WO2024104485A1 - Multi-fidelity network construction method and apparatus for nuclear reactor simulation test - Google Patents

Abstract

The present application relates to a multi-fidelity network construction method and apparatus for a nuclear reactor simulation test. The method comprises: acquiring a first fidelity network according to first fidelity data of a sample nuclear reactor, and acquiring at least one second fidelity network according to second fidelity data of the sample nuclear reactor (102); training the at least one second fidelity network by using the second fidelity data to obtain at least one trained second fidelity network (104); and combining the at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network, and training the multi-fidelity network by using the first fidelity data to obtain a trained multi-fidelity network, wherein the trained multi-fidelity network is used for performing a simulation test on a target nuclear reactor (106). By using the method, a final simulation result can be output according to the coupling between different fidelity data, thereby improving the simulation efficiency while ensuring the simulation precision.

Description

Multi-fidelity network construction method and device for nuclear reactor simulation testing

Related Applications

This application claims priority to Chinese patent application number 202211447961.8, filed on November 18, 2022, and entitled “Multi-fidelity network construction method and device for nuclear reactor simulation testing”, the entire text of which is hereby incorporated by reference.

Technical Field

The present application relates to the technical field of nuclear power plant reactor core design and operation, and in particular to a multi-fidelity network construction method, apparatus, computer equipment, storage medium and computer program product for nuclear reactor simulation testing.

Background technique

During the design and operation of nuclear reactors, simulations need to be performed for the actual or assumed operating conditions of the reactor to verify the safety of the designed or operated reactor. On the one hand, for the designed reactor, it is necessary to quantitatively evaluate the various operating boundaries and operating consequences of the reactor under various assumed accident conditions to ensure the safety of the designed reactor; on the other hand, in the operating reactor, various simulation calculations need to be performed to ensure that the design calculation parameters are consistent with the actual operating parameters within a certain error range, thereby ensuring the consistency between the designed reactor and the actual reactor, and thus ensuring that the operating reactor has sufficient safety margins under various accident conditions. In addition, some parameters directly related to the safety of reactor operation (for example, fuel rod cladding temperature, etc.) cannot be measured directly, but must be derived from some measurable reactor fluid parameters (such as temperature, pressure, etc.) or neutron detector readings (such as characterizing fission reaction rate) combined with the simulation model of the reactor. These models are approximate abstractions of the real reactor process and can only partially simulate the neutron behavior and burnup behavior in the reactor. For example, the nuclear design software package PCM uses the equivalent homogenization assumption and neutron diffusion approximation to realize the simulation of the three-dimensional core. Although point reactor equations have no spatial distribution, they are often used for inversion monitoring of reactivity based on power changes and xenon poisoning. These models are collectively referred to as state transition models, which represent the process of core state changing from the state at the next moment due to control actions. State transition models are used for both the state distribution of non-measurable variables at the current moment and the prediction of core state at subsequent moments, and there are unknown errors.

At present, there are a large number of reactor theory research results, including high-fidelity models (such as Monte Carlo method, transport theory or diffusion approximation, etc.) and low-fidelity models (such as point reactor dynamics, etc.), but these models are all obtained by using single-fidelity data modeling, and there are major defects in the process of use. The computational efficiency of the high-fidelity mathematical equation model is difficult to meet the real-time requirements, and the low-fidelity introduces too many simplified assumptions and is difficult to meet the accuracy requirements.

Therefore, current nuclear reactor simulations cannot improve simulation efficiency while ensuring accuracy.

Summary of the invention

In a first aspect, the present application provides a method for constructing a multi-fidelity network for nuclear reactor simulation testing. The method comprises:

Acquire a first fidelity network based on first fidelity data of the sample nuclear reactor, and acquire at least one second fidelity network based on second fidelity data of the sample nuclear reactor;

Using the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network;

At least one trained second fidelity network is combined with a first fidelity network to obtain a multi-fidelity network, and the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

In one embodiment, before acquiring at least one second fidelity network according to the second fidelity data of the sample nuclear reactor, the method further includes:

Acquire multiple data sets of sample nuclear reactors;

Determine the accuracy of each data set, and use the data in a data set with the highest accuracy as the first fidelity data;

Among the multiple data sets, data other than the first fidelity data is used as the second fidelity data.

In one embodiment, obtaining at least one second fidelity network according to second fidelity data of a sample nuclear reactor includes:

Classifying the second fidelity data into different levels to obtain at least one fidelity level and sub-data corresponding to each fidelity level;

Obtaining corresponding second fidelity networks according to the sub-data corresponding to each fidelity level; the number of second fidelity networks is the same as the number of fidelity levels;

Using the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network includes:

Each second fidelity network is trained using the sub-data corresponding to each fidelity level to obtain at least one trained second fidelity network.

In one embodiment, at least one trained second fidelity network is combined with a first fidelity network to obtain a multi-fidelity network, comprising:

Connecting the output of each trained second fidelity network to one of the inputs of the first fidelity network;

The input end of the first fidelity network and the input ends of each trained second fidelity network are used together as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain a multi-fidelity network.

