is constructed. Its joint probability decomposition relationship is expressed as: in, Let be the state prior distribution for time slice t, and represent the probability distribution of thermal stratification states under the influence of the running boundary. Let represent the observation probability relationship between the observed variables and the thermal stratification latent state variables at time slice t, and let represent the observation probability relationship between the thermal stratification latent state and service monitoring information. Let be the joint probability of time slice t.

[0035] S22: Establish the state transition relationship between the hidden states of adjacent time slices, and use the spatiotemporal continuity of the temperature field, the consistency of evolution, and the laws of thermal stress as physical constraints, embedding the state transition relationship and conditional probability distribution to form a physical-guided dynamic Bayesian network.

[0036] For example, the above network is expanded along the time axis to establish the hidden state transition relationships between adjacent time slices, forming a dynamic Bayesian network. The hot hierarchical state transition model and the observation model are expressed as follows: In the formula, , These are the thermally layered hidden state variables for time slices t and t-1, respectively. Let be the observed variable for time slice t. For the execution boundary variables of time slice t, , These represent state evolution error and observation noise, respectively. The state transition function is established based on the causal dependency between the thermal stratification state and the operating boundary. This is an observation mapping function established based on the observation relationship between thermal stratification stealth state and service monitoring information. and These are the state transition parameters and the observation parameters, respectively.

[0037] By embedding physical constraints such as the spatiotemporal continuity of the temperature field, energy transfer laws, and thermal stress response into the state transition relationship, a feasible domain of thermally stratified states can be constructed. in, To satisfy the feasible domain of thermal stratification state that meets the constraints of temperature field continuity, energy transfer law, and thermal stress response, Let be the i-th physical constraint function, used to characterize the physical constraint relationship that the hidden states of thermal stratification and the operational boundary variables should satisfy. Here, i is the number of the physical constraint function, and n is the total number of physical constraints set. For state transition results exceeding the feasible region, their conditional probability is set to zero, and the remaining state probabilities are renormalized. This limits the evolution range of temperature difference, temperature gradient, and stratification interface, avoiding purely data-driven deductions that do not conform to the thermal stratification evolution mechanism.

[0038] S23: The observation parameters and state transition parameters of the network are calibrated using historical service data, and the initial probability relationships under different operating boundaries are calibrated using multiphysics simulation samples to obtain the jointly calibrated physical-guided dynamic Bayesian network.

[0039] For example, maximum likelihood estimation, expectation-maximization algorithms, or Bayesian parameter learning methods are used to calibrate the parameters in the state transition distribution and the observation distribution. Multiphysics simulation samples are used to cover operating condition combinations missing from the measured data, and historical service data are used to correct deviations between simulation and actual operation. In online applications, the latent state posterior distribution is updated based on real-time monitoring information to recursively predict future thermal stratification states, and the uncertainty of the analysis results is characterized by the dispersion of the predicted distribution, the confidence interval, and the probability of exceeding the risk threshold.

[0040] In another exemplary embodiment, the above-mentioned service characteristic dynamic perception intelligent agent includes a working condition semantic parsing module, a knowledge retrieval module, a candidate feature generation module, a numerical feature evaluation module, and a deduction task configuration module.

[0041] In another exemplary embodiment, step 103 above is based on the dynamic sensing agent of service characteristics, which analyzes the current operating condition and retrieves the thermal stratification operating condition knowledge base in each analysis cycle, autonomously selects monitoring features and configures thermal stratification state inference tasks, and extracts historical similar operating conditions from the thermal stratification operating condition knowledge base to construct the prior distribution of the current operating condition.

[0042] In another exemplary embodiment, the specific implementation process of parsing the current operating condition and retrieving the thermal stratification knowledge base in each analysis cycle, autonomously selecting monitoring features, and configuring thermal stratification state inference tasks is as follows: S31: Semantically encode the current operating events of nuclear power equipment and state encode the monitoring data of the same service period, and integrate the two to form a current operating condition representation.

[0043] Specifically, within each analysis cycle, the unit operation logs, control commands, equipment action records, and current service monitoring data are acquired. The operation logs and control commands are transformed into semantic descriptions representing the operation phase, operating condition changes, and adjustment events. Statistical features, trends, and time-series fluctuation features are extracted from the monitoring data to form a joint representation of the current operating condition.

