Method, medium and electronic device for determining a chf prediction model

Abstract

The application provides a CHF prediction model determination method, a medium and an electronic device, and relates to the technical field of data processing. The method comprises the following steps: introducing a first norm penalty term and a second norm penalty term into a composite loss function simultaneously, the first norm penalty term being associated with the sum of the absolute values of each coefficient included in a coefficient set in a CHF initial prediction model, and the second norm penalty term being associated with the sum of squares of each coefficient in the coefficient set. In the process of optimizing the CHF initial prediction model with the optimization target of minimizing the composite loss function, the first norm penalty term is used to induce the sparsity of the corresponding polynomial function of the model, and tends to accurately compress the coefficients corresponding to the working condition parameters with weak contribution to the predicted CHF to zero, which is equivalent to automatically removing the redundant terms in the polynomial function. The second norm penalty is used to control the complexity of the model, prevent the value of a certain coefficient in the coefficient set from becoming too large, and effectively improve the stability and anti-overfitting ability of the model.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for determining a CHF prediction model, a computer-readable medium, and an electronic device. Background Technology

[0002] Critical heat flux (CHF) is a critical safety parameter for the operation of thermal equipment. When the heat flux of the heating surface reaches the CHF value, the heat transfer mechanism between the heating surface and the working fluid will change abruptly, which can easily lead to safety accidents such as wall overheating and equipment burnout. Therefore, constructing a high-precision CHF prediction model to accurately predict the CHF value under different operating conditions is a core requirement for ensuring the safe and stable operation of thermal equipment. Summary of the Invention

[0003] In view of this, this application provides a method for determining a CHF prediction model, a computer-readable medium, and an electronic device.

[0004] Firstly, this application provides a method for determining a CHF prediction model, comprising:

[0005] Obtain the initial CHF prediction model, which is a polynomial function that weights the feature matrix of the input sample by a set of coefficients. The output parameter of the initial CHF prediction model is the CHF prediction value. The set of coefficients includes multiple coefficients, and each element in the feature matrix is ​​associated with different operating condition parameters.

[0006] Obtain a composite loss function, which includes a goodness-of-fit term and a composite penalty term. The goodness-of-fit term is used to characterize the deviation between the CHF predicted value and the CHF measured value. The composite penalty term includes a first norm penalty term and a second norm penalty term. The first norm penalty term is associated with the sum of the absolute values ​​of all coefficients in the coefficient set, and the second norm penalty term is associated with the sum of the squares of all coefficients in the coefficient set.

[0007] With minimizing the composite loss function as the optimization objective, the values ​​of each coefficient in the coefficient set are updated in combination with the first CHF test sample set to optimize the initial CHF prediction model, thereby obtaining the CHF prediction model. The first CHF test sample set includes the data corresponding to the feature matrix.

[0008] Secondly, this application provides an electronic device, comprising:

[0009] At least one processor; and

[0010] At least one memory storing instructions that, when executed individually or jointly by the at least one processor, cause the electronic device to perform the method as described in the first aspect.

[0011] Thirdly, this application provides a computer-readable medium storing computer program code that, when executed by a processor, implements the method described in the first aspect.

[0012] This application proposes a method for determining a CHF prediction model, comprising: simultaneously introducing a first-norm penalty term and a second-norm penalty term into a composite loss function. The first-norm penalty term is associated with the sum of the absolute values ​​of all coefficients in the coefficient set of the initial CHF prediction model, and the second-norm penalty term is associated with the sum of the squares of all coefficients in the coefficient set. In optimizing the initial CHF prediction model with the goal of minimizing the composite loss function, the first-norm penalty term induces sparsity in the polynomial function corresponding to the model, tending to precisely compress the coefficients corresponding to operating parameters that contribute little to CHF prediction to zero, effectively eliminating redundant terms in the polynomial function. Furthermore, the second-norm penalty term controls the model's complexity, preventing the value of any coefficient in the coefficient set from becoming excessively large, effectively improving the model's stability and resistance to overfitting. Based on this, the CHF prediction model obtained after optimization using the aforementioned composite loss function exhibits higher stability, stronger resistance to overfitting, and lower complexity compared to known CHF prediction models, resulting in higher CHF prediction accuracy. Attached Figure Description

[0013] The accompanying drawings are included to provide a further understanding of this application; they are incorporated into and constitute a part of this application. The drawings illustrate embodiments of this application and, together with this specification, serve to explain the principles of this application. In the drawings:

[0014] Figure 1 This is a flowchart illustrating a method for determining a CHF prediction model provided in an embodiment of this application;

[0015] Figure 2 This is a schematic diagram of a CHF prediction model optimization process provided in an embodiment of this application;

[0016] Figure 3 This is a flowchart illustrating another method for determining a CHF prediction model provided in an embodiment of this application;

[0017] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this application. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0019] As indicated in this application, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0020] Furthermore, this application uses specific terms to describe embodiments of the application. For example, "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic related to at least one embodiment of the application. Therefore, it should be emphasized and noted that "an embodiment," "one embodiment," or "an alternative embodiment" mentioned twice or more in different locations in this specification do not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of the application can be appropriately combined.

[0021] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0022] Furthermore, although the terminology used in this application is selected from commonly known and used terms, some terms mentioned in this application's specification may have been chosen by the applicant according to his or her judgment, and their detailed meanings are explained in the relevant sections of the description herein. Moreover, this application is to be understood not only by the actual terms used, but also by the meaning implied by each term.

[0023] This application uses flowcharts to illustrate the operations performed by an apparatus or device according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously. Furthermore, other operations may be added to these processes, or one or more steps may be removed from these processes.

[0024] As described above, constructing a high-precision CHF prediction model to accurately predict CHF values ​​under different operating conditions is a core requirement for ensuring the safe and stable operation of thermal equipment. Existing methods for determining CHF prediction models often suffer from the following problems:

[0025] First, the selection of parameters is fragmented, often only focusing on a single operating condition parameter (such as pressure or mass flow rate) and relating it to the CHF value, ignoring the coupling effect between multiple parameters, resulting in poor adaptability of the relational formula;

[0026] Second, the model fitting algorithm was not selected properly. The traditional linear regression algorithm is prone to overfitting and cannot effectively screen key parameters, resulting in low prediction accuracy of the final CHF prediction model.

[0027] Third, the development process lacks standardization, and the experimental data processing, model training and verification are disconnected, making it difficult to form a reusable integrated development system.

[0028] To alleviate at least one of the above problems, see Figure 1 One embodiment of this application proposes a method for determining a CHF prediction model, comprising the following steps:

[0029] S10: Obtain the initial CHF prediction model. This initial CHF prediction model is a polynomial function that uses a coefficient set to weightedly combine the feature matrix of the input samples. The output parameter of the initial CHF prediction model is the predicted CHF value. The coefficient set includes multiple coefficients. Each element in the feature matrix is ​​associated with different operating parameters, which may include core operating parameters affecting CHF prediction as well as other operating parameters. For example, core operating parameters may include three thermal parameters: the pressure P at the outlet of the rod bundle CHF experimental specimen, the mass flow rate G at the CHF occurrence location, and the local vapor content x at the CHF occurrence location. Other operating parameters may include pipe diameter Dh, heating length L, etc.

[0030] In some embodiments, the polynomial function corresponding to the initial prediction model for CHF is as follows:

[0031]

[0032] CHF predicted value, For the characteristic matrix, For the intercept term of the model, This is a set of coefficients. For example, assuming that each element in the feature matrix is ​​associated with a core operating condition parameter affecting CHF prediction, including pressure P, mass flow rate G, and local vapor content x, the feature matrix can be a row vector: = including higher-order terms (such as...) ) and interactive items (such as ), corresponding It can be a column vector, represented as:

[0033] .

[0034] One of them It includes n coefficients ( , , ..., The number of n is the same as the total number of elements in the feature matrix.

[0035] S11: Obtain the composite loss function, which includes a goodness-of-fit term and a composite penalty term. The goodness-of-fit term is used to characterize the deviation between the CHF predicted value and the CHF measured value. The composite penalty term includes a first-norm penalty term and a second-norm penalty term. The first-norm penalty term is associated with the sum of the absolute values ​​of all coefficients in the coefficient set, and the second-norm penalty term is associated with the sum of the squares of all coefficients in the coefficient set.

[0036] For example, the composite loss function The corresponding formula is as follows:

[0037]

[0038] in, The goodness-of-fit term is a mean squared error function, representing the average squared error between the model's predicted CHF values ​​and the measured CHF values. It is used to drive the model to reproduce the input-output mapping relationship implied in the data as accurately as possible. First norm penalty term: Second norm penalty term: The first norm penalty term induces sparsity when using the composite loss function. In optimizing the initial CHF prediction model, the first norm penalty term tends to precisely compress the coefficients of operating parameters that contribute little to CHF prediction to zero, essentially automatically eliminating redundant terms in the polynomial function and achieving feature selection during model training. The second norm penalty term controls the model's complexity and prevents coefficient settling. When the value of a certain coefficient becomes too large, it effectively improves the stability of the model and its resistance to overfitting.

[0039] S12: With minimizing the composite loss function as the optimization objective, the values ​​of each coefficient in the updated coefficient set of the first CHF test sample set are combined to optimize the initial CHF prediction model, thus obtaining the CHF prediction model.

[0040] The first CHF test sample set includes the data corresponding to the feature matrix, specifically, it can be understood as including the data corresponding to each element in the feature matrix. For example, assuming that each element in the feature matrix is ​​associated with the core operating condition parameters affecting CHF prediction, and these core operating condition parameters include pressure P, mass flow rate G, and local vapor content x, then each sample in the first CHF test sample set includes the data corresponding to pressure P, mass flow rate G, and local vapor content x, as well as the measured CHF value. Taking the i-th sample in the first CHF test sample set as an example, it can be represented as follows: , Let these represent the pressure, mass flow rate, and local vapor content of the i-th sample, respectively. This represents the measured CHF value of the i-th sample.

