Temperature instrument position determination method and temperature field prediction model training method

Abstract

The invention provides a method for determining the position of a temperature instrument of a solid-state reactor core and a method for training a temperature field prediction model. The method for determining the position of the temperature instrument comprises the following steps: determining a coordinate range in which the temperature instrument is allowed to be arranged according to a solid-state reactor core design scheme; in the coordinate range, p instrument coordinates are selected according to system sampling, and p is an integer larger than 1; calculating the importance degree of each instrument coordinate; and an instrument with high importance is selected as the arrangement position of the temperature instrument. The training method of the temperature field prediction model comprises the step that a training loss function comprises a regression loss item for a reactor core temperature field feature vector and a classification loss item for judging whether safety protection is triggered or not. Aiming at the problems of large temperature gradient and limited measurable position of a solid reactor core, the invention provides a method for evaluating the instrument coordinate importance degree through a deep learning neural network model, the instrument coordinates which are relatively important for temperature field prediction are screened out, and a corresponding temperature field prediction model is trained and formed.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of nuclear reactors, and in particular to a method for determining the position of a temperature instrument for a solid core and a method for training a temperature field prediction model. Background Art

[0002] For reactors with solid core design, the core material temperature limit is generally used as the safety criterion in reactor design and the monitoring and protection signal during reactor operation. However, there are many difficulties in measuring and predicting the temperature of solid cores. First, due to the nonlinear distribution of nuclear reaction heat generation and the higher thermal resistance in the solid core (relative to the core with coolant), the temperature difference at different positions in the solid core is very large; second, due to the compact layout in the solid core, it is usually impossible to set temperature measuring instruments at any position in the fuel assembly. Temperature instruments can only be set near the outer surface of the assembly, and the number of instruments is strictly limited; finally, the temperature response inside the solid core is a three-dimensional heat conduction problem with complex multi-boundary conditions (multiple internal heat sources, multiple cooling boundaries), and in actual operation, it is necessary to consider the disturbances caused by the random and unknowable decrease in local heat transfer performance in the solid core or the failure of several heat pipe elements to the temperature field. Its temperature response mechanism and distribution law are difficult to predict by general methods.

[0003] Existing temperature field prediction methods in solids are generally unable to solve the three-dimensional temperature response problem under complex boundary conditions, especially in the field of solid reactors where there is no relevant application experience. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for determining the position of a temperature instrument for a solid core and a method for training a temperature field prediction model, so as to solve the problem that the existing solid internal temperature field prediction method is not suitable for a solid core.

[0005] In order to solve the above technical problems, the present invention provides a method for determining the position of a temperature instrument for a solid-state core, comprising: determining a coordinate range within which the temperature instrument is allowed to be arranged according to a solid-state core design scheme; within the coordinate range, selecting p instrument coordinates by systematic sampling, where p is an integer greater than 1; calculating the importance of each instrument coordinate; sorting the instrument coordinates according to the importance of the instrument coordinates, deleting the z instrument coordinates with the lowest importance, where z < p, and using the remaining instrument coordinates as the arrangement position of the temperature instrument; wherein calculating the importance of each instrument coordinate comprises: according to the instrument coordinates, obtaining each temperature instrument under different working conditions according to the coordinate correspondence in the core temperature field data sample set; The temperature readings of the table are collected to establish an instrument sample set; the temperature readings of the j-th instrument coordinate are removed in turn from the instrument sample set, where 1≤j≤p, to form p first sample sets; the p first sample sets are taken as independent variables and the core temperature field data sample set is taken as the dependent variable, and deep learning neural network training is performed in turn to obtain the loss function of the p instrument coordinate combinations; the instrument sample set is taken as the independent variable and the core temperature field data sample set is taken as the dependent variable, and deep learning neural network training is performed to obtain the loss function of all instrument samples; the importance of each instrument coordinate is calculated according to the loss function of the p instrument coordinate combinations and the loss function of all instrument samples.

[0006] Optionally, the importance of each instrument coordinate is calculated by the following formula:

[0007]

[0008] Among them, W j is the importance of the jth instrument coordinate, is the loss function for eliminating the jth instrument coordinate combination, is the loss function of all instrument samples.

