Neural network-based hybrid energy spectrum reactor neutron solving method and system

Abstract

The invention provides a hybrid energy spectrum reactor neutron solving method and system based on a neural network, and relates to the technical field of hybrid energy spectrum reactors, and the method comprises the steps: obtaining spatial-temporal characteristic data of a material in different energy spectrum regions; and inputting the spatial-temporal characteristic data into a pre-trained deep learning model, and predicting to obtain coupling response data and a neutron diffusion mode between the thermal spectrum region and the fast spectrum region. According to the embodiment of the invention, a deep learning model combining two deep neural networks is utilized, the limitation of a traditional numerical simulation method is broken through, and the method has relatively high spatial-temporal feature capturing capability, nonlinear mapping capability and generalization capability; the method has a remarkable technical effect in the aspect of accurately predicting the coupling response and the neutron diffusion mode in the mixed energy spectrum reactor, the calculation complexity and the experiment dependence are effectively reduced, and the application universality and the real-time prediction capability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hybrid energy spectrum reactors, and in particular to a hybrid energy spectrum reactor neutron solution method and system based on a neural network. Background Art

[0002] With the continuous development of nuclear energy technology, mixed energy spectrum reactors have gradually become a key research direction. Such reactors have the advantages of thermal neutrons and fast neutrons. However, in actual operation, there is a complex neutron correlation between the thermal spectrum and the fast spectrum, which leads to uneven power distribution and fuel utilization, and the thermodynamic performance of the core is limited. In addition, in the mixed energy spectrum reaction of the reactor, the diffusion law of materials (such as fission materials and fertile materials) also directly affects the safety and efficiency of the reactor reaction.

[0003] Most existing technologies use numerical methods to solve the mixed energy spectrum reactor, which has high computational complexity and is difficult to predict in real time. In particular, Monte Carlo simulation can accurately simulate the coupling response in the reactor, but because it needs to process complex equations in multi-dimensional space, the amount of calculation is extremely large and it is difficult to make real-time predictions. For actual reactor design and optimization needs, this high computational complexity leads to a long calculation cycle and cannot quickly respond to design adjustments or parameter optimization. However, the state changes during the operation of the reactor are dynamic, and the existing numerical simulation requires repeated large-scale calculations, which is time-consuming and has poor real-time performance. Summary of the invention

[0004] To solve the above problems, an embodiment of the present invention provides a hybrid energy spectrum reactor neutron solution method based on a neural network, including: acquiring the spatiotemporal characteristic data of the material in different energy spectrum regions; inputting the spatiotemporal characteristic data into a pre-trained deep learning model to predict the coupling response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern; the deep learning model includes a first deep neural network and a second deep neural network; the first deep neural network is used to extract spatial features based on the spatiotemporal characteristic data, and the second deep neural network is used to extract temporal features based on the spatial features and the spatiotemporal characteristic data, and predict the coupling response data and the neutron diffusion pattern.

[0005] The neural network-based hybrid energy spectrum reactor neutron solution method provided in the embodiment of the present invention breaks through the limitations of traditional numerical simulation methods, has strong spatiotemporal feature capture capabilities, nonlinear mapping capabilities and generalization capabilities, and shows significant technical effects in accurately predicting coupled responses and neutron diffusion patterns in hybrid energy spectrum reactors. It also effectively reduces computational complexity and experimental dependence, and improves the breadth of application and real-time prediction capabilities.

[0006] Optionally, the first deep neural network is a convolutional neural network, and the second deep neural network is a long short-term memory network.

[0007] In the embodiment of the present invention, a CNN-LSTM combined model is adopted, in which CNN is responsible for extracting the spatial distribution pattern in the reactor, and LSTM is responsible for processing the evolution of the connection response and diffusion pattern at different time points in the time series.

[0008] Optionally, the spatiotemporal characteristic data include distribution data of temperature, pressure, and neutron density in different energy spectrum regions, and data recording changes of the distribution data over time.

[0009] In the embodiment of the present invention, the training model and the model-based prediction are performed using the spatial characteristic data and time series characteristic data of the material in different energy spectrum regions, so as to achieve efficient capture and comprehensive modeling of spatiotemporal characteristics, with strong generalization ability and wide adaptability.

