Temperature field prediction method and device, server and computer readable storage medium
By using a temperature prediction model based on a long short-term memory network and single-step training with core temperature field information and operating parameters, the problem of real-time prediction of the helium temperature field of the coolant in a pebble bed high-temperature gas-cooled reactor was solved, achieving fast and accurate temperature field prediction and meeting the needs of online operating condition optimization.
Patent Information
- Application Number
- CN202611233917.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to predict the spatial distribution of the helium temperature field in a pebble bed high-temperature gas-cooled reactor in real time, making it impossible to meet the needs of online operating condition optimization and rapid analysis of multi-parameter sensitivity. Furthermore, existing shallow intelligent prediction models do not fully utilize the thermal inertia time-series correlation characteristics of the core heat transfer process.
A temperature prediction model based on a long short-term memory network is adopted. By acquiring the core temperature field information and operating parameters of the previous refueling frame for single-step training, a loss function is constructed to characterize the fluid active zone, thereby realizing real-time prediction of the coolant temperature field.
It reduced the computational resource utilization rate, enabled real-time state prediction of the coolant temperature field, met the timeliness requirements of online monitoring, and preserved the engineering integrity of the temperature distribution throughout the space.
Smart Images

Figure CN123116792A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and more specifically, to a temperature field prediction method, apparatus, server, and computer-readable storage medium. Background Technology
[0002] In the continuous refueling operation of a pebble bed high-temperature gas-cooled reactor (HTGR), the fuel spheres and graphite spheres inside the core continuously move downwards and exchange positions. This physical movement leads to continuous changes in the core power distribution, and the in-core heat transfer conditions also dynamically evolve accordingly. Due to the complex convective and conductive heat transfer processes between the helium coolant flowing through the core and these moving spheres, the helium coolant temperature field exhibits significant spatiotemporal differences at different refueling cycles. In other words, the coolant temperature at the same spatial location may differ significantly at different refueling times, and there is also a large temperature gradient between different spatial locations at the same refueling time. Therefore, the spatial distribution of the helium coolant temperature field becomes a core fundamental parameter for determining the core heat transfer safety margin, analyzing the core thermal stress state, and determining the operating parameters of the primary loop hot end. It is also an indispensable key state quantity for conducting core safety analysis, operating condition optimization, and full life-cycle state assessment of HTGRs.
[0003] Currently, in related technologies, solving the spatial distribution of the helium temperature field in a pebble bed high-temperature gas-cooled reactor mainly relies on deterministic numerical simulation calculations performed by thermal-hydraulic calculation programs (such as HUAKUN) suitable for pebble bed high-temperature gas-cooled reactors. These simulation programs are based on first-principles modeling, with fine mesh generation and complete physical models, and their calculation accuracy can meet the actual engineering requirements.
[0004] However, the HUAKUN computation process is complex. A single complete calculation needs to cover the entire core geometry and multiphase flow heat transfer coupling process. Solving a set of coolant helium temperature field calculations that includes multiple refueling cycles is usually time-consuming and consumes a lot of computing resources. It is difficult to support the real-time state prediction requirements of coolant helium temperature field at the reactor site, and thus cannot meet the application scenarios with strong timeliness requirements such as online operation condition optimization and adjustment and rapid analysis of multi-parameter sensitivity. Summary of the Invention
[0005] In view of this, the purpose of this application is to provide a temperature field prediction method, apparatus, server and computer-readable storage medium to reduce resource occupancy and ensure the real-time state prediction of coolant temperature field.
[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, this application provides a method for predicting a temperature field, the method comprising: Obtain the core temperature field information corresponding to the previous refueling frame of the pebble bed high-temperature gas-cooled reactor and the operating parameters corresponding to the current refueling frame; the core temperature field information corresponding to each refueling frame represents the grid distribution information of the coolant temperature of the pebble bed high-temperature gas-cooled reactor after a refueling cycle is completed. The core temperature field information and the operating parameters are input into a pre-trained temperature prediction model for processing to obtain the core temperature field prediction information corresponding to the current refueling frame. The temperature prediction model is obtained by single-step training based on sample core temperature field information, sample operating parameters and mask information corresponding to multiple refueling frames. The mask information represents the fluid active region in the temperature field of the pebble bed high-temperature gas-cooled reactor, and the loss function of the temperature prediction model is constructed based on the mask information.
[0007] In an optional implementation, the temperature prediction model is obtained through the following steps: Obtain the mask information of the pebble bed high-temperature gas-cooled reactor, the sample core temperature field information and sample operating parameters corresponding to multiple feed frames; Training samples are constructed based on the mask information, the sample core temperature field information corresponding to multiple reflow frames, and the sample operation parameters. The training samples are input into a pre-built temperature prediction model for single-step learning training to obtain a trained temperature prediction model.