[0009] Optionally, the loss function of the instrument coordinate combination and the loss function of all instrument samples both include: a regression loss term for the core temperature field feature vector and a classification loss term for determining whether to trigger safety protection based on the temperature field.

[0010] Optionally, the loss function of the instrument coordinate combination and the loss function of all instrument samples adopt the following loss function:

[0011]

[0012] Among them, w r and w c is the hyperparameter of the predefined loss function, L tpr represents the regression loss term for the core temperature field feature vector, L ptcIt represents the classification loss item that determines whether to trigger the safety protection according to the temperature field, and n is the number of working condition samples.

[0013] Optionally, the regression loss term L is calculated using the following formula: tpr :

[0014]

[0015] Among them, n is the number of working condition samples, m is the number of field feature dimensions, x ij is the predicted value of the jth field characteristic dimension under the i-th working condition, is the original value of the jth field feature dimension under the i-th working condition.

[0016] Optionally, calculate the classification loss term L ptc The method comprises: labeling each working condition sample with a label indicating whether safety protection is triggered, and obtaining original labels of n working condition samples; labeling the working condition sample with a label indicating whether safety protection is triggered according to the predicted temperature reading, and obtaining predicted labels of n working condition samples; and calculating the classification loss term L according to the original label and the predicted label. ptc .

[0017] Optionally, the classification loss term L is calculated by the following formula: ptc :

[0018]

[0019] Where n is the number of working condition samples, T i is the predicted label of the i-th working condition, is the original label of the i-th working condition.

[0020] To solve the above technical problems, the present invention provides a training method for a temperature field prediction model, comprising: using three-dimensional heat transfer response analysis software to establish a core thermal analysis model, analyzing n operating conditions through the core thermal analysis model to obtain a core temperature field data sample set, where n is an integer greater than 1; marking each operating condition sample with a label indicating whether safety protection is triggered, and obtaining the original labels of the n operating condition samples; using the method described in any one of claims 1 to 7 to determine the position of a temperature instrument, arranging the temperature instrument at the position, and obtaining the temperature reading of each temperature instrument; using the temperature readings of all temperature instruments as independent variables and the core temperature field data sample set as dependent variables to perform deep learning neural network training to obtain a temperature field prediction loss function, wherein the temperature field prediction loss function includes a regression loss term for the core temperature field feature vector and a classification loss term for judging whether safety protection is triggered according to the temperature field; minimizing the temperature field prediction loss function to obtain a trained temperature field prediction model.

[0021] Optionally, it also includes: reducing the dimension of the core temperature field data in the core temperature field data sample set, and using the reduced-dimensional core temperature field data sample set as the dependent variable.

[0022] Optionally, dimensionality reduction of the core temperature field data includes: dividing the core temperature field data into multiple categories based on a clustering algorithm, calculating the category center of each category, and selecting a node closest to the category center in each category to retain in the data set.

[0023] Optionally, it also includes: further reducing the dimension of the remaining core temperature field data using principal component analysis.

[0024] Compared with the prior art, the present invention has the following advantages:

[0025] The method for determining the position of temperature instruments for a solid core of the present invention addresses the problems of large temperature gradients and limited measurable positions in solid cores. A method for evaluating the importance of instrument coordinates through a deep learning neural network model is proposed to screen out instrument coordinates with a high correlation with the core temperature field information as a basis for formulating a core measurement instrument layout plan.

[0026] The training method of the temperature field prediction model of the present invention, on the one hand, establishes the mapping relationship between the instrument measurement data and the temperature field information through the deep learning neural network technology, combines the recovery and restoration method of dimensionality reduction compression, and forms a temperature field prediction model for predicting the three-dimensional temperature field of the core; on the other hand, directly targeting the demand that the reactor should perform the protection function under abnormal conditions, the temperature field prediction model is designed, including introducing the evaluation of whether the protection function is correctly performed in the loss function, so that the temperature field prediction model has a higher sensitivity to the working conditions where the reactor should trigger safety protection. In addition, the present invention performs dimensionality reduction compression preprocessing on high-dimensional three-dimensional temperature field data, greatly reduces the data dimension at the cost of a very small amount of information loss, and significantly reduces the training overhead of the neural network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings are included to provide a further understanding of the present application. They are included and constitute a part of the present application. The accompanying drawings illustrate embodiments of the present application and together with the present specification serve to explain the principles of the present invention. In the accompanying drawings:

