Method for predicting burnup distribution of fuel spheres in dynamic pebble bed high temperature gas cooled reactor

Abstract

The application discloses a kind of dynamic decomposition pebble bed type high temperature gas cooled reactor fuel ball burnup distribution prediction methods, it is related to nuclear engineering and artificial intelligence cross technical field, wherein method includes: obtaining initial burnup distribution and unloading instruction of core, normalizing and constructing dynamic mask matrix to distinguish fuel and non-fuel ball layer for burnup data;Multiple cycle burnup data is constructed time series snapshot matrix, feature is extracted by modal decomposition and dynamic operator is calculated;Based on dynamic operator, subsequent period burnup distribution is iteratively predicted by autoregression, is denormalized after covering target period, and then the value of non-fuel layer is zeroed by mask matrix, and the result meeting physical constraint is output.The present application can realize multi-cycle prediction of second level, greatly shorten the calculation time, meet the real-time monitoring demand;Through physical constraint and adjustable mode number, the prediction accuracy and calculation efficiency are considered, to ensure that the result is consistent with the core physical law.

Description

Technical Field

[0001] This invention relates to the field of nuclear engineering and artificial intelligence, and in particular to a method and apparatus for predicting the burnup distribution of fuel spheres in a dynamic decomposition pebble bed type high-temperature gas-cooled reactor. Background Technology

[0002] The pebble bed high-temperature gas-cooled reactor employs a continuous refueling design, where tens of thousands of fuel spheres within the core slowly descend under gravity, while new fuel spheres are continuously loaded at the top and high-burnup spheres are periodically discharged from the bottom. This dynamic refueling mechanism results in a highly non-uniform burnup distribution of the fuel spheres in the core, which evolves complexly over time. This is a core parameter for analyzing core power distribution, temperature field, and safety margin.

[0003] Traditional fuel sphere burnup distribution calculations mainly rely on Monte Carlo methods or deterministic methods, achieving high-precision calculations based on refined neutron physics modeling. However, they suffer from the bottleneck of long calculation time. A single full-core three-dimensional burnup distribution calculation often takes 1 to 2 hours. If multi-cycle coupled evolution calculations are performed to simulate the continuous refueling process, the total time will increase significantly, failing to meet emerging demands such as real-time reactor operation monitoring, online optimization, and digital twins. Summary of the Invention

[0004] The main objective of this invention is to provide a method for predicting the burnup distribution of fuel spheres in a dynamic decomposition pebble bed type high-temperature gas-cooled reactor.

[0005] Another objective of this invention is to propose a dynamic decomposition pebble bed type high-temperature gas-cooled reactor fuel sphere burnup distribution prediction device.

[0006] The third objective of this invention is to provide an electronic device.

[0007] A fourth objective of this invention is to provide a non-transitory computer-readable storage medium.

[0008] To achieve the above objectives, a first aspect of the present invention proposes a method for predicting the burnup distribution of fuel spheres in a dynamically decomposed pebble bed type high-temperature gas-cooled reactor, comprising:

[0009] S1: Obtain the fuel consumption distribution data of the core in the initial cycle and the current refueling status command, normalize the fuel consumption distribution data, construct a dynamic masking matrix based on the fuel consumption threshold, and construct a ball type status update mechanism based on the core refueling rules. S2 expands the normalized fuel consumption distribution data of multiple consecutive cycles along the time dimension to construct a time-series snapshot matrix, performs modal decomposition on the time-series snapshot matrix, extracts modal features, and calculates dynamic operators that characterize the evolution law of fuel consumption. S3, based on the dynamic operator, performs autoregressive iterative calculation on the fuel consumption distribution characteristics of the current cycle to obtain the predicted fuel consumption distribution data for the next cycle, and uses the predicted fuel consumption distribution data as the new current cycle data, repeating the process until the target prediction cycle is covered; S4 performs inverse normalization on the final predicted burnup distribution data and uses a dynamic masking matrix to force the predicted values ​​corresponding to non-fuel spheres to zero, outputting the burnup distribution prediction results that conform to the core physical constraints.

[0010] Optionally, acquire the core burnup distribution data during the initial cycle and the current refueling status command, and normalize the burnup distribution data, including: Read the initial cycle burnup distribution data conforming to the 20×6 core grid format, where the height corresponds to the core Layer 28 to Layer 9, and the width corresponds to the 6 radial regions. Simultaneously acquire the core global operation parameters including power, cumulative operating days, coolant outlet temperature, and control rod position. The initial cyclic fuel consumption distribution data is normalized using the global minimum and global maximum values ​​to generate normalized fuel consumption distribution data that eliminates the influence of dimensions.

[0011] in, For the 20×6 grid, the first Line 1 The original fuel consumption value of the column, , These are the global minimum and maximum values ​​of fuel consumption for all samples, respectively. DIV_EPS is used to avoid the abnormal case where the denominator is zero. This is the normalized fuel consumption value.

