A method for establishing a model for predicting the landing area of ​​foreign objects in the lower chamber

By constructing a backpropagation neural network model and utilizing foreign object parameters and operating condition parameters, the problem of rapid and accurate prediction of the foreign object landing area in the lower chamber of a nuclear reactor was solved. This model is applicable to foreign object risk assessment under different reactor types and operating conditions.

CN123115789APending Publication Date: 2026-11-17CHONGQING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611246379.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-17
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies are insufficient to quickly and accurately determine the final landing area of ​​foreign objects in the lower chamber of a nuclear reactor, especially when faced with various types of foreign objects and complex operating conditions, and cannot meet the needs of rapid assessment of foreign object risks in engineering applications.

Method used

A backpropagation neural network model is used to establish a mapping relationship between foreign object parameters and the final landing area by constructing a training sample set and iterative training. Predictions are made using parameters such as foreign object density, mass, shape factor, and lower chamber inlet flow rate.

Benefits of technology

It enables rapid and accurate prediction of the foreign object landing area in the lower chamber of a nuclear reactor, reduces computational costs, improves analysis efficiency, and is applicable to foreign object risk assessment under different reactor types and operating conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115789A_ABST
    Figure CN123115789A_ABST
Patent Text Reader

Abstract

This invention discloses a method for establishing a foreign object landing area prediction model for the lower chamber of a nuclear reactor. First, based on the structural characteristics of the lower chamber and the coolant flow characteristics, control parameters affecting the movement behavior of foreign objects are determined. The landing area of ​​foreign objects in the lower chamber of the nuclear reactor, along with the corresponding control parameters, is obtained, thereby constructing a training sample set. Second, an initial backpropagation neural network model is constructed. The training sample set is divided into a training set, a validation set, and a test set. The initial backpropagation neural network model is iteratively trained using the training set to obtain a trained backpropagation neural network model. Finally, the trained backpropagation neural network model is validated and tested using the validation set and the test set, respectively. After successful validation, the trained backpropagation neural network model is output as a foreign object landing area prediction model for the lower chamber, thus realizing the prediction of the foreign object landing area in the lower chamber of the nuclear reactor.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of nuclear reactor thermal-hydraulic engineering, specifically a method for establishing a model to predict the landing area of ​​foreign objects in the lower chamber. Background Technology

[0002] During long-term operation, nuclear reactors may accumulate foreign objects such as bolts and metal fragments due to factors such as equipment manufacturing, installation, maintenance, structural component aging, and overhaul operations. These foreign objects can enter the reactor coolant system and may flow into the pressure vessel, eventually settling in the lower chamber of the pressure vessel, the core inlet area, or other critical locations. If these foreign objects enter the core area or remain near critical structures, they can cause problems such as fuel assembly wear, flow channel blockage, and damage to internal components, potentially impacting the safe operation of the reactor.

[0003] Currently, research on foreign object (FOO) issues inside nuclear reactors mainly focuses on FEO detection, FEO movement analysis, and risk assessment. Existing studies have shown that the final location of a FEO is influenced by multiple factors, including the FEO's own physical properties and reactor operating parameters. Specifically, parameters such as FEO density, mass, and geometry affect the stress state of the FEO in the coolant environment, while operating parameters such as the lower chamber inlet flow rate alter the flow environment of the FEO, thus affecting its final landing area.

[0004] In existing technologies, determining the final location of foreign objects (FOOs) typically relies on numerical simulation methods. By establishing a reactor structural model and performing fluid calculations, the final location of FEOs under different FEO parameters and operating conditions can be obtained. However, such methods usually require recalculation for each combination of FEO parameters and operating conditions, which is complex and time-consuming. When dealing with multiple types of FEOs or numerous operating conditions, it is difficult to make a rapid determination in a timely manner, failing to meet the needs of rapid FEO risk assessment in engineering applications. On the other hand, because the FEO landing area is affected by multiple factors, the relationship between FEO parameters and the final landing location usually exhibits a complex nonlinear relationship, making it difficult for traditional empirical methods to establish an accurate correspondence.

