Reactor core evaluation and source item estimation method suitable for heavy water reactor unit accident condition
By establishing a simulation model for heavy water reactor accidents and training a deep learning neural network, combining expert knowledge base and logical algorithms, the accuracy of core damage and radioactive source items estimation in heavy water reactor accidents is solved, and automated core damage reports and source items estimation are realized, improving the automation and accuracy of accident response.
Patent Information
- Application Number
- CN202510417534.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-03
- Publication Date
- 2025-07-25
AI Technical Summary
In the event of a serious accident in a heavy water reactor unit, it is difficult for the prior art to accurately evaluate the degree of core damage and radioactive source terms, resulting in large errors in the estimation results, and the filling of the report depends on manual judgment and lacks automation support.
Serious accident analysis software is used to establish a heavy water reactor accident simulation model, generate a data sample library, train a neural network model through deep learning, combine expert knowledge base and logical algorithms to realize automated evaluation and report filling of core damage and radioactive source items.
The core status online evaluation and rapid prediction of radioactive source items in the case of heavy water reactor units in the accident situation are realized, and the core damage and source items are automatically filled, which improves the estimation accuracy and efficiency.
Smart Images
Figure CN120372909A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of core physical analysis of nuclear power plants, and particularly relates to a method for core evaluation and source term estimation under accident conditions of heavy water reactor units. Background Art
[0002] When a severe accident occurs in a heavy water reactor unit, accompanied by the loss of power supply of the power plant or extensive damage to instruments, the power plant operators and emergency response technicians can obtain very limited information about the power plant status. Evaluating the core damage degree of the unit based on this limited available information is an extremely important and very difficult task in the industry.
[0003] Currently, when technical support personnel of heavy water reactor units perform nuclear accident emergency response operations, the calculation of the core damage fraction is only estimated based on the contact dose rate at the severe accident dose rate measurement points outside the containment, and the source term estimation also completely depends on manually judging the accident type and then referring to the final safety analysis report or the severe accident potential release source term report, which has little relation to the core damage fraction. This will lead to large errors and limitations in the source term estimation results. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for core evaluation and source term estimation under accident conditions of heavy water reactor units, which solves the problem of low accuracy in core damage evaluation and radioactive source term estimation when an accident occurs in a heavy water reactor unit. At the same time, it also automatically fills the core damage report and the source term report, providing support for the emergency technology of heavy water reactor nuclear power plants.
[0005] The technical solution of the present invention is as follows: A method for core evaluation and source term estimation under accident conditions of heavy water reactor units, comprising the following steps:
[0006] Step 1: Establish an accident simulation model of a heavy water reactor unit using severe accident analysis software;
[0007] Step 2: Generate a nuclear accident data sample library by adjusting the model parameters, accident scenarios, and accident intervention measures in the software;
[0008] Step 3: Use deep learning artificial intelligence to train a neural network model on the nuclear accident data sample library generated in Step 2;
[0009] Step 4: Generate a big data deep neural network model applicable to accident conditions of heavy water reactor units after training;
[0010] Step 5: Based on an expert knowledge base and deterministic logic, build an identification logic algorithm for the initiating accident type and system equipment action identification under accident conditions of heavy water reactor units;
[0011] Step 6: Package the big data deep neural network model in Step 4 and the recognition logic algorithms for the initial accident type and system equipment action recognition in Step 5;
[0012] Step 7: Obtain the monitoring parameters of the accident conditions of the heavy water reactor unit;
[0013] Step 8: Perform unit conversion preprocessing on the unit parameters obtained in Step 7. The preprocessed parameters can be directly used as the input for the model packaged in Step 6;
[0014] Step 9: Through the packaged big data deep neural network model with the unit input parameters in Step 8, the fuel cladding damage fraction, fuel pellet overheating fraction, and radionuclide release fraction under the accident conditions of the heavy water reactor unit can be calculated in real time;
[0015] Step 10: The data in Step 9 is used for the core damage assessment of the unit and automatically fills the core damage report form of the heavy water reactor unit;
[0016] Step 11: Through the packaged recognition logic algorithms for the initial accident type and system equipment action recognition with the unit input parameters in Step 8, identify the accident occurrence type of the unit and the system action status;
[0017] Step 12: Using the accident occurrence type and system action status in Step 11 as the input, utilize the fast-time calculation characteristics of the severe accident analysis software to quickly predict the future radionuclide source term release under the accident of the heavy water reactor unit;
[0018] Step 13: The radionuclide source term in Step 12 is used for the estimation of the potential release source of the unit and automatically fills the source term report of the heavy water reactor unit.