[0028] Figure 1 is a flow chart of a method for determining a temperature instrument position for a solid-state core according to an embodiment of the present invention;

[0029] Figure 2 yes Figure 1 Flow chart of an embodiment of step S13;

[0030] Figure 3is a flowchart of a method for calculating a classification loss term according to an embodiment of the present invention;

[0031] Figure 4 is a flow chart of a method for training a temperature field prediction model according to an embodiment of the present invention;

[0032] Figure 5 It is a flow chart of a method for reducing the dimension of core temperature field data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0034] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously. At the same time, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0035] Figure 1 FIG. 1 is a flow chart of a method for determining the position of a temperature instrument for a solid core according to an embodiment of the present invention. Figure 1 As shown, the method for determining the location of the temperature instrument for the solid core includes:

[0036] Step S11: Determine the coordinate range within which the temperature instrument can be arranged according to the solid-state core design plan.

[0037] Due to the compact layout inside the solid core, temperature measuring instruments cannot usually be installed at any position inside the fuel assembly. Temperature instruments can only be installed near the outer surface of the assembly. The coordinate range within which the temperature instruments are allowed to be installed is determined based on the solid core design plan.

[0038] Step S12: Within the coordinate range, select p instrument coordinates by systematic sampling, where p is an integer greater than 1.

[0039] Step S13: Calculate the importance of each instrument coordinate.

[0040] Figure 2 yes Figure 1 Flow chart of step S13 in an embodiment. Figure 2 As shown, the importance of calculating each instrument coordinate includes:

[0041] Step S131: According to the instrument coordinates, the temperature readings of each temperature instrument under different working conditions are obtained in the core temperature field data sample set according to the coordinate correspondence, and an instrument sample set is established.

[0042] In some embodiments, before entering step S131, the process further includes establishing a core temperature field data sample set. The process of establishing a core temperature field data sample set includes:

[0043] 1) According to the reactor operating condition classification and accident list, determine the reactor state range that needs to be covered by the solid core temperature field prediction method as a constraint condition of the data sample space, that is, the scope of application of the prediction method.

[0044] 2) Within the reactor state range, a certain number of typical operating conditions are selected by systematic sampling method, and a core thermal analysis model is established using 3D heat transfer response analysis software. These typical operating conditions are analyzed to obtain the corresponding fine 3D temperature field, thereby establishing a core temperature field data sample set covering different reactor states. There are n samples under different working conditions in the core temperature field data sample set, and the characteristic dimension of each sample is equal to the number of temperature field nodes M obtained by software analysis. Optionally, the three-dimensional heat transfer response analysis software includes but is not limited to ANSYS Mechanical, COMSOL and STARCCM+.

[0045] In step S131, according to the instrument coordinates, in the core temperature field data sample set X Mn The temperature response of each instrument under different working conditions is obtained by coordinate correspondence, and an instrument sample set is established. in Represents the temperature reading at each selected instrument coordinate in operating condition n.

[0046] Step S132: Remove the temperature reading of the j-th instrument coordinate in the instrument sample set in turn, where 1≤j≤p, to form p first sample sets.

[0047] In the instrument sample set Q pn The temperature readings of the jth instrument coordinates are removed in turn to form a sample set {Q1, Q2, ... Q p},in Represents the temperature readings of the selected p-1 instrument coordinates under each working condition. For example, the temperature reading of the first instrument coordinate is eliminated to obtain the first sample set Q1; the temperature reading of the second instrument coordinate is eliminated to obtain the first sample set Q2; and so on, the temperature reading of the pth instrument coordinate is eliminated to obtain the first sample set Q p .

[0048] Step S133: Taking the p first sample sets as independent variables and the core temperature field data sample set as dependent variable, deep learning neural network training is performed in turn to obtain the loss function of the p instrument coordinate combinations.