[0012] Optionally, a dynamic masking matrix is ​​constructed based on a burnup threshold, and a sphere type state update mechanism is constructed based on core refueling rules, including: Based on a preset fuel consumption threshold FUEL_MASK_THRESH, the initial cyclic fuel consumption distribution data is compared grid-by-grid. Regions with fuel consumption values ​​higher than the threshold are identified as fuel spheres and marked as 1, while regions with fuel consumption values ​​lower than or equal to the threshold are identified as non-fuel spheres and marked as 0. The specific formula is as follows:

[0013] in, It is a dynamic masking matrix with a 20×6 grid, where 1 indicates that the corresponding position is a fuel sphere and 0 indicates that it is a non-fuel sphere. Obtain the type of the newly loaded ball at the top, determined by the DIY_Charge2 instruction, and the type of the ball discharged at the bottom, controlled by the DIY_Discharge instruction; A ball type state update mechanism is constructed based on the core dumping rules, which updates the ball type matrix before dumping. All rows except the top row are moved down one level, and the top row position is replaced with the newly loaded ball type at the top:

[0014] in, This is the ball type matrix before unloading. This is the ball type matrix after material pouring. For the type of ball newly loaded at the top; The formula updates the sphere type distribution map and new loading mark in real time, ensuring that the core dumping dynamics during the prediction process are completely synchronized with the actual physical process.

[0015] Optionally, the normalized fuel consumption distribution data from multiple consecutive cycles are expanded along the time dimension to construct a time-series snapshot matrix. Modal decomposition is then performed on the time-series snapshot matrix to extract modal features and calculate dynamic operators characterizing the evolution of fuel consumption, including: Obtain multi-cycle fuel consumption time-series data with shape T×20×6, where T is the cycle number. Expand and flatten the multi-cycle fuel consumption time-series data along the time dimension into a high-dimensional vector, and construct a time-series snapshot matrix X with dimension D×T, where D=120 corresponds to the total number of grids and all 120 rows of data are retained without filtering out rows with all zeros. Perform singular value decomposition on the time-series snapshot matrix X to obtain the left singular matrix U and the singular value diagonal matrix. and right singular matrix The specific formula is as follows:

[0016] The last column vector of the time series is removed from the time series snapshot matrix X to generate the forward snapshot matrix X1, and the first column vector of the time series is removed from the time series snapshot matrix X to generate the backward snapshot matrix X2. Based on the left singular matrix U and the singular value diagonal matrix Right singular matrix The forward snapshot matrix X1 and the backward snapshot matrix X2 are used to calculate a dynamic operator characterizing the evolution of fuel consumption through low-dimensional projection. .

[0017] Optionally, based on the dynamic operator, an autoregressive iterative calculation is performed on the fuel consumption distribution characteristics of the current cycle to obtain the predicted fuel consumption distribution data for the next cycle. This predicted fuel consumption distribution data is then used as the new data for the current cycle, and the process is repeated until the target prediction cycle is covered. This includes: by The 120-dimensional vector obtained by normalizing and flattening the real fuel consumption distribution data is used as the initial input and projected onto the DMD low-dimensional modal space to obtain the initial modal features. Based on the dynamic operator Matrix multiplication is performed on the initial modal features to calculate the low-dimensional fuel consumption vector for the next cycle, and the low-dimensional fuel consumption vector is restored to high-dimensional predicted fuel consumption distribution data in 20×6 format through inverse projection. The high-dimensional predicted fuel consumption distribution data obtained in the current cycle is used as the input data for the next round of autoregressive iteration. The projection, dynamic operator calculation and inverse projection restoration steps are repeatedly executed, and the fuel consumption distribution results from cycle=2 to cycle=12 are output sequentially until the target prediction cycle from the initial start-up to the complete conversion of the pure graphite sphere layer into a hybrid layer is covered.

[0018] To achieve the above objectives, a second aspect of the present invention provides a dynamic decomposition pebble bed type high-temperature gas-cooled reactor fuel sphere burnup distribution prediction device, comprising: The normalization module is used to obtain the fuel consumption distribution data of the core in the initial cycle and the current refueling status command, normalize the fuel consumption distribution data, construct a dynamic masking matrix based on the fuel consumption threshold, and construct a ball type status update mechanism based on the core refueling rules. The decomposition module is used to expand the normalized fuel consumption distribution data of multiple consecutive cycles along the time dimension, construct a time-series snapshot matrix, perform modal decomposition on the time-series snapshot matrix, extract modal features, and calculate dynamic operators that characterize the evolution law of fuel consumption. The calculation module is used to perform autoregressive iterative calculations on the fuel consumption distribution characteristics of the current cycle based on dynamic operators, obtain the predicted fuel consumption distribution data for the next cycle, and use the predicted fuel consumption distribution data as the new current cycle data, repeating the process until the target prediction cycle is covered. The output module is used to perform inverse normalization on the final predicted burnup distribution data and use a dynamic masking matrix to force the predicted values ​​corresponding to non-fuel spheres to zero, outputting the burnup distribution prediction results that conform to the core physical constraints.