[0010] Optionally, the method also includes: acquiring multidimensional physical quantity data through a reactor physics simulation tool, the multidimensional physical quantity data including: distribution data of temperature, pressure, and neutron density in the thermal spectrum and fast spectrum, and data recording the changes of the distribution data over time; preprocessing the multidimensional physical quantity data; inputting the preprocessed data into the deep learning model for training, adjusting the model parameters based on minimizing the loss function until the loss function drops to a set threshold, and obtaining a trained deep learning model.

[0011] In the embodiment of the present invention, training data is obtained based on existing simulation tools, and training is performed based on the existing simulation tools until the model is trained. The coupled response and diffusion pattern of any material under quasi-critical conditions can be predicted by inputting the element characteristic curve of the material, and there is no need to rebuild a new numerical model.

[0012] Optionally, the preprocessing of the multidimensional physical quantity data includes: standardizing or normalizing the multidimensional physical quantity data; and representing the processed data as a multidimensional space-time tensor.

[0013] In the embodiment of the present invention, the standardization or normalization processing ensures that each physical quantity is within the same numerical range, avoiding the uneven impact of different physical quantities on model training. The representation as a multi-dimensional space-time tensor can separate the space-time features, making it easier to extract space and time features separately.

[0014] Optionally, the inputting the preprocessed data into the deep learning model for training includes: inputting the spatial tensor of the multidimensional space-time tensor into a convolutional neural network to output spatial features; converting the spatial features into time series features, and inputting the time series features into a long short-term memory network to output coupled response data; calculating a loss function based on the coupled response data and real response data; and iteratively optimizing until the loss function drops to a set threshold, completing the training of the deep learning model.

[0015] The embodiment of the present invention provides a specific training method for the deep learning model, which can effectively fit the coupling response of the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern.

[0016] Optionally, the method also includes: acquiring multidimensional physical quantity data accumulated during the operation of the reactor, and training and updating the deep learning model based on the accumulated multidimensional physical quantity data.

[0017] After the network is trained, the model of the embodiment of the present invention can process input data in real time, is suitable for real-time monitoring and prediction of the operating status in the reactor, and can realize online learning and continuous optimization.

[0018] Optionally, the method further includes: using a cross-validation method to divide the multidimensional physical quantity data into a training set, a validation set and a test set; the validation set is used to monitor the performance of the model during the training process, and the test set is used to evaluate the performance of the model on unseen data.

[0019] The embodiment of the present invention defines that the training data acquired by the simulation tool is divided into a training set, a validation set and a test set, which can verify and evaluate the training effect of the model.

[0020] An embodiment of the present invention provides a hybrid energy spectrum reactor neutron solution system based on a neural network, including: a data acquisition module, used to obtain the spatiotemporal characteristic data of materials in different energy spectrum regions; a prediction module, used to input the spatiotemporal characteristic data into a pre-trained deep learning model, and predict the coupling response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern; the deep learning model includes a first deep neural network and a second deep neural network; the first deep neural network is used to extract spatial features based on the spatiotemporal characteristic data, and the second deep neural network is used to extract temporal features based on the spatial features and the spatiotemporal characteristic data, and predict the coupling response data and the neutron diffusion pattern.

[0021] Optionally, the first deep neural network is a convolutional neural network, and the second deep neural network is a long short-term memory network.

[0022] The hybrid energy spectrum reactor neutron solution system based on neural network in the embodiment of the present invention can achieve the same technical effect as the hybrid energy spectrum reactor neutron solution method based on neural network mentioned above. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0024] Figure 1 A schematic flow chart of a hybrid energy spectrum reactor neutron solution method based on a neural network provided in an embodiment of the present invention; Figure 2 It is a schematic diagram of the CNN-LSTM combined model in an embodiment of the present invention; Figure 3 A schematic structural diagram of a hybrid energy spectrum reactor neutron solution system based on a neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0026] The embodiment of the present invention utilizes the efficient computing power of a deep neural network. After a one-time training, it can quickly perform forward propagation after inputting material properties, and output coupling response and diffusion mode prediction results in real time, which greatly reduces the computing time.

[0027] The deep learning model in this embodiment adopts a combination of two deep neural networks. Exemplarily, a model combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) is adopted. CNN is suitable for processing spatial features. For data in a mixed energy spectrum reactor, such as neutron density distribution, which has significant spatial correlation, CNN can effectively capture these local features in space, and it is responsible for extracting the pattern of spatial distribution (such as neutron density field) in the reactor. LSTM is suitable for mixed data with processing time dependence. For a mixed energy spectrum reactor, the relationship response between the thermal spectrum region and the fast spectrum region not only has spatial distribution characteristics, but also changes dynamically over time. LSTM can effectively capture this information. It is responsible for processing the evolution of connection responses and diffusion patterns at different time points in the time series.

