A method, system and device for diagnosing the migration of a nuclear power plant

Abstract

The present invention provides a method, system and device for diagnosing the migration of a nuclear power plant. The method includes: collecting the operation data of the nuclear power plant under various faults; dividing the operation data into different domains, and selecting the operation data corresponding to two different types of operation conditions as the source domain data and the target domain data; generating a time convolutional capsule feature extraction network; using a label predictor, a domain discriminator and a domain adaptation module to correct the time convolutional capsule feature extraction network to generate a time convolutional capsule feature extraction network with locked parameters; inputting the source domain data into the time convolutional capsule feature extraction network with locked parameters to extract the source domain features; inputting the target domain data into the time convolutional capsule feature extraction network to extract the target domain features; inputting the source domain features and the target domain features into the domain discriminator and the domain adaptation module to determine the shared feature domain. The present invention can improve the feature migration effect when the domain differences are large.

Description

Technical Field

[0001] The present invention relates to the field of nuclear power plant migration diagnosis, and particularly to a nuclear power plant migration diagnosis method, system and device. Background Art

[0002] Nuclear power plants pose a risk of nuclear leakage, which causes short-term irreversible damage to the ecological environment and extremely serious harm to human health in case of contact during an accident. Given the potentially severe consequences of nuclear accidents, high attention should be paid to the safety issues of nuclear power plants.

[0003] With the rise of the computer intelligent industry, data-driven fault diagnosis models have also shown many advantages. For example, they do not rely on the prior knowledge of the system, such as mathematical models and expert experience; they use monitored data from different sources and of different types as the basis, and utilize various data mining techniques to obtain the useful information hidden therein, characterize the normal and fault modes of system operation, and thus achieve the purpose of detection and diagnosis. Among them, traditional machine learning methods are typical of data-driven fault diagnosis models.

[0004] For the fault diagnosis of nuclear power plants, the application of traditional machine learning methods requires a large number of labeled data samples for the classifier to learn. By finding the mapping relationship between the data samples and the labels, a prediction model is constructed, and test data is input into this model to predict the corresponding fault labels, so as to detect the faults. This requires the ideal situation that the test set and the training set must be of the same distribution. However, in actual situations, since the operating conditions (temperature, rotation speed, etc.) faced by nuclear power plants change with time, the obtained fault data does not satisfy the same distribution as the training data used in traditional machine learning, resulting in the failure of the classifier trained with the fault data of the training set with known labels to diagnose the fault data under actual working conditions, and thus there are potential safety hazards. Therefore, traditional machine learning methods are not applicable to actual fault diagnosis problems.

[0005] Transfer learning can find a mapping to map the highly relevant training set data and fault data into a high-dimensional space, minimize the distribution difference between the data, and at the same time retain their internal features to the greatest extent, so that the training set data and the fault data tend to the same distribution. In this case, retraining the classifier can well achieve the fault diagnosis of nuclear power rotating machinery. Therefore, the application of transfer learning methods can solve the drawbacks of data-driven fault diagnosis models in the field of nuclear power rotating machinery.

[0006] The shallow transfer model measures the marginal distribution difference and conditional distribution difference between the source domain and the target domain through the Maximum Mean Discrepancy (MMD), and uses the kernel function mapping to reduce the domain feature distribution difference. Lan Yutao conducts cross-condition fault diagnosis based on calculating the marginal distribution difference between domains through transfer component analysis, and Liu Yingdong conducts transfer diagnosis based on the joint distribution of the margin and conditions.

[0007] With the development of deep learning theory, the deep transfer model formed by combining the domain adaptation module with the deep learning model has been widely used. It deeply abstracts data features and combines domain adaptation for distribution adaptation. Lu combines the Deep Neural Networks (DNN) with marginal distribution adaptation to achieve fault diagnosis of rotating machinery; Shen Changqing uses an improved ResNet-50 network to construct a multi-scale feature extractor combined with conditional distribution adaptation to achieve feature transfer diagnosis of train bearings. Li conducts transfer fault diagnosis on rotating machinery based on a deep convolutional neural network. Due to the mutual confrontation inspiration of the Generative Adversarial Network (GAN), the adversarial deep transfer model has become mainstream in the task of transfer learning. Among them, Guo Liang proposed the Deep Convolutional Transfer Learning Network (DCTLN), which is based on the adversarial training of a convolutional neural network and a domain discriminator, and at the same time adds a domain adaptation module to achieve fault diagnosis of rotating machinery; Zhu Yan proposed an improved joint distribution difference based on DCTLN to optimize the fault diagnosis model. However, in the problem of large domain differences, the transfer accuracy of such models decreases significantly, and the feature transfer effect is poor. Summary of the Invention