[0019] The software in Step 1 includes RELAP5 and MAAP5-CANDU.
[0020] In Step 2, adjusting the model-related factors in the software, including the heat transfer tube friction coefficient, the selection of the core melt debris bed model, and the melt friction coefficient in the calandria vessel cooling model, the generated nuclear data sample library data includes: reactor power, primary heat transport system pressure, pressurizer pressure, pressurizer water level, steam generator pressure, steam generator water level, containment pressure, containment water level, fuel cladding damage fraction, fuel pellet overheating fraction, and radionuclide release parameters.
[0021] In step 5 mentioned above, the right end of the logic represents the algorithm output, which is used to indicate whether the LOCA time outside the initiating event heap is triggered. 1 indicates triggering, and 0 indicates non-triggering. The left end of the logic represents the algorithm input, which is used to judge the conditions for triggering LOCA outside the heap. 1 indicates that the condition is met, and 0 indicates that the condition is not met. When the input conditions on the left meet the corresponding judgment conditions, combined with relevant OR, AND, and NOT logics, the LOCA event outside the heap is judged.
[0022] In step 6 mentioned above, the encapsulation process is to encapsulate the deep neural network model into an executable program and develop API input and output interfaces.
[0023] In step 7 mentioned above, through the data acquisition program, the data of the unit is collected and stored in the local database. The monitoring parameters include: main system pressure, moderator level, steam generator pressure, secondary side level of the steam generator, containment pressure, and sump level.
[0024] The preprocessing in step 8 mentioned above includes unit conversion, including: converting the unit of the unit power from %, to MW, converting the unit of the unit level from %, to M, and converting the unit of the unit from Kpa to Mpa.
[0025] The beneficial effects of the present invention are as follows: 1) realizing the online evaluation of the core state and radioactive source term of the accident unit; 2) realizing the rapid prediction of the radioactive source term of the accident unit; 3) realizing the automatic filling of the core damage report and source term report. Brief Description of the Drawings
[0026] Figure 1 It is a schematic diagram of the principle for core evaluation and source term estimation under accident conditions of a heavy water reactor unit provided by the present invention;
[0027] Figure 2 It is a schematic diagram for establishing a big data deep neural network model;
[0028] Figure 3 It is a recognition logic diagram of the initiating accident type (example of LOCA outside the heap);
[0029] Figure 4 It is a recognition logic diagram of system equipment action (example of end shield cooling failure);
[0030] Figure 5 It is a flow chart for core evaluation and source term estimation under accident conditions of a heavy water reactor unit provided by the present invention. Detailed Embodiments
[0031] The present invention will be further described in detail below with reference to the drawings and specific embodiments.
[0032] A method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit. When an accident occurs in a heavy water reactor nuclear power unit, key information of the unit is obtained online from the emergency management platform, including main system pressure, outlet header pressure, containment pressure, containment radioactivity, secondary side radioactivity of the steam generator, and moderator level, etc. The damage fraction of the fuel cladding, the overheating fraction of the fuel pellets, and the release fraction of radionuclides are quickly evaluated through a big data neural network model trained by a computer. At the same time, a preset algorithm logic is used to judge the type of accident and equipment action information, and the fast-time characteristics of accident analysis software are introduced to quickly predict the potential source term of the accident.
[0033] Based on the integrated severe accident analysis program, a model of the heavy water reactor unit is established, and accident sample data is generated by adjusting model parameters, accident scenarios, accident intervention measures, etc. The deep recurrent neural network algorithm is used to learn a large amount of sample data to establish a big data neural network model (as shown in Figure 2 ). By obtaining the key information of the unit online from the emergency management platform, including main system pressure, outlet header pressure, containment pressure, containment radioactivity, secondary side radioactivity of the steam generator, and moderator level, etc., the damage fraction of the fuel cladding, the overheating fraction of the fuel pellets, and the release fraction of radionuclides are calculated and quickly predicted in real time online through the big data neural network model and the preset algorithm logic, and the calculation results will automatically fill the core damage report and the source term report.