[0028] Figure 1 A schematic flow chart of a hybrid energy spectrum reactor neutron solution method based on a neural network provided by an embodiment of the present invention is shown, and the method comprises: S102, obtaining the spatiotemporal characteristic data of the material in different energy spectrum regions.

[0029] Among them, materials refer to fission materials, breeding materials, etc. in the reactor, and the spatiotemporal characteristic data include the distribution data of temperature, pressure, and neutron density of the above materials in different energy spectrum regions (such as thermal spectrum region and fast spectrum region), as well as data recording the change of the above distribution data over time. The above spatiotemporal characteristic data can be data acquired in real time during the operation of the monitoring mixed energy spectrum reactor, and the pre-trained deep learning model can make predictions based on it to obtain coupled response data and neutron diffusion patterns.

[0030] S104, inputting the spatiotemporal feature data into a pre-trained deep learning model to predict the coupling response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern.

[0031] The deep learning model includes a first deep neural network and a second deep neural network. The first deep neural network is used to extract spatial features based on spatiotemporal feature data, and the second deep neural network is used to extract temporal features based on spatial features and the above-mentioned spatiotemporal feature data, and predict the coupling response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern. In this embodiment, the response data and the diffusion pattern are obtained by spatiotemporal feature-classification (mapping). Among them, the neutron diffusion pattern refers to the phenomenon of neutron migration in the medium, which can be determined by the distribution of neutrons.

[0032] Specifically, the first deep neural network may be a convolutional neural network, and the second deep neural network may be a long short-term memory network.

[0033] Exemplarily, the CNN-LSTM neural network architecture is designed as follows: 1.CNN module (for extracting spatial features): Input: Multidimensional tensor (such as neutron density distribution in core space).

[0034] Structure layers: Set up several matrix layers, each using a different matrix structure size, to capture different structure layers.

[0035] Pooling layer: Use maximum pooling or average pooling to reduce the dimension of the feature map and retain the most important spatial information.

[0036] Waveform output: The feature map after the subsequent surface layer will be shallower and serve as the input of the subsequent LSTM module.

[0037] 2.LSTM module (used to extract time features): Input: spatial features from CNN, and other physical quantities in the time series (such as the response change between time steps). The response is the change process of parameters such as neutron density inside the reactor.

[0038] LSTM layer: Set up a multi-layer LSTM network to capture the unit association relationship at different time steps through its memory.

[0039] Dropout layer: prevents the model from being excessive and overfitting to improve the generalization ability of the model.

[0040] Fully connected layer: The spatiotemporal features processed by LSTM are mapped to the output space through a fully connected layer.

[0041] Output layer: Generates prediction results, specifically the correlation response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern.

[0042] 3. Data interaction and flow Spatial feature extraction and flattening: The original input data is processed by CNN to obtain spatial features. After the spatial features are extracted, these features are flattened into a one-dimensional feature vector. This process can convert the output of the convolution layer into the LSTM input format through the Flatten operation. At this stage, CNN converts the spatial distribution data of neutrons into time series features so that the subsequent LSTM can capture time-dependent information.

[0043] Temporal feature capture and output: The LSTM module learns the dynamic changes in time from the flattened features and captures the complex dependencies between time steps through the structure of the recurrent neural network. Ultimately, the output of the LSTM will be passed to the fully connected layer to generate the final prediction result, completing the mapping from space to time and then to the final physical quantity.

[0044] Figure 2 A schematic diagram of a CNN-LSTM combined model provided in an embodiment of the present invention is shown.

[0045] When using deep learning models, simulation tools can be used to obtain training data. Input data can be collected and preprocessed. Specifically, data collection is as follows: through Monte Carlo and other high-precision reactor physics simulation tools, multi-dimensional physical quantity data is obtained, including: the spatial distribution of data such as temperature and neutron flux density in the thermal spectrum and fast spectrum; time series data, recording the response of each physical quantity over time.

