CN112819053A - Model library establishing method and device, diagnosis method and device, and prediction method - Google Patents
Model library establishing method and device, diagnosis method and device, and prediction method Download PDFInfo
- Publication number
- CN112819053A CN112819053A CN202110095211.8A CN202110095211A CN112819053A CN 112819053 A CN112819053 A CN 112819053A CN 202110095211 A CN202110095211 A CN 202110095211A CN 112819053 A CN112819053 A CN 112819053A
- Authority
- CN
- China
- Prior art keywords
- model
- physical parameters
- diagnosis
- regression
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 106
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000013145 classification model Methods 0.000 claims abstract description 115
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 29
- 238000011156 evaluation Methods 0.000 claims description 23
- 230000000977 initiatory effect Effects 0.000 claims description 23
- 238000012549 training Methods 0.000 claims description 12
- 238000013136 deep learning model Methods 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- 239000001257 hydrogen Substances 0.000 claims description 7
- 229910052739 hydrogen Inorganic materials 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 2
- 125000004435 hydrogen atom Chemical class [H]* 0.000 claims 2
- 230000000694 effects Effects 0.000 claims 1
- 239000002826 coolant Substances 0.000 description 4
- 150000002431 hydrogen Chemical class 0.000 description 4
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 2
- 241000321453 Paranthias colonus Species 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012351 Integrated analysis Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Strategic Management (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Tourism & Hospitality (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a method for establishing a diagnosis model library, which comprises the following steps: determining a diagnosis target of a serious accident; acquiring time sequences of M physical parameters under N initial events to form N sequence sets; sequentially selecting T time sequences with the same physical parameter from the N sequence sets respectively to form a data set, wherein T is more than or equal to 2 and is less than M; and acquiring a classification model and a regression model corresponding to the T physical parameters, and storing the T physical parameters, the classification model corresponding to the T physical parameters, the regression model and the corresponding relation among the classification model, the regression model and the regression model to a model library to form a diagnosis model library. Further, a device for establishing a diagnosis model base, a method and a device for diagnosing the serious accident of the nuclear power plant and a method for predicting the progress of the serious accident of the nuclear power plant are also provided. The diagnosis model obtained based on the establishment method of the diagnosis model library can accurately diagnose the initial events of the serious accidents and the parameters of the corresponding initial events according to a limited number of instruments when the serious accidents occur.
Description
Technical Field
The invention belongs to the technical field of nuclear power serious accident analysis, and particularly relates to a method and a device for establishing a diagnosis model base, a method and a device for diagnosing a nuclear power plant serious accident and a process prediction method for the nuclear power plant serious accident.
Background
The major Accident Management Guidelines (SAMG) is an instructional Management document used by staff in a main control room and a technical support center to take measures to alleviate the consequences of a core damage Accident when a major Accident occurs in a nuclear power plant.
One of the key factors for the successful implementation of SAMG is the need to timely, comprehensively and accurately acquire the physical parameters of the state of the power plant through the instrument. However, when a serious accident occurs, the instrument for monitoring the physical parameters may be damaged or have inaccurate data, so that the accuracy of the existing serious accident diagnosis result is not high.
Disclosure of Invention
The invention aims to solve the technical problems in the prior art, and provides a method and a device for establishing a diagnosis model base, a method and a device for diagnosing a severe accident of a nuclear power plant and a method for predicting the progress of the severe accident of the nuclear power plant, so that when the severe accident occurs, the initial event and corresponding parameters of the severe accident can be accurately diagnosed according to the physical parameters monitored by a limited number of instruments, and accurate data is provided for the effective implementation of SAMG.
In a first aspect, an embodiment of the present invention provides a method for building a diagnostic model library, including: determining a diagnosis target of the serious accident, wherein the diagnosis target of the serious accident comprises mapping relations between N initial events and corresponding parameters thereof; acquiring time sequences of M physical parameters under N starting events to form N sequence sets, wherein each sequence set comprises M lines of data, each line of data is a single time sequence of the physical parameters, the M physical parameters comprise steam generator water level, steam generator pressure, primary circuit pressure, reactor core temperature, water level in a pressure vessel, containment water level, containment hydrogen, containment pressure and containment temperature measured by an instrument, and N and M are positive integers; sequentially selecting T time sequences with the same physical parameter from N sequence sets to form a data set, wherein the data set comprises N first sequence sets, each first sequence set comprises T columns of data, T is more than or equal to 2 and is less than M, and T is a positive integer; the method comprises the steps of obtaining classification models and regression models corresponding to T physical parameters, and storing the T physical parameters, the classification models corresponding to the T physical parameters, the regression models and the corresponding relations among the classification models, the regression models and the regression models to a model base to form a diagnosis model base, wherein the number of the classification models is one.
Preferably, after the storing the T physical parameters, the classification model corresponding to the T physical parameters, the regression model, and the correspondence relationship among the three into the model library to form the diagnosis model library, the method for establishing the diagnosis model library further includes: and sequentially and respectively selecting T + T time sequences with the same physical parameter from the N sequence sets, wherein T is 1,2, 3.
Preferably, the obtaining a time sequence of the M physical parameters under the N origination events forms N sequence sets, which specifically includes: and calling a serious accident integration analysis program to respectively carry out simulation calculation on the parameters corresponding to the N initial events, and acquiring time sequences of the M physical parameters under the N initial events to form N sequence sets.
Preferably, the sequentially selecting T time sequences with the same physical parameter from the N sequence sets to form a data set specifically includes: and sequentially and respectively selecting T time sequences with the same physical parameters, wherein the correlation with the diagnosis target is greater than a first correlation threshold value, and the availability of the instrument is greater than the first instrument threshold value, from the N sequence sets to form a data set.
Preferably, the obtaining of the classification model and the regression model corresponding to the T physical parameters specifically includes: deep learning model training and performance evaluation are carried out on the data set by adopting a long-short term memory algorithm, and a classification model set and a regression model set corresponding to the T physical parameters are obtained; and performing performance evaluation on each model in the classification model set and the regression model set corresponding to the T physical parameters, acquiring the classification model and the regression model corresponding to the performance evaluation value exceeding a performance threshold value, and determining the classification model and the regression model corresponding to the T physical parameters. The method includes the steps of sequentially and respectively selecting T + T time sequences with the same physical parameters from N sequence sets, wherein T is 1,2,3, M-T, and traversing the value of T to obtain a classification model and a regression model corresponding to each T + T physical parameters, and specifically includes the steps of: sequentially and respectively selecting T + T time sequences with the same physical parameters from the N sequence sets, wherein T is 1,2,3, M-T, traversing the value of T, and respectively carrying out deep learning model training and performance evaluation on the time sequences selected each time by adopting a long-short term memory algorithm to obtain a classification model set and a regression model set corresponding to the T + T physical parameters; and performing performance evaluation on each model in the classification model set and the regression model set corresponding to each T + T physical parameters, acquiring the classification model and the regression model corresponding to the performance evaluation value exceeding a performance threshold value, and determining the classification model and the regression model corresponding to each T + T physical parameter, wherein the number of the classification models is the same as the selection times of T.
In a second aspect, an embodiment of the present invention further provides a method for diagnosing a severe accident of a nuclear power plant, including: obtaining a diagnosis model library established by the method for establishing the diagnosis model library in the first aspect; after a serious accident occurs, a diagnosis result of the serious accident initial event and corresponding parameters thereof is obtained according to real-time data of physical parameters monitored by available instruments on site and a predetermined classification model and regression model in a diagnosis model library.
Preferably, the obtaining of the diagnosis result of the serious accident initiating event and the corresponding parameters thereof according to the real-time data of the physical parameters monitored by the available meters on the site and the predetermined classification model and regression model in the diagnosis model library specifically includes: selecting a classification model and a regression model corresponding to the physical parameters monitored by the on-site available instruments from a diagnosis model library as the preset classification model and regression model; and inputting real-time data of the physical parameters monitored by the on-site available instruments into the preset classification model and the regression model to obtain the diagnosis result of the serious accident initial event and the corresponding parameters of the serious accident initial event, wherein the classification model is used for diagnosing the serious accident initial event, and the regression model is used for diagnosing the corresponding parameters of the serious accident initial event.
In a third aspect, an embodiment of the present invention further provides a method for predicting a process of a severe accident of a nuclear power plant, including: and inputting the diagnosis results of the serious accident initiating event and the corresponding parameters thereof obtained by the diagnosis method for the serious accident of the nuclear power plant according to the second aspect into a serious accident integrated analysis program to obtain a predicted value of the progress of the serious accident, wherein the predicted value comprises the positive and negative influences of the intervention means.
In a fourth aspect, an embodiment of the present invention further provides an apparatus for building a diagnostic model library, including a determining module, an obtaining module, a first selecting module, and a building module. The system comprises a determining module, a judging module and a judging module, wherein the determining module is used for determining the diagnosis target of the serious accident, and the diagnosis target of the serious accident comprises the mapping relation between N initial events and corresponding parameters thereof. The acquisition module is connected with the determination module and used for acquiring time sequences of M physical parameters under N starting events to form N sequence sets, wherein each sequence set comprises M rows of data, each row of data is a single time sequence of the physical parameters, the M physical parameters comprise steam generator water level, steam generator pressure, primary circuit pressure, reactor core temperature, pressure vessel water level, containment hydrogen, containment pressure and containment temperature measured by an instrument, and N and M are positive integers. The first selection module is connected with the acquisition module and used for sequentially selecting T time sequences with the same physical parameter from the N sequence sets to form a data set, wherein the data set comprises N first sequence sets, each first sequence set comprises T lines of data, T is more than or equal to 2 and less than M, and T is a positive integer. And the construction module is connected with the first selection module and is used for acquiring the classification models and the regression models corresponding to the T physical parameters and storing the T physical parameters, the classification models corresponding to the T physical parameters, the regression models and the corresponding relations among the classification models, the regression models and the regression models to the model base to form a diagnosis model base, wherein the number of the classification models is one.
