Three-dimensional radiation field inversion method and device based on graph neural network method

Abstract

The invention relates to a three-dimensional radiation field inversion method and device based on a graph neural network method. Performing field measurement through a radiation detector to obtain radiation field local measurement data of three-dimensional radiation field inversion; performing feature extraction on the radiation field local measurement data by using a feature extractor, and constructing a radiation field graph neural network structure of three-dimensional radiation field inversion; deducing three-dimensional radiation field distribution of three-dimensional radiation field inversion according to the radiation field graph neural network structure; training and optimizing the graph neural network model according to the radiation field local measurement data and the three-dimensional radiation field distribution to obtain a radiation field graph neural network model of three-dimensional radiation field inversion; and performing model prediction adjustment of three-dimensional radiation field inversion according to the actually measured data set and the radiation field graph neural network model. According to the invention, a high-resolution and omnibearing three-dimensional radiation field is recovered from radiation field local measurement data, and the inversion of the distribution of the whole three-dimensional radiation field is realized; and the accuracy and precision of radiation field inversion are improved.

Description

Technical Field

[0001] The present invention relates to the field of radiation field evaluation and visualization technology, and more specifically, to a three-dimensional radiation field inversion method and device based on a graph neural network method. Background Art

[0002] In the field of nuclear engineering, obtaining accurate and real-time three-dimensional radiation field distribution information is essential for maintaining facility safety and effective monitoring. Traditional radiation field measurement methods usually rely on dense sensor networks, which have a series of challenges in cost, layout and real-time performance. With the continuous development of technology, Graph Neural Network (GNN), as an emerging deep learning method, provides a new way to solve these problems.

[0003] Traditional radiation field acquisition methods require a large number of high-density sensor networks to obtain high-resolution data in space. However, in practical applications, such densely arranged sensor networks are often difficult to implement because they involve a large number of equipment deployments and high costs. At the same time, due to the particularity of nuclear facilities, traditional methods may not be able to flexibly deploy sensors in certain environments, limiting the practical feasibility of radiation field measurements. Summary of the invention

[0004] The embodiments of the present application provide a three-dimensional radiation field inversion method and device based on a graph neural network method. In order to have a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary is not a general review, nor is it intended to identify key / important components or describe the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a preface to the detailed description that follows.

[0005] In a first aspect, the present application provides a three-dimensional radiation field inversion method based on a graph neural network method, the method comprising:

[0006] Through on-site measurement with radiation detectors, local measurement data of the radiation field for three-dimensional radiation field inversion is obtained;

[0007] Using a feature extractor to extract features from the local measurement data of the radiation field, and constructing a radiation field graph neural network structure for the three-dimensional radiation field inversion;

[0008] Inferring the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure;

[0009] The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain the radiation field graph neural network model for the three-dimensional radiation field inversion;

[0010] The model prediction adjustment of the three-dimensional radiation field inversion is performed according to the measured data set and the radiation field graph neural network model.

[0011] According to a preferred embodiment, the feature extractor is used to extract features from the local measurement data of the radiation field to construct a radiation field graph neural network structure for three-dimensional radiation field inversion, including:

[0012] Using a graph convolution layer, a residual connection layer, a nonlinear transformation layer, a pooling layer and a fully connected layer to extract features from the local measurement data of the radiation field to obtain global feature data of the three-dimensional radiation field inversion;

[0013] Constructing a radiation field graph neural network structure for three-dimensional radiation field inversion according to the global feature data;

[0014] The graph convolution layer, the residual connection layer, the nonlinear transformation layer, the pooling layer and the fully connected layer are used as the feature extractor.

[0015] According to a preferred embodiment, the inferring the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure includes:

[0016] The adjacency matrix construction operation, feature propagation operation and radiation field reconstruction operation are performed on the radiation field graph neural network structure to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion.

[0017] According to a preferred embodiment, the graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain the radiation field graph neural network model for the three-dimensional radiation field inversion, including:

[0018] The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, the loss metric function and the gradient descent method to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion.

[0019] According to a preferred embodiment, the model prediction adjustment of the three-dimensional radiation field inversion is performed based on the measured data set and the radiation field graph neural network model, including:

[0020] Determining a model prediction result of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model;

[0021] Adjusting the model performance of the three-dimensional radiation field inversion according to the model prediction results and the evaluation index of the actual radiation field distribution;

[0022] Determining the model prediction results and adjusting the model performance are used as model prediction adjustments for the three-dimensional radiation field inversion.

[0023] In a second aspect, the present application provides a three-dimensional radiation field inversion device based on a graph neural network method, the device comprising:

[0024] An acquisition module is used to obtain local measurement data of the radiation field for three-dimensional radiation field inversion through field measurement with a radiation detector;

[0025] A structure construction module, used for extracting features from the local measurement data of the radiation field using a feature extractor, and constructing a radiation field graph neural network structure for the three-dimensional radiation field inversion;

[0026] An inference module, used to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the neural network structure of the radiation field graph;

[0027] A model training and optimization module, used to train and optimize the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, to obtain the radiation field graph neural network model of the three-dimensional radiation field inversion;

[0028] The model prediction adjustment module is used to perform model prediction adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model.

