High temperature gas cooled reactor safety analysis method and device, electronic equipment and storage medium

Abstract

The application discloses a high-temperature gas cooled reactor safety analysis method and device, electronic equipment and storage medium, relates to the nuclear reactor technical field, and includes: constructing a neural network model; pre-training and fine-tuning the model based on physical simulation data and experimental data; generating virtual training samples for extreme accident conditions to enhance the model generalization ability; inputting the real-time monitored irradiation loop operating parameters into the trained model, and outputting the target channel temperature prediction value and safety margin evaluation result. The problems of high calculation cost caused by dependence on complex physical models, difficulty in real-time prediction, insufficient experimental data leading to weak model generalization ability, and inability to quickly provide emergency decision support under extreme conditions are solved. The technical effects of accurately predicting temperature response, improving the adaptability of the model to different conditions, and providing efficient and real-time data support for high-temperature gas cooled reactor irradiation loop safety analysis and emergency decision-making are achieved.

Description

Technical Field

[0001] This application relates to the field of nuclear reactor technology, and in particular to a method, apparatus, electronic equipment and storage medium for safety analysis of high-temperature gas-cooled reactors. Background Technology

[0002] As an important development direction of fourth-generation advanced nuclear energy systems, the high-temperature gas-cooled reactor (HTGR) has its irradiation circuit as the core device for isotope production. Its safety is directly related to the core stability and the economy and reliability of isotope production.

[0003] Currently, safety analysis of irradiation circuits mainly relies on physical model simulations or experimental data, which has many limitations: traditional simulations require the establishment of complex heat transfer-fluid coupling models, resulting in high computational costs and difficulty in predicting transient temperature changes in real time; the special operating environment of irradiation circuits makes it difficult to obtain experimental data, thus limiting the model's generalization ability; temperature response is affected by multiple factors such as coolant flow rate and power level, and traditional methods are unable to efficiently extract nonlinear correlations between multiple variables; in the face of extreme conditions such as loss of flow accidents, existing methods cannot quickly assess safety margins and provide emergency decision support, making it difficult to meet actual engineering needs. Summary of the Invention

[0004] This application provides a method, apparatus, electronic device, and storage medium for safety analysis of high-temperature gas-cooled reactors. It addresses the problems in related technologies, such as high computational costs due to reliance on complex physical models, difficulty in real-time prediction, limited model generalization ability due to insufficient experimental data, difficulty in extracting multi-factor nonlinear correlations, and inability to quickly provide emergency decision support under extreme conditions.

[0005] According to a first aspect of this application, a method for safety analysis of a high-temperature gas-cooled reactor is provided, comprising: A neural network model is constructed, comprising an input layer, a hidden layer, and an output layer. The input layer is used to receive operating parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model. The neural network model is pre-trained and fine-tuned based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal and accident conditions, and the fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target condition. Virtual training samples are generated for extreme accident conditions, and the virtual samples are input into the neural network model to enhance its generalization ability to extreme conditions; The real-time monitored irradiation circuit operating parameters are input into the trained neural network model, and the target channel temperature prediction value and safety margin assessment result output by the neural network model are received to provide real-time data support for emergency decision-making.

[0006] According to a second aspect of this application, a high-temperature gas-cooled reactor safety analysis device is provided, comprising: The building module is configured to build a neural network model, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive running parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model. The optimization module is configured to pre-train and fine-tune the neural network model based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal working conditions and accident working conditions, and the fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target working condition. The generation module is configured to generate virtual training samples for extreme accident conditions and input the virtual samples into the neural network model to enhance its generalization ability to extreme conditions. The receiving module is configured to input the real-time monitored irradiation circuit operating parameters into the trained neural network model, and receive the target channel temperature prediction value and safety margin assessment result output by the neural network model, so as to provide real-time data support for emergency decision-making.

[0007] According to a third aspect of this application, an electronic device is provided, comprising: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, which are executed by at least one processor to enable the at least one processor to perform the high-temperature gas-cooled reactor safety analysis method described in the first aspect above.

[0008] According to a fourth aspect of this application, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute the high-temperature gas-cooled reactor safety analysis method described in the first aspect above.

[0009] According to a fifth aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements the high-temperature gas-cooled reactor safety analysis method as described in the first aspect above.

[0010] This application provides a method, apparatus, electronic device, and storage medium for safety analysis of a high-temperature gas-cooled reactor, comprising: constructing a neural network model, wherein the neural network model includes an input layer, a hidden layer, and an output layer, wherein the input layer is used to receive operating parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model; The neural network model is pre-trained and fine-tuned based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal and accident conditions, and the fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target condition. Virtual training samples are generated for extreme accident conditions, and the virtual samples are input into the neural network model to enhance its generalization ability to extreme conditions; The real-time monitored operating parameters of the irradiation circuit are input into the trained neural network model, which then outputs the target channel temperature prediction and safety margin assessment results, providing real-time data support for emergency decision-making. This application constructs a neural network model containing an input layer, hidden layers, and an output layer, enhancing its nonlinear expression capabilities. Pre-training and output layer parameter fine-tuning are performed based on physical simulation and experimental data. Virtual training samples are generated for extreme accident conditions to improve the model's generalization ability. Simultaneously, it supports inputting real-time irradiation circuit operating parameters and outputting target channel temperature predictions and safety margin assessment results. Therefore, it solves the problems in related technologies, such as high computational costs due to reliance on complex physical models, difficulty in real-time prediction, limited model generalization due to insufficient experimental data, difficulty in extracting multi-factor nonlinear correlations, and inability to quickly provide emergency decision support under extreme conditions. This achieves the technical effects of accurately predicting the irradiation circuit temperature response, improving the model's adaptability to different operating conditions, and providing efficient real-time data support for the safety analysis and emergency decision-making of high-temperature gas-cooled reactor irradiation circuits.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the embodiments of this application, the accompanying 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.