[0041] In some embodiments, the specific implementation process of S12 is as follows: substituting each sample in the first CHF test sample set into the polynomial function corresponding to the initial CHF prediction model to minimize the composite loss function. To optimize the objective, the coefficient set is continuously iterated and updated. The values ​​of each coefficient and the intercept term The final output is the optimal set of coefficients. and intercept term The initial CHF prediction model has been optimized, resulting in a CHF prediction model. For example, the CHF prediction model can be a polynomial function as follows: Coefficient set ( , , ..., The parameters are determined through iterative optimization of the model. Their higher-order terms and interaction terms quantitatively characterize the nonlinearity and coupling effects between parameters. Thanks to the regularization induced by the first-norm penalty term, the generated polynomial is structurally sparse, automatically eliminating unimportant higher-order terms and ensuring the simplicity of the formula. This allows the CHF prediction model to generate explicit coupled prediction relationships with clear physical meaning and direct application, significantly enhancing its engineering practical value. Compared with traditional "black box" machine learning models, it has the following advantages: transforming the "black box" machine learning model into an explicit mathematical formula with clear physical meaning not only facilitates prediction but also directly reveals the coupling mechanism between multiple parameters, greatly enhancing the interpretability and engineering acceptance of the results.

[0042] This embodiment executes steps S10-S12 as described above, simultaneously introducing a first norm penalty term and a second norm penalty term into the composite loss function. The first norm penalty term is associated with the sum of the absolute values ​​of all coefficients in the coefficient set of the initial CHF prediction model, and the second norm penalty term is associated with the sum of the squares of all coefficients in the coefficient set. In optimizing the initial CHF prediction model with the goal of minimizing the composite loss function, the first norm penalty term induces sparsity in the polynomial function corresponding to the model, tending to precisely compress the coefficients corresponding to operating parameters that contribute little to CHF prediction to zero, effectively eliminating redundant terms in the polynomial function. Furthermore, the second norm penalty is used to control the model's complexity, preventing the value of a certain coefficient in the coefficient set from becoming excessively large, effectively improving the model's stability and resistance to overfitting. Based on this, the CHF prediction model obtained after optimization using the composite loss function is more stable, has stronger resistance to overfitting, and lower complexity than known CHF prediction models; correspondingly, its CHF prediction accuracy is also higher.

[0043] To further balance the complexity and stability of the final CHF prediction model and simplify the corresponding polynomial function, another embodiment of this application introduces hyperparameters into the composite penalty term, including an intensity parameter λ and a mixing ratio parameter ρ. The composite penalty term is a weighted combination of the first norm penalty term and the second norm penalty term adjusted by the intensity parameter λ and the mixing ratio parameter ρ. For example, the composite penalty term is given by the following formula:

[0044] .

[0045] in, This is the first norm penalty term (LASSO part). This is the second norm penalty term (ridge regression part).

[0046] Hyperparameter description:

[0047] λ (intensity parameter): λ>0, controls the strength of the influence of the entire compound penalty term on the loss function. The larger λ is, the simpler the model tends to be (the weights tend to be smaller or approach zero); the smaller λ is, the more the model tends to fit the training data.

[0048] ρ (mixing ratio parameter): 0≤ρ≤1 is the most innovative design parameter in this application embodiment. It finely adjusts the balance between the first norm penalty term and the second norm penalty term: when ρ=1, the model degenerates into pure LASSO regression, focusing on feature selection; when ρ=0, the model degenerates into pure ridge regression, focusing on stabilizing weights and handling collinearity; when 0<ρ<1, the model is a true "elastic network", possessing both the feature selection capability of LASSO and the stabilization capability of ridge regression.

[0049] In some embodiments, the implementation process of step S12 above may include: determining the current optimal intensity parameter λ and the current optimal mixing ratio parameter ρ using a search algorithm and a first CHF test sample set, with minimizing the composite loss function as the optimization objective and a predefined search space. The predefined search space characterizes the numerical search range for the current optimal λ and ρ. Further, based on the current optimal intensity parameter λ, the current optimal mixing ratio parameter ρ, and the first CHF test sample set, with minimizing the composite loss function as the optimization objective, the values ​​of each coefficient in the current coefficient set are further optimized to obtain the CHF prediction model. The predefined search space is λ > 0, 0 ≤ ρ ≤ 1.

[0050] For example, suppose the polynomial function corresponding to the initial prediction model for CHF is: The predefined search space is 0.001≤λ≤0.1, 0≤ρ≤1. The process of optimizing the initial CHF prediction model based on the composite loss function can be found in [reference needed]. Figure 2 As shown, it is divided into two stages. The first stage can be understood as the optimal hyperparameter search stage, in which the goal is to minimize the composite loss function. To optimize the objective, the first CHF experimental sample set was used in conjunction with 5-fold cross-validation and a grid / Bayesian search. Within a search space of 0.001 ≤ λ ≤ 0.1 and 0 ≤ ρ ≤ 1, the current optimal intensity parameter λ and the current optimal mixing ratio parameter ρ were searched to ensure the model's generalization performance under different CHF conditions (e.g., high / low pressure, high / low vapor content). It is understandable that during the search for the optimal λ and ρ, the initial CHF prediction model is also continuously optimized, i.e., the coefficient set... The coefficients in the model are also constantly being updated and iterated. After the first stage is completed, the second stage is executed. Based on the optimal λ and ρ, and using the first CHF experimental sample set as the optimization objective, the current coefficient set is further optimized to minimize the composite loss function. Given the values ​​of each coefficient, output the optimal set of coefficients. and intercept term The CHF prediction model was obtained. Efficient synergistic optimization of model accuracy, simplicity, and generalization ability was achieved, improving the overall performance of the prediction model. Specifically, a weighted composite penalty term of the first norm penalty term and the second norm penalty term was innovatively introduced into the loss function, enabling the model to automatically possess both the feature selection capability of LASSO and the stability of ridge regression. A mixing ratio parameter ρ was set for optimization, allowing the model to adapt to data characteristics and find the optimal balance between sparsity and stability. Furthermore, cross-validation and intelligent search strategies were used to optimize the hyperparameters (λ, ρ), ensuring the model's optimal generalization performance on unknown data. In addition, through a single unified optimization objective, three key objectives were simultaneously achieved: (1) minimizing prediction error; (2) automatically selecting the most relevant features; and (3) ensuring the model's stability under complex multi-condition scenarios. This design is particularly suitable for engineering problems like CHF prediction, which have a small number of features but complex relationships and high requirements for model interpretability, providing a solid mathematical foundation for extracting accurate, robust, and physically meaningful weight coefficients from the data.

[0051] In some embodiments, the above feature matrix The elements included are related to the core operating parameters that affect CHF prediction (such as pressure P, mass flow rate G, and local vapor content x). The process of determining the core operating parameters includes the following steps:

[0052] Step 1: Obtain the CHF test dataset for the CHF test specimen. Each data sample in the CHF test dataset includes multiple corresponding operating parameters (such as pressure P, mass flow rate G, vapor content x, pipe diameter D). h (e.g., heating length L) and measured CHF value.

[0053] For example, the test conditions cover multiple operating environments with pressures ranging from 0.5 to 15 MPa, mass flow rates from 200 to 1600 kg / (m²·s), and local vapor content ranging from 0 to 1. For each operating condition, the outlet pressure of the rod bundle CHF test specimen is simultaneously acquired. P Mass flow rate at the location where CHF occurs G Local vapor content at the location of CHF occurrence x Record the operating conditions parameters and the corresponding CHF measured values. Organize the records into an N1-row, N2-column structured data matrix, where each row represents a CHF test data sample. To form the CHF experimental dataset D raw .

[0054] Furthermore, continuing with the above example, after obtaining... D raw After that, you can... D rawPerform data preprocessing before proceeding to step 2. Data preprocessing aims to clean the data and improve its quality. For example, under the assumption of a normal distribution, 99.7% of the data should fall within (…). μ -3 σ , μ +3 σ Points outside this range are most likely caused by sensor malfunction, transient interference, or recording errors. The data preprocessing process can be as follows: using 3... σ The criteria identify and remove outliers caused by sensor fluctuations or operational errors. The specific removal process can be as follows: First, calculate the mean value for the CHF value column. μ and standard deviation σ, Further, one by one, each CHF i If |CHF i - μ |>3 σ If it is an outlier, the entire row of outlier data (including its corresponding...) will be removed. P, G, x )from D raw Remove from the middle. In addition, data preprocessing may also include removing redundant data from duplicate collections: comparing the remaining data row by row and removing all features ( P, G, x The CHF dataset is cleaned by creating identical duplicate rows to ensure the uniqueness of each sample. D clean No abnormalities, no duplicates, and no missing data.

[0055] Step 2: Calculate the MIC value between each operating condition parameter and CHF using the Maximum Information Coefficient (MIC) method, and filter the candidate operating condition parameter set associated with CHF from multiple operating condition parameters based on the MIC value.

[0056] Step 3: Use the random forest feature importance method to determine the importance score of each candidate operating condition parameter in the candidate operating condition parameter set for predicting CHF, and select the core operating condition parameters from the candidate operating condition parameter set based on the importance score.

[0057] By implementing steps 1-3 above to screen core operating parameters, this approach aims to overcome the limitations of traditional linear correlation analysis and solve the technical problem of comprehensively and accurately identifying CHF influencing factors under complex thermal-hydraulic conditions. A dual-mode feature screening method, "generalized correlation measurement-prediction contribution verification," is provided to robustly identify the core physical parameters (i.e., core operating parameters) that have a decisive impact on CHF from a candidate set containing multiple types of operating parameters. This lays a solid foundation for building a high-precision and concise prediction model. The training samples used in subsequent model training, such as the first, second, and third CHF experimental sample sets, can all contain only the core operating parameters and their corresponding measured CHF values. This provides the most concise and relevant feature set for subsequent modeling, fundamentally avoiding the curse of dimensionality and overfitting risks, and directly improving the prediction accuracy and generalization ability of the final prediction model.