[0044] S32: Based on the current working condition characterization, retrieve the thermal stratification working condition knowledge base to obtain historical similar working conditions and their monitoring experience, and generate candidate monitoring feature combinations by combining the large language model with the retrieval results; Specifically, the agent identifies the operating stage of the fluctuating pipe, the conditions for thermal stratification, and potential state changes based on the joint representation. It then generates operating condition descriptions and retrieval conditions, and jointly retrieves historically similar operating conditions with similar semantics and matching operating parameters from the knowledge base to obtain corresponding effective monitoring features, state information, and analytical experience. A pre-constructed library of thermal stratification monitoring feature operators is used, including temperature differences between the upper and lower pipe walls, temperature differences between different axial sections, temperature change rate, fluctuation amplitude, and temperature difference between the voltage regulator and the hot section. The agent combines the retrieval results with currently available sensor data to select and combine candidate features from the feature operator library.

[0045] S33: Based on the correlation between candidate monitoring features and thermal stratification state, the redundancy between features and the reliability of corresponding data, evaluate the candidate monitoring feature combination, determine the target monitoring feature combination, and configure the corresponding perception time window and state inference time domain to form a thermal stratification state inference task.

[0046] Specifically, the evaluation is based on the correlation between each feature and the thermal stratification state, the redundancy between features, and the data reliability. Features with low correlation, high repetition, or insufficient quality are eliminated to determine the target monitoring feature combination. Based on the current operating condition change rate, the evolution duration of similar historical operating conditions, and the data status of the target monitoring features, the sensing time window, data reading interval, and state inference time domain are configured. The thermal stratification state variables that need to be estimated and predicted are identified, forming a structured thermal stratification state inference task. This task is then passed to a physically guided dynamic Bayesian network to limit the range of input data, the state inference object, and the prediction time domain within the current analysis cycle.

[0047] In another exemplary embodiment, the specific process of extracting historical similar operating conditions from the thermal stratification operating condition knowledge base to construct the prior distribution of the current operating condition is as follows.

[0048] S41: Based on the degree of matching between the historical similar operating conditions and the current operating conditions, the posterior information of the thermal stratification state of each historical similar operating condition extracted from the thermal stratification operating condition knowledge base is weighted and fused to form the prior distribution of the current operating condition, which serves as the initial prior distribution of the state.

[0049] For example, based on the current operating condition characterization and deduction task of the wave tube, historical operating conditions with similar operating stages, boundary parameter change trends and monitoring feature distributions are retrieved from the knowledge base. The posterior distribution of the thermal stratification state of each historical operating condition is weighted and fused according to the degree of matching to form the state prior distribution at the beginning of the current cycle, i.e. the initial state prior distribution. If there is no matching operating condition, the initial state distribution of the Bayesian network is used as the prior.

[0050] In another exemplary embodiment, in step 104 above, the physically guided dynamic Bayesian network is used to fuse current service monitoring data, estimate the thermal stratification state, and predict its evolution and uncertainties. The specific implementation process is as follows.

[0051] S51: Extract the current service monitoring data according to the thermal stratification state inference task, and input it into the physical-guided dynamic Bayesian network to obtain the posterior distribution of the current thermal stratification state through Bayesian inference.

[0052] In this step, based on the target monitoring characteristics and time window determined by the simulation task, the wall temperature, pressure, flow rate and related operating parameters within the current period are extracted. After data preprocessing, they are used as observation evidence and input into the physical-guided dynamic Bayesian network. Based on the state prior, Bayesian updates are performed to obtain the posterior distribution and estimated values ​​of the latent states such as the temperature difference between the upper and lower fluids, the location of the layer interface, the axial temperature gradient and thermal stress at the current moment.

[0053] Specifically, let the thermal stratification hidden state be... Current service monitoring information is The running boundary conditions are Then the Bayesian update relation is: in, For the thermal stratification hidden state variables of time slice t, Let be the observed variable for time slice t. For the execution boundary variables of time slice t, The posterior distribution of the state after fusing the observed variables at time slice t, Let represent the observation probability relationship between the observed variables and the latent state variables of the thermal stratification at time slice t. Let be the prior distribution of the state in time slice t. The marginal probability of the observed variable is defined by the boundary variables of the operation at time slice t. It represents the state prior distribution obtained by fusing historical similar operating conditions and is used to normalize the state posterior distribution.

