Method, system and equipment for repairing missing data and constructing missing data model

Abstract

The invention provides a missing data restoration method, system and equipment and a missing data restoration model construction method, system and equipment, a missing data restoration model comprises a generator and a discriminator, and the construction method comprises the steps of performing preprocessing operation on a sample missing data set to obtain a sample to-be-restored data set, and performing preprocessing operation on an initial time sequence data set to obtain a sample real data set; inputting the sample to-be-repaired data set into a generator to output a sample repair data set corresponding to the sample to-be-repaired data set, inputting the sample to-be-repaired data set and the sample real data set into a discriminator to output a discrimination result, feeding back the discrimination result to the generator, and performing mutual adversarial training on the generator and the discriminator to obtain a sample repair data set corresponding to the sample to-be-repaired data set. According to the method, the missing data restoration model is constructed, the input actual missing data set is restored through the missing data restoration model, the actual restoration data set corresponding to the actual missing data set is output, and the missing time series data is comprehensively, accurately and effectively restored.

Description

Technical Field

[0001] The present disclosure relates to the field of machine learning technology, and in particular to a method, system, and device for repairing missing data and constructing a model thereof. Background Art

[0002] Most existing thermal power plants, hydropower plants, nuclear power plants and other factories rely on sensors to collect time series data. By monitoring and analyzing the time series data, they can monitor the factory's operating status and analyze operating accidents.

[0003] Taking nuclear power plants as an example, emergency preparedness and response for nuclear power plant accidents is the last barrier of in-depth defense of nuclear power safety. In order to ensure the scientific nature of nuclear emergency decision-making and minimize the harm of nuclear accidents to the public and the environment, it is necessary to scientifically diagnose the cause of the accident, predict, analyze and evaluate the development process of the accident and the impact of mitigation measures, and then formulate feasible accident intervention plans. At present, most of my country's nuclear power plants still use traditional threshold monitoring methods, which not only lack an effective information classification mechanism, but also have difficulty in providing the root cause of the accident.

[0004] In addition, due to the harsh operating environment of nuclear power plant sensors, they need to work under high temperature and high radiation conditions for a long time. In the event of a loss of coolant accident or a main steam pipeline rupture accident, the signals transmitted by the sensors may be distorted or missing. Therefore, it is necessary to develop a data repair method for the sensor's time series data, which can correct and supplement the missing monitoring signals of the sensors based on historical data, so as to more comprehensively and objectively reflect the operating status of the nuclear power plant. In extreme cases such as nuclear accidents, the accident status can also be portrayed.

[0005] The current data repair methods mainly include mean interpolation, median interpolation, linear time series interpolation, matrix decomposition, machine learning-based KNN (K-Nearest Neighbor, K nearest neighbor), EM (Expectation-Maximum), RNN (Recurrent Neural Network) and other repair methods; however, nuclear power plant sensor data has the characteristics of path dependence, irreversibility and uncertainty, and the characteristics of time series data vary greatly. General interpolation methods are basically ineffective for data completion; existing machine learning methods can only repair small-scale time series data missing, and cannot further comprehensively, accurately and effectively repair time series data missing. Summary of the invention

[0006] The technical problem to be solved by the present disclosure is to overcome the defect in the prior art that time series data cannot be repaired comprehensively, accurately and effectively, and to provide a method, system and device for repairing missing data and building a model thereof.

[0007] The present invention solves the above technical problems through the following technical solutions:

[0008] In a first aspect, a method for constructing a missing data repair model is provided, wherein the missing data repair model includes a generator and a discriminator; the construction method includes:

[0009] Obtaining an original time series data set within a preset time period as an initial time series data set;

[0010] Performing a data deletion operation on the initial time series data set to obtain a sample missing data set;

[0011] Performing a preprocessing operation on the sample missing data set to obtain a sample data set to be repaired, and performing the preprocessing operation on the initial time series data set to obtain a sample real data set;

[0012] Inputting the sample data set to be repaired into the generator to output a sample repair data set corresponding to the sample data set to be repaired;

[0013] Inputting the sample data set to be repaired and the sample real data set into the discriminator to output a discrimination result;

[0014] Feeding the discrimination result back to the generator, and performing adversarial training on the generator and the discriminator to construct the missing data repair model;

[0015] The discrimination result is used to characterize the closeness between the sample repair data set and the sample real data set, and the missing data repair model is used to repair the actual repair data set corresponding to the input actual missing data set.