[0008] In an optional implementation, the sample core temperature field information corresponding to the plurality of reflow frames is obtained through multiple calculations of a pre-built simulation model, each calculation including a plurality of consecutive reflow frames; The step of constructing training samples based on the mask information, the sample core temperature field information corresponding to multiple reflow frames, and the sample operation parameters includes: For each of the above calculation cases, each dumping frame in the calculation case is taken as the target dumping frame. The sample core temperature field information corresponding to the previous dumping frame of the target dumping frame and the sample operating parameters corresponding to the target dumping frame are determined as input data. The sample core temperature field information corresponding to the target dumping frame is determined as the input label. The input data and the input labels are standardized based on the mask information, and the standardized input data and the input labels are determined as training samples corresponding to the target pouring frame.
[0009] In an optional implementation, the temperature prediction model is a long short-term memory network model, and the loss function of the temperature prediction model is:
[0010] Where L is the loss value, For mask information, For input labels, The predicted value output by the temperature prediction model.
[0011] In an optional implementation, after obtaining the trained temperature prediction model, the method further includes: Acquire test cases; each test case includes test core temperature field information and test operation parameters corresponding to multiple reflow frames; The temperature prediction model is tested based on the test case to obtain the capability index corresponding to the temperature prediction model. The performance of the temperature prediction model is determined based on the capability index, and if the model performance does not meet the preset performance conditions, the temperature prediction model is optimized and trained based on the capability index.
[0012] In an optional implementation, testing the temperature prediction model based on the test examples to obtain the capability indicators corresponding to the temperature prediction model includes: Based on the test case, the temperature prediction model is subjected to rolling single-step prediction to obtain the first capability index corresponding to the rolling prediction; the first capability index characterizes the autonomous evolution capability of the temperature prediction model without external correction constraints.
[0013] In an optional implementation, testing the temperature prediction model based on the test examples to obtain the capability indicators corresponding to the temperature prediction model includes: Based on the test case, rolling single-step prediction and teacher-mandated inference are performed on the temperature prediction model respectively to obtain the first capability index corresponding to rolling prediction and the second capability index corresponding to teacher-mandated inference; the second capability index represents the upper limit of the single-step mapping capability of the temperature prediction model.
[0014] Secondly, this application provides a temperature field prediction device, the device comprising: The acquisition module is used to acquire the core temperature field information and operating parameters corresponding to the previous refueling frame of the pebble bed high-temperature gas-cooled reactor; the core temperature field information corresponding to each refueling frame represents the grid distribution information of the coolant temperature of the pebble bed high-temperature gas-cooled reactor after a refueling cycle. The processing module is used to input the core temperature field information and the operating parameters into a pre-trained temperature prediction model for processing, so as to obtain the core temperature field prediction information corresponding to the current unloading frame. The temperature prediction model is obtained by single-step training based on sample core temperature field information, sample operating parameters and mask information corresponding to multiple refueling frames. The mask information represents the fluid active region in the temperature field of the pebble bed high-temperature gas-cooled reactor, and the loss function of the temperature prediction model is constructed based on the mask information.
[0015] Thirdly, this application provides a server including a processor and a memory, the memory storing a computer program executable by the processor, the processor being able to execute the computer program to implement the method described in any of the foregoing embodiments.
[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the foregoing embodiments.
[0017] The temperature field prediction method, apparatus, server, and computer-readable storage medium provided in this application embodiment allow the server to acquire the core temperature field information and operating parameters corresponding to the previous refueling frame of a pebble bed high-temperature gas-cooled reactor. This core temperature field information and operating parameters are then input into a pre-trained temperature prediction model for processing, outputting the core temperature field prediction information corresponding to the current refueling frame. Since the temperature prediction model is obtained through single-step training of a loss function constructed based on sample core temperature field information, sample operating parameters, and mask information characterizing the fluid active zone corresponding to multiple refueling frames, coolant temperature field prediction can be transformed into a single-step mapping task based on the temperature field and operating parameters of the previous frame, outputting the full-stack temperature distribution of the current frame. This eliminates the need for real-time calls to physical simulation programs, effectively reducing computational resource consumption and meeting real-time state prediction requirements.
[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 A block diagram of a server provided in an embodiment of this application is shown; Figure 2 This paper illustrates a flowchart of a temperature field prediction method provided in an embodiment of this application. Figure 3A schematic diagram of a single-step LSTM network is shown. Figure 4 A schematic diagram of the LSTM layer structure is shown; Figure 5 This diagram illustrates the frame-by-frame changes in RMSE. Figure 6 This shows a schematic diagram of the changes in MAPE frame by frame; Figure 7 A scatter plot comparing the average true value and the predicted value of the test current collector area is shown; Figure 8 The spatial heatmap of the average RMSE within the grid is shown; Figure 9 This diagram illustrates the functional block diagram of a temperature field prediction device provided in an embodiment of this application. Detailed Implementation
[0021] 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. The components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations.
[0022] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0023] It should be noted that 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. Without further limitations, 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 said element.
[0024] Currently, the simulation-based temperature field analysis method in related technologies consumes a large amount of computing resources, making it difficult to support the real-time state prediction requirements of the coolant helium temperature field at the reactor site. Consequently, it cannot meet the application scenarios with strong timeliness requirements, such as online operating condition optimization and adjustment and rapid analysis of multi-parameter sensitivity.