[0046] Data preprocessing is as follows: standardize or normalize the multidimensional physical quantity data; represent the processed data as a multidimensional space-time tensor. The first step, standardization / normalization: standardize or normalize the input data to ensure that each physical quantity is within the same numerical range to avoid the uneven impact of different physical quantities on model training. The second step, separation of space-time features: represent the data in the reactor as a multidimensional space-time tensor (such as three-dimensional space plus time) so that CNN and LSTM can extract spatial and temporal features respectively.

[0047] like Figure 2 As exemplified, the neutron density distribution (multi-dimensional tensor) in the core space is input into the CNN module, which outputs spatial features, and the response changes (one-dimensional feature vectors) between time steps are input into the LSTM module, which outputs temporal features.

[0048] Optionally, the training process of the above deep learning model is as follows: First, multi-dimensional physical quantity data are obtained through a reactor physics simulation tool. The multi-dimensional physical quantity data may include: distribution data of temperature, pressure, and neutron density in the thermal spectrum region and the fast spectrum region, and data recording the change of the distribution data over time; Secondly, preprocess the multi-dimensional physical quantity data; Then, the preprocessed data is input into the deep learning model for training, and the model parameters are adjusted based on the minimization of the loss function until the loss function drops to a set threshold, thereby obtaining a trained deep learning model. Specifically, the spatial tensor of the multidimensional spatiotemporal tensor can be input into the convolutional neural network to output spatial features; then, the spatial features are converted into time series features, and the time series features are input into the long short-term memory network to output coupled response data; the loss function is calculated based on the coupled response data and the real response data; after iterative optimization until the loss function drops to a set threshold, the training of the deep learning model is completed.

[0049] The embodiment of the present invention utilizes a deep learning model that combines two deep neural networks, breaks through the limitations of traditional numerical simulation methods, and possesses strong spatiotemporal feature capture capabilities, nonlinear mapping capabilities, and generalization capabilities. It demonstrates significant technical effects in accurately predicting coupled responses and neutron diffusion patterns in mixed energy spectrum reactors, and effectively reduces computational complexity and experimental dependence, thereby enhancing the breadth of application and real-time prediction capabilities.

[0050] In order to effectively fit the coupled response of the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern, the embodiment of the present invention can use the CNN-LSTM combined model for training and optimization. The specific steps are as follows: First, input data CNN branch network input: spatial characteristic data of materials in different energy spectrum regions (such as temperature field, pressure field and neutron density distribution). These features are used as input to the CNN branch network, and the spatial distribution features are extracted through multiple convolution layers.

[0051] LSTM backbone network input: time series characteristic data (such as the response data of the reactor at different time points). LSTM captures the time evolution pattern of the material through its memory units.

[0052] Second, the specific training process Training of CNN module: First, the convolutional layer of the convolutional neural network is trained to capture the spatial local features of the material through different convolution kernels. The output after convolution is flattened and used as the input of the LSTM backbone network.

[0053] Training of LSTM modules: LSTM networks process time series data and capture the temporal dependencies of materials during operation. LSTM learns temporal coupling responses and diffusion patterns by gradually adjusting the states of memory cells.

[0054] Loss function and optimization: The model output is compared with the actual response curve in the experimental data, and the mean square error (MSE) is calculated as the loss function. A back-propagation algorithm (such as the Adam optimizer) is used to adjust the weights of the CNN and LSTM to reduce the prediction error.

[0055] Third, training iteration and convergence With the iterative optimization of the loss function, the network weights gradually converge and the prediction accuracy of the model continues to improve. Batch gradient descent is used in the training process to accelerate convergence and ensure that the model can efficiently capture the coupling patterns in space and time.

[0056] Fourth, model validation Use cross-validation to divide the data into training set, validation set and test set to ensure that the model has good generalization ability. The validation set is used to monitor the performance of the model during the training process to prevent overfitting, while the test set is used to evaluate the performance of the model on unseen data.

[0057] Fifth, model optimization: When the model's loss function drops to the set threshold, the model's training process ends, and the model can accurately predict the coupling response and diffusion pattern of different materials. The model's fitting performance can be further improved by further adjusting hyperparameters, such as convolution kernel size, number of LSTM layers, and learning rate.

[0058] In practical applications, the trained CNN-LSTM model can be deployed in the core computing system to monitor the responses of the thermal spectrum and fast spectrum during operation and predict the diffusion pattern of neutrons in the subsequent time.