In a fifth aspect, an embodiment of the present invention further provides a diagnostic apparatus for a severe accident of a nuclear power plant, including the establishing apparatus for the diagnostic model library described in the fourth aspect, a second selecting module, and a diagnostic module. And the device for establishing the diagnosis model library is used for establishing the diagnosis model library. And the second selection module is connected with the establishing device of the diagnosis model base and is used for selecting the classification model and the regression model corresponding to the physical parameters monitored by the on-site available instruments from the diagnosis model base as the preset classification model and the regression model. And the diagnosis module is connected with the second selection module and used for inputting the real-time data of the physical parameters monitored by the on-site available instruments into the preset classification model and the regression model to obtain the diagnosis results of the serious accident initiating event and the corresponding parameters of the serious accident initiating event, wherein the classification model is used for diagnosing the serious accident initiating event, and the regression model is used for diagnosing the corresponding parameters of the serious accident initiating event.
According to the method and the device for establishing the diagnosis model base, the method and the device for diagnosing the serious accident of the nuclear power plant and the method for predicting the progress of the serious accident of the nuclear power plant, the diagnosis target of the serious accident is determined; acquiring time sequences of M physical parameters under N initial events according to a diagnosis target to form N sequence sets; sequentially selecting T time sequences with the same physical parameter from N sequence sets to form a data set, wherein N and M are positive integers, T is more than or equal to 2 and less than M, and T is a positive integer; and acquiring classification models and regression models corresponding to the T physical parameters according to the data set, and storing the classification models and the regression models into a model library to form a diagnosis model library. After a serious accident occurs, the initial event and corresponding parameters of the serious accident can be accurately diagnosed according to a limited number of physical parameters which can be monitored by instruments and classification models and regression models in a diagnosis model library, so that accurate data can be provided for the effective implementation of SAMG.
Drawings
FIG. 1: a flow chart of a method for establishing a diagnostic model library according to embodiment 1 of the present invention;
FIG. 2: a structure diagram of a diagnostic model library creation apparatus according to embodiment 4 of the present invention;
FIG. 3: a configuration diagram of a diagnosis apparatus for a serious accident of a nuclear power plant according to embodiment 5 of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.
Example 1:
as shown in fig. 1, the present embodiment provides a method for building a diagnostic model library, including:
In this embodiment, the diagnosis target of the serious accident is determined, that is, the mapping relationship between the N originating events and their corresponding parameters is determined. As shown in the initial event list of table 1 (not limited to the initial events and their parameters listed in the table), the initial event of "primary coolant system breach" in the table has corresponding parameter values such as breach size, breach location, and blow-off factor, and the specific parameter values corresponding to N initial events of a severe accident can be determined by looking up the literature.
TABLE 1
| Serial number | Initiating events | Parameters of an originating event |
| 1 | Primary circuit coolant system break | Size of the break, position of the break, spray coefficient, etc |
| 2 | Water supply pipeline break | Size of the break, location of the break, etc |
| 3 | Main steam pipeline break | Size of the break, location of the break, etc |
| 4 | Steam generator tube rupture | Location of rupture, number of ruptures, etc |
| ... | ...... | ...... |
| N-1 | ...... | ...... |
| N | ...... | ...... |
In this embodiment, the physical parameters refer to parameters that can still be monitored by the instruments after a serious accident occurs, such as steam generator water level, steam generator pressure, primary circuit pressure, core temperature, pressure vessel water level, containment hydrogen, containment pressure, containment temperature, and other physical parameters. In the embodiment, simulation calculation is performed on the parameters corresponding to the N initial events by calling a severe accident integration analysis program, so that time sequences of the M physical parameters under the N initial events are obtained, and N sequence sets are formed. For example, a time series of M physical parameters developed at the initial event of a circuit coolant system breach is obtained to obtain a sequence set, and the sequence set includes M columns of data, wherein each column of data is a time series of physical parameters. Common integrated analysis programs for serious accidents comprise MAAP developed by Fauske & Associates in the United states, MELCOR developed by Sandia national laboratory in the United states, ASTEC developed by IRSN in France and GRS in Germany in combination, and the like, and physical parameters can be set in an input card of the integrated analysis program for serious accidents.
And 103, sequentially selecting T time sequences with the same physical parameter from the N sequence sets to form a data set, wherein the data set comprises N first sequence sets, each first sequence set comprises T rows of data, T is more than or equal to 2 and is less than M, and T is a positive integer.
In the embodiment, a time sequence of four physical parameters, namely, a steam generator water level, a primary circuit pressure, a core temperature and a pressure vessel water level, is selected from a 1 st sequence set, wherein T is 4; then selecting the time sequences of the four same physical parameters from the 2 nd sequence set; and sequentially selecting the time sequences until the four same physical parameters are selected from the Nth sequence set to obtain a data set, wherein the data set comprises N first sequence sets, and each first sequence set comprises the time sequences of the water level of the steam generator, the pressure of a primary circuit, the temperature of the reactor core and the water level in the pressure vessel.