[0025] Based on this, numerous intelligent prediction studies have been conducted both domestically and internationally regarding high-temperature gas-cooled reactors. However, most of these studies focus on single-target prediction directions, such as estimating the temperature at local single points within the reactor core. These studies can only predict the temperature values at discrete points, such as estimating the coolant temperature on the surface of a fuel element or at the center of a channel. They cannot fully reconstruct the overall spatial distribution of the coolant temperature field across the entire reactor core.
[0026] Furthermore, existing shallow intelligent prediction models generally employ a static fitting modeling approach, directly using the current operating parameters as input to fit the coolant temperature field data at the corresponding moment. This modeling process completely ignores the inherent thermal inertia and temporal correlation characteristics of the core heat transfer process itself. This modeling method fails to fully utilize the temporal evolution information of the coolant temperature field formed during the preceding refueling cycle, leading to a severe disconnect between the model's deduction logic and engineering reality. Therefore, existing shallow intelligent prediction models have significant shortcomings in terms of engineering practicality and scenario adaptability. They cannot reflect the dynamic evolution of the coolant temperature field over time, nor can they meet the basic requirements of digital twin systems for temporal consistency and physical interpretability.
[0027] Based on this, embodiments of this application provide a temperature field prediction method, apparatus, server, and computer-readable storage medium to solve the above-mentioned problems.
[0028] Figure 1 Please refer to the block diagram of the server provided in the embodiments of this application. Figure 1 The server includes a memory, a processor, and a communication module. These components are electrically connected directly or indirectly to enable data transmission or interaction. For example, they can be electrically connected via one or more communication buses or signal lines.
[0029] The memory is used to store computer programs or data that can be executed by the processor. The memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc.
[0030] The processor is used to read / write data or computer programs stored in the memory and execute the computer program to implement the temperature field prediction method provided in the embodiments of this application.
[0031] The communication module is used to establish communication connections between the server and other communication terminals through the network, and to send and receive data through the network.
[0032] It should be understood that, Figure 1 The structure shown is only a schematic diagram of the server structure; the server may also include components such as... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown. Figure 1 The components shown can be implemented using hardware, software, or a combination thereof.
[0033] The following is based on the above. Figure 1 As the implementing entity, the temperature field prediction method provided in this application embodiment will be described executively with reference to the flowchart. Specifically, Figure 2 For a flowchart illustrating the temperature field prediction method provided in this application embodiment, please refer to [link / reference]. Figure 2 The method includes: Step S20: Obtain the core temperature field information corresponding to the previous refueling frame of the pebble bed high-temperature gas-cooled reactor and the operating parameters corresponding to the current refueling frame.
[0034] The core temperature field information corresponding to each refueling frame represents the grid distribution information of the coolant temperature of the pebble bed high-temperature gas-cooled reactor after a refueling cycle.
[0035] Understandably, pebble bed high-temperature gas-cooled reactors undergo continuous refueling during operation, with fuel spheres and graphite spheres moving slowly from top to bottom and constantly replacing each other, causing the core power distribution and heat transfer state to change continuously over time. This dynamic characteristic means that the temperature distribution of the coolant helium is not static, but evolves frame by frame with the refueling cycle as the basic time unit. In this case, the temperature field information corresponding to each refueling frame is the temperature field state after the end of a refueling cycle, and there is a clear temporal sequence relationship between adjacent frames, that is, the temperature field corresponding to the previous refueling frame constitutes the initial state basis for the prediction of the next frame.
[0036] In this embodiment, the core temperature field information can be a data carrier in the form of a grid. That is, the core temperature field information corresponding to each refueling frame can be the set of temperature values of the coolant at multiple computational grid points in the core at the end of the refueling cycle.
[0037] In one possible implementation, these grids can be arranged in 18 radial layers and 41 axial layers, forming a two-dimensional discrete temperature field with a spatial resolution of 18×41. After being flattened in row-first order, this temperature field information forms a 738-dimensional vector, which fully characterizes the spatial temperature distribution of the coolant throughout the entire core.
[0038] In this embodiment, the operating parameters are the condition parameters under the current refueling frame cycle. In one possible implementation, these operating parameters may include coolant inlet temperature, coolant pressure, coolant flow rate, current cycle duration, core rated power, and control rod insertion position. Furthermore, these operating parameters may also include the current refueling cycle index and the cumulative number of days since the start of operation, where the current refueling cycle index refers to the frame index position of the current frame within a complete core refueling cycle.
[0039] When performing predictions, the server can read the 738-dimensional vector representation of the temperature field in the previous frame and the 8-dimensional operating parameters of the current frame as joint inputs for subsequent single-step inferences.
[0040] In one possible implementation, the server can input the temperature prediction information output by the temperature prediction model for the previous refueling frame as the core temperature field information of the previous frame into the temperature prediction model to predict the temperature field prediction information corresponding to the current refueling frame. At the same time, the server can also obtain the operating parameters corresponding to the current refueling frame through the reactor real-time monitoring system or the operation scheduling platform.