[0054] S52: Based on the posterior distribution and the temporal transition relationship of the physically guided dynamic Bayesian network, the predicted distribution of the thermal stratification state in the preset extrapolation time domain is obtained recursively, and the state evolution trend and its uncertainty are determined according to the predicted distribution.

[0055] Starting with the current posterior distribution, the predicted distributions of state variables for each heat stratification within the preset extrapolation time domain are obtained by recursively extrapolating along the time slices of the dynamic Bayesian network. Specifically, this is based on the state transition relationships in the dynamic Bayesian network: The predicted distribution of thermal stratification states at future times is obtained recursively. Let represent all observed variables up to time slice t, indicating the service monitoring information up to time slice t. Let these be the hidden state variables of the hot stratification at future time t+k. Let these be the hidden state variables of the hot stratification at future time t+k-1. Let be the boundary variables for operation at a future time t+k. Predict the distribution of thermal stratification states for future time t+k. Hidden state variables for thermal stratification are determined by Transfer to The probability, Predict the distribution of thermal stratification states for future time t+k-1.

[0056] The evolution trend of the state is determined based on the expected prediction, and the uncertainty is quantified based on the dispersion of the predicted distribution, the confidence interval, and the probability of exceeding the risk threshold, so as to provide a basis for the judgment and analysis results of thermal stratification risk.

[0057] In another exemplary embodiment, the specific implementation process of step 105 above is as follows.

[0058] S51: Based on the posterior distribution and predicted distribution of the current thermal stratification state, output the state estimate, evolution trend, confidence interval, and probability of exceeding the risk threshold to form a structured analysis result.

[0059] Specifically, based on the posterior distribution of the current thermal stratification state of the fluctuating pipe and the state prediction distribution within the preset extrapolation time domain, the system outputs state estimation results such as the temperature difference between the upper and lower fluids, the location of the stratification interface, the axial temperature gradient, and the thermal stress level, and provides the corresponding evolution trend, confidence interval, and probability of exceeding the risk threshold. The system determines the thermal stratification risk level of the current operating condition according to preset risk classification rules, forming a structured analysis result that includes operating condition information, target monitoring characteristics, state estimation results, evolution prediction results, and uncertainty characterization.

[0060] S52: Based on the deviation between subsequent service monitoring data and prediction results, correct the state transition parameters and observation parameters of the physically guided dynamic Bayesian network.

[0061] Specifically, as subsequent service monitoring data arrives, the thermal stratification state at the corresponding time point is re-estimated, and the updated state estimation result is compared with the prediction result of the previous analysis period to obtain the state prediction deviation. When the prediction deviation exceeds a preset update threshold and the monitoring data quality meets the requirements, the state transition parameters and observation parameters of the physical-guided dynamic Bayesian network are corrected while maintaining the network topology and physical constraints, so as to adapt to the changes in the service characteristics of nuclear power equipment.

[0062] S53: Write the current operating conditions, target monitoring features, and thermal stratification analysis results into the thermal stratification operating condition knowledge base, and adjust the intelligent agent's operating condition retrieval and monitoring feature optimization configuration for use in the next analysis cycle.

[0063] The operating conditions, target monitoring characteristics, thermal stratification state estimation results, evolution prediction results, and subsequent verification results of this analysis cycle are correlated and written into the thermal stratification operating condition knowledge base to form a new case study for thermal stratification analysis of the fluctuating pipe. The service characteristic dynamic sensing agent adjusts the historical operating condition retrieval weight and monitoring feature validity score based on the verification results, increases the priority of calling valid features, and updates the target monitoring features, sensing time window, and extrapolation time domain configuration for the next analysis cycle.

[0064] Compared with the prior art, the present invention has the following advantages: 1) This invention analyzes the current operating conditions through a dynamic sensing agent of service characteristics, and autonomously selects monitoring features and configuration status to deduce tasks in combination with historical similar operating conditions, thereby avoiding the problem that fixed monitoring features are difficult to adapt to changes in operating conditions and improving the effective utilization of multi-source service data.

