Rapid prediction method for three-dimensional transient temperature field of passive residual heat removal heat exchanger

Through random sampling and four-dimensional tensor convolution technology of artificial neural network, the problem of three-dimensional transient temperature field prediction of non-energy waste heat discharge heat exchanger is solved, and the rapid and accurate temperature field prediction is achieved, which meets the development of digital twin models and structural optimization. The design requirement improves the safety and stability of nuclear power plants.

CN119004953APending Publication Date: 2024-11-22XI AN JIAOTONG UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410990210.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The prior art is difficult to achieve fast and high-precision prediction of the three-dimensional transient temperature field of non-active waste heat discharge heat exchanger, and cannot meet the needs of digital twin model development and structural design optimization.

Method used

Random sampling is used to obtain simulation conditions, combine OpenFOAM and pyTorch to build an artificial neural network for four-dimensional tensor convolution, and achieve rapid prediction of three-dimensional temperature field.

Benefits of technology

It greatly reduces the demand for computing resources, realizes fast and accurate prediction of three-dimensional transient temperature fields, supports the development of digital twin models and structural optimization design, and improves the waste heat discharge capacity and safety margin of the dedicated safety system of nuclear power plants.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004953A_ABST
    Figure CN119004953A_ABST
Patent Text Reader

Abstract

The invention discloses a rapid prediction method for a three-dimensional transient temperature field of a passive residual heat removal heat exchanger, and the method comprises the following steps: 1, carrying out the random sampling in a range of the operation condition of the passive residual heat removal heat exchanger, and obtaining a plurality of groups of simulation conditions; 2, carrying out the three-dimensional transient simulation of the passive residual heat removal heat exchanger based on the sampling working condition; 3, performing post-processing to obtain a four-dimensional tensor containing the three-dimensional temperature field data of the passive residual heat removal heat exchanger at a certain moment; 4, building an artificial neural network; 5, taking each group of simulation working condition parameters as input, taking a four-dimensional tensor obtained by post-processing as target output, and training to obtain an artificial neural network at the moment; 6, repeating the steps 3-5 to obtain an artificial neural network at each moment; 7, inputting the new working condition parameters into the artificial neural network at each moment, and quickly predicting a three-dimensional transient temperature field of the passive residual heat removal heat exchanger; according to the method, the three-dimensional transient temperature field of the passive residual heat removal heat exchanger under different working conditions can be quickly predicted, simulation resources are greatly saved, and digital twin model development and optimization design research of the passive residual heat removal heat exchanger are supported.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of virtual real-time simulation of nuclear power system equipment, and specifically relates to a method for quickly predicting a three-dimensional transient temperature field of a passive waste heat removal heat exchanger. Background Art

[0002] Passive safety is one of the important concepts introduced in the third-generation advanced pressurized water reactor nuclear power plant. The passive residual heat removal heat exchanger is a key equipment in the passive dedicated safety facilities of the third-generation advanced pressurized water reactor nuclear power plant. Under the transient or accident shutdown conditions of the reactor, the core coolant enters the passive residual heat removal heat exchanger through the pipe on the primary hot leg by natural drive, and removes the core residual heat to the external heat sink without external force, ensuring the safety and integrity of the reactor core. The passive residual heat removal heat exchanger has a complex structure and many heat transfer tubes. The three-dimensional numerical simulation of it using numerical simulation methods consumes a lot of computing resources. Therefore, it is urgent to develop a three-dimensional transient temperature field prediction method for passive residual heat removal heat exchanger based on machine learning methods to save computing resources, improve the efficiency of the method, and support the three-dimensional transient thermal stratification and natural circulation characteristics analysis, digital twin model development, and structural optimization design research of passive residual heat removal heat exchanger.