[0016] Preferably, the preprocessing operation includes a grayscale transformation algorithm operation;

[0017] The sample real data set and the sample missing data set are both image data sets.

[0018] Preferably, the generator comprises a first convolutional layer, a first pooling layer, a first fully connected layer, a first recurrent layer and a second fully connected layer which are connected in sequence;

[0019] The first convolutional layer is used to extract the spatial features of the sample data set to be repaired to obtain first spatial feature data;

[0020] The first pooling layer is used to perform a downsampling operation on the first spatial feature data to obtain second spatial feature data;

[0021] The first fully connected layer is used to perform a feature fusion operation on the second spatial feature data to obtain first fused data;

[0022] The first circulation layer is used to extract the time series characteristics of the sample data set to be repaired to obtain first time characteristic data;

[0023] The second fully connected layer is used to obtain the sample repair data based on the first fusion data and the first time feature data mapping.

[0024] Preferably, the discriminator comprises a second convolutional layer, a second pooling layer, a third fully connected layer, a second recurrent layer and a fourth fully connected layer connected in sequence;

[0025] The second convolutional layer is used to extract the spatial features of the sample repair data set to obtain third spatial feature data, and is used to extract the spatial features of the sample real data set to obtain fourth spatial feature data;

[0026] The second pooling layer is used to perform a downsampling operation on the third spatial data to obtain fifth spatial feature data, and is used to perform a downsampling operation on the fourth spatial feature data to obtain sixth spatial feature data;

[0027] The third fully connected layer is used to perform a feature fusion operation on the fifth spatial feature data to obtain second fused data, and is used to perform a feature fusion operation on the sixth spatial feature data to obtain third fused data;

[0028] The second circulation layer is used to extract the time series features of the sample repair data set to obtain second time feature data, and is used to extract the time series features of the sample real data set to obtain third time feature data;

[0029] The fourth fully connected layer is used to determine the degree of closeness between the sample repair data set and the sample real data set based on the second fusion data, the third fusion data, the second time feature data and the third time feature data.

[0030] Preferably, the initial time series data set includes at least one of the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature;

[0031] The initial time series data set includes historical operation data of the nuclear power plant actually collected;

[0032] And / or, the initial time series data set includes simulation operation data of the nuclear power plant generated based on simulation software.

[0033] In a second aspect, a missing data repair method is provided, wherein the missing data repair method is implemented based on a missing data repair model constructed by the missing data repair model construction method described above;

[0034] The missing data repair method comprises:

[0035] Get the actual time series data set;

[0036] Performing a preprocessing operation on the actual time series data set to obtain an actual missing data set;

[0037] The actual missing data set is input into the generator of the missing data repair model to generate an actual repair data set corresponding to the actual missing data set.

[0038] In a third aspect, a system for constructing a missing data repair model is provided, wherein the missing data repair model includes a generator and a discriminator; the construction system includes:

[0039] An initial data acquisition module is used to acquire an original time series data set within a preset time period as an initial time series data set;

[0040] A sample missing data acquisition module is used to perform a data deletion operation on the initial time series data set to obtain a sample missing data set;

[0041] A sample preprocessing module is used to perform a preprocessing operation on the sample missing data set to obtain a sample data set to be repaired, and to perform the preprocessing operation on the initial time series data set to obtain a sample real data set;

[0042] A generator module, used for inputting the sample data set to be repaired into the generator to output a sample repair data set corresponding to the sample data set to be repaired;

[0043] A discriminator module, used for inputting the sample repair data set and the sample real data set into the discriminator to output a discrimination result;

[0044] An adversarial training module, used for feeding back the discrimination result to the generator, performing adversarial training on the generator and the discriminator to construct the missing data repair model;

[0045] The discrimination result is used to characterize the closeness between the sample repair data set and the sample real data set, and the missing data repair model is used to repair the actual repair data set corresponding to the input actual missing data set.

[0046] In a fourth aspect, a missing data repair system is provided, wherein the missing data repair system is implemented based on a missing data repair model constructed by the missing data repair model construction system described above;

[0047] The missing data repair system comprises:

[0048] An actual missing data acquisition module is used to acquire an actual missing data set;

[0049] An actual preprocessing module, used to perform a preprocessing operation on the actual missing data set to obtain an actual data set to be repaired;

[0050] The data generation module is used to input the actual data to be repaired into the generator of the missing data repair model to generate actual repair data corresponding to the actual data to be repaired.