[0041] Step S21: Input the core temperature field information and operating parameters into the pre-trained temperature prediction model for processing to obtain the core temperature field prediction information corresponding to the current refueling frame.
[0042] The temperature prediction model is obtained by single-step training based on sample core temperature field information, sample operating parameters and mask information corresponding to multiple refueling frames. The mask information represents the fluid active region in the temperature field of the pebble bed high-temperature gas-cooled reactor, and the loss function of the temperature prediction model is constructed based on the mask information.
[0043] In this embodiment, the server can input the above-mentioned core temperature field information and current operating parameters into a pre-trained temperature prediction model. The model can output the core temperature field prediction information corresponding to the current refueling frame. This prediction information is also represented by the same grid structure to completely restore the overall spatial distribution of the core temperature field, that is, it is still a set of temperature values on 18×41 grid points. After flattening, it is a 738-dimensional vector, which can be directly mapped back to the three-dimensional core space for thermal stress analysis or digital twinning.
[0044] In one possible implementation, the temperature prediction model can be a single-step mapping model built on a Long Short-Term Memory (LSTM) network. This model can be trained step-by-step based on sample core temperature field information, sample operating parameters, and mask information corresponding to multiple reflow frames. In other words, the model can learn during training how to obtain the temperature prediction data from the k-th reflow frame. The temperature field of frame 1 and the corresponding operating parameters of frame k are used to predict the overall temperature distribution of frame k. This single-step training method allows the network to focus on local temporal transfer patterns, which is more in line with the physical nature of heat conduction and slow accumulation in reactor core heat transfer.
[0045] In this process, the mask information can be used to characterize the fluid active region in the temperature field of the pebble bed high-temperature gas-cooled reactor. In this embodiment, the loss function is constructed based on the mask information, and the mean square error is calculated only at the corresponding position of the fluid active region, so as to avoid the model deviating from the optimization direction of the real heat transfer region due to fitting a large number of non-active points that are always zero.
[0046] In one possible implementation, the server can predict the temperature field for each frame of a complete core refueling cycle using the method described above. In this case, the server first loads the initial core temperature field information obtained from actual measurements or simulations at the initial moment—that is, frame 0—and simultaneously reads the operating parameters corresponding to the first frame and inputs them into the temperature prediction model. Thereafter, the model no longer relies on the actual temperature fields of subsequent frames, but instead enters an open-loop rolling prediction process. That is, the prediction results of the previous frame and the operating parameters of the current frame are used as input data to the model to predict the temperature field information of the current frame, and so on, until the core temperature field prediction information for each frame is obtained.
[0047] For example, if a complete core refueling cycle consists of 13 frames, where frame 0 is the initial state and frames 1-12 are the 12 refueling cycles executed sequentially, the server can generate the temperature fields of all 12 frames within seconds based on the initial core temperature field information and the operating parameters corresponding to the first frame. The output is a two-dimensional temperature matrix of 18 radial × 41 axial dimensions, which can be directly mapped to the core physical coordinates for calculating fuel element surface temperatures, identifying local hotspot areas, verifying reflector thermal stress, or connecting to upper-level safety analysis modules to conduct transient accident simulations. In this way, the HUAKUN full-reactor thermal simulation, which originally took one to two hours, is compressed to a second-level response, meeting the timeliness requirements of online monitoring while preserving the engineering integrity of the entire space temperature distribution.
[0048] The temperature field prediction method provided in this application allows the server to obtain the core temperature field information and operating parameters corresponding to the previous refueling frame of a pebble bed high-temperature gas-cooled reactor. This core temperature field information and operating parameters are then input into a pre-trained temperature prediction model for processing, outputting the core temperature field prediction information for the current refueling frame. Since the temperature prediction model is obtained through single-step training of a loss function constructed based on sample core temperature field information, sample operating parameters, and mask information characterizing the fluid active zone corresponding to multiple refueling frames, coolant temperature field prediction can be transformed into a single-step mapping task based on the temperature field and operating parameters of the previous frame, outputting the full-reactor temperature distribution for the current frame. This eliminates the need for real-time calls to physical simulation programs, effectively reducing computational resource consumption and meeting real-time state prediction requirements.
[0049] The following section provides a possible approach to training a temperature prediction model.
[0050] Specifically, the server can obtain the mask information of the pebble bed high-temperature gas-cooled reactor, the sample core temperature field information and sample operating parameters corresponding to multiple refueling frames; construct training samples based on the mask information, the sample core temperature field information corresponding to multiple refueling frames and the sample operating parameters; input the training samples into the pre-constructed temperature prediction model for single-step learning training to obtain the trained temperature prediction model.
[0051] In this embodiment, the server can determine whether the temperature value of each 1×18×41 grid location has ever been non-zero, based on all refueling frames. Grid locations with non-zero values are then included in the fluid active zone index set Ω, thereby obtaining the mask information of the pebble bed high-temperature gas-cooled reactor. This mask information is essentially the fluid active zone index set Ω, which characterizes the spatial region within the reactor core where helium flow and heat exchange actually occur. It excludes grid points without fluid participation, such as solid structures in the reflector layer and boundary channels, thus ensuring that subsequent model training and evaluation always focus on physically meaningful temperature evolution regions.
[0052] In practical applications, the training samples include a 738-dimensional temperature vector and 8-dimensional operating parameters, totaling 746 dimensions. The temperature prediction model can be an LSTM, and its structural design is highly compatible with the thermal evolution characteristics of the reactor core.
[0053] In one possible implementation method Figure 3 This is a schematic diagram of a single-step LSTM network. Figure 4 For a schematic diagram of the LSTM layer structure, please refer to [link / reference]. Figure 3 and Figure 4 The model's input layer receives a 746-dimensional vector, which is then reshaped by a reshape layer into a (1,746) form, treated as a single time step sequence. This is then fed into an LSTM layer containing 256 hidden units. This layer uses forgetting gates, input gates, and output gates to collaboratively regulate cell states, achieving dynamic capture of the inertia of temperature changes between feeding cycles. This represents the hidden state output by the LSTM at frame t-1. A Dropout layer (with a dropout rate of 0.10) is then applied to suppress overfitting, followed by a fully connected layer with 512 ReLU activation units to enhance nonlinear expressive power. Finally, a 738-unit linear output layer reconstructs the full-field temperature distribution for the next frame. The entire network has approximately one million parameters, and its output structure corresponds one-to-one with the original HUAKUN output grid (1×18×41). It can efficiently complete single-step mapping training of 12,000 examples on a single GPU.
[0054] In this embodiment, once the model training is complete, the following can be obtained: LSTM weight file (Keras format), parameters of the normalization tool (mean and standard deviation), test set prediction array (test_predictions.npz), frame-by-frame metric table (metrics_by_frame.csv), and rolling prediction summary (metrics_rollout_summary.json). Offline reproduction and plotting of results are supported.
[0055] In one possible implementation, the sample core temperature field information corresponding to the plurality of reflow frames is obtained through multiple calculations of a pre-built simulation model, each calculation corresponding to a plurality of consecutive reflow frames.
[0056] In this embodiment, the server can sequentially, for each of the computational examples, take each dumping frame in the computational example as the target dumping frame, determine the sample core temperature field information corresponding to the previous dumping frame of the target dumping frame and the sample operating parameters corresponding to the target dumping frame as input data, and determine the sample core temperature field information corresponding to the target dumping frame as the input label; perform standardization processing on the input data and the input label according to the mask information, and determine the standardized input data and the input label as the training sample corresponding to the target dumping frame.
[0057] In this embodiment, each simulation case corresponds to a complete core refueling cycle simulation, including multiple consecutive refueling frames (e.g., 13), where the 0th frame... The initial core temperature field information is shown in frames 1 through 12, which represent the coolant temperature field after 12 refueling cycles. to Each frame is output as an 18 (radial) × 41 (axial) grid, which is flattened in row priority to form a 738-dimensional vector.
[0058] When constructing training samples, the server can process each frame in each example. (k ranges from 1 to 12), the previous frame is... The corresponding temperature field vector As part of the input data, and at the same time, this frame The corresponding 8-dimensional operating parameters Together, they form a single-step input data with a dimension of 746. ,Right now And that frame Its own temperature field vector These labels serve as input for supervision. Thus, one example can generate 12 independent single-step training samples, with the entire dataset totaling approximately 1.4 × 10⁻⁶. 5 Sample.
[0059] Simultaneously, all input data and labels undergo Z-score normalization within the fluid active region Ω. This normalization process only applies to the grid index positions covered by Ω, excluding invalid grid points where the values are always zero. During training, the model receives the 746-dimensional input normalized within this masked region, while the supervision signal consists of the 738-dimensional temperature field labels normalized within the same masked region. The loss function is calculated as the mean square error over Ω. In this way, the model learning process consistently focuses on the actual fluid space involved in heat transfer, ensuring both the rationality of the physical meaning and improving the numerical stability of high-dimensional field predictions.
[0060] In this embodiment, the temperature prediction model is a long short-term memory network model, and the loss function of the temperature prediction model is:
[0061] Where L is the loss value, For mask information, For input labels, The predicted value output by the temperature prediction model.
[0062] Optionally, after obtaining the trained temperature prediction model, performance testing of the model is required before formal application. Specifically, the server can obtain test cases; test the temperature prediction model based on the test cases to obtain the capability index corresponding to the temperature prediction model; determine the model performance based on the capability index; and optimize the temperature prediction model based on the capability index if the model performance does not meet the preset performance conditions.
[0063] Each of the test cases includes test core temperature field information and test operation parameters corresponding to multiple reflow frames.
[0064] The capability indicators include the root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination for each feeding frame.
[0065] In practical applications, the server can obtain multiple simulation cases and divide them into training set, validation set and test set. The cases in the training set can be used to build training samples and participate in model training. The cases in the validation set can be used to verify whether the current training state has reached the optimal balance point. The cases in the test set are used to test the model performance.