[0034] The principle of a method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit is shown in Figure Figure 1 as follows, where the business flow chart for building the neural network model is shown in Figure 2 , the recognition logic for the initial accident type (example of LOCA outside the reactor) is shown in Figure 3 , the recognition logic for system equipment actions (example of end shield cooling failure) is shown in Figure 4 , and the business flow chart for core evaluation and source term estimation of the heavy water reactor unit is shown in Figure 5 .
[0035] A method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit includes the following steps:
[0036] Step 1: Use severe accident analysis software (such as RELAP5, MAAP5-CANDU, etc.) to establish an accident simulation model of the heavy water reactor unit;
[0037] Step 2: Generate a nuclear accident data sample library by adjusting model parameters, accident scenarios, accident intervention measures, etc. in the software;
[0038] Specific adjustments include: adjusting model-related factors in the software, such as: the friction coefficient of heat transfer tubes, the selection of the core melt debris bed model, the friction coefficient of the melt in the calandria vessel cooling model, etc. The data in the generated sample library mainly includes parameters such as reactor power, primary heat transfer system pressure, pressurizer pressure, pressurizer water level, steam generator pressure, steam generator water level, containment pressure, containment water level, fuel cladding damage fraction, fuel pellet overheating fraction, and radionuclide release.
[0039] Step 3: Use deep learning artificial intelligence to train the neural network model with the nuclear accident data sample library generated in Step 2;
[0040] In the present invention, the long short-term memory network (LSTM) model is adopted. The long short-term memory network is a deep learning method and is currently the most widely used model in the field of machine learning, and has numerous applications in the scientific and technological fields.
[0041] Step 4: After training, generate a big data deep neural network model applicable to the accident conditions of the heavy water reactor unit (such as Figure 2 );
[0042] Step 5: Based on technologies such as the expert knowledge base and deterministic logic, build an identification logic algorithm for the initiating accident type and system equipment actions under the accident conditions of the heavy water reactor unit (such as Figure 3 and Figure 4 );
[0043] Such as Figure 3 shown, the right end of the logic represents the algorithm output, which is used to indicate whether the out-of-core LOCA event of the initiating event is triggered (1 means triggered, 0 means not triggered), while the left end of the logic represents the algorithm input, which is used to judge the conditions for triggering the out-of-core LOCA (1 means the condition is satisfied, 0 means the condition is not satisfied). When the input conditions on the left meet the corresponding judgment conditions, combined with relevant OR, AND, and NOT logics, the out-of-core LOCA event is judged.
[0044] Step 6: Package the big data deep neural network model in Step 4 and the identification logic algorithm for the initiating accident type and system equipment actions in Step 5;
[0045] The said packaging process is to package the deep neural network model into an executable program and develop API input and output interfaces.
[0046] Step 7: Obtain the monitoring parameters of the accident conditions of the heavy water reactor unit
[0047] Specifically, through the data acquisition program, collect the data of the unit and store it in the local database.
[0048] The monitoring parameters described above include: main system pressure, moderator level, steam generator pressure, secondary side level of the steam generator, containment pressure, sump level, etc.;
[0049] Step 8: Perform unit conversion preprocessing on the unit parameters obtained in Step 7. The preprocessed parameters can be directly used as the input for the model after encapsulation in Step 6;
[0050] The preprocessing described above includes unit conversion. For example: the unit of unit power is %, which needs to be converted to MW; the unit of unit level is %, which needs to be converted to M; the unit of unit is Kpa, which needs to be converted to Mpa, and so on.
[0051] Step 9: For the unit input parameters in Step 8, through the encapsulated big data deep neural network model, the fuel cladding damage fraction, fuel pellet overheating fraction, and radionuclide release fraction under accident conditions of the heavy water reactor unit can be calculated in real time;
[0052] Among them, the big data deep neural network model learns the model characteristics of the main severe accident analysis software, and continuously learns the relationship characteristics between input and output during the model training process.
[0053] The input mainly involves the data collected by the unit, the fuel cladding damage fraction, the fuel pellet overheating fraction, and the radionuclide release fraction.