[0051] In a fifth aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, wherein when the processor executes the computer program, it implements the method for constructing a missing data repair model described above, or implements the missing data repair method described above.

[0052] In a sixth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for constructing a missing data repair model described above is implemented, or the missing data repair method described above is implemented.

[0053] On the basis of being in accordance with the common sense in the art, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present disclosure.

[0054] The positive and progressive effects of this disclosure are:

[0055] The missing data repair and model construction method, system, and device disclosed in the present invention, wherein the missing data repair model includes a generator and a discriminator. The missing data repair model is constructed by training the generator and the discriminator against each other, and then the input actual missing data set is repaired by the missing data repair model, and the actual repaired data set corresponding to the actual missing data set is output. A larger range of missing time series data can be repaired, thereby achieving comprehensive, accurate, and effective repair of missing time series data. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A first flow chart of a method for constructing a missing data repair model provided in Embodiment 1 of the present disclosure;

[0057] Figure 2 A schematic diagram of an initial time series data set in the method for constructing a missing data repair model provided in Example 1 of the present disclosure;

[0058] Figure 3 A schematic diagram of a sample missing data set in the method for constructing a missing data repair model provided in Example 1 of the present disclosure;

[0059] Figure 4 A schematic diagram of a sample data set to be repaired after a sample missing data set provided in Embodiment 1 of the present disclosure is operated by a grayscale transformation algorithm;

[0060] Figure 5 A second flow chart of the method for constructing a missing data repair model provided in Embodiment 1 of the present disclosure;

[0061] Figure 6 A schematic diagram of the model structure of the missing data repair model provided in Example 1 of the present disclosure;

[0062] Figure 7 A schematic diagram of a process for repairing missing data provided in Embodiment 2 of the present disclosure;

[0063] Figure 8 A schematic diagram of the structure of a system for constructing a missing data repair model provided in Embodiment 3 of the present disclosure;

[0064] Fig. 9 A schematic diagram of the structure of a missing data repair system provided in Embodiment 4 of the present disclosure;

[0065] Fig.10 This is a schematic diagram of the structure of an electronic device provided in Example 5 of the present disclosure. DETAILED DESCRIPTION

[0066] The present disclosure is further described below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0067] Prefixes such as "first" and "second" are used in the embodiments of the present disclosure only to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers to distinguish description objects in the embodiments of the present disclosure does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the claims or embodiments, and no unnecessary limitation should be constituted due to the use of such prefixes. In addition, in the description of the present embodiment, unless otherwise specified, the meaning of "plurality" is two or more.

[0068] In the embodiments of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0069] This paper uses the Generative Adversarial Network (GAN) as the basic architecture to construct a missing data repair model. The following is a brief introduction to the Generative Adversarial Network.

[0070] A Generative Adversarial Network is a deep learning architecture that creates two different, opposing networks with the goal of having one network generate samples that are different enough from the training set to be difficult for the other network to distinguish between real and fake. The architecture trains two neural networks, the Generator and the Discriminator, to compete with each other to generate new, more realistic data from a given training dataset. GANs are called adversarial networks because the architecture trains two different networks and pits them against each other. The Generator generates new data by taking an input data sample and modifying it as much as possible, and the Discriminator tries to predict whether the generated data output belongs to the original dataset. In other words, the prediction network decides whether the generated data is fake or real, and the Discriminator generates updated, improved versions of the fake data values ​​until the Discriminator can no longer distinguish between the fake data values ​​and the original data values.

[0071] The whole idea of ​​GAN is based on game theory, where the Generator and Discriminator compete against each other and eventually both are at their peak relative to the other network, reaching a Nash equilibrium.

[0072] Example 1

[0073] This embodiment provides a method for constructing a missing data repair model, wherein the missing data repair model includes a generator and a discriminator; Figure 1 As shown, the method for constructing the missing data repair model includes:

[0074] S1. Obtain an original time series data set within a preset time period as an initial time series data set.

[0075] S2. Perform data deletion on the initial time series data set to obtain a sample missing data set.

[0076] S3. Preprocess the sample missing data set to obtain the sample data set to be repaired, and preprocess the initial time series data set to obtain the sample real data set.

[0077] S4. Input the sample data set to be repaired into the generator to output a sample repair data set corresponding to the sample data set to be repaired.

[0078] S5. Input the sample data set to be repaired and the sample real data set into the discriminator to output the discrimination result.

[0079] S6. Feed the discrimination result back to the generator, and conduct adversarial training between the generator and the discriminator to construct a missing data repair model.

[0080] The discriminant result is used to characterize the closeness between the sample repair data set and the sample real data set. Finally, the discriminator cannot distinguish between the sample repair data set and the sample real data set. The missing data repair model is used to repair the actual repair data set corresponding to the input actual missing data set.

[0081] Time series data refers to data collected at different time points, which is used to describe the changes of phenomena over time. This type of data reflects the state or degree of change of a thing or phenomenon over time.

[0082] The original time series data set is a data set consisting of parameters of various types at various times. Data is selected by time period in the original time series data set to obtain the initial time series data set. The initial time series data set is the data within a complete time period (for example, 1-100 seconds).

[0083] Perform data deletion operations on the initial time series data set, such as deleting data from 1 to 10 seconds to obtain sample missing data sets of different time periods, or perform random deletion at different ratios of 10%, 20%, and 30% to obtain sample missing data sets of different ratios. That is, the initial time series data set is the data obtained by slicing the original time series data set according to the preset time period. The initial time series data set is expressed as:

[0084] ;

[0085] in , represents the initial time series data at the tth moment, N is the time sampling length corresponding to the original time series data set, M is the total number of parameter types corresponding to the original time series data set, represents the initial time series data corresponding to the kth parameter type at the tth time, T is the total number of time period slices, Represents the original time series dataset.

[0086] Figure 2 is a schematic diagram of an exemplary initial time series data set, Figure 3 is a schematic diagram of an exemplary sample missing data set, such as Figure 2 and Figure 3 As shown, both the initial time series data set and the sample missing data set are data with time dimension and parameter type dimension.

[0087] The method for constructing a missing data repair model in this embodiment uses a generative adversarial network as a basic architecture, and constructs a missing data repair model through mutual adversarial training of a generator and a discriminator. The missing data repair model is then used to repair the actual missing data set that is input, and an actual repaired data set corresponding to the actual missing data set is output. This can repair a larger range of missing time series data, thereby achieving comprehensive, accurate, and effective repair of missing time series data.

[0088] In an optional implementation, the preprocessing operation includes a grayscale transformation algorithm operation, and the sample real data set and the sample missing data set are both image data sets.

[0089] The grayscale transformation algorithm is used to map the two-dimensional array data into image data for processing. The specific processing is carried out using the following formula:

[0090] ;

[0091] in, The coordinate point is The corresponding image data, The coordinate point in the collected two-dimensional array data is The normalized value at the position, and It is a hyperparameter and can be set based on experience.

[0092] Figure 4 This is a schematic diagram of a sample data set to be repaired after an exemplary sample missing data set is operated by a grayscale transformation algorithm; a typical initial time series data set is multiple sets of one-dimensional time series data with a time dimension (or called a two-dimensional array data with a time dimension and a parameter type dimension), and the correlation and data characteristics between the data are not obvious. If the data is directly processed into a two-dimensional list, data correlation may be lost. Simply mapping the data to an image will result in low contrast, excessive noise in the mapped image, or poor effect. For this reason, the present disclosure uses a grayscale transformation algorithm to map multiple sets of one-dimensional time series data to image data for training, processes the sample real data set and the sample missing data set into an image data set, retains the periodicity and autocorrelation of the data over time, and obtains the overall feature representation of the parameters.

[0093] By using the generative adversarial network as the basic architecture and operating the grayscale transformation algorithm, the sample real data set and the sample missing data set are processed into image data sets, retaining the periodicity and autocorrelation of the data changing over time, making the sample repair data set output by the generator more accurate.

[0094] like Figure 4 As shown, the sample missing data set in this embodiment is mapped into a grayscale image of 64*64 size by the image grayscale transformation algorithm, so the Valid Padding that does not cross the edge sampling is selected (referring to not padding in the convolution operation, and directly performing the convolution operation on the input data), and the step size is 1. After the above transformation, the spatial feature extraction of the data can be completed.