[0003] At present, the main rapid prediction and analysis methods for passive waste heat removal heat exchangers of nuclear power systems include lumped parameter method, traditional neural network method, model reduction method, etc. Among them, the lumped parameter method uses one-dimensional simplification on both sides of the entire passive waste heat removal heat exchanger, and performs one-dimensional calculations on the heat exchanger based on the mass, energy conservation relationship and basic heat transfer laws. The advantage is fast calculation speed, but it cannot obtain the three-dimensional temperature field inside the heat exchanger and its transient changes, and is mostly used for thermal calculations in the design stage. For traditional neural network methods: BP neural network only contains fully connected layers, so it is mostly used to predict a limited number of key operating parameters of heat exchangers. For temperature field data prediction, a large number of hidden layers and neurons need to be introduced, and the training cost is high and difficult; due to the convolution operation characteristics, convolutional neural network is suitable for image prediction problems such as simulation physical field prediction, but the input of convolutional neural network must also be converted into an image first, and it is difficult to achieve transient prediction; recurrent neural network has a good prediction effect on transient serialized data, but as time steps advance, the continuously accumulated training error restricts the prediction accuracy; the model reduction method first uses the intrinsic orthogonal decomposition algorithm to perform modal decomposition on the three-dimensional temperature field of the heat exchanger, and minimizes the complexity of the three-dimensional temperature field through modal analysis and order reduction processing. Combined with neural network, it can effectively predict the steady-state temperature field of the two-dimensional section, but it is difficult to predict the three-dimensional and transient temperature field.

[0004] In summary, the existing rapid prediction and analysis methods for passive waste heat removal heat exchangers are difficult to achieve high-precision rapid prediction of three-dimensional and transient temperature fields, and cannot meet the real-time virtual simulation requirements for the development of digital twin models of important industrial equipment and the structural design optimization of heat exchangers. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a method for rapid prediction of the three-dimensional transient temperature field of a passive waste heat removal heat exchanger. This method can greatly reduce the repeated simulation cost of the passive waste heat removal heat exchanger in the three-dimensional transient thermal-hydraulic characteristics analysis and research, and quickly and accurately predict the three-dimensional transient temperature field of the heat exchanger under any parameters within the operating range of the heat exchanger, thereby supporting the development of digital twin models and structural optimization design research of passive waste heat removal heat exchangers.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for rapid prediction of three-dimensional transient temperature field of a passive waste heat removal heat exchanger, characterized in that it comprises the following steps:

[0008] Step 1: Randomly sample within the operating range of the passive waste heat removal heat exchanger to obtain several groups of simulation conditions. The specific steps are as follows:

[0009] Step 1-1: Determine the operating parameters required for the simulation of the passive waste heat removal heat exchanger: primary side inlet flow a, primary side inlet temperature b, secondary side initial temperature c; determine the value range of each parameter under the normal operating conditions of the passive waste heat removal heat exchanger, that is:

[0010] a∈A=[a min , a max ] (1)

[0011] b∈B=[b min , b max ] (2)

[0012] c∈C=[c min , c max ] (3)

[0013] Wherein, A, B, and C represent the value range of the primary side inlet flow rate a (unit kg / s), the value range of the primary side inlet temperature b (unit K), and the value range of the secondary side initial temperature c (unit K), respectively; the subscript min represents the minimum value, and max represents the maximum value;

[0014] Step 1-2: Use the random function in Excel software to randomly sample each key operating parameter within the range of each parameter value to obtain N groups of simulation conditions, namely:

[0015] a i =a min +RAND()·(a max -a min ), i=1,2,3...N (4)

[0016] b i =b min +RAND()·(b max -b min ), i=1,2,3...N (5)

[0017] c i =c min +RAND()·(c max -c min ), i=1,2,3...N (6)

[0018] Among them, a i , b i , c i They represent the primary inlet flow, primary inlet temperature, and secondary initial temperature of the i-th group of simulation conditions respectively; N represents the total number of simulation conditions; RAND() is a random function that takes random values ​​between 0 and 1 and is dimensionless;

[0019] Step 2: Based on the N groups of simulation conditions sampled in step 1, set the simulation boundary conditions of the primary and secondary sides of the passive waste heat removal heat exchanger in the open source numerical simulation software OpenFOAM, and carry out three-dimensional transient numerical simulation calculations of the passive waste heat removal heat exchanger under N groups of simulation conditions. Each group of simulation conditions can obtain T calculation result files at different times, totaling N×T calculation result files;

[0020] Step 3: Post-process the N calculation result files at a certain time t (t = 1, 2, 3 ... T) to obtain a four-dimensional tensor containing the three-dimensional temperature field data of the passive residual heat removal heat exchanger at a certain time. The specific steps are as follows:

[0021] Step 3-1: Import the N calculation result files at time t into the post-processing software ParaView, and use the slicing function to establish M two-dimensional planes of the same size inside the passive waste heat removal heat exchanger. The number of grids in the length and width directions of the two-dimensional planes are X and Y respectively;

[0022] Step 3-2: Export the temperature field data of M two-dimensional planes under N working conditions from ParaView in the format of csv files, and obtain a total of N×M csv files;

[0023] Step 3-3: Use the open source machine learning tool pyTorch to read the data of N×M csv files and arrange them into a four-dimensional tensor D0 with a size of N×M×X×Y; the tensor contains the three-dimensional temperature field of the passive waste heat removal heat exchanger at the same time and under different working conditions;

[0024] Step 4: Build an artificial neural network, including input layer, linear layer, convolutional layer and output layer:

[0025] Step 4-1: Establish an input layer, which contains 3 neurons to receive the 3 key operating parameters contained in the i-th group of simulation conditions: a i , b i , c i , arranged as a one-dimensional tensor D1 = [a i , b i , c i ], size is 1×3;

[0026] Step 4-2: Establish a fully connected layer. The linear layer is used to receive the one-dimensional tensor D1, transform it linearly and output it as a one-dimensional tensor D2 with a size of 1×(M×X×Y), and then concatenate it into a three-dimensional tensor D3 with a size of M×X×Y;

[0027] Step 4-3: Establish a convolutional layer, which is used to receive the three-dimensional tensor D3, and output a three-dimensional tensor D4 of size M×X×Y after convolution;

[0028] Step 4-4: Establish an output layer, which is used to receive the three-dimensional tensor D4 and obtain the actual output of the artificial neural network after linear transformation: a three-dimensional tensor D5 of size M×X×Y, which contains the three-dimensional temperature field data of the heat exchanger at the i-th working condition and time t;

[0029] Step 5: Take the key operating parameters of N groups of simulation conditions as the input of the artificial neural network, take the four-dimensional tensor obtained in step 3-3 as the target output, take the root mean square error between the actual output D5 and the target output D0 of the artificial neural network as the convergence criterion, train the artificial neural network built in step 4 until convergence, and obtain the artificial neural network at time t;

[0030] Step 6: Repeat steps 3 to 5 for each moment t=1, 2, 3…T, and T trained artificial neural networks can be obtained, which can be used to predict the three-dimensional temperature field of the passive waste heat removal heat exchanger under different working conditions and at specific times;

[0031] Step 7: Input the three parameters of the passive waste heat removal heat exchanger under the new operating condition: primary side inlet flow, primary side inlet temperature and secondary side initial temperature into T trained and converged artificial neural networks respectively, so as to quickly obtain the three-dimensional temperature field distribution of the passive waste heat removal heat exchanger at T moments under the new operating condition, and realize the rapid prediction of the three-dimensional transient temperature field of the heat exchanger.

[0032] By adopting the rapid prediction method of the present invention, the three-dimensional temperature field of the passive residual heat removal heat exchanger under any operating condition and at any time can be obtained, and the three-dimensional transient thermal stratification and natural circulation characteristics of the passive residual heat removal heat exchanger under different operating conditions can be analyzed accordingly, and digital twin model development and structural optimization design research can be carried out to improve the residual heat removal capacity and safety margin of the dedicated safety system of the third-generation pressurized water reactor nuclear power plant, so as to ensure the safe and stable operation of the entire nuclear power plant.