[0054] After the big data deep neural network model is trained, the characteristics learned by the model can be output through the input parameters.
[0055] Step 10: The data in Step 9 can be used for the evaluation of the core damage of the unit, and automatically fill in the core damage report form of the heavy water reactor unit;
[0056] The fuel cladding damage fraction, the fuel pellet overheating fraction, and the radionuclide release fraction are the outputs of the neural network model. These three parameters can be used to judge the state of the core and assist the core physics group to make a correct judgment on the core state of the accident unit. For example: whether the core is intact, whether the core has melted, etc.;
[0057] When the fuel cladding damage fraction is greater than 0, it indicates that the core has been damaged; when the fuel pellet overheating fraction is greater than 0, it indicates that the fuel pellet has melted. The radionuclide release fraction is used as the input data for the environmental consequence evaluation under power plant accidents and provides support for emergency decision-making.
[0058] Step 11: For the unit input parameters in Step 8, through the encapsulated logic algorithm for identifying the type of initiating accident and the action state of system equipment, the type of accident that occurred in the unit and the action state of the system can be identified;
[0059] Build logic using technologies such as expert knowledge bases, fuzzy recognition, and fault trees to identify the accident types of the unit. System action status: Mainly track whether the system is in operation. For example: start of the high-pressure safety injection system, start of the low-pressure safety injection system, etc.;
[0060] Take Figure 3 the logic algorithm for identifying the LOCA event outside the reactor vessel of the unit as an example. The main process is as follows:
[0061] Input judgment 1: The pressure of the outlet header of the main heat transfer system is less than 5.9 Mpa;
[0062] Input judgment 2: Triggered by the low main system pressure signal;
[0063] Input judgment 3: Containment pressure < 3.25 Kpa;
[0064] Input judgment 4: Containment radioactivity > 3000 cps;
[0065] Input judgment 5: Sump level > 0.1;
[0066] Input judgment 6: The radioactivity of the secondary side of the steam generator = 0;
[0067] Input judgment 7: Moderator level < 8.6 m.
[0068] For the input parameters of the above judgment conditions, the parameters can be obtained by collecting the corresponding monitoring instruments of the unit. According to Figure 3 the logical relationship, once input judgments 1 / 2 / 5 / 6 / 7 are satisfied, and at least one of input judgments 3 and 4 is satisfied, the unit can be automatically identified as having an LOCA accident outside the reactor vessel currently.
[0069] Step 12: Take the accident occurrence type and system action status in Step 11 as inputs, and use the fast-time calculation characteristics of the severe accident analysis software to quickly predict the future radioactive source term release under the accident of the CANDU unit;
[0070] Step 13: The radioactive source term in Step 12 can be used for estimating the potential release source of the unit and automatically filling the source term report of the CANDU unit.
[0071] First: This report of the CANDU unit has a fixed template file in the power plant. Automatically generating this report is to collect the data of the unit, combine the calculation results of the present invention, and automatically fill and generate this report of the power plant by developing a data interface program. It can be considered that: the automatic generation of the report is just a digital automatic filling of the previous manual filling process of the report based on the calculation results of the system.
[0072] A method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit generally includes establishing an accident simulation model of a heavy water reactor unit using internationally recognized severe accident analysis software; generating a nuclear accident data sample library by adjusting model parameters, accident scenarios, accident intervention measures, etc., and training to generate a big data deep neural network model; building a logic algorithm for identifying the initial accident type and system equipment action under accident conditions of a heavy water reactor unit based on technical means such as an expert knowledge base and deterministic logic; by obtaining the monitoring parameters of the accident conditions of a heavy water reactor unit, using the big data deep neural network model to calculate in real time the fuel cladding damage fraction, fuel pellet overheating fraction and radionuclide release fraction under the accident conditions of a heavy water reactor unit, and automatically filling in the core damage report form; at the same time, using the logic algorithm for identifying the initial accident type and system equipment action and the fast-time calculation characteristics of the severe accident analysis software, estimating the potential release sources of the unit, and automatically filling in the source term report form. The method for core evaluation and source term estimation of a heavy water reactor unit implemented based on this method has certain similarities with other methods applicable to core damage evaluation and source term estimation under accidents of a heavy water reactor unit. If the products of others only include some technical features, they do not belong to infringement, and only covering all the above technical features belongs to infringement.
Claims
1. A method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit, characterized in that It includes the following steps: Step 1: Establish an accident simulation model of the heavy water reactor unit using severe accident analysis software; Step 2: Generate a nuclear accident data sample library by adjusting the model parameters, accident scenarios, and accident intervention measures in the software; Step 3: Use deep learning artificial intelligence to train a neural network model on the nuclear accident data sample library generated in Step 2; Step 4: After training, a big data deep neural network model applicable to the accident conditions of the heavy water reactor unit is generated; Step 5: Based on the expert knowledge base and deterministic logic, build a recognition logic algorithm for the initial accident type and system equipment action under the accident conditions of the heavy water reactor unit; Step 6: Package the big data deep neural network model in Step 4 and the recognition logic algorithm for the initial accident type and system equipment action in Step 5; Step 7: Obtain the monitoring parameters of the accident conditions of the heavy water reactor unit; Step 8: Perform unit conversion preprocessing on the unit parameters obtained in Step 7, and the preprocessed parameters can be directly used as the input of the model packaged in Step 6; Step 9: The input parameters of the unit in Step 8, through the packaged big data deep neural network model, can calculate in real time the fuel cladding damage fraction, fuel pellet overheating fraction, and radionuclide release fraction under the accident conditions of the heavy water reactor unit; Step 10: The data in Step 9 is used for the core damage evaluation of the unit and automatically fills the core damage report form of the heavy water reactor unit; Step 11: The input parameters of the unit in Step 8, through the packaged recognition logic algorithm for the initial accident type and system equipment action, identify the accident type that occurs in the unit and the system action status; Step 12: Using the accident type that occurs and the system action status in Step 11 as the input, utilize the fast-time calculation characteristics of the severe accident analysis software to quickly predict the future radionuclide source term release under the accident of the heavy water reactor unit; Step 13: The radionuclide source term in Step 12 is used for the estimation of the potential release source of the unit and automatically fills the source term report of the heavy water reactor unit.
2. The method for evaluating the reactor core and estimating the source term under accident conditions of a heavy water reactor unit according to claim 1, wherein: The software in Step 1 includes RELAP5 and MAAP5-CANDU.
3. A method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit as claimed in claim 1, characterized in that: In Step 2, adjusting the model-related factors in the software, including the heat transfer tube friction coefficient, the selection of the core melt debris bed model, and the melt friction coefficient in the calandria vessel cooling model, the generated nuclear data sample library data includes: reactor power, primary heat transport system pressure, pressurizer pressure, pressurizer water level, steam generator pressure, steam generator water level, containment pressure, containment water level, fuel cladding damage fraction, fuel pellet overheating fraction, and radionuclide release parameters.
4. A method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit as claimed in claim 1, characterized in that: The right end of the logic in Step 5 represents the algorithm output, which is used to indicate whether the out-of-core LOCA event is triggered. 1 indicates triggered, and 0 indicates not triggered. The left end of the logic represents the algorithm input, which is used to judge the conditions for triggering the out-of-core LOCA. 1 indicates that the condition is satisfied, and 0 indicates that the condition is not satisfied. When the input conditions on the left meet the corresponding judgment conditions, combined with relevant OR, AND, and NOT logics, the out-of-core LOCA event is judged.
5. A method for evaluating the reactor core and estimating the source term under accident conditions of a heavy water reactor unit, as claimed in claim 1, wherein: In the said step 6, the encapsulation process is to encapsulate the deep neural network model into an executable program and develop the API input and output interfaces.
6. A method for evaluating the reactor core and estimating the source term under accident conditions of a heavy water reactor unit as claimed in claim 1, characterized in that: In the said step 7, the data of the unit is collected through the data acquisition program and stored in the local database. The said monitoring parameters include: main system pressure, moderator level, steam generator pressure, secondary side level of the steam generator, containment pressure, and sump level.
7. A method for core evaluation and source term estimation under accident conditions of a heavy water reactor unit as claimed in claim 1, characterized in that: The preprocessing in the said step 8 includes unit conversion, including: converting the unit of unit power from %, to MW, converting the unit of unit level from %, to M, and converting the unit of unit from Kpa, to Mpa.