[0029] According to a preferred embodiment, the structural building module is specifically used for:

[0030] Using a graph convolution layer, a residual connection layer, a nonlinear transformation layer, a pooling layer and a fully connected layer to extract features from the local measurement data of the radiation field to obtain global feature data of the three-dimensional radiation field inversion;

[0031] Constructing a radiation field graph neural network structure for three-dimensional radiation field inversion according to the global feature data;

[0032] The graph convolution layer, the residual connection layer, the nonlinear transformation layer, the pooling layer and the fully connected layer are used as the feature extractor.

[0033] According to a preferred embodiment, the inference module is specifically used for:

[0034] The adjacency matrix construction operation, feature propagation operation and radiation field reconstruction operation are performed on the radiation field graph neural network structure to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion.

[0035] According to a preferred embodiment, the model training optimization module is specifically used for:

[0036] The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, the loss metric function and the gradient descent method to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion.

[0037] According to a preferred embodiment, the model prediction adjustment module is specifically used to:

[0038] Determining a model prediction result of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model;

[0039] Adjusting the model performance of the three-dimensional radiation field inversion according to the model prediction results and the evaluation index of the actual radiation field distribution;

[0040] Determining the model prediction results and adjusting the model performance are used as model prediction adjustments for the three-dimensional radiation field inversion.

[0041] In a third aspect, the present application provides a computer storage medium, which stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the above-mentioned method steps.

[0042] In a fourth aspect, the present application provides a terminal, which may include: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the above-mentioned method steps.

[0043] The technical solution provided by this application may have the following beneficial effects:

[0044] In the present application, the three-dimensional radiation field inversion method based on the graph neural network method obtains the local measurement data of the radiation field for the three-dimensional radiation field inversion through field measurement of the radiation detector; uses a feature extractor to extract features from the local measurement data of the radiation field, and constructs the radiation field graph neural network structure for the three-dimensional radiation field inversion; infers the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure; trains and optimizes the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, and obtains the radiation field graph neural network model for the three-dimensional radiation field inversion; and performs model prediction and adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model. The present application makes full use of the graph neural network, and through the synergy of feature extraction and data upsampling, recovers the high-resolution and all-round three-dimensional radiation field from the local measurement data of the radiation field measured by the limited radiation detector, and realizes the inversion of the entire three-dimensional radiation field distribution; the present application processes the local measurement data of the radiation field of the limited measurement in an intelligent way, realizes the intelligent processing of the limited measurement data, and provides a new efficient and accurate method for radiation field monitoring in the field of nuclear engineering. This application has broad application prospects in the field of nuclear engineering. It not only improves the accuracy and precision of radiation field inversion, but also provides an innovative solution for the safety monitoring of nuclear facilities.

[0045] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0047] Figure 1 It is a schematic flow chart of a three-dimensional radiation field inversion method based on a graph neural network method provided in an embodiment of the present application;

[0048] Figure 2 It is a schematic diagram of the change trend of the model loss value during the training process of a graph neural network model of a three-dimensional radiation field inversion method based on a graph neural network method provided in an embodiment of the present application;

[0049] Figure 3 It is a schematic diagram of a randomly distributed measured data set calculated and verified by a radiation field graph neural network model of a three-dimensional radiation field inversion method based on a graph neural network method provided in an embodiment of the present application;

[0050] Figure 4 It is a schematic diagram of the model prediction results of a radiation field graph neural network model of a three-dimensional radiation field inversion method based on a graph neural network method provided in an embodiment of the present application after inputting a measured data set;

[0051] Figure 5 It is a structural schematic diagram of a three-dimensional radiation field inversion device based on a graph neural network method provided in an embodiment of the present application;

[0052] Figure 6 It is a structural diagram of a terminal provided in an embodiment of the present application.

[0053] Reference numerals:

[0054] 10000, acquisition module, 20000, structure construction module, 30000, inference module, 40000, model training optimization module, 50000, model prediction adjustment module;

[0055] 1001. Processor, 1002. Communication bus, 1003. User interface, 1004. Network interface, 1005. Memory. DETAILED DESCRIPTION

[0056] The following description and the drawings sufficiently illustrate specific embodiments of the invention to enable those skilled in the art to practice them.

[0057] It should be clear that the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] When the following description refers to the 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 present invention. Instead, they are only examples of systems and methods consistent with some aspects of the present invention as detailed in the attached claims.

[0059] In the description of the present invention, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In addition, in the description of the present invention, unless otherwise specified, "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previously associated objects are in an "or" relationship.

[0060] The following will be combined with the attached Figure 1 -Attached Figure 4, a three-dimensional radiation field inversion method based on a graph neural network method provided in an embodiment of the present application is introduced in detail.

[0061] See also Figure 1-4 The method of the embodiment of the present application may include the following steps:

[0062] By effectively modeling sensor locations and measurement data, the limitations of traditional methods are overcome. This technology abstracts sensor locations as nodes in a graph, and radiation field measurements as information on the nodes, thus forming a complex graph structure. By learning this graph structure, the graph neural network can infer the three-dimensional radiation field distribution in the entire nuclear facility from a limited number of sensor data, achieving efficient inversion calculations.

[0063] Compared with traditional methods, the three-dimensional radiation field inversion method based on graph neural network has significant advantages. First, this method can achieve high-quality inversion of the three-dimensional radiation field distribution in the entire nuclear facility with a small number of sensor arrangements, reducing the cost of implementation and maintenance. Secondly, through the powerful fitting ability of graph neural network, this method not only adapts to the complex changes in sensor positions, but also improves the accuracy of inversion. Most importantly, the method based on graph neural network has a faster inversion speed, which provides unprecedented possibilities for real-time monitoring and emergency response in nuclear engineering.