[0049] For example, the first sample set Q1 is used as the independent variable, and the core temperature field data sample set X Mn As the dependent variable, deep learning neural network training is performed to obtain the loss function of the instrument coordinate combination The first sample set Q2 is used as the independent variable, and the core temperature field data sample set X Mn As the dependent variable, deep learning neural network training is performed to obtain the loss function of the instrument coordinate combination Similarly, with the first sample set Q p As the independent variable, the core temperature field data sample set X Mn As the dependent variable, deep learning neural network training is performed to obtain the loss function of the instrument coordinate combination

[0050] Optionally, the core temperature field data sample set X can be first Mn Dimensionality reduction is performed to obtain the temperature field feature vector. Taking the p first sample sets as independent variables and the temperature field feature vector as the dependent variable, deep learning neural network training is performed in turn to obtain the loss function of the p instrument coordinate combinations.

[0051] Step S134: Using the instrument sample set as the independent variable and the core temperature field data sample set as the dependent variable, deep learning neural network training is performed to obtain the loss function of all instrument samples.

[0052] For example, the instrument sample set Q pn As the independent variable, the core temperature field data sample set X Mn As the dependent variable, deep learning neural network training is performed to obtain the loss function of all instrument samples

[0053] Optionally, the loss function of the instrument coordinate combination and the loss function of all instrument samples adopt the loss function L DNN Both include: regression loss term L for the core temperature field feature vector tpr and the classification loss term L that determines whether to trigger safety protection based on the temperature field ptc Optionally, the deep neural network training uses the following loss function L DNN :

[0054]

[0055] Among them, w r and w cis the hyperparameter of the predefined loss function, L tpr represents the regression loss term for the core temperature field feature vector, L ptc It represents the classification loss item that determines whether to trigger the safety protection according to the temperature field, and n is the number of working condition samples.

[0056] Optionally, the regression loss term L is calculated using the following formula tpr :

[0057]

[0058] Wherein, n is the number of working condition samples, and m is the number of field feature dimensions. In this embodiment, m may be the original feature dimension M of the sample, or the feature dimension after the sample is reduced in dimension. ij is the predicted value of the jth field characteristic dimension under the i-th working condition, is the original value of the jth field feature dimension under the i-th working condition.

[0059] Figure 3 FIG. 1 is a flowchart of a method for calculating a classification loss term according to an embodiment of the present invention. Figure 3 As shown, the method 300 for calculating the classification loss term includes:

[0060] Step S31: label each working condition sample with a label indicating whether safety protection is triggered, and obtain original labels of n working condition samples.

[0061] Since the main purpose of this method is to achieve safety protection of the reactor, the n operating condition samples are labeled with whether the safety protection is triggered T = {T1, T2, ... T n}, for working condition i, if the peak temperature exceeds the required temperature for safety protection, then T i =1, otherwise T i =0.

[0062] Step S32: label the operating condition sample with a label indicating whether safety protection is triggered according to the predicted temperature reading, and obtain predicted labels of n operating condition samples.

[0063] For example, based on whether the maximum temperature prediction value exceeds the set safety protection limit, a label can be added to the working condition sample to predict whether safety protection is triggered. Together with the original label T of the working condition sample, the loss term L is completed ptc Calculation.

[0064] Step S33: Calculate the classification loss term L based on the original label and the predicted label ptc .

[0065] Optionally, the classification loss term L is calculated by the following formula ptc :

[0066]

[0067] Where n is the number of working condition samples, T i is the predicted label of the i-th working condition, is the original label of the i-th working condition.

[0068] Step S135: Calculate the importance of each instrument coordinate according to the loss function of the p instrument coordinate combinations and the loss function of all instrument samples.

[0069] After the training of p deep learning neural networks is completed, the loss function of p instrument coordinate combinations is obtained, which is recorded as Where j = 1, 2, 3, ..., p. Use a set Q consisting of p instrument samples pn and temperature field information sample collection Perform deep neural network training to obtain the loss function of all instrument samples

[0070] Optionally, the importance of each instrument coordinate is calculated by the following formula:

[0071]

[0072] Among them, W j is the importance of the jth instrument coordinate, is the loss function for eliminating the jth instrument coordinate combination, is the loss function of all instrument samples.

[0073] Step S14: sorting the instrument coordinates according to their importance, deleting the z instrument coordinates with the lowest importance, where z is a positive integer less than p, and using the remaining instrument coordinates as the layout positions of the temperature instruments.

[0074] In some embodiments, the method further includes: sorting the instrument coordinates according to their importance, deleting the z instrument coordinates with the lowest importance, wherein the value of z can be adjusted as needed, and re-establishing a new instrument sample set Q p-z,n , repeat steps S13 to S14 until the number of remaining instrument coordinates is no greater than the upper limit required by the core instrument design.