[0019] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0020] To achieve the above objectives, a third aspect of this application provides an electronic device, including a processor and a memory; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, for implementing the dynamic decomposition pebble bed type high-temperature gas-cooled reactor fuel sphere burnup distribution prediction method as described in the first aspect embodiment.

[0021] To achieve the above objectives, a fourth aspect of this application provides a non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements the dynamic decomposition pebble bed high-temperature gas-cooled reactor fuel sphere burnup distribution prediction method as described in the first aspect embodiment.

[0022] The embodiments of the present invention have the following beneficial effects: This invention significantly reduces the calculation time for burnup distribution to the second level, meeting the timeliness requirements of real-time reactor monitoring and digital twins. Through physical constraint mechanisms and adjustable mode design, it achieves a flexible balance between calculation accuracy and efficiency while ensuring that the prediction results conform to the physical laws of the reactor core. Attached Figure Description

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart illustrating a method for predicting the burnup distribution of fuel spheres in a dynamic decomposition pebble bed type high-temperature gas-cooled reactor, provided in an embodiment of the present invention; Figure 2 A flowchart of another method for predicting the burnup distribution of fuel spheres in a dynamic decomposition spherical bed type high-temperature gas-cooled reactor provided in an embodiment of the present invention; Figure 3 This is a DMD singularity decay curve provided in an embodiment of the present invention; Figure 4 The cumulative energy percentage curve of DMD singularities provided in the embodiments of the present invention; Figure 5 The sensitivity analysis curves of DMD mode number provided in the embodiments of the present invention; Figure 6 This is a comparison chart of the average fuel consumption time series for all examples provided in the embodiments of the present invention; Figure 7 This is the overall average MAPE heatmap for all test cases provided in the embodiments of the present invention; Figure 8 This is a structural diagram of a dynamic decomposition pebble bed type high-temperature gas-cooled reactor fuel sphere burnup distribution prediction device provided in an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] The following describes, with reference to the accompanying drawings, a method and apparatus for predicting the burnup distribution of fuel spheres in a dynamic decomposition pebble bed type high-temperature gas-cooled reactor according to an embodiment of the present invention.

[0027] Example 1 This invention provides a method for predicting the burnup distribution of fuel spheres in a dynamically decomposed pebble bed type high-temperature gas-cooled reactor, such as... Figure 1 As shown, the method includes the following steps: S1: Obtain the fuel consumption distribution data of the core in the initial cycle and the current refueling status command, normalize the fuel consumption distribution data, construct a dynamic masking matrix based on the fuel consumption threshold, and construct a sphere type status update mechanism based on the core refueling rules.

[0028] In the embodiments of this application, S1, as the initial preparation step for model prediction, is the foundation for ensuring the accuracy and physical rationality of subsequent predictions. Its core lies in completing the initial data loading, standardization processing, and initialization of the physical constraint mechanism, providing input data that meets the requirements for subsequent snapshot matrix construction and DMD decomposition.

[0029] First, this embodiment of the application needs to obtain the burnup distribution data of the reactor core during the initial cycle (cycle=1). This data is strictly adapted to a 20×6 reactor core grid format, where the height H=20 corresponds to Layer 28 to Layer 9 of the reactor core, and the width W=6 corresponds to 6 radial regions. This grid format is the native adaptation format of the model in this application. The model is directly calculated based on this original grid data, removing the resize operation in traditional models, which can effectively eliminate the accuracy loss caused by data format conversion. At the same time, this embodiment of the application synchronously obtains the global operating parameters of the reactor core and the refueling status instructions. The global operating parameters of the reactor core include power, cumulative operating days, coolant outlet temperature, and control rod position. The refueling status instructions include DIY_Charge2 (top ball loading instruction) and DIY_Discharge (bottom ball loading instruction). The above parameters together provide data support for the operation of the subsequent physical constraint mechanism.

[0030] In this embodiment of the application, in order to eliminate the influence of dimensions and ensure the consistency of data calculation, the initial cyclic fuel consumption distribution data (BU) in 20×6 format needs to be normalized using Min-Max, and the specific formula is as follows:

[0031] in, This represents the original fuel consumption value (unit: GWd / tU) in the i-th row and j-th column of a 20×6 grid. , These are the global minimum and maximum values ​​of fuel consumption for all samples, respectively. DIV_EPS is used to avoid the abnormal case where the denominator is zero. This is the normalized fuel consumption value.