[0064] This application breaks the shackles of traditional radiation field acquisition methods by introducing advanced theories of graph neural networks, bringing a more intelligent and efficient solution to the field of nuclear engineering. Through deep learning, graph neural networks can not only cope with the complexity of sensor layout, but also complete radiation field inversion in a short time, providing more reliable technical support for nuclear safety. The above-mentioned sensors include radiation detectors, and the corresponding sensor data include local measurement data of the radiation field.

[0065] The three-dimensional radiation field inversion method based on the graph neural network method described in the present application aims to achieve high-precision and detailed inversion reconstruction of the entire three-dimensional radiation field distribution through the local measurement data of the radiation field of a small number of radiation detectors. The present application adopts the graph neural network as the core algorithm, and realizes the complete inversion process from the local measurement data of the radiation field included in the local information to the three-dimensional radiation field distribution included in the overall scene through the end-to-end learning of the graph neural network. It mainly includes two modules: the feature extraction module and the data upsampling module. The present invention obtains the local measurement data of the three-dimensional radiation field through a limited number of radiation detectors. Considering the high nonlinearity and spatial correlation of the radiation field, the graph neural network GNN is introduced, and the complex three-dimensional radiation field information is effectively processed through the feature extraction module, and the local measurement data of the radiation field is mapped to the global feature space, providing a high-quality radiation field graph neural network structure representation for subsequent inversion. The graph neural network, with its flexible modeling ability for complex relationships, better restores the overall characteristics of the radiation field, that is, the three-dimensional radiation field distribution, from the limited local measurement data of the radiation field. The present application can accurately predict the three-dimensional radiation field (space) distribution, thanks to the graph structure modeling ability of the graph neural network and the effectiveness of the data upsampling module.

[0066] Due to the vanishing gradient problem that may occur in traditional recurrent neural networks due to long-term dependence, the gradient descent method is adopted in the three-dimensional radiation field inversion technology based on graph neural networks in this application. The radiation field graph neural network model of this application fully utilizes the advantages of graph neural networks in spatial information modeling and its adaptability to graph structure data. This application further verifies the significant advantages and practicality of the three-dimensional radiation field inversion method based on graph neural networks proposed in this application by comparing and analyzing the model prediction results of the radiation field graph neural network model with the actual radiation field distribution.

[0067] This application adopts graph neural network technology to overcome the limitations of traditional methods in processing three-dimensional radiation field inversion. It can accurately restore the entire three-dimensional radiation field distribution when there are only limited local measurement data of the three-dimensional radiation field, providing strong technical support for radiation monitoring and emergency response. The radiation field graph neural network model of this application is suitable for application in the field of nuclear engineering, especially in the inversion of three-dimensional radiation field distribution based on limited local measurement data of the radiation field, providing researchers with efficient and high-resolution three-dimensional radiation field visualization and reconstruction technology.

[0068] S100, obtaining local measurement data of the radiation field for three-dimensional radiation field inversion through field measurement with a radiation detector.

[0069] In the embodiment of the present application, the data acquisition module acquires limited (three-dimensional) local measurement data of the radiation field through a small number of precisely arranged radiation detectors, covering local information of the target area.

[0070] For example, there are M radiation detectors, and the local measurement data of the radiation field is expressed as a matrix Where T is the number of time steps, is the set of real numbers.

[0071] S200, using a feature extractor to extract features from the local measurement data of the radiation field, and constructing a radiation field graph neural network structure for the three-dimensional radiation field inversion, including:

[0072] A graph convolution layer, a residual connection layer, a nonlinear transformation layer, a pooling layer and a fully connected layer are used to extract features from the local measurement data of the radiation field to obtain global feature data of the three-dimensional radiation field inversion; a radiation field graph neural network structure of the three-dimensional radiation field inversion is constructed based on the global feature data; the graph convolution layer, the residual connection layer, the nonlinear transformation layer, the pooling layer and the fully connected layer are used as the feature extractor.

[0073] In the embodiment of the present application, the feature extraction module uses a graph neural network to learn and extract features from the local measurement data of the radiation field, capture the spatial structure and mutual relationship of the radiation field, and ensure high-quality representation of the radiation field features. The radiation field graph neural network structure represents the spatial structure and mutual relationship of the radiation field, also known as the graph structure.

[0074] Specifically, each node represents a radiation detector, and the edge represents the relationship between adjacent radiation detectors. The graph convolutional neural network (GCN) structure is introduced. By superimposing multiple layers of graph convolutional neural networks, feature information is learned from local to global to generate global feature data representation. Suppose the parameters of the graph convolutional neural network GCN are θ GCN , the feature extraction process is expressed as:

[0075] H global =GCN(D measured ,θ GCN )

[0076] Among them, H global It is the global feature data.

[0077] Furthermore, the fully connected layer MLP is used to extract the global feature data to obtain the final global feature data representation:

[0078] H global =MLP(H global )

[0079] A new feature extractor is designed, called Graph Residual Network, or GRNets for short. The feature extraction module corresponding to this feature extractor combines graph convolutional neural network with residual connected ResNet to design a new graph neural network model. This graph neural network model can not only effectively process local measurement data of radiation fields with graph structure, but also utilize the ResNet characteristics of residual connections to achieve deeper network learning.