[0013] Figure 1 A flowchart illustrating a safety analysis method for a high-temperature gas-cooled reactor provided in an embodiment of this application; Figure 2 A flowchart illustrating another high-temperature gas-cooled reactor safety analysis method provided in this application embodiment; Figure 3 A flowchart illustrating another high-temperature gas-cooled reactor safety analysis method provided in this application embodiment; Figure 4 This is a schematic diagram of a high-temperature gas-cooled reactor safety analysis device provided in an embodiment of this application. Detailed Implementation

[0014] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0015] The following description, with reference to the accompanying drawings, outlines a method, apparatus, electronic device, and storage medium for safety analysis of high-temperature gas-cooled reactors according to embodiments of this application.

[0016] Figure 1 This is a flowchart illustrating a safety analysis method for a high-temperature gas-cooled reactor provided in an embodiment of this application.

[0017] like Figure 1 As shown, the method includes the following steps: Step 101: Construct a neural network model. The neural network model includes an input layer, a hidden layer, and an output layer. The input layer is used to receive operating parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model.

[0018] In some embodiments, the constructed neural network model adopts a multilayer perceptron architecture, which has the ability to efficiently handle multivariate inputs and complex mapping relationships, and is the core foundation for achieving accurate safety analysis of the irradiation loop of a high-temperature gas-cooled reactor. The model consists of an input layer, a hidden layer, and an output layer, with each layer working together to complete the entire process from inputting operating parameters to outputting prediction results. The input layer, as the interface between the model and actual operating data, has multiple neurons, each corresponding to a key parameter in the operation of the irradiation loop. These parameters cover the core factors affecting the temperature response of the irradiation loop, including relevant coolant parameters, core power level, target channel geometry, ambient temperature and pressure, target heat release characteristics, and operating status indicators, comprehensively capturing key information in the operation of the irradiation loop. The hidden layer is a key part to enhance the nonlinear expression capability of the model. Through reasonable layer and node design, the model can deeply explore the complex relationship between input parameters and target channel temperature, breaking away from the dependence on empirical formulas in traditional methods and efficiently handling the analytical challenges caused by the coupling of multiple factors. The output layer focuses on accurately outputting prediction results, directly providing core data support for the safety analysis of the irradiation loop. The construction of this neural network model effectively solves the problem that traditional analysis methods have difficulty in extracting multivariate nonlinear correlations, laying a solid foundation for subsequent accurate prediction of temperature response and significantly improving the scientificity and efficiency of safety analysis of the irradiation loop of high-temperature gas-cooled reactors.

[0019] Step 102: The neural network model is pre-trained and fine-tuned based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal working conditions and accident working conditions. The fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target working condition.

[0020] In some embodiments, the neural network model is optimized through collaborative training using dual data sources, laying the foundation for the model to accurately adapt to the safety analysis requirements of the high-temperature gas-cooled reactor irradiation circuit. The data sources used for training include physical simulation data and experimental data. The physical simulation data covers temperature-time series data under normal and accident conditions of the irradiation circuit. This type of data comprehensively covers various states that may occur during the operation of the irradiation circuit, without being limited by the harsh conditions of the actual operating environment, and provides rich and diverse training samples. The pre-training phase is based on this simulation data, initializing the parameters of the neural network model in a data-driven manner. This allows the model to capture the general patterns of the irradiation circuit's temperature response in the early stages of training, avoiding inefficient training or convergence difficulties caused by random initial parameters, and laying a solid foundation for subsequent model performance optimization. The experimental data comes from historical operating records of the laboratory or prototype reactor irradiation circuit, realistically reflecting the mapping relationships of various parameters in actual operating scenarios, and has extremely high engineering practicality. The fine-tuning phase utilizes experimental data specific to these target operating conditions, optimizing only the output layer parameters of the neural network model. This preserves the general patterns learned by the model during pre-training while ensuring the model accurately reflects the actual characteristics of the specific target operating conditions. This eliminates the need to retrain the entire model, significantly improving training efficiency and reducing computational costs. This pre-training followed by fine-tuning strategy fully integrates the advantages of broad and readily available physical simulation data with the reliability of experimental data. It effectively addresses the problem of limited model generalization ability caused by insufficient experimental data in related technologies, allowing the model to maintain versatility while possessing precise target operating condition adaptation capabilities, significantly improving the accuracy and reliability of subsequent temperature predictions.

[0021] Step 103: Generate virtual training samples for extreme accident conditions and input the virtual samples into the neural network model to enhance its generalization ability to extreme conditions.