[0005] Neural network-based prediction methods can learn the mapping relationship between input parameters and output results through a large amount of sample data, achieving rapid prediction of target results without repeating complex calculations. Therefore, applying artificial intelligence prediction methods to determine the landing area of ​​foreign objects in nuclear reactors is expected to improve prediction efficiency and reduce computational costs. However, current technology lacks a rapid prediction method for the landing area of ​​foreign objects in the lower chamber of nuclear reactors, and cannot directly determine the possible final location of foreign objects based on their own characteristic parameters and operating conditions. Summary of the Invention

[0006] The purpose of this invention is to provide a method for establishing a prediction model of the landing point area of ​​foreign objects in the lower cavity, comprising the following steps:

[0007] Step 1) Based on the structural characteristics of the lower chamber of the nuclear reactor and the flow characteristics of the coolant, determine the control parameters that affect the movement behavior of foreign objects;

[0008] Obtain the foreign object landing area in the lower chamber of the nuclear reactor and the corresponding control parameters, thereby constructing a training sample set containing multiple sets of control parameter-foreign object landing area sample pairs in the lower chamber of the nuclear reactor.

[0009] Step 2) Construct the initial backpropagation neural network model;

[0010] Step 3) Divide the training sample set into a training set, a validation set, and a test set;

[0011] The initial backpropagation neural network model is iteratively trained using the training set to obtain the trained backpropagation neural network model.

[0012] Step 4) Use the validation set and test set to verify and test the performance of the trained backpropagation neural network model. After the verification is passed, output the trained backpropagation neural network model as the prediction model for the landing area of ​​foreign objects in the lower chamber.

[0013] A foreign object impact zone prediction model for the lower chamber is used to predict the impact zone of foreign objects in the lower chamber of a nuclear reactor.

[0014] Furthermore, in step 1), the control parameters include foreign matter density, foreign matter mass, shape factor, and lower chamber inlet flow rate.

[0015] The shape factor is shown below:

[0016] (1)

[0017] In the formula, To have the same surface area as spherical particles with the same volume, This represents the actual surface area of ​​the solid particles.

[0018] Furthermore, the foreign object landing area in the lower chamber of the nuclear reactor includes the foreign object being retained inside the lower chamber and the foreign object flowing out of the lower chamber with the fluid.

[0019] Furthermore, in step 2), the initial backpropagation neural network model includes an input layer, a hidden layer, and an output layer;

[0020] The input layer is used to receive control parameters that affect the movement behavior of foreign objects;

[0021] The output layer combines the actual movement of foreign objects in the lower chamber to predict the landing area of ​​foreign objects in the lower chamber of the nuclear reactor.

[0022] The range of the number of hidden layer nodes is determined by the number of input layer nodes and the number of output layer neurons in the neural network.

[0023] Furthermore, the range of the number of hidden layer nodes is as follows:

[0024] (2)

[0025] In the formula, m is the number of nodes in the input layer of the neural network. This represents the number of neurons in the hidden layer, where a is a constant and n is the number of neurons in the output layer.

[0026] Furthermore, the output layer transfer function is a sigmoid function logsig;

[0027] The hidden layer transfer function is set to the sigmoid function tansig.

[0028] Furthermore, in step 3), the iterative training steps for the initial backpropagation neural network model are as follows:

[0029] Step 3.1) Initialize the backpropagation neural network, training times ;

[0030] Step 3.2) Let Input the training set and calculate the output value layer by layer;

[0031] Step 3.3) Calculate the output error of the backpropagation neural network;

[0032] Step 3.4) Determine whether the output error meets the preset threshold requirement;

[0033] If the output error meets the preset threshold requirement, the training ends and the trained backpropagation neural network model is obtained.

[0034] If the output error does not meet the preset threshold requirement, update the network weights and return to step 3.2.

[0035] Furthermore, in step 3.2), the steps for calculating the output value layer by layer are as follows:

[0036] Enter the command in the command line window ,in, For foreign object density, For foreign object mass, For shape factor, The inlet flow rate of the lower chamber;

[0037] The net input of the hidden layer is shown below:

[0038] (3)

[0039] In the formula, As weight; The bias is denoted by j; j is the number of hidden layers.