[0080] More specifically, the feature extractor consists of two parts: a graph convolutional layer and a residual block. The graph convolutional layer is the graph convolutional neural network mentioned above, and the residual block is the ResNet with residual connection mentioned above. The graph convolutional layer is used to extract node features, and the residual block is used to implement jump connections to improve the expressiveness of the graph neural network model. This new graph neural network model is the feature extractor, and the basic structural network layer design of the feature extractor includes:

[0081] Graph convolution layer: This layer performs convolution operation on the input graph structure data, and uses the graph convolution layer to extract features from the input graph structure data to capture the node and edge information in the graph structure data. Specifically, the graph convolution neural network uses the neighbor node information on the graph structure data to update the local feature representation of each node. In this way, the graph convolution neural network can learn and extract useful local features from the graph structure data; the graph structure data is the local measurement data of the radiation field.

[0082] Residual connection layer: The residual network is used to further process the local features of the output of the graph convolution layer through residual connection, that is, more original graph structure data is retained through skip connections. This residual connection method helps to alleviate the gradient vanishing problem, allowing the three-dimensional radiation field distribution to perform deeper learning on the graph neural network model;

[0083] Nonlinear transformation layer: This layer usually uses nonlinear activation functions such as ReLU to increase the expressiveness of the three-dimensional radiation field distribution to the graph neural network model. The nonlinear transformation layer can introduce more nonlinear characteristics into the graph neural network model, so that the graph neural network model can better process complex graph structure data;

[0084] Pooling layer / fully connected layer: Perform global average pooling on the local features of each channel. The pooling layer can effectively reduce the data dimension and reduce the amount of calculation. The fully connected layer can integrate and map the output local features after global average pooling of the previous layers, and the obtained global feature data is used for subsequent upsampling and inversion reconstruction tasks.

[0085] S300, inferring the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure, including:

[0086] The adjacency matrix construction operation, feature propagation operation and radiation field reconstruction operation are performed on the radiation field graph neural network structure to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion.

[0087] In an embodiment of the present application, the data upsampling module uses graph neural network technology to invert and reconstruct the three-dimensional radiation field of the entire target area of ​​the radiation field graph neural network structure. The radiation field graph neural network structure obtained based on limited local measurement data of the radiation field infers the three-dimensional radiation field distribution of the radiation field in the entire area, thereby achieving high-precision inversion.

[0088] Specifically, based on the radiation field graph neural network structure, a data upsampling module is designed. In order to fill the radiation field distribution in the unmeasured area, on the basis of the measured area, the data upsampling module is based on the radiation field graph neural network structure of the graph neural network. By learning and propagating the local measurement data of the radiation field, the data of the nodes in the unmeasured area are upsampled, and the radiation field distribution in the unmeasured area is effectively restored, and a complete three-dimensional radiation field distribution is generated, achieving high-precision inversion of the entire three-dimensional space.

[0089] Let the parameter of the data upsampling module be θ Upsampling , the upsampling process is expressed as:

[0090] H complete =GCN_Upsampling(H global ,D measured ,θ Upsampling )

[0091] Among them, H complete It is the three-dimensional radiation field distribution data.

[0092] More specifically, the refinement of the data upsampling module can be decomposed into the following sub-steps:

[0093] Adjacency matrix construction operation: Based on the measured nodes, construct the adjacency relationship between the unmeasured nodes and the measured nodes to form an adjacency matrix M represents M radiation detectors, one radiation detector represents a measured node, and N represents N unmeasured nodes.

[0094] Feature propagation operation: Through graph convolution operation, the global feature H global Propagate to unmeasured nodes to obtain the upsampled global feature representation.

[0095] Reconstruct radiation field operation: Use the upsampled global feature representation and combine it with the real data of the measured area to reconstruct the complete three-dimensional radiation field.

[0096] The specific implementation of the upsampling operation is to use NodeShuffle to effectively upsample the point cloud. NodeShuffle is a graph convolution upsampling layer. Given a node feature ν with a shape of N×C ι , the NodeShuffle operator will output a new node feature ν with a shape of rN×C ι+1 as follows:

[0097] ν l+1 =PS(W l+1 ×V l +b l+1 )

[0098] Among them, PS is a periodic shuffling operator, which rearranges the point feature map of shape N×rC into rN×C. ι is the ιth node. To realize the rearrangement transformation of the point feature map, the parameter C used by the program is the count, and W l+1 is the weight parameter of the ι+1th node, V l is the node feature of the ιth node, b l+1 is the bias parameter of the ι+1th node.

[0099] The NodeShuffle operation can be divided into two steps: (1) Channel expansion: Use a 1-layer graph convolutional neural network GCN to transform the node feature ν ι Expanded to shape N×rC. (2) Periodic shuffling: Using learnable parameters W l+1 and b l+1 , rearrange the output of channel expansion into shape rN×C; use the graph convolutional neural network GCN to expand the number of input global features by r times, and then apply the shuffle operation to rearrange the feature map; r is the upsampling ratio.

[0100] S400, training and optimizing a graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain a radiation field graph neural network model for three-dimensional radiation field inversion, including:

[0101] The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, the loss metric function and the gradient descent method to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion.