[0095] like Figure 5As shown, a grayscale transformation algorithm is performed on the sample missing data set to obtain a sample data set to be repaired, and a grayscale transformation algorithm is performed on the initial time series data set to obtain a sample real data set. The sample data set to be repaired is input into the generator to output a sample repair data set corresponding to the sample data set to be repaired. The sample data set to be repaired and the sample real data set are input into the discriminator to output a discrimination result, and the discrimination result is fed back to the generator. The generator and the discriminator are trained against each other to construct a missing data repair model.

[0096] The method for constructing the missing data repair model of this embodiment uses a generative adversarial network as the basic architecture, and processes the sample real data set and the sample missing data set into an image data set through a grayscale transformation algorithm operation, thereby retaining the periodicity and autocorrelation of the data over time, making the sample repair data set output by the generator more accurate, and improving the discrimination ability of the discriminator, so that the missing data repair model can repair the missing time series data more comprehensively, accurately and effectively.

[0097] In an optional embodiment, the preprocessing operation includes a normalization operation.

[0098] In an optional embodiment, if Figure 6 As shown, the generator includes a first convolutional layer, a first pooling layer, a first fully connected layer, a first recurrent layer, and a second fully connected layer which are sequentially connected;

[0099] The first convolutional layer is used to extract the spatial features of the sample data set to be repaired to obtain the first spatial feature data;

[0100] The first pooling layer is used to perform a downsampling operation on the first spatial feature data to obtain second spatial feature data;

[0101] The first fully connected layer is used to perform a feature fusion operation on the second spatial feature data to obtain first fused data;

[0102] The first circulation layer is used to extract the time series features of the sample data set to be repaired to obtain the first time feature data;

[0103] The second fully connected layer is used to obtain sample repair data based on the first fusion data and the first time feature data mapping.

[0104] In an optional embodiment, if Figure 6 As shown, the discriminator includes a second convolutional layer, a second pooling layer, a third fully connected layer, a second recurrent layer and a fourth fully connected layer connected in sequence;

[0105] The second convolutional layer is used to extract the spatial features of the sample repair data set to obtain third spatial feature data, and is used to extract the spatial features of the sample real data set to obtain fourth spatial feature data;

[0106] The second pooling layer is used to perform a downsampling operation on the third spatial data to obtain fifth spatial feature data, and is used to perform a downsampling operation on the fourth spatial feature data to obtain sixth spatial feature data;

[0107] The third fully connected layer is used to perform a feature fusion operation on the fifth spatial feature data to obtain second fused data, and is used to perform a feature fusion operation on the sixth spatial feature data to obtain third fused data;

[0108] The second circulation layer is used to extract the time series features of the sample repair data set to obtain the second time feature data, and is used to extract the time series features of the sample real data set to obtain the third time feature data;

[0109] The fourth fully connected layer is used to determine the degree of closeness between the sample repair data set and the sample real data set based on the second fusion data, the third fusion data, the second time feature data and the third time feature data.

[0110] Specifically, more recurrent layers and fully connected layers can be designed in the generator or discriminator to extract data features more effectively.

[0111] That is, the convolution layer of the generator is used to measure the spatial correlation of the sample data set to be repaired, and the recurrent layer of the generator is used to capture the temporal correlation of the sample data set to be repaired. Batch normalization is used in each layer of the generator or discriminator, and Dropout (a function) is used in the fully connected layer to randomly inactivate neurons to prevent overfitting of the model.

[0112] For the recurrent layer, a long short-term memory neural network can be used to implement it, which can effectively prevent gradient disappearance and gradient explosion while effectively extracting the temporal correlation of information before and after the data.

[0113] For the convolutional layer, the convolution calculation formula is:

[0114] ;

[0115] in, Represents the input feature map (i.e. the position of the sample to be repaired dataset) The convolution value on is the position in the weight matrix The corresponding weights, the convolution calculation in the formula is calculated by adding the convolution kernel With the input feature map Perform convolution operation to get the position in the output feature map The convolution value on .

[0116] Specifically, the calculation process of the convolution operation is to convert the convolution kernel Overlay on input feature map The double summation in the formula means that the convolution kernel is traversed in the convolution operation. And the input feature map All positions of . m and n are the convolution kernels The row and column coordinates of and They are convolution kernels. Current position and input feature map The row and column coordinates of the corresponding position. During the calculation process, each element of the convolution kernel The pixel value at the corresponding position in the input feature map Multiply them, and then add these product values ​​to get the position in the output feature map The pixel value on .

