JP7590264B2 - 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 the reactor water radioactivity concentration and water discharge information of a nuclear power plant.

Known examples of nuclear power plants (also simply referred to as plants) include boiling water nuclear power plants and pressurized water nuclear power plants. In these plants, stainless steel or nickel-based alloys are used in the main components, such as the reactor pressure vessel, in the water-contacting parts to prevent corrosion. Furthermore, in these plants, a portion of the cooling water (hereafter also referred to as reactor water) present in the reactor pressure vessel is purified by a reactor water purification system, and small amounts of metallic impurities present in the reactor water are actively removed.

Even with the above-mentioned corrosion prevention measures, it is unavoidable that trace amounts of metal impurities remain in the reactor water, and some of these metal impurities adhere to the outer surface of the fuel rods in the form of metal oxides. The metal elements contained in the metal impurities that adhere to the outer surface of the fuel rods undergo nuclear reactions when irradiated with neutrons emitted from the nuclear fuel material in the fuel rods, and become radioactive nuclides such as cobalt-60, cobalt-58, chromium-51, and manganese-54. Some of the radioactive nuclides that adhere to the outer surface of the fuel rods in the form of oxides dissolve into the reactor water as ions depending on the solubility of the oxides that have been incorporated. Radioactive nuclides are also released into the reactor water as insoluble solids called clad.

Some of the radionuclides in the reactor water are removed by the reactor water purification system. However, the radionuclides that are not removed accumulate on the surfaces of structural components that come into contact with the reactor water while circulating through the recirculation system together with the reactor water. As a result, radiation is emitted from the surfaces of the structural components, causing radiation exposure to personnel performing periodic inspections. The radiation exposure dose of personnel is managed so that it does not exceed a specified value. However, in recent years, this specified value has been lowered, and it has become necessary to make the radiation exposure dose of each employee as low as economically possible.

In this situation, predicting the radiation exposure dose at the next regular inspection and then formulating shielding and workforce plans and determining the need for decontamination is an effective measure to reduce the total radiation exposure dose. Predicting the radiation exposure dose requires predicting the piping dose, and because the piping dose is strongly dependent on the reactor water radioactivity concentration during operation, it is important to predict the trends in the reactor water radioactivity concentration during the plant's operation period. In addition, since predicting the radiation exposure dose at the next regular inspection will be used in the regular inspection plan, the prediction must be made as soon as possible.

As a prediction device for reactor water radioactivity concentration, for example, there is a water quality diagnostic system for reducing the dose rate of the primary reactor system that uses a physical and chemical simulation model as described in Patent Document 1. This water quality diagnostic system predicts future plant dose rates using a simulation model (mass balance model) that estimates changes in radioactivity in the cooling water using current water quality conditions as input, and diagnoses the quality of the current water quality conditions based on the results.

The technology in Patent Document 1 requires that a simulation model and model parameters exist and be optimized. However, in reality, model parameters change over time, and as operation continues, equipment and materials are replaced, which can lead to a large difference between model calculation values and actual measured values.

The self-learning diagnosis and prediction device described in Patent Document 2 automatically corrects model parameters in response to changes over time in the plant's specifications and characteristics, preventing deterioration of prediction accuracy, and has the ability to improve the model through self-learning.

Japanese Patent Application Publication No. 64-063894 Japanese Patent Application Publication No. 06-289179

The technology described in Patent Document 2 optimizes model parameters of a model set based on a physical model or a chemical model and adjusts the contribution of the model parameters, and the model must be prepared in advance. For this reason, the correlation between the state quantity of the predicted target and the state quantity used as input must be described as a mathematical formula.
The correlations between the state quantities expressed in the mass balance model are optimized by optimizing the model parameters. However, although correlations are possible, if the correlations are complex and difficult to express mathematically, it is not possible to build an appropriate model. For this reason, there is a demand for accurate prediction of the state quantities of a plant even when correlations that are complex and cannot be expressed mathematically are included.

In addition to predicting the radioactivity concentration of reactor water, plants are also required to manage the temperature of warm wastewater (discharged water). More specifically, at plants, seawater is used to cool the steam that is generated from the heat produced by nuclear fission and turns the turbines, and it is necessary to comply with regulations regarding the range of temperature rise of seawater (warm wastewater, discharged water) as well as reducing the radioactivity concentration of the reactor water. Since the temperature depends on the seawater temperature at the time of water intake and the thermal output of the reactor, it is required to predict the temperature at the time of discharge (or the temperature difference from when water was taken in, also written as discharged water information) along with the radioactivity concentration of the reactor water.

The present invention has been made in consideration of the above problems, and aims to provide a prediction model construction device and a prediction device that construct a prediction model that enables highly accurate prediction of reactor water radioactivity concentration and water discharge information in a nuclear power plant.

In order to solve the above-mentioned problems, the prediction model construction device of the present invention has a learning unit that constructs a prediction model by training a machine learning model with training data including, as input data, measurable plant state quantities and seawater information, and including, as first output data, a first reactor water radioactive metal corrosion product concentration related to a first reactor water radioactive metal, which is an actual value, and water discharge information, and also has a learning unit that constructs a verification index model by training a machine learning model with training data including, as input data, plant state quantities and seawater information, and including, as second output data, a second reactor water radioactive metal corrosion product concentration related to a second reactor water radioactive metal different from the first reactor water radioactive metal, which is an actual value, and water discharge information.

The present invention provides a prediction model construction device and a prediction device that construct a prediction model that enables highly accurate prediction of reactor water radioactivity concentration and water discharge information in a nuclear power plant.

1 is an overall system configuration diagram of a nuclear power plant according to an embodiment; 1 is a diagram for explaining the principle of a prediction device according to an embodiment; FIG. 2 is a diagram for explaining a mass balance model of the transfer behavior of metal corrosion products into reactor water according to the embodiment. 1 is a functional configuration diagram of a reactor water radioactive concentration/discharge water temperature prediction device according to an embodiment. FIG. FIG. 4 is a diagram illustrating an example of objective variable data according to the embodiment. FIG. 11 is a diagram illustrating an example of verification index data according to the embodiment. 1 is a diagram for explaining the configuration of data that is the source of teacher data for a prediction model in a learning process according to the present embodiment. FIG. 1A and 1B are diagrams for explaining the configuration of data that is the source of input data and output data of a prediction model in a prediction process according to an embodiment. 11 is a flowchart for explaining a process of creating a prediction model in the prediction process according to the embodiment. 11 is a flowchart for explaining a process of creating a validation index prediction model in the prediction process according to the embodiment. 11 is a flowchart for explaining a process of creating a consistency determination model in a prediction process according to the embodiment. 11 is a flowchart for explaining a prediction result consistency verification process in the prediction process according to the embodiment. 11 is a flowchart for explaining a consistency determination process in the prediction process according to the embodiment.

The following describes an embodiment of the present invention with reference to the drawings. The following description and drawings are examples for explaining the present invention, and some parts have been omitted or simplified as appropriate for clarity of explanation. The present invention can also be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

In addition, in the figures explaining the embodiments, parts having the same functions are given the same reference numerals, and repeated explanations are omitted.

The position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc., in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings.

When there are multiple components with the same or similar functions, they may be described using the same reference numerals with different subscripts. However, when there is no need to distinguish between these multiple components, the subscripts may be omitted.

Before describing the reactor water radioactivity concentration/discharge water temperature prediction device (prediction model construction method and prediction method) in the form (embodiment) for implementing the present invention, the nuclear power plant that is the subject of the prediction will be described.

Figure 1 is a diagram showing the overall system configuration of a nuclear power plant P100 according to an embodiment. The schematic configuration of a nuclear power plant P100 (e.g., a boiling water nuclear power plant) to which the reactor water radioactivity concentration/discharge water temperature prediction device according to this embodiment is applied will be described with reference to Figure 1.

The nuclear power plant P100 includes a reactor P1, a turbine P3, a condenser P4, a reactor purification system, and a water supply system. The reactor P1, which is installed in a reactor containment vessel P11, has a reactor pressure vessel P12 that houses a reactor core P13. A cylindrical core shroud P15 installed in the reactor pressure vessel P12 surrounds the reactor core P13. A plurality of fuel assemblies (not shown) are loaded into the reactor core P13. Each fuel assembly includes a plurality of fuel rods 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 core shroud P15. A plurality of internal pumps P21 are installed at the bottom of the reactor pressure vessel P12. The impellers of the internal pumps P21 are arranged at the bottom of the downcomer P17.

The feedwater system is composed of a condensate pump P5, a condensate purification device P6, a feedwater pump P7, a low-pressure feedwater heater P8, and a high-pressure feedwater heater P9, which are installed in this order on a feedwater piping P10 that connects the condenser P4 and the reactor pressure vessel P12, from the condenser P4 toward the reactor pressure vessel P12. A hydrogen injection device P16 is connected to the feedwater piping P10 between the condensate purification device P6 and the feedwater pump P7 by a hydrogen injection piping P18. An opening/closing valve P19 is provided on the hydrogen injection piping P18.

The reactor cleanup system is composed of a cleanup system isolation valve P23, a cleanup system pump P24, a regenerative heat exchanger P25, a non-regenerative heat exchanger P26, and a reactor water cleanup device P27, which are installed in this order in the cleanup system piping (stainless steel member) P20 made of stainless steel that connects the reactor pressure vessel P12 and the feedwater piping P10. The residual heat removal system installed in the nuclear power plant P100 has a residual heat removal system piping P28, one end of which is connected to the reactor pressure vessel P12 and connected to the downcomer P17, and the other end of which is connected to the inside of the reactor pressure vessel P12 above the core P13. A residual heat removal system pump P29 and a heat exchanger (cooling device) P30 are installed in this residual heat removal system piping P28. One end of the cleanup system piping P20 is connected to the residual heat removal system piping P28 upstream of the residual heat removal system pump P29.

The cooling water (reactor water) present in the downcomer P17 inside the reactor pressure vessel P12 is pressurized by the internal pump P21 and guided to the lower plenum below the reactor core P13. The reactor water is supplied from the lower plenum to the reactor core P13 and heated by heat generated by nuclear fission of the nuclear fuel material contained in the fuel rods of the fuel assemblies. A part of the heated reactor water becomes steam. This steam is guided from the reactor pressure vessel P12 through the main steam pipe P2 to the turbine P3, rotating the turbine P3. A generator (not shown) connected to the turbine P3 is rotated, generating electricity. The steam discharged from the turbine P3 is condensed in the condenser P4 to become water.

Condenser P4 is a type of heat exchanger that uses seawater taken in at the intake port to cool the steam and turn it back into water. The heated wastewater, whose temperature has increased by cooling the steam, is discharged from the discharge port. The difference between the temperature at the time of intake (intake temperature) and the temperature at the time of discharge (discharge temperature) is specified to be 7 to 8°C. For this reason, the plant is operated while adjusting the seawater flow rate and thermal output.

The water condensed in the condenser P4 is supplied as feedwater through the feedwater piping P10 into the reactor pressure vessel P12. The feedwater flowing through the feedwater piping P10 is pressurized by the condensate pump P5, impurities are removed by the condensate purification device P6, the pressure is further increased by the feedwater pump P7, and the low-pressure feedwater heater P8 and the high-pressure feedwater heater P9 are heated. Extracted steam extracted from the main steam piping P2 and the turbine P3 by the extraction piping P14 is supplied to the low-pressure feedwater heater P8 and the high-pressure feedwater heater P9, respectively, and serves as a heating source for the feedwater.

Because the reactor water in the reactor pressure vessel P12 contains metal corrosion products contained in the feed water and products resulting from corrosion of structural materials in the reactor pressure vessel P12, a certain percentage of the reactor water is purified by the reactor water 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 off from the residual heat removal system piping P28 by the drive of the purification system pump P24, and is cooled to about 50°C by these heat exchangers. The cooled reactor water passes through the reactor water purification device P27 to remove the metal corrosion products contained in the reactor water, is heated in the regenerative heat exchanger P25, and is then merged with the feed water flowing in the feed water piping P10 and supplied to the reactor pressure vessel P12.

When the operation of the reactor P1 is stopped, all the control rods (not shown) are inserted into the core. The nuclear fission reaction of the nuclear fuel material is stopped by inserting all the control rods, and the operation of the reactor P1 is stopped. The heat remaining in the equipment in the reactor core P13 and the reactor pressure vessel P12 is removed by the evaporation of the reactor water, but when the temperature drops to a certain level, the efficiency of heat removal by evaporation of the reactor water decreases, so when the reactor water temperature drops to about 150°C, the residual heat removal system is used to cool the equipment in the reactor core P13 and the reactor pressure vessel P12. That is, the residual heat removal system pump P29 is driven to supply the reactor water in the reactor pressure vessel P12 to the heat exchanger P30 through the residual heat removal system piping P28, and the reactor water is then cooled in the heat exchanger P30 and returned to the reactor pressure vessel P12.

When a nuclear reactor is in operation, the reactor water contains radioactive metal corrosion products, including cobalt-60, which adhere to structural materials depending on their concentration. Radiation from the radionuclides that have adhered to the structural materials causes radiation exposure for personnel involved in periodic inspections.

Next, with reference to FIG. 2, the principle of the reactor water radioactivity concentration/discharge water temperature prediction device of this embodiment (hereinafter simply referred to as the "prediction device") will be explained.

It is known that a physical model shown in the mass balance model in Figure 3 holds true between the metallic corrosion products contained in the feed water into the reactor P1 and the metallic corrosion products generated due to corrosion of structural materials inside and outside the reactor that come into contact with the reactor water.

The mass balance model is a physical model that uses the macroscopic law of conservation of mass to describe the dynamic behavior of metallic corrosion products contained in the feedwater and metallic corrosion products generated due to corrosion of structural materials inside and outside the reactor that come into contact with the reactor water, such as redepositing onto the surfaces of fuel rods and structural materials inside and outside the reactor via the reactor water, or being removed from the system by the reactor water purification system. The solid and dashed arrows in Figure 2 indicate migration by crud and ions.

The mass balance model of metal corrosion products is described by the group of simultaneous differential equations shown in (Equation 1) to (Equation 8) below.

In the above formula, the variables and parameters have the following meanings:
C: Metal corrosion product concentration in reactor water (metal corrosion product concentration in reactor water, e.g., concentration of iron, nickel, cobalt, etc.)
t: time V: amount of retained reactor water F _f : feedwater flow rate C _f : concentration of metal corrosion products in feedwater (concentration of metal corrosion products in feedwater)
X: Generation rate of metal corrosion products generated by corrosion of in-core structural materials ζ: Dissolution or peeling constant of fuel rod deposits ζ _p ¹ : Dissolution or peeling constant of in-core structural materials ζ _p ² : Dissolution or peeling constant of outside-core structural materials

M: amount of metal corrosion products accumulating on fuel rods _m1 : amount of metal corrosion products adhering to and accumulating on the surfaces of structural materials inside the reactor _m2 : amount of metal corrosion products adhering to and accumulating on the surfaces of structural materials outside the reactor δ: adhesion constant to fuel rods β: removal rate in the reactor cleanup system _δp1 : adhesion constant to structural materials inside the ^reactor _δp2 : adhesion constant to structural materials outside the reactor _S1 : surface area of structural materials inside the reactor ^S2 : surface area of structural materials outside the _reactor

R: Reactor water radioactive metal corrosion product concentration (reactor water radioactive metal corrosion product concentration, for example, concentration of cobalt 60, cobalt 58, manganese 54, etc.)
Y: Generation rate of radioactive metal corrosion products generated by corrosion of reactor internal structural materials A: Amount of radioactive metal corrosion products accumulated on fuel rods _Γ1 : Amount of radioactive metal corrosion products adhering to and accumulating on the surface of structural materials inside the reactor _Γ2 : Amount of radioactive metal corrosion products adhering to and accumulating on the surface of structural materials outside the reactor λ: Decay constant of radioactive metal corrosion products G: Generation rate of radioactive nuclides on fuel rods _G1 : Generation rate of radioactive nuclides on structural materials inside the reactor

Among the above variables, C, _Cf , _Ff , R, and _Γ2 are state quantities that can be measured during operation, and M, A, and _Γ1 are state quantities that can be measured when the fuel rods are removed from the reactor during shutdown such as periodic inspection. Also, V, _S1 , and _S2 are plant parameters specific to the plant. _{λ, G, and G1 are physical constants determined according to the nuclide of the radioactive metal corrosion product, and X, Y, ζ, ζp1, ζp2 , δ} ^, _δp1 , ^δp2 , and β are model parameters in principle ^. Note that _{m1 and m2} are state quantities that are usually difficult to measure because they cannot be distinguished between those attached from the water side ^and those generated by corrosion of the structural material.

In the prediction device of this embodiment, the metal corrosion product to be predicted is cobalt 60. On the other hand, it is considered that metal corrosion products other than the prediction target can also follow the above-mentioned mass balance model. Therefore, it is expected that there is a strong correlation between the metal corrosion products contained in the feedwater and cobalt 60, which is a metal corrosion product generated due to corrosion of structural materials inside and outside the reactor that are in contact with the reactor water, while it is also expected that there is a strong correlation between the metal corrosion products contained in the feedwater and metal corrosion products other than cobalt 60 that are generated due to corrosion of structural materials inside and outside the reactor that are in contact with the reactor water (cobalt 58 and iron are given as examples in Figure 2). And, assuming these correlations, it is expected that there is also a strong correlation between the concentration of cobalt 60 and the concentrations of cobalt 58 and iron.

In this embodiment, the prediction device generates a first learning model that learns the correlation between the metal corrosion products contained in the feedwater and cobalt 60, which is a metal corrosion product generated due to corrosion of structural materials inside and outside the reactor that are in contact with the reactor water, and generates a second learning model that learns the correlation between the metal corrosion products contained in the feedwater and metal corrosion products other than cobalt 60 that are generated due to corrosion of structural materials inside and outside the reactor that are in contact with the reactor water. Then, (1) the first learning model predicts the concentration of the metal corrosion products of cobalt 60 based on the metal corrosion products contained in the actual feedwater, (2) the second learning model predicts the concentration of the metal corrosion products other than cobalt 60 based on the metal corrosion products contained in the actual feedwater, and (3) predicts the concentration of the metal corrosion products of cobalt 60 predicted from the concentration of the metal corrosion products other than cobalt 60, and judges the validity of the result of (1) based on the difference between the result of (1) and the result of (3) (both of which predict the concentration of the metal corrosion products of cobalt 60).

Below, a reactor water radioactivity concentration/discharge water temperature prediction device that predicts the reactor water radioactivity concentration and discharge water temperature in the nuclear power plant P100 according to this embodiment will be described. The reactor water radioactivity concentration/discharge water temperature prediction device predicts the concentrations of cobalt 60, cobalt 58, chromium 51, manganese 54, etc. in the reactor water, and the discharge water temperature. Machine learning technology is used for the prediction.

The reactor water radioactivity concentration/discharge water temperature prediction device according to this embodiment can predict the concentration of radioactive metal corrosion products and the discharge water temperature with high accuracy. Using this prediction, plant operators can create and implement plans to reduce radiation exposure by lowering the reactor water radioactivity concentration during operation, installing shielding, or carrying out chemical decontamination, while adhering to regulations for temperature management of discharge water (warm wastewater).

Figure 4 is a functional configuration diagram of the prediction device 100 according to this embodiment. The prediction device 100 is a computer, and includes a control unit 110, a model storage unit 120, a data storage unit 130, and an input/output unit 140. The prediction device 100 receives operational data such as thermal output from a process computer P110 used in the nuclear power plant P100, and output data from a measuring instrument P120 installed in the nuclear power plant P100.

The model storage unit 120 is composed of RAM (Random Access Memory), ROM (Read Only Memory), SSD (Solid State Drive), etc., and has a prediction model storage unit 121 in which a prediction model is stored, a verification index model storage unit 122 in which a verification index model is stored, and a consistency judgment model storage unit 123 in which a consistency judgment model is stored.

The prediction model stored in the prediction model storage unit 121 is a machine learning learning model, such as a neural network, that uses plant state quantities and seawater information as input data (explanatory variables) and the concentration of metal corrosion products of cobalt-60 as output data (objective variable).

The verification index model stored in the verification index model storage unit 122 is a machine learning learning model that uses plant state quantities and seawater information as input data (explanatory variables) and metal corrosion product concentrations other than cobalt 60 (in this embodiment, cobalt 58, iron, and nickel, as described later) as output data (objective variables), and is, for example, a neural network.

The consistency determination model stored in the consistency determination model storage unit 123 is a machine learning learning model, for example a neural network, that uses metal corrosion product concentrations other than cobalt-60 as input data (explanatory variables) and metal corrosion product concentrations of cobalt-60 as output data (objective variables).

The plant state quantities referred to in this embodiment include plant data, feed water data, and reactor water quality data of the nuclear power plant P100. The plant state quantities also include plant state quantities such as the in-reactor residence time of the fuel assemblies loaded in the core P13 (see FIG. 1) and electrical output. The seawater information is the intake water temperature (seawater temperature at the intake port) and the discharge water temperature (warm discharge water temperature at the discharge port). The seawater information may also include information such as the temperature and position of ocean currents that affect the intake water temperature.

The data storage unit 130, like the model storage unit 120, is composed of RAM, ROM, SSD, etc., and has an explanatory variable data storage unit 131 in which explanatory variable data is stored, an objective variable data storage unit 132 in which objective variable data is stored, and a verification index data storage unit 133 in which verification index data is stored.

The explanatory variable data stored in the explanatory variable data storage unit 131 consists of the plant state quantities and seawater information described above.

The objective variable data stored in the objective variable data storage unit 132 has, as an example, a data structure as shown in FIG. 5. As shown in FIG. 5, the objective variable data 200 has, as entries, a date 201 and a reactor water cobalt-60 ion concentration 202.

The verification index data stored in the verification index data storage unit 133 has, as an example, a data structure as shown in FIG. 6. As shown in FIG. 6, the verification index data 210 has, as entries, a date 211, a reactor water cobalt-58 ion concentration 212, a reactor water iron crud concentration 213, and a reactor water nickel ion concentration 214.

The input/output unit 140 receives data from the process computer P110 and the measuring instrument P120 and stores it in the plant state quantity/seawater information database 150. The input/output unit 140 also includes a display, keyboard, and 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 is composed of a CPU (Central Processing Unit), and has a learning unit 111 , a prediction unit 112 , a consistency determination model creation unit 113 , and a verification unit 114 .
The learning unit 111 performs a learning process using the explanatory variable data stored in the explanatory variable data storage unit 131 as teacher data (learning data) and generates a prediction model for predicting the radioactive concentration and discharge water temperature of the reactor water and a verification index prediction model.

The prediction unit 112 inputs explanatory variable data into the generated prediction model and the verification index model, and executes a prediction process to predict the radioactive concentration and discharge water temperature of the reactor water.

The consistency determination model creation unit 113 performs a learning process using the verification index data stored in the verification index data storage unit 133 as input data, and generates a consistency determination model that predicts the objective variable data.

The verification unit 114 compares the concentration of metal corrosion products of cobalt 60 in the reactor water, which is the output result of the prediction model, with the concentration of metal corrosion products of cobalt 60 in the reactor water, which is the output result of the consistency determination model, and verifies the validity of the concentration of metal corrosion products of cobalt 60 in the reactor water, which is the output result of the prediction model.

Details of the processing performed by the learning unit 111, prediction unit 112, consistency determination model creation unit 113, and verification unit 114 will be described later with reference to Figures 9 to 13.

Figure 7 is a diagram for explaining the configuration of data that is the source of the training data for the prediction model in the learning process according to this embodiment. In this embodiment, an example is explained in which predictions are made for the next three years (36 months) from data for the past three years (36 months).

Dataset 230 in FIG. 7 is a dataset that is the source of one set of training data starting from the first month. Dataset 230 includes plant state quantities and seawater information (intake water temperature and discharge water temperature) from the first month to the 36th month as input data (original data for the input data). Dataset 230 also includes reactor water radioactivity concentration of cobalt-60 (also referred to as cobalt-60 concentration) and discharge water temperature from the 37th month to the 72nd month as output data (original data for the output data). These data are stored in explanatory variable data storage unit 131.

The plant state quantities that can be measured (actual data) included in each of the plant state quantities from 1st to 36th month, such as feedwater flow rate, feedwater metal corrosion product concentration, reactor water metal corrosion product concentration, reactor water radioactive metal corrosion product concentration, reactor water purification flow rate, duration of fuel assemblies in the reactor, electrical output, intake water temperature, and discharge water temperature, are included as input data for the prediction model.

The cobalt-60 concentration and discharge water temperature (actual measured values) from the 37th to 72nd month are included as output data of the prediction model (reactor water radioactive metal corrosion product concentration and discharge water temperature).

The input data and output data described above constitute one set of training data for the prediction model. In the same manner, one set of training data is generated from data set 240 starting from the second month. By repeating this process, 30 sets of training data (training data sets) for the prediction model are generated from plant state quantities including cobalt-60 and seawater information from the first to the 101st month. The training data that is the input of the prediction model includes measurable plant state quantities, seawater information (intake water temperature and discharge water temperature), and predicted values of plant state quantities. The training data that is the output of the prediction model includes the actual measured values of the reactor water radioactive metal corrosion product concentration (cobalt-60 concentration) and discharge water information (discharge water temperature).

Figure 8 is a diagram illustrating the configuration of the data that is the source of input data for the prediction model and the output data in the prediction process according to this embodiment.

In Figure 7, the base month is the month in which the prediction process is performed, and the prediction unit 112 predicts the cobalt-60 concentration and discharge water temperature from January to 36 months after the base month based on the plant state quantities and seawater information for the past 36 months, including the base month.

In detail, the input data for the prediction model includes the plant state quantities that can be measured, such as the feedwater flow rate, feedwater metal corrosion product concentration, reactor water metal corrosion product concentration, reactor water radioactive metal corrosion product concentration, reactor water purification flow rate, duration of fuel assemblies in the reactor, and electrical output, which are included in each of the plant state quantities from 35 months prior to the reference month through the reference month, as well as seawater information including the intake water temperature and discharge water temperature (actual measurement data).

The cobalt-60 concentration and discharge water temperature from 1 month to 36 months after the reference date are included as output data (prediction results) of the prediction model.

Note that the data shown in Figures 7 and 8 is related to the prediction model, but the data related to the validation index model is similar, so we will not explain it here.

Next, the operation of the prediction device 100 of this embodiment will be described with reference to the flowcharts in Figures 9 to 13.

Figure 9 is a flowchart for explaining the process of creating a prediction model in the prediction process according to the embodiment.

First, the learning unit 111 of the prediction device 100 reads the objective variable data 200 from the objective variable data storage unit 132 (S901). Next, the learning unit 111 reads the explanatory variable data from the explanatory variable data storage unit 131 (S902). Furthermore, the learning unit 111 creates a regression model in which the explanatory variable data is input data and the objective variable data 200 is output data (S903). Then, the learning unit 111 stores the regression model created in S903 in the prediction model storage unit 121 as a prediction model (S904).

Figure 10 is a flowchart for explaining the process of creating a validation index prediction model in the prediction process according to the embodiment.

First, the learning unit 111 of the prediction device 100 reads the verification index data 210 from the verification index data storage unit 133 (S1001). Next, the learning unit 111 reads the explanatory variable data from the explanatory variable data storage unit 131 (S1002). Furthermore, the learning unit 111 creates a regression model in which the explanatory variable data is input data and the verification index data 210 is output data (S1003). Then, the learning unit 111 stores the regression model created in S1003 in the verification index model storage unit 122 as a verification index model (S1004).

Figure 11 is a flowchart for explaining the process of creating a consistency determination model in the prediction process according to the embodiment.

First, the consistency determination model creation unit 113 of the prediction device 100 reads the objective variable data 200 from the objective variable data storage unit 132 (S1101). Next, the consistency determination model creation unit 113 reads the verification index data 210 from the verification index data storage unit 133 (S1102). Furthermore, the consistency determination model creation unit 113 creates a regression model in which the verification index data 210 is input data and the objective variable data 200 is output data (S1103). Then, the consistency determination model creation unit 113 stores the regression model created in S1103 in the consistency determination model storage unit 123 as a consistency determination model (S1104).

Figure 12 is a flowchart for explaining the prediction result consistency verification process in the prediction process according to the embodiment.

First, the prediction unit 112 of the prediction device 100 reads explanatory variable data from the explanatory variable data storage unit 131 (S1201).

Next, the prediction unit 112 reads the prediction model from the prediction model storage unit 121 (S1202). Furthermore, the prediction unit 112 inputs the explanatory variable data read in S1201 to the prediction model read in S1202 to obtain a predicted value (prediction result) of the objective variable data (S1203).

In parallel with this, the prediction unit 112 reads the verification index model from the verification index model storage unit 122 (S1204). Furthermore, the prediction unit 112 inputs the explanatory variable data read in S1201 to the prediction model read in S1204 to obtain a predicted value (prediction result) of the verification index data (S1205).

Next, the verification unit 114 of the prediction device 100 performs a consistency determination process (S1206). Details of the consistency determination process will be described later. Then, the verification unit 114 outputs the result of the consistency determination process of S1206.

Figure 13 is a flowchart for explaining the consistency determination process in the prediction process according to the embodiment, and is a flowchart showing the details of the consistency determination process of S1206 in Figure 12.

First, the verification unit 114 of the prediction device 100 reads the consistency judgment model from the consistency judgment model storage unit 123 (S1301). Next, the verification unit 114 inputs the predicted value of the verification index data acquired in S1205 of FIG. 12 into the consistency judgment model to obtain an estimate of the objective variable data (S1302).

Then, the verification unit 114 compares the estimated value of the objective variable data acquired in S1302 (the estimated result of Cobalt 60) with the predicted value of the objective variable data acquired in S1203 of FIG. 12 (the predicted result of Cobalt 60), and determines whether the difference between them is equal to or smaller than a predetermined error value (S1203). Note that it is preferable to set the error value in advance.

If the result of the judgment in S1203 is that the error is below the predetermined error value (YES in S1203), the verification unit 114 sets the consistency judgment result to "YES" (S1204), and if the result of the judgment in S1203 is that the error is above the predetermined error value (NO in S1203), the verification unit 114 sets the consistency judgment result to "NO" (S1205).

As described above in detail, the prediction device 100 of this embodiment makes it possible to verify the consistency and validity of the prediction results of the concentration of metal corrosion products of cobalt 60.

It is possible to create a learning model based on a mass balance model and predict the metal corrosion product concentration of cobalt 60 using this learning model. On the other hand, it is difficult to guarantee the accuracy of the metal corrosion product concentration of cobalt 60 based on the learning model using the learning model alone. According to the prediction device 100 of this embodiment, it is possible to verify the predicted value of the metal corrosion product concentration of cobalt 60 obtained by the prediction model using the predicted value of the metal corrosion product concentration other than cobalt 60 obtained by the verification index model and the estimated value of the metal corrosion product concentration of cobalt 60 obtained by the consistency judgment model. This makes it possible to verify the consistency and validity of the metal corrosion product concentration of cobalt 60 obtained by the prediction model, and to obtain a more accurate metal corrosion product concentration of cobalt 60.

The present invention is not limited to the above-mentioned examples, but includes various modified examples. The above-mentioned examples have been described in detail to clearly explain the present invention, and are not necessarily limited to those having all of the configurations described. It is also possible to replace part of the configuration of one example with the configuration of another example, and it is also possible to add the configuration of another example to the configuration of one example. It is also possible to add, delete, or replace part of the configuration of each example with other configurations.

For example, in the above embodiment, it is possible to retrain the prediction model using the estimated results of the cobalt-60 metal corrosion product concentration obtained by the consistency determination model.

Furthermore, the above-mentioned configurations, functions, processing units, processing means, etc. may be realized in part or in whole in hardware, for example by designing them as integrated circuits. Furthermore, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information on the programs, tables, files, etc. that realize each function can be stored in a memory, a recording device such as a hard disk or SSD, or a recording medium such as an IC card, SD card, or DVD.

In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other.
(Appendix 1)
A prediction model construction device for constructing a prediction model for predicting a concentration of radioactive metal corrosion products in reactor water and seawater discharge information of a nuclear reactor in a nuclear power plant , comprising:
A learning unit that constructs the prediction model by training a machine learning model with training data including, as input data, a measurable plant state quantity and seawater information, and including , as first output data, a first reactor water radioactive metal corrosion product concentration related to a first reactor water radioactive metal, which is an actual value, and the discharge water information, and that constructs a verification index model by training a machine learning model with training data including, as input data, the plant state quantity and the seawater information, and including, as second output data, a second reactor water radioactive metal corrosion product concentration related to a second reactor water radioactive metal different from the first reactor water radioactive metal, which is an actual value, and the discharge water information.
A predictive model construction device comprising:
(Appendix 2)
A prediction model construction device for constructing a prediction model for predicting a concentration of radioactive metal corrosion products in reactor water and seawater discharge information of a nuclear reactor in a nuclear power plant , comprising:
a learning unit which constructs the prediction model by training a machine learning model with teacher data including, as input data, a measurable plant state quantity and seawater information, and including, as first output data , a first reactor water radioactive metal corrosion product concentration related to a first reactor water radioactive metal, which is an actual value, and the discharge water information, and which constructs a verification index model by training a machine learning model with teacher data including , as input data, the plant state quantity and the seawater information, and including, as second output data, a second reactor water radioactive metal corrosion product concentration related to a second reactor water radioactive metal different from the first reactor water radioactive metal, which is an actual value, and the discharge water information ;
a prediction unit which inputs the measurable plant state quantity and the seawater information as input data into the prediction model and the verification index model, and calculates the first output data consisting of the first reactor water radioactive metal corrosion product concentration and the discharge water information, and the second output data consisting of the second reactor water radioactive metal corrosion product concentration and the discharge water information;
a verification unit that verifies the validity of the first output data based on the first and second output data;
A prediction device comprising:
(Appendix 3)
The prediction device described in (Appendix 2) is characterized in that there is a strong correlation between the first reactor water radioactive metal corrosion product concentration and the second reactor water radioactive metal corrosion product concentration .
(Appendix 4)
a consistency determination model creation unit that constructs a consistency determination model by training a machine learning model on teacher data in which the second output data is used as input data and the first output data is used as output data ;
The verification unit verifies validity of the first output data by comparing the first output data obtained by inputting the second output data to the consistency determination model with the first output data obtained by inputting the second output data to the prediction model.
The prediction device according to (Appendix 2) is characterized in that
(Appendix 5)
The prediction device described in (Appendix 4) is characterized in that the verification unit determines that the first output data is valid if a difference between the first output data obtained by inputting the second output data into the consistency judgment model and the first output data obtained by inputting the second output data into the prediction model is equal to or less than a predetermined value.
(Appendix 6)
The measurable plant state quantity includes at least one of the reactor water feedwater flow rate, the reactor water metal corrosion product concentration, the reactor water metal corrosion product concentration, the reactor water radioactive metal corrosion product concentration, the reactor water purification flow rate, the in-reactor residence time of the fuel assembly , and the electrical output.
The prediction device according to (Appendix 2) is characterized in that
(Appendix 7)
The seawater information included in the input data includes an intake temperature and a discharge temperature of the seawater, and the discharge information included in the first and second output data includes a discharge temperature of the seawater.
The prediction device according to (Appendix 2) is characterized in that
(Appendix 8)
The first reactor water radioactive metal corrosion product concentration is a reactor water radioactive metal corrosion product concentration of any one of cobalt 60, cobalt 58, and manganese 54.
The prediction device according to (Appendix 2) is characterized in that
(Appendix 9)
The second reactor water radioactive metal corrosion product concentration is any one of cobalt-58, iron, and nickel.
The prediction device according to (Appendix 2) is characterized in that

LIST OF SYMBOLS 100: Prediction device 110: Control unit 111: Learning unit 112: Prediction unit 113: Consistency determination model creation unit 114: Verification unit 120: Model storage unit 121: Prediction model storage unit 122: Verification index model storage unit 123: Consistency determination model storage unit 130: Data storage unit 131: Explanatory variable data storage unit 132: Objective variable data storage unit 133: Verification index data storage unit 140: Input/output unit 150: Seawater information database 200: Objective variable data 210: Verification index data 230: Data set 240: Data set

Claims (9)

A prediction model construction device for constructing a prediction model for predicting a concentration of radioactive metal corrosion products in reactor water and seawater discharge information of a nuclear reactor in a nuclear power plant, comprising:
a learning unit that constructs the prediction model by training a machine learning model with teacher data including, as input data, a measurable plant state quantity and seawater information, and including, as first output data, a first reactor water radioactive metal corrosion product concentration related to a first reactor water radioactive metal, which is an actual value, and the discharge water information, and that constructs a verification index model by training a machine learning model with teacher data including, as input data, the plant state quantity and the seawater information, and including, as second output data, a second reactor water radioactive metal corrosion product concentration related to a second reactor water radioactive metal different from the first reactor water radioactive metal, which is an actual value, and the discharge water information. A prediction device, comprising:
a learning unit for constructing a prediction model for predicting a reactor water radioactive metal corrosion product concentration and seawater discharge information of a reactor in a nuclear power plant , the learning unit including, as input data, a measurable plant state quantity and seawater information, and constructing the prediction model by having a machine learning model learn training data including, as first output data, a first reactor water radioactive metal corrosion product concentration related to a first reactor water radioactive metal, which is an actual value, and the discharge information, and also, the learning unit including, as input data, the plant state quantity and the seawater information, and constructing a verification index model by having a machine learning model learn training data including, as second output data, a second reactor water radioactive metal corrosion product concentration related to a second reactor water radioactive metal different from the first reactor water radioactive metal, which is an actual value, and the discharge information;
a prediction unit which inputs the measurable plant state quantity and the seawater information as input data into the prediction model and the verification index model, and calculates the first output data consisting of the first reactor water radioactive metal corrosion product concentration and the discharge water information, and the second output data consisting of the second reactor water radioactive metal corrosion product concentration and the discharge water information;
and a verification unit that verifies the validity of the first output data based on the first and second output data. 3. The prediction device according to claim 2, wherein there is a strong correlation between the first reactor water radioactive metal corrosion product concentration and the second reactor water radioactive metal corrosion product concentration. a consistency determination model creation unit that constructs a consistency determination model by training a machine learning model on teacher data in which the second output data is used as input data and the first output data is used as output data;
The prediction device according to claim 2, characterized in that the verification unit verifies the validity of the first output data by comparing the first output data obtained by inputting the second output data into the consistency judgment model with the first output data obtained by the prediction unit by inputting the measurable plant state quantities and the seawater information into the prediction model. The prediction device according to claim 4, characterized in that the verification unit determines that the first output data is valid if a difference between the first output data obtained by inputting the second output data to the consistency judgment model and the first output data obtained by the prediction unit by inputting the measurable plant state quantities and the seawater information to the prediction model is equal to or less than a predetermined value. 3. The prediction device according to claim 2, wherein the measurable plant state quantities include at least one of a feedwater flow rate of reactor water, a feedwater metal corrosion product concentration, a reactor water metal corrosion product concentration, a reactor water radioactive metal corrosion product concentration, a reactor water purification flow rate, a residence time of a fuel assembly in the reactor, and an electrical output of the reactor. The seawater information included in the input data includes an intake temperature and a discharge temperature of the seawater,
The prediction device according to claim 2 , wherein the discharge water information included in the first and second output data includes a discharge water temperature of the seawater. 3. The prediction device according to claim 2, wherein the first reactor water radioactive metal corrosion product concentration is any one of cobalt-60, cobalt-58, and manganese-54. 3. The prediction device according to claim 2, wherein the second reactor water radioactive metal corrosion product concentration is any one of cobalt-58, iron, and nickel.

Images