[0102] In an embodiment of the present application, the training and optimization module adopts appropriate optimization strategies, such as Dropout, BatchNormalization, etc., to appropriately adjust and optimize the parameters of the graph neural network, namely the local measurement data of the radiation field and the three-dimensional radiation field distribution, thereby enhancing the prediction ability and generalization performance of the radiation field graph neural network model.

[0103] The graph neural network model is trained using the existing local measurement data of the radiation field and the three-dimensional radiation field distribution. The model prediction result is defined as, and the mean square error MSE is used as the loss function. The parameters of the radiation field graph neural network model are adjusted by optimization algorithms such as the gradient descent method to improve the prediction accuracy of the radiation field graph neural network model. In this process, regularization strategies such as L2 regularization can be introduced to control the complexity of the radiation field graph neural network model and enhance the generalization performance.

[0104] The loss function is expressed as:

[0105]

[0106] Where N is the number of samples. is the radiation field prediction data, corresponding to the model prediction result of this application, is the actual data of the radiation field, corresponding to the actual radiation field distribution of this application.

[0107] The training process of neural networks in deep learning is essentially an optimization problem, whose goal is to minimize the gap between the model prediction results and the actual radiation field distribution. In this process, the loss function is used to measure the prediction accuracy of the radiation field graph neural network model, quantify the error between the model prediction results of the radiation field graph neural network model and the actual radiation field distribution, and find a set of weight parameters w of the radiation field graph neural network model through gradient descent and its variant algorithms. i and the bias parameter b i , where i represents any node, so that the value of the loss function is minimized, providing direction for the optimization of the radiation field graph neural network model. The selection and design of the loss function has an important impact on the performance of the radiation field graph neural network model. Different loss functions may produce different results for the same data set, depending on how the loss function measures the prediction error. For example, the mean square error loss function is sensitive to outliers, while the cross entropy loss function is more suitable for dealing with discrete correlation analysis problems. In addition, the shape of the loss function (such as convexity or non-convexity) will also affect the difficulty of the neural network model optimization process. The present application sets the loss metric function for the inversion model, i.e., the radiation field graph neural network model training, as follows. Compared with ordinary MSE, MAE and other loss functions, it has stronger robustness and is insensitive to noise and outliers. It focuses more on the distribution of the overall vector in high-dimensional space, can better handle individual noise and outliers in point cloud data, and reduce the impact on the training of the radiation field graph neural network model.

[0108] The loss metric function is:

[0109]

[0110] Wherein, P is the predicted distribution, i.e., the three-dimensional radiation field distribution of the present application; Q is the basic truth value, i.e., the local measurement data of the radiation field of the present application; e is the eigenvector corresponding to the three-dimensional radiation field distribution point of P; k is the eigenvector corresponding to the three-dimensional radiation field distribution point of Q; the operator is the square of the Euclidean norm.

[0111] S500, performing model prediction adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model, including:

[0112] According to the measured data set and the radiation field map neural network model, the model prediction results of the three-dimensional radiation field inversion are determined; according to the model prediction results and the evaluation index of the actual radiation field distribution, the model performance of the three-dimensional radiation field inversion is adjusted; and the determination of the model prediction results and the adjustment of the model performance are used as the model prediction adjustment of the three-dimensional radiation field inversion.

[0113] In the embodiment of the present application, when the model prediction is performed, the radiation field graph neural network model receives a small amount of local measurement data of the radiation field from the radiation detector, and performs inversion calculation through the radiation field graph neural network model to generate a detailed distribution of the entire three-dimensional radiation field. Specifically, the existing measured data set, i.e., the local measurement data of the radiation field measured by the new radiation detector, is used, and after being processed by the above-mentioned feature extraction module and data upsampling module, a complete three-dimensional radiation field distribution is obtained, which is referred to as the actual radiation field distribution.

[0114] Step S500 uses the existing measured data set and inputs the measured data set into the radiation field graph neural network model, so that the radiation field graph neural network model can accurately predict the unmeasured area in actual application, and obtain the three-dimensional radiation field distribution predicted by the model, which is referred to as the model prediction result. The mean square error MSE and other evaluation indicators between the model prediction result and the actual radiation field distribution are calculated to verify the accuracy and generalization ability of the radiation field graph neural network model; by repeatedly adjusting the structure and parameters of the radiation field graph neural network model, the model performance is gradually improved.