[0117] The generator transforms and maps the input low-dimensional vector (such as random noise) to output high-dimensional data (such as images, text, etc.). During the training process of GAN, the goal of the generator is to generate data that is as real as possible to deceive the discriminator. The generator continuously optimizes its own parameters through adversarial training with the discriminator, so that the distribution of the generated data (sample repair data) gradually approaches the distribution of the real data (sample real data). The main task of the discriminator is to distinguish whether the input data comes from a real sample real data set or fake data (sample repair data set) generated by the generator.

[0118] By continuously training against the generator, the discriminator improves its discrimination ability to more accurately identify real data and repaired data. The continuous confrontation between the two allows the data generated by the generator to continuously approach the real number.

[0119] The dynamic adversarial training process of the generator and the discriminator is as follows:

[0120] S701. Randomly initialize the GAN model parameters, assign random small values ​​to the model and bias, break the model symmetry, and prevent the network from falling into a local optimal solution.

[0121] S702, input the sample data set to be repaired into the generator to obtain a sample repair data set, and then input the sample repair data set and the sample real data set into the discriminator together;

[0122] S703, comparing and calculating the sample real data set and the sample repaired data set through the loss function to obtain the loss function value;

[0123] Assume that the true distribution of the sample real data set is , there is group A data in the data set, then . is the real data of each group of samples. The distribution of sample repair data set is ,in is the distribution parameter. Since the neural network itself does not have randomness, it is assumed that the generated data Generated by a random variable z, which itself follows a simple distribution (such as a Gaussian distribution), thus introducing randomness. .

[0124] Through the training confrontation between the generator and the discriminator, the loss function of the missing data repair model is finally:

[0125] ;

[0126] Where D is the discriminator, G is the generator, log(D(x)) represents the probability that the discriminator judges that x belongs to the true distribution, and log(D(G(z))) represents the probability that the discriminator judges that the data G(z) generated by the generator from the noise z belongs to the true distribution. Then It means that the discriminator transforms the generator from the random variable The probability of the samples generated in being identified as false samples. For the generator G, it is hoped to minimize the probability that its generated samples are identified as false samples; while for the discriminator D, it is hoped to maximize the probability of distinguishing real data from generated data.

[0127] S704, the gradient value is back-propagated and chain-derived through the network, so that the GAN model

[0128] The parameters of the model are adjusted in the direction of reducing the loss value; the loss function of the model is recorded as:

[0129] ;

[0130] Then the derivative of the loss function is:

[0131] ;

[0132] ;

[0133] S705. When the generator loss and the discriminator loss both decrease to convergence, the model training process is completed. After the above training, when the discriminator cannot distinguish whether the sample repair data is real data or fake data, the generator reaches the global optimum, that is, When the equal sign is obtained, the training ends and the missing data repair model is obtained. Otherwise, steps S702 to S704 are repeated;

[0134] S706: Use the missing data repair model to repair the actual missing data set.

[0135] In order to evaluate the actual performance of the missing data repair model, RMSE (root mean square error) is used as the model evaluation indicator:

[0136] ;

[0137] Among them, h is the number of grayscale images input into the generative adversarial network, The network determines the grayscale image data. If it is real data, it outputs 1, otherwise it outputs 0. is the real label of grayscale image data, if it is real data, it is recorded as 1, otherwise it is recorded as 0.

[0138] In an optional embodiment, the initial time series data set includes but is not limited to the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature, and the initial time series data set includes the historical operation data of the nuclear power plant actually collected.

[0139] In a specific nuclear power plant application environment, the initial time series data set corresponds to the operating status of the nuclear power plant. The initial time series data set includes the core outlet temperature, the cold pipe section temperature, the hot pipe section temperature, etc. The initial time series data set is the historical operating data of the nuclear power plant actually collected by the sensor.

[0140] The method for constructing a missing data repair model in this embodiment uses the historical operating data of the nuclear power plant to construct the model, which can effectively repair the operating data of the nuclear power plant when the sensor fails, making the actual repair data set output by the missing data repair model more realistic and able to more realistically reflect the daily operating status of the nuclear power plant.

[0141] In an optional embodiment, the initial time series data set includes but is not limited to the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature, and the initial time series data set includes simulation operation data of the nuclear power plant generated based on simulation software.

[0142] In a specific nuclear power plant application environment, the initial time series data set includes core outlet temperature, cold pipe section temperature, hot pipe section temperature, etc. The initial time series data set is simulation operation data of the nuclear power plant generated based on simulation software.