[0033] Compared with the prior art, the method of the present invention avoids the need for high-performance computing resources when carrying out repetitive three-dimensional refined simulation of non-passive residual heat removal heat exchangers with complex structures and numerous tube bundles, overcomes the deficiency of the traditional one-dimensional lumped parameter method in capturing the three-dimensional thermal-hydraulic characteristics of the heat exchanger, and solves the problem that the model reduction method and the traditional neural network method are difficult to quickly predict the transient and three-dimensional temperature field of the heat exchanger. The present invention combines the advantages of existing methods, uses multi-channel four-dimensional tensor convolution technology to expand the traditional temperature field prediction from two dimensions to three dimensions, and discretizes the difficult time series prediction into artificial neural networks at several moments, which can greatly reduce the computing cost while realizing the rapid prediction of the three-dimensional transient temperature field of the non-passive residual heat removal heat exchanger, and at the same time meets the engineering requirements of high computing speed and three-dimensional transient resolution. The present invention can support the three-dimensional transient thermal-hydraulic characteristics analysis of non-passive residual heat removal heat exchangers and the development of digital twin models, which is crucial to improving the residual heat removal capacity and safety margin of the dedicated safety system of the third-generation pressurized water reactor nuclear power plant and ensuring the safe and stable operation of the entire nuclear power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagram of the artificial neural network built in step 4.

[0035] Figure 2 It is a comparison chart of the three-dimensional numerical simulation results and the temperature field fast prediction results of the method of the present invention.

[0036] Figure 3 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0037] The following combination Figure 3 The flow chart of the method of the present invention shown in the figure takes the rapid prediction of the three-dimensional transient temperature field of the passive residual heat removal heat exchanger of the AP1000 reactor as an example to further describe the present invention in detail.

[0038] The present invention provides a method for quickly predicting the three-dimensional transient temperature field of a passive waste heat removal heat exchanger, comprising the following steps:

[0039] Step 1: Randomly sample within the operating range of the AP1000 passive waste heat removal heat exchanger to obtain several groups of simulation conditions. The specific steps are as follows:

[0040] Step 1-1: Determine the operating parameters required for the simulation of the passive waste heat removal heat exchanger: primary side inlet flow a, primary side inlet temperature b, secondary side initial temperature c; determine the value range of each parameter under the normal operating conditions of the passive waste heat removal heat exchanger, that is:

[0041] a∈A=[40, 70] (1)

[0042] b∈B=[470, 570] (2)

[0043] c∈C=[300, 330] (3)

[0044] Among them, A, B, and C represent the value range of the primary side inlet flow rate a (unit kg / s), the value range of the primary side inlet temperature b (unit K), and the value range of the secondary side initial temperature c (unit K), respectively;

[0045] Step 1-2: Use the random function in Excel software to randomly sample each key operating parameter within the range of each parameter value to obtain 200 sets of simulation conditions, namely:

[0046] a i =40+RAND()·(70-40), i=1,2,3...200 (4)

[0047] b i = +RAND()·(570- i= 3 0 (5)

[0048] c i =300+RAND()·(330-300), i=1,2,3...200 (6)

[0049] Among them, a i , b i , c i They represent the primary side inlet flow, primary side inlet temperature, and secondary side initial temperature of the i-th group of simulation conditions respectively; RAND() is a random function, which takes random values ​​between 0 and 1 and is dimensionless;

[0050] Step 2: Based on the 200 sets of simulation conditions sampled in step 1, set the simulation boundary conditions of the primary and secondary sides of the AP1000 reactor passive residual heat removal heat exchanger in the open source numerical simulation software OpenFOAM, and carry out three-dimensional transient numerical simulation calculations of the heat exchanger under 200 sets of simulation conditions. Each set of simulation conditions can obtain 20 calculation result files at different times, totaling 200×20 calculation result files;

[0051] Step 3: Post-process 200 calculation result files at a certain time t (t = 1, 2, 3 ... 20) to obtain a four-dimensional tensor containing the three-dimensional temperature field data of the passive residual heat removal heat exchanger at a certain time. The specific steps are as follows:

[0052] Step 3-1: Import the 200 calculation result files at time t into the post-processing software ParaView, and use the slicing function to establish three two-dimensional planes of the same size inside the passive waste heat removal heat exchanger. The number of grids in the length and width directions of the two-dimensional planes is 20 and 48 respectively;

[0053] Step 3-2: Export the temperature field data of 3 two-dimensional planes under 200 working conditions from ParaView in the format of csv files, and obtain a total of 200×3 csv files;