Optionally, T time sequences with the same physical parameter, which have a correlation with the diagnostic target greater than the first correlation threshold and meter availability greater than the first meter threshold, are sequentially selected from the N sequence sets, respectively, to form a data set.
In the embodiment, the physical parameters which are relevant to the diagnosis target and larger than the first relevant threshold value and the meter availability of which is larger than the first meter threshold value are selected, so that the classification models and regression models corresponding to the T physical parameters are matched with the important meter combination with the best availability. It should be noted that, the correlations between the M physical parameters and the diagnostic target respectively are obtained in advance according to experiments or experiences, and a list sorted according to the correlations is obtained, where the list includes the physical parameters, the correlations, and the mapping relationship between the physical parameters and the correlations. In addition, an ordered list of the availability of the meters is obtained in advance according to experiments or experiences, and the meters have a corresponding relationship with the physical parameters.
And 104, acquiring the classification models and the regression models corresponding to the T physical parameters, and storing the T physical parameters, the classification models corresponding to the T physical parameters, the regression models and the corresponding relations among the classification models, the regression models and the regression models to a model base to form a diagnosis model base, wherein the number of the classification models is one.
In the embodiment, deep learning model training and performance evaluation are performed on the data set obtained in the step 103, and classification models corresponding to 4 physical parameters are established for diagnosing the serious accident initiating event, wherein the number of the classification models is one; for the diagnosis of the parameters of the serious accident initiating event, regression models corresponding to 4 physical parameters are established, the number of the regression models exceeds N, for example, the number of the regression models of the corresponding parameters for diagnosing the initiating event of 'primary coolant system break' is three, and the regression models are respectively used for diagnosing the break position, the break size and the blow-off coefficient. And storing 4 physical parameters (namely steam generator water level, primary circuit pressure, reactor core temperature and water level in the pressure vessel), the classification model corresponding to the 4 physical parameters, the regression model and the corresponding relation among the classification model, the regression model and the three into a model base to form a diagnosis model base. By using the classification model and the regression model in the diagnosis model library, diagnosis of the initial event of the serious accident and the parameters thereof can be realized.
Optionally, step 104 specifically includes: deep learning model training and performance evaluation are carried out on the data set obtained in the step 103 by adopting a long-short term memory algorithm, and a classification model set and a regression model set corresponding to the T physical parameters are obtained; and performing performance evaluation on each model in the classification model set and the regression model set corresponding to the T physical parameters, acquiring the classification model and the regression model corresponding to the performance evaluation value exceeding a performance threshold value, and determining the classification model and the regression model corresponding to the T physical parameters.
Optionally, after the T physical parameters, the classification model corresponding to the T physical parameters, the regression model, and the correspondence relationship between the classification model and the regression model are stored in the model library to form a diagnosis model library, the method for establishing the diagnosis model library further includes:
and 105, sequentially and respectively selecting T + T time sequences with the same physical parameters from the N sequence sets, wherein T is 1,2, 3.
In this embodiment, when a serious accident occurs, because the situation that the instrument on the site for monitoring the physical parameters is damaged or the data is inaccurate is uncertain, and the more the types of the physical parameters in the training data set are, the higher the accuracy of the corresponding classification model and regression model is, the classification model and regression model corresponding to various situations are established to perfect the model in the diagnosis model library. On the basis of the embodiment, 4+1 time sequences with the same physical parameters (such as steam generator water level, primary circuit pressure, reactor core temperature, water level in a pressure vessel and containment water level) are respectively selected from the N sequence sets to form a data set, deep learning model training and performance evaluation are carried out on the data set, a classification model and a regression model corresponding to the 4+1 physical parameters are obtained and stored in a diagnosis model library; then respectively selecting 4+2 time sequences with the same physical parameters (such as steam generator water level, primary circuit pressure, reactor core temperature, water level in a pressure vessel, containment water level and containment hydrogen) from the N sequence sets to form a data set, carrying out deep learning model training and performance evaluation on the data set, obtaining a classification model and a regression model corresponding to the 4+2 physical parameters, and storing the classification model and the regression model into a diagnosis model library; and traversing the value of t until obtaining the classification model and the regression model corresponding to the M physical parameters. It should be noted that, in this embodiment, the classification model and the regression model corresponding to each physical parameter combination are obtained in a manner that one physical parameter is increased incrementally, or the classification model and the regression model corresponding to each physical parameter combination may be obtained in a manner that a plurality of physical parameters are increased incrementally.
Optionally, step 105 specifically includes: sequentially and respectively selecting T + T time sequences with the same physical parameters from the N sequence sets, wherein T is 1,2,3, M-T, traversing the value of T, and respectively carrying out deep learning model training and performance evaluation on the time sequences selected each time by adopting a long-short term memory algorithm to obtain a classification model set and a regression model set corresponding to the T + T physical parameters; and performing performance evaluation on each model in the classification model set and the regression model set corresponding to each T + T physical parameters, acquiring the classification model and the regression model corresponding to the performance evaluation value exceeding a performance threshold value, and determining the classification model and the regression model corresponding to each T + T physical parameter, wherein the number of the classification models is the same as the number of selection times (namely traversal times) of T. In addition, the deep learning model training and performance evaluation uses the commonly used deep learning framework, such as deep learning4J, TensorFlow, Pythrch, etc., and uses the LSTM algorithm integrated in these frameworks.
In the method for establishing the diagnostic model library of this embodiment, a data set corresponding to T physical parameters is obtained by simulating an originating event and corresponding parameters thereof, deep learning model training and performance evaluation are performed on the data set to obtain classification models and regression models corresponding to T physical parameters, and the classification models and regression models are stored in the model library to form the diagnostic model library. After a serious accident occurs, the initial event and corresponding parameters of the serious accident are accurately diagnosed according to the physical parameters monitored by the T limited instruments and the classification models and regression models in the diagnosis model library, so that accurate data are provided for the effective implementation of SAMG. In addition, T + T time sequences with the same physical parameters are sequentially selected from the N sequence sets respectively, wherein T is 1,2,3, M-T, values of T are traversed, classification models and regression models corresponding to the T + T physical parameters are obtained and stored in a diagnosis model library, and therefore the number and the quality of the models in the diagnosis model library are improved, a model matched with the physical parameters monitored by the instruments can be selected according to the available instruments on site for diagnosis, the available instruments on site are utilized to the maximum extent, and accuracy of diagnosis results of the originating events and the corresponding parameters based on the models can be further improved.
Example 2:
the embodiment provides a method for diagnosing a severe accident of a nuclear power plant, which comprises the following steps:
step 201, a diagnosis model library established by the method for establishing a diagnosis model library described in embodiment 1 is obtained.
Step 202, after a serious accident occurs, obtaining a diagnosis result of the serious accident initiating event and corresponding parameters thereof according to the real-time data of the physical parameters monitored by the available instruments on site and the preset classification model and regression model in the diagnosis model library.
Optionally, step 202 specifically includes: selecting a classification model and a regression model corresponding to physical parameters monitored by on-site available instruments from a diagnosis model library as a preset classification model and a preset regression model; and inputting real-time data of the physical parameters monitored by the on-site available instruments into a preset classification model and a regression model to obtain the diagnosis result of the serious accident initial event and the corresponding parameters of the serious accident initial event, wherein the classification model is used for diagnosing the serious accident initial event, and the regression model is used for diagnosing the corresponding parameters of the serious accident initial event.
In this embodiment, after a severe accident occurs, when T +5 physical parameters are monitored by the on-site available instrument, a classification model and a regression model corresponding to the T +5 physical parameters are selected from the diagnostic model library as a predetermined classification model and a predetermined regression model, and real-time data of the T +5 physical parameters monitored by the on-site available instrument is input to the predetermined classification model and the predetermined regression model, so as to obtain a diagnostic result of the severe accident initiating event and the corresponding parameters thereof.
In the method for diagnosing the severe accident of the nuclear power plant, the diagnosis result of the initial event and the parameters thereof is obtained according to the perfect diagnosis model in the diagnosis model library and the real-time data of the physical parameters monitored by the on-site available instruments. Because the diagnosis model base is provided with the diagnosis models matched with different physical parameter combinations, the data monitored by the on-site available instruments are utilized to the maximum extent for diagnosis, and the accuracy of the diagnosis result is higher.
Example 3:
the embodiment provides a method for predicting the progress of a serious accident of a nuclear power plant, which comprises the following steps: the diagnosis results of the serious accident initiating event and the corresponding parameters thereof obtained by the diagnosis method for the serious accident of the nuclear power plant described in the embodiment 2 are input into a serious accident integrated analysis program to obtain a predicted value of the progress of the serious accident, wherein the predicted value includes positive and negative influences of an intervention means. The integrated serious accident analysis program comprises MAAP developed by Fauske & Associates in the United states, MELCOR developed by Sandia national laboratory in the United states, ASTEC developed by IRSN in France and GRS in Germany in combination, and the like.
Example 4:
as shown in fig. 2, the present embodiment provides a device 4 for building a diagnostic model library, which includes a determining module 41, an obtaining module 42, a first selecting module 43, and a constructing module 44.
The determining module 41 is configured to determine a diagnosis target of the serious accident, where the diagnosis target of the serious accident includes mapping relationships between the N original events and corresponding parameters thereof. The diagnostic targets are for example as described in table 1 in example 1.
And an obtaining module 42, connected to the determining module 41, configured to obtain a time sequence of M physical parameters under N initiation events, to form N sequence sets, where each sequence set includes M columns of data, each column of data is a time sequence of a single physical parameter, the M physical parameters include a steam generator level, a steam generator pressure, a primary circuit pressure, a core temperature, a pressure vessel water level, a containment hydrogen gas, a containment pressure, and a containment temperature measured by an instrument, and N and M are positive integers.
Optionally, the obtaining module 42 is configured to invoke a severe accident integration analysis program to perform simulation calculation on parameters corresponding to the N initial events, respectively, and obtain a time sequence of the M physical parameters under the N initial events.
And the first selecting module 43 is connected to the obtaining module 42, and is configured to sequentially select T time sequences with the same physical parameter from the N sequence sets, so as to form a data set, where the data set includes N first sequence sets, each first sequence set includes T rows of data, T is greater than or equal to 2 and is less than M, and T is a positive integer.
Optionally, the selecting module 43 is configured to sequentially select, from the N sequence sets, T time sequences with the same physical parameter, where correlation with the diagnostic target is greater than a first correlation threshold and reliability of the meter is greater than the first meter threshold, respectively, to form a data set.
And a building module 44 connected to the first selecting module 43, configured to obtain the classification models and regression models corresponding to the T physical parameters, and store the T physical parameters, the classification models corresponding to the T physical parameters, the regression models, and correspondence between the classification models and the regression models to a model library to form a diagnostic model library, where the number of the classification models is one, and the number of the regression models exceeds N.
Optionally, the building module 44 is further configured to sequentially select T + T time sequences with the same physical parameter from the N sequence sets, where T is 1,2, 3.
Example 5:
as shown in fig. 3, the present embodiment provides a diagnostic apparatus for a severe accident of a nuclear power plant, which includes a building apparatus 4 of a diagnostic model library described in embodiment 4, a second selecting module 51 and a diagnostic module 52,
and the building device 4 of the diagnosis model library is used for building the diagnosis model library.
And a second selecting module 51, connected to the establishing device 4 of the diagnostic model library, for selecting the classification model and the regression model corresponding to the physical parameters monitored by the on-site available instruments from the diagnostic model library as the predetermined classification model and regression model.
And the diagnosis module 52 is connected to the second selection module 51, and is configured to input real-time data of the physical parameters monitored by the on-site available meters into a predetermined classification model and a regression model, and obtain a diagnosis result of the severe accident initiating event and corresponding parameters of the severe accident initiating event, where the classification model is used for diagnosing the severe accident initiating event, and the regression model is used for diagnosing corresponding parameters of the severe accident initiating event.
It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.
Claims (10)
1. A method for building a diagnostic model library, comprising:
determining a diagnosis target of the serious accident, wherein the diagnosis target of the serious accident comprises mapping relations between N initial events and corresponding parameters thereof;
acquiring time sequences of M physical parameters under N starting events to form N sequence sets, wherein each sequence set comprises M lines of data, each line of data is a single time sequence of the physical parameters, the M physical parameters comprise steam generator water level, steam generator pressure, primary circuit pressure, reactor core temperature, water level in a pressure vessel, containment water level, containment hydrogen, containment pressure and containment temperature measured by an instrument, and N and M are positive integers;
sequentially selecting T time sequences with the same physical parameter from N sequence sets to form a data set, wherein the data set comprises N first sequence sets, each first sequence set comprises T columns of data, T is more than or equal to 2 and is less than M, and T is a positive integer;
the method comprises the steps of obtaining classification models and regression models corresponding to T physical parameters, and storing the T physical parameters, the classification models corresponding to the T physical parameters, the regression models and the corresponding relations among the classification models, the regression models and the regression models to a model base to form a diagnosis model base, wherein the number of the classification models is one.
2. The method for building a diagnostic model library according to claim 1, wherein after storing the T physical parameters, the classification models corresponding to the T physical parameters, the regression models, and the correspondence relationship between the T physical parameters and the regression models in the model library to form the diagnostic model library, the method further comprises:
and sequentially and respectively selecting T + T time sequences with the same physical parameter from the N sequence sets, wherein T is 1,2, 3.
3. The method for building a diagnostic model library according to claim 2, wherein the obtaining of time series of M physical parameters under N origination events to form N series sets specifically comprises:
and calling a serious accident integration analysis program to respectively carry out simulation calculation on the parameters corresponding to the N initial events, and acquiring time sequences of the M physical parameters under the N initial events to form N sequence sets.
4. The method for building a diagnostic model library according to claim 2, wherein the sequentially selecting T time sequences having the same physical parameter from the N sequence sets to form a data set comprises:
and sequentially and respectively selecting T time sequences with the same physical parameters, wherein the correlation with the diagnosis target is greater than a first correlation threshold value, and the availability of the instrument is greater than the first instrument threshold value, from the N sequence sets to form a data set.
5. The method for building a diagnostic model library according to claim 2, wherein the obtaining of classification models and regression models corresponding to the T physical parameters specifically comprises:
deep learning model training and performance evaluation are carried out on the data set by adopting a long-short term memory algorithm, and a classification model set and a regression model set corresponding to the T physical parameters are obtained;
performing performance evaluation on each model in the classification model set and the regression model set corresponding to the T physical parameters, acquiring the classification model and the regression model corresponding to the performance evaluation value exceeding the performance threshold value, determining the classification model and the regression model corresponding to the T physical parameters,
the method includes the steps of sequentially and respectively selecting T + T time sequences with the same physical parameters from N sequence sets, wherein T is 1,2,3, M-T, and traversing the value of T to obtain a classification model and a regression model corresponding to each T + T physical parameters, and specifically includes the steps of:
sequentially and respectively selecting T + T time sequences with the same physical parameters from the N sequence sets, wherein T is 1,2,3, M-T, traversing the value of T, and respectively carrying out deep learning model training and performance evaluation on the time sequences selected each time by adopting a long-short term memory algorithm to obtain a classification model set and a regression model set corresponding to the T + T physical parameters;
and performing performance evaluation on each model in the classification model set and the regression model set corresponding to each T + T physical parameters, acquiring the classification model and the regression model corresponding to the performance evaluation value exceeding a performance threshold value, and determining the classification model and the regression model corresponding to each T + T physical parameter, wherein the number of the classification models is the same as the selection times of T.
6. A method for diagnosing a severe accident in a nuclear power plant, comprising:
obtaining a diagnosis model library created by the method for creating a diagnosis model library according to any one of claims 1 to 5;
after a serious accident occurs, a diagnosis result of the serious accident initial event and corresponding parameters thereof is obtained according to real-time data of physical parameters monitored by available instruments on site and a predetermined classification model and regression model in a diagnosis model library.
7. The method for diagnosing the serious accident of the nuclear power plant according to claim 6, wherein the obtaining of the diagnosis result of the serious accident initiating event and the corresponding parameters according to the real-time data of the physical parameters monitored by the available instruments on the site and the predetermined classification model and regression model in the diagnosis model library specifically comprises:
selecting a classification model and a regression model corresponding to the physical parameters monitored by the on-site available instruments from a diagnosis model library as the preset classification model and regression model;
and inputting real-time data of the physical parameters monitored by the on-site available instruments into the preset classification model and the regression model to obtain the diagnosis result of the serious accident initial event and the corresponding parameters of the serious accident initial event, wherein the classification model is used for diagnosing the serious accident initial event, and the regression model is used for diagnosing the corresponding parameters of the serious accident initial event.
8. A method for predicting the progress of a serious accident of a nuclear power plant is characterized by comprising the following steps:
inputting the diagnosis results of the serious accident initial events and the corresponding parameters thereof obtained by the diagnosis method for the serious accident of the nuclear power plant according to claim 6 or 7 into a serious accident integrated analysis program to obtain the predicted value of the serious accident progress, wherein the predicted value comprises the positive and negative effects of the intervention means.
9. The device for establishing the diagnosis model library is characterized by comprising a determining module, an obtaining module, a first selecting module and a constructing module,
the system comprises a determining module, a judging module and a judging module, wherein the determining module is used for determining a diagnosis target of the serious accident, and the diagnosis target of the serious accident comprises N initial events and mapping relations among corresponding parameters of the N initial events;
the acquisition module is connected with the determination module and used for acquiring time sequences of M physical parameters under N starting events to form N sequence sets, wherein each sequence set comprises M rows of data, each row of data is a single time sequence of the physical parameters, the M physical parameters comprise steam generator water level, steam generator pressure, primary circuit pressure, reactor core temperature, pressure vessel water level, containment hydrogen, containment pressure and containment temperature measured by an instrument, and N and M are positive integers;
the first selection module is connected with the acquisition module and is used for sequentially selecting T time sequences with the same physical parameter from the N sequence sets to form a data set, wherein the data set comprises N first sequence sets, each first sequence set comprises T lines of data, T is more than or equal to 2 and less than M, and T is a positive integer;
and the construction module is connected with the first selection module and is used for acquiring the classification models and the regression models corresponding to the T physical parameters and storing the T physical parameters, the classification models corresponding to the T physical parameters, the regression models and the corresponding relations among the classification models, the regression models and the regression models to the model base to form a diagnosis model base, wherein the number of the classification models is one.