[0032] In this embodiment, the model training and testing data are derived from typical operating examples generated by the HUAKUN program for physical and thermal calculations of pebble-bed high-temperature gas-cooled reactors. The global range of fuel sphere burnup depth (BU) is 0.000 to 30.501 GWd / tU, and the dataset contains a total of 144,000 samples. The training set contains 9,600 examples and 115,200 samples, while the validation and test sets each contain 1,200 examples and 14,400 samples, ensuring that the normalization process uses... , It is sufficiently representative, ensuring the normalization effect.

[0033] In this embodiment of the application, to ensure that the prediction results conform to the physical operation law of the reactor core, step S1 needs to construct the first constraint in the dual physical constraint mechanism, which is the dynamic distinction constraint between fuel spheres and non-fuel spheres. Specifically, by setting a burnup threshold (FUEL_MASK_THRESH), a dynamic masking matrix is ​​constructed, and the region with a burnup value higher than the threshold is determined as a fuel sphere, and the region with a burnup value lower than the threshold is determined as a non-fuel sphere. The specific formula is as follows:

[0034] in, The dynamic masking matrix is ​​a 20×6 grid, where 1 indicates a fuel sphere and 0 indicates a non-fuel sphere (graphite sphere). As the cycle progresses, the number of rows for fuel spheres gradually increases while the number of rows for non-fuel spheres decreases accordingly. This dynamic masking matrix can accurately match the physical process of the transformation from pure graphite spheres to a hybrid layer. Furthermore, in subsequent model calculations, only the prediction error for fuel spheres is calculated, while the predicted values ​​for non-fuel spheres are forcibly set to zero, effectively avoiding false burnout.

[0035] In this embodiment, step S1 simultaneously constructs the second constraint in the dual physical constraint mechanism: the refueling state adaptation constraint. Using the HTR-PM core refueling rules, a sphere type state update mechanism is constructed. Initially, the top 9 layers of the core are a mixture of fuel spheres and graphite spheres, and the bottom 11 layers are pure graphite spheres. After each refueling cycle, all spheres in the core move down one layer, a specified number of spheres are discharged from the bottom, and a new layer of mixed spheres is loaded into the top. The model updates the sphere type distribution map and the newly loaded markers in real time through this mechanism, ensuring that the prediction process is completely synchronized with the actual core refueling dynamics. The specific formula is as follows:

[0036] in, This is the ball type matrix before unloading. This is the ball type matrix after material pouring. The type of ball newly loaded at the top is determined by the current cycle's DIY_Charge2 instruction; the type of ball discharged at the bottom is controlled by the DIY_Discharge class instruction.

[0037] Furthermore, in this embodiment, step S1 also requires initializing the model's runtime environment. To ensure consistency of results under different hardware environments (CPU / GPU) and multi-threaded computing scenarios, this application constructs a multi-level random seed fixing mechanism, namely, initializing the Python built-in random library, NumPy random number generator, and TensorFlow random number generator respectively, and setting a unified fixed seed value (SEED=42); at the same time, configuring the TensorFlow deterministic computing environment, enabling TF_DETERMINISTIC_OPS and TF_CUDNN_DETERMINISTIC modes, forcing all operators to use deterministic algorithms, and avoiding result deviations caused by parallel computing.

[0038] S2 expands the normalized fuel consumption distribution data of multiple consecutive cycles along the time dimension to construct a time-series snapshot matrix. Modal decomposition is performed on the time-series snapshot matrix to extract modal features and calculate dynamic operators that characterize the evolution law of fuel consumption.

[0039] In this embodiment, S2 is the core computational preparation step of the model. Its core purpose is to extract the temporal evolution characteristics of fuel consumption distribution through the construction of a time-series snapshot matrix and DMD mode decomposition, calculate dynamic operators, and provide core support for subsequent autoregressive prediction. At the same time, through the adjustable mode number mechanism, a balance between prediction accuracy and computational efficiency is achieved.

[0040] First, this embodiment of the application acquires multi-cycle burnup time-series data with a shape of T×20×6, where T is the cycle number. This multi-cycle burnup time-series data is expanded and flattened along the time dimension into a high-dimensional vector, constructing a time-series snapshot matrix X. The matrix dimension is D×T, where D=20×6=120, corresponding to the total number of grids. Unlike traditional models, the snapshot matrix constructed in this embodiment retains all 120 rows of data, without filtering rows containing all zeros, thus completely preserving the core spatial distribution information. Its mathematical expression is:

[0041] in, For the first The high-dimensional vector of fuel consumption distribution for each cycle is obtained by flattening the 20×6 two-dimensional fuel consumption distribution grid for the corresponding cycle.

