Topological structure generation method and system based on variational auto-encoder

Abstract

The invention provides a topological structure generation method and system based on a variational auto-encoder, and relates to the technical field of topological structure generation, and the method comprises the steps: inputting an initial topological structure of a nuclear reactor into an encoder structure of the variational auto-encoder to obtain a first hidden variable; inputting the target performance of the nuclear reactor into the mapping model to obtain a second hidden variable; fusing the first hidden variable and the second hidden variable to obtain a fused hidden variable; performing random operation on the fused hidden variables to obtain a plurality of random hidden variables; and inputting the fused hidden variable and the plurality of random hidden variables into a decoder structure of a variational auto-encoder to obtain a target topological structure of the nuclear reactor.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear reactor topological structures, and in particular to a method and system for generating a topological structure based on a variational autoencoder. Background Art

[0002] As an important energy facility, nuclear reactors have high design complexity and strict safety requirements. The topology of a nuclear reactor directly affects the performance and efficiency of the reactor. Therefore, optimizing the topology is the key to improving the performance of a nuclear reactor. Traditional nuclear reactor design methods rely on expert experience and trial and error, which are time-consuming and costly. In addition, traditional methods may have limitations when exploring the design space, making it difficult to achieve the optimal solution.

[0003] Therefore, how to automatically generate topological structures that meet specific performance requirements and improve design efficiency is a problem that needs to be solved urgently. Summary of the invention

[0004] The main technical problem solved by the present invention is how to automatically generate a topological structure that meets specific performance requirements and improve design efficiency.

[0005] According to a first aspect, the present invention provides a method for generating a topological structure based on a variational autoencoder, comprising: inputting an initial topological structure of a nuclear reactor into the encoder structure of the variational autoencoder to obtain a first latent variable; inputting the target performance of the nuclear reactor into a mapping model to obtain a second latent variable; fusing the first latent variable and the second latent variable to obtain a fused latent variable; performing random operations on the fused latent variable to obtain multiple random latent variables; and inputting the fused latent variable and the multiple random latent variables into the decoder structure of the variational autoencoder to obtain the target topological structure of the nuclear reactor.

[0006] Furthermore, the random operation includes convolution, crossover, and mutation.

[0007] Furthermore, the step of inputting the initial topological structure of the nuclear reactor into the encoder structure of the variational autoencoder to obtain the first latent variable includes: representing the initial topological structure of the nuclear reactor as 80*80*80 pixels and inputting the pixels into the encoder structure of the variational autoencoder to obtain the first latent variable.

[0008] Furthermore, the encoder includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, and a fully connected layer. The convolution kernel of the first convolutional layer is 2, 2, 80, 48, and the step size is 2. The output is a feature mapping group of 40*40*48, which is then activated by the RELU function. The first pooling layer uses a maximum pooling operation of size 2*2, with a step size of 2, to obtain a feature mapping group of 20*20*48. The second convolutional layer uses convolution kernels of 2, 2, 48, 128, with a step size of 1, and zero padding is obtained. A feature mapping group of 19*19*128 is obtained, and then activated by the RELU function. The second pooling layer uses a maximum pooling operation of size 3*3 with a step size of 2 to obtain a feature mapping group of 9*9*128. The third convolutional layer uses a convolution kernel of 2,2,128,128 with a step size of 1 to obtain a feature mapping group of 8*8*128, which is then activated by the RELU function. The third pooling layer uses a maximum pooling operation of size 2*2 with a step size of 3 to obtain a feature mapping group of 2*2*128. The fully connected layer outputs a 1*512 dimensional tensor.

[0009] Furthermore, the method of inputting the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the target topological structure of the nuclear reactor includes: inputting the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the fused topological structure and multiple random topological structures of the nuclear reactor; judging the target performance of the nuclear reactor, the fused topological structure of the nuclear reactor, and multiple random topological structures based on the convolutional neural network model to obtain the degree of difference between the fused topological structure of the nuclear reactor and the target performance of the nuclear reactor, and the degree of difference between each random topological structure and the target performance of the nuclear reactor; and taking the topological structure with the smallest degree of difference between the fused topological structure and each random topological structure as the target topological structure.

[0010] Furthermore, the mapping model is a bidirectional recurrent neural network.

[0011] According to a second aspect, the present invention provides a system for generating a topological structure based on a variational autoencoder, comprising: an acquisition module, for inputting an initial topological structure of a nuclear reactor into an encoder structure of the variational autoencoder to obtain a first latent variable; A mapping module, used for inputting the target performance of the nuclear reactor into the mapping model to obtain a second latent variable; A fusion module, used for fusing the first latent variable and the second latent variable to obtain a fused latent variable; A random latent variable determination module is used to perform random operations on the fused latent variables to obtain multiple random latent variables; The target topology structure determination module is used to input the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the target topology structure of the nuclear reactor.

[0012] Furthermore, the random operation includes convolution, crossover, and mutation.

[0013] Furthermore, the encoder includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, and a fully connected layer. The convolution kernel of the first convolutional layer is 2, 2, 80, 48, and the step size is 2. The output is a feature mapping group of 40*40*48, which is then activated by the RELU function. The first pooling layer uses a maximum pooling operation of size 2*2, with a step size of 2, to obtain a feature mapping group of 20*20*48. The second convolutional layer uses convolution kernels of 2, 2, 48, 128, with a step size of 1, and zero padding is obtained. A feature mapping group of 19*19*128 is obtained, and then activated by the RELU function. The second pooling layer uses a maximum pooling operation of size 3*3 with a step size of 2 to obtain a feature mapping group of 9*9*128. The third convolutional layer uses a convolution kernel of 2,2,128,128 with a step size of 1 to obtain a feature mapping group of 8*8*128, which is then activated by the RELU function. The third pooling layer uses a maximum pooling operation of size 2*2 with a step size of 3 to obtain a feature mapping group of 2*2*128. The fully connected layer outputs a 1*512 dimensional tensor.

[0014] Furthermore, the target topology structure is determined to be further used for: Inputting the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the fused topological structure and multiple random topological structures of the nuclear reactor; Based on the convolutional neural network model, the target performance of the nuclear reactor, the fusion topology of the nuclear reactor, and multiple random topology structures are judged to obtain the difference between the fusion topology of the nuclear reactor and the target performance of the nuclear reactor, and the difference between each random topology structure and the target performance of the nuclear reactor; The topology with the smallest difference between the fusion topology and each random topology is taken as the target topology.

[0015] The present invention provides a method and system for generating a topological structure based on a variational autoencoder. The method comprises inputting an initial topological structure of a nuclear reactor into an encoder structure of a variational autoencoder to obtain a first latent variable; inputting a target performance of the nuclear reactor into a mapping model to obtain a second latent variable; fusing the first latent variable and the second latent variable to obtain a fused latent variable; performing random operations on the fused latent variable to obtain a plurality of random latent variables; and inputting the fused latent variable and the plurality of random latent variables into a decoder structure of a variational autoencoder to obtain a target topological structure of the nuclear reactor. The method can automatically generate a topological structure that meets specific performance requirements, thereby improving design efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a method for generating a topological structure based on a variational autoencoder provided in an embodiment of the present invention; Figure 2 A schematic diagram of an encoder provided by an embodiment of the present invention; Figure 3 A schematic diagram of a process for generating a target topology structure provided by an embodiment of the present invention; Figure 4 A schematic diagram of a system for generating a topological structure based on a variational autoencoder provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In an embodiment of the present invention, there is provided Figure 1 A method for generating a topological structure based on a variational autoencoder is shown, and the method for generating a topological structure based on a variational autoencoder comprises steps S1 to S5: Step S1, inputting the initial topological structure of the nuclear reactor into the encoder structure of the variational autoencoder to obtain the first latent variable.

[0018] The initial topology of the nuclear reactor refers to the initial layout or structure of the nuclear reactor system. The encoder structure is a part of the VAE that encodes the input data into a representation in the latent space. The layout of the nuclear reactor is represented as a pixelated three-dimensional image, which is then converted into the first latent variable through the encoder.

[0019] In some embodiments, the initial topological structure of the nuclear reactor can be represented as 80*80*80 pixels and input into the encoder structure of the variational autoencoder to obtain the first latent variable.

[0020] Figure 2A schematic diagram of an encoder provided for an embodiment of the present invention. In some embodiments, the encoder includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, and a fully connected layer. The convolution kernel of the first convolutional layer is 2, 2, 80, 48, and the step size is 2. The output is a feature mapping group of 40*40*48, which is then activated by a RELU function. The first pooling layer uses a maximum pooling operation of 2*2 with a step size of 2 to obtain a feature mapping group of 20*20*48. The second convolutional layer uses a convolution kernel of 2, 2, 48, 128. , the step size is 1, zero padding is used to obtain a feature mapping group of 19*19*128, and then the RELU function is used for activation. The second pooling layer uses a maximum pooling operation of size 3*3, with a step size of 2, to obtain a feature mapping group of 9*9*128. The third convolutional layer uses a convolution kernel of 2,2,128,128, with a step size of 1, to obtain a feature mapping group of 8*8*128, and then the RELU function is used for activation. The third pooling layer uses a maximum pooling operation of size 2*2, with a step size of 3, to obtain a feature mapping group of 2*2*128. The fully connected layer outputs a 1*512 dimensional tensor.

[0021] Step S2, inputting the target performance of the nuclear reactor into the mapping model to obtain a second latent variable.

[0022] The mapping model is a bidirectional recurrent neural network. The mapping model includes a forward RNN hidden layer and a backward RNN hidden layer. The two hidden layers are combined to extract the target performance characteristics and the second latent variable characteristics. The hidden state of the forward RNN hidden layer and the hidden state of the backward RNN hidden layer are updated as follows: represents the state at time t when forward hiding, represents the forward hidden state at time t-1, , , , , , Indicates deviation, Represents input.

[0023] The forward hidden state and the reverse hidden state are concatenated to obtain the hidden state Passed to the next bidirectional layer.

[0024] Target performance refers to the expected behavior or performance index of a nuclear reactor system under specific conditions. A bidirectional recurrent neural network (Bi-RNN) is used as a mapping model, and its forward and backward hidden layers are used to extract the characteristics of the target performance. Through the mapping model, the complex characteristics of the target performance can be extracted and converted into the second latent variable, so as to better associate it with the topological structure.

[0025] Step S3, fusing the first latent variable and the second latent variable to obtain a fused latent variable.

[0026] The fused latent variable may be a mixed Gaussian model. The fused latent variable is a new representation formed by combining the first latent variable and the second latent variable, and is usually used to comprehensively consider information from different sources. By using a mixed Gaussian model to fuse the first latent variable and the second latent variable, a richer and more complex latent variable representation can be obtained.

[0027] Step S4, performing random operations on the fused latent variables to obtain multiple random latent variables.

[0028] Random operations are a series of random transformations or operations on the fused latent variables to increase their diversity and richness. The random operations include convolution, crossover, and mutation. Increasing randomness can introduce more changes and diversity, which helps to generate more diverse and realistic topological structures.

[0029] Step S5, inputting the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the target topological structure of the nuclear reactor.

[0030] In some embodiments, the variational autoencoder is obtained by training. The variational autoencoder can be trained by maximizing the loss function and thereby minimizing the KL divergence.

[0031] in represents the loss function, where represents the KL divergence, is the reconstruction term, which represents the approximate posterior distribution of the potential distribution z Under this condition, the log-likelihood expected value of the reconstructed data x is obtained. is the approximate distribution of the latent variable z after given observed data x, is a regularization term that prevents overfitting and improves the generalization ability of the model, representing the approximate posterior distribution and the prior distribution of the latent variable z The KL divergence between . The KL divergence measures the difference between two probability distributions.

[0032] The decoder structure is a part of VAE that decodes the representation in the latent space into the original data. The fused latent variables and multiple random latent variables are input into the decoder to generate a topology corresponding to the target performance.

[0033] In some embodiments, step S5 may be performed by Figure 3 The method shown is implemented as follows, Figure 3 A schematic flow chart of a method for generating a target topology structure provided in an embodiment of the present invention, wherein the method for generating a target topology structure comprises steps S31 to S33: Step S31, inputting the fused latent variables and the multiple random latent variables into the decoder structure of the variational autoencoder to obtain the fused topological structure and the multiple random topological structures of the nuclear reactor.

[0034] The fusion latent variable and multiple random latent variables are used as input, and multiple possible nuclear reactor topologies are generated through the decoder structure, each latent variable corresponds to a different topology. By generating multiple random topologies, different possibilities can be explored, diversity can be increased, and more alternatives can be obtained to meet the target performance.

[0035] Step S32, based on the convolutional neural network model, the target performance of the nuclear reactor, the fusion topology structure of the nuclear reactor, and multiple random topology structures are judged to obtain the degree of difference between the fusion topology structure of the nuclear reactor and the target performance of the nuclear reactor, and the degree of difference between each random topology structure and the target performance of the nuclear reactor.

[0036] The convolutional neural network (CNN) model is used to judge the target performance, fusion topology, and multiple random topologies to obtain the degree of difference from the target performance. Each topology is evaluated using the trained CNN model to obtain the degree of difference from the target performance, thereby measuring the degree to which it meets the target performance.

[0037] Step S33: taking the topology with the smallest difference between the fused topology and each random topology as the target topology.

[0038] From the evaluation results of step S32, the topology with the smallest difference from the target performance among the fusion topology and the multiple random topologies is selected as the final target topology.

[0039] Based on the same inventive concept, Figure 4 A schematic diagram of a system for generating a topological structure based on a variational autoencoder provided in an embodiment of the present invention, wherein the system for generating a topological structure based on a variational autoencoder comprises: An acquisition module 41 is used to input the initial topological structure of the nuclear reactor into the encoder structure of the variational autoencoder to obtain a first latent variable; A mapping module 42, used for inputting the target performance of the nuclear reactor into the mapping model to obtain a second latent variable; A fusion module 43, used for fusing the first latent variable and the second latent variable to obtain a fused latent variable; A random latent variable determination module 44 is used to perform random operations on the fused latent variables to obtain multiple random latent variables; The target topology structure determination module 45 is used to input the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the target topology structure of the nuclear reactor.

Claims

1. A method for generating a topological structure based on a variational autoencoder, characterized in that: include: Inputting the initial topological structure of the nuclear reactor into the encoder structure of the variational autoencoder to obtain the first latent variable; Inputting the target performance of the nuclear reactor into the mapping model to obtain the second latent variable; The first latent variable and the second latent variable are fused to obtain a fused latent variable; Perform random operations on the fused latent variables to obtain multiple random latent variables; The fused latent variables and multiple random latent variables are input into the decoder structure of the variational autoencoder to obtain the target topology of the nuclear reactor.

2. The method for generating a topological structure based on a variational autoencoder according to claim 1, characterized in that: The random operations include convolution, crossover, and mutation.

3. The method for generating a topological structure based on a variational autoencoder according to claim 1, characterized in that: The step of inputting the initial topological structure of the nuclear reactor into the encoder structure of the variational autoencoder to obtain the first latent variable includes: representing the initial topological structure of the nuclear reactor as 80*80*80 pixels and inputting the pixels into the encoder structure of the variational autoencoder to obtain the first latent variable.

4. The method for generating a topological structure based on a variational autoencoder according to claim 3, characterized in that: The encoder includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, and a fully connected layer. The convolution kernel of the first convolutional layer is 2, 2, 80, 48, and the step size is 2. The output is a feature mapping group of 40*40*48, which is then activated by the RELU function. The first pooling layer uses a maximum pooling operation of size 2*2 and a step size of 2 to obtain a feature mapping group of 20*20*48. The second convolutional layer uses convolution kernels of 2, 2, 48, 128 and a step size of 1. Zero padding is used to obtain a feature mapping group of 19* A 19*128 feature mapping group is obtained, and then the RELU function is used for activation. The second pooling layer uses a maximum pooling operation of size 3*3 with a step size of 2 to obtain a 9*9*128 feature mapping group. The third convolutional layer uses a convolution kernel of 2,2,128,128 with a step size of 1 to obtain a 8*8*128 feature mapping group, which is then activated with the RELU function. The third pooling layer uses a maximum pooling operation of size 2*2 with a step size of 3 to obtain a 2*2*128 feature mapping group. The fully connected layer outputs a 1*512 dimensional tensor.

5. The method for generating a topological structure based on a variational autoencoder according to claim 1, characterized in that: The target topology structure of the nuclear reactor obtained by inputting the fused latent variables and the multiple random latent variables into the decoder structure of the variational autoencoder includes: Inputting the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the fused topological structure and multiple random topological structures of the nuclear reactor; Based on the convolutional neural network model, the target performance of the nuclear reactor, the fusion topology of the nuclear reactor, and multiple random topology structures are judged to obtain the difference between the fusion topology of the nuclear reactor and the target performance of the nuclear reactor, and the difference between each random topology structure and the target performance of the nuclear reactor; The topology with the smallest difference between the fusion topology and each random topology is taken as the target topology.

6. The method for generating a topological structure based on a variational autoencoder according to claim 1, characterized in that: The mapping model is a bidirectional recurrent neural network.

7. A system for generating topological structures based on variational autoencoders, characterized in that: include: An acquisition module, used for inputting the initial topological structure of the nuclear reactor into the encoder structure of the variational autoencoder to obtain a first latent variable; A mapping module, used for inputting the target performance of the nuclear reactor into the mapping model to obtain a second latent variable; A fusion module, used for fusing the first latent variable and the second latent variable to obtain a fused latent variable; A random latent variable determination module is used to perform random operations on the fused latent variables to obtain multiple random latent variables; The target topology structure determination module is used to input the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the target topology structure of the nuclear reactor.

8. The system for generating a topological structure based on a variational autoencoder according to claim 7, characterized in that: The random operations include convolution, crossover, and mutation.

9. The system for generating a topological structure based on a variational autoencoder according to claim 7, characterized in that: The encoder includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, and a fully connected layer. The convolution kernel of the first convolutional layer is 2, 2, 80, 48, and the step size is 2. The output is a feature mapping group of 40*40*48, which is then activated by the RELU function. The first pooling layer uses a maximum pooling operation of size 2*2 and a step size of 2 to obtain a feature mapping group of 20*20*48. The second convolutional layer uses convolution kernels of 2, 2, 48, 128 and a step size of 1. Zero padding is used to obtain a feature mapping group of 19* A 19*128 feature mapping group is obtained, and then the RELU function is used for activation. The second pooling layer uses a maximum pooling operation of size 3*3 with a step size of 2 to obtain a 9*9*128 feature mapping group. The third convolutional layer uses a convolution kernel of 2,2,128,128 with a step size of 1 to obtain a 8*8*128 feature mapping group, which is then activated with the RELU function. The third pooling layer uses a maximum pooling operation of size 2*2 with a step size of 3 to obtain a 2*2*128 feature mapping group. The fully connected layer outputs a 1*512 dimensional tensor.

10. The system for generating a topological structure based on a variational autoencoder according to claim 7, characterized in that: The target topology determination is also used for: Inputting the fused latent variables and multiple random latent variables into the decoder structure of the variational autoencoder to obtain the fused topological structure and multiple random topological structures of the nuclear reactor; Based on the convolutional neural network model, the target performance of the nuclear reactor, the fusion topology of the nuclear reactor, and multiple random topology structures are judged to obtain the difference between the fusion topology of the nuclear reactor and the target performance of the nuclear reactor, and the difference between each random topology structure and the target performance of the nuclear reactor; The topology with the smallest difference between the fusion topology and each random topology is taken as the target topology.

Images