In one embodiment, the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network, including:

Use the multi-fidelity network as the generator network, and obtain the corresponding discriminator network based on the generator network;

Construct a generative adversarial network based on the generator network and the discriminator network;

Using the first fidelity data to train the generative adversarial network to obtain a trained generative adversarial network;

Obtain the trained generator network from the trained generative adversarial network as the trained multi-fidelity network.

In one embodiment, the method further comprises:

Acquiring control parameters and first moment state parameters of a target nuclear reactor;

Inputting the control parameters and the state parameters at the first moment into the trained multi-fidelity network to obtain the state parameters of the target nuclear reactor at the second moment;

The simulation test result of the target nuclear reactor is obtained based on the state parameters at the second moment.

In a second aspect, the present application also provides a multi-fidelity network construction device for nuclear reactor simulation testing. The device comprises:

an acquisition module, configured to acquire a first fidelity network according to first fidelity data of the sample nuclear reactor, and to acquire at least one second fidelity network according to second fidelity data of the sample nuclear reactor;

A training module, configured to train at least one second fidelity network using second fidelity data to obtain at least one trained second fidelity network;

A combination module is used to combine at least one trained second fidelity network with a first fidelity network to obtain a multi-fidelity network, and use the first fidelity data to train the multi-fidelity network to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

Acquire a first fidelity network based on first fidelity data of the sample nuclear reactor, and acquire at least one second fidelity network based on second fidelity data of the sample nuclear reactor;

Using the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network;

At least one trained second fidelity network is combined with a first fidelity network to obtain a multi-fidelity network, and the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

Acquire a first fidelity network based on first fidelity data of the sample nuclear reactor, and acquire at least one second fidelity network based on second fidelity data of the sample nuclear reactor;

Using the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network;

At least one trained second fidelity network is combined with a first fidelity network to obtain a multi-fidelity network, and the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

Acquire a first fidelity network based on first fidelity data of the sample nuclear reactor, and acquire at least one second fidelity network based on second fidelity data of the sample nuclear reactor;

Using the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network;

At least one trained second fidelity network is combined with a first fidelity network to obtain a multi-fidelity network, and the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

The above-mentioned multi-fidelity network construction method, device, computer equipment, storage medium and computer program product for nuclear reactor simulation testing obtains a first fidelity network according to the first fidelity data of the sample nuclear reactor, and obtains at least one second fidelity network according to the second fidelity data of the sample nuclear reactor; uses the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network; combines at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network, and uses the first fidelity data to train the multi-fidelity network to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to simulate the target nuclear reactor. In this way, when the target nuclear reactor is simulated by the multi-fidelity network, the second fidelity network can first obtain a reference simulation result according to the input parameters, and input the input parameters and the simulation result into the first fidelity network together, and the first fidelity network can output the final simulation result according to the coupling between different fidelity data. Details of one or more embodiments of the present application are presented in the following drawings and descriptions. Other features, objects, and advantages of the application will be apparent from the description and drawings, and from the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions will be briefly introduced below. Obviously, the drawings described below are merely embodiments of the present application, and for ordinary technicians in this field, other drawings can be obtained based on the disclosed drawings without paying any creative work.

FIG1 is a schematic diagram of a flow chart of a method for constructing a multi-fidelity network for nuclear reactor simulation testing in one embodiment;

FIG2 is a schematic diagram of the structure of a second fidelity network in one embodiment;

FIG3 is a schematic diagram of the structure of a multi-fidelity network in one embodiment;

FIG4 is a schematic diagram of a process of generating adversarial network training in another embodiment;

FIG5 is a structural block diagram of a multi-fidelity network construction device for nuclear reactor simulation testing in one embodiment;

FIG. 6 is a diagram showing the internal structure of a computer device in one embodiment.

Detailed ways

The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

In one embodiment, as shown in FIG1 , a multi-fidelity network construction method for nuclear reactor simulation testing is provided. This embodiment uses the method applied to a computer device as an example. It is understandable that the computer device may be a terminal or a server. The terminal may be, but is not limited to, various industrial computers. The server may be, for example, a computer. The method can be implemented by a separate server or a server cluster composed of multiple servers. In this embodiment, the method includes the following steps:

Step 102: acquiring a first fidelity network according to first fidelity data of a sample nuclear reactor, and acquiring at least one second fidelity network according to second fidelity data of the sample nuclear reactor.

Among them, the first fidelity data refers to high-fidelity data, which can be generated by high-fidelity software; the second fidelity data refers to low-fidelity data with lower data accuracy than the first fidelity data, which can be quickly generated by low-fidelity software. Both the first fidelity data and the second fidelity data include multiple control parameters and state parameters of the input and output of the sample nuclear reactor, including: 1) reactivity parameters, such as control rod positions, etc.; 2) power parameters, such as power level, power distribution; 3) nuclear density parameters, such as the nuclear density at the axial height of each component (including fissile nuclides, minor actinide nuclides, light nuclides, etc.); 4) macroscopic or microscopic reaction cross sections; 5) thermal parameters, such as coolant temperature, pressure, flow, etc., fuel or material temperature, etc. Among them, reactivity parameters and power parameters can be used as input parameters, and other state parameters as output parameters.