[0042] In this embodiment of the application, after constructing the time-series snapshot matrix X, singular value decomposition (SVD) is performed on it to obtain the left singular matrix U, the singular value diagonal matrix Σ, and the right singular matrix Vn, which satisfy:

[0043] The core function of singular value decomposition (SVD) is to extract key modal features of fuel consumption distribution. Different singular values ​​correspond to different modes of contribution; the larger the singular value, the more significant the influence of the corresponding mode on the evolution of fuel consumption distribution. For example... Figure 3 As shown, the horizontal axis represents the singular value index (1-20), and the vertical axis represents the singular value value. The figure marks the currently configured DMD_MODES parameter value of 20 with a red dashed line, clearly showing the decay law of singular values ​​as the index increases. This intuitively reflects the contribution of different modes to the fuel consumption distribution information and provides an intuitive basis for the selection of the number of modes in the future.

[0044] In this embodiment of the application, after the singular value decomposition is completed, the snapshot matrix is ​​split into a forward snapshot matrix X1 (with the last column removed) and a backward snapshot matrix X2 (with the first column removed). The dynamic operator Φ is obtained by low-dimensional projection calculation. This dynamic operator is the core parameter characterizing the time-series evolution law of fuel consumption distribution. Subsequent autoregressive predictions are all based on this dynamic operator.

[0045] In this embodiment, to achieve a precise balance between prediction accuracy and computational efficiency, step S2 incorporates an adjustable mode number mechanism. The core model parameter DMD_MODES can be flexibly adjusted within the range of 1-120, and the optimal mode number is determined through singular value energy analysis. Specifically, the cumulative energy percentage of singular values ​​is calculated. When the cumulative energy reaches a target threshold, the corresponding minimum mode number is the optimal configuration. The formula for calculating the cumulative energy percentage is as follows:

[0046] in, For the number of candidate modes, when the cumulative energy percentage When the target threshold is reached, the corresponding minimum The value represents the optimal number of modes. The core advantage of this mechanism is that a higher number of modes leads to higher prediction accuracy but also increases computation time, while a lower number increases computation speed and moderately reduces accuracy, allowing the model to meet the needs of different engineering scenarios. For example... Figure 4 As shown, the horizontal axis represents the number of DMD modes, and the vertical axis represents the percentage of cumulative energy. The orange dotted line marks the current DMD_MODES parameter value of 20, which quantifies the percentage of singular cumulative energy corresponding to different numbers of modes. This allows for the accurate determination of the optimal number of modes to cover the target energy. In this embodiment, 20 modes (DMD_MODES=20) are used for calculation. This configuration can cover 99.999821% of the singular cumulative energy, balancing prediction accuracy and computational efficiency.

[0047] At the same time, such as Figure 5 As shown, the horizontal axis represents the number of DMD modes, and the vertical axis represents the mean square error (MSE) of fuel zone reconstruction. The figure marks the current DMD_MODES parameter value of 20 with a red dashed line, showing the trend of fuel consumption distribution reconstruction error corresponding to different modes, quantifying the influence of the number of modes on the model accuracy, and further verifying the rationality of DMD_MODES=20 as the optimal number of modes.

[0048] In this embodiment, by constructing a time-series snapshot matrix, executing the DMD singular value decomposition calculation dynamic operator, and setting an adjustable mode number mechanism, the time-series evolution characteristics of fuel consumption distribution are extracted and the optimal mode number is determined, laying the foundation for subsequent autoregressive iterative calculation.

[0049] S3, based on the dynamic operator, performs autoregressive iterative calculation on the fuel consumption distribution characteristics of the current cycle to obtain the predicted fuel consumption distribution data for the next cycle, and uses the predicted fuel consumption distribution data as the new current cycle data, repeating the process until the target prediction cycle is covered.

[0050] In this embodiment, S3 is the core prediction step of the model, based on the dynamic operator obtained from step S2. An autoregressive prediction framework was constructed to achieve full-process prediction from the initial start-up state to the complete formation of the hybrid layer. The core lies in completing the time-series prediction of burnup distribution in multiple cycles through iterative calculations, ensuring that the prediction process is completely synchronized with the actual dynamic operation of the reactor core.

[0051] First, perform initial state initialization, to The actual fuel consumption distribution data is used as the initial input. After normalization, the data is flattened into a 120-dimensional high-dimensional vector, which is then projected onto the DMD low-dimensional modal space to obtain the initial modal features. These initial modal features retain the core information of the initial fuel consumption distribution and provide an initial benchmark for subsequent iterative predictions.

[0052] In this embodiment of the application, after initial state initialization is completed, single-step prediction is performed: based on the dynamic operator obtained in step S2. Perform matrix multiplication on the initial modal features to calculate the next iteration ( The low-dimensional burnup vector is then converted into high-dimensional burnup distribution data in 20×6 format through inverse projection. This conversion process ensures that the prediction results are consistent with the core mesh format, which facilitates subsequent physical constraint post-processing and result output.