[0065] 2) This invention integrates the thermal stratification formation mechanism, historical operating condition knowledge and real-time monitoring data through a physically guided dynamic Bayesian network to achieve probability estimation and evolution prediction of thermal stratification state, and characterize the uncertainty of the prediction results, thereby reducing the limitations of simply relying on data models or offline simulation.

[0066] 3) This invention uses subsequent service monitoring data to correct the state inference model and updates the thermal stratification condition knowledge base and monitoring feature configuration, thereby improving the continuous adaptability and reliability of thermal stratification analysis under long-term complex operating conditions.

[0067] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0068] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent for service characteristics, characterized in that, include: Construct a thermally hierarchical operating condition knowledge base for nuclear power equipment, a physically guided dynamic Bayesian network, and a service characteristic dynamic perception agent based on a large language model; Obtain the current operating parameters of nuclear power equipment; The current operating parameters include: operation log, control commands, equipment action records, and current service monitoring data; Based on the current operating parameters, the service characteristics dynamic sensing agent is used to retrieve the thermal stratification operating condition knowledge base to form a target monitoring feature combination, thermal stratification state inference task, and initial state prior distribution. Based on the target monitoring feature combination, the thermal stratification state inference task, and the initial state prior distribution, a physical-guided dynamic Bayesian network is used for analysis to obtain the thermal stratification state prediction distribution, state evolution trend, and uncertainty within the preset inference time domain.

2. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 1, characterized in that, Based on the target monitoring feature combination, the thermal stratification state inference task, and the initial state prior distribution, a physical-guided dynamic Bayesian network is used for analysis to obtain the predicted distribution of thermal stratification states, state evolution trends, and uncertainties within the preset inference time domain. This is followed by: The physical-guided dynamic Bayesian network is corrected based on the service monitoring data obtained in the preset time domain, and the thermal stratification knowledge base and the service characteristic dynamic sensing agent are updated.

3. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 1, characterized in that, Based on the current operating parameters, the service characteristic dynamic sensing agent is used to retrieve the thermal stratification operating condition knowledge base, forming a target monitoring feature combination, a thermal stratification state deduction task, and an initial state prior distribution, specifically including: Transform operation logs and control commands into semantic descriptions that represent operation phases, changes in operating conditions, and adjustment events; Extract statistical features, trends, and time-series fluctuation characteristics from current service monitoring data; Based on the semantic description, the statistical features, the changing trends, and the temporal fluctuation features, a joint representation of the current operating condition is formed. Retrieve historical operating conditions that match the joint representation of the current operating condition from the thermal stratification operating condition knowledge base, and use them as the first historical matching operating condition; Based on the effective monitoring features corresponding to each first historical matching working condition and the available sensor data under the current working condition, candidate features are selected from the feature operator library. Based on the correlation between each candidate feature and the hot stratification state, the redundancy between candidate features, and the credibility of candidate features, one or more target monitoring features are selected from each candidate feature to form a target monitoring feature combination; A structured thermal stratification state deduction task is formed based on the rate of change of current operating condition parameters, the evolution duration of each first historical matching operating condition, and the data status of each target monitoring feature in the target monitoring feature combination. Based on the current joint characterization of operating conditions and the thermal stratification state inference task, retrieve historical operating conditions that match both the current joint characterization of operating conditions and the thermal stratification state inference task from the thermal stratification operating condition knowledge base, and use them as the second historical matching operating conditions. The posterior distribution of the hot-layered state of each second historical matching condition is weighted and fused according to the degree of matching between each second historical matching condition and the current condition to form the prior distribution of the initial state.

4. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 1, characterized in that, Based on the target monitoring feature combination, the thermal stratification state inference task, and the initial state prior distribution, a physical-guided dynamic Bayesian network is used for analysis to obtain the predicted distribution of thermal stratification states, state evolution trends, and uncertainties within the preset inference time domain, specifically including: The task parameters of the physical-guided dynamic Bayesian network are set based on the thermal stratification state inference task, and the composition of the thermal stratification hidden state variables of the physical-guided dynamic Bayesian network is determined based on the combination of target monitoring features. Starting from the initial state prior distribution, the thermal stratification state prediction distribution within the preset extrapolation time domain is obtained by recursively extrapolating along the time slices of the physically guided dynamic Bayesian network. The expected prediction is calculated based on the predicted distribution of thermal stratification state within the preset simulation time domain. The state evolution trend is determined based on the predicted expectation; Based on the degree of dispersion and confidence interval of the predicted distribution of thermal stratification state within the preset extrapolation time domain, uncertainty is quantified using the probability of exceeding the risk threshold.

5. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 1 or 4, characterized in that, The probability relationship of a single time slice in the physically guided dynamic Bayesian network is expressed as follows: ； in, For the thermal stratification hidden state variables of time slice t, Let be the observed variable for time slice t. For the execution boundary variables of time slice t, The posterior distribution of the state after fusing the observed variables at time slice t, Let represent the observation probability relationship between the observed variables and the latent state variables of the thermal stratification at time slice t. Let be the prior distribution of the state in time slice t. The marginal probabilities of observed variables are determined under the constraints of the boundary variables at time slice t, and are used to normalize the posterior distribution of the state.

6. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 1 or 4, characterized in that, The joint probability distribution of time-slice recursion in the physically guided dynamic Bayesian network is expressed as follows: ； in, For all observed variables up to time slice t, Let these be the hidden state variables of the hot stratification at future time t+k. Let these be the hidden state variables of the hot stratification at future time t+k-1. Let be the boundary variables for operation at a future time t+k. Predict the distribution of thermal stratification states for future time t+k. Hidden state variables for thermal stratification are determined by Transfer to The probability, Predict the distribution of thermal stratification states for future time t+k-1.

7. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 6, characterized in that, The state transition relationship between adjacent time slices in a physics-guided dynamic Bayesian network is as follows: ； ； in, , These are the thermally layered hidden state variables for time slices t and t-1, respectively. Let be the observed variable for time slice t. For the execution boundary variables of time slice t, , These represent state evolution error and observation noise, respectively. The state transition function is established based on the causal dependency between the thermal stratification state and the operating boundary. This is an observation mapping function established based on the observation relationship between the thermal stratification hidden state and service monitoring information.

8. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 1, characterized in that, The thermal stratification operating condition knowledge base for nuclear power equipment includes: Thermal stratification knowledge ternary set under different historical operating conditions; Each thermal stratification condition knowledge triplet corresponds to an attribute feature; the thermal stratification condition knowledge triplet is associated with and stored with its corresponding attribute feature, which includes historical monitoring sequences, state estimation results, and evolutionary features; Each thermal stratified operating condition knowledge triple corresponds to a semantic index for operating condition description and a numerical index for operating parameters.

9. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 1, characterized in that, The service characteristic dynamic perception intelligent agent includes: a working condition semantic parsing module, a knowledge retrieval module, a candidate feature generation module, a numerical feature evaluation module, and a deduction task configuration module.

10. The method for thermal stratification analysis of nuclear power equipment driven by a dynamic sensing intelligent agent according to claim 2, characterized in that, The physical-guided dynamic Bayesian network is corrected based on the service monitoring data obtained subsequently within the preset time domain, and the thermally hierarchical operating condition knowledge base and the service characteristic dynamic sensing agent are updated, specifically including: In subsequent monitoring, service monitoring data within the preset simulation time domain will be acquired; Thermal stratification state estimation is performed on the service monitoring data obtained within the preset simulation time domain to obtain the estimated distribution of thermal stratification state within the preset simulation time domain. Compare the estimated distribution of thermal stratification states with the predicted distribution of thermal stratification states in the preset extrapolation time domain, and calculate the state prediction deviation; When the absolute value of the state prediction deviation exceeds a preset threshold, the state transition parameters and observation parameters of the physically guided dynamic Bayesian network are corrected. The state transition parameters are the parameters of the thermally hierarchical state transition function, and the observation parameters are the parameters in the observation mapping function. The system associates the current operating conditions, target monitoring characteristics, thermal stratification state estimation results, evolution prediction results, and comparative verification results of the current analysis cycle, and updates them to the thermal stratification operating condition knowledge base and adjusts the optimal configuration of the operating condition retrieval and monitoring characteristics of the dynamic sensing agent for service characteristics.

Images