[0022] In some embodiments, extreme accident conditions in the irradiation circuit of a high-temperature gas-cooled reactor include loss-of-current accidents. These conditions have a low probability of occurrence, extremely harsh operating environments, and are accompanied by complex conditions such as high temperature and high radiation. Obtaining real experimental data is not only extremely difficult but also poses safety risks and high costs, resulting in insufficient training data for the model under these conditions and limited generalization ability. To fill this gap, this step generates virtual training samples by simulating the changes in core parameters under extreme accident conditions. The generation process strictly follows the physical operating logic. For example, by simulating the evolution trends of key parameters under extreme scenarios such as sudden drops in coolant flow and abnormal power fluctuations, virtual data that matches the characteristics of actual extreme conditions is constructed to ensure that the samples can truly reflect the correlation of the irradiation circuit's operating state under extreme conditions. These virtual training samples are then input into the neural network model, allowing the model to fully learn the complex mapping relationship between various operating parameters and the target channel temperature response under extreme conditions during training, filling the training gap caused by insufficient real data. This approach effectively solves the problem of poor adaptability of traditional methods to extreme accident conditions, greatly enhances the model's generalization ability in extreme scenarios, and enables the model to accurately capture temperature change patterns under extreme conditions, providing reliable assurance for safety analysis when extreme accidents occur.

[0023] Step 104: Input the real-time monitored irradiation circuit operation parameters into the trained neural network model, and receive the target channel temperature prediction value and safety margin assessment result output by the neural network model to provide real-time data support for emergency decision-making.

[0024] In some embodiments, during actual operation, the system continuously collects real-time operating parameters of the irradiation loop. These parameters cover key information such as coolant flow rate, core power level, target channel geometry, ambient temperature and pressure, target heat release characteristics, and operating conditions, ensuring that the data comprehensively and accurately reflects the real-time operating status of the irradiation loop. This real-time monitoring data is directly input into a neural network model that has been pre-trained, fine-tuned, and augmented with extreme condition samples. Leveraging its learned multi-factor nonlinear correlation laws and adaptability to different operating conditions, the model can quickly process and analyze the input data without relying on complex physical model simulations. Subsequently, the model accurately outputs predicted target channel temperatures and safety margin assessment results. The predicted target channel temperatures visually represent the temperature change trend in the core area of ​​the loop, while the safety margin assessment clearly defines the difference between the current operating state and the safety threshold, clarifying the potential risk level. These outputs are fed back to relevant personnel and the monitoring system in real time, providing timely and reliable data support for emergency decision-making. When encountering abnormal temperatures or extreme accident conditions, it helps personnel quickly assess the risk level and formulate response strategies. This step effectively solves the problems of high computational cost, difficulty in real-time prediction, and inability to quickly provide emergency decision support in traditional methods. It significantly improves the timeliness of irradiation loop operation monitoring and the scientific nature of emergency response, providing strong support for the safe and stable operation of high-temperature gas-cooled reactors.

[0025] Compared with related technologies, in this embodiment, a neural network model is constructed, comprising an input layer, a hidden layer, and an output layer. The input layer receives operating parameters, the hidden layer enhances the nonlinear expression capability of the neural network model, and the output layer outputs the prediction results of the neural network model. The neural network model is pre-trained and fine-tuned based on physical simulation data and experimental data. The pre-training initializes the neural network model parameters using simulation data from normal and accidental operating conditions, while the fine-tuning optimizes the output layer parameters of the neural network model using experimental data from the target operating condition. Virtual training samples are generated for extreme accidental operating conditions and input into the neural network model to enhance its generalization ability to extreme conditions. Real-time monitored irradiation circuit operating parameters are input into the trained neural network model, and the target channel temperature prediction value and safety margin assessment results output by the neural network model are received, providing real-time data support for emergency decision-making. It can solve the problems in related technologies, such as high computational costs and difficulty in real-time prediction due to reliance on complex physical models, limited model generalization ability due to insufficient experimental data, difficulty in extracting nonlinear correlations of multiple factors, and inability to quickly provide emergency decision support under extreme conditions. It achieves the technical effects of accurately predicting the temperature response of the irradiation circuit, improving the model's adaptability to different operating conditions, and providing efficient real-time data support for the safety analysis and emergency decision-making of the irradiation circuit of the high-temperature gas-cooled reactor.

[0026] Figure 2 A flowchart illustrating another high-temperature gas-cooled reactor safety analysis method provided in this application embodiment includes the following steps: Step 201: The input layer is configured with multiple neurons to receive input data such as coolant flow rate, core power level, target channel geometry parameters, ambient temperature and pressure, target identification, and accident condition identification.