[0053] In this embodiment, after a single-step prediction is completed, a multi-loop iteration phase is entered: the prediction result of the current loop is used as the input data for the next loop, and the projection, dynamic operator calculation, and inverse projection restoration steps are repeatedly executed, outputting sequentially. to The burnup distribution results are used to predict the target cycle from initial startup to the complete conversion of the pure graphite sphere layer to the hybrid layer. During this iteration, the embodiments of this application call the constructed sphere type status update mechanism in real time. According to the discharging status instructions (DIY_Charge2, DIY_Discharge), the sphere type matrix and the new loading mark are updated in real time to ensure that each iteration prediction is dynamically synchronized with the actual core discharging, avoiding prediction errors caused by the disconnection of discharging status.

[0054] like Figure 6 As shown, the horizontal axis represents the number of core refueling cycles, and the vertical axis represents the average burnup value of the fuel zone (GWd / tU). The blue solid line in the figure represents the actual burnup value, and the red dashed line represents the DMD predicted burnup value. This figure intuitively demonstrates the effect of autoregressive prediction. It can be seen from the figure that the predicted value is in excellent agreement with the actual value, which verifies the rationality and effectiveness of the prediction logic of step S3.

[0055] In this embodiment, the autoregressive prediction process operates at a time consumption of only seconds, eliminating the need for complex deep learning training procedures. Compared to traditional Monte Carlo and deterministic burnup calculation methods that take 1 to 2 hours per calculation, this significantly improves prediction efficiency and meets the engineering requirements for real-time core monitoring. Furthermore, during iterative prediction, the model incorporates a real-time adjustable mode number mechanism, performing calculations based on the optimal mode number (DMD_MODES=20) to ensure that each iterative prediction achieves an optimal balance between accuracy and efficiency.

[0056] S4 performs inverse normalization on the final predicted burnup distribution data and uses a dynamic masking matrix to force the predicted values ​​corresponding to non-fuel spheres to zero, outputting the burnup distribution prediction results that conform to the core physical constraints.

[0057] In this embodiment, S4 is the post-processing and output step of the model. Its core purpose is to optimize and correct the obtained prediction results to ensure that the output results conform to the physical laws and format specifications of the reactor core and can be directly applied to engineering practice, providing reliable support for reactor core fuel cycle management and operation safety analysis.

[0058] First, the obtained predicted fuel consumption distribution data is denormalized to restore the normalized fuel consumption value to its original dimensions (GWd / tU), eliminating the influence of the normalization operation on the fuel consumption value and ensuring that the physical meaning of the prediction result is clear and can be directly used for fuel consumption distribution assessment and safety margin calculation.

[0059] In this embodiment of the application, after the denormalization process is completed, the dynamic masking matrix constructed in step S1 is called. Physical constraint correction is applied to the predicted burnup distribution data: the predicted values ​​corresponding to the regions marked as 0 in the masking matrix (non-fuel spheres, i.e. graphite spheres) are forced to be zero, and only the predicted values ​​of the regions marked as 1 (fuel spheres) are retained. This effectively eliminates false burnup and ensures that the prediction results conform to the distribution patterns of fuel spheres and graphite spheres in the reactor core, further improving the engineering usability of the prediction results.

[0060] In this embodiment, after inverse normalization and physical constraint correction, a burnup distribution prediction result conforming to the core physical constraints is output. This result still maintains a 20×6 core grid format (height corresponds to core Layer 28 to Layer 9, width corresponds to 6 radial regions), and can be directly transmitted to the core monitoring system or edge computing device for burnup distribution assessment, safety margin calculation, and digital operation and maintenance decision-making. Figure 1 As shown, step S4 corresponds to the "post-processing" and "output" stages in the flowchart. It follows the prediction results of step S3 and outputs the final usable fuel consumption distribution data through inverse normalization and physical constraint correction, fully covering the entire process from raw data to final output.

[0061] To verify the prediction accuracy of the post-processing and the entire model, this application's embodiments validated the model using 1200 test cases. The tests showed that the model performed excellently in autoregressive prediction: the overall average MAE was 0.2773 GWd / tU, the overall average MAPE was 9.672%, and the average R² from cycle=2 to cycle=12 remained between 0.9629 and 0.9911. In the early cycles (cycle2-cycle4), the MAPE was below 5.3% and the R² was above 0.982, demonstrating excellent prediction accuracy. Although the error increased slightly in the later cycles, it still maintained a high degree of fit, reflecting long-term prediction stability. Figure 7 As shown, a heatmap summarizes the average relative error distribution of all test cases in a 20×6 core grid. The vertical axis represents Layer 28 to Layer 9, and the horizontal axis represents the radial regions 1-6. Each grid cell is labeled with a specific MAPE value. This heatmap corresponds to the overall average MAE (0.2773 GWd / tU) and MAPE (9.672%) results, clearly presenting the error differences at different levels and in different radial regions, demonstrating the model's accurate predictive ability for the overall burnup distribution of the core.

[0062] In the embodiments of this application, the model realizes rapid prediction of burnup distribution throughout the entire process from initial reactor start-up to the complete conversion of pure graphite spherical layers into hybrid layers. Through the coordinated cooperation of each step, it solves the pain points of traditional methods being time-consuming and existing deep learning models having weak interpretability. At the same time, by fixing the global random seed, it ensures the reliability of the results, significantly improving the engineering usability of burnup distribution calculation for pebble bed reactors. It has important industrial application value in scenarios such as real-time reactor monitoring, digital twins, and fuel management optimization.

[0063] Example 2 This invention provides a dynamic decomposition pebble bed type high-temperature gas-cooled reactor fuel sphere burnup distribution prediction device 10, such as... Figure 8 As shown, the device includes: The normalization module is used to obtain the fuel consumption distribution data of the core in the initial cycle and the current refueling status command, normalize the fuel consumption distribution data, construct a dynamic masking matrix based on the fuel consumption threshold, and construct a ball type status update mechanism based on the core refueling rules. The decomposition module is used to expand the normalized fuel consumption distribution data of multiple consecutive cycles along the time dimension, construct a time-series snapshot matrix, perform modal decomposition on the time-series snapshot matrix, extract modal features, and calculate dynamic operators that characterize the evolution law of fuel consumption. The calculation module is used to perform autoregressive iterative calculations on the fuel consumption distribution characteristics of the current cycle based on dynamic operators, obtain the predicted fuel consumption distribution data for the next cycle, and use the predicted fuel consumption distribution data as the new current cycle data, repeating the process until the target prediction cycle is covered. The output module is used to perform inverse normalization on the final predicted burnup distribution data and use a dynamic masking matrix to force the predicted values ​​corresponding to non-fuel spheres to zero, outputting the burnup distribution prediction results that conform to the core physical constraints.

[0064] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0065] Example 3 To implement the methods of the above embodiments, the present invention also provides an electronic device, which includes a memory and a processor; wherein the processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the various steps of the methods described above.

[0066] Example 4 To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in the foregoing embodiments.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0068] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0069] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for predicting the burnup distribution of fuel spheres in a dynamic decomposition pebble bed type high-temperature gas-cooled reactor, characterized in that, include: S1: Obtain the fuel consumption distribution data of the core in the initial cycle and the current refueling status command, normalize the fuel consumption distribution data, construct a dynamic masking matrix based on the fuel consumption threshold, and construct a ball type status update mechanism based on the core refueling rules. S2 expands the normalized fuel consumption distribution data of multiple consecutive cycles along the time dimension to construct a time-series snapshot matrix, performs modal decomposition on the time-series snapshot matrix, extracts modal features, and calculates dynamic operators that characterize the evolution law of fuel consumption. S3, based on the dynamic operator, performs autoregressive iterative calculation on the fuel consumption distribution characteristics of the current cycle to obtain the predicted fuel consumption distribution data for the next cycle, and uses the predicted fuel consumption distribution data as the new current cycle data, repeating the process until the target prediction cycle is covered; S4 performs inverse normalization on the final predicted burnup distribution data and uses a dynamic masking matrix to force the predicted values ​​corresponding to non-fuel spheres to zero, outputting the burnup distribution prediction results that conform to the core physical constraints.

2. The method according to claim 1, characterized in that, The process of acquiring the core burnup distribution data during the initial cycle and the current refueling status command, and normalizing the burnup distribution data, includes: Read the initial cycle burnup distribution data conforming to the 20×6 core grid format, where the height corresponds to the core Layer 28 to Layer 9, and the width corresponds to the 6 radial regions. Simultaneously acquire the core global operation parameters including power, cumulative operating days, coolant outlet temperature, and control rod position. The initial cyclic fuel consumption distribution data is normalized using the global minimum and global maximum values ​​to generate normalized fuel consumption distribution data that eliminates the influence of dimensions. in, For the 20×6 grid, the first Line number The original fuel consumption value of the column, , These are the global minimum and maximum values ​​of fuel consumption for all samples, respectively. DIV_EPS is used to avoid the abnormal case where the denominator is zero. This is the normalized fuel consumption value.