[0058] Traditional feature selection often relies on linear measures such as the Pearson correlation coefficient, which struggles to capture the complex nonlinear dependencies commonly found between thermal parameters and CHF (chromatic scattering). This can easily lead to the omission of key features or the introduction of irrelevant features. This application's embodiment, by implementing steps 1-3 above, first introduces the maximum information coefficient (MIC) as a generalized association probe. Based on information theory, MIC calculates and normalizes the maximum mutual information between two variables by dynamically optimizing the grid partitioning on the variable scatter plot. Its core advantage lies in its ability to unbiasedly measure the statistical association strength of arbitrary functional forms (including highly nonlinear and non-functionalized forms). For example, for a dataset containing n samples... In the dataset, the MIC value between the operating condition parameter X and the target CHF is defined as:

[0059]

[0060] in, G A grid division on a scatter plot, | G | represents the total number of grid cells. G X | and | G Y | These represent the number of intervals into which the X-axis and Y-axis are divided. I ( X , Y | G ) is in the grid G Mutual information under computation B ( n) is the upper limit of the grid size. By calculating the MIC values ​​of various operating parameters (such as pressure P, mass flow rate G, vapor content x, pipe diameter Dh, heating length L, etc.) in the CHF test dataset with CHF, and comparing them with the preset threshold, a subset of parameters that have a significant generalized statistical correlation with CHF (i.e., the above candidate operating parameter set) can be preliminarily screened out, ensuring that important influencing factors are not missed due to the nonlinearity of the relationship.

[0061] To further verify the reliability of the preliminary screening results and eliminate the spurious interference that pure statistical associations might introduce, a Random Forest model was introduced as a predictive contribution validator based on the aforementioned candidate operating condition parameter set. Using the candidate operating condition parameter set selected by MIC as features, a Random Forest regression model was trained, and feature importance scores (i.e., the aforementioned importance scores) based on Gini impurity reduction or ranking importance were extracted. These scores quantify the marginal contribution of each feature to accurately predicting CHF from a practical perspective of "improving model predictive performance." Finally, combining the generalized association strength revealed by MIC and the predictive contribution evaluated by the Random Forest, parameters that performed well in both (e.g., operating condition parameters with MIC values ​​greater than a preset threshold and ranking in the top 3 importance scores) were determined as core operating condition parameters. This method of determining core operating condition parameters has the following advantages:

[0062] 1. Overcoming the challenge of nonlinear correlation identification: The maximum information coefficient (MIC) is used to replace the traditional linear correlation analysis, achieving "blind-zone-free" capture of the complex nonlinear dependence between thermal parameters and CHF.

[0063] 2. Establish a dual-mode closed-loop validation framework: Through the synergistic verification of "Statistical Association Measure (MIC)" and "Model Performance Driven Validation (Random Forest)," a robust screening logic is formed, ensuring that the selected features have both statistical significance and predictive contribution.

[0064] 3. Optimize the model input base: Provide the most concise and relevant feature set for subsequent modeling, fundamentally avoiding the curse of dimensionality and the risk of overfitting, and directly improving the prediction accuracy and generalization ability of the final relation.

[0065] See Figure 3 To verify the reliability of the CHF determined above and ensure that it meets the required engineering practical standards, one embodiment of this application proposes another method for determining the CHF prediction model. Figure 1 In addition, the following steps are also included:

[0066] S30: Obtain the second CHF test sample set, and divide each second CHF test sample in the second CHF test sample set into multiple sub-intervals based on the core operating condition parameters and the preset operating condition sub-domain partitioning rules. The second CHF test sample set is different from the first CHF test sample set and is an independent validation set completely distinct from the first CHF test sample set.

[0067] S31: Based on the CHF prediction model, calculate the interval performance index corresponding to each sub-interval. The interval performance index includes any one or more of the following: sample size, mean absolute percentage error, coefficient of determination, relative error distribution, mean and standard deviation.

[0068] Each sub-interval corresponds to a specific operating condition. For example, the core operating condition parameters include pressure P, mass flow rate G, and local vapor content x. Based on a preset sub-interval partitioning rule, each second CHF test sample in the second CHF test sample set is divided into multiple sub-intervals. The sub-interval partitioning process is as follows:

[0069] a. Divide the pressure P into 3 zones: low pressure zone: P∈[0.5, 5.0) MPa; medium pressure zone: P∈[5.0, 10.0) MPa; high pressure zone: P∈[10.0, 15.0] MPa.

[0070] b. Divide the flow rate into three intervals according to the mass flow rate G: low velocity region: G∈[200, 600) kg / (m²·s); medium velocity region: G∈[600, 1000) kg / (m²·s); high velocity region: G∈[1000, 1600] kg / (m²·s).

[0071] c. Divide the region into four zones based on local vapor content α: subcooled zone: α<0; low vapor content zone: 0≤α<0.3; medium vapor content zone: 0.3≤α<0.7; high vapor content zone: α≥0.7.