[0059] The above CNN-LSTM model can also perform online learning. As new data is accumulated during the operation of the reactor, the model can continue to perform online learning, update the weights of the neural network, and further improve the prediction accuracy.

[0060] The specific application examples based on the above CNN-LSTM model are as follows: First, numerical simulation experiments were carried out. Using Monte Carlo, a reactor physics simulation tool, 12 core power levels, including 0%, 3%, 5%, 10%, 20%, 30%, 50%, 70%, 90%, 95%, 100%, and 105%, were selected as the initial core nuclear power required by the existing high-precision core steady-state calculation software, and the spatial distribution of data such as temperature and neutron flux density in the thermal spectrum and fast spectrum regions, as well as their changes over time, were calculated.

[0061] Secondly, perform deep neural network training and fitting. Input the data into the CNN-LSTM network for network training.

[0062] Then, the trained CNN-LSTM model is deployed in the core computing system, which can monitor the responses of the thermal spectrum and fast spectrum during operation and predict the diffusion pattern of neutrons in the subsequent time.

[0063] The hybrid energy spectrum reactor neutron solution method based on neural network provided in the embodiment of the present invention has the following beneficial effects: First, efficient capture and comprehensive modeling of spatiotemporal features. Most existing technologies rely on single-dimensional models, such as using numerical solution methods to simulate spatial and temporal features respectively. Such methods often have problems of high computational complexity and limited accuracy when dealing with complex coupled responses. The present invention combines the advantages of CNN and LSTM to achieve global modeling, so that the model can accurately reflect the complex coupling relationship between the thermal spectrum region and the fast spectrum region.

[0064] Second, it has strong generalization ability and wide adaptability. Existing models based on physical laws usually need to rely on a large amount of experimental data for tuning, and complex parameter adjustments are required for the characteristics of the materials one by one, so the scope of application is limited. The present invention is trained through a neural network, which can learn and extract the characteristics of different materials in different spectral regions and time periods, so it has a strong generalization ability. Once the model is trained, the coupled response and diffusion pattern of any material under quasi-critical conditions can be predicted by inputting the element characteristic curve of the material, and there is no need to rebuild a new numerical model.

[0065] Third, accurate nonlinear mapping capabilities. In existing numerical simulations and empirical modeling methods, the processing of nonlinear responses often has large errors, especially when the coupled response and diffusion pattern change nonlinearly with material properties and time. Traditional methods have difficulty capturing these complex relationships. The present invention uses the nonlinear mapping capability of the deep learning model. This nonlinear mapping capability enables the model to have higher accuracy in processing complex coupled reactions, especially when there is a complex nonlinear relationship between material properties and responses, and can provide more accurate prediction results than traditional models.

[0066] Fourth, the computational efficiency is high and the real-time prediction capability is strong. The prior art usually needs to rely on complex finite element or differential equation numerical solution methods for simulation, and these methods have large computational complexity, especially when processing spatiotemporal coupling characteristics, and may require long-term iterative calculations, which are difficult to apply in real time. The present invention can complete the calculation in a relatively short time through the forward propagation process of a deep neural network, and generate the prediction results of the coupling response and diffusion pattern. After training through the network, the model can process the input data in real time, is suitable for real-time monitoring and prediction of the operating status in the reactor, and can realize online learning and continuous optimization.

[0067] In summary, the embodiment of the present invention uses a deep neural network, a CNN-LSTM combined model, breaks through the limitations of traditional numerical simulation methods, and has strong spatiotemporal feature capture capabilities, nonlinear mapping capabilities, and generalization capabilities. It shows significant technical effects in accurately predicting the coupling response and neutron diffusion pattern in a mixed energy spectrum reactor, and effectively reduces the computational complexity and experimental dependence, and improves the breadth of application and real-time prediction capabilities.

[0068] Figure 3 The structure diagram of the hybrid energy spectrum reactor neutron solution system based on neural network provided by an embodiment of the present invention is shown, and the system includes: The data acquisition module 301 is used to acquire the spatiotemporal characteristic data of the material in different energy spectrum regions; A prediction module 302 is used to input the spatiotemporal feature data into a pre-trained deep learning model to predict the coupling response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern; the deep learning model includes a first deep neural network and a second deep neural network; The first deep neural network is used to extract spatial features based on the spatiotemporal feature data, and the second deep neural network is used to extract temporal features based on the spatial features and the spatiotemporal feature data, and predict the coupled response data and the neutron diffusion pattern.

