CN113076684B - Intelligent calculation method for transient parameters in rod adjusting process of nuclear reactor core - Google Patents
Intelligent calculation method for transient parameters in rod adjusting process of nuclear reactor core Download PDFInfo
- Publication number
- CN113076684B CN113076684B CN202110203032.1A CN202110203032A CN113076684B CN 113076684 B CN113076684 B CN 113076684B CN 202110203032 A CN202110203032 A CN 202110203032A CN 113076684 B CN113076684 B CN 113076684B
- Authority
- CN
- China
- Prior art keywords
- reactor core
- nuclear reactor
- parameters
- adjusting process
- rod adjusting
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
-
- 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
Abstract
The invention particularly relates to an intelligent calculation method for transient parameters in a nuclear reactor core rod adjusting process, which comprises the following steps of obtaining an intelligent calculation model for the transient parameters in the nuclear reactor core rod adjusting process by deep machine learning of data pairs matched with input and output related to the nuclear reactor core rod adjusting process; input parameters related to the nuclear reactor core rod adjusting process are calculated through the intelligent transient parameter calculation model in the nuclear reactor core rod adjusting process so as to obtain corresponding output parameters, and the rapid calculation, simulation and prediction of the nuclear reactor core rod adjusting process are realized. The intelligent calculation method provided by the invention is based on an artificial intelligence deep machine learning technology, can realize rapid calculation, analysis and prediction of the nuclear reactor core rod adjusting transient process, and meets the application requirements of nuclear reactor core design, rapid verification of a nuclear reactor system design scheme, real-time simulation and the like.
Description
Technical Field
The invention relates to the technical field of nuclear reactor engineering and computer science, in particular to an intelligent calculation method for transient parameters in a nuclear reactor core rod adjusting process.
Background
Nuclear reactor core design and safety analysis are indispensable key joints for nuclear reactor design. Because of the large number of complex phenomena and processes involved in a nuclear reactor core, various core computing software is often relied upon to implement simulations of various phenomena and processes in a reactor core. In the light water reactor, the control rod is an important means for reactivity control, and clear design requirements are put forward in relevant laws and regulations and guidelines in China so as to ensure the safety of the reactor during the process that the control rod is ejected out of the reactor core under the working conditions of normal putting-out, insertion and accident. In order to accurately simulate the reactor core flux, power and reactor core heat distribution of a control rod in the reactor core moving process, the traditional reactor core calculation software adopts two-dimensional reactor core physical calculation and one-dimensional axial calculation to perform reactor core neutron physical calculation and thermal hydraulic calculation.
The traditional method and model of two-dimensional core physics calculation plus one-dimensional axial calculation have low precision when evaluating accidents with space effect (such as rod ejection accidents, etc.), and in recent years, the method and model of three-dimensional core high-precision calculation are gradually developed to realize the simulation of control rod movement process. However, these three-dimensional core high-precision calculation models and their supporting software usually consume large computing resources, are slow in calculation speed and low in calculation efficiency, and even on a high-performance computer system, the transient parameter calculation in the core rod adjusting process also requires a lot of time, and the high-precision calculation in the transient process often does not have the real-time capability. Meanwhile, the three-dimensional reactor core high-precision calculation method enables errors of calculation results and experimental results to be difficult to effectively feed back to the optimization of the numerical model, and the assimilation capability of data is weak.
In the fields of downstream professional design of nuclear reactors, application simulation of nuclear power systems and the like, the method has the requirement of quickly calculating transient parameters in the rod adjusting process of the nuclear reactors. In actual engineering work, in order to meet the requirement of engineering on rapid calculation of transient parameters in the rod adjusting process of a nuclear reactor core, a method and a model different from three-dimensional core high-precision calculation have to be used in the engineering field, and reactor core simplified calculation models such as a point reactor and a coarse grid node method are adopted to realize rapid calculation of the transient parameters in the rod adjusting process of the nuclear reactor core. Because the three-dimensional dynamic characteristic description of the reactor core by the simplified calculation model of the reactor core has the inherent defect, the calculation precision is greatly reduced compared with the method and the model for calculating the three-dimensional reactor core with high precision.
Disclosure of Invention
Based on this, it is necessary to provide an intelligent calculation method for transient parameters in a nuclear reactor core rod adjusting process based on deep machine learning, aiming at the problem that the existing core calculation method and model cannot meet the requirements of high calculation speed, high precision and low resource consumption of transient parameters in the nuclear reactor core rod adjusting process, and the intelligent calculation method can meet the requirements of high calculation speed, high precision and low resource consumption of transient parameters in the nuclear reactor core rod adjusting process (the process of changing the positions of control rods such as a lifting rod, a rod inserting rod, a rod ejecting rod, a rod falling rod and the like).
In order to achieve the above purpose, the invention provides the following technical scheme:
an intelligent calculation method for transient parameters in a nuclear reactor core rod adjusting process comprises the following steps:
s1: determining data pairs required for deep machine learning and matched with input and output related to a nuclear reactor core rod adjusting process;
s2: acquiring data pairs which are required by deep machine learning and are matched with input and output related to a nuclear reactor core rod adjusting process;
s3: acquiring an intelligent calculation model of transient parameters in the rod adjusting process of the reactor core of the nuclear reactor by learning data pairs matched with input and output related to the rod adjusting process of the reactor core of the nuclear reactor through a deep machine;
s4: calculating input parameters related to the rod adjusting process of the reactor core of the nuclear reactor through an intelligent calculation model of transient parameters in the rod adjusting process of the reactor core of the nuclear reactor to obtain corresponding output parameters, and realizing rapid calculation, simulation and prediction of the rod adjusting process of the reactor core of the nuclear reactor;
the data pairs include input parameters related to a nuclear reactor core rod adjusting process, and corresponding output parameters.
Further, step S1 specifically includes the following steps: selecting calculation input parameters which are required by deep machine learning and are related to the rod adjusting process of the reactor core of the nuclear reactor by analyzing key influence factors of the transient state process of the rod adjusting of the reactor core of the nuclear reactor; selecting calculation output parameters which are required by deep machine learning and are related to the rod adjusting process of the reactor core of the nuclear reactor by analyzing transient parameters of the rod adjusting process of the reactor core of the nuclear reactor; thereby determining data pairs required for deep machine learning that match inputs and outputs associated with a nuclear reactor core rod tuning process.
Further, in step S2, a data pair matching the input and output of the nuclear reactor core rod adjusting process required for the deep machine learning is acquired by calculation with existing high-precision core calculation software or extraction of actual operation data.
Further, the method for acquiring the data pair matched with the input and the output related to the nuclear reactor core rod adjusting process required by the deep machine learning through the calculation of the existing high-precision core calculation software specifically comprises the following steps:
1. the existing high-precision reactor core steady-state calculation software obtains the nuclear reactor core initial state parameters required by the existing high-precision reactor core transient calculation software through calculation; 2. the method comprises the steps of calculating and acquiring transient parameters of the nuclear reactor core rod adjusting process by utilizing initial state parameters of the nuclear reactor core through the conventional high-precision reactor core transient calculation software, thereby acquiring data pairs which are required by deep machine learning and are matched with input and output related to the nuclear reactor core rod adjusting process.
Further, the step 2 specifically comprises the following steps: dividing a nuclear reactor core into a plurality of spatial grids by using the existing high-precision reactor core transient calculation software; calculating the initial state parameters of the nuclear reactor core of each space grid through the existing high-precision reactor core transient calculation software so as to obtain the corresponding transient parameters of the rod adjusting process of the nuclear reactor core; and selecting transient parameters of the nuclear reactor core rod adjusting process of the spatial grid where the detector is located to establish a data pair which is required by deep machine learning and is matched with input and output related to the nuclear reactor core rod adjusting process.
Further, the input parameters related to the nuclear reactor core rod regulating process include one or more combinations of nuclear reactor core initial state parameters such as core initial nuclear power, core initial burnup, core inlet coolant temperature, coolant flow, coolant density, coolant boron concentration, power adjustment control rod position, temperature adjustment control rod position, and core poison concentration.
Further, the output parameters related to the nuclear reactor core rod adjusting process comprise one or more combinations of transient parameters of the nuclear reactor core rod adjusting process, such as core neutron flux distribution, core power distribution and temperature field distribution.
Further, the core neutron flux distribution comprises a core neutron density distribution, the core power distribution comprises core power, core power hot spot factor, core power enthalpy rise factor, core power level and core axial power distribution, and the temperature field distribution comprises cladding temperature, fuel temperature and core outlet coolant temperature.
Further, step S3 specifically includes the following steps: and performing regression learning on the input and output matched data pair related to the rod adjusting process of the nuclear reactor core through an artificial intelligent neural network model to realize supervised deep learning, thereby obtaining an intelligent calculation model of transient parameters in the rod adjusting process of the nuclear reactor core.
Further, the artificial intelligent neural network model comprises one or more of a Fully Connected Neural Network (FCNN), a Convolutional Neural Network (CNN), a long-short term memory artificial neural network (LSTM), or a Recurrent Neural Network (RNN).
Further, when the input and output parameters related to the nuclear reactor core rod adjusting process are one or more static (non-time sequence) variables, selecting a Fully Connected Neural Network (FCNN) model for deep machine learning;
further, when the input and output parameters related to the nuclear reactor core rod adjusting process are one-dimensional or multidimensional time sequence variables, an LSTM or RNN neural network structure model aiming at the time sequence characteristics is selected for deep machine learning.
The invention has the beneficial technical effects that:
the intelligent calculation method for transient parameters in the nuclear reactor core rod adjusting process is based on an artificial intelligent deep machine learning technology, aims at the existing high-precision reactor core calculation software and actual measurement experiment data, and establishes an intelligent calculation model based on data driving through a machine learning method.
1) The intelligent calculation model obtained by the method can give consideration to the calculation precision and speed of the reactor core rod adjusting transient process of the nuclear reactor. On one hand, the input and output data pair used by the method is obtained through the existing high-precision reactor core calculation software or the actual measurement data, the calculation precision of the intelligent calculation model can well approach the input-output characteristics, and the calculation precision of the intelligent calculation model can be effectively ensured. On the other hand, the intelligent calculation model obtained by the method has high calculation efficiency, can realize the rapid calculation of the transient state of the reactor core rod control of the nuclear reactor, has good calculation real-time performance, and can improve the calculation efficiency by more than 10000 times compared with the existing high-precision reactor core calculation software under the same granularity and scale.
2) The neural network model based on data driving established by the invention has good generalization capability on the measured data, and can realize the quick correction of the intelligent calculation model by relearning the measured data under the condition that the neural network structure model is not changed.
3) The intelligent calculation model obtained by the invention has relatively little consumption on calculation resources, can realize high-precision rapid prediction of the reactor core rod adjusting transient process on a desktop computer and embedded equipment, and achieves the effect of the existing high-precision reactor core calculation software on a large-scale supercomputing system, thereby greatly reducing the calculation cost.
Drawings
FIG. 1 is a flow chart of an intelligent calculation method for transient parameters in a nuclear reactor core rod adjusting process according to the present invention;
FIG. 2 is a graph of core power level versus coolant temperature;
FIG. 3 is a design view of a nuclear reactor core rod-adjusting transient process database;
FIG. 4 is a schematic diagram of a static parameter prediction machine learning model;
FIG. 5 is a schematic diagram of a time series transient parameter prediction machine learning model.
In FIG. 2, T 0 Core inlet coolant temperature (deg.C) at zero core power; t is 1 Core inlet coolant temperature (deg.C) at 100% core power level; t is 2 Core outlet coolant average temperature (deg.C) at 100% core power level; t is 3 Core outlet coolant temperature (deg.C) at 100% core power level; the three curves represent, in order from top to bottom: a core power level versus core outlet coolant temperature curve, a core power level versus core outlet coolant average temperature curve, and a core power level versus core inlet coolant temperature curve.
Detailed Description
The intelligent calculation method for transient parameters in the rod adjusting process of the nuclear reactor core is applied to a certain pressurized water reactor core, the pressurized water reactor core consists of 157 fuel assemblies, and each assembly comprises 264 fuel rods arranged in a 17 x 17 square mode, 24 guide pipes capable of placing control rods/burnable poison rods/neutron sources and 1 instrument pipe. The control rod assemblies are functionally divided into control rod groups and shutdown rod groups. The control rod group is composed of power regulating control rods (G1, G2, N1 and N2) and a temperature regulating control rod (R). The power regulating control rod is used for compensating the reactivity change in load tracking. The temperature control rods are used for adjusting the average temperature of the reactor core, compensating the slight change of reactivity and controlling the axial power deviation. The function of the shutdown rod sets (SA, SB and SC) is to ensure the negative reactivity necessary for reactor shutdown. The height of the active section of the core (cold state) is 365.76cm, and the base statepoint is selected as the average core burn-up of 150 MWd/tU.
The intelligent calculation method for the transient parameters in the rod adjusting process of the reactor core of the nuclear reactor comprises the following steps:
s1: selecting calculation input parameters which are required by deep machine learning and are related to the rod adjusting process of the reactor core of the nuclear reactor by analyzing key influence factors of the transient state process of the rod adjusting of the reactor core of the nuclear reactor; selecting calculation output parameters which are required by deep machine learning and are related to the rod adjusting process of the reactor core of the nuclear reactor by analyzing transient parameters of the rod adjusting process of the reactor core of the nuclear reactor; thereby determining data pairs required for deep machine learning that match inputs and outputs associated with a nuclear reactor core rod tuning process.
The key influencing factors of the nuclear reactor core rod regulating transient process comprise nuclear reactor core initial state parameters such as core initial nuclear power, core initial fuel consumption, core inlet coolant temperature, coolant flow, coolant density, coolant boron concentration, power regulation control rod position, temperature regulation control rod position, core poison concentration and the like.
Selecting calculation input parameters related to the nuclear reactor core rod adjusting process required by the depth machine learning, wherein the calculation input parameters include but are not limited to one or more combinations of nuclear reactor core initial state parameters such as core initial nuclear power, core initial fuel consumption, core inlet coolant temperature, coolant flow, coolant density, coolant boron concentration, power adjusting control rod position, temperature adjusting control rod position, core poison concentration and the like.
The transient parameters of the rod adjusting process of the nuclear reactor core include, but are not limited to, reactor core neutron flux distribution such as reactor core neutron density distribution, reactor core power distribution such as reactor core power, reactor core power hot spot factor, reactor core power enthalpy rise factor, reactor core power level and reactor core axial power distribution, and temperature field distribution such as cladding temperature, fuel temperature and reactor core outlet coolant temperature.
Selecting calculation output parameters required by deep machine learning and related to the reactor core rod adjusting process of the nuclear reactor, wherein the calculation output parameters include, but are not limited to, one or more combinations of reactor core neutron flux distribution such as reactor core neutron density distribution, reactor core power hot spot factor, reactor core power enthalpy rise factor, reactor core power level, reactor core axial power distribution and other reactor core power distributions, and temperature field distribution such as cladding temperature, fuel temperature and reactor core outlet coolant temperature.
Detecting reactor core neutron flux distribution through an in-reactor neutron measurement channel in the actual operation process of the nuclear reactor, and obtaining a Deviating Nucleate Boiling Ratio (DNBR) and a reactor core Linear Power Density (LPD) through online calculation; providing the temperature of the coolant at the outlet of the reactor core through a reactor core temperature measuring system, and calculating the highest temperature of the coolant at the outlet of the reactor core, the average temperature of the coolant at the outlet of the reactor core and the lowest supercooling margin of the coolant at the outlet of the reactor core according to the temperature of the coolant at the outlet of the reactor core; the reactor core power level, the change of the reactor core power level and the axial power distribution of the reactor core are continuously monitored by an out-of-core measurement system.
S2: the existing high-precision reactor core steady-state calculation software utilizes reactor core geometric parameters such as component size, fuel rod size, grid division and the like, reactor core arrangement, reactor core power level, reactor core inlet coolant temperature, coolant pressure, coolant flow rate, initial boron concentration, convergence criterion and the like as calculation input parameters, the method comprises the steps of obtaining reactor core steady state parameters such as neutron flux distribution, coolant temperature distribution, fuel temperature distribution, nuclide concentration distribution, power distribution, burnup distribution, critical boron concentration, power regulation control rod position, coolant outlet temperature and the like through calculation, and using high-precision reactor core steady state calculation software to calculate input parameters (such as reactor core power level, reactor core inlet coolant temperature, coolant pressure, coolant flow rate and the like) and the calculated steady state parameters as the reactor core initial state parameters required by a high-precision reactor core transient calculation program.
In this embodiment, 12 core power levels, such as 0%, 3%, 5%, 10%, 20%, 30%, 50%, 70%, 90%, 95%, 100%, 105%, are selected as the initial core nuclear power required by the existing high-precision core steady-state calculation software. Referring to fig. 2, a relationship curve of the core power level and the core inlet coolant temperature is plotted with the core power level as a horizontal axis and the core inlet coolant temperature as a horizontal axis, and the core inlet coolant temperature parameter at different core power levels is determined by the relationship curve of the core power level and the core inlet coolant temperature using the following relationship: tin ═ T 0 +Pn(T 1 -T 0 ) Wherein, Tin: core inlet coolant temperature (° c); t is 0 : core inlet coolant temperature (deg.C) at zero power; t is 1 : core inlet coolant temperature (° c) at 100% core power level; pn: core power level (%).
Based on the selection method of the core power level and the core inlet coolant temperature, 12 groups of steady state parameters of the core at different core power levels, such as 0%, 3%, 5%, 10%, 20%, 30%, 50%, 70%, 90%, 95%, 100%, 105% and the like, are calculated and obtained by using the existing high-precision core steady state calculation program, and the 12 groups of steady state parameters are used as the initial state parameters of the core of the nuclear reactor required by the existing high-precision core transient calculation software.
S3: the method comprises the following steps of utilizing steady state parameters obtained by calculation of existing high-precision reactor core steady state calculation software to serve as initial state parameters of the existing high-precision reactor core transient calculation software, and obtaining the transient parameters of the nuclear reactor core rod adjusting process through calculation of the existing high-precision reactor core transient calculation software, so as to obtain input and output matched data pairs required by deep machine learning and related to the nuclear reactor core rod adjusting process, and specifically comprises the following steps:
dividing a nuclear reactor core into a plurality of spatial grids by using the existing high-precision reactor core transient calculation software; calculating the initial state parameters of the nuclear reactor core of each space grid through the existing high-precision reactor core transient calculation software so as to obtain the corresponding transient parameters of the rod adjusting process of the nuclear reactor core; and selecting transient parameters of the nuclear reactor core rod adjusting process of the spatial grid where the detector is located to establish a data pair which is required by deep machine learning and is matched with input and output related to the nuclear reactor core rod adjusting process.
In this embodiment, with the lifting power schemes shown in fig. 3, each lifting power scheme may adopt a transient scheme in which the control rod lifting/inserting speed (8 steps/min, 24 steps/min, 36 steps/group, 40 steps/min, 56 steps/min, 72 steps/min) is different from the control rod lifting/inserting speed to the designated rod position (the control rod moving step from the bottom 5 steps to the top 225 steps is divided into 20 position points), and the nuclear reactor core rod regulating process transient data 240 caused by the power regulating rod change at different power levels is calculated and obtained by the existing high-precision core transient calculation software, namely 1440 groups.
Utilizing the existing high-precision reactor core transient calculation software to perform three-dimensional space-time neutron dynamics and thermal hydraulic coupling calculation on the 1440 groups of nuclear reactor core rod adjusting process transient processes to obtain reactor core rod adjusting process transient parameters including reactor core neutron flux distribution such as reactor core neutron density distribution, reactor core power distribution such as reactor core power, reactor core power hotspot factor, reactor core power enthalpy rise factor, reactor core power level and reactor core axial power distribution, and temperature distribution such as cladding temperature, fuel temperature and reactor core outlet coolant temperature, so that input and output matched data pairs required by deep machine learning and related to the nuclear reactor core rod adjusting process are obtained.
The above data pairs required for deep machine learning that match the inputs and outputs associated with the nuclear reactor core rod adjusting process may be established based on the operating data of the actual nuclear reactor.
S4: and performing regression learning on the input and output matched data pair related to the rod adjusting process of the nuclear reactor core through an artificial intelligent neural network model, realizing supervised deep learning, and obtaining an intelligent calculation model of transient parameters in the rod adjusting process of the nuclear reactor core.
The data learning and intelligent calculation model establishment can be carried out in various ways, for example, according to application requirements, one or more combinations of initial nuclear power, initial core burnup, core inlet coolant temperature, coolant flow, coolant density, coolant boron concentration, control rod position, poison concentration and other nuclear reactor core initial state parameters of a nuclear reactor can be selected as calculation input parameters; selecting one or more combinations of reactor core power distribution such as reactor core neutron flux distribution, reactor core power hot spot factor, reactor core power enthalpy rise factor, reactor core power level, reactor core axial power distribution and the like of the nuclear reactor and temperature distribution such as cladding temperature, fuel temperature, reactor core outlet coolant temperature and the like as calculation output parameters; and extracting a sufficient amount of data from the large data pool to be used as a data pair which is required by the intelligent calculation model for the transient parameter during the rod-adjusting process of the nuclear reactor core and is matched with the input and output related to the rod-adjusting process of the nuclear reactor core, wherein experiments show that the precision can be higher after the number of the data pairs is more than 1000.
In the network model building mode, a specific neural network structure model is constructed to serve as a basic model of machine learning, when the input variable is one or more static (non-time sequence) variable changes and steady-state result calculation is carried out, a fully-connected network is selected to serve as an input-output neural network model, and stable reactor core state parameters are selected to serve as the fully-connected network for output. Referring to fig. 4, the network structure may select the static parameter input layer as a first layer, the full connection layer as a second layer, and the regression output layer as a third layer. Specifically, for example, the power regulation control rod changes from the position a to the position B under a certain reactor core initial state condition, the reactor core initial state parameter and the power regulation control rod change from the position a to the position B as input, the state parameter when the reactor core reaches the stable state after the power regulation control rod changes from the position a to the position B as output, 64, 128, 256 and other hidden neurons are selected (adaptively modified according to different applications), so that an intelligent calculation model of the steady state result after the reactor core rod regulation of the nuclear reactor is obtained, and the prediction calculation of the steady state parameters such as the reactor core power, the coolant outlet temperature and the like of the nuclear reactor after the power regulation control rod changes from the position a to the position B under a certain initial condition is realized.
Referring to fig. 5, for the case of inputting a neural network including one-dimensional or multi-dimensional time-series variables, the LSTM is selected as the core, the time-series input layer is selected as the first layer, the LSTM layer is selected as the second layer, the active layer is selected as the third layer, the full-link layer is the fourth layer, and the regression output layer is the fifth layer. The main technical parameters comprise: the LSTM input is multidimensional, and the output is a sequence variable; 256, 512, etc. (adaptively modified according to different applications) hidden neurons are selected from the full connection layer. The network structures and parameters can be adaptively modified according to different applications and different computer system performances are optimized, so that an intelligent calculation model of the transient parameters of the nuclear reactor core rod adjusting process, which is just aligned to the sequence values, is obtained.
After the intelligent calculation model of the transient parameters in the rod adjusting process of the nuclear reactor core is established, supervised regression learning is carried out by combining input and output data, and when the error of the result is smaller than a preset value, the learning is stopped. And then checking by using part of data, if the error is greater than a given value, continuously re-learning, otherwise, solidifying the intelligent calculation model of the transient parameters of the nuclear reactor core rod adjusting process, outputting the intelligent calculation model of the transient parameters of the nuclear reactor core rod adjusting process into function libraries (lib \ dll and other various forms) in the forms of C, C + + and the like, and supporting subsequent application programs to be used.
S5: input parameters related to the nuclear reactor core rod adjusting process are calculated through the intelligent transient parameter calculation model in the nuclear reactor core rod adjusting process so as to obtain corresponding output parameters, and the rapid calculation, simulation and prediction of the nuclear reactor core rod adjusting process are realized.
And calling the intelligent calculation model of the transient parameters in the nuclear reactor core rod adjusting process through specific application software (main programs) aiming at different applications. The method comprises the steps of changing the rod position of a control rod by using software, sending information such as relevant boundary condition parameters, rod position sequences and the like into an intelligent calculation model of transient parameters in the rod adjusting process of the reactor core of the nuclear reactor, and rapidly calculating and obtaining the transient parameters in the rod adjusting process of the reactor core of the nuclear reactor by the intelligent calculation model of the transient parameters in the rod adjusting process of the reactor core of the nuclear reactor, wherein the transient parameters include reactor core neutron flux distribution such as reactor core neutron density distribution, reactor core power hot spot factors, reactor core power enthalpy rise factors, reactor core power levels, reactor core axial power distribution and other reactor core power distributions, and temperature distributions such as cladding temperature, fuel temperature and reactor core outlet coolant temperature, so that rapid calculation and simulation prediction of the transient parameters in the rod adjusting process of the reactor core of the nuclear reactor are realized, and the model utilization is realized.
Meanwhile, the intelligent calculation model for transient parameters in the nuclear reactor core rod adjusting process can realize high-precision rapid prediction of the nuclear reactor core rod adjusting transient process on lightweight hardware equipment such as a desktop computer and embedded equipment, achieves the effect of the existing high-precision reactor core calculation software on a large-scale supercomputing system, is applied to scheme design, multidisciplinary optimization design, a DCS system design verification system, a personnel training simulation system, a real-time demonstration system and the like of downstream specialties of a nuclear reactor, and provides an efficient and rapid nuclear reactor core rod adjusting transient calculation module and related calculation functions for the intelligent calculation model.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (6)
1. An intelligent calculation method for transient parameters in a nuclear reactor core rod adjusting process is characterized by comprising the following steps:
s1: determining data pairs required for deep learning and matched with input and output related to a nuclear reactor core rod adjusting process;
s2: acquiring data pairs which are required by deep learning and are matched with input and output related to a nuclear reactor core rod adjusting process;
s3: obtaining an intelligent calculation model of transient parameters in the rod adjusting process of the reactor core of the nuclear reactor by deeply learning data pairs matched with input and output related to the rod adjusting process of the reactor core of the nuclear reactor;
s4: calculating input parameters related to the rod adjusting process of the reactor core of the nuclear reactor through an intelligent calculation model of transient parameters in the rod adjusting process of the reactor core of the nuclear reactor so as to obtain corresponding output parameters;
the data pairs comprise input parameters related to a nuclear reactor core rod adjusting process and corresponding output parameters;
in step S2, obtaining a data pair that matches input and output related to a nuclear reactor core rod adjusting process required for deep learning by calculation or actual operation data extraction using existing high-precision core calculation software;
the method for acquiring the data pair which is required by deep learning and is matched with the input and output related to the nuclear reactor core rod adjusting process through the calculation of the existing high-precision core calculation software specifically comprises the following steps:
(1) acquiring initial state parameters of the reactor core of the nuclear reactor required by the existing high-precision reactor core transient calculation software through calculation by the existing high-precision reactor core steady-state calculation software;
(2) calculating and acquiring transient parameters of a nuclear reactor core rod adjusting process by using initial state parameters of the nuclear reactor core through the conventional high-precision reactor core transient calculation software so as to acquire input and output matched data pairs required by deep learning and related to the nuclear reactor core rod adjusting process;
the step (2) specifically comprises the following steps: dividing a nuclear reactor core into a plurality of spatial grids by using the existing high-precision reactor core transient calculation software; calculating the initial state parameters of the nuclear reactor core of each space grid through the existing high-precision reactor core transient calculation software so as to obtain the corresponding transient parameters of the rod adjusting process of the nuclear reactor core; selecting transient parameters of a nuclear reactor core rod adjusting process of a spatial grid where a detector is located to establish a data pair which is required by deep learning and is matched with input and output related to the nuclear reactor core rod adjusting process;
step S3 specifically includes the following steps: and performing regression learning on the input and output matched data pair related to the rod adjusting process of the nuclear reactor core through an artificial intelligent neural network model to realize supervised deep learning, thereby obtaining an intelligent calculation model of transient parameters in the rod adjusting process of the nuclear reactor core.