[0027] In some embodiments, the input layer serves as the core of interaction between the neural network model and the irradiation loop operating data. By configuring multiple neurons, it comprehensively receives and initially transmits key operating parameters. The received coolant flow rate includes parameters directly affecting heat transfer efficiency, such as inlet flow rate and velocity. The core power level includes local power density and total power, reflecting the overall and local state of core heat release. Target channel geometry parameters focus on structural characteristics affecting heat transfer paths, such as channel length and diameter. Ambient temperature and pressure include key indicators reflecting the overall operating environment, such as primary loop outlet temperature and system pressure. Target identifiers correspond to different target types with varying heat release rates. Accident condition identifiers are used to determine whether the current operating mode is a special operating mode such as a loss-of-flow accident. These parameters comprehensively cover the core factors affecting the irradiation loop temperature response, ensuring that the input layer provides the model with complete and accurate raw data. Its multi-neuron configuration allows each parameter to be processed independently and efficiently, avoiding data confusion or omission of key information. The beneficial effect of this design is that it achieves comprehensive and accurate input of key operating parameters, providing a complete and reliable data foundation for subsequent feature learning and prediction, and improving the relevance and completeness of the input data.

[0028] Step 202: Configure 3 hidden layers with 64, 32 and 16 nodes respectively, and use a preset activation function to enhance the nonlinear expressive ability of the neural network model.

[0029] In some embodiments, the hidden layer adopts a three-layer progressive structure design, with the number of nodes set to 64, 32, and 16 respectively. This decreasing node configuration enables gradual filtering of input data and focus on core features. First, 64 nodes perform comprehensive preliminary processing of multi-dimensional input data; then, 32 nodes extract key correlation features; and finally, 16 nodes focus on the core mapping relationship, effectively filtering redundant information and improving the efficiency and accuracy of feature extraction. The selected preset activation function is the ReLU function, which has piecewise linear properties and can maintain gradient stability during forward propagation, effectively alleviating the gradient vanishing problem during model training and allowing deep networks to learn features efficiently. Simultaneously, the ReLU function has a simple calculation logic, requiring no complex exponential or logarithmic operations, which can accelerate the model's training convergence speed. Compared to traditional activation functions, it is more suitable for complex feature learning needs under multi-factor coupling. The beneficial effect of this design is that, through reasonable layer and node configuration and efficient activation functions, it significantly enhances the model's non-linear expressive ability, allowing the model to deeply explore the complex correlations between multiple parameters, laying the algorithmic foundation for accurate prediction.

[0030] Step 203: The output layer is configured with a single neuron to map the features learned by the hidden layer to specific engineering prediction values.

[0031] In some embodiments, the output layer, configured with a single neuron, achieves centralized mapping and transformation of the abstract features learned by the hidden layer. Its core objective is to output intuitive and usable engineering predictions. The single-neuron design avoids interference from multiple outputs, ensuring the focus of the prediction target and allowing the model to concentrate its computational power on precise output of core indicators. The mapped engineering predictions are specifically the target channel temperature or temperature change rate. These two indicators are the core basis for evaluating the safety of the irradiation circuit, directly reflecting the thermal operating state of the target channel and providing clear reference for personnel to determine whether the circuit is within a safe range. The output layer, through specific mapping logic, transforms the multi-dimensional, abstract feature information extracted by the hidden layer into concrete numerical results, eliminating the abstractness of the features and giving them direct engineering application value. The beneficial effect of this design is that it achieves precise transformation from abstract features to specific engineering indicators, providing intuitive and targeted output results. It provides direct and core judgment criteria for irradiation circuit safety analysis, improving the practicality and operability of the model results.

[0032] Step 204: Optimize the gap between the predicted value and the true value of the neural network model through the loss function, and dynamically adjust the learning rate through the adaptive moment estimation optimizer.

[0033] In some embodiments, to accurately reduce the deviation between the predicted values ​​and the true values ​​of the neural network model and ensure the reliability of the model output, this step achieves efficient optimization of model parameters by reasonably selecting a loss function and optimizer. The selected loss function is the mean squared logarithmic error. The core advantage of this function is that it measures the prediction error in logarithmic space, which can effectively focus on the relative error problem in the temperature prediction scenario, avoiding the dominance of the entire loss calculation process by the existence of some large numerical samples. This allows the model to focus more on the accurate fitting of temperature change trends under different operating conditions, rather than simply pursuing the matching of absolute values. The corresponding Adaptive Moment Estimation (Adam) optimizer integrates the core advantages of the momentum method and the RMSProp algorithm. During the parameter update process, it not only considers the first moment estimation of the gradient but also takes into account the second moment estimation, and can adaptively adjust the learning rate of each parameter. This adaptive adjustment mechanism makes it particularly suitable for sparse gradient and noisy data scenarios, which can accelerate the convergence speed of the model and avoid the problem of parameter oscillation caused by an excessively large learning rate or training stagnation caused by an excessively small learning rate. Through the synergy of the loss function and the Adam optimizer, the model can continuously and dynamically correct its parameters during training, constantly narrowing the gap between the predicted and actual values. This significantly improves the stability and efficiency of model training, providing an algorithmic guarantee for the subsequent accurate output of target channel temperature-related prediction results.

[0034] Step 205: Apply L2 regularization constraints to the output layer parameters using regularization coefficients, and prevent overfitting of the neural network model using early stopping.