[0072] Based on the above division, theoretically, 3×3×4=36 sub-intervals can be formed by combining 3 different pressure zones, 3 different flow velocity zones, and 4 different vapor content zones. For example, sub-interval 1 is a combination of low pressure zone, low flow velocity zone, and subcooled zone; sub-interval 2 is a combination of medium pressure zone, low flow velocity zone, and subcooled zone. Each sub-interval represents a combination of the three core operating parameters under different operating conditions, with one sub-interval corresponding to one operating condition. However, in practical applications, as an optional approach, based on data distribution and project importance, key sub-intervals can be selected for focused analysis, such as: high-risk operating condition combinations: high pressure zone, low flow velocity zone, and high vapor content zone combination; typical operating condition combinations: medium pressure zone, medium flow velocity zone, and medium vapor content zone combination; boundary operating condition combinations: extreme value combinations of each parameter. Furthermore, for each sub-interval... Sk (k=1,2,...,K), calculate the following performance metrics:

[0073] Sample size N k : N k = The number of samples within this sub-interval.

[0074] Mean Absolute Percentage Error (MAPE) k :

[0075] .

[0076] Relative error distribution Calculate the statistical properties of the relative error of all samples within this sub-interval:

[0077] .

[0078] mean : .

[0079] Standard deviation : .

[0080] It should be noted that each of the above sub-intervals S k It can be any of the above 36 sub-intervals, or any of the above key sub-intervals, without limitation.

[0081] S32: Based on the CHF prediction model and the second CHF experimental sample set, determine the global performance index corresponding to the CHF prediction model. The global performance index includes any one or more of the following: mean absolute percentage error (MAPE) and coefficient of determination (R²). R² is the proportion of the model's ability to explain data variation, reflecting the goodness of fit between the predicted value and the actual value. The closer the value is to 1, the more actual data variation the model can explain and the better the fit.

[0082] S33: If the global performance index and the interval performance index corresponding to each sub-interval all meet the preset standards, then the CHF prediction model is confirmed to have passed the verification.

[0083] In some embodiments, the preset criteria may include global preset criteria and local preset criteria. Global preset criteria may be, for example, MAPE ≤ 5% and R² ≥ 0.92. Local preset criteria are any two of the following: performance achievement criteria, relative performance criteria, and trend consistency criteria. If, based on the interval performance indicators corresponding to each sub-interval, any two of the performance criteria, relative performance criteria, and trend consistency criteria are achieved for that sub-interval, then it is determined that the local preset criteria are met. When the global performance indicator meets the global preset criteria, and the interval performance indicators of each sub-interval meet the local preset criteria, then it is determined that both the global performance indicator and the interval performance indicators corresponding to each sub-interval meet the preset criteria.

[0084] For example, the above absolute performance criterion is: the performance of all sub-intervals. All must not exceed a preset threshold (e.g., 12%), and at least 90% of the sub-intervals must be within this threshold. ≤8%. The relative performance criterion is: for each sub-interval The ratio to the global MAPE should be within a reasonable range, meaning that no subinterval should be allowed to exceed this range. More than twice the global MAPE, and all sub-intervals The ratio to the global MAPE should be within the range of [0.5, 2.0]. The trend consistency criterion is that the coefficient of determination of each sub-interval should be greater than 0.85, and the error distribution should not show a significant systematic bias (| |Should be less than 5%, It should be within the range of [-1, 1].

[0085] In some embodiments, if the global performance metric and the interval performance metric corresponding to each sub-interval do not meet the preset criteria, the following optimization steps are performed:

[0086] Step 1: Based on the weak area diagnosis rule, locate at least one performance weak area from multiple sub-intervals, and analyze the common operating condition characteristics and error distribution characteristics of the at least one performance weak area. For example, the weak area diagnosis rule can be based on MAPE... k The top three sub-intervals are identified as weak performance areas. Common operating conditions can be, for example, a combination of specific operating conditions such as high steam content and high pressure. Error distribution characteristics can be, for example, positive or negative deviation.

[0087] Step 2: Obtain target samples that match the common operating condition characteristics from the second CHF test sample set, and construct a CHF correction model based on the target samples.

[0088] Step 3: Using weighting coefficients A weighted combination of the CHF correction model and the CHF prediction model yields the first target CHF prediction model. Among them, The numerical range is determined based on the error distribution characteristics. The final value of the weighting coefficients is determined by iteratively optimizing the first-target CHF prediction model using the third CHF experimental sample set, with the goal of minimizing the composite loss function. For example, assuming the error distribution has a positive bias, then 0 < If the error is ≤0.5 and the error distribution is a negative deviation, then... >0.5. The third CHF test sample set mentioned above may be the same as or different from the second CHF test sample set mentioned above, and there is no limitation on this.

[0089] For example, the formula corresponding to the first target CHF prediction model is as follows:

[0090]

[0091] in, For the first objective CHF prediction model, ω(X) is the weighting coefficient. For CHF prediction models, This is a CHF correction model. Assuming the common operating condition is characterized by high vapor content, all samples corresponding to high vapor content can be obtained from the second CHF test sample set as target samples. Based on these target samples, the polynomial function is then modified. A separate optimization was performed to obtain a CHF correction model.