[0066] In one possible implementation, the simulation examples obtained by the server can be organized in the form of example folders. The training set, validation set and test set are randomly divided at the example index level, with a ratio of 7:1:2, and a fixed random seed (e.g. 42) is used to ensure that all 13 frames of temperature field and parameters of the same example always belong to the same subset, avoiding information leakage caused by cross-subset splitting and ensuring the authenticity of the model generalization ability evaluation.
[0067] In this embodiment, the capability indicators may include root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²). These indicators reflect both the overall prediction accuracy of the model and the evolution trend of error as the material reloading cycle progresses. For example, the change in RMSE from the first frame to the last frame can reflect the physical nature of the gradual accumulation of error under open-loop rolling prediction.
[0068] In this case, if any of the above-mentioned capability indicators fails to reach its corresponding preset threshold, it can be determined that the model performance does not meet the engineering application conditions. At this time, based on the trend of the capability indicators gradually evolving with each frame, the weak links of the model in a specific material pouring stage or a specific area can be located. The training sample composition can be adjusted in a targeted manner, the corresponding working condition coverage can be enhanced, or the network structure parameters can be optimized and single-step training can be carried out again until all capability indicators are stably converged to the expected range.
[0069] If all performance indicators reach the corresponding preset thresholds, a high-performance temperature prediction model can be obtained and deployed for subsequent application.
[0070] The following provides a possible implementation method for testing the temperature prediction model based on the test cases and obtaining the corresponding capability indicators of the temperature prediction model.
[0071] In one possible implementation, the server can perform only rolling single-step prediction, that is, the server can perform rolling single-step prediction on the temperature prediction model based on the test case to obtain a first capability index corresponding to the rolling prediction; the first capability index characterizes the autonomous evolution capability of the temperature prediction model without external correction constraints.
[0072] In this embodiment, rolling single-step prediction refers to the process of generating the coolant temperature field prediction results after the end of the material unloading cycle in each subsequent frame by repeatedly calling the trained single-step prediction model, relying only on the actual temperature field information at the initial moment and the corresponding operating parameters of each frame.
[0073] During this process, each time the model performs forward inference, its input consists of the predicted output of the previous frame and the operating parameters of the current unloading cycle, without introducing any subsequent real temperature field data; for example, when predicting the first frame, the input is the initial true value FT. 00 The output is F, which corresponds to the running parameters of the first frame loop. 01 When predicting the second frame, the prediction result F from the first frame is used. 01 As input data, combined with the running parameters corresponding to the second frame loop, F is obtained. 02 And so on, until the completion of the process from FT 00 To F 12 12 consecutive iterations.
[0074] This open-loop autoregressive approach realistically simulates the actual working conditions of digital twin systems or online monitoring scenarios where there is a lack of real-time correction signals, thus objectively reflecting the model's autoevolution capability under conditions without external intervention.
[0075] The first capability index obtained from this test refers to several quantitative error statistics calculated for all test cases under the rolling single-step prediction mode, including the root mean square error RMSE, mean absolute error MAE, mean absolute percentage error MAPE, and coefficient of determination R², which are statistically analyzed frame by frame in the fluid active region Ω, and can be further summarized into the average value of all test cases.
[0076] Optionally, the server can also perform further analysis of the model's performance based on these capability metrics.
[0077] In one example Figure 5 For a frame-by-frame illustration of RMSE changes, please refer to [link / reference]. Figure 5 The diagram shows the RMSE changing frame by frame, with the rewinding frame number (0 to 12) on the horizontal axis and the RMSE value (in degrees Celsius) on the vertical axis. It can intuitively show the trend of the error gradually increasing as the rewinding cycle progresses, reflecting the inherent characteristics of error accumulation in the model during long-term autonomous simulation. Figure 6 For a schematic diagram of MAPE frame-by-frame changes, please refer to [link / reference]. Figure 6 This corresponds to showing the MAPE variation pattern under the same set of prediction sequences, which helps to illustrate the distribution characteristics of the relative error.
[0078] Figure 7 For a scatter plot comparing the average true value and the predicted value of the collector area, please refer to [link / reference]. Figure 7 The predicted temperature of all fluid grid points is plotted one-to-one with the actual temperature, and a 1:1 reference line is superimposed to determine whether there is a systematic shift in the overall fit. Figure 8 For a spatial heatmap of the average RMSE within the grid, please refer to [link / reference]. Figure 8 Furthermore, the RMSE is averaged grid by grid in an 18×41 space and rendered as a heat map (the upper part of the diagram shows the thermal distribution within all grids, and the lower part shows the thermal distribution within the fluid active zone), which allows for the visualization and identification of local prediction reliability differences in different physical regions such as the main channel of the ball bed, the transition zone of the graphite reflective layer, and the boundary channel.
[0079] In another possible implementation, the server can also superimpose teacher-mandated inference on top of rolling single-step prediction. That is, based on the test case, the server can perform rolling single-step prediction and teacher-mandated inference on the temperature prediction model respectively to obtain a first capability index corresponding to rolling prediction and a second capability index corresponding to teacher-mandated inference; the second capability index represents the upper limit of the single-step mapping capability of the temperature prediction model.