3. The method according to claim 1, characterized in that, The mechanism for constructing a dynamic masking matrix based on a burnup threshold and for updating the sphere type state based on core dumping rules includes: Based on a preset fuel consumption threshold FUEL_MASK_THRESH, the initial cyclic fuel consumption distribution data is compared grid-by-grid. Regions with fuel consumption values ​​higher than the threshold are identified as fuel spheres and marked as 1, while regions with fuel consumption values ​​lower than or equal to the threshold are identified as non-fuel spheres and marked as 0. The specific formula is as follows: in, It is a dynamic masking matrix with a 20×6 grid, where 1 indicates that the corresponding position is a fuel sphere and 0 indicates that it is a non-fuel sphere. Obtain the type of the newly loaded ball at the top, determined by the DIY_Charge2 instruction, and the type of the ball discharged at the bottom, controlled by the DIY_Discharge instruction; A ball type state update mechanism is constructed based on the core dumping rules, which updates the ball type matrix before dumping. All rows except the top row are moved down one level, and the top row position is replaced with the newly loaded ball type at the top: in, This is the ball type matrix before unloading. This is the ball type matrix after material pouring. For the type of ball newly loaded at the top; The formula updates the sphere type distribution map and new loading mark in real time, ensuring that the core dumping dynamics during the prediction process are completely synchronized with the actual physical process.

4. The method according to claim 1, characterized in that, The process of expanding multiple consecutive, cyclically normalized fuel consumption distribution data along the time dimension to construct a time-series snapshot matrix, performing modal decomposition on the time-series snapshot matrix, extracting modal features, and calculating dynamic operators characterizing the evolution of fuel consumption includes: Obtain multi-cycle fuel consumption time-series data with shape T×20×6, where T is the cycle number. Expand and flatten the multi-cycle fuel consumption time-series data along the time dimension into a high-dimensional vector, and construct a time-series snapshot matrix X with dimension D×T, where D=120 corresponds to the total number of grids and all 120 rows of data are retained without filtering out rows with all zeros. Perform singular value decomposition on the time-series snapshot matrix X to obtain the left singular matrix U and the singular value diagonal matrix. and right singular matrix The specific formula is as follows: The last column vector of the time series is removed from the time series snapshot matrix X to generate the forward snapshot matrix X1, and the first column vector of the time series is removed from the time series snapshot matrix X to generate the backward snapshot matrix X2. Based on the left singular matrix U and the singular value diagonal matrix Right singular matrix The forward snapshot matrix X1 and the backward snapshot matrix X2 are used to calculate a dynamic operator characterizing the evolution of fuel consumption through low-dimensional projection. .

5. The method according to claim 1, characterized in that, The process of performing autoregressive iterative calculations on the fuel consumption distribution characteristics of the current cycle based on dynamic operators to obtain the predicted fuel consumption distribution data for the next cycle, and then using the predicted fuel consumption distribution data as the new current cycle data, repeating the process until the target prediction cycle is covered, includes: by The 120-dimensional vector obtained by normalizing and flattening the real fuel consumption distribution data is used as the initial input and projected onto the DMD low-dimensional modal space to obtain the initial modal features. Based on the dynamic operator Matrix multiplication is performed on the initial modal features to calculate the low-dimensional fuel consumption vector for the next cycle, and the low-dimensional fuel consumption vector is restored to high-dimensional predicted fuel consumption distribution data in 20×6 format through inverse projection. The high-dimensional predicted fuel consumption distribution data obtained in the current cycle is used as the input data for the next round of autoregressive iteration. The projection, dynamic operator calculation and inverse projection restoration steps are repeatedly executed, and the fuel consumption distribution results from cycle=2 to cycle=12 are output sequentially until the target prediction cycle from the initial start-up to the complete conversion of the pure graphite sphere layer into a hybrid layer is covered.

6. A dynamic decomposition pebble bed type high-temperature gas-cooled reactor fuel pelt burnup distribution prediction device, characterized in that, include: The normalization module is used to obtain the fuel consumption distribution data of the core in the initial cycle and the current refueling status command, normalize the fuel consumption distribution data, construct a dynamic masking matrix based on the fuel consumption threshold, and construct a ball type status update mechanism based on the core refueling rules. The decomposition module is used to expand the normalized fuel consumption distribution data of multiple consecutive cycles along the time dimension, construct a time-series snapshot matrix, perform modal decomposition on the time-series snapshot matrix, extract modal features, and calculate dynamic operators that characterize the evolution law of fuel consumption. The calculation module is used to perform autoregressive iterative calculations on the fuel consumption distribution characteristics of the current cycle based on dynamic operators, obtain the predicted fuel consumption distribution data for the next cycle, and use the predicted fuel consumption distribution data as the new current cycle data, repeating the process until the target prediction cycle is covered. The output module is used to perform inverse normalization on the final predicted burnup distribution data and use a dynamic masking matrix to force the predicted values ​​corresponding to non-fuel spheres to zero, outputting the burnup distribution prediction results that conform to the core physical constraints.

7. An electronic device, characterized in that, Including processor and memory; The processor reads executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method as described in any one of claims 1-5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

Images