[0054] Step 3-3: Use the open source machine learning tool pyTorch to read the data of the 200×3 csv files and arrange them into a four-dimensional tensor D0 with a size of 200×3×20×48. This tensor contains the three-dimensional temperature field of the passive waste heat removal heat exchanger at the same time and under different working conditions;

[0055] Step 4: Build Figure 1 The artificial neural network shown includes input layer, linear layer, convolutional layer and output layer:

[0056] Step 4-1: Establish an input layer, which contains 3 neurons to receive the 3 key operating parameters contained in the i-th group of simulation conditions: a i , b i , c i , arranged as a one-dimensional tensor D1 = [a i , b i , c i ], size is 1×3;

[0057] Step 4-2: Establish a fully connected layer. The linear layer is used to receive the one-dimensional tensor D1, transform it linearly and output it into a one-dimensional tensor D2 with a size of 1×2880, and then concatenate it into a three-dimensional tensor D3 with a size of 3×20×48;

[0058] Step 4-3: Establish a convolutional layer, which is used to receive the three-dimensional tensor D3, and output a three-dimensional tensor D4 of size 3×20×48 after convolution;

[0059] Step 4-4: Establish an output layer, which is used to receive the three-dimensional tensor D4 and obtain the actual output of the artificial neural network after linear transformation: a three-dimensional tensor D5 of size 3×20×48, which contains the three-dimensional temperature field data of the heat exchanger at the i-th working condition and time t;

[0060] Step 5: Take the key operating parameters of N groups of simulation conditions as the input of the artificial neural network, take the four-dimensional tensor obtained in step 3-3 as the target output, take the root mean square error between the actual output D5 and the target output D0 of the artificial neural network as the convergence criterion, train the artificial neural network built in step 4 until convergence, and obtain the artificial neural network at time t;

[0061] Step 6: Repeat steps 3 to 5 for each moment t=1, 2, 3…20, and 20 trained artificial neural networks can be obtained, which can be used to predict the artificial neural network of the three-dimensional temperature field of the passive waste heat removal heat exchanger under different working conditions and specific moments;

[0062] Step 7: Input the three parameters of the passive waste heat removal heat exchanger under the new operating condition: primary side inlet flow, primary side inlet temperature and secondary side initial temperature into 20 trained artificial neural networks respectively, and the three-dimensional temperature field distribution of the passive waste heat removal heat exchanger at each moment under the new operating condition can be quickly obtained, such as Figure 2 As shown in the figure, it is a comparison diagram between the three-dimensional numerical simulation results and the temperature field fast prediction results of the method of the present invention. Figure 2 It can be seen that the method of the present invention can quickly predict the three-dimensional transient temperature field of the AP1000 reactor passive residual heat removal heat exchanger at different operating times. The prediction results of the method of the present invention are in good agreement with the three-dimensional numerical simulation results, and the method of the present invention greatly reduces the large amount of computing time and computing resources required for three-dimensional simulation, which shows the correctness and advantages of the method of the present invention.

[0063] Parts of the present invention that are not described in detail belong to common knowledge among those skilled in the art.