[0040] The hidden layer output is shown below:

[0041] (4)

[0042] The net input of the output layer is shown below:

[0043] (5)

[0044] In the formula, As weight; For bias; k is the number of output layers;

[0045] The final output value is shown below:

[0046] (6)

[0047] In the formula, This is the net input for the output layer.

[0048] Furthermore, in step 3.3), when calculating the output error of the backpropagation neural network, for a single sample, the target output vector... The loss function is as follows:

[0049] (7)

[0050] The overall error of processing N sample batches is shown below:

[0051] (8)

[0052] In the formula, The error of the nth sample;

[0053] The output layer error sensitivity for the backpropagation error term is as follows:

[0054] (9)

[0055] The hidden layer error sensitivity is shown below:

[0056] (10)

[0057] The output layer weights and biases are updated as follows:

[0058] (11)

[0059] (12)

[0060] The hidden layer weights and biases are updated as follows:

[0061] (13)

[0062] (14)

[0063] In the formula, This is the learning rate.

[0064] Furthermore, in step 3.4), the preset threshold requirement refers to the output error decreasing to the convergence accuracy target or the number of training rounds reaching the upper limit.

[0065] This invention utilizes existing numerical simulation results to construct training samples and establishes a mapping relationship between foreign object parameters and the final landing area. After the model training is completed, only a few key parameters such as foreign object density, mass, shape factor, and inlet flow rate need to be input to quickly obtain the foreign object landing point prediction results, avoiding repeated high-cost numerical calculations and improving analysis efficiency.

[0066] This invention does not rely on empirical judgment using a single parameter, but comprehensively considers the physical characteristics of the foreign object, such as its density, mass, and shape factor, as well as flow conditions such as the flow rate at the lower chamber inlet. By extracting the complex nonlinear relationships between multiple parameters through a BP neural network, it can more accurately describe the motion patterns of different types of foreign objects in complex flow environments and improve the reliability of landing area prediction.

[0067] The predictive model established in this invention has good engineering applicability. Employing a data-driven predictive framework, the model input parameters can be adjusted according to different reactor structure characteristics and operating conditions, and can be further extended to predict the landing points of foreign objects of different sizes, materials, and shapes. In practical applications, staff do not need to rebuild complex calculation models or perform lengthy simulations; they only need to input the basic parameters of the discovered foreign object and the current operating conditions to quickly determine whether the foreign object may remain inside the reactor cavity or enter other areas with the coolant. This provides rapid technical support for foreign object risk assessment, maintenance decisions, and retrieval plan formulation in nuclear power plants. Attached Figure Description

[0068] Figure 1 This is a flowchart of the algorithm of the present invention;

[0069] Figure 2 This is a flowchart of the BP neural network parameter adjustment process of the present invention;

[0070] Figure 3 The number of hidden layer neurons in this invention is The training process of the network; where (a) is (b) is (c) is (d) is ;

[0071] Figure 4 The results are the grid training results of the present invention; where (a) is the change of training mean square error with the number of training iterations, and (b) is the error histogram;

[0072] Figure 5 This is the confusion matrix result of the training, verification, and testing of this invention;

[0073] Figure 6 This is a schematic diagram of the model prediction results of the present invention. Detailed Implementation

[0074] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0075] Example 1:

[0076] See Figures 1 to 6 A method for establishing a prediction model of the landing point area of ​​a foreign object in the lower cavity, comprising the following steps:

[0077] Step 1) Based on the structural characteristics of the lower chamber of the nuclear reactor and the flow characteristics of the coolant, determine the control parameters that affect the movement behavior of foreign objects;

[0078] Obtain the foreign object landing area in the lower chamber of the nuclear reactor and the corresponding control parameters, thereby constructing a training sample set containing multiple sets of control parameter-foreign object landing area sample pairs in the lower chamber of the nuclear reactor.

[0079] Step 2) Construct the initial backpropagation neural network model;

[0080] Step 3) Divide the training sample set into a training set, a validation set, and a test set;

[0081] The initial backpropagation neural network model is iteratively trained using the training set to obtain the trained backpropagation neural network model.

