CN115166811A - Method for quickly constructing neutron spectrum in simulation working site - Google Patents
Method for quickly constructing neutron spectrum in simulation working site Download PDFInfo
- Publication number
- CN115166811A CN115166811A CN202210527270.2A CN202210527270A CN115166811A CN 115166811 A CN115166811 A CN 115166811A CN 202210527270 A CN202210527270 A CN 202210527270A CN 115166811 A CN115166811 A CN 115166811A
- Authority
- CN
- China
- Prior art keywords
- neural network
- network model
- neutron
- spectrum
- sample data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01T—MEASUREMENT OF NUCLEAR OR X-RADIATION
- G01T3/00—Measuring neutron radiation
- G01T3/001—Spectrometry
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E30/00—Energy generation of nuclear origin
- Y02E30/30—Nuclear fission reactors
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Spectroscopy & Molecular Physics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- High Energy & Nuclear Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a method for quickly constructing a neutron spectrum in a simulation working site. The method adopts a Monte Carlo method to simulate neutron moderation, scattering and shielding, obtains a moderated neutron spectrum, combines a neutron source energy spectrum, a moderation configuration scheme and a moderated neutron spectrum result which are adopted during simulation to manufacture a data set, then trains a neural network model by using the data set in combination with a deep learning method, predicts the moderation configuration scheme according to a required work site neutron spectrum and the neutron source energy spectrum combined network model, and finally searches a global optimal configuration scheme by using a multi-objective optimization algorithm. The method can quickly calculate the neutron spectrum slowing configuration scheme of a simulation working site aiming at any standard neutron source in a laboratory, the obtained configuration scheme has the calculation precision similar to that of a Monte Carlo method, and meanwhile, the calculation speed can be greatly improved. The method can carry out multi-objective optimization on the slowing-down configuration scheme according to the requirements of economy, shielding material quality and volume, and finds out a global optimal scheme.
Description
Technical Field
The invention belongs to the fields of radiation safety evaluation, neutron detection instrument calibration field construction and deep learning, and relates to a rapid neutron spectrum construction method in a simulation working field.
Background
Accurate measurement of neutron dose levels in a work site is particularly important for radiation safety assessment and worker radiation protection. The neutron energy spectrum is wide in the normal working site, and the energy range is 10 -9 -20 MeV; the neutron ambient dose equivalent rate H (10) is closely related to the neutron energy density. Measurement research carried out by the European Union shows that the neutron dosimeter based on GB/T14055 series (international standard ISO 8259 series) standard reference radiation field scale has larger field measurement error, and the difference between the measurement result of most common neutron dosimeters and the reference value is about 2 times in different fields such as reactors, spent fuel transport containers, spent fuel post-treatment and the like. The energy spectrum of the calibration source is different from the energy spectrum of the job site where the dosimeter is to be used, and the dosimeter reading is considered to be incorrectly determined.
The neutron spectrum technology of the simulation working site establishes a calibration field which can better represent the energy spectrum of the working field by utilizing a laboratory neutron source through configuring a slowing body, a scattering body and a shielding body, and is an effective solution for solving the problem of large errors of reference values and measured values of dosimeters.
The construction of neutron spectrum calibration fields in simulation work sites requires designing neutron moderation. The physical process of neutron moderation is complex, the error of moderation calculation based on an empirical formula or an attenuation coefficient is large, and a large moderation margin needs to be reserved usually, so that the neutron spectrum in a simulation working site is inaccurate. The Monte Carlo Method (MC) can accurately calculate neutron moderation to obtain a simulated on-site neutron spectrum, but extremely occupies computer resources and takes a long time for calculation. In addition, the neutron energy spectrum characteristics are different in different workplaces, so that the slowing-down materials and the structural configuration schemes are different, and the design of the slowing-down configuration schemes on specific sites one by one is complex, so that a calculation method is needed to achieve high calculation accuracy, and meanwhile, the occupation of computer resources is small, and the consumed time is short.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a method for quickly constructing a neutron spectrum in a simulation working site, which can realize the quick design of a slowing configuration scheme according to a neutron source spectrum and a working site neutron spectrum by utilizing a deep learning technology and combining a Monte Carlo method, and can ensure that the obtained simulation site neutron spectrum and the actual working site neutron spectrum meet the set accuracy requirement.