[0092] Furthermore, after obtaining the first target CHF prediction model Subsequently, the third CHF experimental sample set can be used to iteratively optimize the first-objective CHF prediction model with the goal of minimizing the composite loss function, thus determining the second-objective CHF prediction model. The second-objective CHF prediction model can then be validated again using the validation set, repeating this process until one of the following convergence conditions is met: a) all sub-intervals MAPE k All are ≤8% and global MAPE ≤5b); the global MAPE improvement after two consecutive iterations is less than 0.1%; c) the maximum number of iterations is reached (e.g., 10 times). An automatic cycle of "prediction → verification → diagnosis → optimization" is formed until the preset standards are met in all key operating conditions, fundamentally ensuring the robustness of the delivered results.

[0093] In this embodiment, the CHF prediction model is validated by introducing a completely independent validation set (i.e., the second CHF test sample set mentioned above). This not only evaluates the global prediction accuracy but also performs local performance diagnosis, comprehensively assessing the performance of the CHF prediction model. If the model validation fails, a targeted CHF correction model can be created by combining the common operating conditions and error distribution characteristics of the weak performance areas to correct the original CHF prediction model, achieving targeted optimization of the model rather than simple repetitive training.

[0094] This application also provides a chip, including a circuit system configured to perform the life assessment method for nuclear power materials described above. The chip includes a Field Programmable Gate Array (FPGA) chip, a Complex Programmable Logic Device (CPLD) chip, and an Application Specific Integrated Circuit (ASIC) chip, etc.

[0095] Figure 4 This is a simplified block diagram of an electronic device 400 suitable for implementing embodiments of this application. For example, the life assessment method for nuclear power materials described above can be implemented by the electronic device 400. As shown, the electronic device 400 includes one or more processors 410, one or more memories 420 coupled to the processors 410, and one or more communication modules 440 coupled to the processors 410.

[0096] Communication module 440 is used for bidirectional communication. Communication module 440 has at least one antenna to facilitate communication. The communication interface can represent any interface necessary for communication with other network elements.

[0097] Processor 410 can be of any type suitable for a local technology network, and as a non-limiting example, can include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture. Electronic device 400 can have multiple processors, such as application-specific integrated circuit (ASIC) chips, which are timely driven to a clock that synchronizes with the main processor.

[0098] Memory 420 may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, read-only memory (ROM), electrically programmable read-only memory (EPROM), flash memory, hard disk, optical disc (CD), digital video disc (DVD), and other magnetic and / or optical storage. Examples of volatile memories include, but are not limited to, random access memory (RAM) and other volatile memories that do not persist during power-off periods.

[0099] Computer program 430 includes computer-executable instructions that are executed by the associated processor 410. Program 430 may be stored in ROM 424. Processor 410 may perform any appropriate actions and processes by loading program 430 into RAM 422.

[0100] The embodiments of this application can be implemented by program 430, enabling electronic device 400 to execute the reference. Figure 1 or Figure 3 Any process disclosed in the discussion. Embodiments of this application may also be implemented by hardware or by a combination of software and hardware.

[0101] In some embodiments, program 430 may be tangibly contained in a computer-readable medium, which may be contained in an electronic device 400 (e.g., memory 420) or other storage device accessible to the electronic device 400. The electronic device 400 may load program 430 from the computer-readable medium into RAM 422 for execution. The computer-readable medium may include any type of tangible non-volatile memory, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc. Program 430 is stored on the computer-readable medium.

[0102] Generally, the various embodiments of this application can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects may be implemented in hardware, while others may be implemented in firmware or software, which may be executed by a controller, microprocessor, or other electronic device. Although various aspects of the embodiments of this application are shown and described as block diagrams, flowcharts, or other graphical representations, it should be understood that, as non-limiting examples, the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other electronic devices, or some combination thereof.

[0103] This application also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the aforementioned references. Figure 1 or Figure 3 The method described herein. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of a program module can be combined or separated among program modules as needed. The machine-executable instructions used in the program module can execute on a local or distributed device. In a distributed device, the program module can reside on both local and remote storage media.

[0104] The program code used to perform the methods of this application can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program code is executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code can be executed entirely on a machine, partially on a machine, partially on a remote machine, partially on a remote machine, or entirely on a remote machine or server as a standalone software package.

[0105] In the context of this application, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc.

[0106] Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or apparatuses, or any suitable combination thereof. More specific examples of computer-readable storage media include electrical connections having one or more wires, portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0107] Furthermore, although the operations are described in a specific order, this should not be construed as requiring that these operations be performed in the specific order or sequence shown, or that all of the operations shown be performed to obtain the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the foregoing discussion, these details should not be construed as limiting the scope of this application, but rather as descriptions of features specific to particular embodiments. Certain features described in the context of a single embodiment may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0108] Although this application has been described in language specific to structural features and / or methodological behavior, it should be understood that the application as defined in the appended claims is not necessarily limited to the specific features or behaviors described above. Rather, the specific features and actions described above are disclosed as exemplary forms for implementing the claims.

Claims

1. A method for determining a CHF prediction model, characterized in that, include: Obtain the initial CHF prediction model, which is a polynomial function that weights the feature matrix of the input sample by a set of coefficients. The output parameter of the initial CHF prediction model is the CHF prediction value. The set of coefficients includes multiple coefficients, and each element in the feature matrix is ​​associated with different operating condition parameters. Obtain a composite loss function, which includes a goodness-of-fit term and a composite penalty term. The goodness-of-fit term is used to characterize the deviation between the CHF predicted value and the CHF measured value. The composite penalty term includes a first norm penalty term and a second norm penalty term. The first norm penalty term is associated with the sum of the absolute values ​​of all coefficients in the coefficient set, and the second norm penalty term is associated with the sum of the squares of all coefficients in the coefficient set. With minimizing the composite loss function as the optimization objective, the values ​​of each coefficient in the coefficient set are updated in combination with the first CHF test sample set to optimize the initial CHF prediction model, thereby obtaining the CHF prediction model. The first CHF test sample set includes the data corresponding to the feature matrix. Obtain the second CHF test sample set, and divide each second CHF test sample in the second CHF test sample set into multiple sub-intervals based on the core operating condition parameters and the preset operating condition sub-domain division rules. Each sub-interval corresponds to one operating condition. Based on the CHF prediction model, the interval performance index corresponding to each sub-interval is calculated. The interval performance index includes any one or more of the following: sample size, mean absolute percentage error, relative error distribution, mean and standard deviation. Based on the CHF prediction model and the second CHF test sample set, the global performance index corresponding to the CHF prediction model is determined. The global performance index includes any one or more of the following: mean absolute percentage error and coefficient of determination. If the global performance index and the interval performance index corresponding to each sub-interval both meet the preset standard, then the CHF prediction model is confirmed to have passed the verification. If the global performance index and the interval performance index corresponding to each sub-interval do not meet the preset standard, then at least one performance weak area is located from the multiple sub-intervals based on the weak area diagnosis rule; Analyze the common operating condition characteristics and error distribution characteristics of the at least one performance weakness area; Obtain target samples that match the common operating condition characteristics from the second CHF test sample set, and construct a CHF correction model based on the target samples; The CHF correction model and the CHF prediction model are weighted and combined by weighting coefficients to obtain the first target CHF prediction model. The numerical range of the weighting coefficients is determined according to the error distribution characteristics. The final value of the weighting coefficients is determined by iterative optimization of the first target CHF prediction model in combination with the third CHF test sample set, with the goal of minimizing the composite loss function.

2. The determination method as described in claim 1, characterized in that, The composite penalty term is a weighted combination of the first norm penalty term and the second norm penalty term, adjusted by the intensity parameter λ and the mixing ratio parameter ρ. The optimization objective is to minimize the composite loss function. The initial CHF prediction model is optimized by updating the values ​​of each coefficient in the coefficient set using the first CHF experimental sample set, resulting in the CHF prediction model, which includes: Using a search algorithm and the first CHF test sample set, with minimizing the composite loss function as the optimization objective and a predefined search space, the current optimal intensity parameter λ and the current optimal mixing ratio parameter ρ are determined. The predefined search space represents the numerical search range for the current optimal λ and ρ. Based on the current optimal intensity parameter λ, the current optimal mixing ratio parameter ρ, and the first CHF test sample set, with minimizing the composite loss function as the optimization objective, the values ​​of each coefficient in the current coefficient set are further optimized to obtain the CHF prediction model.

3. The determination method as described in claim 2, characterized in that, The composite penalty term is given by the following formula: in, For the first norm penalty term, The second norm penalty term is defined as λ > 0, 0 ≤ ρ ≤ 1.

4. The determination method as described in claim 1, characterized in that, The formula corresponding to the first target CHF prediction model is as follows: in, For the first target CHF prediction model, ω(X) is the weight coefficient. For the CHF prediction model, This is the CHF correction model.

5. The determination method as described in claim 1, characterized in that, Each element in the feature matrix is ​​associated with a core operating condition parameter that affects CHF prediction. The process of determining the core operating condition parameter includes: Obtain the CHF test dataset of the CHF test specimen. Each data sample in the CHF test dataset includes multiple corresponding operating condition parameters and measured CHF values. The maximum mutual information coefficient method is used to calculate the MIC value between each operating condition parameter and CHF, and a set of candidate operating condition parameters associated with CHF is selected from the plurality of operating condition parameters based on the MIC value; The importance score of each candidate operating condition parameter in the candidate operating condition parameter set for predicting CHF is determined by the random forest feature importance method, and the core operating condition parameter is selected from the candidate operating condition parameter set based on the importance score.

6. The determination method as described in claim 1, characterized in that, The elements in the feature matrix are associated with core operating condition parameters that affect CHF prediction, including pressure. Mass flow rate and local vapor content The feature matrix is .

7. A computer-readable medium storing computer program code that, when executed by a processor, implements the determination method as described in any one of claims 1-6.

8. An electronic device, characterized in that, include: At least one processor; as well as At least one memory storing instructions that, when executed individually or jointly by the at least one processor, cause the electronic device to perform the determination method as described in any one of claims 1-6.

Images