5g-based nuclear power plant data augmentation method and apparatus

Abstract

The application discloses a 5G-based nuclear power equipment data augmentation method and device. The method comprises the following steps: obtaining a first training sample set; determining a similar equipment of a target nuclear power equipment; obtaining a second training sample set; converting the second training sample set into a third training sample set; the third training sample set comprises a plurality of third training samples; each third training sample comprises a target augmented data with the same characteristic distribution as the first sampling data, which is converted from the second sampling data, and a second health state category label corresponding to the second sampling data; and training a health state evaluation model by using the first training sample set and the third training sample set. The application expands the training data samples of the target nuclear power equipment by means of a small amount of target nuclear power equipment related data and a large amount of complete data of similar equipment, so as to meet the training sample data demand of the health state evaluation model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a nuclear power equipment data augmentation method and device based on 5G. BACKGROUND

[0002] A nuclear power plant is composed of nearly 300 systems, with more than 6000 sets of mechanical equipment. Rotating equipment such as pumps, motors, and fans exists in multiple process systems, is widely distributed, and has a large variety and quantity. At the same time, rotating equipment is the main carrier of energy conversion and transmission, and plays an important role in the operation of the system.

[0003] Rotating equipment is prone to performance degradation under long-term operating conditions. In order to ensure the continuous and reliable operation of the equipment and avoid permanent damage caused by faults, it is necessary to monitor and evaluate the state of the equipment. However, due to the extreme environment of nuclear power scenarios such as radiation, high temperature, and high pressure, it is difficult to install sensors, and there are few detection points. The monitoring data available in the current nuclear power scenario is relatively scarce, and cannot meet the data volume requirements for monitoring and evaluating the state of the equipment. SUMMARY

[0004] The present application aims to provide a nuclear power equipment data augmentation method and device based on 5G, which can expand the training data samples of the target nuclear power equipment with a small amount of target nuclear power equipment related data and a large amount of complete data of similar equipment, and meet the training sample data requirements of the health state evaluation model.

[0005] The nuclear power equipment data augmentation method based on 5G according to the first aspect of the present application comprises:

[0006] Obtain a first training sample set; the first training sample set includes a plurality of first training samples of a target nuclear power equipment to be detected for health state, each first training sample includes a first sampling data of the target nuclear power equipment, and a first health state category label corresponding to the first sampling data;

[0007] Determine a similar equipment of the target nuclear power equipment;

[0008] Obtain a second training sample set; the second training sample set includes a plurality of second training samples of the similar equipment, each second training sample includes a second sampling data of the similar equipment, and a second health state category label corresponding to the second sampling data;

[0009] convert the second training sample set into a third training sample set; the third training sample set includes a plurality of third training samples, each of the third training samples including target augmented data converted from the second sampling data and having the same feature distribution as the first sampling data, and a second health state category label corresponding to the second sampling data;

[0010] train a health state evaluation model by using the first training sample set and the third training sample set; the health state evaluation model is used to evaluate the health state of the target nuclear power equipment.

[0011] According to some embodiments of the present application, before the second training sample set is converted into the third training sample set, the method further includes:

[0012] obtain a fourth training sample set; the fourth training sample set includes a plurality of fourth training samples, each of the fourth training samples including a third sampling data of the similar equipment and a fourth sampling data of the target nuclear power equipment corresponding to the third sampling data;

[0013] train a first neural network by using the fourth training sample set until a training stop condition of the first neural network is met, to obtain the trained first neural network;

[0014] The conversion of the second training sample set into the third training sample set includes:

[0015] input the second training sample into the trained first neural network to obtain a third training sample corresponding to the second training sample.

[0016] According to some embodiments of the present application, the training of the first neural network by using the fourth training sample set until the training stop condition of the first neural network is met to obtain the trained first neural network includes:

[0017] for an i-th fourth training sample in the fourth training sample set, input the i-th fourth training sample into the first neural network to obtain i-th augmented data corresponding to the i-th fourth training sample; wherein i is a positive integer;

[0018] determine a first loss condition corresponding to the i-th fourth training sample according to the fourth sampling data of the i-th fourth training sample and the i-th augmented data;

[0019] In a case where the first loss condition does not satisfy a first training stop condition of the first neural network, parameters of the first neural network are adjusted to obtain an updated first neural network, i is updated to i+1, and the second training sample set is input into the first neural network until the first training stop condition is satisfied.

[0020] According to some embodiments of the present application, the first loss condition corresponding to the i th fourth training sample is determined according to the fourth sampling data of the i th fourth training sample and the i th augmented data, comprising:

[0021] The fourth sampling data of the i th fourth training sample and the i th augmented data are input into a first discriminant neural network to obtain a corresponding first discriminant result.

[0022] According to the first discriminant result, a first discriminant accuracy and a first loss value are obtained, and the first discriminant accuracy represents a feature distribution similarity condition of a plurality of fourth sampling data and a plurality of i th augmented data.

[0023] Wherein, the first loss condition satisfies the first training stop condition, including that the first discriminant accuracy is within a first accuracy range.

[0024] According to some embodiments of the present application, in a case where the target loss value satisfies the first training stop condition of the first neural network, the fourth training sample set is used to train the first neural network until the training stop condition of the first neural network is satisfied to obtain a trained first neural network, further comprising:

[0025] For a j th fourth training sample in the fourth training sample set, the j th fourth training sample is input into the first neural network to be trained to obtain j th augmented data corresponding to the j th fourth training sample; wherein j is a positive integer.

[0026] The j th augmented data is input into a second neural network to obtain j th restored data of the similar device; wherein the network structure of the second neural network is mirror-symmetrical to the network structure of the first neural network.

[0027] According to the third sampling data of the i th fourth training sample and the j th restored data, a second loss condition corresponding to the i th fourth training sample is determined.