[0008] The purpose of the present invention is to provide a transfer diagnosis method, system and device for a nuclear power plant to solve the problem of poor feature transfer effect when the domain difference is large.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] A transfer diagnosis method for a nuclear power plant, comprising:

[0011] Collect the operation data of the nuclear power plant under various faults corresponding to the full-range simulator simulation;

[0012] Divide the operation data into different domains according to different types of operating conditions, and select the operation data corresponding to two different types of operating conditions as the source domain data and the target domain data; the fault label categories of the source domain data and the target domain data are the same;

[0013] Train a deep learning model using the source domain data and fault labels to generate a temporal convolutional capsule feature extraction network; the deep learning model is a temporal convolutional capsule network;

[0014] Use a label predictor, a domain discriminator, and a domain adaptation module to correct the temporal convolutional capsule feature extraction network to generate a parameter-locked temporal convolutional capsule feature extraction network; the label predictor is a support vector machine model; the domain discriminator is constructed by a temporal convolutional capsule network, where the source domain label is 0 and the target domain label is 1 in the domain discriminator; the domain adaptation module includes marginal distribution adaptation based on maximum mean discrepancy and conditional distribution adaptation based on conditional maximum mean discrepancy;

[0015] Input the source domain data into the parameter-locked temporal convolutional capsule feature extraction network to extract source domain features;

[0016] Input the target domain data into the temporal convolutional capsule feature extraction network to extract target domain features;

[0017] Input the source domain features and the target domain features into the domain discriminator and the domain adaptation module to determine the shared feature domain.

[0018] Optionally, the using a label predictor, a domain discriminator, and a domain adaptation module to correct the temporal convolutional capsule feature extraction network to generate a parameter-locked temporal convolutional capsule feature extraction network specifically includes:

[0019] Input the source domain data and the target domain data into the temporal convolutional capsule feature extraction network to determine the output layer result; the output layer result is the loss function of the mapping between the source domain data features and the source domain labels;

[0020] Input the output layer result into the label predictor to predict the fault mode to which the source domain data belongs and output the prediction loss function;

[0021] Input the prediction loss function into the domain discriminator to lock the parameters of the temporal convolutional capsule feature extraction network, change the parameters of the domain discriminator, and output the discriminant loss function;

[0022] Based on the domain adaptation module, fit the capsule network layer of the penultimate layer and the temporal convolutional layer of the second-to-last layer of the temporal convolutional capsule feature extraction network, and quantitatively calculate the marginal distribution and conditional distribution differences of the source domain features and the target domain features in the deep network layer;

[0023] Use the output layer result, the prediction loss function, and the discriminant loss function to correct the fitted temporal convolutional capsule feature extraction network to generate a parameter-locked temporal convolutional capsule feature extraction network.

[0024] Optionally, before using the output layer result, the prediction loss function, and the discriminant loss function to correct the fitted temporal convolutional capsule feature extraction network to generate a parameter-locked temporal convolutional capsule feature extraction network, it further includes:

[0025] Adding a penalty factor before the output layer result, the prediction loss function, and the discriminant loss function to determine the total loss function.

[0026] Optionally, the marginal distribution adaptation based on the maximum mean distance specifically includes:

[0027] Using the formula to calculate the domain marginal distribution difference; where X is the source domain data; Y is the target domain data; n is the number of source domain data; t is the number of target domain data; Φ(·) is the mapping function; x i is the i-th data feature of the source domain; y i is the i-th data feature of the target domain; H is the marginal distribution distance measured by mapping the data into the reproducing Hilbert space by Φ(·).

[0028] Optionally, the conditional distribution adaptation based on the conditional maximum mean distance specifically includes:

[0029] Using the formula to calculate the domain conditional distribution difference; where X is the source domain data; Y is the target domain data; C is the fault label domain category; ω c is the prior probability under different fault modes of the nuclear power plant, that is, the proportion of the fault label c; n s (c) is the number of source domains under the fault label c; n t (c) is the number of target domains under the fault label c; Φ(·) is the kernel function mapping; x s is the s-th data feature of the source domain; x t is the t-th data feature of the source domain; is the source domain under the fault label c; is the target domain under the fault label c.

[0030] A nuclear power plant migration diagnosis system includes:

[0031] An operation data acquisition module, configured to acquire operation data of the nuclear power plant under various faults corresponding to the full-scope simulator simulation;

[0032] A domain division module, which is used to divide the operation data into different domains according to different types of operating conditions, and select the operation data corresponding to two different types of operating conditions as the source domain data and the target domain data; the fault label categories of the source domain data and the target domain data are the same;

[0033] A time convolutional capsule feature extraction network generation module, which is used to train a deep learning model using the source domain data and fault labels to generate a time convolutional capsule feature extraction network; the deep learning model is a time convolutional capsule network;

[0034] A parameter locking module, which is used to correct the time convolutional capsule feature extraction network using a label predictor, a domain discriminator, and a domain adaptation module to generate a parameter-locked time convolutional capsule feature extraction network; the label predictor is a support vector machine model; the domain discriminator is constructed by a time convolutional capsule network, in which the source domain label is 0 and the target domain label is 1; the domain adaptation module includes marginal distribution adaptation based on maximum mean discrepancy and conditional distribution adaptation based on conditional maximum mean discrepancy;

[0035] A source domain feature extraction module, which is used to input the source domain data into the parameter-locked time convolutional capsule feature extraction network to extract source domain features;

[0036] A target domain feature extraction module, which is used to input the target domain data into the time convolutional capsule feature extraction network to extract target domain features;

[0037] A shared feature domain determination module, which is used to input the source domain features and the target domain features into the domain discriminator and the domain adaptation module to determine the shared feature domain.

[0038] Optionally, the parameter locking module specifically includes:

[0039] An output layer result determination unit, which is used to input the source domain data and the target domain data into the time convolutional capsule feature extraction network to determine the output layer result; the output layer result is the loss function of the mapping between the source domain data features and the source domain labels;

[0040] A predicted loss function output unit, which is used to input the output layer result into the label predictor to predict the fault mode to which the source domain data belongs and output the predicted loss function;

[0041] A discriminant loss function output unit, which is used to input the predicted loss function into the domain discriminator to lock the parameters of the time convolutional capsule feature extraction network, change the parameters of the domain discriminator, and output the discriminant loss function;

[0042] A fitting unit, configured to fit the capsule network layer of the last layer and the temporal convolutional layer of the second last layer of the temporal convolutional capsule feature extraction network based on the domain adaptation module, and quantitatively calculate the differences in the marginal distributions and conditional distributions of the source domain features and target domain features in the deep network layer;

[0043] A parameter-locked temporal convolutional capsule feature extraction network generation unit, configured to correct the fitted temporal convolutional capsule feature extraction network by using the output layer result, the prediction loss function, and the discriminant loss function, and generate a parameter-locked temporal convolutional capsule feature extraction network.

[0044] Optionally, it further includes:

[0045] A total loss function determination module, configured to add penalty factors before the output layer result, the prediction loss function, and the discriminant loss function, and determine a total loss function.

[0046] An electronic device, including a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above nuclear power plant migration diagnosis method.

[0047] A computer-readable storage medium, characterized in that it stores a computer program, and when the computer program is executed by a processor, the above nuclear power plant migration diagnosis method is implemented.

[0048] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention: The present invention provides a nuclear power plant migration diagnosis method, system and device, which uses a domain adversarial network for pre-training and locks the source domain network parameters to generate a parameter-locked temporal convolutional capsule feature extraction network. The source domain features and target domain features of the source domain data and target data are extracted through the parameter-locked temporal convolutional capsule feature extraction network and the temporal convolutional capsule feature extraction network, realizing secondary domain confrontation. The present invention enhances the deep features when the domain differences are too large through step-by-step domain confrontation, and improves the feature migration effect when the domain differences are large. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0050] Figure 1 It is a flowchart of the nuclear power plant migration diagnosis method provided by the present invention;

[0051] Figure 2 The flowchart of fault diagnosis based on distribution domain adversarial provided by the present invention. Specific implementation mode

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0053] The purpose of the present invention is to provide a nuclear power plant migration diagnosis method, system and device to improve the feature migration effect when the domain differences are large.

[0054] To make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation modes.

[0055] Embodiment 1

[0056] As Figure 1 shown, the present invention provides a nuclear power plant migration diagnosis method, including:

[0057] Step 101: Collect the operation data of the nuclear power plant corresponding to various faults in the full-scope simulator simulation.

[0058] Step 102: Divide the operation data into different domains according to different types of operation conditions, and select the operation data corresponding to two different types of operation conditions as the source domain data and the target domain data; the fault label categories of the source domain data and the target domain data are the same.

[0059] In practical applications, the data feature dimension under the source domain conditions is m dimensions, and the number of samples is n, obtaining an m×n data matrix. The data feature dimension under the target domain conditions is m dimensions, and the number of samples is t (t << n), obtaining an m×t data matrix.

[0060] Step 103: Use the source domain data and fault labels to train a deep learning model to generate a time convolutional capsule feature extraction network; the deep learning model is a time convolutional capsule network.

[0061] In practical applications, the deep learning model in the present invention uses a time convolutional capsule network. Except for the input layer and the output layer, the model also has 3 layers of time convolutional layers and 1 layer of capsule network layer. The output signal of each layer of the network serves as the input signal of the next layer. The input layer is the input data of m rows and n columns, and the output layer is the fault diagnosis result of m rows and 1 column.

[0062] Step 104: Use the label predictor, domain discriminator, and domain adaptation module to correct the temporal convolutional capsule feature extraction network to generate a temporal convolutional capsule feature extraction network with locked parameters; the label predictor is a support vector machine model; the domain discriminator is constructed by a temporal convolutional capsule network, where the source domain label is 0 and the target domain label is 1 in the domain discriminator; the domain adaptation module includes marginal distribution adaptation based on the maximum mean discrepancy and conditional distribution adaptation based on the conditional maximum mean discrepancy.

[0063] In practical applications, Step 104 specifically includes: Input the source domain data and the target domain data into the temporal convolutional capsule feature extraction network to determine the output layer result; the output layer result is the loss function of the mapping between the source domain data features and the source domain labels; input the output layer result into the label predictor to predict the fault mode to which the source domain data belongs and output the prediction loss function; input the prediction loss function into the domain discriminator to lock the parameters of the temporal convolutional capsule feature extraction network, change the parameters of the domain discriminator, and output the discriminant loss function; based on the domain adaptation module, fit the capsule network layer of the penultimate layer and the temporal convolutional layer of the second-to-last layer of the temporal convolutional capsule feature extraction network, and quantitatively calculate the marginal distribution and conditional distribution differences between the source domain features and the target domain features in the deep network layer; use the output layer result, the prediction loss function, and the discriminant loss function to correct the fitted temporal convolutional capsule feature extraction network to generate a temporal convolutional capsule feature extraction network with locked parameters.

[0064] In practical applications, before using the output layer result, the prediction loss function, and the discriminant loss function to correct the fitted temporal convolutional capsule feature extraction network to generate a temporal convolutional capsule feature extraction network with locked parameters, it also includes: adding penalty factors before the output layer result, the prediction loss function, and the discriminant loss function to determine the total loss function.

[0065] Adding penalty factors before each loss is used to assign the weight coefficients of each loss, and the total loss is used to correct the parameters of the feature extraction network. This makes the output source domain features have an excellent mapping relationship with the labels, the source domain features and the target domain features are shared, and the overall difference between the two domains becomes smaller.

[0066] In practical applications, in calculating the domain marginal distribution difference, the marginal distribution distance metric considers using MMD. The marginal distribution adaptation based on the maximum mean discrepancy specifically includes:

[0067] Using the formula to calculate the domain marginal distribution difference; where X is the source domain data; Y is the target domain data; n is the number of source domain data; t is the number of target domain data; Φ(·) is the mapping function; xi is the i-th data feature of the source domain; y i is the i-th data feature of the target domain; H is the marginal distribution distance, which is measured by mapping the data into a reproducing Hilbert space by Φ(·).

[0068] In practical applications, for calculating the domain conditional distribution difference, it is proposed to use the Conditional Maximum Mean Discrepancy (CMMD) as the conditional difference metric. The conditional distribution adaptation based on the conditional maximum mean distance specifically includes:

[0069] Using the formula to calculate the domain conditional distribution difference; where, X is the source domain data; Y is the target domain data; C is the fault label domain category; ω c is the prior probability under different fault modes of the nuclear power plant, that is, the proportion of the fault label c; n s (c) is the number of source domains under the fault label c; n t (c) is the number of target domains under the fault label c; Φ(·) is the kernel function mapping; x s is the s-th data feature of the source domain; x t is the t-th data feature of the source domain; is the source domain under the fault label c; is the target domain under the fault label c.

[0070] Statistically calculate the prior probability ω under different fault modes of the nuclear power plant c and incorporate it into the CMMD distance. Compared with the MMD distance, the conditional distribution can be optimized through domain experience and the quality of shared feature extraction can be enhanced.

[0071] In practical applications, to minimize the fault mode prediction error rate, a relevant loss function is obtained to correct the temporal convolutional capsule feature extraction network. The label predictor selects a traditional machine learning model, and in the present invention, a support vector machine model is adopted.

[0072] In practical applications, in the present invention, the domain discriminator is constructed by a temporal convolutional capsule network. The source domain label is 0, and the target domain label is 1. Lock the parameters of the feature extraction network, change the parameters of the domain discriminator, effectively classify the input data features to output the domain label, and lock the network parameters of the domain discriminator after the result is stable. To make the domain discriminator unable to accurately distinguish the feature domain category, an output loss function is used to correct the feature extraction network.

[0073] In practical applications, the capsule network layer of the penultimate layer of the model and the temporal convolutional layer of the antepenultimate layer are fitted with a domain adaptation module, and the differences in the marginal distributions and conditional distributions of the source domain and target domain features in the deep network layer are quantitatively calculated. To minimize the differences between domains, the output loss function corrects the feature extraction network.

[0074] Figure 2 The flowchart of fault diagnosis based on distribution domain adversariality provided by the present invention is as Figure 2 shown. Through continuous iterative loops in step 104, the fault diagnosis rate is finally guaranteed within extremely small fluctuations, and the parameter representations of the feature extraction network and the domain discriminator are highly convergent, and the training of the source domain feature extraction network in the first step is completed.

[0075] Since the debugging of the network parameters of the above formulas has reached the optimal and stable state, the feature extraction network trained in the first step is parameter-locked and only the source domain features are input, and then the network is copied for feature extraction of the target domain, and feature extraction is separately implemented for data in different domains.

[0076] In practical applications, the source domain features and target domain features extracted by the two networks are input into the domain discriminator and the domain adaptation module for training. Step 104 is repeated, and the target domain network is corrected until the feature extraction network for the target domain is highly convergent, indicating that the second step is completed; the target domain network is a temporal convolutional capsule feature extraction network.

[0077] The first step is to input the source domain data and the target domain data, but only the source domain labels are available. The domain discriminator makes the features extracted by the network shared, and the domain adaptation module makes the differences between the two domains smaller (it also performs a mapping transformation on the features, which will also make the features shared between the two domains). Therefore, the network model trained in the first step itself outputs a shared feature domain for the input source domain and target domain features.

[0078] The second step is an optimization of the first step. Since the first step uses one network to map the source domain and target domain features, the effect of predicting the target domain by training a classifier using the migrated source domain features and the corresponding labels after migration is not so good. Therefore, two networks are used to separately extract features from data in different domains, and then domain discrimination and domain adaptation correction are performed to obtain a better shared feature domain.

[0079] Step 105: Input the source domain data into the parameter-locked temporal convolutional capsule feature extraction network to extract source domain features.

[0080] Step 106: Input the target domain data into the temporal convolutional capsule feature extraction network to extract target domain features.

[0081] Step 107: Input the source domain features and the target domain features into the domain discriminator and the domain adaptation module to determine the shared feature domain.

[0082] In practical applications, the source domain features and the target domain features extracted by different networks can form a shared feature domain with better migration effect, and the classifier is trained with the migrated source domain features and the corresponding labels to predict the fault mode to which the target domain data belongs.

[0083] The present invention uses a temporal convolutional capsule network to extract features from the source domain and target domain data, extracts the temporal information of the data through the temporal convolutional kernel and reduces the computational difficulty, and then mines the deep vector features of the data through the capsule network, which can maximize the utilization of the data features in the operation information and highlight the temporality and vectority of the data on the basis of the feature extraction by the neural network.

[0084] The present invention uses a label predictor to test the effectiveness of the extracted source domain features, a domain discriminator to test the sharing of the extracted features, and a domain adaptation module to measure the similarity degree of the extracted features. The above three modules respectively output loss functions to optimize the feature extraction network.

[0085] The present invention proposes a CMMD distance formula. By considering the situation that the conditional distribution weights are uneven due to different probabilities of different fault modes in engineering, the prior probability ω of different fault modes of the nuclear power plant is statistically calculated c and incorporated into the CMMD distance, and the accuracy is effectively improved in the calculation of the data conditional distribution difference.

[0086] In the present invention, the domain discriminator and the feature extraction network are in a state of mutual confrontation, and the network parameters of one party are stabilized to optimize the network parameters of the other party, and iterative confrontation is realized for strengthening.

[0087] The present invention proposes a method of step-by-step domain confrontation, and uses exclusive feature extraction networks for the source domain and target domain data respectively. On the basis of initially pre-training the source domain feature extraction network, the network is copied for feature extraction of the target domain data, and the network parameters of the source domain feature extraction network are locked, and the domain discriminator and the domain adaptation module are used to train the target domain feature extraction network, so as to effectively improve the migration accuracy in the case of large domain differences by separately mapping the domain data.

[0088] Embodiment 2

[0089] In order to execute the method corresponding to the above Embodiment 1 to achieve the corresponding functions and technical effects, a nuclear power plant migration diagnosis system is provided below.

[0090] A nuclear power plant migration diagnosis system includes:

[0091] An operation data acquisition module, configured to acquire operation data under various faults corresponding to the full-scope simulator simulation of a nuclear power plant.

[0092] A domain division module, configured to divide the operation data into different domains according to different types of operation conditions, and select operation data corresponding to two different types of operation conditions as source domain data and target domain data; the fault label categories of the source domain data and the target domain data are the same.

[0093] A temporal convolutional capsule feature extraction network generation module, configured to train a deep learning model by using the source domain data and fault labels to generate a temporal convolutional capsule feature extraction network; the deep learning model is a temporal convolutional capsule network.

[0094] A parameter locking module, configured to correct the temporal convolutional capsule feature extraction network by using a label predictor, a domain discriminator, and a domain adaptation module to generate a parameter-locked temporal convolutional capsule feature extraction network; the label predictor is a support vector machine model; the domain discriminator is constructed by a temporal convolutional capsule network, in which the source domain label is 0 and the target domain label is 1; the domain adaptation module includes marginal distribution adaptation based on maximum mean discrepancy and conditional distribution adaptation based on conditional maximum mean discrepancy.

[0095] In practical applications, the parameter locking module specifically includes: an output layer result determination unit, configured to input the source domain data and the target domain data into the temporal convolutional capsule feature extraction network to determine an output layer result; the output layer result is a loss function of the mapping between the source domain data features and the source domain labels; a predicted loss function output unit, configured to input the output layer result into the label predictor to predict the fault mode to which the source domain data belongs and output a predicted loss function; a discriminant loss function output unit, configured to input the predicted loss function into the domain discriminator to lock the parameters of the temporal convolutional capsule feature extraction network, change the parameters of the domain discriminator, and output a discriminant loss function; a fitting unit, configured to perform fitting on the capsule network layer of the last layer but one and the temporal convolutional layer of the second last layer of the temporal convolutional capsule feature extraction network based on the domain adaptation module, and quantitatively calculate the differences in the marginal distribution and conditional distribution of the source domain features and the target domain features in the deep network layer; a parameter-locked temporal convolutional capsule feature extraction network generation unit, configured to correct the fitted temporal convolutional capsule feature extraction network by using the output layer result, the predicted loss function, and the discriminant loss function to generate a parameter-locked temporal convolutional capsule feature extraction network.

[0096] In practical applications, it further includes: a total loss function determination module, which is used to add penalty factors before the output layer result, the prediction loss function, and the discriminant loss function to determine the total loss function.

[0097] A source domain feature extraction module, which is used to input the source domain data into the time convolutional capsule feature extraction network with parameters locked to extract source domain features.

[0098] A target domain feature extraction module, which is used to input the target domain data into the time convolutional capsule feature extraction network to extract target domain features.

[0099] A shared feature domain determination module, which is used to input the source domain features and the target domain features into the domain discriminator and the domain adaptation module to determine the shared feature domain.

[0100] Embodiment III

[0101] An embodiment of the present invention provides an electronic device including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the nuclear power plant migration diagnosis method provided in Embodiment I.

[0102] In practical applications, the above-mentioned electronic device may be a server.

[0103] In practical applications, the electronic device includes: at least one processor, a memory, a bus, and a communication interface.

[0104] Wherein: the processor, the communication interface, and the memory complete communication with each other through the communication bus.

[0105] The communication interface is used to communicate with other devices.

[0106] The processor is used to execute the program, specifically, it can execute the method described in the above-mentioned embodiment.

[0107] Specifically, the program may include program code, and the program code includes computer operation instructions.

[0108] The processor may be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention. One or more processors included in the electronic device may be of the same type of processor, such as one or more CPUs; or they may be of different types of processors, such as one or more CPUs and one or more ASICs.

[0109] A memory for storing programs. The memory may include high-speed RAM memory and may also include non-volatile memory, such as at least one disk memory.

[0110] Based on the description of the above embodiments, an embodiment of the present application provides a storage medium having computer program instructions stored thereon, and the computer program instructions can be executed by a processor to implement the method described in any embodiment.

[0111] The nuclear power plant migration diagnosis system provided by the embodiments of the present application exists in various forms, including but not limited to:

[0112] (1) Mobile communication devices: The characteristics of such devices are that they have mobile communication functions and mainly aim to provide voice and data communication. Such terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones, etc.

[0113] (2) Ultra-mobile personal computer devices: Such devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access performance. Such terminals include: PDA, MID, and UMPC devices, etc., such as iPad.

[0114] (3) Portable entertainment devices: Such devices can display and play multimedia content. Such devices include: audio and video players (such as iPod), handheld game consoles, e-books, and smart toys and portable vehicle navigation devices.

[0115] (4) Other electronic devices with data interaction functions.

[0116] So far, specific embodiments of the present subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing may be advantageous.

[0117] The systems, devices, modules, or units illustrated in the above embodiments may be specifically implemented by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0118] For the convenience of description, when describing the above device, it is divided into various units according to functions and described separately. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware. Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0120] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0122] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0123] The memory may include non-permanent memory in the form of computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory is an example of computer-readable media.

[0124] Computer-readable media includes both permanent and non-permanent, removable and non-removable media implemented by any method or technology for storing information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM),

[0125] digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices

[0126] or any other non-transitory medium that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0127] It should also be noted that the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising the element.

[0128] This application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. This application may also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

[0129] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0130] In this article, specific examples are used to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method of the present invention and its core idea. At the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A migration diagnosis method for a nuclear power plant, characterized in that, it includes: Collect the operation data of the nuclear power plant under various faults corresponding to the full - range simulator simulation; Divide the operation data into different domains according to different types of operation conditions, and select the operation data corresponding to two different types of operation conditions as the source - domain data and the target - domain data; The fault label categories of the source - domain data and the target - domain data are the same; Use the source - domain data and the fault labels to train a deep - learning model to generate a temporal convolutional capsule feature extraction network; The deep - learning model is a temporal convolutional capsule network; Use a label predictor, a domain discriminator, and a domain adaptation module to correct the temporal convolutional capsule feature extraction network to generate a temporal convolutional capsule feature extraction network with locked parameters; The label predictor is a support vector machine model; the domain discriminator is constructed by a temporal convolutional capsule network, where the source - domain domain label is 0 and the target - domain domain label is 1 in the domain discriminator; the domain adaptation module includes marginal distribution adaptation based on the maximum mean discrepancy and conditional distribution adaptation based on the conditional maximum mean discrepancy; Input the source - domain data into the temporal convolutional capsule feature extraction network with locked parameters to extract source - domain features; Input the target - domain data into the temporal convolutional capsule feature extraction network to extract target - domain features; Input the source - domain features and the target - domain features into the domain discriminator and the domain adaptation module to determine the shared feature domain.

2. The migration diagnosis method for a nuclear power plant according to claim 1, characterized in that, The step of using a label predictor, a domain discriminator, and a domain adaptation module to correct the temporal convolutional capsule feature extraction network to generate a temporal convolutional capsule feature extraction network with locked parameters specifically includes: Input the source - domain data and the target - domain data into the temporal convolutional capsule feature extraction network to determine the output - layer result; the output - layer result is the loss function of the mapping between the source - domain data features and the source - domain domain labels; Input the output - layer result into the label predictor to predict the fault mode to which the source - domain data belongs and output the prediction loss function; Input the prediction loss function into the domain discriminator to lock the parameters of the temporal convolutional capsule feature extraction network, change the parameters of the domain discriminator, and output the discriminant loss function; Based on the domain adaptation module, fit the capsule network layer of the penultimate layer and the temporal convolutional layer of the second - last layer of the temporal convolutional capsule feature extraction network, and quantitatively calculate the differences in the marginal distribution and conditional distribution of the source - domain features and the target - domain features in the deep - network layer; Use the output - layer result, the prediction loss function, and the discriminant loss function to correct the fitted temporal convolutional capsule feature extraction network to generate a temporal convolutional capsule feature extraction network with locked parameters.

3. The migration diagnosis method for a nuclear power plant according to claim 2, characterized in that, Before using the output layer result, the prediction loss function, and the discriminant loss function to correct the fitted temporal convolutional capsule feature extraction network to generate a temporal convolutional capsule feature extraction network with locked parameters, the following steps are also included: Add a penalty factor before the output layer result, the prediction loss function, and the discriminant loss function to determine the total loss function.

4. The nuclear power plant migration diagnosis method according to claim 1, wherein, the marginal distribution adaptation based on the maximum mean discrepancy specifically includes: Using the formula to calculate the difference in the marginal distribution of the domain edge; where X is the source domain data; Y is the target domain data; n is the number of source domain data; t is the number of target domain data; Φ(·) is a mapping function; x i is the i-th data feature of the source domain; y j is the j-th data feature of the target domain; H is the marginal distribution distance, which is measured by mapping the data into the reproducing Hilbert space by Φ(·).

5. The nuclear power plant migration diagnosis method according to claim 1, wherein, the conditional distribution adaptation based on the conditional maximum mean discrepancy specifically includes: Using the formula calculate the difference in the conditional distribution of domain conditions; where X is the source domain data; Y is the target domain data; C is the category of the fault label domain; ω c is the prior probability under different fault modes of the nuclear power plant, that is, the proportion of the fault label c; n s (c) is the number of source domains under the fault label c; is the number of target domains under the fault label c; Φ(·) is the kernel function mapping; x s is the s-th data feature of the source domain; x t is the t-th data feature of the source domain; is the source domain under the fault label c; is the target domain under the fault label c.

6. A nuclear power plant migration diagnosis system, wherein, it includes: An operation data acquisition module, configured to acquire operation data of the nuclear power plant under various faults corresponding to a full-scope simulator simulation; A domain division module, configured to divide the operation data into different domains according to different types of operation conditions, and select operation data corresponding to two different types of operation conditions as source domain data and target domain data; the fault label categories of the source domain data and the target domain data are the same; A temporal convolutional capsule feature extraction network generation module, configured to train a deep learning model using the source domain data and the fault label to generate a temporal convolutional capsule feature extraction network; The deep learning model is a temporal convolutional capsule network; A parameter locking module, configured to correct the temporal convolutional capsule feature extraction network using a label predictor, a domain discriminator, and a domain adaptation module to generate a temporal convolutional capsule feature extraction network with locked parameters; The label predictor is a support vector machine model; the domain discriminator is constructed by a temporal convolutional capsule network, where the source domain label is 0 and the target domain label is 1 in the domain discriminator; the domain adaptation module includes marginal distribution adaptation based on the maximum mean discrepancy and conditional distribution adaptation based on the conditional maximum mean discrepancy; A source domain feature extraction module, configured to input the source domain data into the temporal convolutional capsule feature extraction network with locked parameters to extract source domain features; A target domain feature extraction module, configured to input the target domain data into the temporal convolutional capsule feature extraction network to extract target domain features; A shared feature domain determination module, configured to input the source domain features and the target domain features into the domain discriminator and the domain adaptation module to determine the shared feature domain.

7. The nuclear power plant migration diagnosis system according to claim 6, wherein, the parameter locking module specifically includes: An output layer result determination unit, configured to input the source domain data and the target domain data into the temporal convolutional capsule feature extraction network to determine the output layer result; the output layer result is the loss function of the mapping between the source domain data features and the source domain labels; A prediction loss function output unit, configured to input the output layer result into the label predictor to predict the fault mode to which the source domain data belongs and output the prediction loss function; A discriminative loss function output unit, configured to input the prediction loss function into a domain discriminator, lock the parameters of the temporal convolutional capsule feature extraction network, change the parameters of the domain discriminator, and output a discriminative loss function; A fitting unit, configured to perform fitting on the capsule network layer of the last layer and the temporal convolutional layer of the second last layer of the temporal convolutional capsule feature extraction network based on the domain adaptation module, and quantitatively calculate the differences in the marginal distributions and conditional distributions of the source domain features and target domain features in the deep network layer; A parameter-locked temporal convolutional capsule feature extraction network generation unit, configured to use the output layer result, the prediction loss function, and the discriminative loss function to correct the fitted temporal convolutional capsule feature extraction network, and generate a parameter-locked temporal convolutional capsule feature extraction network.

8. The nuclear power plant migration diagnosis system according to claim 7, wherein, it further comprises: A total loss function determination module, configured to add penalty factors before the output layer result, the prediction loss function, and the discriminative loss function, and determine a total loss function.

9. An electronic device, wherein, it comprises a memory and a processor, the memory is used for storing a computer program, and the processor runs the computer program to enable the electronic device to execute the nuclear power plant migration diagnosis method according to any one of claims 1-5.

10. A computer-readable storage medium, wherein, it stores a computer program, and when the computer program is executed by a processor, it implements the nuclear power plant migration diagnosis method according to any one of claims 1-5.

Images