[0069] The embodiment of the present invention utilizes a deep learning model that combines two deep neural networks, breaks through the limitations of traditional numerical simulation methods, and possesses strong spatiotemporal feature capture capabilities, nonlinear mapping capabilities, and generalization capabilities. It demonstrates significant technical effects in accurately predicting coupled responses and neutron diffusion patterns in mixed energy spectrum reactors, and effectively reduces computational complexity and experimental dependence, thereby enhancing the breadth of application and real-time prediction capabilities.

[0070] Optionally, the first deep neural network is a convolutional neural network, and the second deep neural network is a long short-term memory network.

[0071] The embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is read and executed by a processor, the method provided in the above embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not described here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0072] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the control device through a computer, and the program can be stored in a computer-readable storage medium. When the program is executed, it may include the processes of the above-mentioned method embodiments, wherein the storage medium may be a memory, a disk, an optical disk, etc.

[0073] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the scope defined by the claims.

[0074] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements. The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0075] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A hybrid spectrum reactor neutron solution method based on neural network, characterized in that: include: Obtain the spatiotemporal characteristic data of materials in different energy spectrum regions; Inputting the spatiotemporal feature data into a pre-trained deep learning model to predict the coupling response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern; the deep learning model includes a first deep neural network and a second deep neural network; The first deep neural network is used to extract spatial features based on the spatiotemporal feature data, and the second deep neural network is used to extract temporal features based on the spatial features and the spatiotemporal feature data, and predict the coupled response data and the neutron diffusion pattern.

2. The method according to claim 1, characterized in that: The first deep neural network is a convolutional neural network, and the second deep neural network is a long short-term memory network.

3. The method according to claim 1, characterized in that The spatiotemporal characteristic data include distribution data of temperature, pressure, and neutron density in different energy spectrum regions, and data recording changes of the distribution data over time.

4. The method according to claim 1, characterized in that: The method further comprises: Acquire multi-dimensional physical quantity data through a reactor physical simulation tool, wherein the multi-dimensional physical quantity data includes: distribution data of temperature, pressure, and neutron density in a thermal spectrum region and a fast spectrum region, and data recording changes of the distribution data over time; Preprocessing the multidimensional physical quantity data; The preprocessed data is input into the deep learning model for training, and the model parameters are adjusted based on minimizing the loss function until the loss function drops to a set threshold, thereby obtaining a trained deep learning model.

5. The method according to claim 4, characterized in that The preprocessing of the multi-dimensional physical quantity data includes: Standardizing or normalizing the multidimensional physical quantity data; Represent the processed data as a multi-dimensional spatiotemporal tensor.

6. The method according to claim 5, characterized in that The step of inputting the preprocessed data into the deep learning model for training includes: Inputting the spatial tensor of the multidimensional space-time tensor into a convolutional neural network and outputting spatial features; Converting the spatial features into time series features, inputting the time series features into a long short-term memory network, and outputting coupled response data; Calculating a loss function based on the coupled response data and the real response data; After iterative optimization until the loss function drops to a set threshold, the training of the deep learning model is completed.

7. The method according to claim 6, characterized in that The method further comprises: Acquire multidimensional physical quantity data accumulated during the operation of the reactor, and train and update the deep learning model based on the accumulated multidimensional physical quantity data.

8. The method according to claim 6, characterized in that The method further comprises: The multidimensional physical quantity data is divided into a training set, a validation set and a test set using a cross-validation method; the validation set is used to monitor the performance of the model during the training process, and the test set is used to evaluate the performance of the model on unseen data.

9. A hybrid spectrum reactor neutron solution system based on neural network, characterized in that: include: A data acquisition module, used to acquire the spatiotemporal characteristic data of materials in different energy spectrum regions; A prediction module, used for inputting the spatiotemporal feature data into a pre-trained deep learning model to predict the coupling response data between the thermal spectrum region and the fast spectrum region and the neutron diffusion pattern; the deep learning model includes a first deep neural network and a second deep neural network; The first deep neural network is used to extract spatial features based on the spatiotemporal feature data, and the second deep neural network is used to extract temporal features based on the spatial features and the spatiotemporal feature data, and predict the coupled response data and the neutron diffusion pattern.

10. The system according to claim 9, characterized in that The first deep neural network is a convolutional neural network, and the second deep neural network is a long short-term memory network.

Images