In order to achieve the purpose, the invention provides a method for quickly constructing a neutron spectrum in a simulation working site, which comprises the following steps:
s1, making a sample data set of a neural network model;
s2, training to obtain a learned neural network model;
s3, deploying the learned neural network model to hardware;
and S4, searching a global optimal slowing-down configuration scheme by adopting a multi-objective optimization algorithm.
Further, the manufacturing process of the sample data set comprises the following two steps:
s11, simulating a neutron energy spectrum of neutrons after slowing, scattering and shielding based on a Monte Carlo method to obtain sample data and a sample label;
the sample data comprises energy spectrum sample data and a moderation configuration scheme, wherein the energy spectrum sample data comprises a neutron source energy spectrum and simulated on-site neutron energy spectrum data, and the moderation configuration scheme is a sample label;
s12, preprocessing a plurality of groups of energy spectrum sample data and sample labels to manufacture a sample data set;
the sample data set comprises a training set and a test set, wherein the training set is used for training the neural network model, and the test set is used for verifying the accuracy of the neural network model for predicting the slowing configuration scheme and the sample label.
Further, in the step S11, a monte carlo simulation particle transport tool is used to perform moderation structure modeling on an isotope neutron source or an accelerator neutron source, and a series of simulation calculations are performed for a plurality of moderation configuration schemes to obtain simulation field neutron energy spectrum data under different moderation configuration schemes;
the moderator arrangement includes moderator features in various moderator materials, various size structures and shapes, and various moderator material placement sequences that form a sample label.
Further, in the step 12, the method for preprocessing the plurality of sets of energy spectrum sample data and sample labels includes: and normalizing the energy spectrum sample data, performing numerical representation on the sample label, and making a sample data set for training and testing a neural network model.
Further, in the step 12, the method for preprocessing the spectrum sample data and the sample label further includes: before the data is input into the neural network model, the process of random disturbance and batch processing of the sample data in the sample data set is performed.
Further, the step S2 includes the following four steps:
s21, constructing a multilayer neural network model, wherein the neural network model comprises but is not limited to a full-connection neural network model;
s22, constructing a loss function, processing input data through a multilayer network structure, and calculating a loss value;
s23, model training, namely performing iterative computation on the input training set, searching a minimum value of a loss value variance through logistic regression gradient descent, reversely transmitting a gradient to a parameter of the neural network model, and updating a weight value in the network according to an updating rule; adjusting and optimizing the hyper-parameters of the neural network model to obtain a learned neural network model;
and S24, verifying the accuracy of the learned neural network model through the test set.
Further, in step S21, the neural network model is a fully-connected neural network model, and the fully-connected neural network model adopts a 5-layer structure: the system comprises 1 input layer, 3 hidden layers and 1 output layer, wherein the input layer is used for acquiring the energy spectrum sample data, the hidden layers are used for further abstracting the energy spectrum sample data, and the output layer outputs a prediction slowing-down configuration scheme according to the energy spectrum sample data.
Further, in step S22, the loss function is used to calculate a difference value between the predicted slowing-down configuration scheme output by the neural network model and the sample label, where the difference value is a loss value;
if the number of parameters of the predicted slowing-down configuration scheme output by the neural network model is 2, adopting a logarithmic loss function as the loss function;
and if the number of parameters of the predicted slowing-down configuration scheme output by the neural network model is more than 2, adopting a cross entropy loss function as the loss function.