[0082] Step 4) Use the validation set and test set to verify and test the performance of the trained backpropagation neural network model. After the verification is passed, output the trained backpropagation neural network model as the prediction model for the landing area of ​​foreign objects in the lower chamber.

[0083] A foreign object impact zone prediction model for the lower chamber is used to predict the impact zone of foreign objects in the lower chamber of a nuclear reactor.

[0084] Example 2:

[0085] The main structure of this embodiment is the same as that of embodiment 1. Further, in step 1), the control parameters include foreign matter density, foreign matter mass, shape factor and lower chamber inlet flow rate.

[0086] When using this model for actual prediction, the foreign object density, foreign object mass, and shape factor are all obtained through actual measurement and require no additional processing.

[0087] The shape factor is shown below:

[0088] (1)

[0089] In the formula, To have the same surface area as spherical particles with the same volume, This represents the actual surface area of ​​the solid particles.

[0090] Example 3:

[0091] The main structure of this embodiment is the same as any one of embodiments 1-2. Furthermore, the foreign object landing area in the lower chamber of the nuclear reactor includes the foreign object being retained inside the lower chamber and the foreign object flowing out of the lower chamber with the fluid.

[0092] Example 4:

[0093] The main structure of this embodiment is the same as any one of embodiments 1-3. Further, in step 2), the initial backpropagation neural network model includes an input layer, a hidden layer, and an output layer.

[0094] The input layer is used to receive control parameters that affect the movement behavior of foreign objects;

[0095] The output layer combines the actual movement of foreign objects in the lower chamber to predict the landing area of ​​foreign objects in the lower chamber of the nuclear reactor.

[0096] The range of the number of hidden layer nodes is determined by the number of input layer nodes and the number of output layer neurons in the neural network.

[0097] Example 5:

[0098] The main structure of this embodiment is the same as any one of embodiments 1-4. Furthermore, the range of the number of hidden layer nodes is as follows:

[0099] (2)

[0100] In the formula, m is the number of nodes in the input layer of the neural network. This represents the number of neurons in the hidden layer, where a is a constant between 0 and 10, and n is the number of neurons in the output layer.

[0101] Example 6:

[0102] The main structure of this embodiment is the same as any one of embodiments 1-5. Furthermore, the output layer transfer function is an S-shaped function logsig.

[0103] The hidden layer transfer function is set to the sigmoid function tansig.

[0104] Example 7:

[0105] The main structure of this embodiment is the same as any one of embodiments 1-6. Further, in step 3), the iterative training steps for the initial backpropagation neural network model are as follows:

[0106] Step 3.1) Initialize the backpropagation neural network, training times ;

[0107] Step 3.2) Let Input the training set and calculate the output value layer by layer;

[0108] Step 3.3) Calculate the output error of the backpropagation neural network;

[0109] Step 3.4) Determine whether the output error meets the preset threshold requirement;

[0110] If the output error meets the preset threshold requirement, the training ends and the trained backpropagation neural network model is obtained.

[0111] If the output error does not meet the preset threshold requirement, update the network weights and return to step 3.2.

[0112] Example 8:

[0113] The main structure of this embodiment is the same as any one of embodiments 1-7. Further, in step 3.2), the steps for calculating the output value layer by layer are as follows:

[0114] Enter the command in the command line window ,in, Foreign matter density (kg / m³) Foreign matter mass (g) For shape factor, The inlet flow rate of the lower chamber is (kg / s).

[0115] The net input of the hidden layer is shown below:

[0116] (3)

[0117] In the formula, As weight; The bias is denoted by j, where j is the number of hidden layers. ; ;

[0118] The hidden layer output is shown below:

[0119] (4)

[0120] The net input of the output layer is shown below:

[0121] (5)

[0122] In the formula, As weight; For bias; k is the number of output layers. ;

[0123] The final output value is shown below:

[0124] (6)

[0125] In the formula, This is the net input for the output layer.