[0143] Considering that real nuclear accident data is extremely scarce and difficult to obtain, mainstream simulation software and platforms are used to simulate the operating status of the accident nuclear power plant. For example, 3KeyMaster (a simulation software) can be used for data simulation to replace the historical operating data of the nuclear power plant actually collected.

[0144] The missing data repair model construction method of this embodiment uses simulation operation data to construct the model, so that the missing data repair model can output the actual repair data set under the nuclear accident state, which is convenient for the staff to accurately analyze the cause of the nuclear accident.

[0145] In an optional embodiment, the initial time series data set includes but is not limited to the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature. The initial time series data set includes the historical operating data of the nuclear power plant actually collected, and the simulated operating data of the nuclear power plant generated based on the simulation software.

[0146] The method for constructing a missing data repair model in this embodiment uses the historical operating data and simulated operating data of the nuclear power plant to construct the model. It can not only effectively repair the operating data of the nuclear power plant when the sensor fails, but also more realistically reflect the daily operating status of the nuclear power plant. It can also output the actual repair data set under the nuclear accident state, which is convenient for the staff to accurately analyze the causes of the nuclear accident and improve the application scenarios of the model.

[0147] Example 2

[0148] This embodiment provides a missing data repair method, which is implemented based on the missing data repair model constructed by the missing data repair model construction method in Embodiment 1; Figure 7 As shown, the missing data repair method includes:

[0149] S8. Obtain an actual time series data set.

[0150] S9. Perform preprocessing operations on the actual time series data set to obtain an actual missing data set.

[0151] S10. Inputting the actual missing data set into a generator of a missing data repair model to generate an actual repaired data set corresponding to the actual missing data set.

[0152] The actual time series data set is a data set with missing data. Specifically, the missing data may be caused by sensor failure.

[0153] The preprocessing operation in this embodiment is the same as the preprocessing operation in Embodiment 1, for example, it may be a grayscale transformation algorithm operation.

[0154] The missing data repair method of this embodiment is based on the missing data repair model constructed in Example 1. It obtains the actual missing data set by performing preprocessing operations on the actual time series data set, inputs the actual missing data set into the generator of the missing data repair model, and then outputs the actual repaired data set corresponding to the actual missing data set, thereby realizing data repair of the actual time series data set, and can repair a larger range of missing time series data, thereby realizing comprehensive, accurate and effective repair of missing time series data.

[0155] Example 3

[0156] The present disclosure also provides an embodiment of a system for constructing a missing data repair model, wherein the missing data repair model includes a generator and a discriminator, such as Figure 8 As shown, the missing data repair model includes a generator and a discriminator; the construction system of the missing data repair model includes:

[0157] An initial data acquisition module 1 is used to acquire an original time series data set within a preset time period as an initial time series data set;

[0158] The sample missing data acquisition module 2 is used to perform a data deletion operation on the initial time series data set to obtain a sample missing data set;

[0159] The sample preprocessing module 3 is used to perform preprocessing operations on the sample missing data set to obtain the sample to-be-repaired data set, and to perform preprocessing operations on the initial time series data set to obtain the sample real data set;

[0160] Generator module 4, used for inputting the sample data set to be repaired into the generator to output the sample repair data corresponding to the sample data set to be repaired;

[0161] The discriminator module 5 is used to input the sample repair data and the sample real data set into the discriminator to output the discrimination result;

[0162] The adversarial training module 6 is used to feed back the discrimination result to the generator, and perform adversarial training on the generator and the discriminator to construct a missing data repair model;

[0163] Among them, the discrimination result is used to characterize the closeness between the sample repair data set and the sample real data set, and the missing data repair model is used to repair the actual repair data set corresponding to the input actual missing data set.

[0164] In an optional implementation, the preprocessing operation includes a grayscale transformation algorithm operation; the sample real data set and the sample missing data set are both image data sets.

[0165] In an optional embodiment, the generator includes a first convolutional layer, a first pooling layer, a first fully connected layer, a first recurrent layer, and a second fully connected layer connected in sequence;

[0166] The first convolutional layer is used to extract the spatial features of the sample data set to be repaired to obtain the first spatial feature data;

[0167] The first pooling layer is used to perform a downsampling operation on the first spatial feature data to obtain second spatial feature data;

[0168] The first fully connected layer is used to perform a feature fusion operation on the second spatial feature data to obtain first fused data;

[0169] The first circulation layer is used to extract the time series features of the sample data set to be repaired to obtain the first time feature data;

[0170] The second fully connected layer is used to obtain a sample repair data set based on the first fusion data and the first time feature data mapping.