[0035] In some embodiments, to ensure the generalization ability of the neural network model and avoid overfitting during training, this step optimizes and controls model training through the synergistic application of L2 regularization constraints and early stopping. When fine-tuning the output layer parameters of the neural network model, L2 regularization constraints are introduced. By setting a reasonable regularization coefficient, the weights of the output layer parameters are limited, preventing excessive growth of some parameter weights that could lead to the model over-reliance on details or even noise in the training data. This encourages the model to learn more universal core patterns from the data, maintaining the rationality and simplicity of parameter configuration. Simultaneously, early stopping is used to monitor the training process. The model's performance on the validation set is continuously tracked during training. When the model's prediction accuracy on the validation set no longer improves or even shows a downward trend, the training process is stopped promptly to prevent the model from continuing to learn redundant information and noise in the training set, thus curbing overfitting from the perspective of training duration. The combined use of L2 regularization constraints and early stopping provides dual protection from both parameter constraints and training regulation dimensions, effectively solving the problem of model overfitting and ensuring that the model maintains good performance on both the training and validation sets. This significantly improves the model's adaptability to different working conditions and the stability of prediction results.

[0036] Step 206: Disturb the historical operating parameters under normal operating conditions to simulate the equipment state under the extreme accident conditions.

[0037] In some embodiments, real-world operational data under extreme accident conditions are difficult to use directly for model training due to their low probability of occurrence, stringent acquisition conditions, and safety risks. However, historical operational parameters under normal operating conditions encompass the core indicators of a stable irradiation circuit, possessing reliable physical laws and serving as a high-quality foundation for generating virtual training samples. This step simulates the equipment state under extreme accident conditions by scientifically and reasonably perturbing these historical operational parameters. The perturbation process strictly follows the physical operating logic of the irradiation circuit, targeting key parameters affecting temperature response, such as a sudden drop in coolant flow during a loss-of-flow accident, abnormal fluctuations in core power under extreme loads, or sudden changes in system pressure. This ensures that the parameter changes after perturbation conform to the actual evolution of extreme conditions, rather than being random modifications. In this way, normal operating condition data is transformed into virtual samples that reflect the equipment's operating characteristics under extreme accidents, ensuring the authenticity and effectiveness of the samples while avoiding the high costs and safety risks of real extreme condition experiments. The beneficial effect of this processing method is the efficient generation of virtual extreme condition samples that conform to physical laws, filling the gap of scarce extreme accident data and providing ample material for the model to learn the characteristics of extreme conditions.

[0038] Step 207: The virtual training samples generated after the perturbation process are merged with the original training data to form an enhanced training dataset.

[0039] In some embodiments, the original training data includes physical simulation data and experimental data, covering the operational characteristics of normal and general accident conditions. This data serves as the core basis for the model to grasp the basic operational laws of the irradiation circuit. The virtual training samples generated in step 206 focus on the specific characteristics of extreme accident conditions, thus effectively complementing each other. This step merges the perturbation-processed virtual training samples with the original training data to construct an enhanced training dataset. During the merging process, it is necessary to ensure that the two types of data maintain consistency in format and dimensions. Unified data standardization is used to eliminate potential differences and avoid interfering with model training. Simultaneously, the ratio of original data to virtual samples is reasonably allocated to ensure that the model consolidates its understanding of basic operating conditions while allowing for sufficient learning of extreme operating condition characteristics. The enhanced training dataset not only expands the data scale but, more importantly, improves the coverage of operating conditions, allowing the model to access data from all scenarios, from normal operation to extreme accidents, during training. This enriches the diversity and completeness of the training data, providing data support for the model to build a comprehensive understanding of operating condition characteristics and laying the foundation for subsequently improving the model's generalization ability to extreme conditions.

[0040] Step 208: Input the enhanced training dataset into the neural network model, and freeze some hidden layers of the neural network model during training, updating the parameters only for the unfrozen layers to adapt them to the characteristics of the extreme accident conditions.

[0041] In some embodiments, after inputting the enhanced training dataset into the neural network model, this step employs a selective parameter update training strategy, i.e., freezing some hidden layers of the model and updating parameters only for the unfrozen layers. The frozen hidden layers are typically those that have fully learned the basic operating rules of the irradiation circuit during the pre-training phase with the original training data. These layers possess universal knowledge of multi-parameter nonlinear correlations and basic temperature response logic; freezing them prevents the loss of existing knowledge due to retraining and reduces the computational cost of parameter updates, thus improving training efficiency. The unfrozen layers are mainly some hidden layers or output layers near the output layer. These layers are directly responsible for the final feature mapping and prediction output. Updating their parameters using the enhanced dataset allows the model to specifically learn the special characteristics of extreme accident conditions and quickly adapt to the parameter correlation rules under extreme conditions. This training method retains the basic capabilities already learned by the model while achieving accurate adaptation to extreme condition features. While improving training efficiency and reducing computational costs, it significantly enhances the model's generalization ability to extreme accident conditions, ensuring that the model can output accurate prediction results in extreme scenarios.

[0042] Step 209: Input the real-time collected coolant flow rate, core power, target channel geometric parameters, system pressure and temperature, and accident condition identifier into the trained neural network model and obtain the real-time temperature prediction value or temperature change rate of the target channel output by the neural network model.