[0126] Example 9:

[0127] The main structure of this embodiment is the same as any one of embodiments 1-8. Further, in step 3.3), when calculating the output error of the backpropagation neural network, for a single sample, the target output vector... The loss function is as follows:

[0128] (7)

[0129] The overall error of processing N sample batches is shown below:

[0130] (8)

[0131] In the formula, The error of the nth sample;

[0132] The output layer error sensitivity for the backpropagation error term is as follows:

[0133] (9)

[0134] The hidden layer error sensitivity is shown below:

[0135] (10)

[0136] The hyperparameter update uses the gradient descent method. In each iteration, the gradient is calculated for all training samples and averaged. Then, all weights and biases are updated.

[0137] The output layer weights and biases are updated as follows:

[0138] (11)

[0139] (12)

[0140] The hidden layer weights and biases are updated as follows:

[0141] (13)

[0142] (14)

[0143] In the formula, This is the learning rate. .

[0144] Example 10:

[0145] The main structure of this embodiment is the same as any one of embodiments 1-9. Further, in step 3.4), the preset threshold requirement means that the output error decreases to the convergence accuracy target or the number of training rounds reaches the upper limit.

[0146] Example 11:

[0147] This invention proposes a method for predicting the landing point of foreign objects in the lower chamber of a nuclear reactor based on a backpropagation (BP) neural network. The complete implementation path of the BP neural network is systematically divided into three closely linked core stages: model architecture construction, network parameter learning and training, and model prediction result output. Each stage proceeds sequentially and works in concert, fully covering the entire computational logic from model initialization to final prediction result output. The specific BP neural network prediction process used in this study is as follows: Figure 1 As shown, this method establishes a nonlinear mapping relationship between foreign object characteristic parameters, operating condition parameters, and the final landing area of ​​the foreign object, enabling rapid prediction of the movement outcome of foreign objects entering the reactor lower chamber.

[0148] To ensure the residual between the model output and the target value meets a preset threshold, the neural network needs to undergo multiple iterative optimizations. Each training round aims to drive the network weights towards the desired output direction. When the global error decreases to a predetermined accuracy range, the network is considered to have converged, and the training process terminates; if the error metric still does not meet the target, sample data needs to be reloaded, and a new round of parameter update loops begins. The flowchart for BP algorithm parameter adjustment is as follows: Figure 2 As shown.

[0149] This method first conducts numerical simulation analysis of the foreign object movement process in the lower chamber of a nuclear reactor, obtaining the final movement results of the foreign object under different foreign object parameters and operating conditions, and using these as training samples for a neural network model. Based on the sample data, key parameters affecting the foreign object movement results are extracted, including foreign object density, foreign object mass, shape factor, and lower chamber inlet flow rate. These parameters are used as input features of the prediction model, and the final landing area of ​​the foreign object is used as the model output, establishing a mapping relationship between foreign object parameters and landing location.

[0150] The specific implementation process is as follows: First, based on the structural characteristics of the lower chamber of the nuclear reactor and the flow characteristics of the coolant, the main control parameters affecting the movement behavior of foreign objects are determined. Considering the combined influence of the physical properties of the foreign object itself and the external flow environment on its movement state, foreign object density, foreign object mass, shape factor, and lower chamber inlet flow rate are selected as model input parameters. In terms of output layer design, combined with the actual movement results of the foreign object in the lower chamber, the prediction results are divided into two typical states: foreign object remaining inside the lower chamber and foreign object flowing out of the lower chamber with the fluid. For this binary classification prediction task, the model output layer is set to have two neurons, corresponding to the two landing point results mentioned above, thereby completing the mapping from multiple physical parameters to the landing point region.

[0151] This paper uses an empirical formula to initially estimate the candidate range for the number of neurons. During network training, the number of neurons is gradually increased from the minimum value in the range to the maximum value. The training results of networks with different numbers of neurons are compared to select an appropriate number. The following formula is typically used to determine the range of hidden layer nodes.