10. A diagnostic device for a severe accident of a nuclear power plant, comprising a building device of a diagnostic model library according to claim 9, a second selection module and a diagnostic module,
the building device of the diagnosis model base is used for building the diagnosis model base;
a second selection module connected with the establishment device of the diagnosis model base and used for selecting the classification model and the regression model corresponding to the physical parameters monitored by the on-site available instruments from the diagnosis model base as the preset classification model and the regression model,
and the diagnosis module is connected with the second selection module and used for inputting the real-time data of the physical parameters monitored by the on-site available instruments into the preset classification model and the regression model to obtain the diagnosis results of the serious accident initiating event and the corresponding parameters of the serious accident initiating event, wherein the classification model is used for diagnosing the serious accident initiating event, and the regression model is used for diagnosing the corresponding parameters of the serious accident initiating event.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110095211.8A CN112819053B (en) | 2021-01-25 | 2021-01-25 | Model library establishing method and device, diagnosis method and device and prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110095211.8A CN112819053B (en) | 2021-01-25 | 2021-01-25 | Model library establishing method and device, diagnosis method and device and prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112819053A true CN112819053A (en) | 2021-05-18 |
| CN112819053B CN112819053B (en) | 2024-04-16 |
Family
ID=75859269
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110095211.8A Active CN112819053B (en) | 2021-01-25 | 2021-01-25 | Model library establishing method and device, diagnosis method and device and prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112819053B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114996821A (en) * | 2022-06-28 | 2022-09-02 | 中建八局装饰工程有限公司 | Curtain wall cavity air tightness judgment method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103366047A (en) * | 2013-06-24 | 2013-10-23 | 中国核电工程有限公司 | Severe accident countermeasure computing and analyzing method for nuclear power plant |
| CN104915768A (en) * | 2015-06-02 | 2015-09-16 | 中科华核电技术研究院有限公司 | Method and system for diagnosing and responding and supporting nuclear power plant severe accident |
| CN109597396A (en) * | 2018-11-26 | 2019-04-09 | 国网湖北省电力有限公司电力科学研究院 | A kind of distribution transforming on-line fault diagnosis method based on high amount of traffic and transfer learning |
| CN112001440A (en) * | 2020-08-20 | 2020-11-27 | 苏州鸿哲智能科技有限公司 | Fault diagnosis logic algorithm and system |
-
2021
- 2021-01-25 CN CN202110095211.8A patent/CN112819053B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103366047A (en) * | 2013-06-24 | 2013-10-23 | 中国核电工程有限公司 | Severe accident countermeasure computing and analyzing method for nuclear power plant |
| CN104915768A (en) * | 2015-06-02 | 2015-09-16 | 中科华核电技术研究院有限公司 | Method and system for diagnosing and responding and supporting nuclear power plant severe accident |
| CN109597396A (en) * | 2018-11-26 | 2019-04-09 | 国网湖北省电力有限公司电力科学研究院 | A kind of distribution transforming on-line fault diagnosis method based on high amount of traffic and transfer learning |
| CN112001440A (en) * | 2020-08-20 | 2020-11-27 | 苏州鸿哲智能科技有限公司 | Fault diagnosis logic algorithm and system |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114996821A (en) * | 2022-06-28 | 2022-09-02 | 中建八局装饰工程有限公司 | Curtain wall cavity air tightness judgment method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112819053B (en) | 2024-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hofer et al. | An approximate epistemic uncertainty analysis approach in the presence of epistemic and aleatory uncertainties | |
| Prošek et al. | Review of quantitative accuracy assessments with fast Fourier transform based method (FFTBM) | |
| US6195624B1 (en) | Corrosion analysis system and method | |
| Fantoni et al. | Multiple-failure signal validation in nuclear power plants using artificial neural networks | |
| CN103279657B (en) | A kind of product accelerated degradation test Design Method based on engineering experience | |
| CN110046379A (en) | A kind of structure entirety damnification recognition method based on space-frequency information | |
| Di Maio et al. | Condition-based probabilistic safety assessment of a spontaneous steam generator tube rupture accident scenario | |
| CN112819053B (en) | Model library establishing method and device, diagnosis method and device and prediction method | |
| Lei et al. | Prediction of crucial nuclear power plant parameters using long short‐term memory neural networks | |
| CN109359803B (en) | Reliability evaluation method and device | |
| Fleming et al. | Database development and uncertainty treatment for estimating pipe failure rates and rupture frequencies | |
| CN109545408A (en) | Judgement and prediction technique based on artificial intelligence technology to nuclear power plant reactor fuel element cladding breakage | |
| Freixa et al. | Validation of a BEPU methodology through a blind benchmark activity at the PKL test facility | |
| CN105741184A (en) | Transformer state evaluation method and apparatus | |
| Park et al. | Smart sensing of the RPV water level in NPP severe accidents using a GMDH algorithm | |
| CN114298226A (en) | Fault diagnosis method, device and medium for nuclear power plant | |
| CN115564247A (en) | Intelligent identification and decision-making method and system for nuclear power plant accidents | |
| Farber et al. | Using multiple-model adaptive estimation and system identification for fault detection in nuclear power plants | |
| CN110688152B (en) | Software reliability quantitative evaluation method combining software development quality information | |
| Wu et al. | Online fault diagnosis of nuclear power plants using signed directed graph and fuzzy theory | |
| Santhosh et al. | Transient identification & operator decision support system for PHWR (Diagnostic System) | |
| CN117791856B (en) | Power grid fault early warning method and device based on inspection robot | |
| CN113283070B (en) | Intelligent diagnosis method and system for intrinsic safety of technological process | |
| CN114355850B (en) | Atmospheric and vacuum device fault diagnosis method based on queuing competition algorithm | |
| Grelle et al. | Development and Testing of Fault-Diagnosis Algorithms for Reactor Plant Systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |