Method and device for verifying data of multi-nucleus neutron nuclear reaction based on deep learning

Abstract

The application discloses a multi-nucleus neutron nuclear reaction data verification method and device based on deep learning, comprising the following steps: determining a target nucleus to be evaluated, and obtaining multi-dimensional physical characteristic data of the target nucleus from a preset evaluation nuclear database and an experimental database; performing physical perception preprocessing and feature splicing on the multi-dimensional physical characteristic data to obtain a physical characteristic vector of the target nucleus; constructing and training a nuclear data correction model with uncertainty quantification capability; determining evaluation cross-section data and uncertainty data of the target nucleus according to the nuclear data correction model and the physical characteristic vector, and encapsulating the evaluation data of the target nucleus according to the evaluation cross-section data and the uncertainty data; verifying the evaluation data based on particle transport simulation, and optimizing the evaluation nuclear database according to the verified evaluation data. The application improves the efficiency and reliability of neutron nuclear reaction data evaluation and verification, and can be applied to the technical field of data processing.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and apparatus for verifying multi-nucleon neutron nuclear reaction data based on deep learning. Background Technology

[0002] Nuclear data is fundamental to nuclear energy development, nuclear device design, and radiation protection. Among these, cross-section data of neutron-nuclear interactions are particularly crucial. Currently, internationally accepted nuclear evaluation databases (such as the US ENDF / B series, Japan's JENDL series, and China's CENDL series) are generated based on a combination of theoretical model calculations and experimental measurement data.

[0003] Existing nuclear data evaluation processes typically include the following steps:

[0004] 1. Experimental data collection: Retrieve experimental measurements for specific nuclides from EXFOR (Experimental Nuclear Reaction Data and Related Parameters Database).

[0005] 2. Theoretical model calculation: Theoretical cross-sectional values ​​are calculated using nuclear reaction modeling programs (such as EMPIRE, TALYS, etc.).

[0006] 3. Data Fitting and Evaluation: Statistical methods (such as generalized least squares, GLS) are used to fuse experimental and theoretical values ​​to obtain recommended evaluation data.

[0007] 4. Benchmark verification: The evaluated data are compiled into an application library, and the critical benchmark experiment is simulated using a Monte Carlo particle transport program (such as MCNP) to verify the accuracy of the data.

[0008] Although the aforementioned traditional evaluation methodologies are relatively mature, they still face the following significant technical bottlenecks and shortcomings when dealing with large-scale, multi-nucleon data evaluation tasks:

[0009] (1) The processing efficiency of multi-nucleoside data is low, making it difficult to achieve batch evaluation: Traditional evaluation work is usually carried out independently for a single nuclide (such as only for U-235 or Fe-56). Due to the huge differences in the physical properties, energy range division and resonance structure of different nuclides, evaluators often need to adjust parameters, screen experimental data and fit them individually for each nuclide. This "manual workshop" evaluation mode is inefficient and cannot meet the needs of modern nuclear engineering for the rapid updating and iteration of hundreds or thousands of nuclide data.

[0010] (2) The experimental data itself contains systematic biases and contradictions: The experimental data in the EXFOR database comes from a wide range of sources, and the measurement methods and standards used by different laboratories and in different eras are inconsistent, which often leads to significant differences between experimental data for the same nuclide (i.e., "data contradictions"). The traditional least squares method (GLS) is easily interfered with when dealing with data with serious outliers or systematic biases, causing the evaluation results to deviate from the true physical laws. In addition, manually removing bad points is not only time-consuming, but also highly subjective.

[0011] (3) Traditional statistical methods are insufficient for modeling nonlinear relationships: the cross-section of nuclear reactions changes with energy in a highly nonlinear manner (especially in the resonance region). Traditional statistical evaluation methods are mostly based on linear or Gaussian distribution assumptions, which make it difficult to accurately capture complex nonlinear physical laws. Although some studies have attempted to introduce neural networks in recent years, most of them are limited to simple fitting of a single nuclide, lack a universal feature extraction mechanism, and cannot transfer learning laws between multiple nuclides.

[0012] (4) The evaluation and verification processes are disconnected, lacking closed-loop feedback: In the existing process, "nuclear data evaluation" and "macroscopic benchmark verification (particle transport simulation)" are usually two independent stages. After the evaluators generate the data, it often takes a long time to obtain downstream verification feedback. If the simulation results are not ideal, it is necessary to return to the evaluation stage to find the cause, resulting in a long iterative cycle. There is a lack of an automated closed-loop mechanism that directly feeds back the macroscopic integral (benchmark experimental results) for the correction of microscopic components (cross-sectional data).

[0013] The above problems urgently need to be addressed. Summary of the Invention

[0014] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0015] Therefore, one objective of this invention is to provide a method for verifying multi-nucleon nuclear reaction data based on deep learning. This method can utilize the powerful nonlinear mapping capability of deep neural networks to uniformly process multi-source physical data of multiple nuclides, batch learn and correct complex deviations between theoretical and experimental values, automatically generate evaluation data files in standard format and perform validity verification, thereby improving the efficiency and reliability of neutron nuclear reaction data evaluation and verification.

[0016] Another objective of this invention is to provide a deep learning-based multi-nucleon neutron nuclear reaction data verification device.

[0017] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of the present invention include:

[0018] On one hand, embodiments of the present invention provide a method for verifying multi-nucleon neutron nuclear reaction data based on deep learning, comprising the following steps:

[0019] The target nuclide to be evaluated is determined, and multi-dimensional physical characteristic data of the target nuclide are obtained from the preset evaluation nuclide database and experimental database;

[0020] The multi-dimensional physical feature data is subjected to physical perception preprocessing and feature concatenation to obtain the physical feature vector of the target nuclide;

[0021] Construct and train a kernel data correction model with uncertainty quantification capability;

[0022] The evaluation cross-sectional data and uncertainty data of the target nuclide are determined based on the nuclear data correction model and the physical feature vector, and the evaluation data of the target nuclide is obtained by encapsulating the evaluation cross-sectional data and the uncertainty data.

[0023] The evaluation data is validated based on particle transport simulation, and the evaluation kernel database is optimized based on the validated evaluation data.

[0024] Furthermore, in one embodiment of the present invention, the multi-dimensional physical characteristic data includes neutron energy data, multi-channel reaction cross-section data, and physical and chemical property data. The neutron energy data includes data points of the target nuclide at different incident neutron energies in the thermal neutron region, resonance region, and fast neutron region. The multi-channel reaction cross-section data includes the total cross-section data of the target nuclide and the cross-section values ​​of different interactions with neutrons, protons, and gamma rays. The physical and chemical property data includes the mass number, atomic number, isotopic abundance information, half-life, excited state energy, energy level density parameter, spin, parity, chemical binding energy, and nuclear binding energy of the target nuclide.

[0025] Furthermore, in one embodiment of the present invention, the step of performing physical perception preprocessing and feature concatenation on the multi-dimensional physical feature data to obtain the physical feature vector of the target nuclide specifically includes:

[0026] A non-uniform mixed energy grid is constructed based on the neutron energy data;

[0027] The multi-channel reaction cross-section data are interpolated using a logarithmic space interpolation algorithm to obtain discrete experimental cross-section data.

[0028] The discrete experimental cross-section data is mapped onto the non-uniform mixed energy grid, and the weighted average method is used to calculate the weighted average of the experimental cross-sections at each energy point in the full energy range to obtain the dynamic energy characteristics.

[0029] The dynamic energy features are concatenated with the physical and chemical property data to obtain the physical feature vector.

[0030] Furthermore, in one embodiment of the present invention, the construction and training of the kernel data correction model with uncertainty quantification capability specifically includes:

[0031] A surrogate model for simulating the mapping relationship between cross sections and multiplication factors is trained based on a lightweight multilayer perceptron.

[0032] A deep neural network was constructed to predict the cross-sectional distribution of the full-range region, and physical feature samples, cross-sectional distribution labels, and true proliferation factors of the sample nuclides were obtained.

[0033] The physical feature samples are input into the deep neural network to obtain the cross-sectional predicted distribution of the sample nuclides;

[0034] The predicted cross-sectional distribution is input into the surrogate model to obtain the predicted proliferation factor;

[0035] The total loss value is determined based on the predicted cross-sectional distribution, the cross-sectional distribution label, the predicted proliferation factor, the actual proliferation factor, and a preset hybrid loss function.

[0036] The deep neural network is optimized based on the total loss value to obtain the kernel data correction model.

[0037] Furthermore, in one embodiment of the present invention, the hybrid loss function is:

[0038]

[0039] in, This represents the total loss value. This represents the negative log-likelihood loss between the predicted cross-sectional distribution and the cross-sectional distribution label. This represents the mean square error between the predicted and actual proliferation factors. Represents the smoothing constraint term. Represents the uncertainty constraint term. , , as well as All of these represent optimizable weight coefficients.

[0040] Further, in one embodiment of the present invention, the step of determining the evaluation cross-sectional data and uncertainty data of the target nuclide based on the nuclear data correction model and the physical feature vector, and encapsulating the evaluation data of the target nuclide based on the evaluation cross-sectional data and the uncertainty data, specifically includes:

[0041] The physical feature vector is input into the nuclear data correction model to obtain the target cross-sectional distribution data of the target nuclide;

[0042] Monte Carlo sampling is performed on the target cross-section distribution data to obtain a set of predicted cross-sections of the target nuclide at each energy point in the full energy range;

[0043] Calculate the cross-sectional mean and corresponding uncertainty of the target nuclide at each energy point in the full energy range based on the cross-sectional prediction set;

[0044] The evaluation cross-sectional data is generated based on the cross-sectional mean of the target nuclide at each energy point in the full energy range, and the corresponding uncertainty data is generated based on the uncertainty.

[0045] The evaluation cross-sectional data and the uncertainty data are written into a preset kernel data file format to obtain the evaluation data.

[0046] Furthermore, in one embodiment of the present invention, the step of verifying the evaluation data based on particle transport simulation and optimizing the evaluation kernel database based on the verified evaluation data specifically includes:

[0047] Based on the evaluation data, a particle transport simulation experiment was conducted on the target nuclide to obtain the experimental measurement cross-sectional values ​​of the target nuclide at each energy point in the full energy range, as well as the macroscopic physical quantities of the target nuclide.

[0048] The chi-square value per unit degree of freedom is calculated based on the evaluation cross-sectional data and the experimental measurement cross-sectional values, and the evaluation data is verified by micro-statistics based on the degree of closeness of the chi-square value per unit degree of freedom to 1.