2. The intelligent calculation method for transient parameters in a nuclear reactor core rod adjusting process according to claim 1, wherein the step S1 specifically comprises the following steps: selecting calculation input parameters which are required by deep learning and are related to the rod adjusting process of the reactor core of the nuclear reactor by analyzing key influence factors of the transient state process of the rod adjusting of the reactor core of the nuclear reactor; selecting calculation output parameters required by deep learning and related to the rod adjusting process of the reactor core of the nuclear reactor by analyzing transient parameters of the rod adjusting process of the reactor core of the nuclear reactor; thereby determining data pairs required for deep learning that match inputs and outputs associated with a nuclear reactor core rod-tuning process.
3. The intelligent calculation method for transient parameters in nuclear reactor core rod regulating process according to claim 1, wherein the input parameters related to the nuclear reactor core rod regulating process comprise one or more combinations of initial state parameters of the nuclear reactor core, and the output parameters related to the nuclear reactor core rod regulating process comprise one or more combinations of transient parameters of the nuclear reactor core rod regulating process.
4. The intelligent calculation method for transient parameters in the nuclear reactor core rod-tuning process of claim 3, wherein the initial state parameters of the nuclear reactor core comprise initial nuclear power, initial burnup, coolant inlet temperature, coolant flow, coolant density, coolant boron concentration, control rod position, and core poison concentration of the nuclear reactor core; the transient parameters of the nuclear reactor core rod adjusting process comprise reactor core neutron flux distribution, reactor core power distribution and temperature field distribution of the nuclear reactor.
5. The intelligent calculation method for transient parameters of the nuclear reactor core rod adjusting process of claim 1, wherein the artificial intelligent neural network model comprises one or a combination of fully-connected neural networks, convolutional neural networks or cyclic neural networks.
6. The intelligent calculation method for transient parameters of the nuclear reactor core rod adjusting process of claim 5, wherein when the input and output parameters related to the nuclear reactor core rod adjusting process are one or more static variables, the fully-connected neural network model is selected for deep learning and generalization prediction capability; and when the input and output parameters related to the nuclear reactor core rod adjusting process are one-dimensional or multi-dimensional time sequence variables, selecting a circulating neural network structure model aiming at the time sequence characteristics to carry out deep learning and generalization prediction capability.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110203032.1A CN113076684B (en) | 2021-02-23 | 2021-02-23 | Intelligent calculation method for transient parameters in rod adjusting process of nuclear reactor core |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110203032.1A CN113076684B (en) | 2021-02-23 | 2021-02-23 | Intelligent calculation method for transient parameters in rod adjusting process of nuclear reactor core |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113076684A CN113076684A (en) | 2021-07-06 |
| CN113076684B true CN113076684B (en) | 2022-09-20 |
Family
ID=76609542
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110203032.1A Active CN113076684B (en) | 2021-02-23 | 2021-02-23 | Intelligent calculation method for transient parameters in rod adjusting process of nuclear reactor core |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113076684B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113901029A (en) * | 2021-10-15 | 2022-01-07 | 中国核动力研究设计院 | Unified modeling method and system for supporting reactor core multi-professional computing software |
| CN114446496B (en) * | 2022-02-17 | 2024-04-23 | 中国核动力研究设计院 | Ultra-high flux reactor core based on annular fuel elements |
| CN114694861A (en) * | 2022-03-28 | 2022-07-01 | 哈尔滨工程大学 | Nuclear reactor core parameter simulation method and device and electronic equipment |
| CN114462336B (en) * | 2022-04-11 | 2022-06-24 | 四川大学 | Method for calculating average temperature of coolant of main pipeline of nuclear reactor |
| CN114547988B (en) * | 2022-04-25 | 2022-07-08 | 西安交通大学 | Neutron transport solving method for reactor with uniformly distributed materials |
| CN115146545A (en) * | 2022-08-01 | 2022-10-04 | 中国核动力研究设计院 | Intelligent analysis method and system for critical steady state parameters of nuclear reactor core |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103474098A (en) * | 2012-06-06 | 2013-12-25 | 中国核动力研究设计院 | Core using machinery control rod combination boron injection system as second reactor shutdown system |
| CN104102782A (en) * | 2014-07-18 | 2014-10-15 | 上海电力学院 | Method for controlling temperatures of reactors of pressurized water reactor nuclear power stations by aid of RBF (radial basis function) neural networks |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101651893B1 (en) * | 2015-04-13 | 2016-08-29 | 한전원자력연료 주식회사 | Method of synthesizing axial power distribution of reactor core using neural network and the In-Core Monitoring System(ICOMS) using the same |
| CN107301884B (en) * | 2017-07-24 | 2019-07-16 | 哈尔滨工程大学 | A kind of hybrid nuclear power station method for diagnosing faults |
-
2021
- 2021-02-23 CN CN202110203032.1A patent/CN113076684B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103474098A (en) * | 2012-06-06 | 2013-12-25 | 中国核动力研究设计院 | Core using machinery control rod combination boron injection system as second reactor shutdown system |
| CN104102782A (en) * | 2014-07-18 | 2014-10-15 | 上海电力学院 | Method for controlling temperatures of reactors of pressurized water reactor nuclear power stations by aid of RBF (radial basis function) neural networks |
Non-Patent Citations (3)
| Title |
|---|
| Deep learning-based unsupervised representation clustering methodology for automatic nuclear reactor operating transient identification;XiangLi等;《Knowledge-Based Systems》;20200927;第204卷;1-14 * |
| 小型压水堆功率神经网络预测控制研究;肖凯等;《核动力工程》;20201220;第41卷(第s2期);50-53 * |
| 结合神经网络的压水堆堆芯动态模型的研究;高明明;《中国优秀博硕士学位论文全文数据库(硕士)(工程科技Ⅱ辑)》;20130715(第7期);C040-24 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113076684A (en) | 2021-07-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113076684B (en) | Intelligent calculation method for transient parameters in rod adjusting process of nuclear reactor core | |
| CN107066745B (en) | Method for obtaining three-dimensional neutron flux density distribution in fast neutron reactor core transient process | |
| Ivanov et al. | Challenges in coupled thermal–hydraulics and neutronics simulations for LWR safety analysis | |
| US7461038B2 (en) | Method and apparatus for evaluating robustness of proposed solution to constraint problem and considering robustness in developing a constraint problem solution | |
| CN104102782B (en) | Method for controlling temperatures of reactors of pressurized water reactor nuclear power stations by aid of RBF (radial basis function) neural networks | |
| US20190392959A1 (en) | Predicting Multiple Nuclear Fuel Failures, Failure Locations and Thermal Neutron Flux 3D Distributions Using Artificial Intelligent and Machine Learning | |
| CN115146545A (en) | Intelligent analysis method and system for critical steady state parameters of nuclear reactor core | |
| CN111737869A (en) | High-precision prediction method for burnup distribution of pressurized water reactor core | |
| Kianpour et al. | Optimal design of a VVER-1000 nuclear reactor core with dual cooled annular fuel based on the reactivity temperature coefficients using thermal hydraulic and neutronic analysis by implementing the genetic algorithms | |
| Guo et al. | Challenges and progress of uncertainty analysis for the pebble-bed high-temperature gas-cooled reactor | |
| Yamamoto et al. | AEGIS: an advanced lattice physics code for light water reactor analyses | |
| Sun et al. | Control system design and validation platform development for small pressurized water reactors (SPWR) by coupling an engineering simulator and MATLAB/Simulink | |
| CN107122564A (en) | A kind of presurized water reactor pin by pin calculate few group constant parametric method | |
| Wang et al. | A feasibility study of load-follow operation without boron adjustment for CPR1000 | |
| Liu et al. | Multi-objective optimal design of the nuclear reactor pressurizer | |
| Blakely et al. | Demonstration of LOTUS multiphysics BEPU analysis framework for LB-LOCA simulations | |
| CN112632873B (en) | Method for determining position arrangement mode of measuring points of temperature thermocouple of nuclear reactor hot leg | |
| Canuti et al. | Sensitivity Studies for the Exercise I-1 of the OECD/UAM Benchmark | |
| Pelykh et al. | Theory of Fuel Life Control Methods at Nuclear Power Plants (NPP) with Water-Water Energetic Reactor (WWER) | |
| Villarino | Numerical methods used in CONDOR code | |
| Hassan et al. | Simulation of BEAVRS benchmark at hot zero power using MCNP6 | |
| McDonald | Fuel and Core Physics Considerations for a Pressure Tube Supercritical Water Cooled Reactor | |
| Mercatali et al. | Propagation of nuclear data uncertainties in PWR pin-cell burnup calculations via stochastic sampling | |
| Shahmirzaei et al. | Assessment of gadolinium concentration effects on the NuScale reactor parameters and optimizing the fuel composition via machine learning method | |
| Badrun et al. | Modeling of SPERT IV reactivity initiated transient tests in EUREKA-2/RR code |
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 |