[0043] In some embodiments, during the actual operation of the high-temperature gas-cooled reactor irradiation loop, the real-time data acquisition system continuously captures various key operating parameters, which are the core basis for reflecting the current operating status of the loop. The collected coolant flow rate is directly related to the heat transfer efficiency, and its dynamic changes can instantly reflect the fluctuations in the loop's heat dissipation capacity; the core power parameters accurately reflect the core heat release intensity and are the core factors affecting the target channel temperature; the target channel geometric parameters, as fixed structural features, determine the inherent path and efficiency of heat transfer; system pressure and temperature intuitively present the stability of the overall operating environment of the loop, and their abnormal changes may indicate potential risks; the accident condition indicator clarifies whether the current operating mode is under special conditions such as a loss-of-current accident, providing a background reference for the model. These real-time collected parameters are synchronously input into the trained neural network model. Leveraging its learned multi-parameter nonlinear correlation laws and adaptability to different operating conditions, the model can quickly perform in-depth processing of the input data without complex physical model simulation calculations, directly outputting the real-time temperature prediction value or temperature change rate of the target channel. The beneficial effect of this process is that it enables the rapid conversion of operating parameters into core temperature indicators, providing timely and accurate basic data for subsequent safety assessments and ensuring the real-time and targeted nature of safety analysis.

[0044] Step 210: Based on the comparison between the predicted temperature value or the rate of temperature change and the preset safety threshold, calculate the safety margin under the current operating conditions.

[0045] In some embodiments, the preset safety threshold is established based on the design standards of the high-temperature gas-cooled reactor irradiation loop, the tolerance limits of equipment materials, long-term operating experience, and safety regulations. It serves as a key benchmark for determining the safety of the loop operation, and its value has undergone multiple rounds of verification and calibration to ensure its scientific validity and reliability. After obtaining the predicted temperature value or temperature change rate of the target channel, a quantitative analysis logic is constructed by systematically comparing this data with the preset safety threshold: if the predicted temperature value is used, the difference between the threshold and the predicted value is calculated; the magnitude of the difference directly reflects the distance between the current operating condition and a dangerous state. If the temperature change rate is used, the threshold is combined to determine whether the temperature change trend exceeds the safe evolution range, avoiding safety hazards caused by rapid temperature increases. Through this comparison process, the safety margin under the current operating condition is accurately calculated. This margin clearly quantifies the safety status of the loop, allowing operators to intuitively grasp the risk level of the current operating condition. Its beneficial effect lies in transforming abstract temperature data into clear quantitative safety indicators, providing a clear and operable basis for subsequent risk assessment and decision-making, and improving the objectivity and accuracy of safety analysis.

[0046] Step 211: When the safety margin is lower than the set threshold, a temperature anomaly warning is triggered and a corresponding emergency response suggestion is generated. The suggestion is used to guide the operators to perform intervention operations.

[0047] In some embodiments, when the calculated safety margin is lower than a set threshold, it indicates that the irradiation circuit is approaching its safety boundary and there is a risk of a safety accident caused by abnormal temperature. At this point, the system automatically triggers a temperature anomaly warning mechanism. The warning signal can be delivered to operators through intuitive methods such as monitoring platform pop-ups and audible / visual prompts, ensuring that risk information can be quickly perceived. Simultaneously, the system generates corresponding emergency response recommendations based on a large amount of operating data, accident handling patterns, and engineering practice experience learned during the neural network model training process. These recommendations are highly targeted; for example, if the safety margin is low due to insufficient coolant flow, it is recommended to adjust the cooling system valves to increase the flow rate; if the core power is too high, it is recommended to reasonably reduce the power output, providing clear operational guidance for operators. Operators can quickly implement intervention operations based on these recommendations to promptly curb the escalation of risks. This step achieves timely risk warning and precise response, significantly shortening emergency response time, effectively reducing the probability of accidents, and providing strong protection for the safe and stable operation of the high-temperature gas-cooled reactor irradiation circuit.

[0048] Figure 3 A flowchart illustrating another high-temperature gas-cooled reactor safety analysis method provided in this application embodiment includes the following steps: Step 301: Construct a neural network model, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive operating parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model.

[0049] For a description of step 301, please refer to the description of step 101 in the above embodiment. This embodiment will not repeat the description in detail.

[0050] Step 302: Standardize the input data to eliminate dimensional differences, and use a sliding window to divide the time series data, wherein the window length and step size are dynamically adjusted according to the core power change cycle.

[0051] In some embodiments, the input data encompasses various operating parameters such as coolant flow rate, core power, temperature, and pressure. These parameters have significantly different units of measurement and numerical ranges, i.e., different dimensions. If directly input into the model, parameters with larger numerical scales may excessively dominate the model training process, while the influence of smaller but crucial parameters may be weakened, thus interfering with the model's learning of the correlation between multiple parameters. Therefore, standardizing the input data can uniformly transform parameters with different dimensions to the same data scale, eliminating the interference caused by differences in dimensions, allowing each parameter to play a balanced role in model training, and ensuring that the model learns the true correlation between parameters rather than differences in numerical scale. Furthermore, the operating parameters of the irradiation loop are time-series data, and their variation patterns are closely related to the time dimension. Using a sliding window to divide the time-series data can segment the continuous time series into multiple subsequences containing information within specific time ranges. Each subsequence fully preserves the dynamic change characteristics of the parameters within that time period, helping the model capture the evolution patterns and interdependencies of the parameters over time. Crucially, the window length and step size are not fixed but dynamically adjusted according to the core power variation cycle. When the core power is in a stable variation cycle, a longer window length and appropriate step size can be set to comprehensively capture the parameter correlation features under stable operating conditions. When the core power fluctuates frequently and the variation cycle shortens, the window length and step size are reduced accordingly to ensure accurate capture of key parameter details that change rapidly in the short term, avoiding the omission of dynamic features or the introduction of excessive redundant information. This significantly improves the consistency and effectiveness of the input data, enabling the model to extract operating condition features more accurately, reducing irrelevant interference, and providing strong data support for the efficiency of subsequent model training and the accuracy of prediction results.

[0052] Step 303: The neural network model is pre-trained and fine-tuned based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal working conditions and accident working conditions. The fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target working condition.

[0053] Step 304: Generate virtual training samples for extreme accident conditions and input the virtual samples into the neural network model to enhance its generalization ability to extreme conditions.

[0054] Step 305: Input the real-time monitored irradiation circuit operation parameters into the trained neural network model, and receive the target channel temperature prediction value and safety margin assessment result output by the neural network model to provide real-time data support for emergency decision-making.

[0055] For a description of steps 303-305, please refer to the description of steps 102-104 in the above embodiment. This embodiment will not repeat these steps in detail.

[0056] Figure 4 This is a schematic diagram of the structure of a high-temperature gas-cooled reactor safety analysis device provided in an embodiment of this application, as shown below. Figure 4 As shown, it includes: a construction module 401, an optimization module 402, a generation module 403, and a receiving module 404.

[0057] The construction module 401 is configured to construct a neural network model, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive running parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model. The optimization module 402 is configured to pre-train and fine-tune the neural network model based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal working conditions and accident working conditions, and the fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target working condition. The generation module 403 is configured to generate virtual training samples for extreme accident conditions and input the virtual samples into the neural network model to enhance its generalization ability to extreme conditions. The receiving module 404 is configured to input the real-time monitored irradiation circuit operating parameters into the trained neural network model, and receive the target channel temperature prediction value and safety margin assessment result output by the neural network model to provide real-time data support for emergency decision-making.

[0058] In some examples of this embodiment, the construction module 401 is specifically configured such that the input layer has multiple neurons to receive input data such as coolant flow rate, core power level, target channel geometry parameters, ambient temperature and pressure, target identification, and accident condition identification; three hidden layers are configured with 64, 32, and 16 nodes respectively, and a preset activation function is used to enhance the nonlinear expressive ability of the neural network model; the output layer is configured with a single neuron to map the features learned by the hidden layers into specific engineering prediction values.

[0059] In some examples of this embodiment, the optimization module 402 is specifically configured to optimize the gap between the predicted value and the true value of the neural network model through a loss function, and dynamically adjust the learning rate through an adaptive moment estimation optimizer; to impose L2 regularization constraints on the output layer parameters through a regularization coefficient, and to prevent the neural network model from overfitting through an early stopping method.

[0060] In some examples of this embodiment, the generation module 403 is specifically configured to perturb the historical operating parameters under normal operating conditions to simulate the equipment state under the extreme accident conditions; merge the virtual training samples generated after the perturbation processing with the original training data to form an enhanced training dataset; input the enhanced training dataset into the neural network model, and freeze some hidden layers of the neural network model during training, updating the parameters only for the unfrozen layers to adapt them to the characteristics of the extreme accident conditions.

[0061] In some examples of this embodiment, the receiving module 404 is specifically configured to input the real-time collected coolant flow rate, core power, target channel geometric parameters, system pressure and temperature, and accident condition identifier into the trained neural network model and obtain the real-time temperature prediction value or temperature change rate of the target channel output by the neural network model; calculate the safety margin under the current operating condition by comparing the temperature prediction value or the temperature change rate with a preset safety threshold; when the safety margin is lower than the set threshold, trigger a temperature anomaly warning and generate corresponding emergency response suggestions, which are used to guide operators to perform intervention operations.

[0062] It should be noted that other corresponding descriptions of the functional units involved in the high-temperature gas-cooled reactor safety analysis device provided in this embodiment can be found in [reference]. Figure 1 , Figure 2 and Figure 3 The corresponding description in [the document] will not be repeated here.

[0063] Based on the above, Figure 1 , Figure 2 and Figure 3The embodiment illustrates a safety analysis method for a high-temperature gas-cooled reactor. Correspondingly, this embodiment also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method. Figure 1 , Figure 2 and Figure 3 This paper presents a safety analysis method for a high-temperature gas-cooled reactor.

[0064] Based on the above, Figure 1 , Figure 2 and Figure 3 The embodiment illustrates a safety analysis method for a high-temperature gas-cooled reactor. Correspondingly, this embodiment also provides a computer program product storing a computer program that, when executed by a processor, implements the aforementioned method. Figure 1 , Figure 2 and Figure 3 This paper presents a safety analysis method for a high-temperature gas-cooled reactor.