[0080] In this embodiment, during teacher-mandated inference, the server can use the real temperature field from the previous moment as input at each inference step. At this time, the model is always in an "ideal supervised" state, and each prediction is based on the true value, thus reflecting the theoretical optimal performance boundary of the model on the single-step mapping task.
[0081] For example, when predicting the first frame, the input is the initial true Fourier transform (FT). 00 The output is F, which corresponds to the running parameters of the first frame loop. 01 When predicting the second frame, the true temperature field value of the first frame is used (FT). 01 As input data, combined with the running parameters corresponding to the second frame loop, F is obtained. 02 And so on, until the completion of the process from FT 00 To F 12 12 consecutive iterations.
[0082] In this embodiment, the second capability index may also include frame-by-frame predicted temperature field arrays, root mean square error calculated under fluid active region mask constraints, mean absolute error, mean absolute percentage error, and coefficient of determination, among other numerical indices. This index itself does not reflect the time accumulation effect, but rather focuses on the upper limit of the model's fitting accuracy to the temperature change relationship between adjacent frames.
[0083] In one possible implementation, this reasoning approach can provide a diagnostic benchmark. By comparing the error levels of corresponding frames in the second capability metric with those in the first capability metric, the server can identify the main sources of model error growth. For example, if the error in a frame remains significantly high under teacher-mandated conditions, it indicates an inherent defect in the model's single-step mapping capability; if the error is minimal under teacher-mandated conditions but rises rapidly in rolling prediction, it suggests that the error mainly stems from multi-step accumulation and feedback bias.
[0084] Thus, the synergistic analysis of the two types of indicators provides quantifiable technical criteria for determining whether a model is engineering-ready.
[0085] To perform the corresponding steps in the above embodiments and various possible methods, an implementation of a temperature field prediction device is given below. Optionally, the temperature field prediction device can adopt the above-described... Figure 1 The server's device structure is shown. For further details, please refer to... Figure 9 , Figure 9This is a functional block diagram of a temperature field prediction device provided in an embodiment of this application. It should be noted that the basic principle and technical effects of the temperature field prediction device provided in this embodiment are the same as those in the above embodiments. For the sake of brevity, any parts not mentioned in this embodiment can be referred to the corresponding content in the above embodiments. The temperature field prediction device includes: an acquisition module and a processing module.
[0086] The acquisition module is used to acquire the core temperature field information and operating parameters corresponding to the previous refueling frame of the pebble bed high-temperature gas-cooled reactor; the core temperature field information corresponding to each refueling frame represents the grid distribution information of the coolant temperature of the pebble bed high-temperature gas-cooled reactor after a refueling cycle. The processing module is used to input the core temperature field information and the operating parameters into a pre-trained temperature prediction model for processing, so as to obtain the core temperature field prediction information corresponding to the current refueling frame. The temperature prediction model is obtained by single-step training based on sample core temperature field information, sample operating parameters and mask information corresponding to multiple refueling frames. The mask information represents the fluid active region in the temperature field of the pebble bed high-temperature gas-cooled reactor, and the loss function of the temperature prediction model is constructed based on the mask information.
[0087] Optionally, the temperature field prediction device further includes a training module, which is used to acquire the mask information of the pebble bed high-temperature gas-cooled reactor, the sample core temperature field information and sample operating parameters corresponding to multiple refueling frames; construct training samples based on the mask information, the sample core temperature field information corresponding to multiple refueling frames and the sample operating parameters; input the training samples into the pre-constructed temperature prediction model for single-step learning training to obtain the trained temperature prediction model.
[0088] Optionally, the training module is further configured to, for each of the computational examples, take each dumping frame in the computational example as a target dumping frame, determine the sample core temperature field information corresponding to the previous dumping frame of the target dumping frame and the sample operating parameters corresponding to the target dumping frame as input data, and determine the sample core temperature field information corresponding to the target dumping frame as input labels; perform standardization processing on the input data and the input labels according to the mask information, and determine the standardized input data and the input labels as training samples corresponding to the target dumping frame.
[0089] Optionally, the training module is further used to acquire test cases; each test case includes test core temperature field information and test operation parameters corresponding to multiple refueling frames; the temperature prediction model is tested based on the test cases to obtain the capability index corresponding to the temperature prediction model; the model performance of the temperature prediction model is determined based on the capability index, and if the model performance does not meet the preset performance conditions, the temperature prediction model is optimized and trained based on the capability index.
[0090] Optionally, the training module is further configured to perform rolling single-step prediction on the temperature prediction model based on the test case to obtain a first capability index corresponding to the rolling prediction; the first capability index characterizes the autonomous evolution capability of the temperature prediction model without external correction constraints.
[0091] Optionally, the training module is further configured to perform rolling single-step prediction and teacher-mandated inference on the temperature prediction model based on the test examples, respectively, to obtain a first capability index corresponding to rolling prediction and a second capability index corresponding to teacher-mandated inference; the second capability index characterizes the upper limit of the single-step mapping capability of the temperature prediction model.
[0092] Optionally, the above modules can be stored in the form of software or firmware. Figure 1 The memory shown is either stored in or embedded in the server's operating system (OS), and can be used by... Figure 1 The processor executes the commands. Meanwhile, the data and program code required to execute these modules can be stored in memory.
[0093] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, can implement the temperature field prediction method provided in this application.
[0094] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0095] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0096] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0097] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A temperature field prediction method characterized by, The method includes: Obtain the core temperature field information corresponding to the previous refueling frame of the pebble bed high-temperature gas-cooled reactor and the operating parameters corresponding to the current refueling frame; the core temperature field information corresponding to each refueling frame represents the grid distribution information of the coolant temperature of the pebble bed high-temperature gas-cooled reactor after a refueling cycle is completed. The core temperature field information and the operating parameters are input into a pre-trained temperature prediction model for processing to obtain the core temperature field prediction information corresponding to the current refueling frame. The temperature prediction model is obtained by single-step training based on sample core temperature field information, sample operating parameters and mask information corresponding to multiple refueling frames. The mask information represents the fluid active region in the temperature field of the pebble bed high-temperature gas-cooled reactor, and the loss function of the temperature prediction model is constructed based on the mask information.
2. The method of claim 1, wherein, The temperature prediction model is obtained through the following steps: Obtain the mask information of the pebble bed high-temperature gas-cooled reactor, the sample core temperature field information and sample operating parameters corresponding to multiple feed frames; Training samples are constructed based on the mask information, the sample core temperature field information corresponding to multiple reflow frames, and the sample operation parameters. The training samples are input into a pre-built temperature prediction model for single-step learning training to obtain a trained temperature prediction model.
3. The method of claim 2, wherein, The sample core temperature field information corresponding to the multiple reloading frames is obtained through multiple calculation cases of a pre-built simulation model, and each calculation case includes multiple consecutive reloading frames; The step of constructing training samples based on the mask information, the sample core temperature field information corresponding to multiple reflow frames, and the sample operation parameters includes: For each of the above calculation cases, each dumping frame in the calculation case is taken as the target dumping frame. The sample core temperature field information corresponding to the previous dumping frame of the target dumping frame and the sample operating parameters corresponding to the target dumping frame are determined as input data. The sample core temperature field information corresponding to the target dumping frame is determined as the input label. The input data and the input labels are standardized based on the mask information, and the standardized input data and the input labels are determined as training samples corresponding to the target pouring frame.
4. The method of claim 3, wherein, The temperature prediction model is a long short-term memory network model, and the loss function of the temperature prediction model is: wherein L is a loss value, is mask information, is an input label, is a predicted value output by the temperature prediction model.
5. The method of claim 2, wherein, After obtaining the trained temperature prediction model, the method further includes: Acquire test cases; each test case includes test core temperature field information and test operation parameters corresponding to multiple reflow frames; The temperature prediction model is tested based on the test case to obtain the capability index corresponding to the temperature prediction model. The performance of the temperature prediction model is determined based on the capability index, and if the model performance does not meet the preset performance conditions, the temperature prediction model is optimized and trained based on the capability index.
6. The method of claim 5, wherein, The step of testing the temperature prediction model based on the test examples to obtain the capability indicators corresponding to the temperature prediction model includes: Based on the test case, the temperature prediction model is subjected to rolling single-step prediction to obtain the first capability index corresponding to the rolling prediction; the first capability index characterizes the autonomous evolution capability of the temperature prediction model without external correction constraints.
7. The method of claim 5, wherein, The step of testing the temperature prediction model based on the test examples to obtain the capability indicators corresponding to the temperature prediction model includes: Based on the test case, rolling single-step prediction and teacher-mandated inference are performed on the temperature prediction model respectively to obtain the first capability index corresponding to rolling prediction and the second capability index corresponding to teacher-mandated inference; the second capability index represents the upper limit of the single-step mapping capability of the temperature prediction model.
8. A temperature field prediction device, characterized by comprising: The device includes: The acquisition module is used to acquire the core temperature field information and operating parameters corresponding to the previous refueling frame of the pebble bed high-temperature gas-cooled reactor; the core temperature field information corresponding to each refueling frame represents the grid distribution information of the coolant temperature of the pebble bed high-temperature gas-cooled reactor after a refueling cycle. The processing module is used to input the core temperature field information and the operating parameters into a pre-trained temperature prediction model for processing, so as to obtain the core temperature field prediction information corresponding to the current unloading frame. The temperature prediction model is obtained by single-step training based on sample core temperature field information, sample operating parameters and mask information corresponding to multiple refueling frames. The mask information represents the fluid active region in the temperature field of the pebble bed high-temperature gas-cooled reactor, and the loss function of the temperature prediction model is constructed based on the mask information.
9. A server, characterized by It includes a processor and a memory, the memory storing a computer program executable by the processor, the processor being able to execute the computer program to implement the method of any one of claims 1-7.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.