[0171] In an optional embodiment, the discriminator includes a second convolutional layer, a second pooling layer, a third fully connected layer, a second recurrent layer and a fourth fully connected layer connected in sequence;

[0172] The second convolutional layer is used to extract the spatial features of the sample repair data set to obtain third spatial feature data, and is used to extract the spatial features of the sample real data set to obtain fourth spatial feature data;

[0173] The second pooling layer is used to perform a downsampling operation on the third spatial data to obtain fifth spatial feature data, and is used to perform a downsampling operation on the fourth spatial feature data to obtain sixth spatial feature data;

[0174] The third fully connected layer is used to perform a feature fusion operation on the fifth spatial feature data to obtain second fused data, and is used to perform a feature fusion operation on the sixth spatial feature data to obtain third fused data;

[0175] The second circulation layer is used to extract the time series features of the sample repair data set to obtain the second time feature data, and is used to extract the time series features of the sample real data set to obtain the third time feature data;

[0176] The fourth fully connected layer is used to determine the degree of closeness between the sample repair data set and the sample real data set based on the second fusion data, the third fusion data, the second time feature data and the third time feature data.

[0177] In an optional embodiment, the initial time series data set includes but is not limited to the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature; the initial time series data set includes the historical operation data of the nuclear power plant actually collected.

[0178] In an optional embodiment, the initial time series data set includes but is not limited to the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature; the initial time series data set includes simulation operation data of the nuclear power plant generated based on simulation software.

[0179] In an optional embodiment, the initial time series data set includes but is not limited to the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature; the initial time series data set includes the historical operating data of the nuclear power plant actually collected, and the simulated operating data of the nuclear power plant generated based on the simulation software.

[0180] The missing data repair model construction system of this embodiment corresponds to the missing data repair model construction method in Example 1. The working principles of the construction system and the construction method are the same, and will not be repeated here.

[0181] The missing data repair model construction system of this embodiment uses a generative adversarial network as its basic architecture, and constructs a missing data repair model through mutual adversarial training of the generator and the discriminator. The missing data repair model is then used to repair the actual missing data set that is input, and the actual repaired data set corresponding to the actual missing data set is output. This can repair a larger range of missing time series data, thereby achieving comprehensive, accurate and effective repair of missing time series data.

[0182] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, in which the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution.

[0183] Example 4

[0184] This embodiment provides a missing data repair system, which is implemented based on the missing data repair model constructed by the missing data repair model construction system in Embodiment 3; Fig. 9 As shown, the missing data repair system includes:

[0185] An actual missing data acquisition module 7 is used to acquire an actual missing data set;

[0186] The actual preprocessing module 8 is used to perform preprocessing operations on the actual missing data set to obtain the actual data set to be repaired;

[0187] The data generation module 9 is used to input the actual data set to be repaired into the generator of the missing data repair model to generate an actual repair data set corresponding to the actual data set to be repaired.

[0188] The missing data repair system of this embodiment corresponds to the missing data repair method in Embodiment 2. The missing data repair system and the missing data repair method have the same working principles, which will not be described in detail here.

[0189] The actual time series data set is a data set with missing data. Specifically, the missing data may be caused by sensor failure.

[0190] The preprocessing operation in this embodiment is the same as the preprocessing operation in Embodiment 1, for example, it may be a grayscale transformation algorithm operation.

[0191] The missing data repair system of this embodiment is based on the missing data repair model constructed by the missing data repair model construction system in Example 3. It obtains the actual missing data set by performing preprocessing operations on the actual time series data set, inputs the actual missing data set into the generator of the missing data repair model, and then outputs the actual repaired data set corresponding to the actual missing data set, thereby realizing data repair of the actual time series data set and realizing comprehensive, accurate and effective repair of missing time series data.

[0192] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, in which the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution.

[0193] Example 5

[0194] Fig.10 This is a structural schematic diagram of an electronic device shown in an example embodiment of the present disclosure, the electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor, and when the processor executes the computer program, it implements the method for constructing a missing data repair model or the missing data repair method of any of the above embodiments. