JP7446960B2 - Prediction model building device and prediction device - Google Patents

Description

The present invention relates to a prediction model construction device and a prediction device for reactor water quality in a nuclear power plant.

As nuclear power plants (also simply referred to as plants), for example, boiling water nuclear power plants (BWR, Boiling Water Reactor) and pressurized water nuclear power plants (PWR, Pressurized Water Reactor) are known. In these plants, stainless steel, nickel-based alloys, and the like are used in the water-contact parts of major structural members such as reactor pressure vessels to suppress corrosion. In these structural materials, stress corrosion cracking (SCC) occurs and progresses when water quality, material, and stress meet certain conditions. Therefore, measures to prevent stress corrosion cracking are being applied to maintain the integrity of nuclear reactors. In addition, in recent years, preventive measures against stress corrosion cracking have been applied from the viewpoint of improving economic efficiency such as improving the capacity utilization rate of nuclear reactors and extending their service life.

As a countermeasure against stress corrosion cracking, stainless steel and nickel-based alloys designed with compositions that are highly resistant to stress corrosion cracking are used, and various stress relaxation techniques are applied to alleviate residual tensile stress generated in welds. . Examples include surface polishing, peening, and high frequency residual stress improvement methods. In addition to the stress corrosion cracking countermeasures mentioned above in terms of materials and stress, we also remove harmful ions (chloride ions, sulfate ions, chromate ions, etc.) that promote stress corrosion cracking that exist in the reactor water in small amounts. Pure water management is carried out by actively removing waste using a purification system.

When oxidizing chemical species such as oxygen and hydrogen peroxide generated by radiolysis of water during operation of a nuclear power plant are present in reactor water, the electrochemical corrosion potential (ECP) increases. At the same time as the corrosion resistance of the material decreases and the presence of residual tensile stress increases, stress corrosion cracking occurs and progresses due to the increase in corrosion potential. Therefore, hydrogen is added to the reactor water and oxidizing species are returned to water through a recombination reaction. Hydrogen injection is applied to reduce the corrosion potential. Hydrogen injection is applied to all PWR plants and to many BWR plants (see Patent Documents 1 and 2).

Corrosion potential is used as an environmental indicator of stress corrosion cracking, and it is known that the occurrence of stress corrosion cracking is suppressed when the corrosion potential is lower than a value of -300 to -200 mV vs. SHE ( See R.L.Cowan, et al., "Experience with hydrogen water chemistry in boiling water reactors," Water chemistry of nuclear reactor systems 4, 1, P29, BNES (1986)).

Furthermore, in recent years, a technology called noble metal injection has been introduced in BWR, in which platinum group noble metals are added to the reactor water and deposited on the surfaces of stainless steel and nickel-based alloys, giving the platinum group noble metals electrochemical catalytic properties for the reaction between oxygen and hydrogen. (see Patent Document 3) is mainly applied to BWRs in the United States. By applying noble metal injection, the amount of hydrogen required to reduce the corrosion potential below the target value can be significantly reduced, and the range in which the corrosion potential can be reduced can also be expanded compared to the case where it is not applied.

In these conventional techniques, it is necessary to accurately know the corrosion potential of structural materials in order to evaluate the effects of stress corrosion cracking. Therefore, in conventional plants, corrosion potential sensors are installed inside the reactor pressure vessel or on piping connected to the reactor pressure vessel to measure the corrosion potential of structural materials.
Life prediction of nuclear reactor primary system structures that predicts crack propagation due to stress corrosion cracking and corrosion fatigue is known (see Patent Document 4). This life prediction is performed using the measured corrosion potential and crack growth characteristic data of the structural material. Furthermore, another method for estimating the pre-life of structural members of a nuclear power plant is known (see Patent Document 5).

The corrosion potential of structural materials is measured by installed sensors, but the corrosion potential at locations other than the sensor installation location can be calculated. The method for calculating corrosion potential will be explained below.
Factors that influence corrosion potential include dissolved oxygen, hydrogen peroxide, and dissolved hydrogen concentrations, as well as flow rate and hydraulic equivalent diameter, which are responsible for the mass transfer rate of these species to the surface of the structural material. The flow velocity and hydraulic equivalent diameter are determined by the core flow rate, the flow path area at each location, and the wetted edge length. Dissolved oxygen, dissolved hydrogen, and hydrogen peroxide concentrations are produced by the radiolysis reaction of water and are therefore affected by the dose rate of gamma rays and neutrons emitted from the reactor core. The dose rate and dose rate distribution within these pressure vessels are influenced by the operating state of the reactor core, ie, the power output of the fuel, and the power distribution in the axial and radial directions.

When hydrogen injection is applied, the concentration of dissolved oxygen, hydrogen peroxide, and dissolved hydrogen in the furnace decreases due to the increase in the hydrogen concentration of the feed water, and the distribution changes, so the distribution of the corrosion potential also changes. Furthermore, when noble metal implantation is applied, the amount of precious metal deposited that contributes to electrochemical reactions on the surface of the structural material and the amount of hydrogen that is excessive in the stoichiometric ratio relative to the amount of oxygen affect the corrosion potential. This is an important factor.

In precious metal injection, precious metals (mostly platinum is used) are injected into the reactor water during shutdown or during periods of power operation. At this time, precious metals adhere to the surfaces of structural materials and fuel rods that come in contact with reactor water. When the injection of precious metals is stopped, a portion of the precious metals attached to the surfaces of the structural materials and fuel rods are melted or peeled off and transferred from the respective surfaces to the reactor water. The precious metals transferred to the reactor water will re-adhere to the surfaces of structural materials and fuel rods depending on their concentration, but some of them will be removed by the purification system.

In order to determine the corrosion potential of the surface of structural materials at any location, it is necessary to calculate the dissolved oxygen concentration and hydrogen peroxide concentration evaluated by radiolysis calculations from the hydrogen injection concentration in the feed water and the dose rate at each site determined by the operating status of the reactor core. First, determine the concentration and dissolved hydrogen concentration. Next, using these as input, the corrosion potential is determined by calculating the flow velocity and electrochemical reaction at each location (see Patent Document 1). Furthermore, under the noble metal injection conditions, the electrochemical reaction of hydrogen is catalyzed by the noble metal attached to the structural material, so the amount of attached noble metal is incorporated into the electrochemical reaction calculation to calculate the corrosion potential.

In order to suppress stress corrosion cracking, it is necessary to maintain the corrosion potential lower than a value of -300 to -200 mV vs. SHE, so it is necessary to keep the corrosion potential of structural materials in the operating plant below the target value. It is necessary to confirm by actual measurement or calculation that the value has decreased. The calculation requires the core power plan, the planned feedwater hydrogen injection amount, and the water quality around the structural materials of each part to be evaluated, that is, the concentrations of oxygen, hydrogen peroxide, and hydrogen. In addition, temperature and mass transfer coefficient are required, and when noble metal implantation is applied, the amount of noble metal deposited on the surface of the structural material is also required.

Among these, the core output plan and the planned values for the feedwater hydrogen injection amount are values that are available in chronological order, but the water quality and precious metal deposits in the areas targeted for stress corrosion cracking protection in the reactor are available in chronological order. There are no actual measured values. Therefore, it is necessary to obtain it by calculation, and it is determined by a physical and chemical simulation model (see Patent Document 6). In Patent Document 6, a simulation model (mass balance model) that estimates radioactivity changes in cooling water using current water quality conditions as input is used to predict future plant dose rates, and based on this prediction result, current water quality conditions are estimated. The amount of radioactivity deposited on structural materials is predicted by diagnosing the quality of the materials. A similar method can be used for precious metals to predict the amount of structural material deposited on the precious metal.

The self-learning diagnosis and prediction device described in Patent Document 7 automatically repairs model parameters in response to temporal changes in plant specifications, characteristics, etc., prevents deterioration of prediction accuracy, and improves the model through self-learning. It has functions.