[0028] In a case where the second loss condition does not satisfy a second training stop condition of the first neural network, parameters of the first neural network and the second neural network are adjusted to obtain updated first neural network and second neural network, j is updated to j+1, and the jth fourth training sample is input into the first neural network to be trained to obtain the jth augmented data corresponding to the jth fourth training sample until the second training stop condition is satisfied.

[0029] According to some embodiments of the present application, the second loss condition corresponding to the ith fourth training sample is determined according to the third sampling data of the ith fourth training sample and the jth restored data, comprising:

[0030] The third sampling data of the ith fourth training sample and the jth restored data are input into a second discriminant neural network to obtain a corresponding second discriminant result;

[0031] According to the second discriminant result, a second discriminant accuracy and a second loss value are obtained, and the second discriminant accuracy represents the feature distribution similarity of a plurality of third sampling data and a plurality of jth restored data;

[0032] Wherein, the second loss condition satisfies the second training stop condition, including that the first discriminant accuracy and the second discriminant accuracy are within a second accuracy range, and the first loss value is less than a first threshold value, and the second loss value is less than a second threshold value.

[0033] According to some embodiments of the present application, the target nuclear power equipment is a rotating equipment with a bearing, and the similar equipment of the target nuclear power equipment is determined, comprising:

[0034] A type of other rotating equipment containing a bearing is determined as a first similar equipment;

[0035] The sampling data amount of the first similar equipment is obtained;

[0036] In a case where the sampling data amount is greater than or equal to a sample amount threshold value, the first similar equipment is determined as the similar equipment; wherein the sample amount threshold value is obtained based on parameters of the health state evaluation model.

[0037] The 5G-based nuclear power equipment data augmentation device according to the second aspect of the present application comprises:

[0038] The first acquisition module is configured to acquire a first training sample set, wherein the first training sample set comprises a plurality of first training samples of a target nuclear power equipment to be detected for health state, each first training sample comprises a first sampling data of the target nuclear power equipment, and a first health state category label corresponding to the first sampling data;

[0039] determining a similar device of the target nuclear power equipment;

[0040] a second obtaining module, configured to obtain a second training sample set; the second training sample set comprises a plurality of second training samples of the similar device, each of the second training samples comprises a second sampling data of the similar device, and a second health state category label corresponding to the second sampling data;

[0041] a data augmentation module, configured to convert the second training sample set into a third training sample set; the third training sample set comprises a plurality of third training samples, each of the third training samples comprises a target augmented data having the same feature distribution as the first sampling data and obtained by converting the second sampling data, and the second health state category label corresponding to the second sampling data;

[0042] a training module, configured to train a health state evaluation model by using the first training sample set and the third training sample set; the health state evaluation model is used to evaluate the health state of the target nuclear power equipment.

[0043] The electronic device according to the third aspect of the present application comprises a processor and a memory, the memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the 5G-based nuclear power equipment data augmentation method according to any one of the second aspect of the present application.

[0044] The computer readable storage medium according to the fourth aspect of the present application stores computer executable instructions for executing the 5G-based nuclear power equipment data augmentation method according to the first aspect of the present application.

[0045] In the embodiments of the present application, by converting the similar device related data with sufficient data amount, augmented data having the same feature distribution as the target nuclear power equipment related data is obtained, which realizes the expansion of the training data samples of the target nuclear power equipment by means of a small amount of target nuclear power equipment related data and a large amount of complete data of similar devices, meets the training sample data demand of the health state evaluation model, and can meet the data amount requirement while ensuring data accuracy. In addition, the 5G-based nuclear power equipment data augmentation method of the present application trains the health state evaluation model based on the expanded training data samples, performs 5G-based nuclear power equipment data augmentation, and the health state evaluation model is supported by sufficient training data, which can achieve good evaluation effect.

[0046] Other features and advantages of the present application will be described in the subsequent description, and some will become apparent from the description, or will be understood by those skilled in the art. BRIEF DESCRIPTION OF DRAWINGS

[0047] The above and / or additional aspects and advantages of the present application will become apparent and rational understood by those skilled in the art from the following description, taken in conjunction with the accompanying drawings, wherein:

[0048] Figure 1 is a flowchart of an embodiment of the 5G-based nuclear power equipment data augmentation method of the present application;

[0049] Figure 2 is a network structure diagram of a first neural network of an embodiment of the 5G-based nuclear power equipment data augmentation method of the present application;

[0050] Figure 3 is a training flowchart of a first neural network of an embodiment of the 5G-based nuclear power equipment data augmentation method of the present application;

[0051] Figure 4 is a structural diagram of an embodiment of the 5G-based nuclear power equipment data augmentation apparatus of the present application;

[0052] Figure 5 is a hardware structure diagram of an embodiment of the electronic device of the present application. DETAILED DESCRIPTION

[0053] Embodiments of the present application are described in detail below with reference to the accompanying drawings, examples of which are shown in the drawings, wherein the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0054] In the description of the present application, if there is a description to the first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features or the sequence of the indicated technical features.

[0055] In the description of the present application, it is to be understood that the orientation description, such as the orientation or position relationship indicated by up, down, etc., is based on the orientation or position relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the indicated device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application.

[0056] In the description of the present application, it is to be noted that, unless otherwise explicitly limited, the words such as setting, installing, connecting, etc. should be broadly understood, and those skilled in the art can reasonably determine the specific meaning of the above words in the present application in conjunction with the specific content of the technical solution.

[0057] The technical solutions of the present application will be described clearly and completely in combination with the drawings. Obviously, the following described embodiments are part of the embodiments of the present application, not all embodiments.

[0058] Figure 1 The flowchart of the embodiment of the 5G-based nuclear power equipment data augmentation method provided by the present application is shown. The following refers to Figure 1 The embodiments of the present application are further described.

[0059] As Figure 1 shown, the present application provides a 5G-based nuclear power equipment data augmentation method, which comprises:

[0060] Step 101, obtaining a first training sample set; the first training sample set comprises a plurality of first training samples of a target nuclear power equipment to be detected for health status, each first training sample comprising a first sampling data of the target nuclear power equipment, and a first health status category label corresponding to the first sampling data;

[0061] Step 102, determining a similar equipment of the target nuclear power equipment;

[0062] Step 103, obtaining a second training sample set; the second training sample set comprises a plurality of second training samples of the similar equipment, each second training sample comprising a second sampling data of the similar equipment, and a second health status category label corresponding to the second sampling data;

[0063] Step 104, converting the second training sample set into a third training sample set; the third training sample set comprises a plurality of third training samples, each third training sample comprising a target augmented data having the same feature distribution as the first sampling data converted from the second sampling data, and the second health status category label corresponding to the second sampling data;

[0064] Step 105, training a health status evaluation model using the first training sample set and the third training sample set; the health status evaluation model is used to evaluate the health status of the target nuclear power equipment.

[0065] In the embodiments of the present application, the similar device related data with sufficient data volume is converted to obtain augmented data with the same characteristic distribution as the target nuclear power equipment related data, thereby realizing the expansion of the training data samples of the target nuclear power equipment with the aid of a small amount of target nuclear power equipment related data and a large amount of complete data of similar equipment, meeting the training sample data demand of the health state evaluation model, and achieving the data volume requirement while ensuring data accuracy. In addition, the 5G-based nuclear power equipment data augmentation method of the present application trains a health state evaluation model based on the expanded training data samples, performs 5G-based nuclear power equipment data augmentation, and the health state evaluation model is supported by sufficient training data, which can achieve good evaluation effect. In addition, it is not necessary to carry out data collection process by building or simulating actual scene, which greatly reduces the difficulty of data acquisition, reduces the application cost of data-driven model, and improves the utilization rate of data.

[0066] The target nuclear power equipment can be any nuclear power equipment that needs to be evaluated for health state, and the corresponding health state evaluation model is used to evaluate the health state of the nuclear power equipment.

[0067] The target nuclear power equipment can be a rotating equipment in a nuclear power plant, which has a bearing structure, and the bearing of the target nuclear power equipment can be a rolling bearing. Specifically, the target nuclear power equipment can be a motor equipment, a fan equipment or a pump equipment which is currently widely used in nuclear power plants.

[0068] For example, the target nuclear power equipment is a motor equipment, and the health state evaluation model is used to evaluate the health state of the motor equipment.

[0069] The similar equipment only refers to a type of rotating equipment with sufficient data volume. It can be a rotating equipment with a bearing, specifically a rotating equipment with a rolling bearing, or a rolling bearing.

[0070] The first sampling data and the second sampling data can refer to vibration data during constant power or constant speed operation of the equipment, and the vibration data is collected by a vibration sensor taking the bearing of the equipment as the measurement object. Specifically, the first sampling data can be collected by a vibration sensor from a nuclear rotating equipment, and the second sampling data can be collected by a vibration sensor from a similar equipment.

[0071] One first sampling data or second sampling data can be a group of vibration data collected by a vibration sensor within a collection duration; and the multiple collection data included in one training sample can be multiple groups of vibration data collected multiple times. The related parameters of the collection process can be set according to the equipment condition to make the collected data more accurate, and the related parameters can include collection duration, collection frequency, etc.; and the vibration data can be amplitude data.

[0072] For example, the data acquisition parameters of the target nuclear power equipment may include: acquisition frequency f tar Duration t of a single data collection tar,sample Sampling interval t tar,break The rotational speed of the target rotating device is w tar Each first training sample set contains the number of vibrations.

[0073] Data acquisition parameters for similar devices may include: acquisition frequency f sim Duration t of a single data collection sim,sample Sampling interval t sim,break The rotational speed of the target rotating device is w sim Each second training sample set contains the number of vibrations.

[0074] The aforementioned health status category labels are used to determine the health status of the corresponding equipment based on vibration data. Specifically, the first health status category label corresponds to the first sampled data, and the second health status category label corresponds to the second sampled data.

[0075] The first training sample set may include multiple first training samples, and the second training sample set may include multiple second training samples. The first and second training samples may consist of vibration data and corresponding labels.

[0076] For example, the first training sample and the second training sample are labeled data samples, both of which can be represented as (X, Y), where X is the vibration data and Y is the health status category label corresponding to the vibration data, which should be divided into at least three categories: "normal", "decline" and "fault".

[0077] For example, the data format of the first training sample of the target nuclear power equipment is: D tar =(X tar Y tar );

[0078] in,

[0079]

[0080] Where, x 1i The first sampling data for the target nuclear power equipment includes a set of amplitude data collected by the target nuclear power equipment within a single sampling duration, N. tar This refers to the sample size corresponding to the target nuclear power equipment, which can be determined based on the number of target nuclear power equipment and the data collection situation. A 1i For the corresponding amplitude data obtained from the collection, N x,tar The number of amplitude data; y 1in is the number of label categories, which can be set to 3, and C1, C2 and C3 are normal, degradation and failure respectively, that is, y 1i ∈(normal, degradation, failure).

[0081] The data form of the second training sample of the similar device is D sim = (X sim , Y sim ).

[0082] wherein, y i ∈(C1,C2,……,C n ), n≥3.

[0083] wherein, x 2i is the second sampling data of the similar device, containing a group of amplitude data collected by the similar device in one collection duration, N sim is the sample quantity corresponding to the similar device, which can be determined according to the number of similar devices and the data collection condition, A 2i is the corresponding amplitude data collected, N x,sim is the number of amplitude data; y 2i is the corresponding second health status category label, n is the number of label categories, which can be set to 3, and C1, C2 and C3 are normal, degradation and failure respectively, that is, y 2i ∈(normal, degradation, failure).

[0084] The above label can be the health status category determined according to the vibration data.

[0085] For example, the maximum amplitude of the “normal” label sample is A norm , the amplitude A d of the “degradation” label is in the range of [A norm , 5A norm ], and the amplitude A f of the “failure” label is greater than 5A norm . The amplitude A d of the “degradation” label and the amplitude A f of the “failure” label can refer to all values of a group of amplitude data contained in one sampling data, or can refer to the average value or the median.

[0086] The above amplitude-related data can be absolute values.

[0087] The above first training sample set can be a vibration data set with labels related to the target nuclear power equipment, and the above second training sample set can be a vibration data set with labels related to the similar device.

[0088] The third training sample set can be an augmented sample set converted from the similar device related data set and having the same feature distribution as the target nuclear power plant related data set, that is, the similarity with the target nuclear power plant related data set is ensured, and the data amount order of magnitude of the similar device related data set is also ensured, so as to meet the requirement of the training amount of the nuclear power rotating equipment health state evaluation model.

[0089] The health state evaluation model can be a deep learning model that can be used to perform a multi-classification task. Specifically, the network layer of the model can be a fully connected network, a convolutional network, a recurrent neural network, etc. The model can be trained in a supervised learning manner, and the training process can be performed by inputting the first training sample set and the third training sample set.

[0090] The health state evaluation model is used to evaluate the health state of the target nuclear power plant. The trained health state evaluation model can output the evaluated health state category of the device according to the input vibration data.

[0091] In some embodiments, before converting the second training sample set into the third training sample set, the method further comprises:

[0092] obtaining a fourth training sample set; the fourth training sample set includes a plurality of fourth training samples, and each fourth training sample includes a third sampling data of a similar device and a fourth sampling data of the target nuclear power plant corresponding to the third sampling data;

[0093] training the first neural network using the fourth training sample set until the training stop condition of the first neural network is met, to obtain a trained first neural network;

[0094] converting the second training sample set into the third training sample set includes:

[0095] inputting the second training sample into the trained first neural network to obtain a third training sample corresponding to the second training sample.

[0096] In this embodiment, the first neural network can be used to convert the second training sample set into the third training sample set, to realize the conversion of the similar device related data set into the augmented data set having the same feature distribution as the target nuclear power plant related data set, and to better realize data augmentation through neural network technology.

[0097] The third sampling data can be a part of the plurality of second sampling data in the second training sample set, and the fourth sampling data can be a part of the plurality of first sampling data in the first training sample set. The fourth sampling data corresponding to the third sampling data can be understood as that the corresponding health state category labels of the third sampling data and the fourth sampling data are the same.

[0098] The aforementioned fourth training sample set can be understood as including relevant vibration data of similar equipment and relevant vibration data of the target nuclear power equipment. Multiple fourth training samples are formed according to the label status, which can be used to train the first neural network.

[0099] In some implementations, the first neural network is trained using a fourth training sample set until the training stopping condition of the first neural network is met, resulting in a trained first neural network, including:

[0100] For the i-th fourth training sample in the fourth training sample set, input the i-th fourth training sample into the first neural network to obtain the i-th augmented data corresponding to the i-th fourth training sample; where i is a positive integer;

[0101] Based on the fourth sampling data and the i-th augmented data of the i-th fourth training sample, determine the first loss situation corresponding to the i-th fourth training sample;

[0102] If the first training stopping condition of the first neural network is not met in the first loss case, the parameters of the first neural network are adjusted to obtain the updated first neural network, and i is updated to i+1. Then, the second training sample set is input into the first neural network until the first training stopping condition is met.

[0103] In this embodiment, the first neural network is trained by inputting the fourth training sample set, and the parameters are continuously adjusted and updated according to the loss situation until the training stopping condition is met, thus obtaining the trained first neural network. Updating the neural network using the loss situation can improve the accuracy and robustness of the model to a certain extent, and can also adapt to different training environments and data types, with good generalization ability and practicality.

[0104] like Figure 2 As shown, in some implementations, the first neural network G1(θ) g1 The network structure can be as follows: it includes three main network structures: an input layer, a hidden layer, and an output layer. The input layer has one layer, and the number of neurons is equal to the number of vibrations N of similar device data samples. x,sim Same, number of hidden layers L h ∈[4,8],L h ∈N + The number of hidden layer neurons is unlimited, the number of output layers is 1, and the number of neurons is proportional to the number of vibrations N in the generated augmented data samples. x,tar same.

[0105] In some implementations, the first neural network G1(θ) g1 This can be a fully connected neural network, with the activation function being act. g1= max(0.01x, x), wherein x is a linear operation output result of the neuron.

[0106] In some embodiments, the first loss condition corresponding to the i-th fourth training sample is determined according to the fourth sampling data of the i-th fourth training sample and the i-th augmented data, comprising:

[0107] The fourth sampling data of the i-th fourth training sample and the i-th augmented data are input into the first discriminant neural network to obtain the corresponding first discriminant result.

[0108] According to the first discriminant result, the first discriminant accuracy and the first loss value are obtained, and the first discriminant accuracy represents the feature distribution similarity of the plurality of fourth sampling data and the plurality of i-th augmented data.

[0109] The first loss condition satisfies the first training stop condition, including that the first discriminant accuracy is within the first accuracy range.

[0110] In this embodiment, the loss condition is obtained by the discriminant neural network. The discriminant neural network not only provides a clear optimization target and an efficient optimization means, but also has good adaptability and flexibility, which helps to improve the performance and generalization ability of the model.

[0111] The above fourth sampling data is the related vibration data of the target nuclear power equipment, and the i-th augmented data is the augmented data output by the first neural network after converting the related vibration data of the similar equipment.

[0112] The above first discriminant neural network can judge the feature distribution similarity between the related vibration data of the target nuclear power equipment and the generated augmented data, and further can judge the loss condition of the first neural network.

[0113] The above first discriminant neural network D1(θ d1 ) can discriminate whether the generated augmented data D tar,gen and the actual target nuclear power equipment data D tar have the same feature distribution; the output result of the network, i.e. the first discriminant result, can be a binary classification value (0, 1), if the distribution is similar, output 0, and if the distribution is different, output 1, the discriminant result C D1 = D1(D tar,gen , θ d1 ) e {0, 1}; the loss function loss(d1) of the first discriminant neural network D1(θ d1 ) can be calculated through the discriminant result, specifically, the loss function is a cross-entropy loss function, further, the loss function is a binary cross-entropy loss function, and then the first discriminant accuracy and the first loss value can be obtained, wherein the first discriminant accuracy is a binary classification discriminant accuracy.

[0114] The first training stop condition can be that the first discrimination accuracy reaches a certain accuracy range, and the training is stopped. Specifically, the first accuracy range can be an interval greater than 90%.

[0115] In some embodiments, in the case that the first loss condition does not satisfy the first training stop condition of the first neural network, the parameters of the first neural network are adjusted to obtain an updated first neural network, which can be updating the parameters of the first neural network by the Adam gradient descent algorithm.

[0116] As shown in FIG. 1, in some embodiments, in the case that the target loss value satisfies the first training stop condition of the first neural network, the first neural network is trained using the fourth training sample set until the training stop condition of the first neural network is satisfied to obtain a trained first neural network, further comprising: Figure 3

[0117] For the jth fourth training sample in the fourth training sample set, the jth fourth training sample is input into the first neural network to be trained to obtain jth augmented data corresponding to the jth fourth training sample; wherein j is a positive integer;

[0118] The jth augmented data is input into the second neural network to obtain jth restored data of the similar device; wherein the network structure of the second neural network is mirror-symmetric to the network structure of the first neural network;

[0119] According to the third sampling data of the ith fourth training sample and the jth restored data, a second loss condition corresponding to the ith fourth training sample is determined;

[0120] In the case that the second loss condition does not satisfy the second training stop condition of the first neural network, the parameters of the first neural network and the second neural network are adjusted to obtain an updated first neural network and second neural network, and j is updated to j+1, and the jth fourth training sample is input into the first neural network to be trained to obtain the jth augmented data corresponding to the jth fourth training sample until the second training stop condition is satisfied.

[0121] In this embodiment, by building the second neural network D2(θ d2 ), the generated augmented data D sim,gen is restored to the restored data similar to the similar device data, i.e. D sim,gen =G2(D tar,gen ,θ g2 ), wherein the network structure of the second neural network G2(θ g2 ) is mirror-symmetric to the first neural network G1(θ g1 ); then by building the second discrimination neural network D2(θ d2 ​, discriminates the reduction data D sim,gen , and the actual similar device data D sim whether they have the same feature distribution. Further, the explicit optimization target and efficient optimization means for the first neural network are provided, which has good adaptability and flexibility, and helps to improve the performance and generalization ability of the model.

[0122] In some embodiments, according to the third sampling data of the i-th fourth training sample and the j-th reduction data, the second loss condition corresponding to the i-th fourth training sample is determined, including:

[0123] The third sampling data of the i-th fourth training sample and the j-th reduction data are input into the second discriminant neural network to obtain the corresponding second discriminant result;

[0124] According to the second discriminant result, the second discriminant accuracy and the second loss value are obtained, and the second discriminant accuracy represents the feature distribution similarity of the plurality of third sampling data and the plurality of j-th reduction data;

[0125] The second loss condition satisfies the second training stop condition, including that the first discriminant accuracy and the second discriminant accuracy are within the second accuracy range, and the first loss value is less than the first threshold value, and the second loss value is less than the second threshold value.

[0126] In the embodiment, the second loss condition satisfies the second training stop condition, which requires that the discriminant accuracy requirement and the loss value requirement are met at the same time, and further optimizes the training of the model.

[0127] The output result of the above-mentioned second discriminant neural network can be a binary classification value (0, 1), if the distribution is similar, output 0, and the distribution difference is large, output 1, the discriminant result C D2 = D2(D sim,gen , θ d2 ) ∈ {0, 1}, the loss function loss(d2) of the neural network D2(θ d2 ) is calculated, specifically, the loss function is the cross-entropy loss function, further, the loss function is the binary cross-entropy loss function, and then the second discriminant accuracy and the second loss value can be obtained, wherein the second discriminant accuracy is the binary discriminant accuracy.

[0128] The above-mentioned second training stop condition includes that the first discriminant accuracy and the second discriminant accuracy are within the second accuracy range, and the first loss value is less than the first threshold value, and the second loss value is less than the second threshold value, which can be that the first discriminant accuracy of the first discriminant neural network D1(θ d1 ) and the second discriminant accuracy of the second discriminant neural network D2(θ d2 ) are both greater than or equal to 95%, and satisfy the following formula:

[0129]

[0130] wherein A tar,min is the minimum vibration amplitude of the target nuclear power equipment data used for training, A sim,min is the minimum vibration amplitude of the similar equipment data used for training.

[0131] In some embodiments, in the case that the second loss condition does not satisfy the second training stop condition of the first neural network, the parameters of the first neural network and the second neural network are adjusted to obtain updated first neural network and second neural network, which can be updating the parameters of the first neural network G1(θ g1 ) and the second neural network G2(θ g2 ) by Adam gradient descent algorithm respectively.

[0132] In some embodiments, the first discriminant neural network D1(θ d1 ) and the second discriminant neural network D2(θ d2 ) are nonlinear classifier networks, and the basic structure of the network can be a fully connected network or a convolutional network. Both training processes are supervised training. The initial training termination condition is that the binary classification discrimination accuracy obtained by the neural network through the test set for 10 consecutive times is in the interval of 75%-85%, wherein the test set is divided from the data sample set used for training.

[0133] In some embodiments, the target nuclear power equipment is a rotating equipment with a bearing, and the similar equipment of the target nuclear power equipment is determined, including:

[0134] determining a type of other rotating equipment containing a bearing as a first similar equipment;

[0135] obtaining the sampling data amount of the first similar equipment;

[0136] In the case that the sampling data amount is greater than or equal to the sample amount threshold, the first similar equipment is determined as a similar equipment; wherein the sample amount threshold is obtained based on the parameters of the health state evaluation model.

[0137] In the embodiment, the other rotating equipment with sufficient sampling data amount is selected to be determined as a similar equipment, which can provide sufficient sample amount basis for the data augmentation process, and by selecting a type of other rotating equipment which can provide sufficient sampling data amount as a similar equipment and using the data of this type of equipment, rather than using the data of multiple types of other rotating equipment to reach the sampling data amount requirement, the data of the similar equipment provided can be of higher quality, and thus the data augmentation can achieve better results.

[0138] The above bearing can be a rolling bearing.

[0139] The sampling data is greater than or equal to a sample quantity threshold, and the sample quantity threshold is shown in the following formula:

[0140]

[0141] wherein n is a category number of the evaluation task, that is, a health state category number, e is a minimum reachable error of the health state evaluation model, e [0, 0.1], is a discrimination accuracy of the estimated health state evaluation model in the training process, and a is a learning rate of the iterative training process of the health state evaluation model, a [0.01, 1], d is a learning decay rate, d [-1, 0], and the related parameters of the health state evaluation model can be obtained in advance.

[0142] The execution subject of the 5G-based nuclear power equipment data augmentation method provided in the embodiments of the present application can be a 5G-based nuclear power equipment data augmentation device 200. In the embodiments of the present application, the 5G-based nuclear power equipment data augmentation method executed by the 5G-based nuclear power equipment data augmentation device 200 is taken as an example to illustrate the 5G-based nuclear power equipment data augmentation device 200 provided in the embodiments of the present application.

[0143] Please refer to Figure 4 , which is a structural schematic diagram of a 5G-based nuclear power equipment data augmentation device 200 provided in the embodiments of the present application. As Figure 4 indicated, the 5G-based nuclear power equipment data augmentation device 200 includes:

[0144] The first acquisition module 201 is configured to acquire a first training sample set. The first training sample set includes a plurality of first training samples of a target nuclear power equipment to be detected in a health state, each first training sample includes a first sampling data of the target nuclear power equipment, and a first health state category label corresponding to the first sampling data;

[0145] The determination module 202 is configured to determine a similar equipment of the target nuclear power equipment.

[0146] The second acquisition module 203 is configured to acquire a second training sample set. The second training sample set includes a plurality of second training samples of the similar equipment, each second training sample includes a second sampling data of the similar equipment, and a second health state category label corresponding to the second sampling data.

[0147] The data augmentation module 204 is configured to convert the second training sample set into a third training sample set. The third training sample set includes a plurality of third training samples, each third training sample includes a target augmented data having the same feature distribution as the first sampling data and obtained by converting the second sampling data, and the second health state category label corresponding to the second sampling data.

[0148] The training module 205 is configured to train a health state evaluation model by using the first training sample set and the third training sample set, and the health state evaluation model is configured to evaluate the health state of the target nuclear power equipment.

[0149] In some embodiments, the data augmentation module 204 can be configured to:

[0150] obtain a fourth training sample set, the fourth training sample set comprising a plurality of fourth training samples, each fourth training sample comprising a third sampling data of a similar equipment and a fourth sampling data of the target nuclear power equipment corresponding to the third sampling data;

[0151] train the first neural network by using the fourth training sample set until a training stop condition of the first neural network is met, to obtain a trained first neural network;

[0152] convert the second training sample set into the third training sample set, comprising:

[0153] input the second training sample into the trained first neural network to obtain a third training sample corresponding to the second training sample.

[0154] In some embodiments, the data augmentation module 204 can be configured to:

[0155] for an i-th fourth training sample in the fourth training sample set, input the i-th fourth training sample into the first neural network to obtain i-augmented data corresponding to the i-th fourth training sample, wherein i is a positive integer;

[0156] determine a first loss condition corresponding to the i-th fourth training sample according to the fourth sampling data of the i-th fourth training sample and the i-augmented data;

[0157] in a case where the first loss condition does not meet a first training stop condition of the first neural network, adjust parameters of the first neural network to obtain an updated first neural network, and update i to i+1, and return to input the second training sample set into the first neural network until the first training stop condition is met.

[0158] In some embodiments, the data augmentation module 204 can be configured to:

[0159] input the fourth sampling data of the i-th fourth training sample and the i-augmented data into the first discriminant neural network to obtain a corresponding first discriminant result;

[0160] obtain a first discriminant accuracy and a first loss value according to the first discriminant result, the first discriminant accuracy representing a feature distribution similarity between the plurality of fourth sampling data and the plurality of i-augmented data;

[0161] The first loss condition satisfies a first training stop condition, including that the first discrimination accuracy is within a first accuracy range.

[0162] In some embodiments, the data augmentation module 204 can also be configured to:

[0163] For the jth fourth training sample in the fourth training sample set, input the jth fourth training sample into the first neural network to be trained to obtain jth augmented data corresponding to the jth fourth training sample; where j is a positive integer;

[0164] Input the jth augmented data into the second neural network to obtain jth restored data of the similar device; where the network structure of the second neural network is mirror-symmetrical to the network structure of the first neural network;

[0165] According to the third sampling data of the ith fourth training sample and the jth restored data, determine a second loss condition corresponding to the ith fourth training sample;

[0166] In the case where the second loss condition does not satisfy a second training stop condition of the first neural network, adjust the parameters of the first neural network and the second neural network to obtain updated first neural network and second neural network, and update j to j+1, return to input the jth fourth training sample into the first neural network to be trained to obtain the jth augmented data corresponding to the jth fourth training sample, until the second training stop condition is satisfied.

[0167] In some embodiments, the data augmentation module 204 can be configured to:

[0168] Input the third sampling data of the ith fourth training sample and the jth restored data into the second discrimination neural network to obtain a corresponding second discrimination result;

[0169] According to the second discrimination result, obtain a second discrimination accuracy and a second loss value, the second discrimination accuracy representing the feature distribution similarity of the plurality of third sampling data and the plurality of jth restored data;

[0170] The second loss condition satisfies a second training stop condition, including that the first discrimination accuracy and the second discrimination accuracy are within a second accuracy range, and the first loss value is less than a first threshold value, and the second loss value is less than a second threshold value.

[0171] In some embodiments, the determination module 202 can be configured to:

[0172] Determine a type of other rotating equipment containing bearings as the first similar device;

[0173] Obtain the sampling data amount of the first similar device;

[0174] In a case that the sampling data amount is greater than or equal to the sample amount threshold, the first similar device is determined as the similar device; wherein the sample amount threshold is obtained based on parameters of the health state evaluation model.

[0175] Since the 5G-based nuclear power equipment data augmentation apparatus 200 adopts all the technical solutions of the 5G-based nuclear power equipment data augmentation method of the above-mentioned embodiments, it at least has all the beneficial effects brought by the technical solutions of the above-mentioned embodiments, and here it will not be repeated.

[0176] Figure 5 The hardware structure schematic diagram of the electronic device provided by the embodiments of the present application is shown in the figure.

[0177] The electronic device can include a processor 301 and a memory 302 storing computer program instructions.

[0178] Specifically, the processor 301 can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or can be configured to implement one or more integrated circuits of the embodiments of the present application.

[0179] The memory 302 can include a mass storage for data or instructions. By way of example and not limitation, the memory 302 can include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 302 can include removable or non-removable (or fixed) media. Where appropriate, the memory 302 can be internal or external to the integrated gateway disaster recovery device. In some embodiments, the memory 302 is a non-volatile solid-state memory.

[0180] In some embodiments, the memory 302 can include read-only memory (ROM), random access memory (RAM), a disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Thus, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software that, when executed (e.g., by one or more processors), is operable to perform operations described with reference to the methods according to an aspect of the present disclosure.

[0181] The processor 301 realizes any one of the above-mentioned embodiments of the 5G-based nuclear power equipment data augmentation method by reading and executing the computer program instructions stored in the memory 302.

[0182] In one example, the electronic device can further include a communication interface 303 and a bus 310. As shown, the processor 301, the memory 302, and the communication interface 303 are connected through the bus 310 and complete communication with each other. Figure 5

[0183] The communication interface 303 is mainly used to realize the communication between the modules, devices, units and / or equipment in the embodiments of the present application.

[0184] The bus 310 includes hardware, software or both to couple components of the online data traffic billing device to each other. By way of example, and not limitation, the bus can include an accelerated graphics port (AGP) or other graphics bus, an enhanced industry standard architecture (EISA) bus, a front-side bus (FSB), a HyperTransport (HT) interconnect, an industry standard architecture (ISA) bus, an InfiniBand (IB) interconnect, a low pin count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standards association local (VLB) bus, or another suitable bus or a combination of two or more of these. Where suitable, the bus 310 can include one or more buses. Although specific buses are described and shown in the embodiments of the present application, the present application contemplates any suitable bus or interconnect.

[0185] The electronic device can execute the 5G-based nuclear power plant data augmentation method in the embodiments of the present application, thereby realizing the 5G-based nuclear power plant data augmentation method and device described in combination Figure 1 and Figure 4 described in the embodiments of the present application.

[0186] In addition, in combination with the 5G-based nuclear power plant data augmentation method in the above embodiments, the embodiments of the present application can provide a computer storage medium to realize. The computer storage medium has computer program instructions stored thereon; the computer program instructions are executed by the processor to realize any one of the 5G-based nuclear power plant data augmentation methods in the above embodiments.

[0187] It needs to be clear that the present application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0188] ​The functions noted in the description of the structural block diagrams above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are program or code segments that are used to perform the required tasks. The program or code segments can be stored in a machine-readable medium, or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. A "machine-readable medium" includes any medium that can store or transport information. Examples of machine-readable media include electronic circuitry, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. The code segments can be downloaded via computer networks such as the Internet, Intranet, etc.

[0189] It is also important to note that the examples mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the steps mentioned in the examples, that is, the steps can be performed in the order mentioned in the examples, or in a different order from the examples, or several steps can be performed simultaneously.

[0190] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks. The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks.

[0191] The above merely describes a specific implementation of the present application. Those skilled in the art can clearly understand the specific working processes of the system, modules and units described above for the convenience and brevity of description, and can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A 5G-based nuclear power plant data augmentation method, characterized in that, The method comprises: obtaining a first training sample set; the first training sample set comprises a plurality of first training samples of a target nuclear power equipment to be detected in a health state, each first training sample comprises a first sampling data of the target nuclear power equipment, and a first health state category label corresponding to the first sampling data; determining a similar device of the target nuclear power equipment; obtaining a second training sample set; the second training sample set comprises a plurality of second training samples of the similar device, each second training sample comprises a second sampling data of the similar device, and a second health state category label corresponding to the second sampling data; convert the second training sample set into a third training sample set; the third training sample set comprises a plurality of third training samples, each third training sample comprises a target augmented data with the same feature distribution as the first sampling data converted from the second sampling data, and a second health state category label corresponding to the second sampling data; training a health state evaluation model using the first training sample set and the third training sample set; the health state evaluation model is used to evaluate the health state of the target nuclear power equipment.

2. The method of claim 1, wherein, Before converting the second training sample set into a third training sample set, the method further comprises: obtaining a fourth training sample set; the fourth training sample set comprises a plurality of fourth training samples, each fourth training sample comprises a third sampling data of the similar device, and a fourth sampling data of the target nuclear power equipment corresponding to the third sampling data; training a first neural network using the fourth training sample set until a training stop condition of the first neural network is met, to obtain a trained first neural network; the step of converting the second training sample set into a third training sample set comprises: inputting the second training sample into the trained first neural network to obtain a third training sample corresponding to the second training sample.

3. The method of claim 2, wherein, The step of training a first neural network using the fourth training sample set until a training stop condition of the first neural network is met, to obtain a trained first neural network, comprises: for the i-th fourth training sample in the fourth training sample set, inputting the i-th fourth training sample into the first neural network to obtain i-th augmented data corresponding to the i-th fourth training sample; wherein i is a positive integer; determining a first loss condition corresponding to the i-th fourth training sample according to the fourth sampling data of the i-th fourth training sample and the i-th augmented data; if the first loss condition does not meet the first training stop condition of the first neural network, adjusting the parameters of the first neural network to obtain an updated first neural network, updating i to i+1, and returning to input the second training sample set into the first neural network until the first training stop condition is met.

4. The method of claim 3, wherein, The step of determining a first loss condition corresponding to the i-th fourth training sample according to the fourth sampling data of the i-th fourth training sample and the i-th augmented data comprises: inputting the fourth sampling data of the i-th fourth training sample and the i-augmented data into a first discriminant neural network to obtain a corresponding first discriminant result; obtaining a first discriminant accuracy and a first loss value according to the first discriminant result, the first discriminant accuracy representing a feature distribution similarity between the plurality of fourth sampling data and the plurality of i-augmented data; wherein the first loss condition satisfies the first training stop condition, including that the first discriminant accuracy is within a first accuracy range.

5. The method of claim 4, wherein, In the case that the target loss value satisfies the first training stop condition of the first neural network, the training of the first neural network using the fourth training sample set until the training stop condition of the first neural network is satisfied to obtain the trained first neural network, further comprises: for a j-th fourth training sample in the fourth training sample set, inputting the j-th fourth training sample into the first neural network to be trained to obtain j-augmented data corresponding to the j-th fourth training sample; wherein j is a positive integer; inputting the j-augmented data into a second neural network to obtain j-restored data of the similar device; wherein the network structure of the second neural network is mirror-symmetrical to the network structure of the first neural network; determining a second loss condition corresponding to the i-th fourth training sample according to the third sampling data of the i-th fourth training sample and the j-restored data; in the case that the second loss condition does not satisfy the second training stop condition of the first neural network, adjusting the parameters of the first neural network and the second neural network to obtain updated first neural network and second neural network, and updating j to j+1, returning to input the j-th fourth training sample into the first neural network to be trained to obtain j-augmented data corresponding to the j-th fourth training sample, until the second training stop condition is satisfied.

6. The method of claim 5, wherein, The determining a second loss condition corresponding to the i-th fourth training sample according to the third sampling data of the i-th fourth training sample and the j-restored data, comprises: inputting the third sampling data of the i-th fourth training sample and the j-restored data into a second discriminant neural network to obtain a corresponding second discriminant result; obtaining a second discriminant accuracy and a second loss value according to the second discriminant result, the second discriminant accuracy representing a feature distribution similarity between the plurality of third sampling data and the plurality of j-restored data; wherein the second loss condition satisfies the second training stop condition, including that the first discriminant accuracy and the second discriminant accuracy are within a second accuracy range, and the first loss value is less than a first threshold value, and the second loss value is less than a second threshold value.

7. The method of claim 1, wherein, The target nuclear power equipment is a rotating equipment with a bearing, and the determining a similar equipment of the target nuclear power equipment comprises: determining a type of other rotating equipment containing a bearing as a first similar equipment; obtaining a sampling data amount of the first similar equipment; In a case where the sampling data amount is greater than or equal to a sample amount threshold, the first similar device is determined as the similar device; wherein the sample amount threshold is obtained based on a parameter of the health status evaluation model. 8.A 5G-based nuclear power plant data augmentation apparatus, characterized by, Comprise: A first obtaining module is configured to obtain a first training sample set; The first training sample set comprises a plurality of first training samples of a target nuclear power plant in a to-be-detected health state, each first training sample comprising a first sampling data of the target nuclear power plant and a first health status category label corresponding to the first sampling data; A determining module is configured to determine a similar device of the target nuclear power plant; A second obtaining module is configured to obtain a second training sample set; the second training sample set comprises a plurality of second training samples of the similar device, each second training sample comprising a second sampling data of the similar device and a second health status category label corresponding to the second sampling data; A data augmentation module is configured to convert the second training sample set into a third training sample set; the third training sample set comprises a plurality of third training samples, each third training sample comprising a target augmented data having the same feature distribution as the first sampling data and converted from the second sampling data, and the second health status category label corresponding to the second sampling data; A training module is configured to train a health status evaluation model by using the first training sample set and the third training sample set; the health status evaluation model is used to evaluate the health status of the target nuclear power plant.

Images