[0115] In the embodiment of the present application, a radiation dose inversion method based on a graph neural network is adopted. First, the graph neural network model is trained and the convergence of the graph neural network model is recorded. Figure 2 The changing trend of the model loss value during the training of the graph neural network model is shown, showing that as the training progresses, the model loss value of the graph neural network model gradually decreases and tends to be stable as the number of iterations increases, indicating that the graph neural network model has successfully converged and achieved the expected accuracy requirements. At this time, the trained graph neural network model is called the radiation field graph neural network model. In order to verify the generalization ability of the radiation field graph neural network model, the measured data set with a random distribution of a radiation field was used for testing. Figure 3 and Figure 4 The spatial position coordinates of each point in the equation can be (x, y, z), and the point in the shape of a small ball is the dose rate value represented by the spatial position coordinates; Figure 3 The randomly input measured data set used is shown in . The measured data set includes local measurement data of the radiation field, which are randomly extracted from a specific distribution. By inputting these measured data sets into the radiation field graph neural network model, a three-dimensional radiation field inversion calculation of the three-dimensional radiation field distribution is performed. The local measurement data of the radiation field includes the radiation field measurement values ​​of several points, and the three-dimensional radiation field distribution includes the radiation dose distribution of each point in the three-dimensional radiation field. Figure 4 The model prediction results of the radiation field graph neural network model after the measured data set is input are shown. The model prediction results represent the radiation dose distribution of each point in the three-dimensional radiation field. By comparing the model prediction results with the actual radiation field distribution, the accuracy and reliability of the model prediction results of the radiation field graph neural network model are verified. The embodiment of the present application inputs the local measurement data of the radiation field, and then the graph neural network-based method can more accurately invert the entire three-dimensional radiation field distribution, so that a small number of radiation detector measurement points in the target scene can obtain the three-dimensional radiation field distribution information of the complete three-dimensional space. The experimental results show that the present application has a high practical value in the prediction of unknown radiation environments.

[0116] In the present application, the three-dimensional radiation field inversion method based on the graph neural network method obtains the local measurement data of the radiation field for the three-dimensional radiation field inversion through field measurement of the radiation detector; uses a feature extractor to extract features from the local measurement data of the radiation field, and constructs the radiation field graph neural network structure for the three-dimensional radiation field inversion; infers the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure; trains and optimizes the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, and obtains the radiation field graph neural network model for the three-dimensional radiation field inversion; and performs model prediction and adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model. The present application makes full use of the graph neural network, and through the synergy of feature extraction and data upsampling, recovers the high-resolution and all-round three-dimensional radiation field from the local measurement data of the radiation field measured by the limited radiation detector, and realizes the inversion of the entire three-dimensional radiation field distribution; the present application processes the local measurement data of the radiation field of the limited measurement in an intelligent way, realizes the intelligent processing of the limited measurement data, and provides a new efficient and accurate method for radiation field monitoring in the field of nuclear engineering. This application has broad application prospects in the field of nuclear engineering. It not only improves the accuracy and precision of radiation field inversion, but also provides an innovative solution for the safety monitoring of nuclear facilities.

[0117] The following are embodiments of the device of the present invention, which can be used to implement the method embodiments of the present invention. For details not disclosed in the device embodiments of the present invention, please refer to the method embodiments of the present invention.

[0118] See also Figure 5 , which shows a schematic diagram of the structure of a three-dimensional radiation field inversion device based on a graph neural network method provided by an exemplary embodiment of the present invention. The device includes: an acquisition module 10000, a structure construction module 20000, an inference module 30000, a model training optimization module 40000 and a model prediction adjustment module 50000.

[0119] An acquisition module 10000 is used to obtain local measurement data of the radiation field for three-dimensional radiation field inversion through field measurement with a radiation detector;

[0120] A structure construction module 20000 is used to extract features from the local measurement data of the radiation field using a feature extractor, and to construct a radiation field graph neural network structure for the three-dimensional radiation field inversion;

[0121] An inference module 30000 is used to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure;

[0122] A model training and optimization module 40000 is used to train and optimize the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion;

[0123] The model prediction adjustment module 50000 is used to perform model prediction adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model.

[0124] According to a preferred embodiment, the structure building module 20000 is specifically used for:

[0125] Using a graph convolution layer, a residual connection layer, a nonlinear transformation layer, a pooling layer and a fully connected layer to extract features from the local measurement data of the radiation field to obtain global feature data of the three-dimensional radiation field inversion;

[0126] Constructing a radiation field graph neural network structure for three-dimensional radiation field inversion according to the global feature data;

[0127] The graph convolution layer, the residual connection layer, the nonlinear transformation layer, the pooling layer and the fully connected layer are used as the feature extractor.

[0128] According to a preferred embodiment, the inference module 30000 is specifically used for:

[0129] The adjacency matrix construction operation, feature propagation operation and radiation field reconstruction operation are performed on the radiation field graph neural network structure to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion.

[0130] According to a preferred embodiment, the model training optimization module 40000 is specifically used for:

[0131] The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, the loss metric function and the gradient descent method to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion.

[0132] According to a preferred embodiment, the model prediction adjustment module 50000 is specifically used for:

[0133] Determining a model prediction result of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model;

[0134] Adjusting the model performance of the three-dimensional radiation field inversion according to the model prediction results and the evaluation index of the actual radiation field distribution;

[0135] Determining the model prediction results and adjusting the model performance are used as model prediction adjustments for the three-dimensional radiation field inversion.

[0136] It should be noted that the three-dimensional radiation field inversion device based on the graph neural network method provided in the above embodiment only uses the division of the above-mentioned functional modules as an example when executing the three-dimensional radiation field inversion method based on the graph neural network method. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the three-dimensional radiation field inversion device based on the graph neural network method provided in the above embodiment and the embodiment of the three-dimensional radiation field inversion method based on the graph neural network method belong to the same concept. The implementation process is detailed in the method embodiment, which will not be repeated here.