[0152]

[0153]

[0154] In the two calculation formulas above, m represents the number of nodes in the input layer of the neural network, l represents the number of neurons in the hidden layer, a is a constant between 0 and 10, and n represents the number of neurons in the output layer. Based on the prediction model parameters established in this paper, m=4 and n=2. According to the above formulas, the reasonable range for the number of neurons in the hidden layer of this network can be preliminarily determined to be between 2 and 12. To select the number of hidden layer neurons with the best learning performance and fitting effect from this range, the number of hidden layer neurons is set to 3, 6, 9, and 12, uniformly covering the entire value range. The training accuracy, i.e., the target error, is set to 10. -6 To ensure that different networks are compared under the same convergence criterion, these training processes are set up as follows: Figure 3 As shown.

[0155] To compare the performance of the different networks, the evaluation was based on two aspects: number of iterations and training accuracy. The network with 9 nodes achieved optimal training at 125 iterations, with a significantly faster convergence speed than the other three networks with 3, 6, and 12 nodes. The network with 9 nodes had the smallest error of 1.4679 × 10⁻⁶. -7 The network with 9 hidden layer nodes achieves the best prediction accuracy. Therefore, the network with 9 hidden layer nodes performs best. In summary, in the prediction model, the BP neural network is determined to be a three-layer network with 4 inputs, 2 outputs, and 9 hidden layer nodes.

[0156] To achieve accurate determination of the foreign object's landing location, this paper constructs an algorithm simulation system based on the Matlab platform. Regarding hyperparameter selection, the upper limit of the number of training epochs, the learning rate, and the convergence accuracy target are defined as 50, 0.1, and 0.01, respectively. Considering the balance between computational efficiency and convergence performance, the model applies a batch gradient descent criterion during training, rather than a per-sample update scheme. The landing point model in this study outputs the landing point position from several input feature vectors. The hidden layer transfer function is set to a sigmoid function (tansig). Since the network output is a prediction of the landing point, a sigmoid function (logsig) is used as the output layer transfer function, limiting the output to between 0 and 1.

[0157] The feature vector extracted in this study consists of four parameters: foreign matter density, mass, shape factor, and lower chamber flow rate. The output landing point location is categorized into two types: inside and outside the lower chamber, defined as 1 and 2 respectively. Therefore, the data samples for this model are shown in the table below. For example, the data in the first row of Table 1 represents a density of 7850 kg / m³. 3 A foreign object with a mass of 64.9g and a shape factor of 0.6 appeared at the inlet of the lower chamber with a flow rate of 12232 kg / s. The final landing point of the foreign object will be inside the lower chamber.

[0158]

[0159] For all data samples, this model will use 70% of the data as the training set, 15% as the validation set, and 15% as the test set.

[0160] Figure 4 (a) shows the change of training mean square error with the number of training iterations. As can be seen from the figure, the mean square error of training, validation and testing decreases with the increase of the number of training iterations, which indicates that the model established by the BP neural network in this paper has good performance. Figure 4 (b) also shows that the model has a small error during training and performs well.

[0161] Figure 5 This is the confusion matrix result for training, validation, and testing. (From...) Figure 5As can be seen, the accuracy of this model is 100% in training, validation and testing, indicating that the landing point prediction model established by the BP neural network in this paper has good performance.

[0162] The foreign object landing point prediction model established in this paper requires the use of Matlab. In the command line window, input the foreign object parameter matrix A to be predicted, for example, in a lower chamber with a mass flow rate of 15290 kg / m³. 3 A nut with a mass of 5 g was found. Enter the following in the command line window: Then type in the command line window At this point, the prediction result y will appear in the command line window, such as... Figure 6 The two data points shown in the output represent two landing locations. The larger value indicates the region where the foreign object landed. The output in the figure indicates that the foreign object landed in the lower chamber. In this example, this means that the mass flow rate in the lower chamber is 15290 kg / m³. 3 A nut with a mass of 5 g was found, and the foreign object ultimately landed in the lower chamber. This result is consistent with the simulation calculation results mentioned above, proving the reliability of the prediction model.