[0065] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0066] Based on the above, Figure 1 , Figure 2 and Figure 3 A safety analysis method for a high-temperature gas-cooled reactor is shown, and Figure 4 To achieve the above objectives, the present application also provides an electronic device, such as a personal computer or a server, in the illustrated virtual device embodiment. This device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to implement the above-described virtual device. Figure 1 , Figure 2 and Figure 3 This paper presents a safety analysis method for a high-temperature gas-cooled reactor.

[0067] In some embodiments, the aforementioned physical device may further include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, an input unit such as a keyboard, etc., and optionally, a USB interface, a card reader interface, etc. In some embodiments, the network interface may include a standard wired interface, a wireless interface (such as a Wi-Fi interface), etc.

[0068] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0069] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0070] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for safety analysis of a high-temperature gas-cooled reactor, characterized in that, include: A neural network model is constructed, comprising an input layer, a hidden layer, and an output layer. The input layer is used to receive operating parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model. The neural network model is pre-trained and fine-tuned based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal and accident conditions, and the fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target condition. Virtual training samples are generated for extreme accident conditions, and the virtual samples are input into the neural network model to enhance its generalization ability to extreme conditions; The real-time monitored irradiation circuit operating parameters are input into the trained neural network model, and the target channel temperature prediction value and safety margin assessment result output by the neural network model are received to provide real-time data support for emergency decision-making.

2. The method for safety analysis of high-temperature gas-cooled reactors according to claim 1, characterized in that, The construction of the neural network model, which includes an input layer, hidden layers, and an output layer, includes: The input layer is configured with multiple neurons to receive input data such as coolant flow rate, core power level, target channel geometry parameters, ambient temperature and pressure, target identification, and accident condition identification. The neural network model is configured with three hidden layers, with 64, 32, and 16 nodes respectively, and a preset activation function is used to enhance its nonlinear expressive power. The output layer is configured with a single neuron, which maps the features learned by the hidden layer to specific engineering prediction values.

3. The method for safety analysis of high-temperature gas-cooled reactors according to claim 1, characterized in that, The pre-training and fine-tuning of the neural network model based on physical simulation data and experimental data includes: The difference between the predicted and actual values ​​of the neural network model is optimized by a loss function, and the learning rate is dynamically adjusted by an adaptive moment estimation optimizer. The output layer parameters are subject to L2 regularization constraints using regularization coefficients, and the neural network model is prevented from overfitting by using early stopping.

4. The method for safety analysis of high-temperature gas-cooled reactors according to claim 1, characterized in that, The process of generating virtual training samples for extreme accident conditions and inputting these virtual samples into the neural network model to enhance its generalization ability to extreme conditions includes: The historical operating parameters under normal operating conditions are perturbed to simulate the equipment state under the extreme accident conditions. The virtual training samples generated after the perturbation process are merged with the original training data to form an enhanced training dataset. The enhanced training dataset is input into the neural network model, and during training, some hidden layers of the neural network model are frozen, while only the parameters of the unfrozen layers are updated to adapt them to the characteristics of the extreme accident conditions.

5. The method for safety analysis of high-temperature gas-cooled reactors according to claim 1, characterized in that, The process of inputting real-time monitored irradiation circuit operating parameters into a trained neural network model, and receiving the target channel temperature prediction and safety margin assessment results output by the neural network model, includes: The real-time collected coolant flow rate, core power, target channel geometric parameters, system pressure and temperature, and accident condition indicators are input into the trained neural network model, and the real-time temperature prediction value or temperature change rate of the target channel output by the neural network model is obtained. Based on the comparison between the predicted temperature value or the rate of temperature change and the preset safety threshold, the safety margin under the current operating condition is calculated. When the safety margin is lower than the set threshold, a temperature anomaly warning is triggered and a corresponding emergency response suggestion is generated. The suggestion is used to guide operators to perform intervention operations.

6. The method for safety analysis of high-temperature gas-cooled reactors according to claim 1, characterized in that, Also includes: The input data is standardized to eliminate dimensional differences, and a sliding window is used to divide the time series data, where the window length and step size are dynamically adjusted according to the core power change cycle.

7. A safety analysis device for a high-temperature gas-cooled reactor, characterized in that, include: The building module is configured to build a neural network model, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive running parameters, the hidden layer is used to enhance the nonlinear expressive power of the neural network model, and the output layer is used to output the prediction results of the neural network model. The optimization module is configured to pre-train and fine-tune the neural network model based on physical simulation data and experimental data. The pre-training initializes the parameters of the neural network model using simulation data of normal working conditions and accident working conditions, and the fine-tuning optimizes the output layer parameters of the neural network model using experimental data of the target working condition. The generation module is configured to generate virtual training samples for extreme accident conditions and input the virtual samples into the neural network model to enhance its generalization ability to extreme conditions. The receiving module is configured to input the real-time monitored irradiation circuit operating parameters into the trained neural network model, and receive the target channel temperature prediction value and safety margin assessment result output by the neural network model to provide real-time data support for emergency decision-making.

8. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the high-temperature gas-cooled reactor safety analysis method according to any one of claims 1-6.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the high-temperature gas-cooled reactor safety analysis method according to any one of claims 1-6.

10. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the high-temperature gas-cooled reactor safety analysis method according to any one of claims 1-6.

Images