[0049] The evaluation data are verified by macroscopic integration based on the difference between the macroscopic physical quantity and the reference true value of the target nuclide;

[0050] When both the micro-statistical verification and the macro-integral verification pass, the evaluation kernel database is updated based on the evaluation data.

[0051] If the micro-statistical verification or macro-integral verification fails, the kernel data correction model is optimized and fine-tuned.

[0052] On the other hand, embodiments of the present invention provide a multi-nucleon neutron nuclear reaction data verification device based on deep learning, comprising:

[0053] The data acquisition module is used to determine the target nuclide to be evaluated and to acquire multi-dimensional physical characteristic data of the target nuclide from a preset evaluation nuclide database and an experimental database.

[0054] The data preprocessing module is used to perform physical perception preprocessing and feature splicing on the multi-dimensional physical feature data to obtain the physical feature vector of the target nuclide.

[0055] The model training module is used to build and train a kernel data correction model with uncertainty quantification capability;

[0056] The data correction module is used to determine the evaluation cross-sectional data and uncertainty data of the target nuclide based on the nuclear data correction model and the physical feature vector, and to encapsulate the evaluation data of the target nuclide based on the evaluation cross-sectional data and the uncertainty data.

[0057] The data verification module is used to verify the evaluation data based on particle transport simulation, and to optimize the evaluation kernel database based on the verified evaluation data.

[0058] On the other hand, embodiments of the present invention provide an electronic device, including:

[0059] At least one processor;

[0060] At least one memory for storing at least one program;

[0061] When the at least one program is executed by the at least one processor, the at least one processor implements the above-described deep learning-based method for verifying multi-nucleon nuclear reaction data.

[0062] On the other hand, embodiments of the present invention also provide a computer-readable storage medium storing a processor-executable computer program that, when executed by a processor, implements the above-described deep learning-based method for verifying multi-nucleon nuclear reaction data.

[0063] On the other hand, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the above-described deep learning-based method for verifying multi-nucleon nuclear reaction data.

[0064] The advantages and beneficial effects of the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention:

[0065] This invention identifies the target nuclide to be evaluated and obtains multi-dimensional physical feature data of the target nuclide from a pre-set evaluation nuclide database and an experimental database. The multi-dimensional physical feature data undergoes physical sensing preprocessing and feature concatenation to obtain the physical feature vector of the target nuclide. A nuclear data correction model with uncertainty quantification capability is constructed and trained. Evaluation cross-sectional data and uncertainty data of the target nuclide are determined based on the nuclear data correction model and the physical feature vector. Evaluation data of the target nuclide is then encapsulated based on the evaluation cross-sectional data and uncertainty data. The evaluation data is verified based on particle transport simulation, and the evaluation nuclide database is optimized based on the verified evaluation data. This invention can utilize the powerful nonlinear mapping capability of deep neural networks to uniformly process multi-source physical data of various nuclides, batch learn and correct complex deviations between theoretical and experimental values, automatically generate standard format evaluation data files, and perform validity verification, thereby improving the efficiency and reliability of neutron nuclear reaction data evaluation and verification. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the embodiments of the present invention are described below. It should be understood that the drawings described below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A flowchart illustrating the steps of a deep learning-based method for verifying multi-nucleon neutron nuclear reaction data, as provided in this embodiment of the invention.

[0068] Figure 2 This is a structural block diagram of a deep learning-based multi-nucleon neutron nuclear reaction data verification device provided in an embodiment of the present invention;

[0069] Figure 3 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0071] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0072] The deep learning-based multi-nucleon nuclear reaction data verification method provided in this invention can be applied to terminals, servers, or software running on either terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the deep learning-based multi-nucleon nuclear reaction data verification method, but is not limited to the above forms.

[0073] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0074] To overcome the shortcomings and deficiencies of existing nuclear data evaluation techniques, such as high reliance on manual labor, the ability to process only a single nuclide individually, low evaluation efficiency, and lack of automated verification loops, this invention provides a method for batch evaluation, correction, and verification of multi-nucleon neutron nuclear reaction data based on deep learning. This method can utilize the powerful nonlinear mapping capability of deep neural networks to uniformly process the physical data of multiple nuclides, batch learn and correct complex deviations between theoretical and experimental data, automatically generate standard format nuclear data files, and complete the validity verification based on reactor physics simulation.

[0075] Reference Figure 1 This invention provides a method for verifying multi-nucleon neutron nuclear reaction data based on deep learning, specifically including the following steps:

[0076] S101. Determine the target nuclide to be evaluated, and obtain multi-dimensional physical characteristic data of the target nuclide from the preset evaluation nuclide database and experimental database;

[0077] S102. Perform physical perception preprocessing and feature splicing on the multi-dimensional physical feature data to obtain the physical feature vector of the target nuclide;

[0078] S103. Construct and train a kernel data correction model with uncertainty quantification capability;

[0079] S104. Determine the evaluation cross-sectional data and uncertainty data of the target nuclide based on the nuclear data correction model and physical eigenvectors, and encapsulate the evaluation data of the target nuclide based on the evaluation cross-sectional data and uncertainty data.

[0080] S105. Verify the evaluation data based on particle transport simulation, and optimize the evaluation kernel database based on the verified evaluation data.

[0081] The embodiments of the present invention can utilize the powerful nonlinear mapping capability of deep neural networks to uniformly process multi-source physical data of various nuclides, batch learn and correct complex deviations between theoretical and experimental values, automatically generate evaluation data files in standard format and perform validity verification, thereby improving the efficiency and reliability of neutron nuclear reaction data evaluation and verification.

