Learning device and neutron measurement device

Abstract

This learning device excludes erroneous detection, as a neutron, of a pulse signal originating from a pile-up phenomenon.　This learning device is for obtaining a trained model for inferring a pulse origin, which is the origin of a pulse signal, in a neutron measurement device using a pulse measurement method, and comprises: a learning data acquisition unit (61) for acquiring learning data including pulse signals outputted from a detector (1) that is an ionization chamber or a proportional counter; and a model generation unit (62) for, by using the learning data, generating the trained model for inferring the pulse origin from a pulse signal outputted from the detector (1).

Description

[Technical Field]

[0001] The present application relates to a learning device and a neutron measurement device. [Background technology]

[0002] Neutron measurement devices used to monitor reactor power detect neutrons generated in the reactor core for reactor control and protection. Neutron measurement methods for monitoring reactor power include the pulse measurement method, which counts the number of individual pulses in the detector signal; the Campbell measurement method, which calculates the root-mean-square value of the detector signal fluctuation; and the current measurement method, which calculates the detector signal as a direct current. In pressurized water reactors (PWRs), neutron detectors are installed outside the pressure vessel containing the fuel core to detect neutrons. Reactor power generally varies over 11 orders of magnitude from start-up to rated power. For measurement purposes, this is conveniently divided into three regions: the source start-up region, which is defined as the region from start-up to a relatively low reactor power range; the power region, which is near rated power; and the intermediate region, which is between the source start-up region and the power region. Neutron detectors and neutron measurement methods are used for each region.

[0003] In the radiation source activation region, a BF3 gas proportional counter tube or a fission chamber is used as the neutron detector, and the pulse measurement method is used as the neutron measurement method. In the pulse signal output from the neutron detector, the peak value of the pulse signal generated by neutrons is generally high, while the peak value of the pulse signal generated by other elements is low. Therefore, in the radiation source activation region, pulse height discrimination processing is performed using the difference in the peak value of the pulse signal to remove the influence of radiation other than neutrons, which becomes signal noise.

[0004] In the intermediate region, the current measurement method is used for the gamma-ray compensated B-10 coated ionization chamber, and the pulse measurement method and Campbell measurement method are used for the fission ionization chamber. Neutrons are detected after noise removal processing to reduce the influence of noise caused by radiation other than neutrons from the reactor core.

[0005] In the power region, gamma-ray uncompensated B-10 coated ionization chambers or fission ionization chambers are used as neutron detectors, and the current measurement method is used for neutron measurement. Note that in the power region, the influence of noise signals other than neutrons from the core becomes relatively small, so the noise removal process performed in the source activation region or intermediate region is not required.

[0006] In the pulse measurement method, noise pulse signals are removed by pulse-height discrimination processing that utilizes differences in the peak values of pulse signals. However, as the frequency of pulse signals increases with increasing reactor power, a pile-up phenomenon occurs, in which two or more pulse signals overlap, resulting in only one peak value for multiple pulse signals. In pulse-height discrimination processing that utilizes differences in the peak values of pulse signals, pulse signals due to the pile-up phenomenon are mistakenly treated as pulse signals due to neutrons. In pulse measurement methods using pulse-height discrimination processing, pulse signals due to the pile-up phenomenon, which should be removed as noise despite their frequency, are mistakenly counted as neutrons, resulting in an overestimation of reactor power. Therefore, the pile-up phenomenon was a factor that reduced the reliability of neutron detectors using the pulse measurement method.

[0007] In order to suppress erroneous counting due to the pile-up phenomenon, a method has been proposed for improving the performance of pulse height discrimination processing by predicting the occurrence rate of noise sources caused by the pile-up phenomenon in advance and setting a threshold value for pulse height discrimination processing based on pulse height differences (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Patent No. 5336934 Summary of the Invention [Problem to be solved by the invention]

[0009] In the neutron measurement device disclosed in Patent Document 1, the maximum value of a pulse signal due to the pile-up phenomenon of alpha rays, which is noise, is used for threshold determination to remove the pulse signal that is noise derived from alpha rays, but this is only a threshold determination for superimposed signals that are generated by the pile-up phenomenon of alpha rays that is expected in advance. In superimposed signals generated by the pile-up phenomenon of radiation other than neutrons, for example, the peak value of a superimposed signal generated by the pile-up phenomenon of multiple gamma rays or the peak value of a superimposed signal generated by the pile-up phenomenon of a combination of gamma rays and alpha rays is higher than the peak value of a superimposed signal generated by the pile-up phenomenon of alpha rays, so the neutron measurement device disclosed in Patent Document 1 has the problem of erroneously detecting these superimposed signals as neutrons.

[0010] The present application has been made to solve the above-mentioned problems, and aims to provide a learning device that prevents pulse signals originating from pile-up phenomena from being mistakenly detected as neutrons. [Means for solving the problem]

[0011] The learning device disclosed in the present application is a learning device that obtains a trained model that infers the pulse origin, which is the origin of a pulse signal, in a neutron measurement device that uses a pulse measurement method. , electric Acquire training data including pulse signals output from a detector such as a radio frequency chamber or proportional counter. study The device includes a data acquisition unit and a model generation unit that generates a trained model for inferring the pulse origin from the pulse signal output from the detector using the training data. The pulse origins inferred by the trained model are characterized as including neutrons, non-neutron radiation, and pile-up phenomena. [Effects of the Invention]

[0012] The learning device disclosed in the present application acquires learning data including a pulse signal output from a detector. study The device includes a data acquisition unit and a model generation unit that generates a trained model for inferring the pulse origin from the pulse signal output from the detector using the training data. The pulse origins inferred by the trained model include neutrons, non-neutron radiation, and pile-up phenomena. Therefore, it is possible to prevent a pulse signal originating from a pile-up phenomenon from being mistakenly detected as a neutron. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a block diagram showing the configuration of a neutron measurement device according to a first embodiment. [Figure 2] FIG. 2 is a diagram for explaining neutron detection by the detector in the first embodiment. [Figure 3] FIG. 2 is a diagram for explaining gamma ray detection by the detector in the first embodiment. [Figure 4] FIG. 2 is a diagram for explaining alpha ray detection by the detector in the first embodiment. [Figure 5] 4 is a diagram showing an example of a pulse signal output from a detector in the first embodiment. FIG. [Figure 6] 4 is a diagram showing an example of a pulse signal due to a pile-up phenomenon output from a detector in the first embodiment. FIG. [Figure 7] FIG. 10 is a diagram for explaining the operation of the neutron measurement device of the comparative example. [Figure 8] FIG. 1 is a diagram showing the relationship between reactor power and the number of neutron detection counts in a neutron measurement device. [Figure 9] FIG. 2 is a block diagram showing the configuration of a signal processing unit in the first embodiment. [Figure 10] FIG. 1 is a block diagram showing a configuration of a learning device according to a first embodiment. [Figure 11] FIG. 2 is a diagram for explaining an example of unsupervised learning in the model generation unit of the first embodiment. [Figure 12] 1 is a flowchart for explaining the processing of the learning device and the learned model storage unit in the first embodiment. [Figure 13] 1 is a block diagram showing the configuration of an inference device according to a first embodiment. [Figure 14]FIG. 4 is a diagram for explaining the processing of an inference unit in the first embodiment. [Figure 15] 4 is a flowchart for explaining the processing of the inference device, the counting device, and the higher-level device in the first embodiment. [Figure 16] FIG. 1 is a diagram showing the time of generation of neutrons resulting from nuclear fission in a nuclear reactor. [Figure 17] FIG. 1 is a diagram showing the generation times of radiation other than neutrons from a nuclear reactor. [Figure 18] FIG. 10 is a block diagram showing the configuration of a neutron measurement device according to a second embodiment. [Figure 19] FIG. 10 is a block diagram showing the configuration of a signal processing unit in the second embodiment. [Figure 20] FIG. 10 is a block diagram showing the configuration of a learning device according to a second embodiment. [Figure 21] FIG. 10 is a diagram illustrating an example of unsupervised learning in the model generation unit of the second embodiment. [Figure 22] 10 is a diagram for explaining the discrimination process of pulse signals due to the pile-up phenomenon in the learning device of the second embodiment. FIG. [Figure 23] FIG. 10 is a block diagram showing the configuration of an inference device according to a second embodiment. [Figure 24] 1 is a schematic diagram showing an example of a hardware configuration of a neutron measurement device according to a first embodiment and a second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, a learning device and a neutron measurement device according to embodiments of the present invention will be described in detail with reference to the drawings. Note that the same reference numerals in each drawing indicate the same or corresponding parts.

[0015] Embodiment 1 1 is a block diagram showing the configuration of a neutron measurement device 100 according to a first embodiment. The neutron measurement device 100 includes a detector 1 that detects neutrons from a nuclear reactor, a signal processing unit 2 that counts and outputs neutrons from the output of the detector 1, and a host device 3 that receives the output of the signal processing unit 2. The host device 3 is a device that processes the neutron count or count rate output from the signal processing unit 2. The host device 3 is, for example, a computer for protecting the nuclear reactor, a computer for controlling the nuclear reactor, or a display device for monitoring the nuclear reactor power output. The neutron measurement device 100 measures neutrons using a pulse measurement method.

[0016] The detector 1 is an ionization chamber or a proportional counter, which detects neutrons from the reactor and outputs a pulse signal to the signal processing unit 2. The detector 1 is equipped with a material that has a large cross section for reaction with thermal neutrons with an energy of approximately 0.025 eV, which are generated in the reactor, and is, for example, a fission ionization chamber coated on the inside with the uranium isotope U-235, which is a nuclear fuel material, or a BF3 proportional counter containing the boron isotope B-10.

[0017] Fig. 2 is a diagram for explaining neutron detection by detector 1 in embodiment 1. Detector 1 shown in Fig. 2 is a fission chamber, and detector housing 11 is filled with gas 12, and the inside of detector housing 11 is provided with a uranium coating layer 13. Electrode 14 is connected to a power supply through a resistor, detector housing 11 is grounded, and electrode 14 and detector housing 11 are insulated by insulator 15. As a result, a high voltage is applied between detector housing 11 and electrode 14. Electronic circuit 16 is connected to electrode 14, and electronic circuit 16 includes a capacitor that blocks the DC voltage of the power supply connected to electrode 14 and an amplifier that amplifies the signal.

[0018] When a neutron 40a is incident on the detector 1, a fission product 41a is generated in the uranium coating layer 13, which in turn generates a group of ions 42 and a group of electrons 43. If the detector 1 is a BF3 proportional counter, when a neutron 40a is incident on the detector 1, a nuclear reaction occurs between the incident neutron 40a and B-10, which in turn generates a fission product 41a, which in turn generates a group of ions 42 and a group of electrons 43. The generated group of ions 42 gathers in the detector housing 11, and the generated group of electrons 43 gathers on the electrode 14, causing the electronic circuit 16 to output a pulse signal.

[0019] In principle, the detector 1 cannot detect only neutrons 40a, but detects gamma rays derived from fission products generated in the reactor or gamma rays derived from activation of structures around the reactor. Fig. 3 is a diagram for explaining gamma ray detection by the detector 1 in the first embodiment. Fig. 3 shows how a gamma ray 40b is incident on the detector 1, generating an electron 41b, an ion group 42, and an electron group 43; at this time, a pulse signal is also output from the electronic circuit 16.

[0020] Furthermore, if the detector 1 is a fission chamber, U-235 in the internal uranium coating layer 13 may be nuclear converted into a radioactive isotope that undergoes alpha decay through a nuclear reaction with neutrons, generating alpha rays inside the detector 1. Fig. 4 is a diagram for explaining the detection of alpha rays by the detector 1 in the first embodiment. Fig. 4 shows how ions 42 and electrons 43 are generated by alpha rays 41c generated inside the detector 1, and a pulse signal is also output from the electronic circuit 16 at this time.

[0021] FIG. 5 is a diagram showing an example of a pulse signal output from detector 1 in embodiment 1. In FIG. 5, the left diagram shows a pulse signal 50a due to neutron detection output when a neutron is incident on detector 1, the center diagram shows a pulse signal 50b due to gamma ray detection output when a gamma ray is incident on detector 1, and the right diagram shows a pulse signal 50c due to alpha ray detection output when alpha rays are generated inside detector 1. In each diagram shown in FIG. 5, the horizontal axis represents time and the vertical axis represents voltage. When detector 1 is a fission ionization chamber, pulse signal 50a due to neutron detection originates from energy of approximately 200 MeV resulting from the nuclear fission reaction between U-235 and neutrons, and has a peak value derived from the energy of generated fission product 41a. The pulse signal 50b generated by gamma ray detection is due to electrons 41b of several tens to several hundreds of keV generated by the interaction of the gamma ray with the detector housing 11 or the interaction of the gamma ray with the uranium coating layer 13, and the peak value of the pulse signal 50b generated by gamma ray detection is lower than the peak value of the pulse signal 50a generated by neutron detection. The energy of the alpha ray 41c emitted by the alpha-decaying radioactive isotope generated by the nuclear reaction of the neutron with U-235 in the uranium coating layer 13 is approximately 5 MeV, and the peak value of the pulse signal 50c generated by alpha ray detection is approximately several percent of the peak value of the pulse signal 50a generated by neutron detection.

[0022] As a result, the pulse signal 50a resulting from neutron detection has the highest peak value, followed by the pulse signal 50c resulting from alpha ray detection, and the pulse signal 50b resulting from gamma ray detection has the lowest peak value.

[0023] When detector 1 is a BF3 proportional counter, the pulse signal generated by neutron detection has a peak value originating from a maximum energy of 2.7 MeV caused by the nuclear reaction between B-10 and neutrons. When detector 1 is a BF3 proportional counter, the pulse signal generated by gamma ray detection has the same peak value as when detector 1 is a fission chamber. When detector 1 is a BF3 proportional counter, no radioactive isotopes that undergo alpha decay are produced as with a fission chamber, so no pulse signal is output due to alpha ray detection.

[0024] Furthermore, when a charge integration circuit is used in the electronic circuit 16, the rise time of the pulse signal varies depending on the source of the pulse signal. The rise time is defined, for example, as the time it takes for the pulse signal to rise from 20% to 80% of its maximum value. The left diagram in Figure 5 shows the rise time 51a of the pulse signal 50a resulting from neutron detection. When the detector 1 is a fission chamber, the gas 12 filled inside is ionized by fission products 41a produced by the nuclear reaction between neutrons 40a and U-235, electrons 41b produced by interactions originating from gamma rays 40b, or alpha rays 41c emitted by alpha-decaying radioactive isotopes produced in the uranium coating layer 13. However, the range of ionization varies depending on the source. As a result, the rise times of the pulse signals 50a resulting from neutron detection, 50b resulting from gamma ray detection, and 50c resulting from alpha ray detection are different, and the rise times of the pulse signals also contain information specific to the source radiation.

[0025] The pulse signals output by the detector 1 include a pulse signal 50a due to neutron detection, a pulse signal 50b due to gamma ray detection, and a pulse signal 50c due to alpha ray detection, as well as a pulse signal due to a pile-up phenomenon. FIG. 6 shows an example of a pulse signal due to a pile-up phenomenon output by the detector 1. The pile-up phenomenon refers to the simultaneous or nearly simultaneous appearance of pulse signals due to neutron detection or pulse signals due to the detection of radiation other than neutrons, resulting in the output of a superimposed signal in which multiple pulse signals are superimposed. The rate of superposition increases as the reactor power increases. FIG. 6 shows an example of a pulse signal 50d due to a pile-up phenomenon in which a pulse signal 50c due to alpha ray detection and a pulse signal 50a due to neutron detection are superimposed. Note that the number of superimposed pulse signals and the timing at which the multiple pulse signals are superimposed are not uniquely determined.

[0026] The pulse signals 50d due to the pile-up phenomenon can be broadly classified into those in which only multiple pulse signals 50a due to neutron detection are superimposed, those in which the pulse signals 50a due to neutron detection and the pulse signals 50b due to gamma ray detection are superimposed, and those in which only multiple pulse signals 50b due to gamma ray detection are superimposed.If the detector 1 is a fission chamber, there is also a further type in which only multiple pulse signals 50c due to alpha ray detection are superimposed.

[0027] FIG. 7 is a diagram illustrating the operation of a comparative neutron measurement device. The comparative neutron measurement device includes the same detector 1 as the neutron measurement device 100 according to the first embodiment shown in FIG. 1. In the signal processing unit of the comparative neutron measurement device, in order to selectively count pulse signals 50a resulting from neutron detection, pulse signals whose peak values exceed a predetermined threshold value 52 are determined to originate from neutrons, as shown in FIG. 7. Here, the threshold value 52 is arbitrarily set as an adjustment factor for measurement, and is set, for example, to a value smaller than the minimum peak value of the pulse signal 50a resulting from neutron detection, as shown in the left diagram of FIG. 7. When the detector 1 is a BF3 proportional counter, the threshold value 52 may be set to the maximum peak value of a pulse signal 50b resulting from the detection of gamma rays, which are radiation other than neutrons, as shown in the right diagram of FIG. 7. When the detector 1 is a fission ionization chamber, the threshold value 52 may be set to the maximum peak value of a pulse signal 50c resulting from the detection of alpha rays.

[0028] When a pulse signal due to the pile-up phenomenon is input to the signal processing unit of the neutron measurement device of the comparative example, and the pulse signal due to the pile-up phenomenon is a superposition of multiple pulse signals 50a due to neutron detection alone, the pulse signal due to the pile-up phenomenon will be counted as a single pulse signal 50a due to neutron detection. As a result, the pulse signal 50a due to neutron detection will be counted incorrectly, resulting in a number that is smaller than the actual number of neutrons being output. Furthermore, for example, if the pulse signal due to the pile-up phenomenon is a superposition of multiple pulse signals 50b due to gamma ray detection alone and the peak value exceeds the threshold value 52, or if the pulse signal due to the pile-up phenomenon is a superposition of multiple pulse signals 50c due to alpha ray detection alone and the peak value exceeds the threshold value 52, the pulse signal due to the pile-up phenomenon will be counted as a single pulse signal 50a due to neutron detection alone, resulting in a number that is larger than the actual number of neutrons being output. In either case, the reliability of the neutron measurement device of the comparative example will be reduced.

[0029] A factor that reduces the reliability of the neutron measurement device of the comparative example is that, as the reactor power increases, not only neutrons but also other radiations increase, resulting in a pile-up phenomenon. Figure 8 shows the relationship between reactor power and the number of neutron detections in the neutron measurement device. When measurements are performed correctly in the neutron measurement device, the reactor power and the number of neutron detections are linear, as shown by line 53a. In the neutron measurement device of the comparative example, when there are more radiations other than neutrons than neutrons, the pile-up phenomenon occurs as the reactor power increases, resulting in an excessive count due to the superposition of pulse signals from the detection of radiations other than neutrons. This disrupts the linearity between the reactor power and the number of neutron counts, as shown by curve 53b. Alternatively, when there is not much radiation other than neutrons, the superposition of pulse signals from the neutron detection alone results in an insufficient count, resulting in a disruption of the linearity between the reactor power and the number of neutron counts, as shown by curve 53c.

[0030] Next, the signal processing unit 2 of the neutron measurement device 100 according to the first embodiment will be described. The signal processing unit 2 outputs the count of pulse signals due to neutrons detected within a preset measurement time. At a preset measurement interval, the signal processing unit 2 may output a counting rate obtained by dividing the count of pulse signals due to neutrons within the measurement interval by the measurement interval. FIG. 9 is a block diagram showing the configuration of the signal processing unit 2 according to the first embodiment. The signal processing unit 2 includes a learning device 60, a learned model storage unit 21, an inference device 70, and a counting device 22.

[0031] 10 is a block diagram showing the configuration of a learning device 60 according to embodiment 1. The learning device 60 includes a learning data acquisition unit 61 that acquires learning data including a pulse signal output from a detector 1 that is an ionization chamber or a proportional counter, and a model generation unit 62 that uses the learning data to generate a trained model for inferring the pulse origin, which is the origin of the pulse signal, from the pulse signal output from the detector 1.

[0032] The training data acquisition unit 61 acquires an analog pulse signal output from the detector 1 and outputs a digital pulse signal obtained by converting the analog pulse signal to a model generation unit 62. The time width for converting the analog signal to a digital signal may be a time width that includes the entire acquired pulse signal, such as a time width that includes the time from when the acquired pulse signal rises from zero to when it returns to zero. The time resolution for converting the analog signal to a digital signal may be sufficient for the model generation unit 62 to generate a trained model. The pulse signal output from the detector 1 has characteristics depending on the origin of the pulse signal. The training data acquisition unit 61 acquires training data including pulse signals generated by neutron detection, i.e., pulse signals originating from neutrons, pulse signals generated by gamma ray detection, i.e., pulse signals originating from radiation other than neutrons, and pulse signals generated by pile-up phenomena, i.e., pulse signals originating from pile-up phenomena. If the detector 1 is a fission chamber, the training data acquisition unit 61 also acquires training data including pulse signals generated by alpha ray detection.

[0033] The model generation unit 62 uses the learning data output from the learning data acquisition unit 61 to generate a trained model for inferring the pulse origin, which is the origin of a pulse signal, from the pulse signal output from the detector 1. The learning algorithm used by the model generation unit 62 can be a known algorithm such as supervised learning, unsupervised learning, or reinforcement learning. As an example, a case where the K-means method, which is unsupervised learning, is applied will be described. Unsupervised learning is a method of learning features in learning data by providing the learning device with learning data that does not include labels, which are the results.

[0034] FIG. 11 is a diagram illustrating an example of unsupervised learning in the model generation unit 62 of the first embodiment. The trained model in the first embodiment infers from the pulse signal output from the detector 1 whether the pulse origin of the pulse signal is a neutron, radiation other than neutrons, or a pile-up phenomenon. The model generation unit 62 learns the origin of the pulse signal by so-called unsupervised learning, for example, in accordance with a grouping method using the K-means method. The K-means method is a non-hierarchical clustering algorithm, and is a method of classifying a given number of clusters into k clusters using the cluster mean.

[0035] Specifically, the K-means algorithm is processed as follows: First, a cluster is randomly assigned to each data item x. Next, the center Vj of each cluster is calculated based on the assigned data. Next, the distance between each x and each vj is calculated, and x is reassigned to the cluster with the closest center. If the cluster assignments for all x remain unchanged through the above process, or if the amount of change falls below a predetermined threshold, it is determined that convergence has occurred and the process is terminated. The learning device 60 of the first embodiment learns the pulse origin, which is the origin of the pulse signal, through so-called unsupervised learning in accordance with learning data created based on pulse signals acquired by the learning data acquisition unit 61. For example, as shown in FIG. 11, the learning device 60 of the first embodiment learns the pulse origin, which is the origin of the pulse signal, through so-called unsupervised learning in accordance with learning data 80 including a pulse signal 50a generated by neutron detection, a pulse signal 50b generated by gamma ray detection, and a pulse signal 50d generated by a pile-up phenomenon, and classifies the data into three groups. Furthermore, for example, simulation data of pulse signals originating from neutrons may be created from numerical calculations that reproduce the inside of a nuclear reactor, and the group assigned when the created simulation data of the pulse signals is input to the trained model 81 may be referred to as group 82 originating from neutrons. Furthermore, simulation data of pulse signals originating from gamma rays may be created from numerical calculations that reproduce the inside of a nuclear reactor, and the group assigned when the created simulation data of the pulse signals is input to the trained model 81 may be referred to as group 83 originating from radiation other than neutrons. Furthermore, simulation data of pulse signals originating from pile-up phenomena may be created from numerical calculations that reproduce the inside of a nuclear reactor, and the group assigned when the created simulation data of the pulse signals is input to the trained model 81 may be referred to as group 84 originating from pile-up phenomena. The model generation unit 62 generates and outputs trained models by performing the above-mentioned learning. The trained model storage unit 21 stores the trained models output from the model generation unit 62.

[0036] Fig. 12 is a flowchart for explaining the processing of the learning device 60 and the learned model storage unit 21 in embodiment 1. Fig. 12 is a flowchart relating to the learning processing of the learning device 60. Step S01 is a learning pulse signal acquisition step, step S02 is a learning processing step, and step S03 is a learned model storage step.

[0037] In step S01, the training data acquisition unit 61 acquires training data including a training pulse signal output from the detector 1, outputs the data to the model generation unit 62, and proceeds to step S02. In step S02, the model generation unit 62 learns the pulse origin, which is the origin of the output pulse signal, by so-called unsupervised training in accordance with the training data created based on the pulse signal, generates a trained model, outputs the generated trained model to the trained model storage unit 21, and proceeds to step S03. In step S03, the trained model storage unit 21 stores the trained model acquired from the model generation unit 62, and ends the process.

[0038] Next, an inference device 70 of the neutron measurement device 100 according to embodiment 1 will be described. Fig. 13 is a block diagram showing the configuration of the inference device 70 according to embodiment 1. The inference device 70 includes a detection data acquisition unit 71 that acquires a detection pulse signal output from a detector 1 that is an ionization chamber or a proportional counter, and an inference unit 72 that outputs a pulse origin from the pulse signal acquired by the detection data acquisition unit 71 using a trained model for inferring the pulse origin, which is the origin of the pulse signal, from the pulse signal.

[0039] The detection data acquisition unit 71 acquires an analog pulse signal output from the detector 1, converts the pulse signal into a digital signal, and outputs the resulting digital pulse signal to the inference unit 72. The time width for converting the analog signal into a digital signal may be a time width that includes the entire acquired pulse signal, such as a time width that includes the rise time to the end time of the acquired pulse signal. Furthermore, the time resolution for converting the analog signal into a digital signal may be a time resolution that is sufficient for the inference unit 72 to infer the pulse origin using the learned model stored in the learned model storage unit 21, and may be the same as the time resolution for converting the analog signal into a digital signal in the learned data acquisition unit 61.

[0040] The inference unit 72 infers the pulse origin, which is the origin of a pulse signal, using the learned model stored in the learned model storage unit 21. That is, by inputting the pulse signal acquired by the detection data acquisition unit 71 into this learned model, it is possible to infer to which cluster the input pulse signal belongs and output the inference result as the pulse origin. FIG. 14 is a diagram for explaining the processing of the inference unit 72 in the first embodiment. For example, the top diagram of FIG. 14 shows that when a pulse signal 50a resulting from neutron detection is input to the inference unit 72, the inference unit 72 assigns the signal to group 82 of the learned model 81, which has a neutron origin, and outputs "neutron" as the pulse origin of the inference result. The center diagram of FIG. 14 shows that when a pulse signal 50b resulting from gamma ray detection is input to the inference unit 72, the inference unit 72 assigns the signal to group 83 of the learned model 81, which has a radiation origin other than neutrons, and outputs "radiation other than neutrons" as the pulse origin of the inference result. The lower diagram in Figure 14 shows that when a pulse signal 50d due to a pile-up phenomenon is input to the inference unit 72, the inference unit 72 assigns it to a group 84 originating from the pile-up phenomenon of the trained model 81, and outputs "pile-up phenomenon" as the pulse origin of the inference result.

[0041] In embodiment 1, the inference unit 72 has been described as outputting the pulse origin as an output using a trained model trained by the model generation unit 62 of the neutron measurement device 100, but it may also be configured to acquire a trained model from an external source, such as another neutron measurement device, and output the pulse origin as an output based on this trained model.

[0042] 15 is a flowchart for explaining the processing of the inference device 70, the counting device 22, and the higher-level device 3 in embodiment 1. Step S11 is a detected pulse signal acquisition step, step S12 is an inference processing step, step S13 is a pulse origin output step, step S14 is a measurement time confirmation step, step S15 is a count calculation step, and step S16 is a higher-level device processing step.

[0043] In step S11, the detection data acquisition unit 71 acquires a detection pulse signal output from the detector 1 and outputs it to the inference unit 72, and the process proceeds to step S12. In step S12, the inference unit 72 inputs the pulse signal acquired from the detection data acquisition unit 71 into the learned model stored in the learned model storage unit 21 to obtain an output pulse origin, and the process proceeds to step S13. In step S13, the inference unit 72 outputs the pulse origin, which is the output of the learned model, to the counter 22, and the process proceeds to step S14. In step S14, the counter 22 checks whether a preset measurement time has elapsed. If the measurement time has elapsed, the process proceeds to step S15. If the measurement time has not elapsed, the process returns to step S11 and repeats the processes from step S11 to step S13. In step S15, the counter 22 counts the number of pulse signals whose pulse origin is a neutron received from the inference unit 72 within the measurement time, outputs the count to the host device 3, and the process proceeds to step S16. In step S16, the higher-level device 3 displays the count obtained from the counter 22, or controls the reactor based on the count obtained from the counter 22, and then ends the process.

[0044] The processing from step S11 to step S13 may be repeated, and the counting device 22 may count the number of pulse signals whose pulse origin is neutrons received from the inference unit 72 within a preset measurement time, and output the count result as a count to the host device 3. Alternatively, the processing from step S11 to step S13 may be repeated, and the counting device 22 may output, at a preset measurement interval, a value obtained by dividing the number of pulse signals whose pulse origin is neutrons within the measurement interval by the measurement interval as a counting rate. The measurement time or measurement interval may be, for example, a time period during which the neutron measurement device can respond from the detection of a signal that triggers a reactor protection operation and its transmission to the host device 3, to the insertion of control rods and the securing of core protection after determination by various calculations, and is generally several msec to several hundred msec.

[0045] The above processing makes it possible to eliminate erroneous detection of pulse signals originating from pile-up phenomena as neutrons, thereby obtaining the count of pulse signals whose pulse origin is neutrons and eliminating erroneous counts due to pulse signals originating from pile-up phenomena. Therefore, reactor power can be monitored with high accuracy without overestimation or underestimation.

[0046] Although the above description concerns a case where unsupervised learning is applied to the learning algorithm used by the model generation unit 62 or the inference unit 72, this is not limiting. In addition to unsupervised learning, reinforcement learning, supervised learning, or semi-supervised learning can also be applied as the learning algorithm. Furthermore, the learning algorithm used by the learning device 60 can be deep learning, which learns to extract features themselves, or other known methods. When implementing unsupervised learning, the method is not limited to the non-hierarchical clustering using the K-means method described above, and any other known method capable of clustering can be used. For example, hierarchical clustering such as a shortest distance method may be used. When implementing supervised learning, for example, simulation data can be generated from numerical calculations that reproduce the inside of a nuclear reactor, such as pulse signals due to neutron detection, pulse signals due to gamma ray detection, pulse signals due to alpha ray detection, and pulse signals due to pileup phenomena, and these pulse signals can be used as learning data whose pulse origins are known.

[0047] Furthermore, in the first embodiment, the detector 1 may output a digital pulse signal, and the learning device 60 and the inference device 70 may be connected to the detector 1 or the counting device 22 via a network. The learning device 60 and the inference device 70 may also be built into the neutron measurement device 100. Furthermore, the detector 1 may output a digital pulse signal, and the learning device 60 and the inference device 70 may exist on a cloud server. When the learning device 60 and the inference device 70 are connected to the detector 1 or the counting device 22 via a network, or when the learning device 60 and the inference device 70 exist on a cloud server, the learning data acquisition unit 61 and the detection data acquisition unit 71 acquire the digital pulse signal.

[0048] The model generation unit 62 may also learn the origin of generating an input pulse signal according to learning data created for multiple neutron measurement devices. The model generation unit 62 may acquire learning data from multiple neutron measurement devices used in the same area, or may learn the origin of generating an input pulse signal using learning data collected from multiple neutron measurement devices operating independently in different areas. Alternatively, the model generation unit 62 may learn the origin of generating an input pulse signal according to learning data created from pulse signals of neutron measurement devices obtained through numerical calculations that reproduce the inside of a nuclear reactor. It is also possible to add or remove neutron measurement devices that collect learning data during the process. Furthermore, the learning device 60 that has learned the origin of generating an input pulse signal for a certain neutron measurement device may be applied to the origin of generating a different input pulse signal, and the origin of generating an input pulse signal for the different input pulse signal may be re-learned and updated.

[0049] As described above, the learning device 60 according to embodiment 1 is a learning device 60 that obtains a trained model for inferring the pulse origin, which is the origin of a pulse signal, in a neutron measurement device 100 that uses a pulse measurement method, and is equipped with a training data acquisition unit 61 that acquires training data including a pulse signal output from a detector 1 that is an ionization chamber or a proportional counter, and a model generation unit 62 that uses the training data to generate a trained model for inferring the pulse origin from the pulse signal output from the detector 1, thereby making it possible to prevent pulse signals originating from pile-up phenomena from being mistakenly detected as neutrons.

[0050] Embodiment 2 First, we will explain the generation time of the pulse signal output from detector 1. Figure 16 shows the generation time of neutrons resulting from nuclear fission in a nuclear reactor. In Figure 16, the horizontal axis represents time, and the shaded areas represent neutrons generated at the times indicated on the horizontal axis. As shown in Figure 16, neutrons are generated discretely within a nuclear reactor due to nuclear fission reactions. In an operating nuclear reactor, neutrons fly around, and a detector captures and measures some of these neutrons. Considering the absence of nuclear fission, the generation and movement of neutrons within the reactor occurs randomly, resulting in independent measurement of neutrons at any given time. On the other hand, considering the presence of nuclear fission, a group of neutrons belonging to the same generation, generated from a single neutron through a nuclear fission chain reaction, flies around within the reactor and is generated discretely. In other words, neutrons can be measured consecutively when a group is generated, but not when a group is not generated. Figure 17 shows the generation time of radiation other than neutrons in a nuclear reactor. In FIG. 17, the horizontal axis represents time, and the shaded areas indicate that radiation other than neutrons is being generated at the times indicated on the horizontal axis. As shown in FIG. 17, radiation other than neutrons is constantly being generated from a nuclear reactor. From what is shown in FIGS. 16 and 17, it can be seen that the pulse signals due to the pile-up phenomenon output from detector 1 at the times indicated by the shaded areas in FIG. 16 contain many pulse signals due to the pile-up phenomenon that include neutrons. Therefore, the neutron measurement device according to embodiment 2 uses a trained model for inferring the pulse origin from the pulse signal and the acquisition time, which is the time when the pulse signal is acquired from detector 1.

[0051] Fig. 18 is a block diagram showing the configuration of a neutron measurement apparatus 100a according to embodiment 2. When comparing the neutron measurement apparatus 100a according to embodiment 2 shown in Fig. 18 with the neutron measurement apparatus 100 according to embodiment 1 shown in Fig. 1, the signal processing unit 2 has been replaced with a signal processing unit 2a. Other configurations of the neutron measurement apparatus 100a according to embodiment 2 are the same as those of the neutron measurement apparatus 100 according to embodiment 1.

[0052] Fig. 19 is a block diagram showing the configuration of a signal processing unit 2a according to embodiment 2. Comparing signal processing unit 2a according to embodiment 2 shown in Fig. 19 with signal processing unit 2 according to embodiment 1 shown in Fig. 9, learning device 60 has become learning device 60a, learned model storage unit 21 has become learned model storage unit 21a, inference device 70 has become inference device 70a, and counting device 22 has become counting device 22a.

[0053] Fig. 20 is a block diagram showing the configuration of a learning device 60a according to embodiment 2. Comparing learning device 60a according to embodiment 2 shown in Fig. 20 with learning device 60 according to embodiment 1 shown in Fig. 10, learning data acquisition unit 61 has become learning data acquisition unit 61a, and model generation unit 62 has become model generation unit 62a.

[0054] The learning data acquiring unit 61a acquires a pulse signal of the analog signal output from the detector 1. Furthermore, the learning data acquiring unit 61a acquires information on the acquisition time, which is the time when the learning data acquiring unit 61a acquired the pulse signal from the detector 1. The learning data acquiring unit 61a may acquire the acquisition time information from an external clock at the timing when the learning data acquiring unit 61a acquired the pulse signal from the detector 1, or may be provided with an internal clock and acquire the acquisition time information. The learning data acquiring unit 61a outputs a pair of the acquired pulse signal and the acquisition time to the model generating unit 62a as learning data.

[0055] The model generation unit 62a generates a trained model for inferring the pulse origin, which is the origin of a pulse signal, from the pulse signal and its acquisition time using training data that is a pair of the pulse signal output from the training data acquisition unit 61a and the acquisition time. FIG. 21 is a diagram for explaining an example of unsupervised learning in the model generation unit 62a of the second embodiment. The trained model 81a in the second embodiment infers, from the pulse signal output from the detector 1 and its acquisition time, whether the pulse origin, which is the origin of the pulse signal, is a neutron, radiation other than neutrons, a pile-up phenomenon including neutrons, or a pile-up phenomenon not including neutrons. The model generation unit 62a learns the origin of the pulse signal by so-called unsupervised learning, for example, in accordance with a grouping method using the K-means method.

[0056] The learning device 60a of the second embodiment learns the pulse origin, which is the origin of the pulse signal, by so-called unsupervised learning in accordance with learning data created based on the pulse signal and the acquisition time acquired by the learning data acquiring unit 61a. For example, as shown in Fig. 21 , the learning device 60a of the second embodiment learns the pulse origin, which is the origin of the pulse signal, by so-called unsupervised learning in accordance with learning data 80a including a pulse signal 56a due to neutron detection including the acquisition time, a pulse signal 56b due to gamma ray detection including the acquisition time, and a pulse signal 56d due to a pile-up phenomenon including the acquisition time, and divides the signals into four groups: a group 82 having a neutron origin, a group 83 having a radiation origin other than neutrons, a group 85 having a pile-up phenomenon origin including neutrons, and a group 86 having a pile-up phenomenon origin not including neutrons. In the learning device 60a of the second embodiment, by using information on the acquisition time, the group 84 originating from the pile-up phenomenon is divided into two groups: the group 85 originating from the pile-up phenomenon including neutrons, and the group 86 originating from the pile-up phenomenon not including neutrons.

[0057] FIG. 22 is a diagram illustrating the discrimination process of a pulse signal due to a pile-up phenomenon in the learning device 60a of the second embodiment. FIG. 22 shows the output signal from the detector 1 of FIG. 18, with the horizontal axis representing time and the vertical axis representing voltage. FIG. 22 shows that the detector 1 outputs signals in the following order: a pulse signal 50a due to neutron detection, a pulse signal due to a pile-up phenomenon, a pulse signal 50c due to alpha ray detection, a pulse signal due to a pile-up phenomenon, a pulse signal 50c due to alpha ray detection, a pulse signal due to a pile-up phenomenon, a pulse signal 50a due to neutron detection, a pulse signal due to a pile-up phenomenon, and a pulse signal 50b due to gamma ray detection. The learning data acquisition unit 61a acquires each pulse signal shown in FIG. 22, and acquires the rising time of the pulse signal as the acquisition time, which is the time when the pulse signal was acquired, and outputs a pair of the acquired pulse signal and the acquisition time as learning data.

[0058] For discrimination of pulse signals due to pile-up phenomena in the model generating unit 62a, a neutron measurement time 55 is set in advance, and a pulse signal due to a pile-up phenomenon detected between the rising time of pulse signal 50a due to neutron detection and neutron measurement time 55 is taken as pulse signal 50e due to a pile-up phenomenon including neutrons, and a group including pulse signal 50e due to a pile-up phenomenon including neutrons is taken as group 85 having origins from a pile-up phenomenon including neutrons. Furthermore, a pulse signal due to a pile-up phenomenon detected at a time other than between the rising time of pulse signal 50a due to neutron detection and neutron measurement time 55 is taken as pulse signal 50f due to a pile-up phenomenon not including neutrons, and a group including pulse signal 50f due to a pile-up phenomenon not including neutrons is taken as group 86 having origins from a pile-up phenomenon not including neutrons.

[0059] The neutron measurement time 55 is, for example, the neutron lifetime. The neutron lifetime is the average time that neutrons remain in the reactor after they are generated by nuclear fission inside the reactor until they are annihilated by absorption or leakage outside the reactor. The neutron lifetime is calculated by dividing the total number of neutrons in the reactor by the number of neutrons annihilated per unit time. Note that some of the annihilated neutrons are absorbed by the fuel, causing nuclear fission and generating neutrons of the next generation. The neutron lifetime, which is the time interval between nuclear reactions of one generation and the next generation, is 0.05 to 0.07 seconds in a typical light water reactor. Therefore, the neutron measurement time 55 is set to a value of 0.05 to 0.07 seconds. Since the majority of pile-up phenomena that occur between the rise time of pulse signal 50a due to neutron detection and neutron measurement time 55 are caused by neutrons, in learning device 60a of embodiment 2, a pulse signal due to a pile-up phenomenon detected between the rise time of pulse signal 50a due to neutron detection and neutron measurement time 55 is considered to be pulse signal 50e due to a pile-up phenomenon including neutrons, and is included in group 85 of phenomena originating from pile-up phenomena including neutrons. Furthermore, since the majority of pile-up phenomena that occur at times other than the rise time of pulse signal 50a due to neutron detection and neutron measurement time 55 are not caused by neutrons, in learning device 60a of embodiment 2, a pulse signal due to a pile-up phenomenon that is detected at a time other than the rise time of pulse signal 50a due to neutron detection and neutron measurement time 55 is considered to be pulse signal 50f due to a pile-up phenomenon not including neutrons, and is included in group 86 of phenomena originating from pile-up phenomena not including neutrons.

[0060] Fig. 23 is a block diagram showing the configuration of an inference device 70a according to embodiment 2. Comparing inference device 70a according to embodiment 2 shown in Fig. 23 with inference device 70 according to embodiment 1 shown in Fig. 13, detection data acquisition unit 71 has become detection data acquisition unit 71a, and inference unit 72 has become inference unit 72a.

[0061] The detection data acquiring unit 71a acquires the pulse signal of the analog signal output from the detector 1, and also acquires information on the acquisition time, which is the time when the detection data acquiring unit 71a acquired the pulse signal from the detector 1. The detection data acquiring unit 71a may acquire the acquisition time information from an external clock at the timing when the detection data acquiring unit 71a acquired the pulse signal from the detector 1, or may be provided with an internal clock and acquire the acquisition time information. The detection data acquiring unit 71a outputs a pair of the acquired pulse signal and the acquisition time to the inference unit 72a.

[0062] The inference unit 72a infers the pulse origin, which is the origin of the pulse signal, using the learned model stored in the learned model storage unit 21a. The pulse origin output from the inference unit 72a is either a neutron, radiation other than neutrons, a pile-up phenomenon including neutrons, or a pile-up phenomenon not including neutrons. The counter 22a counts the number of pulse signals whose pulse origin is a neutron or a pile-up phenomenon including neutrons, received from the inference unit 72a within a predetermined measurement time, and outputs the result as a count to the host device 3. The host device 3 displays the count obtained from the counter 22a, or controls the nuclear reactor based on the count obtained from the counter 22a.

[0063] As described above, the learning device 60a according to embodiment 2 is a learning device 60a that obtains a trained model for inferring the pulse origin, which is the origin of a pulse signal, in a neutron measurement device 100a that uses a pulse measurement method, and is equipped with a training data acquisition unit 61a that acquires training data including a pulse signal output from a detector 1 that is an ionization chamber or a proportional counter, and a model generation unit 62a that uses the training data to generate a trained model for inferring the pulse origin from the pulse signal output from the detector 1. The training data acquisition unit 61a acquires training data including time information that is the time when the pulse signal output from the detector 1 was acquired, and the model generation unit 62a uses the training data to generate a trained model for inferring the pulse origin from the pulse signal and the time information.Therefore, it is possible to count both pulse signals due to neutrons and pulse signals due to pile-up phenomena that include neutrons, and to perform highly accurate neutron detection.

[0064] FIG. 24 is a schematic diagram showing an example of the hardware configuration of the neutron measurement apparatus according to the first and second embodiments. The model generation units 62 and 62a, the inference units 72 and 72a, and the counter 22 are implemented by a processor 201, such as a CPU (Central Processing Unit), that executes programs stored in a memory 202. The learning data acquisition units 61 and 61a and the detection data acquisition units 71 and 71a are implemented by an A / D converter 203 and the processor 201, such as a CPU (Central Processing Unit), that executes programs stored in the memory 202. The memory 202 is also used as a temporary storage device for each process executed by the processor 201. Furthermore, multiple processing circuits may cooperate to execute the above functions. Furthermore, the above functions may be implemented by dedicated hardware. When the above functions are implemented by dedicated hardware, the dedicated hardware may be, for example, a single circuit, a composite circuit, a programmed processor, an ASIC, an FPGA, or a combination thereof. The above functions may be implemented by a combination of dedicated hardware and software, or a combination of dedicated hardware and firmware. The memory 202 is, for example, a non-volatile or volatile semiconductor memory such as RAM, ROM, flash memory, or EPROM, a magnetic disk, an optical disk, or a combination of these. The trained model storage units 21 and 21a are realized by the memory 202. The detector 1 is connected to an A / D converter 203, and the processor 201, memory 202, A / D converter 203, and upper device 3 are connected to one another via a bus.

[0065] Although the present application describes various exemplary embodiments, the various features, aspects, and functions described in one or more embodiments are not limited to application to a particular embodiment, but may be applied to the embodiments alone or in various combinations. Therefore, countless variations not illustrated are conceivable within the scope of the technology disclosed in this application, including, for example, cases where at least one component is modified, added, or omitted, and cases where at least one component is extracted and combined with a component of another embodiment. [Explanation of symbols]

[0066] 1 detector, 2, 2a signal processing unit, 3 upper device, 11 detector housing, 12 gas, 13 uranium coating layer, 14 electrode, 15 insulator, 16 electronic circuit, 21, 21a learned model memory unit, 22, 22a counter, 40a neutron, 40b gamma ray, 41a fission product, 41b electron, 41c alpha ray, 42 ion group, 43 electron group, 50a pulse signal due to neutron detection, 50b pulse signal due to gamma ray detection, 50c pulse signal due to alpha ray detection, 50d pulse signal due to pile-up phenomenon, 50e pulse signal due to pile-up phenomenon including neutron, 50f pulse signal due to pile-up phenomenon not including neutron, 51a rise time, 52 threshold, 53a straight line, 53b, 53c curve, 55 neutron measurement time, 56a a pulse signal due to neutron detection including the acquisition time, 56b a pulse signal due to gamma ray detection including the acquisition time, 56d a pulse signal due to a pile-up phenomenon including the acquisition time, 60, 60a learning device, 61, 61a learning data acquisition unit, 62, 62a model generation unit, 70, 70a inference device, 71, 71a detection data acquisition unit, 72, 72a inference unit, 80, 80a learning data, 81, 81a trained model, 82 a group originating from neutrons, 83 a group originating from radiation other than neutrons, 84 a group originating from a pile-up phenomenon, 85 a group originating from a pile-up phenomenon including neutrons, 86 a group originating from a pile-up phenomenon not including neutrons, 100, 100a neutron measurement device, 201 a processor, 202 a memory, 203 an A / D converter.

Claims

1. A learning device for obtaining a trained model for inferring a pulse origin, which is the origin of a pulse signal, in a neutron measurement device using a pulse measurement method, comprising: a learning data acquisition unit that acquires learning data including the pulse signal output from a detector that is an ionization chamber or a proportional counter; a model generation unit that generates the trained model for inferring the pulse origin from the pulse signal output from the detector using the training data, A learning device characterized in that the pulse origin inferred by the trained model includes neutrons, radiation other than neutrons, and pile-up phenomena.

2. A learning device for obtaining a trained model for inferring the pulse origin, which is the origin of a pulse signal, in a neutron measurement device using a pulse measurement method, comprising: a learning data acquisition unit that acquires learning data including the pulse signal output from a detector that is an ionization chamber or a proportional counter; a model generation unit that generates the trained model for inferring the pulse origin from the pulse signal output from the detector using the training data, the learning data acquisition unit acquires the learning data including time information indicating a time when the pulse signal output from the detector was acquired; The model generation unit uses the learning data to generate the trained model for inferring the pulse origin from the pulse signal and the time information.

3. A neutron measurement device using a pulse measurement method, a detector which is an ionization chamber or a proportional counter; a detection data acquisition unit that acquires a pulse signal output from the detector; an inference unit that outputs a pulse origin from the pulse signal acquired by the detection data acquisition unit, using a trained model for inferring a pulse origin that is an origin of the pulse signal from the pulse signal; a counter that outputs a count of the pulse signal whose pulse origin is a neutron; A neutron measurement device, characterized in that the pulse origin inferred by the trained model includes neutrons, radiation other than neutrons, and pile-up phenomena.

4. A neutron measurement device using a pulse measurement method, a detector which is an ionization chamber or a proportional counter; a detection data acquisition unit that acquires a pulse signal output from the detector; an inference unit that outputs a pulse origin from the pulse signal acquired by the detection data acquisition unit, using a trained model for inferring a pulse origin that is an origin of the pulse signal from the pulse signal; a counter that outputs a count of the pulse signal whose pulse origin is a neutron; the detection data acquisition unit acquires an acquisition time, which is a time when the pulse signal output from the detector was acquired; the inference unit outputs the pulse origin from the pulse signal and the acquisition time acquired by the detection data acquisition unit using a trained model for inferring the pulse origin from the pulse signal and the acquisition time; The neutron measuring device is characterized in that the counter outputs the count of the pulse signal whose origin is a neutron or a pile-up phenomenon including a neutron.

Images