Further, in step S23, the model training is to find a minimum value of the variance of the loss values in the training set by iteratively updating the weight values of the neural network, that is, a minimum value point of the cost function, and then obtain the neural network model with generalization capability based on the minimum value point;
in the iterative computation, the gradient of the cost function is computed and reversely transferred to the neural network model, and the network weight value is updated by combining with the learning rate super-parameter;
the cost function is a variance function of the loss value, and overfitting of the neural network model is prevented through a regularization method.
Further, in step S24, the test set is used to verify the accuracy of the learned neural network model, and the learned neural network model may be deployed after verification passes without updating a network weight value in the verification process.
Further, in step S3, after the learned neural network model is deployed to hardware, the field neutron spectrum and neutron source energy spectrum data to be simulated are input, and then the learned neural network model outputs a prediction slowing configuration scheme meeting the set accuracy requirement.
Further, in step S4, the multi-objective optimization algorithm finds a globally optimal shielding configuration scheme based on actual engineering implementation factors, including but not limited to economic cost, moderating material mass, and moderating volume.
The method for rapidly constructing the neutron spectrum in the simulation working site has the advantages that a Monte Carlo method is adopted to simulate neutron moderation, scattering and shielding, the neutron spectrum after moderation is obtained, a data set is manufactured by combining a neutron source energy spectrum, a moderation configuration scheme and a neutron energy spectrum result after moderation, a neural network model is trained by using the data set in combination with a deep learning method, the moderation configuration scheme is predicted according to the neutron spectrum and the neutron source energy spectrum in the required working site in combination with a network model, and finally a global optimal configuration scheme is searched by using a multi-objective optimization algorithm. By adopting the rapid neutron spectrum construction method for the simulation job site, the neutron spectrum slowing configuration scheme for the simulation job site can be rapidly calculated aiming at any standard neutron source in a laboratory, the obtained configuration scheme has calculation precision similar to that of a Monte Carlo method, and meanwhile, the calculation speed can be greatly improved. The method provided by the invention can carry out multi-objective optimization on the slowing-down configuration scheme according to the requirements of economy, the quality and the volume of the slowing-down material, and find out the global optimal scheme.
Drawings
Fig. 1 is a flowchart of a method for fast constructing a neutron spectrum in a simulation job site according to an embodiment of the present invention.
Fig. 2 is a diagram of a fully-connected neural network model structure according to an embodiment of the present invention.
FIG. 3 is a flow chart of a multi-objective optimization algorithm calculation provided by the embodiment of the present invention.
Detailed Description
The invention is further described with reference to the following figures and detailed description. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments of the present invention, belong to the protection scope of the present invention.
As shown in fig. 1 to 3, in the method for rapidly constructing a neutron spectrum in a simulation work site according to an embodiment of the present invention, moderation, scattering, and shielding of neutrons are simulated based on a monte carlo method, a moderated neutron spectrum is obtained, and a data set is created by combining a neutron source spectrum, a moderation configuration scheme, and a result of the moderated neutron spectrum (i.e., a simulated on-site neutron spectrum); then, deep learning of the relation between the moderation configuration scheme and the neutron energy spectrum is carried out by combining a neural network model; the learnt neural network model is deployed to hardware, and a moderation configuration scheme can be quickly constructed according to the neutron energy spectrum and the neutron source energy spectrum conditions in a working site by utilizing the neural network model. When a plurality of moderation configuration schemes meeting the accuracy requirement exist in the same on-site neutron spectrum, a multi-objective optimization algorithm can be further combined, and a global optimal moderation configuration scheme is comprehensively considered and found from factors such as economy, moderation material quality and volume.
The method mainly comprises the following steps:
s1, making a sample data set of a neural network model;
the making process of the input sample data set comprises the following two steps:
s11, simulating a neutron energy spectrum of neutrons after slowing, scattering and shielding based on a Monte Carlo method to obtain sample data and a sample label;
in this embodiment, the neutron spectrum data of neutrons after being moderated, scattered and shielded based on the monte carlo method in step S11 is simulated, a monte carlo simulation particle transport tool needs to be used for performing moderation structure modeling on an isotope neutron source or an accelerator neutron source, and a series of simulation calculations are performed for various moderation configurations to obtain simulation field neutron spectrum data under different moderation configurations. Wherein the moderation configuration includes characteristics of moderation configuration using various moderating materials, various size structures and shapes, various order of placement of moderating materials, and so on.
The sample data comprises energy spectrum sample data and a slowing-down configuration scheme, wherein the slowing-down configuration scheme is a sample label. The energy spectrum sample data comprises a neutron source energy spectrum and simulated on-site neutron energy spectrum data.
Herein, the neutron source spectrum and the simulated on-site neutron spectrum refer to a neutron source spectrum and a simulated on-site neutron spectrum, respectively.
S12, preprocessing a plurality of groups of energy spectrum sample data and sample labels to manufacture a sample data set;
the sample data set comprises a training set and a test set, wherein the training set is used for training the neural network model, and the test set is used for verifying the accuracy of the neural network model in predicting the slowing-down configuration and the sample label.
In this embodiment, the method for preprocessing the energy spectrum sample data and the sample label in step S12 includes: and normalizing the energy spectrum sample data, performing numerical representation on the sample label, and making a sample data set for training and testing a neural network model. The neutron source spectrum and the neutron energy spectrum data in the simulation field are normalized, the moderation configuration characteristics such as the type, the size structure and the placing sequence of the moderation material are expressed in a numerical mode, and a sample data set and a sample label set for training the neural network model are manufactured.
Further, the method for preprocessing the energy spectrum sample data and the sample label further comprises the following steps: and before the input of the input data into the neural network model, randomly disordering and batch processing the sample data in the sample data set.
S2, training to obtain a learned neural network model;
the step S2 includes the following two steps:
s21, constructing a multilayer neural network model, wherein the neural network model comprises but is not limited to a full-connection neural network model;
in this embodiment, step S21 is to construct a multi-layer fully-connected neural network model, where the fully-connected neural network model has a 5-layer structure: the system comprises 1 input layer, 3 hidden layers and 1 output layer, wherein the input layer has the function of acquiring the energy spectrum sample data, the hidden layers further abstract the energy spectrum sample data, and the output layer outputs a prediction slowing-down configuration scheme according to the energy spectrum sample data.
S22, constructing a loss function, processing input data through a multilayer network structure, and calculating a loss value;
in step S22 of this embodiment, the loss function is used to calculate a difference between a predicted slowing-down configuration scheme (output value) output by the fully-connected neural network model and a sample label (true value), and the difference is a loss value. If the parameter number of the prediction slowing-down configuration scheme output by the fully-connected neural network model is 2, the loss function adopts a logarithmic loss function; and if the number of parameters of the prediction slowing configuration scheme output by the fully-connected neural network model is more than 2, adopting a cross entropy loss function as the loss function. For example, when the slowing-down materials are two, the loss function adopts a logarithmic loss function; when the type of the slowing-down material is more than 2, the loss function adopts a cross entropy loss function. The cross entropy loss function is specifically in the form:
wherein L represents a cross entropy loss function; n represents the output predicted slowdown configuration data dimension; x represents an x-th dimension, wherein 0-n-x-n; y represents the actual digitized sample label value of the x dimension; a represents a predicted value of the x-th dimension data.
S23, model training, namely performing iterative computation on an input training set, searching a minimum value of a loss value variance through logistic regression gradient descent, reversely transmitting a gradient to parameters of the fully-connected neural network model, and updating a weight value in the network according to an updating rule; adjusting the hyper-parameters of the fully-connected neural network model to obtain a learned fully-connected neural network model;
in this embodiment, the step S23 of training the model finds the minimum value of the variance of the loss values in the training set, i.e. the minimum value point of the cost function, by iteratively updating the weight values of the neural network, and obtains the neural network model with generalization capability based on the minimum value point. In the iterative computation, the gradient of the cost function is computed and reversely transferred to the neural network, and the weight value of the neural network is updated by combining with the learning rate over-parameter. The cost function is a variance function of the loss value, and overfitting of the neural network model is prevented through a regularization method.
And S24, verifying the accuracy of the learnt neural network model through the test set.
In step S24 of this embodiment, the test set is used to verify the accuracy of the learned fully-connected neural network model, and the deployment of the learned fully-connected neural network model can be performed after the verification passes without updating the network weight value in the verification process.
S3, deploying the learned neural network model to hardware;
and deploying the learned fully-connected neural network model to hardware, inputting field neutron spectrum and neutron source energy spectrum data to be simulated, and outputting a prediction slowing configuration scheme meeting the set accuracy requirement by the learned fully-connected neural network model.
And S4, searching a global optimal slowing-down configuration scheme by adopting a multi-objective optimization algorithm.
In this embodiment, the objective of using the multi-objective optimization algorithm in step S4 is to perform global optimal search on the predicted slowing-down configuration scheme output by the learned fully-connected neural network model based on factors such as economic cost, slowing-down material quality, and slowing-down body volume.
In a specific embodiment, a work flow chart of the neutron spectrum fast construction method in the simulation work site is shown in fig. 1, and the neutron spectrum fast construction method in the simulation work site combines the advantages of high calculation precision of the monte carlo method and high calculation speed of the deep learning method. Simulating neutron spectrum data under different moderation configuration conditions by adopting a Monte Carlo method, and making a sample data set and a sample label set for deep learning neural network model training and learning by combining a neutron source energy spectrum, a moderation configuration scheme and a moderated neutron energy spectrum result which are adopted during simulation; deploying the learnt fully-connected neural network model to hardware, inputting field neutron spectrum and neutron source energy spectrum data to be simulated, and outputting a predicted slowing configuration scheme meeting a certain confidence requirement by the learnt fully-connected neural network model. Because the fully-connected neural network model is based on probability and statistical theory, there will be multiple predicted slowing configuration schemes that meet the requirements, rather than unique solutions. The optimal moderation scheme needs to meet the mass, volume and economy requirements simultaneously, which is a multi-objective optimization problem.
The quality of the sample data set determines the accuracy of the predicted slowdown configuration scheme result output by the network model. The sample data set manufacturing process is complex, modeling and Monte Carlo simulation particle transport are firstly carried out according to a neutron source and a moderation configuration structure of a neutron spectrum in a simulation field, then normalization processing is carried out on the neutron spectrum output after moderation, and numerical information of characteristics such as the type, thickness and placement sequence of a moderation material and the normalized neutron spectrum are combined into a sample data set and a sample label set after the numeralization. Before the sample data set is input into the fully-connected neural network model, the sample data set and the digitized sample tag set need to be randomly disturbed and batch-processed, for example, 10 sample data sets are defined as one batch, and the weight values of the neurons are updated after network model training is performed on one batch of sample data sets.
The fully-connected neural network model structure provided by the present embodiment is as shown in fig. 2, and the fully-connected neural network model adopts a 5-layer structure: the method comprises the following steps that 1 input layer, 3 hidden layers and 1 output layer are arranged, the input layer receives a sample data set, and the number of neurons in the input layer is the number of frequency of a spectrum histogram; the hidden layer abstracts the sample data set, the number of the hidden layers and the number of neurons are variable, but each hidden layer neuron number is required to be larger than the number of neurons of an input layer; and outputting a prediction slowing-down configuration scheme by an output layer, wherein the neuron number of the output layer is equal to the parameter number after characteristic numeralization, such as the type number, the size structure and the shape of slowing-down materials, the placing sequence of the slowing-down materials and the like. The activation function of the fully-connected neural network model runs on the neurons of the fully-connected neural network model and is responsible for mapping the inputs of the neurons to the outputs. The activation function of the fully-connected neural network model adopts a ReLU function, and the convergence speed is higher than that of a Sigmod function and a tanh function. The training of the fully-connected neural network model adopts a random gradient descent algorithm of small batch processing to find minimum value points of errors between a real value and a predicted value; and a self-adaptive learning rate parameter adjusting method is adopted to achieve the purpose of fast convergence of the training model. Because the number of parameters of the prediction slowing configuration scheme output by the fully-connected neural network model is more than 2, the cross entropy loss function is adopted to evaluate the transmission and accumulation of errors in the training process of the fully-connected neural network model.
The multi-objective optimization algorithm calculation flow chart provided by the embodiment is shown in fig. 3, a neutron spectrum in a simulation working site is obtained, and a moderation configuration scheme predicted by a neutron source is not the only solution, so that the multiple calculation results of a learned fully-connected neural network model are different; in view of practical engineering implementation requirements such as moderator volume, mass, and economic cost, it is desirable to find an optimal moderator configuration that satisfies these several conditions simultaneously. To obtain an optimal slowdown configuration scheme, multiple aspects of volume, mass, and economic cost need to be balanced, as mass and cost may be higher when volume optimization is satisfied. The simplest multi-objective optimization method is a linear weighting method, however, in this way the weights need to be carefully chosen and the result obtained is not necessarily a globally optimal solution. The embodiment of the invention adopts a non-dominated sorting genetic algorithm NSGA-II, firstly, the volume, the quality and the economic cost are calculated according to a candidate slowing configuration scheme to generate an initialized population, the population scale is set to be 100, the iteration times are 1000, then, the non-dominated sorting is carried out, the non-dominated sorting population is subjected to crowding degree comparison, selection, intersection and variation by referring to a biological evolution theory, a filial population is generated according to an elite strategy, whether a termination condition is reached is judged, if so, a Pareto solution set with a global optimal shielding scheme is output, otherwise, the non-dominated sorting step is carried out, and genetic iteration is continued until the termination condition is reached. The algorithm keeps the high-quality solution by gradually eliminating the low-quality solution and in a genetic cross-mixing mode. The filial generation of the high-quality solution is still high-quality, and the formed population is Pareto solution set.
In the embodiment, the inventor utilizes a machine learning technology and combines a Monte Carlo simulation particle transport method, so that the problem that the construction precision and the speed of a neutron spectrum slowing configuration scheme conflict in a simulation working site is effectively solved. A Monte Carlo method is utilized to accurately design a simulation field neutron spectrum moderation configuration scheme, a neutron source energy spectrum, a moderation configuration scheme and a moderated neutron spectrum result data set are combined during simulation, then the data set is utilized to combine deep learning to obtain a neural network model, and therefore the neural network model is utilized to rapidly predict any field neutron spectrum moderation configuration design scheme. The method provided by the embodiment combines the quick design characteristic of machine learning and the characteristic of accurate calculation result of the Monte Carlo method.
The above-described embodiments are merely illustrative of the present invention, and those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (12)
1. A neutron spectrum rapid construction method in a simulation work site is characterized by comprising the following steps:
s1, making a sample data set of a neural network model;
s2, training to obtain a learned neural network model;
s3, deploying the learned neural network model to hardware;
and S4, searching a global optimal slowing-down configuration scheme by adopting a multi-objective optimization algorithm.
2. The method for fast constructing neutron spectrum in a simulation job site according to claim 1, wherein the manufacturing process of the sample data set comprises the following two steps:
s11, simulating a neutron energy spectrum of neutrons after slowing, scattering and shielding based on a Monte Carlo method to obtain sample data and a sample label;
the sample data comprises energy spectrum sample data and a moderation configuration scheme, wherein the energy spectrum sample data comprises a neutron source energy spectrum and simulated on-site neutron energy spectrum data, and the moderation configuration scheme is a sample label;
s12, preprocessing a plurality of groups of energy spectrum sample data and sample labels to manufacture a sample data set;
the sample data set comprises a training set and a test set, wherein the training set is used for training the neural network model, and the test set is used for verifying the accuracy of the neural network model for predicting the slowing configuration scheme and the sample label.
3. The method for rapidly constructing the neutron spectrum of the simulation job site according to claim 2, wherein in the step S11, a Monte Carlo simulation particle transport tool is adopted to perform moderation structure modeling on an isotope neutron source or an accelerator neutron source, and a series of simulation calculations are performed according to various moderation configuration schemes to obtain simulation site neutron spectrum data under different moderation configuration schemes;
the moderator arrangement includes moderator features in various moderator materials, various size structures and shapes, and various moderator material placement sequences that form a sample label.
4. The method according to claim 2, wherein in the step 12, the method for preprocessing the plurality of sets of energy spectrum sample data and sample labels comprises: and normalizing the energy spectrum sample data, performing numerical representation on the sample label, and making a sample data set for training and testing a neural network model.
5. The method for fast constructing a neutron spectrum in a simulation job site according to claim 2, wherein in the step 12, the method for preprocessing the energy spectrum sample data and the sample label further comprises: before the data is input into the neural network model, the process of random disturbance and batch processing of the sample data in the sample data set is performed.
6. The method for fast constructing neutron spectrum in a simulation job site according to claim 2, wherein the step S2 comprises the following four steps:
s21, constructing a multilayer neural network model, wherein the neural network model comprises but is not limited to a full-connection neural network model;
s22, constructing a loss function, processing input data through a multilayer network structure, and calculating a loss value;
s23, model training, namely performing iterative computation on the input training set, searching a minimum value of a loss value variance through logistic regression gradient descent, reversely transmitting a gradient to a parameter of the neural network model, and updating a weight value in the network according to an updating rule; adjusting and optimizing the hyper-parameters of the neural network model to obtain a learned neural network model;
and S24, verifying the accuracy of the learned neural network model through the test set.
7. The method according to claim 6, wherein in step S21, the neural network model is a fully-connected neural network model, and the fully-connected neural network model has a 5-layer structure: the system comprises 1 input layer, 3 hidden layers and 1 output layer, wherein the input layer is used for acquiring the energy spectrum sample data, the hidden layers are used for further abstracting the energy spectrum sample data, and the output layer outputs a prediction slowing-down configuration scheme according to the energy spectrum sample data.
8. The method for fast constructing neutron spectrum in a simulation job site according to claim 6, wherein in the step S22, the loss function is used to calculate a difference value between a predicted slowing-down configuration scheme output by the neural network model and a sample label, and the difference value is a loss value;
if the number of parameters of the predicted slowing-down configuration scheme output by the neural network model is 2, adopting a logarithmic loss function as the loss function;
and if the number of parameters of the predicted slowing-down configuration scheme output by the neural network model is more than 2, adopting a cross entropy loss function as the loss function.
9. The method for fast constructing neutron spectrum in a simulation job site according to claim 6, wherein in the step S23, the model training is to find the minimum value of the variance of the loss value in the training set, i.e. the minimum value point of the cost function, by iteratively updating the weight value of the neural network, and thus obtaining the neural network model with generalization capability;
in the iterative computation, the gradient of the cost function is computed and reversely transferred to the neural network model, and the network weight value is updated by combining with the learning rate super-parameter;
the cost function is a variance function of the loss value, and overfitting of the neural network model is prevented through a regularization method.
10. The method according to claim 6, wherein in step S24, the test set is used to verify the accuracy of the learned neural network model, and the learned neural network model can be deployed after verification passes without updating a network weight value during verification.
11. The method for rapidly constructing the neutron spectrum in the simulation job site according to any one of claims 1 to 10, wherein in step S3, after the learned neural network model is deployed to hardware, the on-site neutron spectrum and neutron source energy spectrum data to be simulated are input, and then the learned neural network model outputs a prediction slowing configuration scheme meeting the requirement of a set accuracy.
12. The method for fast constructing neutron spectrum in a simulation job site according to claim 11, wherein in step S4, the multi-objective optimization algorithm finds a globally optimal shielding configuration scheme based on actual engineering implementation factors, wherein the actual engineering implementation factors include but are not limited to economic cost, moderating material quality and moderating volume.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210527270.2A CN115166811A (en) | 2022-05-16 | 2022-05-16 | Method for quickly constructing neutron spectrum in simulation working site |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210527270.2A CN115166811A (en) | 2022-05-16 | 2022-05-16 | Method for quickly constructing neutron spectrum in simulation working site |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115166811A true CN115166811A (en) | 2022-10-11 |
Family
ID=83484090
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210527270.2A Pending CN115166811A (en) | 2022-05-16 | 2022-05-16 | Method for quickly constructing neutron spectrum in simulation working site |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115166811A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116070533A (en) * | 2023-03-09 | 2023-05-05 | 中国原子能科学研究院 | Neutron energy spectrum determination method |
-
2022
- 2022-05-16 CN CN202210527270.2A patent/CN115166811A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116070533A (en) * | 2023-03-09 | 2023-05-05 | 中国原子能科学研究院 | Neutron energy spectrum determination method |
| CN116070533B (en) * | 2023-03-09 | 2023-12-12 | 中国原子能科学研究院 | Neutron energy spectrum determination method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Haji et al. | Comparison of optimization techniques based on gradient descent algorithm: A review | |
| CN110544011B (en) | Intelligent system combat effectiveness evaluation and optimization method | |
| CN109118013A (en) | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based | |
| CN100520817C (en) | Improved performance of artificial neural network model in the presence of instrumental noise and measurement error | |
| CN102694800A (en) | Gaussian process regression method for predicting network security situation | |
| Li et al. | Heterogeneous ensemble learning with feature engineering for default prediction in peer-to-peer lending in China | |
| CN106991285A (en) | A kind of short-term wind speed multistep forecasting method and device | |
| Alzaeemi et al. | Examining the forecasting movement of palm oil price using RBFNN-2SATRA metaheuristic algorithms for logic mining | |
| CN110111606A (en) | A kind of vessel traffic flow prediction technique based on EEMD-IAGA-BP neural network | |
| CN114969953A (en) | Shield underpass tunnel optimization design method and device based on Catboost-NSGA-III | |
| CN116502455A (en) | Process parameter determination method and system for laser selective melting technology | |
| CN114565021A (en) | Financial asset pricing method, system and storage medium based on quantum circulation neural network | |
| CN112884236A (en) | Short-term load prediction method and system based on VDM decomposition and LSTM improvement | |
| CN115166811A (en) | Method for quickly constructing neutron spectrum in simulation working site | |
| CN115982141A (en) | Characteristic optimization method for time series data prediction | |
| CN113762602A (en) | Short-term wind speed prediction method for wind power plant | |
| Li et al. | A new multi-fidelity surrogate modelling method for engineering design based on neural network and transfer learning | |
| CN110363232B (en) | BP neural network-based millimeter wave detector interference effect evaluation method | |
| Varmantchaonala et al. | Quantum hybrid algorithm for solving SAT problem | |
| Guo et al. | Data mining and application of ship impact spectrum acceleration based on PNN neural network | |
| CN116542382A (en) | Sewage treatment dissolved oxygen concentration prediction method based on mixed optimization algorithm | |
| Dumont et al. | Hyperparameter optimization of generative adversarial network models for high-energy physics simulations | |
| CN113408059A (en) | Multi-objective optimization method for thickness of automobile composite bumper anti-collision beam | |
| CN113642785A (en) | Method, system and equipment for long-term prediction of space debris orbit based on prior information | |
| Chakraborty et al. | Multigoal-oriented dual-weighted-residual error estimation using deep neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |