JPH07260985A - Nuclea reactor core designing support device - Google Patents

Abstract

PURPOSE:To obtain the inferred values of the output distribution and the multiplication factor distribution of the whole body of a reactor core, and the fuel loading pattern, more simply and in a shorter time compared to the calculation depending on a minute physical model. CONSTITUTION:After finding the two-dimensional infinite multification factor distribution in the diameter direction, and the primary infinite multification factor distribution in the axial direction, from the three-dimensional infinite multification factor distribution, the output distribution in the diameter direction is inferred by the neuro network in the two-dimension output distribution inferring means 104 in the diameter direction whose learning has been finished, and on the other hand, the output distribution in the axial direction is inferred by the neuro network in the primary output distribution inferring means in the axial direction 103 whose learning has been finished, and by composing the two inferred results, the inferred value of the three-dimensional output distribution is obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a reactor core design support device for preparing a fuel refueling plan for a nuclear reactor,
The present invention relates to a reactor core design support device that realizes easy and short-time estimation of the power distribution and the infinite multiplication factor distribution of the entire core.

[0002]

2. Description of the Related Art In a refueling plan for a reactor core design,
In order to satisfy various operational limits and target burnup of the nuclear power plant that is the design target, many calculations based on detailed physical models are performed. Further, in Japanese Patent Application Laid-Open No. 4-113297, the core of a nuclear reactor or a part thereof is divided into a number of spatial regions, and various quantities representing the physical properties of each region are input to input the core of the nuclear reactor or a part thereof. A reactor core performance estimating device for outputting various quantities representing core performance is described.

[0003]

The calculation based on a detailed physical model represented by neutron diffusion calculation and fuel combustion calculation usually requires not only a long time for calculation by a super computer but also preparation for execution of calculation. The process usually requires a long time. Further, as the performance of core fuels has increased and the number of plants has increased in recent years, the amount of calculations and other design work is increasing, and there is an urgent need to improve design efficiency. With the recent progress of downsizing, we will contribute to the improvement of core design efficiency by simply finding the approximate amount of these calculations in advance using a workstation or personal computer before entering detailed calculations. Is extremely important.

[0004] Here, Japanese Patent Application Laid-Open No. 4-113297 does not show an embodiment for a large-scale problem such as estimating the power distribution of the entire core. Learning in a large-scale neural network usually takes a lot of time until it converges, otherwise it will diverge, and some kind of ingenuity is required.

An object of the present invention is to quickly and simply calculate an output distribution of the entire core, which is an amount representing core performance, from an infinite multiplication factor distribution of each fuel assembly, which is an amount representing physical properties of a nuclear reactor. To estimate.

Another object of the present invention is to obtain an infinite multiplication factor distribution of each fuel assembly of the entire core, which is an amount representing the physical properties of a nuclear reactor, from the power distribution of the entire core which is an amount representing core performance. It is to estimate easily in a short time.

[0007]

The above-mentioned object is to obtain a radial two-dimensional infinite multiplication factor distribution and an axial one-dimensional infinite multiplication factor distribution from a three-dimensional infinite multiplication factor distribution. A neural circuit that obtains a radial two-dimensional output distribution and an axial one-dimensional output distribution from the output distribution, and then uses the radial two-dimensional infinite multiplication factor distribution as an input signal and the radial two-dimensional output distribution as a teacher signal. A net is constructed to determine the connection weight between each nerve cell. At the time of estimation, the radial two-dimensional infinite multiplication factor distribution is used as the input signal, and the output signal of the neural network is used as the radial two-dimensional output distribution estimated value. After obtaining the radial two-dimensional output distribution and the axial one-dimensional output distribution from the three-dimensional output distribution by the detailed calculation, the axial one-dimensional infinite multiplication distribution is used as the input signal, and the axial one-dimensional output distribution is the teacher signal. Neural network Build and determine the coupling load between each nerve cell, and at the time of estimation, the axial direction one-dimensional infinite multiplication factor distribution as the input signal and the neural network output signal as the axial direction one-dimensional output distribution estimated value. A three-dimensional output distribution estimating means for obtaining a three-dimensional output distribution estimated value by synthesizing values and a radial two-dimensional output distribution and an axial one-dimensional output distribution are obtained from the three-dimensional output distribution. From infinite multiplication factor distribution, radial two-dimensional infinite multiplication factor distribution and axial direction 1
After obtaining the dimensional infinite multiplication factor distribution, the neural network which uses the radial direction two-dimensional infinite multiplication factor distribution as a teacher signal as the input signal and the radial two-dimensional output distribution obtained from the three-dimensional output distribution by the detailed calculation Constructed to determine the connection weight between each nerve cell, and at the time of estimation, the target radial two-dimensional output distribution is used as an input signal, and the neural network output signal is used as a radial two-dimensional infinite multiplication factor distribution estimated value. Occasionally, after obtaining the radial two-dimensional infinite multiplication factor distribution and the axial one-dimensional infinite multiplication factor distribution from the three-dimensional infinite multiplication factor distribution, the axial one-dimensional output distribution obtained from the detailed calculation three-dimensional output distribution is input. In addition to the signal, a neural network that uses the axial one-dimensional infinite multiplication factor distribution as the teacher signal is constructed to determine the coupling load between each nerve cell, and at the time of estimation, the target axial one-dimensional output distribution is input signal. As a nerve A three-dimensional infinite multiplication factor distribution estimating means for obtaining a three-dimensional infinite multiplication factor distribution estimated value by combining the above two estimated values with the output signal of the road network as an axial one-dimensional infinite multiplication factor distribution estimated value, and learning. At times, the radial output value and the loading distance of the target fuel assembly when paying attention to one fuel assembly among the fuel assemblies forming the reactor,
And a radial output value and a loading distance of each fuel assembly near the target fuel assembly as input signals, and a neural network that uses the fuel loading pattern of the target fuel assembly as a teacher signal is constructed to construct a nuclear reactor. All or a part of the fuel assemblies that compose the fuel assembly is used as the target fuel assembly to determine the coupling load between nerve cells, and at the time of estimation, one of the fuel assemblies that compose the reactor was focused on. When the radial output value and the loading distance of the target fuel assembly and the radial output values and the loading distance of the fuel assemblies near the target fuel assembly are input signals, all the fuel assemblies that make up the reactor are The target fuel assembly is achieved by the fuel loading pattern estimation means for obtaining the fuel loading pattern estimation value of all the fuel assemblies forming the reactor.

[0008]

According to the three-dimensional output distribution estimating means of the present invention, 3
When the dimension infinite multiplication factor distribution is given, the radial two-dimensional output distribution estimation means and the axial one-dimensional output distribution estimation means obtain the radial two-dimensional output distribution estimated value and the axial one-dimensional output distribution estimated value. By combining these estimated values, a three-dimensional output distribution estimated value can be obtained. According to the three-dimensional infinite multiplication factor distribution estimating means of the present invention, when the three-dimensional output distribution is given, the radial two-dimensional infinite multiplication factor distribution estimating means and the axial one-dimensional infinite multiplication factor distribution estimating means A radial two-dimensional infinite multiplication factor distribution estimated value and an axial one-dimensional infinite multiplication factor distribution estimated value are obtained, and these estimated values are combined to obtain a three-dimensional infinite multiplication factor distribution estimated value. Further, according to the fuel pattern estimating means of the present invention, when the three-dimensional output distribution or the radial two-dimensional output distribution is given, the fuel loading pattern estimated value can be obtained.

Since the acquisition process of the three-dimensional output distribution estimated value, the three-dimensional infinite multiplication factor estimated value, and the fuel loading pattern estimated value is all performed by the sum-of-products operation, the time required is significantly longer than the calculation based on the detailed physical model. Can save

Further, by dividing the three-dimensional distribution into a two-dimensional distribution and a one-dimensional distribution and simplifying the problem, the learning load in the neural network can be reduced, and compared with the learning of the three-dimensional distribution. The convergence speed can be expected to improve.

Further, the radial direction two-dimensional output distribution estimating means which constitutes the three-dimensional output distribution estimating means, the axial direction one-dimensional output distribution estimating means, and the radial direction two-dimensional infinite multiplication which constitutes the three-dimensional infinite multiplication factor distribution estimating means. Since the magnification distribution estimation means, the axial one-dimensional infinite multiplication factor distribution estimation means, and the fuel loading pattern estimation means all estimate a desired value for each fuel assembly that constitutes the reactor, the core system (core size ) Has the advantage that a neural network that does not depend on can be constructed.

[0012]

Embodiments of the present invention will be described below in detail with reference to the accompanying drawings. FIG. 1 shows an embodiment of the reactor core design support apparatus of the present invention. The data input apparatus 101,
Data storage means 102, three-dimensional output distribution estimation means 10
It comprises a three-dimensional infinite multiplication factor distribution estimating means 106, a fuel loading pattern estimating means 109, an estimation result storage means 110, a display means 111, and a display device 112.
The three-dimensional output distribution estimating means 103 includes a radial two-dimensional output distribution estimating means 104 and an axial one-dimensional output distribution estimating means 10.
It consists of 5. Furthermore, the three-dimensional infinite multiplication factor distribution estimation means 1
Reference numeral 06 includes a radial two-dimensional infinite multiplication factor distribution estimating means 107 and an axial one-dimensional infinite multiplication factor distribution estimating means 108.

FIG. 2 is a flowchart showing the operation procedure of the three-dimensional output distribution estimating means 103 of the present invention during learning. First, one core data to be learned is selected (step 201), and the average value of the axial infinite multiplication factor distribution of each fuel assembly is set as the radial infinite multiplication factor of the fuel assembly. Set the axial infinite multiplication factor distribution of each fuel assembly when the infinite multiplication factor distribution is normalized so that the core average value is 1.0 as the axial one-dimensional infinite multiplication factor distribution of each fuel assembly. The work is performed for all fuel assemblies (step 202). Next, the average value of the axial power distribution of each fuel assembly is set as the radial power distribution of the fuel assembly, and the three-dimensional output distribution is normalized so that the core average value becomes 1.0. The operation of setting the axial output distribution of the fuel assemblies as the axial one-dimensional output distribution of each fuel assembly is executed for all the fuel assemblies (step 203). In step 204, it is determined whether all core data to be learned have been processed, and if there is core data remaining, the process returns to step 201. If all core data has been processed, the radial two-dimensional output distribution is calculated. Learning by the estimating means 104 and the one-dimensional axial direction power distribution estimating means 105 is executed.

FIG. 3 is a flow chart showing a detailed procedure for learning by the radial direction two-dimensional output distribution estimating means 104 of the present invention. First, all the connection weights of the neural network in the radial direction two-dimensional output distribution estimation means 104 are initialized to random values between -1.0 and 1.0 (step 301). Next, one of the core data to be learned is randomly selected, and further one of the fuel assemblies is randomly selected (step 302). Next, using the selected fuel assemblies, that is, the radial infinite multiplication factor and loading distance of the target fuel assembly, and the infinite multiplication factors and loading distances of the fuel assemblies in the vicinity of the target fuel assembly as input signals, the neural network is input. (Step 303). Next, the output value of each nerve cell is calculated, the error between the output signal of the neural network and the output value of the target fuel assembly which is the teacher signal is calculated (step 305), and this error is preset as the end reference value. It is determined whether the value has become smaller than the specified value. As a result, if the criterion is not satisfied, the process proceeds to step 307, and if the criterion is satisfied, the learning is completed and the process ends (step 306). In step 307, since the error has not reached the termination criterion, each connection weight of the neural network is corrected by the error backpropagation method or the like, the process returns to step 302, and the connection weight is repeatedly adjusted until the desired accuracy is obtained. .

Next, the radial direction two-dimensional output distribution is evaluated by the neural network which has completed learning. The details are shown in FIG. Further, FIG. 5 shows the configuration of the neural network in the radial two-dimensional output distribution estimating means. First, one of the fuel assemblies is selected (step 401), and the radial infinite multiplication factor and loading distance of the target fuel assembly and the radial infinite multiplication factor and loading distance of the fuel assembly near the target fuel assembly are set. , Are set as input signals in the neural network which has completed learning, and the output value of the neural network is calculated (step 402). In step 403, it is determined whether the output values corresponding to all the fuel assemblies have been obtained. If there are any fuel assemblies remaining, the process returns to step 401, and if all the fuel assemblies have been processed, the processing ends.

Next, the axial one-dimensional output distribution estimating means 105
Perform neural network learning in. This detailed procedure is shown in FIG. First, all the connection weights of the neural network in the axial one-dimensional output distribution estimation means 105 are −1.
Initialize to a random value between 0 and 1.0 (step 601). Next, one of the core data to be learned is randomly selected, and further one of the fuel assemblies is randomly selected (step 602). Next, the selected fuel assembly, that is, the axial infinite multiplication factor distribution of the target fuel assembly, the axial average infinite multiplication factor distribution of the fuel assemblies in the vicinity of the target fuel assembly, and the average infinite multiplication factor of the target core. The distributions are set as input signals in the neural network (step 603). Next, the output value of each nerve cell is calculated, and the error between the output signal of the neural network and the axial power distribution of the target fuel assembly, which is the teacher signal, is calculated (step 605). It is determined whether the value is smaller than the set value. As a result, if the criterion is not satisfied, the process proceeds to step 607, and if the criterion is satisfied, the learning is completed and the process ends (step 606). In step 607, since the error has not reached the termination criterion, each coupling weight of the neural network is corrected by the error back propagation method or the like,
Returning to step 602, the joint load is repeatedly adjusted until the desired accuracy is obtained.

Next, the axial one-dimensional output distribution is evaluated by the neural network which has completed learning. The details are shown in FIG. FIG. 8 shows the configuration of the neural network in the axial one-dimensional output distribution estimating means. First, one of the fuel assemblies is selected (step 701), and the axial infinite multiplication factor distribution of the target fuel assembly, the axial average infinite multiplication factor distribution of the fuel assemblies in the vicinity of the target fuel assembly, and the target core The average infinite multiplication factor distribution is set as an input signal to the neural network which has completed learning, and the output value of the neural network is calculated (step 702). In step 703, it is determined whether the output values corresponding to all the fuel assemblies have been obtained,
If there are any remaining fuel assemblies, the process returns to step 701, and if all the fuel assemblies have been processed, the process ends.

Next, the operation procedure of the three-dimensional infinite multiplication factor distribution estimating means 106 of the present invention during learning will be described. The details are shown in FIG. First, one core data to be learned is selected (step 901), the average value of the output distribution of each fuel assembly is set as the radial output value of the target fuel assembly, and the three-dimensional output distribution is calculated as the core average. The value is 1.
The output distribution of each fuel assembly when normalized to 0 is set as the axial one-dimensional output distribution of each fuel assembly (step 902). Next, the average value of the infinite multiplication factor distribution of each fuel assembly was set as the radial direction infinite multiplication factor distribution of the fuel assembly, and the three-dimensional infinite multiplication factor distribution was normalized so that the core average value was 1.0. The infinite multiplication factor distribution of each fuel assembly at that time is set as the axial one-dimensional infinite multiplication factor distribution of each fuel assembly (step 903). Step 9
In 04, it is determined whether all core data to be learned have been processed. If there is core data remaining, the process returns to step 901. If all core data has been processed, the radial two-dimensional infinite multiplication factor is obtained. Learning by the distribution estimating unit 107 and the axial one-dimensional infinite multiplication factor distribution estimating unit 108 is executed.

FIG. 10 is a flow chart showing a detailed procedure at the time of learning of the radial direction two-dimensional infinite multiplication factor distribution estimating means 107. Radial two-dimensional infinite multiplication factor distribution estimation means 1
Initialize all connection weights of the neural network at 07 to random values between -1.0 and 1.0 (step 1
001). Next, one of the core data to be learned is randomly selected, and further one of the fuel assemblies is randomly selected (step 1002). Next, the selected fuel assemblies, that is, the radial output value and the loading distance of the target fuel assembly, the output values and the loading distance of the fuel assemblies in the vicinity of the target fuel assembly, are input signals to the neural network. Set (step 1003). Next, the output value of each nerve cell is calculated (step 1004), and the error in the radial infinite multiplication factor of the target fuel assembly, which is the output signal of the neural network and the teacher signal, is calculated (step 1005). Is determined to be smaller than a preset value as the end reference value. As a result, if the criterion is not satisfied, the process proceeds to step 1007, and if it is satisfied, the learning is completed and the process ends (step 1006). Step 1007
Then, since the error has not reached the termination criterion, each connection weight of the neural network is corrected by the error back propagation method or the like, and the process returns to step 1002 to repeatedly adjust the connection weight until the desired accuracy is obtained.

Next, the radial two-dimensional infinite multiplication factor distribution is evaluated by the neural network which has completed learning. This detail is shown in Figure 1.
1 is shown. FIG. 12 shows the configuration of the neural network in the radial two-dimensional infinite multiplication factor distribution estimating means. First, select one of the fuel assemblies (step 11
01), the radial output value and the loading distance of the target fuel assembly, and the radial output value and the loading distance of the fuel assemblies near the target fuel assembly are input to the neural network which has completed learning. Set and calculate the output value of the neural network (step 1102). In step 1103, it is determined whether the output values corresponding to all the fuel assemblies have been obtained. If there are any fuel assemblies remaining, the process returns to step 1101 and if all the fuel assemblies have been processed, the process ends.

Next, the neural network is learned in the axial one-dimensional infinite multiplication factor distribution estimating means 108. This detailed procedure is shown in FIG. First, all the connection weights of the neural network in the axial one-dimensional output distribution estimation means 108 are initialized to random values between -1.0 and 1.0 (step 1301). Next, one of the core data to be learned is randomly selected, and further one of the fuel assemblies is randomly selected (step 1302). Next, the selected fuel assembly, that is, the axial power distribution of the target fuel assembly, the axial average power distribution of the fuel assemblies in the vicinity of the target fuel assembly, and the average output distribution of the target core are respectively input signals, Set to neural network (step 130)
3). Next, the output value of each nerve cell is calculated (step 13
04), the error of the axial infinite multiplication factor distribution of the target fuel assembly, which is the output signal of the neural network and the teacher signal, is calculated (step 1305), and this error is calculated from the value preset as the end reference value. Determine if it has become smaller.
As a result, if the criterion is not satisfied, the process proceeds to step 1307, and if it is satisfied, the learning is completed and the process ends (step 1306). In step 1307, since the error has not reached the termination criterion, each connection weight of the neural network is corrected by the error backpropagation method or the like, the process returns to step 1302, and the connection weight is repeatedly adjusted until the desired accuracy is obtained. .

Next, the one-dimensional infinite multiplication factor distribution in the axial direction is evaluated by the neural network which has completed learning. This detail is shown in Figure 1.
4 is shown. FIG. 15 shows the configuration of the neural network in the axial one-dimensional infinite multiplication factor distribution estimation means. First, select one of the fuel assemblies (step 14
01), the axial output distribution of the target fuel assembly, the axial average power distribution of the fuel assemblies in the vicinity of the target fuel assembly, and the average output distribution of the target core are input signals to the neural network which has completed learning. It is set and the output value of the neural network is calculated (step 1402). In step 1403, it is determined whether the output values corresponding to all the fuel assemblies have been obtained. If there are any fuel assemblies remaining, step 1
Returning to step 401, if all the fuel assemblies have been processed, the process ends.

Next, the operation procedure of the fuel loading pattern estimation means 109 of the present invention during learning will be described. The details are shown in FIG. First, all connection weights of the neural network in the fuel loading pattern estimation means 109 are −1.
Initialize to a random value between 0 and 1.0 (step 1601). Next, one of the core data to be learned is randomly selected, and further one of the fuel assemblies is randomly selected (step 1602). Next, the selected fuel assemblies, that is, the radial output value and the loading distance of the target fuel assembly, the output values and the loading distance of the fuel assemblies in the vicinity of the target fuel assembly, are input signals to the neural network. Set (step 1603). Next, the output value of each nerve cell is calculated (step 1604), the error between the output signal of the neural network and the fuel loading pattern of the target fuel assembly which is the teacher signal is calculated (step 1605), and this error ends. It is determined whether the reference value has become smaller than a preset value. As a result, if the criterion is not satisfied, the process proceeds to step 1607, and if the criterion is satisfied, the learning is completed and the process ends (step 1606). Step 160
In step 7, since the error has not reached the termination criterion, each connection weight of the neural network is corrected by the error back propagation method or the like, the process returns to step 1602, and the connection weight is repeatedly adjusted until the desired accuracy is obtained.

Next, the fuel loading pattern is evaluated by the neural network which has completed learning. The details are shown in FIG. FIG. 18 shows the configuration of the neural network in the fuel loading pattern estimation means. First, one of the fuel assemblies is selected (step 1701), and the radial output distribution and the loading distance of the target fuel assembly, and the radial output distribution and the loading distance of the fuel assemblies near the target fuel assembly are input. Set it as a signal in the neural network that has completed learning,
The output value of the neural network is calculated (step 1702). In step 1703, it is determined whether the output values corresponding to all the fuel assemblies have been obtained. If there are any fuel assemblies remaining, the process returns to step 1701 and if all the fuel assemblies have been processed, the process ends.

Next, an operation example of this embodiment will be described. FIG. 19
FIG. 19A shows an example of the positions and the radial infinite multiplication factors of each fuel assembly in the left half of the reactor practical-scale core.
Similarly, (b) shows an example of the position of each fuel assembly in the right half of the nuclear reactor practical-scale core and the infinite multiplication factor in the radial direction. 20 (a) corresponds to FIG. 19 (a), and shows the output value of each fuel assembly in the left half of the nuclear reactor practical-scale core calculated based on the detailed physical model, and FIG.
0 (b) corresponds to FIG. 19 (b), and shows the radial output value of each fuel assembly in the right half of the nuclear reactor practical scale core calculated based on the detailed physical model. 19 and 2
0 indicates one of the combinations of the radial infinite multiplication factor distribution and the radial output distribution. In this operation example, 83 sets of similar combinations were prepared and used for learning the neural network. Figure 2
FIG. 1 shows an example of a fuel loading pattern of each fuel assembly of the reactor practical-scale core. Further, FIG. 22 (a) shows an axial infinite multiplication factor distribution of one fuel assembly,
FIG. 22B corresponds to FIG. 22A and shows the axial power distribution of the fuel assembly calculated based on the detailed physical model. As shown in FIG. 22, the number of divisions in the axial direction is 15 in this embodiment.

FIG. 23 shows the positional relationship between the target fuel assembly and the neighboring fuel assembly in this embodiment. That is, in the present embodiment, the fuel assemblies near the target fuel assembly are 8 fuel assemblies in the periphery of the target fuel assembly, but the 24 fuel assemblies in the periphery in the 5 × 5 region are considered as the neighboring fuel assemblies. You can also The loading distance of the fuel assembly is the distance from the center of the reactor to the center of the fuel assembly.

As described above, the neural network in the radial two-dimensional output distribution estimating means shown in FIG. 5 and the neural circuit in the radial two-dimensional infinite multiplication factor estimating means shown in FIG. 12 in this embodiment. In each structure of the net and the neural network in the fuel loading pattern estimating means shown in FIG. 18, the first layer has 18 layers, the third layer has 1 layer, and the second layer has 36 layers. Further, the neural network in the axial one-dimensional output distribution estimating means shown in FIG. 8 and the neural network in the axial one-dimensional infinite multiplication factor distribution estimating means shown in FIG. 45 for 1 layer, 1 for 3rd layer
The number is 5, and the number of second layers is 90.

Next, the operation results of the three-dimensional output distribution estimating means of the present invention will be shown.

In the learning of the neural network in the radial direction two-dimensional output distribution estimating means, the convergence was almost zero with a mean square error of 0.77%. FIG. 24 shows the 1/4 core of the learning result at this time by the display means of the present invention. In the learning of the neural network in the axial one-dimensional output distribution estimating means, the convergence was almost complete with a mean square error of 1.00%.
Further, when the evaluation result by the radial two-dimensional output distribution estimating means and the evaluation result by the axial one-dimensional output distribution estimating means were combined, the mean square error was 2.46%. Here, the combination of the two estimation results is the product of the two estimation results.

Next, in the evaluation by the three-dimensional output distribution estimating means using the data not used for learning, the mean square error for the result calculated based on the detailed physical model is
It is 2.42%. With this error, the accuracy is practically sufficient, and the effectiveness of the present invention is shown.

Next, the operation result of the three-dimensional infinite image magnification distribution estimating means of the present invention will be shown.

In the learning of the neural network in the radial two-dimensional infinite multiplication factor distribution estimation means, the mean square error was substantially converged at 0.76%. In the learning of the neural network in the axial one-dimensional infinite multiplication factor distribution estimation means, the mean square error is 0.48%.
Almost converged. Also, when the evaluation result by the radial two-dimensional infinite multiplication factor distribution estimating means and the evaluation result by the axial one-dimensional infinite multiplication factor distribution estimating means were combined, the mean square error was 8.72%.

Next, in the evaluation by the three-dimensional infinite multiplication factor distribution estimation means using the data not used for learning, the mean square error with respect to the result calculated based on the detailed physical model is 5.24%. The accuracy is practically sufficient, and the effectiveness of the present invention is shown.

Next, in the learning of the neural network in the fuel loading pattern estimation means, when the learning data number 36 is 10
The estimation result of 0% converges, which shows the effectiveness of the fuel loading pattern estimation means. Here, since the fuel loading pattern is always an integer value, the output result of the neural network is converted to the nearest integer value, and the converted value is used as the evaluation result of the fuel loading pattern estimation means.

[0035]

As compared with the calculation based on the detailed physical model, the power distribution and the infinite multiplication factor distribution of the entire core and the estimated value of the fuel loading pattern can be obtained in a short time and simply.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a reactor core design support apparatus of the present invention.

FIG. 2 is a flowchart of a three-dimensional output distribution estimating means of the present invention.

FIG. 3 is a flowchart showing a detailed procedure of learning in the radial direction two-dimensional output distribution estimating means of the present invention.

FIG. 4 is a flowchart showing the estimation procedure in the radial direction two-dimensional output distribution estimation means of the present invention.

FIG. 5 is an explanatory diagram of a neural network in the radial direction two-dimensional output distribution estimating means of the present invention.

FIG. 6 is a flowchart of learning in the axial direction one-dimensional output distribution estimation means of the present invention.

FIG. 7 is a flowchart showing an estimation procedure in the axial one-dimensional output distribution estimation means of the present invention.

FIG. 8 is an explanatory diagram of a neural network in the axial one-dimensional output distribution estimation means of the present invention.

FIG. 9 is a flowchart showing the operation procedure of the three-dimensional infinite multiplication factor distribution estimation means of the present invention.

FIG. 10 is a flowchart showing the detailed procedure of learning in the radial two-dimensional infinite multiplication factor distribution estimation means of the present invention.

FIG. 11 is a flowchart showing an estimation procedure in the radial two-dimensional infinite multiplication factor distribution estimation means of the present invention.

FIG. 12 is an explanatory diagram of a neural network in a radial two-dimensional infinite multiplication factor distribution estimation means of the present invention.

FIG. 13 is a flowchart showing a detailed procedure of learning in the axial one-dimensional infinite multiplication factor distribution estimation means of the present invention.

FIG. 14 is a flowchart of the estimation procedure in the axial one-dimensional infinite multiplication factor distribution estimation means of the present invention.

FIG. 15 is an explanatory diagram of a neural network in the axial one-dimensional infinite multiplication factor distribution estimation means of the present invention.

FIG. 16 is a flowchart of learning in the fuel loading pattern estimation means of the present invention.

FIG. 17 is a flowchart of an estimation procedure in the fuel loading pattern estimation means of the present invention.

FIG. 18 is an explanatory diagram of a neural network in the fuel loading pattern estimation means of the present invention.

FIG. 19 is an explanatory view showing an example of an infinite multiplication factor in the radial direction in a practical reactor core.

FIG. 20 is an explanatory view showing a radial power distribution corresponding to FIG. 19 in a nuclear reactor core for practical use.

FIG. 21 is an explanatory diagram of an example of a fuel loading pattern in a reactor core for practical use.

FIG. 22 is an axial infinite multiplication factor distribution example in one fuel assembly and a corresponding axial power distribution map.

FIG. 23 is an explanatory diagram showing a positional relationship between a target fuel assembly and a fuel assembly in the vicinity of the target fuel assembly on the fuel assembly map.

FIG. 24 is an explanatory view showing an embodiment of the display means of the present invention.

[Explanation of symbols]

101 ... Reactor core design support device, 102 ... Data storage means, 103 ... Three-dimensional output distribution estimating means, 104 ... Radial two-dimensional output distribution estimating means, 105 ... Axial one-dimensional output distribution estimating means, 106 ... Three-dimensional Infinite multiplication factor distribution estimating means 107, radial two-dimensional infinite multiplication factor distribution estimating means, 1
08 ... Axial one-dimensional infinite multiplication factor distribution estimation means, 109 ...
Fuel loading pattern estimation means, 110 ... Estimation result storage means, 111 ... Display means, 112 ... Display device.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Takashi Nakao 3-2-1, Sachimachi, Hitachi City, Ibaraki Prefecture Hitachi Engineering Co., Ltd. No. 1 within Hitachi Engineering Co., Ltd.

Claims (10)

[Claims] 1. A three-dimensional infinite multiplication factor distribution of a fuel showing the performance of a nuclear fuel from a three-dimensional infinite multiplication factor distribution in a radial direction to a two-dimensional radial infinite multiplication factor distribution when estimating a three-dimensional power distribution in preparation of a fuel exchange plan for a nuclear reactor core design. And the axial one-dimensional infinite multiplication factor distribution, and at the time of learning, the radial two-dimensional output distribution and the axial one-dimensional output distribution are obtained from the three-dimensional output distribution, and then the radial two-dimensional infinite multiplication factor distribution. As the input signal, and the radial direction 2
A neural network that uses the dimensional output distribution as a teacher signal is constructed to determine the connection weight between the nerve cells. At the time of estimation, the radial two-dimensional infinite multiplication factor distribution is used as an input signal to output the output signal of the neural network. A radial direction two-dimensional output distribution estimating means which is an estimated value of the radial direction two-dimensional output distribution, and at the time of learning, the radial direction two-dimensional output distribution and the axial direction one-dimensional output distribution are obtained from the three-dimensional output distribution. After that, a neural network is constructed by using the axial one-dimensional infinite multiplication factor distribution as an input signal and the axial one-dimensional output distribution as a teacher signal to determine the coupling load between the nerve cells. Axial one-dimensional output distribution estimating means, which uses the one-dimensional infinite multiplication factor distribution in the direction as an input signal, and the output signal of the neural network as an estimated value of the one-dimensional output distribution in the axial direction, and the two-dimensional output distribution estimating means in the radial direction Estimated by , By combining the estimated value by the axial one-dimensional power distribution estimating means, reactor core design support apparatus which comprises a three-dimensional power distribution estimating means for obtaining an estimate of the three-dimensional power distribution. 2. When estimating a three-dimensional infinite multiplication factor distribution from a three-dimensional output distribution in the preparation of a fuel refueling plan for a nuclear reactor, a radial two-dimensional output distribution and an axial one-dimensional distribution are obtained from the three-dimensional output distribution. Obtain the output distribution, and at the time of learning, from the three-dimensional infinite multiplication factor distribution, the radial direction 2
A neural network in which the dimensional infinite multiplication factor distribution and the axial one-dimensional infinite multiplication factor distribution are obtained, and the radial two-dimensional output distribution is used as an input signal, and the radial two-dimensional infinite multiplication factor distribution is used as a teacher signal. To determine the coupling load between the nerve cells, and at the time of estimation, use the radial two-dimensional output distribution as an input signal and the neural network output signal as an estimated value of the radial two-dimensional infinite multiplication factor distribution. A radial direction two-dimensional infinite multiplication factor distribution estimating means, and at the time of learning, after obtaining the radial two-dimensional infinite multiplication factor distribution and the axial one-dimensional infinite multiplication factor distribution from the three-dimensional infinite multiplication factor distribution, A neural network that uses the direction one-dimensional output distribution as an input signal and the axial one-dimensional infinite multiplication factor distribution as a teacher signal is used to determine the coupling load between the nerve cells. Dimensional output distribution input signal As an axial infinite multiplication factor distribution estimating means for setting the output signal of the neural network as the axial one-dimensional infinite multiplication factor distribution estimated value, an estimated value by the radial direction two-dimensional infinite multiplication factor distribution estimating means, And a three-dimensional infinite multiplication factor distribution estimating unit for obtaining an estimated value of the three-dimensional infinite multiplication factor distribution by synthesizing the estimated values by the axial one-dimensional infinite multiplication factor distribution estimation unit. Design support device. 3. In a fuel refueling plan of a nuclear reactor, when a fuel loading pattern of each fuel assembly forming a nuclear reactor is estimated, at the time of learning, one fuel among the fuel assemblies forming each nuclear reactor is selected. The radial output value and loading distance of the target fuel assembly when paying attention to the assembly, and the radial output values and loading distance of the fuel assemblies near the target fuel assembly are used as input signals, and the target fuel assembly A neural network that uses the fuel loading pattern as a teacher signal is constructed to determine the joint load between nerve cells with all or part of the fuel assemblies that make up the reactor as the target fuel assemblies. The radial output value and the loading distance of the target fuel assembly, and the radial output values of the fuel assemblies near the target fuel assembly when paying attention to one of the fuel assemblies forming the And loading distance as input signal A fuel loading pattern estimation means for obtaining a fuel loading pattern estimation value of all fuel assemblies forming the nuclear reactor by setting all fuel assemblies forming the nuclear reactor as target fuel assemblies. Reactor core design support equipment. 4. The target fuel assembly according to claim 1, wherein the radial two-dimensional output distribution estimating means focuses on one fuel assembly among the fuel assemblies forming the reactor during learning. Infinite radial multiplication factor and loading distance of each fuel assembly, and each radial infinite multiplication factor and loading distance of a fuel assembly near the target fuel assembly as input signals, and a radial output value of the target fuel assembly as a teacher signal. A fuel cell that constitutes a nuclear reactor, and determines the coupling load between nerve cells by using all or part of the fuel assemblies that compose the reactor as target fuel assemblies, and at the time of estimation, each fuel assembly that composes the reactor. Enter the radial infinite multiplication factor and loading distance of the target fuel assembly and the radial infinite multiplication factor and loading distance of the fuel assembly near the target fuel assembly when focusing on one of the fuel assemblies. Configure the reactor as a signal By making all fuel assemblies that are the target fuel assemblies,
A reactor core design support device that obtains a radial two-dimensional power distribution estimated value of the entire core. 5. The target fuel according to claim 2, wherein the radial direction two-dimensional infinite multiplication factor distribution estimation means focuses on one fuel assembly among the fuel assemblies forming the reactor during learning. The radial output value and loading distance of the assembly, and the radial output value and loading distance of each fuel assembly near the target fuel assembly are used as input signals, and the infinite radial multiplication factor of the target fuel assembly is taught. A neural network that serves as a signal, and determines all or part of the fuel assemblies that make up the reactor as target fuel assemblies to determine the coupling load between nerve cells, and at the time of estimation, each fuel assembly that makes up the reactor. Input signals of the radial output value and loading distance of the target fuel assembly and the radial output value and loading distance of the fuel assembly near the target fuel assembly when paying attention to one of the fuel assemblies As a whole fuel that constitutes the reactor A reactor core design support device that obtains a radial two-dimensional infinite multiplication factor distribution estimated value of the entire core by using a fuel assembly as a target fuel assembly. 6. The target fuel assembly according to claim 1, wherein the axial one-dimensional output distribution estimating means focuses on one fuel assembly among the fuel assemblies forming the reactor during learning. The axial infinite multiplication factor distribution of the target fuel assembly, the axial average infinite multiplication factor distribution of the fuel assemblies near the target fuel assembly, and the axial average infinite multiplication factor distribution of the target core are input signals, and the target fuel assembly Is a neural network that uses the axial power distribution of as a teacher signal, and determines all or part of the fuel assemblies that make up the reactor as target fuel assemblies, Sometimes, when focusing on one fuel assembly among the fuel assemblies that make up the nuclear reactor, the axial infinite multiplication factor distribution of the target fuel assembly and the axial direction of the fuel assembly near the target fuel assembly Average infinite multiplication factor distribution and target furnace Reactor core design that obtains an axial one-dimensional power distribution estimated value of the entire core by targeting all the fuel assemblies that make up the reactor as the input signal of the axial mean infinite multiplication factor distribution of the core Support device. 7. The fuel according to claim 2, wherein the axial one-dimensional infinite multiplication factor distribution estimating means focuses on one fuel assembly among the fuel assemblies forming the reactor during learning. The axial power distribution of the assembly, the axial average power distribution of the fuel assemblies near the target fuel assembly, and the axial average power distribution of the target core are used as input signals, and the axial output of the target fuel assembly is infinite. A neural network that uses the multiplication factor distribution as a teacher signal, and by determining all or part of the fuel assemblies that make up the reactor as target fuel assemblies, after determining the coupling load between nerve cells, At the time of estimation, by setting all fuel assemblies constituting the reactor as target fuel assemblies, the axial direction 1
A reactor core design support device that obtains an infinite multiplication factor distribution estimate. 8. The reactor core design support device according to claim 1, 2, 4, 5, 6 or 7, wherein the control rod effect is included in the infinite multiplication factor distribution. 9. The radial two-dimensional output distribution estimating means according to claim 1 or 4, when learning, displays a "core map" for displaying all or part of a fuel assembly forming a nuclear reactor. Displayed on the device, for each fuel assembly to be displayed, the estimated value of the radial output value that is the estimated value, the radial output value that is the teacher signal, and the error of the estimated output value with respect to the teacher signal A nuclear reactor core design support apparatus having means for displaying three pieces of information on the ratio at the loading positions of the corresponding fuel assemblies on the core map. 10. The core map according to claim 2 or 5, wherein the radial two-dimensional infinite multiplication factor distribution estimating means targets all or part of a fuel assembly forming a nuclear reactor for display during learning. ”Is displayed on the display device, and for each fuel assembly to be displayed, the estimated value in the radial infinite multiplication factor, the infinite multiplication factor in the radial direction that is the teacher signal, and the estimated infinite multiplication factor A reactor core design support apparatus having means for displaying three pieces of information on the error rate of the teacher signal to the loading position of the corresponding fuel assembly on the core map.