Claims

1. A method for rapid prediction of three-dimensional transient temperature field of passive waste heat removal heat exchanger, characterized in that: The steps include: Step 1: Randomly sample within the operating range of the passive waste heat removal heat exchanger to obtain several groups of simulation conditions. The specific steps are as follows: Step 1-1: Determine the operating parameters required for the simulation of the passive waste heat removal heat exchanger: primary side inlet flow a, primary side inlet temperature b, secondary side initial temperature c; determine the value range of each parameter under the normal operating conditions of the passive waste heat removal heat exchanger, that is: a∈A=[a min , a max ] (1) b∈B=[b min , b max ] (2) c∈C=[c min , c max ] (3) Among them, A, B, and C represent the value range of the primary side inlet flow rate a, the value range of the primary side inlet temperature b, and the value range of the secondary side initial temperature c respectively; the subscript min represents the minimum value, and max represents the maximum value; Step 1-2: Use the random function in Excel software to randomly sample each key operating parameter within the range of each parameter value to obtain N groups of simulation conditions, namely: a i =a min +RAND()·(a max -a min ), i=1,2,3...N (4) b i =b min +RAND()·(b max -b min ), i=1,2,3...N (5) c i =c min +RAND()·(c max -c min ), i=1,2,3...N (6) Among them, a i , b i , c i They represent the primary inlet flow, primary inlet temperature, and secondary initial temperature of the i-th group of simulation conditions respectively; N represents the total number of simulation conditions; RAND() is a random function that takes random values ​​between 0 and 1 and is dimensionless; Step 2: Based on the N groups of simulation conditions sampled in step 1, set the simulation boundary conditions of the primary and secondary sides of the passive waste heat removal heat exchanger in the open source numerical simulation software OpenFOAM, and carry out three-dimensional transient numerical simulation calculations of the passive waste heat removal heat exchanger under N groups of simulation conditions; each group of simulation conditions can obtain T calculation result files at different times, totaling N×T calculation result files; Step 3: Post-process the N calculation result files at a certain time t, t = 1, 2, 3...T, and obtain a four-dimensional tensor containing the three-dimensional temperature field data of the passive residual heat removal heat exchanger at a certain time. The specific steps are as follows: Step 3-1: Import the N calculation result files at time t into the post-processing software ParaView, and use the slicing function to establish M two-dimensional planes of the same size inside the passive waste heat removal heat exchanger. The number of grids in the length and width directions of the two-dimensional planes are X and Y respectively; Step 3-2: Export the temperature field data of M two-dimensional planes under N working conditions from ParaView in the format of csv files, and obtain a total of N×M csv files; Step 3-3: Use the open source machine learning tool pyTorch to read the data of N×M csv files and arrange them into a four-dimensional tensor D0 with a size of N×M×X×Y; the tensor contains the three-dimensional temperature field data of the passive waste heat removal heat exchanger at the same time and under different working conditions; Step 4: Build an artificial neural network, including input layer, linear layer, convolutional layer and output layer: Step 4-1: Establish an input layer, which contains 3 neurons to receive the 3 key operating parameters contained in the i-th group of simulation conditions: a i , b i , c i , arranged as a one-dimensional tensor D1 = [a i , b i , c i ], size is 1×3; Step 4-2: Establish a fully connected layer. The linear layer is used to receive the one-dimensional tensor D1, transform it linearly and output it as a one-dimensional tensor D2 with a size of 1×(M×X×Y), and then concatenate it into a three-dimensional tensor D3 with a size of M×X×Y; Step 4-3: Establish a convolutional layer, which is used to receive the three-dimensional tensor D3, and output a three-dimensional tensor D4 of size M×X×Y after convolution; Step 4-4: Establish an output layer, which is used to receive the three-dimensional tensor D4 and obtain the actual output of the artificial neural network after linear transformation: a three-dimensional tensor D5 of size M×X×Y, which contains the three-dimensional temperature field data of the heat exchanger at the i-th working condition and time t; Step 5: Take the key operating parameters of N groups of simulation conditions as the input of the artificial neural network, take the four-dimensional tensor obtained in step 3-3 as the target output, take the root mean square error between the actual output D5 and the target output D0 of the artificial neural network as the convergence criterion, train the artificial neural network built in step 4 until convergence, and obtain the artificial neural network at time t; Step 6: Repeat steps 3 to 5 for each moment t=1, 2, 3…T, and T trained artificial neural networks can be obtained, which can be used to predict the three-dimensional temperature field of the passive waste heat removal heat exchanger under different working conditions and at specific times; Step 7: Input the three parameters of the passive waste heat removal heat exchanger under the new operating condition: primary side inlet flow, primary side inlet temperature and secondary side initial temperature into T trained artificial neural networks respectively, so as to quickly obtain the three-dimensional temperature field distribution of the passive waste heat removal heat exchanger at T moments under the new operating condition, and realize the rapid prediction of the three-dimensional transient temperature field of the heat exchanger.

Citation Information

Cited By

  • Secondary side passive residual heat removal system physical process failure analysis method based on adaptive neural network

    CN120822426A