[0082] As an optional implementation, the multi-dimensional physical characteristic data includes neutron energy data, multi-channel reaction cross-section data, and physical and chemical property data. The neutron energy data includes data points of the target nuclide at different incident neutron energies in the thermal neutron region, resonance region, and fast neutron region. The multi-channel reaction cross-section data includes the total cross-section data of the target nuclide and the cross-section values ​​of different interactions with neutrons, protons, and gamma rays. The physical and chemical property data includes the mass number, atomic number, isotopic abundance information, half-life, excited state energy, energy level density parameter, spin, parity, chemical binding energy, and nuclear binding energy of the target nuclide.

[0083] Specifically, in step S101, the set of target nuclides to be evaluated is first determined. For each nuclide in the set, multi-dimensional physical feature data is collected from the evaluation nuclide database and the experimental database (EXFOR) through an automated interface. To improve the ability of the deep learning model to distinguish the physical properties of different nuclides, the collected data specifically includes the following three aspects:

[0084] 1. Neutron Energy Data (Energy Grid): Acquires data points of nuclides at different incident neutron energies, covering the entire energy range from the thermal neutron region and resonance region to the fast neutron region.

[0085] 2. Multichannel reaction cross sections: In addition to collecting the total cross section, cross section values ​​of the nuclide interacting with different particles such as neutrons, protons, and gamma rays are also collected separately.

[0086] 3. Inherent Nuclide Properties: To enable the model to learn the intrinsic influence of nuclear structure on the reaction cross section, it is also necessary to extract the fundamental physical parameters of the nuclide as static input features, including:

[0087] Basic identifiers: mass number (A), atomic number (Z), and isotopic abundance information;

[0088] Decay and structural information: half-life, excited state energy (Elevel), energy level density parameter (a);

[0089] Quantum mechanical properties: spin (J), parity (π);

[0090] Binding energy information: chemical binding energy and nuclear binding energy (Bn).

[0091] As an optional implementation, the multi-dimensional physical feature data is further subjected to physical perception preprocessing and feature concatenation to obtain the physical feature vector of the target nuclide, which specifically includes:

[0092] S1021. Construct a non-uniform mixed energy grid based on neutron energy data;

[0093] S1022. Interpolate the multi-channel reaction cross-section data based on the logarithmic space interpolation algorithm to obtain discrete experimental cross-section data;

[0094] S1023. Map the discrete experimental cross-section data to a non-uniform mixed energy grid, and use the weighted average method to calculate the weighted average of the experimental cross-sections at each energy point in the full energy range to obtain the dynamic energy characteristics.

[0095] S1024. Perform feature concatenation processing on the dynamic energy characteristics and physical and chemical property data to obtain the physical feature vector.

[0096] Specifically, in step S102, the acquired multi-dimensional physical feature data undergoes unified cleaning and formatting processing:

[0097] 1. Constructing a non-uniform mixed energy grid: To address the characteristics of nuclear reaction cross-sections that change drastically in the resonance region and smoothly in the fast neutron region, a sparse-dense energy grid is constructed based on neutron energy data.

[0098] 2. High-precision interpolation based on physical laws: Utilizing the physical characteristic that the nuclear cross section is approximately linear in the log-log coordinate system, a logarithmic space interpolation algorithm is used to interpolate the multi-channel reaction cross section data to obtain discrete experimental cross section data.

[0099] 3. Weighted Binning of Multi-Source Experimental Data: To address the sparsity and conflict issues in experimental data, discrete experimental points are mapped to mixed grid intervals, and the optimal experimental baseline value at each energy point i is calculated using a weighted averaging method. The calculation formula is as follows:

[0100]

[0101]

[0102] in, Let represent the weighted mean of the experimental cross section calculated at the i-th energy grid point. This represents the total number of experimental data points falling within the i-th energy grid point. This represents the cross-sectional value of the k-th original experimental data point falling within the i-th energy grid point. This represents the weight of the k-th original experimental data point. This represents the measurement error (Error Bar) at the k-th experimental point. Represents a minimal constant (e.g.) This is used to prevent the denominator from being zero when the error is zero.

[0103] 4. Physical feature concatenation: The dynamic energy features are determined by the weighted average of the experimental cross-sections of each energy point obtained in the previous steps. The static nuclide physical properties and the dynamic energy features are concatenated into vectors to form the high-dimensional input layer vector of the deep neural network, i.e., the physical feature vector.

[0104] As an optional implementation, a kernel data correction model with uncertainty quantification capability is constructed and trained, specifically including:

[0105] S1031, A surrogate model for simulating the mapping relationship between cross sections and multiplication factors, trained based on a lightweight multilayer perceptron;

[0106] S1032. Construct a deep neural network for predicting the cross-sectional distribution of the full-range region, and obtain the physical feature samples, cross-sectional distribution labels, and true proliferation factors of the sample nuclides.

[0107] S1033. Input the physical feature samples into a deep neural network to obtain the cross-sectional predicted distribution of the sample nuclides;

[0108] S1034. Input the cross-sectional predicted distribution into the surrogate model to obtain the predicted proliferation factor;

[0109] S1035. Determine the total loss value based on the cross-sectional predicted distribution, cross-sectional distribution label, predicted proliferation factor, actual proliferation factor, and preset mixed loss function;

[0110] S1036. Optimize the deep neural network based on the total loss value to obtain the kernel data correction model.