Claims

1. A method for establishing a predictive model of the landing point region of a foreign object in the lower chamber, characterized in that, Includes the following steps: Step 1) Based on the structural characteristics of the lower chamber of the nuclear reactor and the flow characteristics of the coolant, determine the control parameters that affect the movement behavior of foreign objects; Obtain the foreign object landing area in the lower chamber of the nuclear reactor and the corresponding control parameters, thereby constructing a training sample set containing multiple sets of control parameter-foreign object landing area sample pairs in the lower chamber of the nuclear reactor. Step 2) Construct the initial backpropagation neural network model; Step 3) Divide the training sample set into a training set, a validation set, and a test set; The initial backpropagation neural network model is iteratively trained using the training set to obtain the trained backpropagation neural network model. Step 4) Use the validation set and test set to verify and test the performance of the trained backpropagation neural network model. After the verification is passed, output the trained backpropagation neural network model as the prediction model for the landing area of ​​foreign objects in the lower chamber. A foreign object impact zone prediction model for the lower chamber is used to predict the impact zone of foreign objects in the lower chamber of a nuclear reactor.

2. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 1, characterized in that, In step 1), the control parameters include foreign matter density, foreign matter mass, shape factor, and lower chamber inlet flow rate; The shape factor is shown below: (1) In the formula, To have the same surface area as spherical particles with the same volume, This represents the actual surface area of ​​the solid particles.

3. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 2, characterized in that, The foreign object landing area in the lower chamber of a nuclear reactor includes the foreign object remaining inside the lower chamber and the foreign object flowing out of the lower chamber with the fluid.

4. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 3, characterized in that, In step 2), the initial backpropagation neural network model includes an input layer, hidden layers, and an output layer; The input layer is used to receive control parameters that affect the movement behavior of foreign objects; The output layer combines the actual movement of foreign objects in the lower chamber to predict the landing area of ​​foreign objects in the lower chamber of the nuclear reactor. The range of the number of hidden layer nodes is determined by the number of input layer nodes and the number of output layer neurons in the neural network.

5. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 3, characterized in that, The range of hidden layer nodes is shown below: (2) In the formula, m is the number of nodes in the input layer of the neural network. This represents the number of neurons in the hidden layer, where a is a constant and n is the number of neurons in the output layer.

6. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 3, characterized in that, The output layer transfer function is a sigmoid function logsig; The hidden layer transfer function is set to the sigmoid function tansig.

7. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 1, characterized in that, In step 3), the iterative training steps for the initial backpropagation neural network model are as follows: Step 3.1) Initialize the backpropagation neural network, training times ; Step 3.2) Let Input the training set and calculate the output value layer by layer; Step 3.3) Calculate the output error of the backpropagation neural network; Step 3.4) Determine whether the output error meets the preset threshold requirement; If the output error meets the preset threshold requirement, the training ends and the trained backpropagation neural network model is obtained. If the output error does not meet the preset threshold requirement, update the network weights and return to step 3.

2.

8. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 7, characterized in that, In step 3.2), the steps for calculating the output value layer by layer are as follows: Enter the command in the command line window ,in, For foreign object density, For foreign object mass, For shape factor, The inlet flow rate of the lower chamber; The net input of the hidden layer is shown below: (3) In the formula, As weight; The bias is denoted by j; j is the number of hidden layers. The hidden layer output is shown below: (4) The net input of the output layer is shown below: (5) In the formula, As weight; For bias; k is the number of output layers; The final output value is shown below: (6) In the formula, This is the net input for the output layer.

9. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 8, characterized in that, In step 3.3), when calculating the output error of the backpropagation neural network, for a single sample, the target output vector... The loss function is as follows: (7) The overall error of processing N sample batches is shown below: (8) In the formula, The error is for the nth sample. The output layer error sensitivity for the backpropagation error term is as follows: (9) The hidden layer error sensitivity is shown below: (10) The output layer weights and biases are updated as follows: (11) (12) The hidden layer weights and biases are updated as follows: (13) (14) In the formula, This is the learning rate.

10. The method for establishing a prediction model of the landing point area of ​​a foreign object in the lower chamber according to claim 7, characterized in that, In step 3.4), the preset threshold requirement refers to the output error decreasing to the convergence accuracy target or the number of training rounds reaching the upper limit.