[0137] In the present application, the three-dimensional radiation field inversion device based on the graph neural network method obtains the local measurement data of the radiation field for the three-dimensional radiation field inversion through field measurement of the radiation detector; uses a feature extractor to extract features from the local measurement data of the radiation field, and constructs the radiation field graph neural network structure for the three-dimensional radiation field inversion; infers the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure; trains and optimizes the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, and obtains the radiation field graph neural network model for the three-dimensional radiation field inversion; and performs model prediction and adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model. The present application makes full use of the graph neural network, and through the synergy of feature extraction and data upsampling, recovers the high-resolution and all-round three-dimensional radiation field from the local measurement data of the radiation field measured by the limited radiation detector, and realizes the inversion of the entire three-dimensional radiation field distribution; the present application processes the local measurement data of the radiation field of the limited measurement in an intelligent way, realizes the intelligent processing of the limited measurement data, and provides a new efficient and accurate method for radiation field monitoring in the field of nuclear engineering. This application has broad application prospects in the field of nuclear engineering. It not only improves the accuracy and precision of radiation field inversion, but also provides an innovative solution for the safety monitoring of nuclear facilities.

[0138] The present invention also provides a computer-readable medium having program instructions stored thereon, which, when executed by a processor, implement the three-dimensional radiation field inversion method based on the graph neural network method provided in the above-mentioned method embodiments.

[0139] The present invention also provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the three-dimensional radiation field inversion method based on the graph neural network method of each of the above-mentioned method embodiments.

[0140] See also Figure 6 , which is a schematic diagram of the structure of a terminal according to an embodiment of the present application. The terminal may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0141] The communication bus 1002 is used to realize the connection and communication between these components.

[0142] The user interface 1003 may include a display screen (Display) and a camera (Camera), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.

[0143] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0144] Among them, the processor 1001 may include one or more processing cores. The processor 1001 uses various interfaces and lines to connect various parts in the entire terminal, and executes various functions of the terminal and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 1005, and calling data stored in the memory 1005. Optionally, the processor 1001 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 1001 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 1001, and it can be implemented separately through a chip.

[0145] Among them, the memory 1005 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read-Only Memory). Optionally, the memory 1005 includes a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 1005 may optionally be at least one storage device located away from the aforementioned processor 1001. As Figure 6 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a three-dimensional radiation field inversion application based on a graph neural network method.

[0146] exist Figure 6In the terminal shown, the user interface 1003 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 1001 can be used to call the three-dimensional radiation field inversion application based on the graph neural network method stored in the memory 1005, and specifically perform the following operations:

[0147] Through on-site measurement with radiation detectors, local measurement data of the radiation field for three-dimensional radiation field inversion is obtained;

[0148] Using a feature extractor to extract features from the local measurement data of the radiation field, and constructing a radiation field graph neural network structure for the three-dimensional radiation field inversion;

[0149] Inferring the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure;

[0150] The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain the radiation field graph neural network model for the three-dimensional radiation field inversion;

[0151] The model prediction adjustment of the three-dimensional radiation field inversion is performed according to the measured data set and the radiation field graph neural network model.

[0152] In one embodiment, when the processor 1001 performs the feature extraction of the local measurement data of the radiation field using the feature extractor and constructs the radiation field graph neural network structure of the three-dimensional radiation field inversion, the following operations are specifically performed:

[0153] Using a graph convolution layer, a residual connection layer, a nonlinear transformation layer, a pooling layer and a fully connected layer to extract features from the local measurement data of the radiation field to obtain global feature data of the three-dimensional radiation field inversion;

[0154] Constructing a radiation field graph neural network structure for three-dimensional radiation field inversion according to the global feature data;

[0155] The graph convolution layer, the residual connection layer, the nonlinear transformation layer, the pooling layer and the fully connected layer are used as the feature extractor.

[0156] In one embodiment, when the processor 1001 executes the three-dimensional radiation field distribution inferred from the three-dimensional radiation field inversion according to the radiation field graph neural network structure, the processor 1001 specifically performs the following operations:

[0157] The adjacency matrix construction operation, feature propagation operation and radiation field reconstruction operation are performed on the radiation field graph neural network structure to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion.

[0158] In one embodiment, when the processor 1001 performs the training and optimization of the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain the radiation field graph neural network model for the three-dimensional radiation field inversion, the following operations are specifically performed:

[0159] The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, the loss metric function and the gradient descent method to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion.

[0160] In one embodiment, when the processor 1001 performs the model prediction adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model, the processor 1001 specifically performs the following operations:

[0161] Determining a model prediction result of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model;

[0162] Adjusting the model performance of the three-dimensional radiation field inversion according to the model prediction results and the evaluation index of the actual radiation field distribution;

[0163] Determining the model prediction results and adjusting the model performance are used as model prediction adjustments for the three-dimensional radiation field inversion.

[0164] In the present application, the three-dimensional radiation field inversion method and device based on the graph neural network method obtains the local measurement data of the radiation field for the three-dimensional radiation field inversion through field measurement of the radiation detector; uses a feature extractor to extract features from the local measurement data of the radiation field, and constructs the radiation field graph neural network structure for the three-dimensional radiation field inversion; infers the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure; trains and optimizes the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, and obtains the radiation field graph neural network model for the three-dimensional radiation field inversion; and performs model prediction and adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model. The present application makes full use of the graph neural network, and through the synergy of feature extraction and data upsampling, recovers the high-resolution and all-round three-dimensional radiation field from the local measurement data of the radiation field measured by the limited radiation detector, and realizes the inversion of the entire three-dimensional radiation field distribution; the present application processes the local measurement data of the radiation field of the limited measurement in an intelligent way, realizes the intelligent processing of the limited measurement data, and provides a new efficient and accurate method for radiation field monitoring in the field of nuclear engineering. This application has broad application prospects in the field of nuclear engineering. It not only improves the accuracy and precision of radiation field inversion, but also provides an innovative solution for the safety monitoring of nuclear facilities.

[0165] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0166] The above disclosure is only the preferred embodiment of the present application, which certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A three-dimensional radiation field inversion method based on a graph neural network method, characterized in that: The following steps are involved: Through on-site measurement with radiation detectors, local measurement data of the radiation field for three-dimensional radiation field inversion is obtained; Using a feature extractor to extract features from the local measurement data of the radiation field, and constructing a radiation field graph neural network structure for the three-dimensional radiation field inversion; Inferring the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure; The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain the radiation field graph neural network model for the three-dimensional radiation field inversion; The model prediction and adjustment of the three-dimensional radiation field inversion are performed according to the measured data set and the radiation field graph neural network model.

2. The three-dimensional radiation field inversion method based on the graph neural network method according to claim 1 is characterized in that: The method of using a feature extractor to extract features from the local measurement data of the radiation field and constructing a radiation field graph neural network structure for three-dimensional radiation field inversion includes: Using a graph convolution layer, a residual connection layer, a nonlinear transformation layer, a pooling layer and a fully connected layer to extract features from the local measurement data of the radiation field to obtain global feature data of the three-dimensional radiation field inversion; Constructing a radiation field graph neural network structure for three-dimensional radiation field inversion according to the global feature data; The graph convolution layer, the residual connection layer, the nonlinear transformation layer, the pooling layer and the fully connected layer are used as the feature extractor.

3. The three-dimensional radiation field inversion method based on the graph neural network method according to claim 1 is characterized in that: The inferring the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the radiation field graph neural network structure comprises: The adjacency matrix construction operation, feature propagation operation and radiation field reconstruction operation are performed on the radiation field graph neural network structure to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion.

4. The three-dimensional radiation field inversion method based on the graph neural network method according to claim 1 is characterized in that: The step of training and optimizing the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution to obtain the radiation field graph neural network model for the three-dimensional radiation field inversion includes: The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, the loss metric function and the gradient descent method to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion.

5. The three-dimensional radiation field inversion method based on the graph neural network method according to claim 1 is characterized in that: The model prediction adjustment of the three-dimensional radiation field inversion is performed according to the measured data set and the radiation field graph neural network model, including: Determining a model prediction result of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model; Adjusting the model performance of the three-dimensional radiation field inversion according to the model prediction results and the evaluation index of the actual radiation field distribution; Determining the model prediction results and adjusting the model performance are used as model prediction adjustments for the three-dimensional radiation field inversion.

6. A three-dimensional radiation field inversion device based on a graph neural network method, the device is used to execute the three-dimensional radiation field inversion method based on a graph neural network method as described in any one of claims 1 to 5, characterized in that: include: An acquisition module is used to obtain local measurement data of the radiation field for three-dimensional radiation field inversion through field measurement with a radiation detector; A structure construction module, used for extracting features from the local measurement data of the radiation field using a feature extractor, and constructing a radiation field graph neural network structure for the three-dimensional radiation field inversion; An inference module, used to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion according to the neural network structure of the radiation field graph; A model training and optimization module, used to train and optimize the graph neural network model according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, to obtain the radiation field graph neural network model of the three-dimensional radiation field inversion; The model prediction adjustment module is used to perform model prediction adjustment of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model.

7. The three-dimensional radiation field inversion device based on the graph neural network method according to claim 6 is characterized in that: The structural building module is specifically used for: Using a graph convolution layer, a residual connection layer, a nonlinear transformation layer, a pooling layer and a fully connected layer to extract features from the local measurement data of the radiation field to obtain global feature data of the three-dimensional radiation field inversion; Constructing a radiation field graph neural network structure for three-dimensional radiation field inversion according to the global feature data; The graph convolution layer, the residual connection layer, the nonlinear transformation layer, the pooling layer and the fully connected layer are used as the feature extractor.

8. The three-dimensional radiation field inversion device based on the graph neural network method according to claim 6 is characterized in that: The inference module is specifically used for: The adjacency matrix construction operation, feature propagation operation and radiation field reconstruction operation are performed on the radiation field graph neural network structure to infer the three-dimensional radiation field distribution of the three-dimensional radiation field inversion.

9. The three-dimensional radiation field inversion device based on the graph neural network method according to claim 6 is characterized in that: The model training optimization module is specifically used for: The graph neural network model is trained and optimized according to the local measurement data of the radiation field and the three-dimensional radiation field distribution, the loss metric function and the gradient descent method to obtain a radiation field graph neural network model for the three-dimensional radiation field inversion.

10. The three-dimensional radiation field inversion device based on the graph neural network method according to claim 6, characterized in that: The model prediction adjustment module is specifically used for: Determining a model prediction result of the three-dimensional radiation field inversion according to the measured data set and the radiation field graph neural network model; Adjusting the model performance of the three-dimensional radiation field inversion according to the model prediction results and the evaluation index of the actual radiation field distribution; Determining the model prediction results and adjusting the model performance are used as model prediction adjustments for the three-dimensional radiation field inversion.

Images