[0075] In some embodiments, the step further includes: determining the classification loss item L for security protection effectiveness ptc Is it not greater than the upper limit required by the safety design? If the safety protection requirement cannot be met, the upper limit required by the core instrument design should be increased, the z instrument coordinates that were recently removed are put back into the sample set, and steps S13 to S14 and this step are repeated.

[0076] The method for determining the position of temperature instruments for a solid core of the present invention addresses the problems of large temperature gradients and limited measurable positions in solid cores. A method for evaluating the importance of instrument coordinates through a deep learning neural network model is proposed to screen out instrument coordinates with a high correlation with the core temperature field information as a basis for formulating a core measurement instrument layout plan.

[0077] Figure 4 FIG. 1 is a flow chart of a method for training a temperature field prediction model according to an embodiment of the present invention. Figure 4 As shown, the training method 400 of the temperature field prediction model includes:

[0078] Step S41: Use three-dimensional heat transfer response analysis software to establish a core thermal analysis model, analyze n operating conditions through the core thermal analysis model, and obtain a core temperature field data sample set, where n is an integer greater than 1.

[0079] Among them, establishing the core temperature field data sample set includes:

[0080] According to the reactor operating condition classification and accident list, the range of reactor states that need to be covered by the solid core temperature field prediction method is determined as a constraint on the data sample space, which is the applicable scope of the prediction method. This application constrains the sample space of the model training data based on the reactor operating condition classification and accident list to maximize the matching of reactor safety design requirements.

[0081] Within the reactor state range, a certain number of typical operating conditions are selected by systematic sampling method, and the core thermal analysis model is established using 3D heat transfer response analysis software. These typical operating conditions are analyzed to obtain the corresponding fine 3D temperature field, thereby establishing a core temperature field data sample set covering different reactor states. There are n samples under different working conditions in the core temperature field data sample set, and the characteristic dimension of each sample is equal to the number of temperature field nodes M obtained by software analysis. Optionally, the three-dimensional heat transfer response analysis software includes but is not limited to ANSYS Mechanical, COMSOL and STARCCM+.

[0082] Step S42: label each working condition sample with a label indicating whether safety protection is triggered, and obtain original labels of n working condition samples.

[0083] Since the main purpose of this method is to achieve the safety protection of the reactor, the n working condition samples are labeled with whether they are protected or not. n}, for working condition i, if the peak temperature exceeds the required temperature for safety protection, then T i =1, otherwise T i =0.

[0084] Step S43: using the temperature instrument position determination method of the present application to determine the position of the temperature instrument, arranging the temperature instrument at the position, and obtaining the temperature reading of each temperature instrument.

[0085] Step S44: Take the temperature readings of all temperature instruments as independent variables and the core temperature field data sample set as the dependent variable, perform deep learning neural network training, and obtain the temperature field prediction loss function. The temperature field prediction loss function includes a regression loss term for the core temperature field feature vector and a classification loss term for determining whether to trigger safety protection based on the temperature field.

[0086] Since the core temperature field data sample set X Mn Each sample is a detailed three-dimensional temperature field, usually containing 10 5 The above data nodes, that is, the feature dimension M of the data samples is up to 10 5 The above will greatly increase the computing resource overhead for subsequent research on temperature field prediction methods. Therefore, in some embodiments, the core temperature field data must be compressed and reduced in dimension before entering step S44, and the sample set of the core temperature field data after dimensionality reduction is used as the dependent variable.

[0087] Figure 5 FIG. 1 is a flow chart of a method for reducing the dimension of core temperature field data according to an embodiment of the present invention. Figure 5 As shown, the core temperature field data dimension reduction method 500 includes:

[0088] Step S51: Based on the clustering algorithm, the core temperature field data is divided into multiple categories, the category center of each category is calculated, and the node closest to the category center is selected in each category and retained in the data set.

[0089] In step S51, the data dimension reduction of the three-dimensional temperature field is specifically performed by the following method:

[0090] Step S511: Generally, the performance of temperature field data features at adjacent nodes is relatively close, so consider classifying and reducing the dimension according to the node coordinates. Reconstruct the sample set X T is the matrix X Mn The transpose of X T The meaning of each vector in is the value of the temperature field node i under n working conditions.

[0091] Step S512: Assume that the M temperature field nodes can be divided into k node categories, and the central temperature vectors of each category are k center temperature vectors form a center vector matrix The number of node categories k can be determined by continuous trial and error from large to small, and a more appropriate number of categories can be directly found. Generally, k≤n is required. Assign to the nearest center temperature vector based on vector distance The corresponding categories:

[0092]

[0093] Where D is the average distance from M temperature field nodes to their classification center, The central temperature vector with the smallest distance to each node vector is the vector that minimizes D in the set G. The value of D is minimized by optimizing the central vector matrix G.

[0094] Step S513: After the classification is completed, the node closest to the category center is selected in each category and retained in the data set to obtain After transposition, it is restored to a sample set representing different working conditions Each sample (the value of i is 1, 2, …, n) represents the temperature field information of working condition i after clustering and dimensionality reduction. At this time, the field feature dimension of the sample is k.

[0095] In general, the data dimension is still relatively high at this time, and principal component analysis (PCA) is further used to analyze the sample set X kn Further dimensionality reduction.

[0096] Step S52: principal component analysis is used to further reduce the dimension of the remaining core temperature field data.

[0097] Principal component analysis believes that in high-dimensional data, there are some features that carry less information and there is repetition between features (manifested as linear correlation), so it is necessary to merge features that carry repeated information and delete invalid features. The idea of ​​principal component analysis is to use sample variance to measure the information carried by features, thereby filtering and merging features. The specific implementation steps include:

[0098] Step S521: Center the sample features and subtract the average value of the features to form an adjusted sample set.

[0099]

[0100] Where n is the number of operating conditions, is the average value of field characteristic dimension j under n working conditions, x ij For the sample set X kn The original value of the field characteristic dimension j under the internal working condition i, It is the value obtained after the centralization of the characteristic dimension j of the working condition i.

[0101] Step S522: Calculate the covariance matrix:

[0102]

[0103] Where n is the number of operating conditions, After centralization The sample set composed of for is the transpose of , and C is the covariance matrix.

[0104] Step S523: Find the k eigenvalues ​​{λ1,λ2,…λ k} and its corresponding eigenvector

[0105] Step S524: Take the first m eigenvalues ​​from largest to smallest, satisfying:

[0106]

[0107] Among them, λ i is the i-th eigenvalue in the covariance matrix C, I s is the information retention coefficient, which represents the ratio of the sum of the eigenvalues ​​of the retained features to the sum of all eigenvalues. s is the information retention coefficient. l The information loss requirement is determined according to user needs, usually 10%. Ensure that the information retention rate of the matrix after principal component analysis compression is greater than 90%, and use its corresponding eigenvectors to form the principal component transformation matrix:

[0108]

[0109] in, are the retained eigenvalues ​​{λ1,λ2,…λ m} corresponding to the eigenvector.

[0110] Step S525: Use the transformation matrix B to reduce the dimension of the original sample set to form a temperature field information sample set after dimension reduction:

[0111]

[0112] in, Represents the temperature field information sample of working condition n, where the field feature dimension of the sample is m.

[0113] After the above two steps of dimensionality reduction, the sample set There are still n working condition samples in it, but the characteristic dimension of each sample is reduced to m. At this time, the sample value has no clear physical meaning.

[0114] The data dimension reduction process of the three-dimensional temperature field includes but is not limited to the above method. Any means of reducing the characteristic dimension of the temperature field information data carried by each working condition sample can be used in the present invention. The dimension reduction and compression method generally has a corresponding restoration method, which can restore most of the information of the original temperature field.

[0115] Step S45: Minimize the temperature field prediction loss function to obtain a trained temperature field prediction model.

[0116] In step S44, the temperature readings of all temperature instruments are used as independent variables, and the compressed core temperature field data sample set is used as the dependent variable to perform deep learning neural network training to obtain the temperature field prediction loss function.

[0117] Optionally, the temperature field prediction loss function is expressed as:

[0118]

[0119]

[0120]

[0121] Among them, w r and w c is the hyperparameter of the predefined loss function, L tpr represents the regression loss term for the core temperature field feature vector, L tpc represents the classification loss item for judging whether to trigger safety protection according to the temperature field, n is the number of working condition samples, m is the feature dimension of the compressed sample, x ij is the predicted value of the jth field characteristic dimension under the i-th working condition, is the original value of the jth field characteristic dimension under the i-th working condition, T i is the predicted label of the i-th working condition, is the original label of the i-th working condition.

[0122] Use the following formula to update the parameter θ in the deep learning neural network:

[0123]

[0124] When the loss function is minimized, the deep learning neural network training is considered complete.

[0125] The training method of the temperature field prediction model of the present invention, on the one hand, establishes the mapping relationship between the instrument measurement data and the temperature field information through the deep learning neural network technology, combines the recovery and restoration method of dimensionality reduction compression, and forms a temperature field prediction model for predicting the three-dimensional temperature field of the core; on the other hand, directly targeting the demand that the reactor should perform the protection function under abnormal conditions, the temperature field prediction model is designed, including introducing the evaluation of whether the protection function is correctly performed in the loss function, so that the temperature field prediction model has a higher sensitivity to the working conditions where the reactor should trigger safety protection. In addition, the present invention performs dimensionality reduction compression preprocessing on high-dimensional three-dimensional temperature field data, greatly reduces the data dimension at the cost of a very small amount of information loss, and significantly reduces the training overhead of the neural network model.

[0126] The basic concepts have been described above. Obviously, for those skilled in the art, the above invention disclosure is only used as an example and does not constitute a limitation of the present application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements and amendments to the present application. Such modifications, improvements and amendments are suggested in the present application, so such modifications, improvements and amendments still belong to the spirit and scope of the exemplary embodiments of the present application.

[0127] At the same time, the present application uses specific words to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.

[0128] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0129] Unless otherwise specifically stated, the relative arrangement, numerical expressions and numerical values ​​of the parts and steps set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to those of ordinary skill in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be considered as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so that once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0130] Some aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". The processor may be one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, or combinations thereof. In addition, various aspects of the present application may be expressed as computer products located in one or more computer-readable media, which include computer-readable program codes. For example, computer-readable media may include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, tapes ...), optical disks (e.g., compact disks CDs, digital versatile disks DVDs ...), smart cards, and flash memory devices (e.g., cards, sticks, key drives ...).

[0131] A computer-readable medium may include a propagated data signal containing computer program code, such as in baseband or as part of a carrier wave. The propagated signal may have a variety of manifestations, including electromagnetic, optical, etc., or a suitable combination. A computer-readable medium may be any computer-readable medium other than a computer-readable storage medium, which may be connected to an instruction execution system, device or apparatus to communicate, propagate or transmit a program for use. The program code on the computer-readable medium may be propagated via any suitable medium, including radio, cable, fiber optic cable, radio frequency signal, or similar medium, or any combination of the above mediums.

[0132] Similarly, it should be noted that in order to simplify the description of the disclosure of this application and thus help understand one or more embodiments of the invention, in the above description of the embodiments of this application, multiple features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the object of this application are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the single embodiment disclosed above.

[0133] Although the present application has been described with reference to the current specific embodiments, ordinary technicians in this technical field should recognize that the above embodiments are only used to illustrate the present application, and various equivalent changes or substitutions may be made without departing from the spirit of the present application. Therefore, as long as the changes and modifications to the above embodiments are within the essential spirit of the present application, they will fall within the scope of the claims of the present application.

Claims

1. A method for determining the position of a temperature instrument for a solid-state core, characterized in that: include: Determine the coordinate range for arranging temperature instruments according to the solid-state core design plan; Within the coordinate range, p instrument coordinates are selected by systematic sampling, where p is an integer greater than 1; Calculate the importance of each instrument coordinate; Sort the instrument coordinates according to their importance, delete the z instrument coordinates with the lowest importance, where z is a positive integer less than p, and use the remaining instrument coordinates as the layout positions of the temperature instruments; The step of calculating the importance of each instrument coordinate includes: According to the instrument coordinates, the temperature readings of each temperature instrument under different working conditions are obtained according to the coordinate correspondence in the core temperature field data sample set, and the instrument sample set is established; In the instrument sample set, the temperature reading of the jth instrument coordinate is removed in turn to form p first sample sets, where 1≤j≤p; Taking the p first sample sets as independent variables and the core temperature field data sample set as dependent variables, deep learning neural network training is performed in turn to obtain a loss function of the p instrument coordinate combinations; Taking the instrument sample set as an independent variable and the core temperature field data sample set as a dependent variable, deep learning neural network training is performed to obtain a loss function of all instrument samples; The importance of each instrument coordinate is calculated according to the loss function of the p instrument coordinate combinations and the loss function of all instrument samples.

2. The method for determining the position of a temperature instrument according to claim 1, characterized in that: The importance of each instrument coordinate is calculated by the following formula: Among them, W j is the importance of the jth instrument coordinate, is the loss function for eliminating the jth instrument coordinate combination, is the loss function of all instrument samples.

3. The method for determining the position of a temperature instrument according to claim 1, characterized in that: The loss function of the instrument coordinate combination and the loss function of all instrument samples both include: a regression loss term for the core temperature field feature vector and a classification loss term for determining whether to trigger safety protection based on the temperature field.

4. The method for determining the position of a temperature instrument according to claim 3, characterized in that: The loss function of the instrument coordinate combination and the loss function of all instrument samples adopt the following loss function: Among them, w r and w c is the hyperparameter of the predefined loss function, L tpr represents the regression loss term for the core temperature field feature vector, L ptc It represents the classification loss item that determines whether to trigger the safety protection according to the temperature field, and n is the number of working condition samples.

5. The method for determining the position of a temperature instrument according to claim 4, characterized in that: The regression loss term L is calculated using the following formula: tpr : Among them, n is the number of working condition samples, m is the number of field feature dimensions, x ij is the predicted value of the jth field characteristic dimension under the i-th working condition, is the original value of the jth field feature dimension under the i-th working condition.

6. The method for determining the position of a temperature instrument according to claim 4, characterized in that: Calculate the classification loss term L ptc include: Label each working condition sample with whether it triggers safety protection, and obtain the original labels of n working condition samples; Label the working condition samples according to the predicted temperature readings to indicate whether safety protection is triggered, and obtain the predicted labels of n working condition samples; The classification loss term L is calculated based on the original label and the predicted label ptc .

7. The method for determining the position of a temperature instrument according to claim 4, characterized in that: The classification loss term L is calculated by the following formula ptc : Where n is the number of working condition samples, T i is the predicted label of the i-th working condition, is the original label of the i-th working condition.

8. A method for training a temperature field prediction model, characterized in that: include: A core thermal analysis model is established using a three-dimensional heat transfer response analysis software, and n operating conditions are analyzed using the core thermal analysis model to obtain a core temperature field data sample set, where n is an integer greater than 1; Label each working condition sample with whether it triggers safety protection, and obtain the original labels of n working condition samples; Determine the position of the temperature instrument using the method according to any one of claims 1 to 7, arrange the temperature instrument at the position, and obtain the temperature reading of each temperature instrument; The temperature readings of all temperature instruments are used as independent variables, and the core temperature field data sample set is used as the dependent variable to perform deep learning neural network training to obtain a temperature field prediction loss function, wherein the temperature field prediction loss function includes a regression loss term for the core temperature field feature vector and a classification loss term for determining whether to trigger safety protection according to the temperature field; The temperature field prediction loss function is minimized to obtain a trained temperature field prediction model.

9. The method for training a temperature field prediction model according to claim 8, characterized in that: The temperature field prediction loss function is as follows: Among them, w r and w c is the hyperparameter of the predefined loss function, L tpr represents the regression loss term for the core temperature field feature vector, L ptc It represents the classification loss item that determines whether to trigger the safety protection according to the temperature field, and n is the number of working condition samples.

10. The method for training a temperature field prediction model according to claim 8, characterized in that: Also includes: The core temperature field data in the core temperature field data sample set are reduced in dimension, and the core temperature field data sample set after the dimension reduction is used as the dependent variable.

11. The method for training a temperature field prediction model according to claim 10, characterized in that: Dimensionality reduction of core temperature field data includes: The core temperature field data are divided into multiple categories based on a clustering algorithm, the category center of each category is calculated, and the node closest to the category center is selected in each category and retained in the data set.

12. The method for training a temperature field prediction model according to claim 11, characterized in that: Also includes: Principal component analysis is used to further reduce the dimension of the remaining core temperature field data.

Images