[0111] Specifically, step S103 is used to construct and train a general kernel data correction deep neural network with uncertainty quantification capability, and the specific process is as follows:

[0112] 1. Introduce a surrogate model to accelerate training: Pre-train a lightweight multilayer perceptron (MLP) as a surrogate model. This is used to simulate the mapping relationship between the cross section and the proliferation factor.

[0113]

[0114] in, This represents the macroscopically effective multiplication factor rapidly predicted by the surrogate model, used to evaluate the physical plausibility of the current cross-section in real time during the training loop. This represents a pre-trained surrogate model mapping function that fits the nonlinear relationship from a high-dimensional cross-sectional space to a one-dimensional keff value. This represents the full-range prediction cross-section vector currently generated by the deep learning model, whose dimension is the same as the size of the energy grid Egrid (e.g., a 1×1000 vector).

[0115] 2. A deep neural network is constructed to predict the cross-sectional distribution of the full-range region. Then, physical feature samples, cross-sectional distribution labels, and true proliferation factors of the sample nuclides are obtained. The physical feature samples of the sample nuclides are obtained through steps similar to those described above for obtaining physical feature vectors, and will not be elaborated here. The cross-sectional distribution labels and true proliferation factors are determined through experiments and manual annotation. The physical feature samples, cross-sectional distribution labels, and true proliferation factors of the sample nuclides will be used as training samples for the deep neural network.

[0116] 3. Input the physical feature samples into a deep neural network to obtain the cross-sectional predicted distribution of the sample nuclides. The cross-sectional prediction distribution includes the predicted cross-sectional values ​​of the sample nuclides at each energy point in the full-energy region.

[0117] 4. Input the cross-sectional predicted distribution into the surrogate model trained in the previous steps to obtain the corresponding predicted proliferation factor.

[0118] 5. Determine the total loss value based on the cross-sectional predicted distribution, cross-sectional distribution label, predicted proliferation factor, actual proliferation factor, and preset hybrid loss function. Optimize the deep neural network based on the total loss value to complete one round of iterative training. When the number of iterations reaches a preset threshold, or the total loss value is lower than a preset threshold, stop training to obtain the kernel data correction model.

[0119] As an optional implementation, the hybrid loss function is:

[0120]

[0121] in, This represents the total loss value. This represents the negative log-likelihood loss between the predicted cross-sectional distribution and the cross-sectional distribution label. This represents the mean square error between the predicted and actual proliferation factors. Represents the smoothing constraint term. Represents the uncertainty constraint term. , , as well as All of these represent optimizable weight coefficients.

[0122] Specifically, This is a micro-fit term (negative log-likelihood loss), used to force the model to predict the distribution. Fit the experimental baseline values ​​calculated in the preceding steps. , For macroscopic physical constraints, defined as , The true effective proliferation factor of the baseline experiment, To smooth the constraint term, Uncertainty constraint terms (such as KL divergence) are used to prevent the uncertainty estimate from being too large or too small.

[0123] In some alternative embodiments, a phased training strategy can be used to train the kernel data correction model, such as a strategy combining "warm-up" and "full training," first optimizing... Item, later added The item was fine-tuned.

[0124] As a further optional implementation, the evaluation cross-sectional data and uncertainty data of the target nuclide are determined based on the nuclear data correction model and physical eigenvectors, and the evaluation data of the target nuclide is obtained by encapsulating the evaluation cross-sectional data and uncertainty data, specifically including:

[0125] S1041. Input the physical feature vector into the nuclear data correction model to obtain the target cross-sectional distribution data of the target nuclide;

[0126] S1042. Perform Monte Carlo sampling on the target cross-section distribution data to obtain the cross-section prediction set of the target nuclide at each energy point in the full energy range;

[0127] S1043. Calculate the mean cross-section value and corresponding uncertainty of the target nuclide at each energy point in the full energy range based on the cross-section prediction set;

[0128] S1044. Generate evaluation cross-sectional data based on the cross-sectional mean of the target nuclide at each energy point in the full energy range, and generate corresponding uncertainty data based on the uncertainty.

[0129] S1045. Write the evaluation section data and uncertainty data into the preset kernel data file format to obtain the evaluation data.

[0130] Specifically, in step S104, the trained nuclear data correction model is used to perform batch inference, reconstruction and encapsulation of the full-energy cross-sectional data of all nuclides in the target nuclide set.

[0131] 1. Probability Sampling and Statistical Extraction: The physical feature vector is input into the nuclear data correction model to obtain the target cross-sectional distribution data of the target nuclide. M forward sampling operations are performed on the target cross-sectional distribution data to obtain the cross-sectional prediction set at the i-th energy point, and the recommended mean is calculated. and uncertainty :

[0132]

[0133]

[0134] in, This represents the total number of Monte Carlo samplings. This represents the predicted cross-sectional value at the i-th energy point in the j-th sampling.

[0135] 2. Standardized Format Encapsulation: An automated format conversion module reads the original evaluation file template and encapsulates the model-corrected evaluation cross-sectional data (mean). ) and uncertainty data (standard deviation) Write the data into a nuclear data file format that conforms to industry standards (such as ENDF-6 format) and automatically generate new evaluation data for multiple target nuclides.

[0136] As a further optional implementation, the evaluation data is validated based on particle transport simulation, and the evaluation kernel database is optimized based on the validated evaluation data. This specifically includes:

[0137] S1051. Based on the evaluation data, a particle transport simulation experiment was conducted on the target nuclide to obtain the experimental measurement cross-sectional values ​​of the target nuclide at each energy point in the full energy range, as well as the macroscopic physical quantities of the target nuclide.

[0138] S1052. Calculate the chi-square value per unit degree of freedom based on the evaluation section data and the experimental measurement section value, and perform micro-statistical verification on the evaluation data based on the degree of closeness of the chi-square value per unit degree of freedom to 1.

[0139] S1053. Perform macroscopic integral verification of the evaluation data based on the difference between the macroscopic physical quantity and the reference true value of the target nuclide;

[0140] S1054. When both micro-statistical verification and macro-integral verification pass, update the evaluation kernel database based on the evaluation data.

[0141] S1055. When the micro-statistical verification or macro-integral verification fails, the kernel data correction model shall be optimized and fine-tuned.

[0142] Specifically, step S105 performs closed-loop verification of the evaluation data of the target nuclide obtained in the aforementioned steps based on particle transport simulation.

[0143] 1. Based on the new evaluation data obtained from the preceding steps, conduct particle transport simulation experiments on the target nuclide to obtain the experimental measurement cross-sectional values ​​of the target nuclide at each energy point in the full energy range, as well as the macroscopic physical quantities of the target nuclide.

[0144] 2. Micro-statistical validation: Calculate the chi-square value per unit degree of freedom of the new evaluation data relative to the experimental measurement cross-sectional values. The formula is as follows:

[0145]

[0146] Where N represents the total number of experimental data points, Indicates the number of free parameters in the model. Indicates the evaluation cross-sectional value. This indicates the experimentally measured cross-sectional value. This represents the total error of the experimental measurement.

[0147] Result criterion: Chi-square value per unit degree of freedom The closer the value is to 1, the better the data consistency; a significant deviation from 1 may indicate that the model needs correction or that there are anomalies in the experimental data. When the chi-square value per unit degree of freedom... When the absolute difference from 1 is less than or equal to a preset threshold (such as 0.1), the microstatistical verification is considered successful.

[0148] 3. Macroscopic Integral Verification: Compare the C / E ratio (calculated / experimental) of the macroscopic physical quantities (such as reaction rate, energy spectrum distribution, critical value, etc.) obtained from the simulation with the reference true values ​​of the target nuclide. The C / E ratio should be close to 1 (usually required to be in the range of 0.95 to 1.05). Excessive deviation indicates that the correction of microscopic data has not been effectively transferred to the macroscopic level. When the C / E ratio is in the range of 0.95 to 1.05, the macroscopic integral verification is considered successful.

[0149] 4. By combining micro-statistical indicators and macro-integral indicators, the effectiveness of data correction is verified. Micro-statistical indicators To ensure the accuracy of local data, the macroscopic C / E value verifies the reliability of the overall physical process. If both the microscopic statistical verification and the macroscopic integral verification pass, the data correction scheme can be confirmed as effective, and the evaluation kernel database can be updated based on the evaluation data. If the microscopic statistical verification or the macroscopic integral verification fails, the kernel data correction model needs to be optimized and fine-tuned.

[0150] The method steps of the embodiments of the present invention have been described above. It can be understood that the embodiments of the present invention can utilize the powerful nonlinear mapping capability of deep neural networks to uniformly process multi-source physical data of various nuclides, batch learn and correct complex deviations between theoretical and experimental values, automatically generate evaluation data files in standard format and perform validity verification, thereby improving the efficiency and reliability of neutron nuclear reaction data evaluation and verification.

[0151] Compared with the prior art, the embodiments of the present invention also have the following advantages:

[0152] 1. By employing a feature input that integrates nuclide physical properties (such as spin and parity) with a hybrid energy grid strategy, this invention achieves significantly improved evaluation accuracy and physical consistency. Compared to traditional methods based solely on numerical fitting, this invention uses the nuclide's microscopic physical parameters (atomic number, mass number, quantum number, etc.) as prior knowledge input to the model, coupled with a hybrid grid technique that refines the mesh for the resonance region. This allows the model to "understand" the physical nature of nuclear reactions, thereby more accurately capturing the intense cross-sectional fluctuations in the resonance energy region and significantly reducing non-physical oscillations and errors.

[0153] 2. The introduction of a surrogate model for acceleration and a multi-nucleus batch processing mechanism brings an exponential improvement in evaluation efficiency. Existing technologies heavily rely on manual evaluation one by one, and macroscopic verification calculations are extremely time-consuming. This invention uses a lightweight surrogate model to replace the time-consuming Monte Carlo simulation for real-time feedback during training, improving computational speed by several orders of magnitude. Simultaneously, a unified deep learning framework supports parallel processing of multiple nuclides, achieving a fundamental shift from a "manual workshop" to an "intelligent pipeline" evaluation process.

[0154] 3. The use of a deep learning architecture with quantifiable uncertainty capabilities brings advantages in data security and assessable confidence levels. This invention overcomes the limitation of traditional deterministic models that can only provide a single predicted value. Through probabilistic neural networks or ensemble sampling techniques, this invention provides accompanying uncertainty (standard deviation or covariance data) along with the recommended cross-sectional value. This provides indispensable quantitative evidence for safety margin analysis in nuclear engineering design, making the generated data more valuable for engineering applications.

[0155] 4. By constructing a multi-objective physical verification closed loop that includes microscopic, macroscopic, and smoothness constraints, this invention offers the advantages of highly self-consistent and reliable data. This invention not only focuses on fitting microscopic experimental data but also innovatively designs a hybrid loss function that includes macroscopic physical constraints (Lmacro) and smoothness constraints (Lsmooth). This ensures that the corrected kernel data not only mathematically fits the EXFOR data but also accurately reproduces the critical benchmark experimental results (e.g., keff≈1.0) in macroscopic physics, with smooth curves conforming to physical laws, thus completely solving the industry pain point of the disconnect between evaluation data and application verification.

[0156] 5. The use of automated standard format packaging technology brings the advantage of good compatibility with downstream applications. This invention integrates an automatic formatting module that can directly write the cross-sectional mean and uncertainty data output by the model into the nuclear industry's commonly used ENDF-6 standard format file, eliminating the need for manual input, reducing the risk of human error, and the generated database can be directly read and used by mainstream nuclear software such as NJOY.

[0157] Reference Figure 2 This invention provides a deep learning-based multi-nucleon neutron nuclear reaction data verification device, comprising:

[0158] The data acquisition module is used to determine the target nuclide to be evaluated and to acquire multi-dimensional physical characteristic data of the target nuclide from the preset evaluation nuclide database and experimental database.

[0159] The data preprocessing module is used to perform physical perception preprocessing and feature stitching on multi-dimensional physical feature data to obtain the physical feature vector of the target nuclide;

[0160] The model training module is used to build and train a kernel data correction model with uncertainty quantification capability;

[0161] The data correction module is used to determine the evaluation cross-sectional data and uncertainty data of the target nuclide based on the nuclear data correction model and physical feature vectors, and to encapsulate the evaluation data of the target nuclide based on the evaluation cross-sectional data and uncertainty data.

[0162] The data validation module is used to validate the evaluation data based on particle transport simulation and optimize the evaluation kernel database based on the validated evaluation data.

[0163] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0164] Reference Figure 3 This invention provides an electronic device, comprising:

[0165] At least one processor;

[0166] At least one memory for storing at least one program;

[0167] When the above-mentioned at least one program is executed by the above-mentioned at least one processor, the above-mentioned at least one processor implements the above-mentioned deep learning-based multi-nucleon nuclear reaction data verification method.

[0168] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0169] This invention also provides a computer-readable storage medium storing a processor-executable computer program that, when executed by a processor, implements the aforementioned deep learning-based method for verifying multi-nucleon nuclear reaction data.

[0170] This invention provides a computer-readable storage medium that can execute a deep learning-based multi-nucleon nuclear reaction data verification method provided in the method embodiments of this invention. It can execute any combination of the implementation steps of the method embodiments and has the corresponding functions and beneficial effects of the method.

[0171] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned deep learning-based method for verifying multi-nucleon nuclear reaction data.

[0172] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0173] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0174] The embodiments described in this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0175] The terms "first," "second," "third," "fourth," etc. (if present) in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0176] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the aforementioned blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0177] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the aforementioned functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0178] If the aforementioned functions are implemented as software functional units 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 invention, 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 invention. 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.

[0179] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0180] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the aforementioned program can be printed, because the aforementioned program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0181] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0182] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, 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.

[0183] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0184] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for verifying multi-nucleoside neutron nuclear reaction data based on deep learning, characterized in that, Includes the following steps: The target nuclide to be evaluated is determined, and multi-dimensional physical characteristic data of the target nuclide are obtained from the preset evaluation nuclide database and experimental database; The multi-dimensional physical feature data is subjected to physical perception preprocessing and feature concatenation to obtain the physical feature vector of the target nuclide; Construct and train a kernel data correction model with uncertainty quantification capability; The evaluation cross-sectional data and uncertainty data of the target nuclide are determined based on the nuclear data correction model and the physical feature vector, and the evaluation data of the target nuclide is obtained by encapsulating the evaluation cross-sectional data and the uncertainty data. The evaluation data is verified based on particle transport simulation, and the evaluation kernel database is optimized based on the verified evaluation data. The construction and training of the kernel data correction model with uncertainty quantification capability specifically includes: A surrogate model for simulating the mapping relationship between cross sections and multiplication factors is trained based on a lightweight multilayer perceptron. A deep neural network was constructed to predict the cross-sectional distribution of the full-range region, and physical feature samples, cross-sectional distribution labels, and true proliferation factors of the sample nuclides were obtained. The physical feature samples are input into the deep neural network to obtain the cross-sectional predicted distribution of the sample nuclides; The predicted cross-sectional distribution is input into the surrogate model to obtain the predicted proliferation factor; The total loss value is determined based on the predicted cross-sectional distribution, the cross-sectional distribution label, the predicted proliferation factor, the actual proliferation factor, and a preset hybrid loss function. The deep neural network is optimized based on the total loss value to obtain the kernel data correction model.

2. The method for verifying multi-nucleoside neutron nuclear reaction data based on deep learning according to claim 1, characterized in that: The multi-dimensional physical characteristic data includes neutron energy data, multi-channel reaction cross-section data, and physical and chemical property data. The neutron energy data includes data points of the target nuclide at different incident neutron energies in the thermal neutron region, resonance region, and fast neutron region. The multi-channel reaction cross-section data includes the total cross-section data of the target nuclide and the cross-section values ​​of different interactions with neutrons, protons, and gamma rays. The physical and chemical property data includes the mass number, atomic number, isotopic abundance information, half-life, excited state energy, energy level density parameter, spin, parity, chemical binding energy, and nuclear binding energy of the target nuclide.

3. The method for verifying multi-nucleoside neutron nuclear reaction data based on deep learning according to claim 2, characterized in that, The step of performing physical perception preprocessing and feature concatenation on the multi-dimensional physical feature data to obtain the physical feature vector of the target nuclide specifically includes: A non-uniform mixed energy grid is constructed based on the neutron energy data; The multi-channel reaction cross-section data are interpolated using a logarithmic space interpolation algorithm to obtain discrete experimental cross-section data. The discrete experimental cross-section data is mapped onto the non-uniform mixed energy grid, and the weighted average method is used to calculate the weighted average of the experimental cross-sections at each energy point in the full energy range to obtain the dynamic energy characteristics. The dynamic energy features are concatenated with the physical and chemical property data to obtain the physical feature vector.

4. The method for verifying multi-nucleoside neutron nuclear reaction data based on deep learning according to claim 1, characterized in that, The hybrid loss function is: in, This represents the total loss value. This represents the negative log-likelihood loss between the predicted cross-sectional distribution and the cross-sectional distribution label. This represents the mean square error between the predicted and actual proliferation factors. Represents the smoothing constraint term. Represents the uncertainty constraint term. , , as well as All of these represent optimizable weight coefficients.

5. A method for verifying multi-nucleoside neutron nuclear reaction data based on deep learning according to any one of claims 1 to 4, characterized in that, The step of determining the evaluation cross-sectional data and uncertainty data of the target nuclide based on the nuclear data correction model and the physical feature vector, and encapsulating the evaluation data of the target nuclide based on the evaluation cross-sectional data and the uncertainty data, specifically includes: The physical feature vector is input into the nuclear data correction model to obtain the target cross-sectional distribution data of the target nuclide; Monte Carlo sampling is performed on the target cross-section distribution data to obtain a set of predicted cross-sections of the target nuclide at each energy point in the full energy range; Calculate the cross-sectional mean and corresponding uncertainty of the target nuclide at each energy point in the full energy range based on the cross-sectional prediction set; The evaluation cross-sectional data is generated based on the cross-sectional mean of the target nuclide at each energy point in the full energy range, and the corresponding uncertainty data is generated based on the uncertainty. The evaluation cross-sectional data and the uncertainty data are written into a preset kernel data file format to obtain the evaluation data.

6. The method for verifying multi-nucleoside neutron nuclear reaction data based on deep learning according to claim 5, characterized in that, The process of validating the evaluation data based on particle transport simulation and optimizing the evaluation kernel database based on the validated evaluation data specifically includes: Based on the evaluation data, a particle transport simulation experiment was conducted on the target nuclide to obtain the experimental measurement cross-sectional values ​​of the target nuclide at each energy point in the full energy range, as well as the macroscopic physical quantities of the target nuclide. The chi-square value per unit degree of freedom is calculated based on the evaluation cross-sectional data and the experimental measurement cross-sectional values, and the evaluation data is verified by micro-statistics based on the degree of closeness of the chi-square value per unit degree of freedom to 1. The evaluation data are verified by macroscopic integration based on the difference between the macroscopic physical quantity and the reference true value of the target nuclide; When both the micro-statistical verification and the macro-integral verification pass, the evaluation kernel database is updated based on the evaluation data. If the micro-statistical verification or macro-integral verification fails, the kernel data correction model is optimized and fine-tuned.

7. A data verification device for multi-nucleoside neutron nuclear reactions based on deep learning, characterized in that, include: The data acquisition module is used to determine the target nuclide to be evaluated and to acquire multi-dimensional physical characteristic data of the target nuclide from a preset evaluation nuclide database and an experimental database. The data preprocessing module is used to perform physical perception preprocessing and feature splicing on the multi-dimensional physical feature data to obtain the physical feature vector of the target nuclide. The model training module is used to build and train a kernel data correction model with uncertainty quantification capability; The data correction module is used to determine the evaluation cross-sectional data and uncertainty data of the target nuclide based on the nuclear data correction model and the physical feature vector, and to encapsulate the evaluation data of the target nuclide based on the evaluation cross-sectional data and the uncertainty data. The data verification module is used to verify the evaluation data based on particle transport simulation, and to optimize the evaluation kernel database based on the verified evaluation data. The model training module is specifically used for: A surrogate model for simulating the mapping relationship between cross sections and multiplication factors is trained based on a lightweight multilayer perceptron. A deep neural network was constructed to predict the cross-sectional distribution of the full-range region, and physical feature samples, cross-sectional distribution labels, and true proliferation factors of the sample nuclides were obtained. The physical feature samples are input into the deep neural network to obtain the cross-sectional predicted distribution of the sample nuclides; The predicted cross-sectional distribution is input into the surrogate model to obtain the predicted proliferation factor; The total loss value is determined based on the predicted cross-sectional distribution, the cross-sectional distribution label, the predicted proliferation factor, the actual proliferation factor, and a preset hybrid loss function. The deep neural network is optimized based on the total loss value to obtain the kernel data correction model.

8. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a deep learning-based method for verifying multi-nucleon nuclear reaction data as described in any one of claims 1 to 6.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements a deep learning-based method for verifying multi-nucleon nuclear reaction data as described in any one of claims 1 to 6.

Images