Patent No. 2687780 Japanese Patent Application Publication No. 2005-043051 Japanese Patent Application Publication No. 04-223299 Japanese Patent Application Publication No. 2006-010428 Japanese Patent Application Publication No. 06-034786 Japanese Patent Application Publication No. 01-063894 Japanese Patent Application Publication No. 06-289179

The technique described in Patent Document 6 optimizes model parameters (model parameters) set based on a physical model or a chemical model and adjusts their contributions, and requires that the model be prepared in advance. Therefore, the correlation between the state quantity of the prediction target and the state quantity used as an input must be described as a mathematical expression.

The correlation between the state quantities expressed by the model is optimized by optimizing the model parameters. However, although correlations are possible, the model parameters may not be technically or economically available directly during plant operation. For example, the water quality or dose rate inside a pressure vessel cannot be measured, or even if it can be measured, the locations are limited. In addition, physical property values such as chemical reaction rate constants used in calculations may not be sufficiently prepared, or there may be parameters that are difficult to express mathematically, such as the characteristics of oxide films on the surface of structural materials, or data may be missing over time. or

In this way, if the correlation data is temporally and spatially discrete, or if the correlation is complex and difficult to express mathematically, an appropriate model cannot be constructed. For this reason, even if there are complex correlations that cannot be expressed mathematically, there is a need to accurately predict reactor water quality and corrosion potential in nuclear plants using available data and physical models, and to manage operations. There is.

The present invention has been made in view of such a background, and an object of the present invention is to provide a predictive model construction device and a prediction device that construct a highly accurate predictive model of reactor water quality in a nuclear power plant.

In order to solve the above-described problems, a predictive model building device is a predictive model building device that builds a predictive model that predicts the reactor water quality of a nuclear reactor in a nuclear power plant. a simulation unit that calculates in-reactor predicted values including predicted values of reactor water quality in the reactor using a physical model from input data including predicted values of rate distribution; a plant state quantity including at least one of the following: flow rate, main steam flow rate, core inlet temperature, core outlet temperature, feedwater hydrogen concentration, feedwater iron concentration, precious metal injection amount, reactor water purification flow rate, thermal output; and the reactor. a learning unit that constructs a predictive model by causing a machine learning model to learn training data that includes an in-furnace predicted value as input data and includes as output data the in-furnace predicted value after a predetermined period from the period in which the input data was acquired; Equipped with.

The prediction device is a prediction device for predicting reactor water quality of a nuclear reactor in a nuclear power plant, and uses a physical model from input data including predicted values of dose rate distribution of gamma rays and neutron rays in the reactor. a simulation unit that calculates predicted values in the reactor including predicted values of reactor water quality in the reactor; linear flow velocity of the reactor, feed water flow rate, core flow rate, main steam flow rate, core inlet temperature, core outlet temperature, and feed water; Input data includes a plant state quantity including at least one of hydrogen concentration, feedwater iron concentration, precious metal injection amount, reactor water purification flow rate, and thermal output, and the predicted value in the reactor, and the input data is acquired. a storage unit that stores a prediction model constructed by making a machine learning model learn training data that includes the in-furnace predicted value after a predetermined period from the period when the in-furnace predicted value is output; and the plant state quantity and the in-furnace predicted value. and a prediction unit that calculates an in-furnace predicted value after the predetermined period from a period in which the plant state quantity and the in-furnace predicted value are acquired using the prediction model.

According to the present invention, it is possible to provide a predictive model construction device and a prediction device that construct a highly accurate predictive model of reactor water quality in a nuclear power plant.

1 is a schematic configuration diagram of a nuclear power plant according to the present embodiment. To explain the input/output relationship (radiolysis/corrosion potential model) of calculations regarding the production of oxygen, hydrogen peroxide, and hydrogen by radiolysis of reactor water in the reactor and the corrosion potential of structural materials of the reactor according to the present embodiment. This is a diagram. FIG. 2 is a functional configuration diagram of a prediction device for predicting reactor water quality and corrosion potential according to the present embodiment. FIG. 3 is a diagram for explaining input data and output data of a predictive model included in teacher data according to the present embodiment. FIG. 3 is a diagram for explaining the configuration of training data of a prediction model in learning processing according to the present embodiment. FIG. 3 is a diagram for explaining input data and output data (prediction results) of the dose rate prediction model according to the present embodiment. FIG. 3 is a diagram for explaining the configuration of training data of the dose rate prediction model according to the present embodiment. It is a flowchart of the learning process which the learning part concerning this embodiment performs. FIG. 3 is a diagram for explaining input data and output data of a prediction model in prediction processing according to the present embodiment. It is a flowchart of the prediction process which the prediction part concerning this embodiment performs. FIG. 7 is a diagram for explaining the configuration of training data of a predictive model in learning processing according to a modification of the present embodiment. It is a figure for explaining input data and output data (prediction result) of a prediction model in prediction processing concerning a modification of this embodiment. FIG. 3 is a diagram for explaining a radiolysis/corrosion potential model that predicts reactor water quality and corrosion potential in an embodiment different from the embodiment described above. It is a figure for explaining the dose rate prediction device concerning the modification of this embodiment. It is a functional block diagram of the driving assistance device concerning the modification of this embodiment. It is a flowchart of driving support processing concerning a modification of this embodiment.

Before explaining the method for predicting the reactor water quality of a nuclear reactor and the corrosion potential of the structural material of the reactor in the mode for carrying out the present invention (embodiment), we will explain the nuclear power plant to be predicted and the nuclear power plant concerned. A physical model (radiolysis/corrosion potential model) for calculating reactor water quality and corrosion potential in a plant will be explained.

≪Overview of nuclear power plant≫
FIG. 1 is a schematic configuration diagram of a nuclear power plant P100 according to this embodiment. Referring to FIG. 1, a schematic configuration of a nuclear power plant P100 (BWR here) to which a device for predicting reactor water quality and corrosion potential of structural materials is applied will be described.
The nuclear power plant P100 includes a nuclear reactor P1, a turbine P3, a condenser P4, a reactor purification system, a water supply system, and the like. The nuclear reactor P1 installed in the reactor containment vessel P11 has a reactor pressure vessel P12 containing a so-called reactor core P13 made up of fuel assemblies (not shown). A cylindrical core shroud P15 is installed inside the reactor pressure vessel P12 and holds the reactor core P13. Each fuel assembly is composed of a plurality of fuel rods in which fuel cladding tubes are filled with a plurality of fuel pellets made of nuclear fuel material.

An annular downcomer P17 is formed between the inner surface of the reactor pressure vessel P12 and the outer surface of the reactor core shroud P15. In the case of an internal pump type ABWR (improved BWR), a plurality of internal pumps P21 are installed at the bottom of the reactor pressure vessel P12 to circulate reactor water. The impeller of the internal pump P21 is arranged at the lower part of the downcomer P17. Although not shown, a non-internal pump type BWR has a recirculation system outside the reactor pressure vessel P12, and a recirculation pump installed in the recirculation system circulates reactor water.

The water supply and condensate system includes a water supply pipe P10 that connects the condenser P4 and the reactor pressure vessel P12, a condensate pump P5, a condensate purification device P6, a water supply pump P7, a low-pressure feedwater heater P8, and a high-pressure feedwater heater. P9 are installed in this order from the condenser P4 toward the reactor pressure vessel P12. The hydrogen injection device P16 is connected to the water supply pipe P10 by a hydrogen injection pipe P18 on the suction side of a condensate boost pump (not shown) located between the condensate purification device P6 and the water supply pump P7. An on-off valve P19 is provided in the hydrogen injection pipe P18. In a plant that performs noble metal injection during operation, a noble metal injection device P31 is connected to a water supply pipe P10 at a position close to the reactor pressure vessel P12 by a noble metal injection pipe P32. An on-off valve P33 is provided in the noble metal injection pipe P32.

The reactor purification system includes a purification system isolation valve P23, a purification system piping P20 (selected from carbon steel, stainless steel, and low alloy steel depending on the application) that connects the reactor pressure vessel P12 and the water supply piping P10. A purification system pump P24, a regenerative heat exchanger P25, a non-regenerative heat exchanger P26, and a reactor water purification device P27 consisting of a filter and an ion exchange resin are installed and configured in this order. The residual heat removal system installed in the nuclear power plant P100 is connected at one end to the reactor pressure vessel P12, and in the case of ABWR, the other end is connected to the reactor pressure vessel P12 above the reactor core P13 for purification. System piping P20 is branched. A residual heat removal system pump P29 and a heat exchanger P30 (cooling device) are installed in the residual heat removal system piping P28 of the residual heat removal system.

The reactor water in the reactor pressure vessel P12 flows down the downcomer P17, is pressurized by the internal pump P21, and flows out to the lower plenum located at the lower part of the reactor pressure vessel P12. Thereafter, the reactor water ascends through the lower plenum and flows into the reactor core P13. In the reactor core P13, reactor water is heated by heat generated by fission of the nuclear fuel material contained in the fuel rods of the fuel assembly. Some of the heated reactor water turns into steam. After flowing up to the upper part of the reactor pressure vessel P12, this steam passes through a separator (not shown) and a dryer (not shown), and then is led to the turbine P3 through the main steam pipe P2. Rotate. A generator (not shown) connected to turbine P3 is rotated to generate electric power. Steam discharged from turbine P3 is condensed in condenser P4 and returned to liquid water.

This water is circulated and supplied as condensate/supply water into the reactor pressure vessel P12 through the water supply pipe P10. The water supply flowing through the water supply pipe P10 (water supply condensate pipe) is pressurized by a condensate pump P5, impurities are removed by a condensate purification device P6 consisting of a filter and ion exchange resin, and the pressure is further increased by a water supply pump P7. It is heated by a low pressure feed water heater P8 and a high pressure feed water heater P9. The extracted steam extracted from the main steam pipe P2 and the turbine P3 through the extraction pipe P14 is supplied to the low pressure feed water heater P8 and the high pressure feed water heater P9, respectively, and serves as a heating source for the feed water. In the ABWR, the drain from the high-pressure feed water heater P9 directly enters the water supply, increasing thermal efficiency. The drain from the low pressure feedwater heater P8 is recovered upstream or downstream of the condensate purification system P6.

Since the reactor water in the reactor pressure vessel P12 contains metal corrosion products contained in the feed water and products generated by corrosion of the structural materials in the reactor pressure vessel P12, a certain proportion of the reactor water is It is purified by the purification system. The reactor water in the reactor pressure vessel P12 is supplied to the regenerative heat exchanger P25 and the non-regenerative heat exchanger P26 through the purification system piping P20 branched from the residual heat removal system piping P28 by driving the purification system pump P24. It is cooled down to about 50°C by a heat exchanger. The cooled reactor water passes through the reactor water purification device P27 to remove metal corrosion products contained in the reactor water, is heated in the regenerative heat exchanger P25, and then merges with the feed water flowing in the water supply pipe P10. and is supplied to the reactor pressure vessel P12.

When shutting down the reactor P1, the recirculation flow rate is lowered and all control rods (not shown) are inserted into the reactor core P13. Insertion of all control rods stops the fission reaction of nuclear fuel material. The heat remaining in the equipment in the reactor core P13 and reactor pressure vessel P12 is removed by evaporation of the reactor water, but if the temperature drops to a certain extent, the heat removal efficiency due to evaporation of the reactor water decreases, so the reactor water temperature is reduced to 150°C. When the residual heat removal system is used to cool down the equipment in the reactor core P13 and reactor pressure vessel P12. That is, by driving the residual heat removal system pump P29, the reactor water in the reactor pressure vessel P12 is supplied to the heat exchanger P30 through the residual heat removal system piping P28, and the reactor water is cooled by the heat exchanger P30. It is returned to the reactor pressure vessel P12.

Hydrogen injection or a combination of hydrogen injection and noble metal injection is carried out from the perspective of suppressing stress corrosion cracking in reactor structural materials. Hydrogen is injected into the water supply from the hydrogen injection device P16 through the hydrogen injection pipe P18 by opening the on-off valve P19. The hydrogen injected into the feed water is ejected into the reactor water from a feed water sparger (not shown) above the downcomer P17 and mixed with the reactor water. The dose rate of gamma rays in the downcomer P17 region is moderate and promotes the recombination reaction, so when the added hydrogen passes through the downcomer P17, it recombines with oxygen and hydrogen peroxide in the reactor water to produce water. That is, the corrosion environment is alleviated by consuming dissolved oxygen and hydrogen added to hydrogen peroxide, which are environmental factors for stress corrosion cracking. As a result, the corrosion potential of structural materials that come into contact with reactor water is reduced, and the development and progression of stress corrosion cracking is suppressed.

In addition, noble metal injection is a technology that promotes the effects of hydrogen injection through catalytic action. For example, an aqueous solution of a platinum (Pt) compound is injected from the noble metal injection device P31 into the water supply through the noble metal injection pipe P32 with the on-off valve P33 opened. The platinum injected into the feed water merges with the reactor water, and a portion of the platinum, in the form of minute platinum particles, adheres to the surfaces of structural materials that come into contact with the reactor water. The rest adheres to the fuel surface. On the surface of the platinum particles, oxygen and hydrogen peroxide in the reactor water catalytically react with hydrogen to form water. When hydrogen is present in a stoichiometric ratio ( _H2O is H:O=2:1) of 2 or more than oxygen (hydrogen peroxide is treated as an equivalent 1/2 amount of oxygen), hydrogen is There is a surplus, and the reaction between hydrogen and oxygen proceeds at a potential near the hydrogen oxidation-reduction potential of -500mV vs. SHE. For this reason, the corrosion potential of the structural materials is mixed and falls to around -500mV vs. SHE, which becomes below the target potential. This can suppress stress corrosion cracking.

A corrosion potential sensor is temporarily or long-term installed in the reactor P1 as a means of confirming the effect of hydrogen injection or noble metal injection on suppressing stress corrosion cracking. The bottom drain line P34 draws out reactor water from the bottom of the reactor pressure vessel P12 and is connected to the purification system piping P20. A branch line is provided from this bottom drain line P34, and a corrosion potential sensor P35 is installed there to measure the corrosion potential of the water at the bottom of the reactor pressure vessel P12. The branch line is connected to a reactor water sampling line (not shown) or the like. Furthermore, in the case of ABWR, the purification system piping P20 can be used to measure the corrosion potential of the water in the area above the downcomer P17. For example, the corrosion potential can be measured by installing a flange P36 on the same pipe and loading a corrosion potential sensor P37. Alternatively, a branch line can be provided to perform measurements similar to bottom drain. In the case of a BWR, reactor water from the lower part of the downcomer P17 flows through recirculation system piping (not shown) and purification system piping P20. Corrosion potential is measured by installing a branch line or flange loaded with a corrosion potential sensor, similar to ABWR.

Also, in some cases, the neutron instrumentation tube P38 may be modified to install a corrosion potential sensor P39 in the lower plenum of the reactor pressure vessel P12 or the reactor core P13, and a corrosion potential sensor P39 in the lower plenum and a corrosion potential sensor P40 in the core P13. It is done at the plant. Specifically, corrosion potential sensors P39 and P40 are installed inside the neutron instrumentation tube P38, and the corrosion potential is measured by drilling a hole in the neutron instrumentation tube P38 so that the reactor water in the target area flows into it. will be held. However, the locations where the corrosion potential sensor can measure are limited as described above, and there are few plants that constantly measure the corrosion potential. The measurement sites for the reactor water quality are further limited, and usually only the sampling line of the reactor water purification system is measured, and the bottom drain P34 is measured in accordance with the installation of the corrosion potential sensor.

According to the prediction device according to this embodiment, the water quality distribution in the nuclear reactor and the corrosion potential distribution of structural materials can be predicted with high accuracy. By using these predicted reactor water quality and corrosion potential, plant operators can control the amount of hydrogen injection to maintain the corrosion potential of reactor structural materials during operation below the target value, and control the injection of precious metals. Students will be able to formulate and implement operational plans such as starting operations.

≪Overview of the calculation model for radiolysis and corrosion potential≫
FIG. 2 shows the relationship between input and output of calculations regarding the generation of oxygen, hydrogen peroxide, and hydrogen by radiolysis of reactor water in the reactor and the corrosion potential of structural materials of the reactor (radiolysis/corrosion) according to this embodiment. FIG. 3 is a diagram for explaining a potential model. The radiolysis/corrosion potential model uses plant design data M1 (equipment dimensions, operating temperature, etc.), the ratio of core thermal output to rating determined by core management M2, core flow rate, and gamma ray and neutron ray doses in the reactor pressure vessel. A model regarding the radiolysis of water (radiolysis model) is calculated using the rate distribution M4 and the feed water quality M3 (that is, the feed water hydrogen concentration at the time of hydrogen injection).

As inputs for solving the chemical reactions of the radiolysis model, in particular, the chemical reaction rate constants for each of the 30 or more elementary reactions involved in the reaction mechanism of water radiolysis M5, and the G of each chemical species constituting the elementary reactions. Value matters. This reduces the concentration M6 of stable molecular products hydrogen, oxygen, hydrogen peroxide, which are water decomposition products, and the concentration M6 of various reactive radical products OH, H, e- ^, etc. The distribution is calculated. Among these, the corrosion potential M9 of the structural material is calculated by a corrosion potential model using the concentrations M6 of oxygen, hydrogen peroxide, and hydrogen as input.

When calculating the corrosion potential M9, in addition to the water quality (concentration M6 of hydrogen, oxygen, and hydrogen peroxide), the mass transfer coefficient M8 (mass transfer rate coefficient ) is used. Furthermore, in the case of a complicated channel shape, the mass transfer coefficient M8 on the surface of the structural material may be determined using a flow analysis code without using the hydraulic equivalent diameter.

Further, when noble metal implantation is applied, the amount M7 of noble metal adhesion (indicated as Pt adhesion amount in FIG. 2) at each location is required as input. The noble metal deposition amount M7 can be determined using, for example, a physical and chemical simulation model (see Patent Document 6). The corrosion potential model requires an electrochemical reaction on the surface of the structural material, reaction parameters for each electrochemical reaction (reaction rate constant, permeability coefficient), and a diffusion coefficient for determining the mass transfer coefficient M8.

Models for radiolysis and corrosion potential are described, for example, in the following paper: "Hydrazine and Hydrogen Co-injection to Mitigate Stress Corrosion Cracking of Structural Materials in Boiling Water Reactors (IV)," Y. Wada et al., Journal of Nuclear Science and Technology, 44, 4, P607-622 (2007).

Among the above input/outputs, the quantities that cannot be measured during plant operation include the gamma ray and neutron dose rate distribution M4 in the pressure vessel, the water quality distribution in the pressure vessel, and in the case of BWR, the water quality distribution in the recirculation system (hydrogen , oxygen, and hydrogen peroxide concentrations M6), and corrosion potential M9 at a location where a sensor cannot be installed. The dose rate distribution M4 is calculated by a shielding calculation code as a distribution of gamma rays and neutrons when the radiation source intensity is placed in the reactor core, taking into account a specific burnup related to the fuel in the reactor core. However, calculation of the dose rate distribution M4 according to the power distribution of the reactor core, which changes from moment to moment over the plant operation cycle, has not been carried out in the past, and response confirmation tests for typical early or late cycles or hydrogen injection have not been carried out. Calculations were made only for specific periods, such as the period when plants were used, and only for representative plants. Here, one operation cycle of a plant is a start-up operation typified by the withdrawal of control rods after closing the lid of the reactor pressure vessel of the plant, an output operation that generates electrical output, and a control rod insertion to reactor water temperature. Refers to the total period of operation consisting of stop operations, which are typified by the lowering of pressure and the opening of the pressure vessel. Periods outside of the operating cycle are periodic inspection periods during which inspections, construction work, and fuel changes are performed. Therefore, current radiolysis/corrosion potential models only perform calculations using the typical dose rate distribution M4.

≪Configuration of prediction device≫
A prediction device for predicting the reactor water quality in the reactor and the corrosion potential of structural materials in the reactor in the nuclear power plant P100 according to the present embodiment will be described below. The prediction device includes radial power distribution based on the axial direction of the core and power distribution for each nuclear fuel assembly, gamma ray and neutron dose rate distribution based on shielding calculation results, actual water quality measurement data, actual corrosion potential measurement data, radiation monitoring data, etc. The reactor water quality and corrosion potential after a predetermined period are predicted (output) from the plant state quantities (input data) acquired up to the time of prediction. For prediction, the above-mentioned radiolysis model and corrosion potential model (referred to as radiolysis/corrosion potential model) and machine learning technology are used. Specifically, the results of a simulation using a radiolysis/corrosion potential model will be used as part of the input to the machine learning model.

FIG. 3 is a functional configuration diagram of a prediction device 100 for predicting reactor water quality and corrosion potential according to this embodiment. The prediction device 100 is a computer and includes a control section 110, a storage section 120, and an input/output section 160. The prediction device 100 receives flow data such as thermal output, power distribution in the axial and radial directions of the core, core flow rate, feed water flow rate, main steam flow rate, reactor water purification system flow rate, etc. from the process computer P110 used in the nuclear power plant P100. Operational data such as radiation monitoring data for the steam system, off-gas system, and each building, and output data from the measuring instrument P120 installed in the nuclear power plant P100 are received.

The storage unit 120 includes a RAM (Random Access Memory), a ROM (Read Only Memory), an SSD (Solid State Drive), and the like. The storage unit 120 stores a prediction model 130, a radiolysis/corrosion potential model 140, a dose rate prediction model 145, and a plant state quantity database 150.
Prediction model 130 is a machine learning model, such as a neural network. The prediction model 130 is a machine learning model that uses plant state quantities as input data and uses as prediction results (output data) reactor water quality and corrosion potential after a predetermined period from the period in which the plant state quantities were acquired.

The radiolysis/corrosion potential model 140 (also referred to as the radiolysis/ECP model in the drawings) is a simulation model of reactor water quality and corrosion potential, and includes equations and parameters that describe the model. Alternatively, the radiolysis/corrosion potential model 140 may be regarded as a program that executes a simulation.
The dose rate prediction model 145 is a machine learning model, such as a neural network. The dose rate prediction model 145 is a machine learning model that predicts the dose rate distribution that is input to the radiolysis/corrosion potential model 140.

The prediction model 130, the radiolysis/corrosion potential model 140, and the dose rate prediction model 145 are machine learning or simulation models, but they may also be regarded as programs that calculate outputs based on inputs, or as calculation entities. For example, it may be written that the prediction model 130 predicts reactor water quality and corrosion potential from plant state quantities.

The plant state quantity database 150 includes plant state quantities including plant data, water supply data, reactor water quality, radioactivity data, and radiation monitoring data of the nuclear power plant P100, which are values of variables and parameters included in the radiolysis/corrosion potential model 140. Remember. The plant state quantity database 150 also stores the types of fuel assemblies loaded in the reactor core P13 (see FIG. 1), burnup, length of stay in the reactor, and arrangement in the core, as well as thermal output and power output. Plant state quantities such as distribution are also stored.

The input/output unit 160 receives data from the process computer P110 and the measuring instrument P120 and stores it in the plant state quantity database 150. Further, the input/output unit 160 includes a display, a keyboard, and a mouse (not shown), and receives operations from the user of the prediction device 100 and displays data such as prediction results.

The control unit 110 includes a CPU (Central Processing Unit), and includes a learning unit 111, a prediction unit 112, and a simulation unit 113. The learning unit 111 performs a learning process (see FIG. 8 described later) using data stored in the plant state quantity database 150 as teacher data (learning data), and creates a prediction model that predicts reactor water quality and corrosion potential after a predetermined period. 130 is generated. Further, the learning unit 111 executes learning processing to generate a dose rate prediction model 145.

The prediction unit 112 inputs data stored in the plant state quantity database 150 into the generated prediction model 130, and performs a prediction process (see the figure described later) to predict the reactor water quality and the corrosion potential of structural materials after a predetermined period. 10)). Further, the prediction unit 112 executes a prediction process of predicting the dose rate using the dose rate prediction model 145.
The simulation unit 113 executes a simulation using the radiolysis/corrosion potential model 140. The simulation execution result becomes input data for the prediction model 130. Details of the processing by the learning unit 111 and the prediction unit 112 will be explained with reference to FIGS. 4 to 10, which will be described later.

≪Prediction model and training data≫
FIG. 4 is a diagram for explaining input data and output data of the prediction model 130 included in the teacher data according to this embodiment. Here, a prediction model 130 that predicts the next day's reactor water quality and corrosion potential from N days' worth of plant state quantities 151 and dose rate distribution 152 will be described. That is, here, after the predetermined period means the next day.
Specifically, using the radiolysis/corrosion potential model 140, the water quality/corrosion potential 154 for N days is determined from the dose rate distribution 152 (see dose rate distribution M4 shown in FIG. 2). The input data of the prediction model 130 is the plant state quantity 151 for N days and the water quality/corrosion potential 154 for N days. The output data (prediction result, correct label) of the prediction model 130 is the next day's reactor water quality and corrosion potential. When learning, using the radiolysis/corrosion potential model 140, the next day's water quality/corrosion potential 154A is obtained from the next day's dose rate distribution 152 and used as output data.

FIG. 5 is a diagram for explaining the structure of the teacher data 161 to 169 of the prediction model 130 in the learning process according to the present embodiment. Here, N teacher data 161 to 169 are generated from the plant state quantities from the first day to the second N days.
The input data of the teacher data 161 includes the plant state quantity 151 and the water quality/corrosion potential 154 for N days from the first day to the Nth day. The output data (correct label) of the teacher data 161 includes the water quality/corrosion potential 154A on the N+1st day.

In detail, the feed water flow rate, core flow rate, main steam flow rate, recirculation flow rate (in the case of a BWR with a recirculation system), and core inlet of reactor water included in each of the plant state quantities 151 from the 1st day to the Nth day. Temperature, core exit temperature of reactor water, linear flow velocity in each region within the pressure vessel, amount of precious metals injected, reactor water metal concentration (Fe, Ni, Cr, Cu, Co), reactor water purification flow rate, thermal output, core fuel assembly Axial power distribution and radial power distribution based on each output, nuclear fuel assembly output, radiation monitoring data (N16, N13, off-gas nuclides), feed water hydrogen concentration, main steam system oxygen and hydrogen concentration, reactor water sampling point radioactivity concentration (Co60, Co58, Cr51, Na24, Au199), reactor water impurity concentration (ions such as chloride ions, sulfate ions, chromate ions, etc., insoluble components), dissolved oxygen concentration, dissolved hydrogen concentration, hydrogen peroxide concentration Plant state quantities 151 that can be actually measured, such as concentration and corrosion potential, or that are already stored in the process computer P110 (see FIG. 3), are included as input data for the prediction model 130.

Also included in the plant state quantities from day 1 to day N are feedwater hydrogen concentration, gamma ray and neutron beam dose rate distribution 152, core inlet temperature of reactor water, core outlet temperature of reactor water, core flow rate, main steam The concentration of oxygen, hydrogen peroxide, and hydrogen in the reactor (reactor water quality) and the corrosion potential of structural materials in the reactor are calculated using the radiolysis/corrosion potential model 140 from the system flow rate and linear flow velocity, and the prediction model 130 is included as input data. Teacher data 161 serving as input data for this prediction model 130 corresponds to water quality/corrosion potential 154 shown in FIG. Note that the water quality in the reactor (reactor water quality) and corrosion potential, which are input data to the prediction model 130, are also referred to as in-reactor predicted values.

The training data 161 includes the concentrations of oxygen, hydrogen peroxide, and hydrogen in the reactor calculated using the radiolysis/corrosion potential model 140 from the plant state quantities on the N+1 day in the same manner as the water quality/corrosion potential 154. The corrosion potential is included as output data (correct label, prediction result) of the prediction model 130. The teacher data 161 that is the output data of this prediction model 130 is the water quality/corrosion potential 154A shown in FIG.

Similar to the teacher data 161, the input data of the teacher data 162 includes plant state quantities 151 from the second day to the N+1 day, and water quality/corrosion potential 154. The output data (correct label) of the teacher data 162 includes the water quality/corrosion potential 154A on the N+2th day.
Similar to the teacher data 161 and 162, the input data of the teacher data 169 includes the plant state quantity 151 from the Nth day to the 2N-1th day, and the water quality/corrosion potential 154. The output data (correct label) of the teacher data 169 includes the water quality/corrosion potential 154A on the 2Nth day.

The predictive model 130 can be generated by learning using the teacher data described above. The prediction model 130 calculates the next day's reactor water quality (the concentration of oxygen, hydrogen peroxide, and hydrogen in the reactor) and corrosion potential from the plant state quantities 151 and dose rate distribution 152 for N days up to prediction. It is a predictive model.

≪Dose rate prediction model and training data≫
The gamma ray and neutron dose rate distribution 152 (see FIG. 4), which is input to the radiolysis/corrosion potential model 140, is calculated by shielding calculation. However, calculating the moment-by-moment dose rate distribution over one operating cycle of a plant is not possible with current computer power due to time and economic considerations. In this embodiment, the dose rate distribution is calculated using the dose rate prediction model 145.
FIG. 6 is a diagram for explaining input data and output data (prediction results) of the dose rate prediction model 145 according to this embodiment. The input data includes plant design data 156 (equipment dimensions, operating temperature, etc., see design data M1 shown in Figure 2), and core power distribution 153 (axial power distribution and diameter obtained from the power distribution for each fuel assembly). direction output distribution). The prediction result, which is the output data of the dose rate prediction model 145, is a dose rate distribution 152.

FIG. 7 is a diagram for explaining the configuration of training data of the dose rate prediction model 145 according to this embodiment. The teacher data is constructed from design data 156 of one or more plant groups, core power distribution 153, and dose rate distribution 152.
Specifically, the calculation result of the dose rate distribution 152 obtained by the shielding calculation of the plant A, the core power distribution 153 used when calculating the dose rate distribution, and the design data 156 of the plant A are used as the teacher data. In FIG. 7, the dose rate distribution 152 is calculated from the design data 156 and the core power distribution 153 at each period from period 1 to period L, and is used as one teacher data, and a total of L pieces of teacher data are constructed. . Teacher data is constructed in the same manner for other plants B, . . . , and plant X. The dose rate prediction model 145 is created by learning these teacher data (the learning unit 111 trains the dose rate prediction model 145 using the teacher data).

The dose rate prediction model 145 created in this way calculates (predicts) the dose rate distribution 152 of gamma rays and neutrons in the reactor based on the design data 156 of the plant to be evaluated and the core power distribution 153. . Specifically, the prediction unit 112 uses the dose rate prediction model 145 to calculate the dose rate distribution 152 from the design data 156 and the core power distribution 153.
The calculated dose rate distribution 152 becomes input data to the radiolysis/corrosion potential model 140 (see FIG. 4). The dose rate distribution 152 is input to the prediction model 130 via the radiolysis/corrosion potential model 140. Further, the design data 156 and the core power distribution 153 are input to the prediction model 130 via the dose rate prediction model 145 and the radiolysis/corrosion potential model 140.

The calculation of the dose rate distribution 152 using the dose rate prediction model 145 has been described above. Core power distribution 153 is data acquired from process computer P110 (see FIG. 3), and is stored in plant state quantity database 150. The prediction device 100 may calculate the dose rate distribution 152 at the time of acquiring the core power distribution 153 and store it in the plant state quantity database 150. By doing so, the dose rate distribution 152 can also be regarded as the plant state quantity 151.

≪Learning process that generates a predictive model≫
FIG. 8 is a flowchart of the learning process executed by the learning unit 111 according to the present embodiment.
In step S11, the learning unit 111 executes steps S12 to S13 for each start date (first day to Nth day, see FIG. 5).

In step S12, upon receiving the instruction from the learning unit 111, the simulation unit 113 uses the plant state quantity 151 as input data and executes a simulation using the radiolysis/corrosion potential model 140. In detail, the simulation unit 113 calculates the dose rate distribution, linear flow velocity, feed water flow rate, core flow rate, main steam flow rate, and reactor core contained in the plant state quantities 151 for N+1 days from the start date (N+1 days including the start date). Using the radiolysis/corrosion potential model 140 from the inlet temperature, core exit temperature, feed water hydrogen concentration, feed water iron concentration, precious metal injection amount, reactor water purification flow rate, thermal output, etc., the reactor water quality (in the reactor Calculate the concentration of oxygen, hydrogen peroxide, and hydrogen) and the corrosion potential.

In step S13, the learning unit 111 generates teacher data. In detail, the learning unit 111 learns the dose rate distribution, linear flow velocity, feed water flow rate, core flow rate, main steam flow rate, which are included in each of the plant state quantities 151 for N days from the start date (N days including the start date). Plant state quantities such as core inlet temperature, core outlet temperature, feed water hydrogen concentration, feed water iron concentration, precious metal injection amount, reactor water purification flow rate, thermal output, reactor water quality and corrosion for N days from the start date calculated in step S12 Teacher data is generated in which the potential is input data and the reactor water quality and corrosion potential N days after the start date (start date + N days) are output data (correct data).

In step S14, if the learning unit 111 has executed steps S12 to S13 for every start date, the process proceeds to step S15, and if there is an unprocessed start date, the process returns to step S12, and for the unprocessed start date, the learning unit 111 proceeds to step S15. ~Process S13.
In step S15, the learning unit 111 trains the predictive model 130 using the teacher data generated in step S13 (makes the predictive model 130 learn the teacher data) to construct the predictive model 130.

Through the above-described prediction processing, it is possible to construct a prediction model 130 that predicts the reactor water quality and the corrosion potential of the structural materials of the reactor on the next day from the plant state quantities for N days. Next, a process for predicting reactor water quality and corrosion potential of structural materials using the prediction model 130 will be explained.

≪Prediction processing using prediction model≫
FIG. 9 is a diagram for explaining input data and output data of the prediction model 130 in the prediction process according to this embodiment.
Dose rate distribution for N days included in plant state quantity 151, linear flow velocity, feed water flow rate, core flow rate, main steam flow rate, core inlet temperature, core outlet temperature, feed water hydrogen concentration, feed water iron concentration, precious metal injection amount, reactor water Plant state quantities, such as plant state quantities such as purification flow rate and heat output, are included as input data for the prediction model 130. This input data corresponds to the arrow pointing from the plant state quantity 151 to the prediction model 130 shown in FIG.

Furthermore, a simulation is executed using the radiolysis/corrosion potential model 140 using the plant state quantity 151 as input data. In detail, the simulation unit 113 calculates the dose rate distribution for N days, linear flow rate, feed water flow rate, core flow rate, main steam flow rate, core inlet temperature, core outlet temperature, feed water hydrogen concentration, and feed water concentration, which are included in the plant state quantity 151. Using the iron concentration, precious metal injection amount, reactor water purification flow rate, heat output, etc. as input, the radiolysis/corrosion potential model 140 is used to calculate water quality/corrosion potential 154 (in-reactor predicted value). This water quality/corrosion potential 154 is included as input data of the prediction model 130.
From these inputs, by using the prediction model 130, it is possible to calculate the water quality/corrosion potential 155 on the next day of N days' worth of input data as a prediction result.

FIG. 10 is a flowchart of the prediction process executed by the prediction unit 112 according to this embodiment. Referring to FIG. 10, a prediction process for predicting the reactor water quality and corrosion potential on the day after the reference date from the plant state quantities for the past N days including the reference date (the execution date of the prediction process) will be described.
In step S21, upon receiving instructions from the prediction unit 112, the simulation unit 113 uses the plant state quantity 151 as input data, executes a simulation using the radiolysis/corrosion potential model 140, and calculates the amount of reactor water for the past N days including the reference date. Calculate water quality and corrosion potential.

In step S22, the prediction unit 112 inputs input data to the prediction model 130. Specifically, the prediction unit 112 inputs into the prediction model 130 the plant state quantities for the past N days including the reference date, the reactor water quality and the corrosion potential calculated in step S21.
In step S23, the prediction unit 112 predicts the reactor water quality and corrosion potential by executing the prediction model 130 and obtaining outputs of the reactor water quality and corrosion potential.

When the reference date is the present, the water quality/corrosion potential 155 can be predicted from the plant state quantity 151 which is the past actual value (actual measurement value). Also, by setting the reference date in the past, the prediction can be verified. Specifically, the accuracy of the prediction can be evaluated by calculating the corrosion potential from the plant state quantity that is the past performance value and comparing it with the corrosion potential of the performance value. For example, the accuracy of the prediction can be evaluated by comparing the prediction result of today's corrosion potential when the reference date is one day before with the actually measured value of today's corrosion potential.

≪Modification: Multiple forecast dates≫
In the embodiment described above, the reactor water quality and corrosion potential on the day after the reference date are predicted. Predictions may be made up to N days ahead, including the day after the reference date. In this case, the water quality/corrosion potential 154A (see FIGS. 4 and 5), which is the teacher data that is the output data of the prediction model 130, is for N days. For example, in the teacher data 161, the water quality/corrosion potential 154A from day N+1 to day 2N is set as output data (correct data).
The period to be predicted is not limited to N days, and may be any other number of days. In addition, it is not necessary to predict the reactor water quality and corrosion potential the day after the reference date (one day after the reference date), but rather to predict the reactor water quality and corrosion potential a predetermined number of days after the reference date (after a predetermined period). It's okay.

≪Modification: Prediction including planned values≫
In the embodiment described above, prediction is made from actual data (plant state quantity 151) up to the reference date, but prediction may also be made including planned data (planned value).
FIG. 11 is a diagram for explaining the configuration of the teacher data of the prediction model 130 in the learning process according to the modification of the present embodiment. The input data for the teacher data are plant state quantities for 2N-1 days, reactor water quality, and corrosion potential. Furthermore, the output data (prediction results, correct data) of the teacher data is the reactor water quality and corrosion potential of the next day. For example, the plant state quantities, reactor water quality, and corrosion potential from day 1 to 2N-1 are input data, and the reactor water quality and corrosion potential from day 2N are output data. In FIG. 11, N teacher data whose input data start dates are from day 1 to day N are described.
In the learning process, a predictive model is generated from this teacher data. The generated prediction model is a prediction model that predicts the next day's reactor water quality and corrosion potential from the plant state quantities, reactor water quality, and corrosion potential for 2N-1 days.

FIG. 12 is a diagram for explaining input data and output data (prediction results) of a prediction model in prediction processing according to a modification of the present embodiment. In the prediction process, in addition to the plant state quantities, reactor water quality, and corrosion potential for the past N days including the reference date, the plant state quantities, reactor water quality, and the planned values for N-1 days from the day after the reference date are calculated. , the corrosion potential is used as input data.
For example, the feed water flow rate, core flow rate, feed water hydrogen concentration, feed water iron concentration, precious metal injection amount, reactor water purification flow rate, and thermal output are determined (predictable) in the operation plan, or the nuclear power plant operation ( settings), so it can be added to the input to the prediction model. Furthermore, the dose rate distribution is calculated from the core power distribution, so it can be predicted. The reactor water quality and corrosion potential can also be calculated from the plant state quantities using the radiolysis/corrosion potential model 140, and can be added to the input to the prediction model. Other input data can also be determined or predicted in the driving plan and added to the input to the prediction model.
By adding input data from days close to the expected date, the prediction device can predict corrosion potential with higher accuracy.

≪Modification: Operation planning using a prediction device≫
In order to formulate an operation plan for suppressing the corrosion potential to a desired value or less, the prediction process may be repeatedly executed while changing the plant state quantity serving as the planned value. For example, the prediction process is repeated while changing the feed water hydrogen concentration, feed water iron concentration, and precious metal injection amount as plant state quantities to find the feed water hydrogen concentration, feed water iron concentration, and precious metal injection amount that will cause the corrosion potential to be equal to or less than a desired value. By operating the plant based on the hydrogen concentration in the feed water, the iron concentration in the feed water, and the amount of noble metals injected, the corrosion potential can be kept below a desired value.

≪Characteristics of learning processing and prediction processing≫
The dose rate distribution of gamma rays and neutrons, which strongly affect reactor water quality and corrosion potential, is determined by the power distribution and thermal output of the reactor core. The amount of noble metal deposited on the surface of the structural material is strongly influenced by the amount of precious metal injected, the rate of deposition on the fuel rod surface, and the rate of deposition on the surface of the structural material. These cannot be measured while the nuclear power plant is operating, and shielding calculations are performed using the power distribution, or fuel rods are removed from the core P13 (see Figure 1) during periodic inspections or other times when the nuclear power plant is shut down. This value can only be measured by accessing the structural materials inside and outside the furnace.

Furthermore, the quality of reactor water (water quality inside the reactor) cannot be measured, and the water quality is measured after it is drawn out of the reactor through a small number of sampling lines. Therefore, since hydrogen peroxide decomposes within the sampling line, the actual concentration inside the furnace is unknown. Furthermore, the corrosion potential has only been measured at limited locations within the furnace. Since the corrosion potential measures the state in the immediate vicinity of the sensor, the corrosion potential in areas far from the sensor is unknown, so it is essential to combine this with calculation-based evaluation.

The prediction device 100 predicts reactor water quality and corrosion potential by combining a radiolysis/corrosion potential model 140 that is a physical model and a prediction model 130 (for example, a neural network) that is a machine learning model. This allows even if there are gaps in plant data that are missing in space or time, uncertainties in the physical model, and deviations due to phenomena that were not captured when modeling reality. By combining plant state variables and analysis (simulation) results, a machine learning model will enable more accurate predictions of reactor water quality and corrosion potential.

In other words, the input data of the prediction model 130 in the above-described embodiment is the dose rate distribution predicted by the dose rate prediction model 145, and the reactor water quality and corrosion potential calculated using the radiolysis/corrosion potential model 140. Contains. By adding plant state quantities that cannot be measured during operation but have a strong influence on reactor water quality and corrosion potential to the input data of the prediction model 130, compared to a model that inputs only actual measured values of plant state quantities, It will be possible to improve the prediction accuracy of reactor water quality and corrosion potential. Furthermore, compared to predictions made by the radiolysis/corrosion potential model 140 alone, the predictions take into account the correlation of plant state quantities that are not included in the radiolysis/corrosion potential model, improving accuracy.

≪Modification example≫
Regarding the dose rate prediction model 145 (see FIG. 6), if the calculation accuracy of the radiolysis/corrosion potential model 140, which is a physical model, is sufficiently high, the calculation results of the radiolysis/corrosion potential model 140 can be used without using the prediction model 130. May be used for plant operation.
FIG. 13 is a diagram for explaining a radiolysis/corrosion potential model 140 that predicts reactor water quality and corrosion potential in an embodiment different from the embodiment described above. The dose rate prediction model 145 receives the core power distribution included in the plant state quantity 151 as input and outputs a dose rate distribution 152. Using the output dose rate distribution 152 as input, the simulation unit 113 executes the radiolysis/corrosion potential model 140 to obtain the reactor water quality and corrosion potential (water quality/corrosion potential 154).

Based on the obtained corrosion potential, the feed water hydrogen concentration is controlled so that the corrosion potential of the target area falls below the target value (for example, -300 to -200 mV vs. SHE), and if necessary, precious metals are added to the reactor water. The amount of injection may be increased.
Based on the obtained dose rate distribution 152 in the furnace, it can also be used to calculate the intensity distribution of Cerenkov light in the furnace. This can be used to evaluate the acceleration and suppression of corrosion of oxides in the furnace due to photovoltaic force. There is shadow corrosion and photocatalyst in the reactor core.

≪Modification: Separation of devices that execute learning processing and prediction processing≫
Note that the present invention is not limited to the embodiments described above, and can be modified without departing from the spirit thereof. For example, the learning process (see FIG. 8) and the prediction process (see FIG. 10) performed by the prediction device 100 may be performed by another device. A predictive model construction device may execute a learning process to construct the predictive model 130, and the predictive device that has acquired the constructed predictive model 130 may execute the predictive process.

Dose rate prediction may be performed with a separate device.
FIG. 14 is a diagram for explaining a dose rate prediction device 170 according to a modification of this embodiment. A dose rate prediction model 145 is constructed by the dose rate prediction device 170 from the respective shielding calculation results held by a plurality of plants. Thereafter, the dose rate distribution 152 is calculated and output by inputting the plant state quantity of the plant to be predicted, specifically, the core power distribution. The output dose rate distribution 152 becomes input data to the radiolysis/corrosion potential model 140 of the prediction device 100, and outputs predicted values of reactor water quality and corrosion potential.

≪Modification: Driving support device≫
By changing the feed water hydrogen concentration, feed water iron concentration, and amount of precious metals injected, the reactor water quality and corrosion potential will change. In other words, by appropriately adjusting the feed water hydrogen concentration, feed water iron concentration, and amount of noble metal injection, desired reactor water quality and corrosion potential can be maintained. Specifically, the prediction device prepares several planning patterns for the feedwater hydrogen concentration, the feedwater iron concentration, and the amount of noble metal injection, and performs prediction processing for each planning pattern. The prediction device selects a planning pattern in which the predicted results of reactor water quality and corrosion potential satisfy desired values, and in which the feed water hydrogen concentration, feed water iron concentration, and precious metal injection amount are small. Further, the prediction device, as an operation support device, may operate the plant by adjusting (setting) the feed water hydrogen concentration, the feed water iron concentration, and the amount of noble metal injection according to the selected planning pattern.

FIG. 15 is a functional configuration diagram of a driving support device 100A according to a modification of this embodiment. Compared to the prediction device 100 (see FIG. 3), the storage unit 120 further includes a planning pattern 121, and the control unit 110 further includes a driving support unit 114. The planned pattern 121 is a pattern of planned values of the above-described feed water hydrogen concentration, feed water iron concentration, and amount of noble metal injection. Patterns of planned values include, for example, a pattern in which the concentration/injection amount is set to a predetermined value at the beginning of the planning period and then set to 0 or lower, or a pattern in which the concentration/injection amount is increased/reduced by a predetermined amount during a predetermined period, and then increased/decreased. There are patterns that maintain the value after reduction. The driving support unit 114 executes a driving support process shown in FIG. 16, which will be described later.

FIG. 16 is a flowchart of driving support processing according to a modification of this embodiment.
In step S31, the driving support unit 114 starts processing to execute step S32 for each planning pattern 121.
In step S32, the operation support unit 114 instructs the prediction unit 112 to execute a prediction process (see FIG. 10) using the plant state quantity shown in the planning pattern 121 as input data.

In step S33, if the driving support unit 114 has executed step S32 for every plan pattern 121, the process proceeds to step S34, and if there is an unprocessed plan pattern 121, the process returns to step S32, and for the unprocessed plan pattern, Process.
In step S34, the driving support unit 114 outputs the result (prediction result) of the prediction process in step S32 (displays it on the display of the input/output unit 160).

In step S35, the driving support unit 114 selects the optimal planning pattern 121. Specifically, the operation support unit 114 selects a planning pattern 121 from among the prediction results that satisfies preset criteria for reactor water quality and corrosion potential and has the optimum feedwater hydrogen concentration, feedwater iron concentration, and precious metal injection amount. do. For example, the operation support unit 114 calculates the cost of operating the plant by setting the feed water hydrogen concentration, the feed water iron concentration, and the precious metal injection amount by weighting the feed water hydrogen concentration, the feed water iron concentration, and the precious metal injection amount. Then, the planning pattern 121 with the minimum cost is selected.
In step S36, the operation support unit 114 operates the plant by setting the feed water hydrogen concentration, the feed water iron concentration, and the amount of noble metal injection according to the plan pattern 121 selected in step S35.

By using the above-described operation support device 100A, an operator can operate a plant while satisfying desired values for reactor water quality and corrosion potential at a low cost. Moreover, the workload of the operator can be reduced.

≪Other variations≫
In the embodiment described above, the input data is daily data, but the input data is not limited thereto, and may be data every two days or every six hours, for example. Furthermore, the interval is not limited to a fixed interval, and for example, the interval for input data far from the predicted date may be set to be longer. By increasing the interval and reducing the number of input data, the prediction device 100 can speed up learning processing and prediction processing.

Although several embodiments of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. The present invention can take various other embodiments, and various changes such as omissions and substitutions can be made without departing from the gist of the present invention. These embodiments and their modifications are included within the scope and gist of the invention described in this specification and the like, as well as within the scope of the invention described in the claims and its equivalents.

100 Prediction device (prediction model construction device)
100A Driving support device (prediction device)
111 Learning section 112 Prediction section 113 Simulation section 114 Operation support section 121 Planning pattern 130 Prediction model 140 Radiolysis/corrosion potential model (physical model)
145 Dose rate prediction model 151 Plant state quantities 152 Dose rate distribution 153 Core power distribution 154, 154A Water quality/corrosion potential (in-reactor predicted value)
155 Water quality/corrosion potential 156 Design data 170 Dose rate prediction device P1 Nuclear reactor P100 Nuclear power plant

Claims (10)

A predictive model construction device for constructing a predictive model for predicting reactor water quality of a nuclear reactor in a nuclear power plant,
a simulation unit that calculates in-reactor predicted values including predicted values of reactor water quality in the reactor using a physical model from input data including predicted values of dose rate distributions of gamma rays and neutron rays in the nuclear reactor;
At least any of the reactor's linear flow velocity, feed water flow rate, core flow rate, main steam flow rate, core inlet temperature, core outlet temperature, feed water hydrogen concentration, feed water iron concentration, precious metal injection amount, reactor water purification flow rate, and thermal output. A machine learning model is provided with training data that includes a plant state quantity including one and the in-furnace predicted value as input data, and includes as output data the in-furnace predicted value after a predetermined period from the period in which the input data was acquired. A predictive model construction device comprising: a learning section that performs learning to construct a predictive model. The physical model is a radiolysis/corrosion potential model,
The in-reactor predicted value further includes a predicted value of corrosion potential of structural materials of the reactor,
The reactor water quality is the concentration of at least one of oxygen, hydrogen, and hydrogen peroxide in the reactor water of the nuclear reactor,
The simulation unit calculates the in-furnace predicted value using the physical model,
The learning unit performs machine learning on teacher data that includes the plant state quantity and the in-furnace predicted value as input data, and includes as output data the in-furnace predicted value after a predetermined period from the period in which the input data was acquired. Train the model and build a predictive model
The predictive model construction device according to claim 1, characterized in that: The learning unit causes a machine learning model to learn training data that includes design data and core power distributions of a plurality of reactors as input data and includes dose rate distributions of the plurality of reactors as output data to predict dose rates. build the model,
2. The predicted value of the dose rate distribution, which is the input data of the physical model, is calculated using the dose rate prediction model using the design data of the nuclear reactor and the core power distribution as input. The predictive model building device described. A prediction device for predicting reactor water quality of a nuclear reactor in a nuclear power plant,
a simulation unit that calculates in-reactor predicted values including predicted values of reactor water quality in the reactor using a physical model from input data including predicted values of dose rate distributions of gamma rays and neutron rays in the nuclear reactor;
At least any of the reactor's linear flow velocity, feed water flow rate, core flow rate, main steam flow rate, core inlet temperature, core outlet temperature, feed water hydrogen concentration, feed water iron concentration, precious metal injection amount, reactor water purification flow rate, and thermal output. A machine learning model is provided with training data that includes a plant state quantity including one and the in-furnace predicted value as input data, and includes as output data the in-furnace predicted value after a predetermined period from the period in which the input data was acquired. a storage unit in which a predictive model built by learning is stored;
A prediction unit that uses the prediction model to calculate an in-core predicted value after the predetermined period from the period in which the plant state quantity and the in-core predicted value are obtained, from the plant state quantity and the in-furnace predicted value. A prediction device comprising: and . The physical model is a radiolysis/corrosion potential model,
The in-reactor predicted value further includes a predicted value of corrosion potential of structural materials of the reactor,
The reactor water quality is the concentration of at least one of oxygen, hydrogen, and hydrogen peroxide in the reactor water of the nuclear reactor,
The simulation unit calculates the in-furnace predicted value using the physical model,
The prediction model performs machine learning on training data that includes the plant state quantities and the in-furnace predicted value as input data, and includes as output data the in-furnace predicted value after a predetermined period from the period in which the input data was acquired. Constructed by training the model
The prediction device according to claim 4, characterized in that: The storage unit includes design data and core power distributions of a plurality of nuclear reactors as input data, and is constructed by causing a machine learning model to learn training data including dose rate distributions of the plurality of reactors as output data. The dose rate prediction model is further stored,
5. The predicted value of the dose rate distribution, which is the input data of the physical model, is calculated using the dose rate prediction model using the design data of the reactor and the core power distribution as input. The prediction device described. The storage unit further stores one or more planning patterns of plan values of plant state quantities that can be set among the plant state quantities,
The prediction device according to claim 4, further comprising an operation support unit that uses the prediction unit to output an in-core predicted value after the predetermined period by using a plant state quantity including the plan pattern as input data. The driving support department is
Outputting a plan pattern that satisfies a predetermined standard and minimizes the operating cost of the nuclear power plant when the plan pattern is set, among the in-core predicted values after the predetermined period corresponding to the plan pattern. The prediction device according to claim 7, characterized by: The prediction device according to claim 8, wherein the settable plant state quantity is set according to a planning pattern that minimizes the operating cost or a selected planning pattern. The prediction device according to claim 7, wherein the settable plant state quantity includes at least one of a feed water hydrogen concentration, a feed water iron concentration, and a precious metal injection amount.

Images