Optionally, according to the simulation requirements of the target nuclear reactor and the data type of the first fidelity data, a suitable neural network is selected as the first fidelity network. Similarly, according to the simulation requirements of the target nuclear reactor and the data type of the second fidelity data, another one or more suitable neural networks are selected as the second fidelity networks.

Step 104: Use the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network.

Optionally, all second fidelity data can be directly used to train a second fidelity network to obtain a trained second fidelity network. The second fidelity data can also be first differentiated by accuracy, and each group of data in the second fidelity data can be scored for accuracy according to the state characteristics of the sample nuclear reactor, and then all data can be divided into multiple fidelity levels according to the accuracy score, and the same number of second fidelity networks as the fidelity level are prepared, and each second fidelity network is trained using data of a fidelity level. For example, the second fidelity data is divided into two parts according to the accuracy, and the data with higher accuracy is used as medium fidelity data, and the data with lower accuracy is used as low fidelity data. Two second fidelity networks are prepared: a medium fidelity network and a low fidelity network. The medium fidelity data is used to train the medium fidelity network to obtain a trained medium fidelity network, and the low fidelity data is used to train the low fidelity network to obtain a trained low fidelity network.

Step 106: combine at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network, and use the first fidelity data to train the multi-fidelity network to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on the target nuclear reactor.

Optionally, the output end of each trained second fidelity network is connected to the input end of the first fidelity network; the input end of the first fidelity network and the input end of each trained second fidelity network are used as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain a multi-fidelity network. When training the multi-fidelity network, the input data is first input into each trained second fidelity network, and each trained second fidelity network outputs its own simulation results. These simulation results are used as reference data and input into the first fidelity network together with the input data. The first fidelity network processes the input data in combination with the reference data, outputs a simulation data, compares the simulation data with the label data corresponding to the input data, and adjusts the weight parameters of the first fidelity network. When the comparison result of the simulation data output by the first fidelity network and the label data corresponding to the input data meets the simulation requirements, the training is completed to obtain a trained multi-fidelity network. During the entire training process, each trained second fidelity network is not adjusted, that is, the first fidelity data is only used to train the first fidelity network.

In a feasible implementation, the control parameters and current state parameters of the target nuclear reactor are obtained, and then the control parameters and current state parameters are input into a trained multi-fidelity network. The multi-fidelity network can output predicted state parameters, and the predicted state parameters can characterize the changes in the state parameters of the target nuclear reactor under the influence of the control parameters. Based on the predicted state parameters, the target nuclear reactor can be simulated and tested to determine the optimal control method and control parameters for the target nuclear reactor.

In the above-mentioned multi-fidelity network construction method for nuclear reactor simulation testing, a first fidelity network is obtained according to the first fidelity data of the sample nuclear reactor, and at least one second fidelity network is obtained according to the second fidelity data of the sample nuclear reactor. A two-fidelity network; using the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network; combining at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network, and using the first fidelity data to train the multi-fidelity network to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to simulate and test the target nuclear reactor. In this way, when simulating the target nuclear reactor through the multi-fidelity network, the second fidelity network can first obtain a reference simulation result based on the input parameters, and input the input parameters and the simulation result into the first fidelity network together. The first fidelity network can output the final simulation result based on the coupling between different fidelity data, thereby improving the simulation efficiency while ensuring the simulation accuracy.

In one embodiment, before acquiring at least one second fidelity network based on the second fidelity data of the sample nuclear reactor, the method further includes: acquiring multiple data sets of the sample nuclear reactor; determining the accuracy of each data set, and using the data in a data set with the highest accuracy as the first fidelity data; and using the data in the multiple data sets, except the first fidelity data, as the second fidelity data.

Optionally, the first fidelity data and second fidelity data of the sample nuclear reactor can be generated by a variety of software with different fidelity. These software are based on mathematical physics equations and are used for derivation and calculation. The measured data under some operating conditions can also be obtained manually. The measured data itself can exist as the highest level of fidelity or as low-fidelity data, depending on the nature of data acquisition and data acquisition conditions. Therefore, it is necessary to divide these data into fidelity levels, with high-fidelity data as first-fidelity data and the rest of the data as second-fidelity data. The data format needs to be unified between data of different fidelity levels.

Specifically, the state space parameters of the sample nuclear reactor are defined as High-fidelity software is used to generate high-fidelity data sets. Because high-fidelity simulation software has high calculation accuracy, but low calculation efficiency. For example, the calculation of a typical reactor state point takes about several minutes or hours. Therefore, high-fidelity data is relatively scarce. Assume that the input of the high-fidelity simulation software is The output is Get high-fidelity input-output pairs: Since the reactor process can be essentially regarded as a Markov process, its input state space and output state space can be essentially consistent. Similarly, construct the medium-fidelity input-output pair: And low-fidelity input-output pairs:

In this embodiment, by acquiring multiple data sets of a sample nuclear reactor, determining the accuracy of each data set, taking the data in a data set with the highest accuracy as first fidelity data, and taking the data in multiple data sets except the first fidelity data as second fidelity data, data of different fidelity can be obtained.

In one embodiment, at least one second fidelity network is obtained based on second fidelity data of a sample nuclear reactor, including: classifying the second fidelity data into levels to obtain at least one fidelity level and sub-data corresponding to each fidelity level; obtaining a corresponding second fidelity network based on the sub-data corresponding to each fidelity level; the number of second fidelity networks is the same as the number of fidelity levels.

Furthermore, the second fidelity data is used to train at least one second fidelity network to obtain at least one trained second fidelity network, including: using sub-data corresponding to each fidelity level to train each second fidelity network to obtain at least one trained second fidelity network.

Optionally, taking the second fidelity network including a medium fidelity network and a low fidelity network as an example, for the two second fidelity networks, a standard neural network is selected to perform fitting training on the input and output to achieve and As shown in Figure 2, the number of layers (or depth) m of the neural network, the number of nodes (or width) n of each layer, the activation function ReLU or leakyReLu, the learning rate, the optimization algorithm Adam or SGD, etc. are the hyperparameters of neural network training, which can be set according to different problems or according to the hyperparameter optimization algorithm HPO. To finally determine the relevant parameters. The methods of building and training neural networks can be selected according to the data type, training requirements, etc., which will not be described here. Some deep learning open source platforms can be used, including PaddlePaddle, Pytorch, Tensorflow, etc., which can easily implement an optimized neural network model to characterize the intrinsic relationship of low-fidelity data or medium-fidelity data.

In this embodiment, by classifying the second fidelity data, at least one fidelity level and sub-data corresponding to each fidelity level are obtained; the corresponding second fidelity network is obtained according to the sub-data corresponding to each fidelity level; the number of second fidelity networks is the same as the number of fidelity levels; the sub-data corresponding to each fidelity level are respectively used to train each second fidelity network to obtain at least one trained second fidelity network. Multiple trained second fidelity networks can be obtained.

In one embodiment, at least one trained second fidelity network is combined with a first fidelity network to obtain a multi-fidelity network, including: connecting the output of each trained second fidelity network to one of the inputs of the first fidelity network; using the input of the first fidelity network and the input of each trained second fidelity network as the input of the multi-fidelity network, and using the output of the first fidelity network as the output of the multi-fidelity network to obtain the multi-fidelity network.

Optionally, take the first fidelity network as a high-fidelity network, and the second fidelity network as an example, since the low-fidelity network itself has a large amount of data, while the amount of high-fidelity data is relatively small, directly training the high-fidelity data source will lead to a large overfitting error. Therefore, the high-fidelity input data x ^high is input into the already trained low-fidelity network, medium-fidelity network, etc., to obtain the output of these non-high-fidelity network levels and In fact, or The difference from the true y ^high characterizes the error of the low-fidelity network extended to the high-fidelity data.

Furthermore, through the learning and training of the first fidelity network, the relevant errors can be corrected. Therefore, the first fidelity network is constructed, as shown in FIG3, and the output of the low-fidelity network and the output network of the medium-fidelity network are mapped to one of the inputs of the high-fidelity network, and a multi-fidelity network is obtained by combining them. Among them, the depth and width m ^high and n ^high of the multi-fidelity network, as well as the activation function, training hyperparameters, etc., need to be determined according to the state parameters of the specific nuclear reactor, which will not be repeated here.

In this embodiment, the output end of each trained second fidelity network is connected to one of the input ends of the first fidelity network; the input end of the first fidelity network and the input end of each trained second fidelity network are used as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain a multi-fidelity network. The multi-fidelity network can combine the output results of different fidelity networks to obtain the final simulation result, and the second fidelity network part in the multi-fidelity network is obtained by training with a large amount of low-fidelity data, which can greatly improve the computational efficiency and accuracy of training to meet real-time requirements.

In one embodiment, a multi-fidelity network is trained using first fidelity data to obtain a trained multi-fidelity network, including: using the multi-fidelity network as a generator network, and obtaining a corresponding discriminator network based on the generator network; constructing a generative adversarial network based on the generator network and the discriminator network; training the generative adversarial network using the first fidelity data to obtain a trained generative adversarial network; and obtaining a trained generator network from the trained generative adversarial network as a trained multi-fidelity network.

Optionally, taking the first fidelity network as a high-fidelity network and the second fidelity network including a medium-fidelity network and a low-fidelity network as an example, as shown in FIG3, based on the output of the generator network and Traditional neural network training directly uses And the deviation of y ^high , such as L1loss loss function: You can directly Then carry out training.

Based on the above approach, adding artificially designed noise to the real samples to synthesize a new input data (or adversarial data) will cause the entire generator network to make prediction errors. That is, in the face of adversarial data, the deep generator network is fragile and can easily confuse these deep neural networks. The prediction error rate of deep neural networks for adversarial samples is very high. When humans think that it is almost impossible to distinguish the difference between the original sample and the adversarial sample (the deviation is very small), the original function of the deep neural network has failed and cannot be used as the basis for the subsequent state prediction of the reactor. In particular, in the actual measurement process of nuclear power plants, all measurement parameters are accompanied by random errors that may be eliminated, and some potential error disturbances may lead to catastrophic errors in the prediction of deep neural network models. Improving the adversarial robustness of neural networks (the ability to resist adversarial samples) is intuitively important for the development of reactor prediction models. Therefore, based on the idea of adversarial learning, a discriminator is added. The adversarial training method is used to improve the ability of neural networks to deceive adversarial samples. The basic idea of adversarial training is to continuously generate and learn adversarial samples during network training. The reactor state label generated by the generator G(X) can deceive the discriminator and be consistent with the real reactor state label. That is, the output of the discriminator D(Y) is the possibility of judging the true label. The discriminator's network structure, learning rate and other hyperparameters are adjusted according to the specific actual scenario.

Furthermore, as shown in Figure 4, for the generator G(X), it is necessary to constantly deceive the discriminator D so that log(D(G(X))) reaches the maximum, that is, As close as possible to y ^high . For the discriminator D, it needs to keep learning to prevent being deceived by the generator. At this time, for the real input y ^high, it needs to maximize the correct judgment as true, and at the same time, for the false input Achieve the maximum false judgment. Maximize log(D(Y))+log(1-D(G(X))). The specific training process is to first train the discriminator D, and then train the generator G, until the discriminator D and the generator G reach a Nash equilibrium.

In this embodiment, a multi-fidelity network is used as a generator network, and a corresponding discriminator network is obtained according to the generator network; a generative adversarial network is constructed according to the generator network and the discriminator network; the generative adversarial network is trained using the first fidelity data to obtain a trained generative adversarial network; and a trained generator network is obtained from the trained generative adversarial network as a trained multi-fidelity network. A trained multi-fidelity network can be obtained, and the trained multi-fidelity network can combine the output results of different fidelity networks to obtain simulation results corresponding to the input data.

The trained multi-fidelity network is used to simulate and test the target nuclear reactor pair, that is, the target nuclear reactor can be simulated and tested through the trained multi-fidelity network to simulate the state change of the target reactor. Specifically, in one embodiment, the step of performing simulation testing on the target nuclear reactor through the trained multi-fidelity network includes: obtaining control parameters and state parameters of the target nuclear reactor at a first moment; inputting the control parameters and the state parameters of the first moment into the trained multi-fidelity network to obtain the state parameters of the target nuclear reactor at a second moment; and obtaining the simulation test results of the target nuclear reactor based on the state parameters at the second moment.

Optionally, taking the first fidelity network as a high-fidelity network and the second fidelity network including a medium-fidelity network and a low-fidelity network as an example, a multi-fidelity network obtained by combining the low-fidelity network, the medium-fidelity network, and the high-fidelity network can realize the change of the reactor state. As shown in Figure 3, in the actual application process, the multi-fidelity network inputs x ^{high ,} which includes the control parameters of the target nuclear reactor and the state parameters at the first moment. The low-fidelity network and the medium-fidelity network first process the input data x ^high and output them respectively. and Then input x ^high into high Fidelity network, and will also and Input high fidelity network, high fidelity network output reactor state simulation change Including the state parameters of the target nuclear reactor at the second moment. The relationship between the first moment and the second moment depends on the moment relationship between x ^high and y ^high in the training data (x ^high , y ^high ) during the training process of the multi-fidelity network.

In this embodiment, the control parameters and the state parameters of the target nuclear reactor at the first moment are obtained; the control parameters and the state parameters of the first moment are input into the trained multi-fidelity network to obtain the state parameters of the target nuclear reactor at the second moment; and the simulation test results of the target nuclear reactor are obtained based on the state parameters at the second moment. The final simulation results can be output according to the coupling between different fidelity data, thereby improving the simulation efficiency while ensuring the simulation accuracy.

In one embodiment, a multi-fidelity network for nuclear reactor simulation testing is constructed, comprising:

Acquire multiple data sets of a sample nuclear reactor; determine the accuracy of each data set, and use the data in a data set with the highest accuracy as first fidelity data; and use the data in the multiple data sets, except the first fidelity data, as second fidelity data.

A first fidelity network is obtained according to the first fidelity data of the sample nuclear reactor. The second fidelity data is graded to obtain at least one fidelity grade and sub-data corresponding to each fidelity grade; a corresponding second fidelity network is obtained according to the sub-data corresponding to each fidelity grade; the number of the second fidelity networks is the same as the number of the fidelity grades.

Each second fidelity network is trained using the sub-data corresponding to each fidelity level to obtain at least one trained second fidelity network.

The output end of each trained second fidelity network is connected to one of the input ends of the first fidelity network; the input end of the first fidelity network and the input end of each trained second fidelity network are used as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain a multi-fidelity network.

The multi-fidelity network is used as a generator network, and the corresponding discriminator network is obtained according to the generator network; a generative adversarial network is constructed according to the generator network and the discriminator network; the generative adversarial network is trained using the first fidelity data to obtain a trained generative adversarial network; the trained generator network is obtained from the trained generative adversarial network as a trained multi-fidelity network, and the trained multi-fidelity network is used to perform simulation tests on a target nuclear reactor.

Among them, the specific steps of performing simulation test on the target nuclear reactor through the trained multi-fidelity network include: obtaining control parameters and first-time state parameters of the target nuclear reactor; inputting the control parameters and first-time state parameters into the trained multi-fidelity network to obtain second-time state parameters of the target nuclear reactor; and obtaining simulation test results of the target nuclear reactor based on the second-time state parameters.

It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

Based on the same inventive concept, the embodiment of the present application also provides a multi-fidelity network construction device for nuclear reactor simulation testing, which is used to implement the multi-fidelity network construction method for nuclear reactor simulation testing involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the multi-fidelity network construction device for nuclear reactor simulation testing provided below can refer to the limitations of the multi-fidelity network construction method for nuclear reactor simulation testing above, and will not be repeated here.

In one embodiment, as shown in FIG. 5 , a multi-fidelity network architecture for nuclear reactor simulation testing is provided. The device 500 includes: an acquisition module 501, a training module 502 and a combination module 503, wherein:

The acquisition module 501 is used to acquire a first fidelity network according to the first fidelity data of the sample nuclear reactor, and to acquire at least one second fidelity network according to the second fidelity data of the sample nuclear reactor.

The training module 502 is used to train at least one second fidelity network using the second fidelity data to obtain at least one trained second fidelity network.

The combination module 503 is used to combine at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network, and use the first fidelity data to train the multi-fidelity network to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

In one embodiment, the acquisition module 501 is also used to acquire multiple data sets of a sample nuclear reactor; determine the accuracy of each data set, and use the data in a data set with the highest accuracy as first fidelity data; and use the data in multiple data sets except the first fidelity data as second fidelity data.

In one embodiment, the acquisition module 501 is further used to grade the second fidelity data to obtain at least one fidelity grade and sub-data corresponding to each fidelity grade; obtain the corresponding second fidelity network according to the sub-data corresponding to each fidelity grade; the number of second fidelity networks is the same as the number of fidelity grades.

The training module 502 is further configured to respectively use the sub-data corresponding to each fidelity level to train each second fidelity network to obtain at least one trained second fidelity network.

In one embodiment, the combination module 503 is also used to connect the output end of each trained second fidelity network to one of the input ends of the first fidelity network; the input end of the first fidelity network and the input end of each trained second fidelity network are used as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain a multi-fidelity network.

In one embodiment, the combination module 503 is also used to use the multi-fidelity network as a generator network, and obtain a corresponding discriminator network based on the generator network; construct a generative adversarial network based on the generator network and the discriminator network; use the first fidelity data to train the generative adversarial network to obtain a trained generative adversarial network; obtain a trained generator network from the trained generative adversarial network as a trained multi-fidelity network.

In one embodiment, the apparatus further comprises:

The test module 504 is used to obtain the control parameters and the state parameters of the target nuclear reactor at the first moment; input the control parameters and the state parameters of the first moment into the trained multi-fidelity network to obtain the state parameters of the target nuclear reactor at the second moment; and obtain the simulation test results of the target nuclear reactor based on the state parameters at the second moment.

Each module in the multi-fidelity network construction device for nuclear reactor simulation testing can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG6. The computer device includes a processor, a memory, an input/output interface (Input/Output, referred to as I/O) and a communication interface. Among them, the processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store neural network data. The input/output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a multi-fidelity network construction method for nuclear reactor simulation testing is implemented.

Those skilled in the art will understand that the structure shown in FIG. 6 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program A computer program is provided for executing the computer program by a processor, wherein the following steps are implemented when the processor executes the computer program: obtaining a first fidelity network according to first fidelity data of a sample nuclear reactor, and obtaining at least one second fidelity network according to second fidelity data of the sample nuclear reactor; training at least one second fidelity network using the second fidelity data to obtain at least one trained second fidelity network; combining at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network, and training the multi-fidelity network using the first fidelity data to obtain a trained multi-fidelity network; and using the trained multi-fidelity network to perform simulation testing on a target nuclear reactor.

In one embodiment, when the processor executes the computer program, the following steps are also implemented: acquiring multiple data sets of a sample nuclear reactor; determining the accuracy of each data set, and using the data in a data set with the highest accuracy as first fidelity data; and using the data in multiple data sets, except the first fidelity data, as second fidelity data.

In one embodiment, when the processor executes the computer program, the following steps are also implemented: the second fidelity data is graded to obtain at least one fidelity level and sub-data corresponding to each fidelity level; a corresponding second fidelity network is obtained according to the sub-data corresponding to each fidelity level; the number of second fidelity networks is the same as the number of fidelity levels.

In one embodiment, when the processor executes the computer program, the following steps are further implemented: using the sub-data corresponding to each fidelity level to train each second fidelity network to obtain at least one trained second fidelity network.

In one embodiment, when the processor executes the computer program, the following steps are also implemented: the output end of each trained second fidelity network is connected to one of the input ends of the first fidelity network; the input end of the first fidelity network and the input end of each trained second fidelity network are used as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain the multi-fidelity network.

In one embodiment, when the processor executes the computer program, the following steps are also implemented: using the multi-fidelity network as a generator network, and obtaining a corresponding discriminator network based on the generator network; constructing a generative adversarial network based on the generator network and the discriminator network; using the first fidelity data to train the generative adversarial network to obtain a trained generative adversarial network; obtaining a trained generator network from the trained generative adversarial network as a trained multi-fidelity network.

In one embodiment, when the processor executes the computer program, the following steps are also implemented: obtaining control parameters and first-moment state parameters of the target nuclear reactor; inputting the control parameters and first-moment state parameters into a trained multi-fidelity network to obtain second-moment state parameters of the target nuclear reactor; and obtaining simulation test results of the target nuclear reactor based on the second-moment state parameters.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: a first fidelity network is obtained based on first fidelity data of a sample nuclear reactor, and at least one second fidelity network is obtained based on second fidelity data of the sample nuclear reactor; at least one second fidelity network is trained using the second fidelity data to obtain at least one trained second fidelity network; at least one trained second fidelity network is combined with the first fidelity network to obtain a multi-fidelity network, and the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: acquiring multiple data sets of a sample nuclear reactor; determining the accuracy of each data set, and using the data in a data set with the highest accuracy as first fidelity data; and using the data in multiple data sets, except the first fidelity data, as second fidelity data.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: the second fidelity data is graded to obtain at least one fidelity level and sub-data corresponding to each fidelity level; a corresponding second fidelity network is obtained according to the sub-data corresponding to each fidelity level; the number of second fidelity networks is the same as the number of fidelity levels.

In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: using the sub-data corresponding to each fidelity level to train each second fidelity network to obtain at least one trained second fidelity network.

In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: connecting the output end of each trained second fidelity network to one of the input ends of the first fidelity network; using the input end of the first fidelity network and the input end of each trained second fidelity network as the input end of the multi-fidelity network; The output of the single-fidelity network is used as the output of the multi-fidelity network to obtain a multi-fidelity network.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: using the multi-fidelity network as a generator network, and obtaining a corresponding discriminator network based on the generator network; constructing a generative adversarial network based on the generator network and the discriminator network; using the first fidelity data to train the generative adversarial network to obtain a trained generative adversarial network; obtaining a trained generator network from the trained generative adversarial network as a trained multi-fidelity network.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining control parameters and first-time state parameters of the target nuclear reactor; inputting the control parameters and first-time state parameters into a trained multi-fidelity network to obtain second-time state parameters of the target nuclear reactor; and obtaining simulation test results of the target nuclear reactor based on the second-time state parameters.

In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

A first fidelity network is obtained according to first fidelity data of a sample nuclear reactor, and at least one second fidelity network is obtained according to second fidelity data of the sample nuclear reactor; at least one second fidelity network is trained using the second fidelity data to obtain at least one trained second fidelity network; at least one trained second fidelity network is combined with the first fidelity network to obtain a multi-fidelity network, and the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.

In one embodiment, when the computer program is executed by a processor, the following steps are also implemented: acquiring multiple data sets of a sample nuclear reactor; determining the accuracy of each data set, and using the data in a data set with the highest accuracy as first fidelity data; and using the data in multiple data sets, except the first fidelity data, as second fidelity data.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: the second fidelity data is graded to obtain at least one fidelity level and sub-data corresponding to each fidelity level; a corresponding second fidelity network is obtained according to the sub-data corresponding to each fidelity level; the number of second fidelity networks is the same as the number of fidelity levels.

In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: using the sub-data corresponding to each fidelity level to train each second fidelity network to obtain at least one trained second fidelity network.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: the output end of each trained second fidelity network is connected to one of the input ends of the first fidelity network; the input end of the first fidelity network and the input end of each trained second fidelity network are used as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain the multi-fidelity network.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: using the multi-fidelity network as a generator network, and obtaining a corresponding discriminator network based on the generator network; constructing a generative adversarial network based on the generator network and the discriminator network; using the first fidelity data to train the generative adversarial network to obtain a trained generative adversarial network; obtaining a trained generator network from the trained generative adversarial network as a trained multi-fidelity network.

In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: obtaining control parameters and first-time state parameters of the target nuclear reactor; inputting the control parameters and first-time state parameters into a trained multi-fidelity network to obtain second-time state parameters of the target nuclear reactor; and obtaining simulation test results of the target nuclear reactor based on the second-time state parameters.

It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other media used in the embodiments provided in this application can include non-volatile computer-readable storage media. At least one of volatile and non-volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, etc., but is not limited thereto.

The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims (15)

1. A method for constructing a multi-fidelity network for nuclear reactor simulation testing, characterized in that the method comprises:

   Acquire a first fidelity network based on first fidelity data of a sample nuclear reactor, and acquire at least one second fidelity network based on second fidelity data of the sample nuclear reactor;

   Using the second fidelity data to train at least one second fidelity network to obtain at least one trained second fidelity network;

   At least one trained second fidelity network is combined with the first fidelity network to obtain a multi-fidelity network, and the multi-fidelity network is trained using the first fidelity data to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.
2. The method according to claim 1, characterized in that before acquiring at least one second fidelity network according to the second fidelity data of the sample nuclear reactor, it also includes:

   acquiring a plurality of data sets of the sample nuclear reactor;

   Determine the accuracy of each data set, and use the data in a data set with the highest accuracy as the first fidelity data;

   The data other than the first fidelity data among the multiple data sets are used as the second fidelity data.
3. The method according to claim 1, characterized in that the acquiring at least one second fidelity network according to the second fidelity data of the sample nuclear reactor comprises:

   Classifying the second fidelity data into different levels to obtain at least one fidelity level and sub-data corresponding to each fidelity level;

   Acquire a corresponding second fidelity network according to the sub-data corresponding to each fidelity level; the number of the second fidelity networks is the same as the number of the fidelity levels;

   The step of training at least one second fidelity network using the second fidelity data to obtain at least one trained second fidelity network comprises:

   Each second fidelity network is trained using the sub-data corresponding to each fidelity level to obtain at least one trained second fidelity network.
4. The method according to claim 1, characterized in that the combining at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network comprises:

   Connecting the output end of each trained second fidelity network to one of the input ends of the first fidelity network;

   The input end of the first fidelity network and the input ends of each trained second fidelity network are used together as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain the multi-fidelity network.
5. The method according to claim 1, characterized in that the using the first fidelity data to train the multi-fidelity network to obtain a trained multi-fidelity network comprises:

   Using the multi-fidelity network as a generator network, and obtaining a corresponding discriminator network according to the generator network;

   Constructing a generative adversarial network based on the generator network and the discriminator network;

   Using the first fidelity data to train the generative adversarial network to obtain a trained generative adversarial network;

   A trained generator network is obtained from the trained generative adversarial network as the trained multi-fidelity network.
6. The method according to claim 1, characterized in that the method further comprises:

   Acquiring control parameters and first moment state parameters of the target nuclear reactor;

   Inputting the control parameter and the first moment state parameter into the trained multi-fidelity network to obtain the second moment state parameter of the target nuclear reactor;

   A simulation test result of the target nuclear reactor is obtained based on the state parameter at the second moment.
7. A multi-fidelity network construction device for nuclear reactor simulation testing, characterized in that the device comprises:

   an acquisition module, configured to acquire a first fidelity network according to first fidelity data of a sample nuclear reactor, and to acquire at least one second fidelity network according to second fidelity data of the sample nuclear reactor;

   A training module, configured to train at least one second fidelity network using the second fidelity data to obtain at least one trained second fidelity network;

   A combination module is used to combine at least one trained second fidelity network with the first fidelity network to obtain a multi-fidelity network, and use the first fidelity data to train the multi-fidelity network to obtain a trained multi-fidelity network; the trained multi-fidelity network is used to perform simulation testing on a target nuclear reactor.
8. The device according to claim 7, characterized in that the acquisition module is also used to acquire multiple data sets of the sample nuclear reactor; determine the accuracy of each data set, and select the data in the data set with the highest accuracy. as the first fidelity data; and among the multiple data sets, data other than the first fidelity data is used as the second fidelity data.
9. The device according to claim 7, characterized in that the acquisition module is further used to classify the second fidelity data into levels to obtain at least one fidelity level and sub-data corresponding to each fidelity level; obtain the corresponding second fidelity network according to the sub-data corresponding to each fidelity level; the number of the second fidelity networks is the same as the number of the fidelity levels;

   The training module is also used to respectively use the sub-data corresponding to each fidelity level to train each second fidelity network to obtain at least one trained second fidelity network.
10. The device according to claim 7 is characterized in that the combination module is also used to connect the output end of each trained second fidelity network to one of the input ends of the first fidelity network; the input end of the first fidelity network and the input end of each trained second fidelity network are used as the input end of the multi-fidelity network, and the output end of the first fidelity network is used as the output end of the multi-fidelity network to obtain the multi-fidelity network.
11. The device according to claim 7 is characterized in that the combination module is also used to use the multi-fidelity network as a generator network, and obtain a corresponding discriminator network based on the generator network; construct a generative adversarial network based on the generator network and the discriminator network; use the first fidelity data to train the generative adversarial network to obtain a trained generative adversarial network; obtain a trained generator network from the trained generative adversarial network as the trained multi-fidelity network.
12. The device according to claim 7, characterized in that the device further comprises:

    A test module is used to obtain control parameters and first-time state parameters of the target nuclear reactor; input the control parameters and the first-time state parameters into the trained multi-fidelity network to obtain second-time state parameters of the target nuclear reactor; and obtain simulation test results of the target nuclear reactor based on the second-time state parameters.
13. A computer device comprises a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of any one of the methods of claims 1 to 6 when executing the computer program.
14. A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.
15. A computer program product, comprising a computer program, characterized in that the computer program is executed by a processor The method of any one of claims 1 to 6 is implemented when executed.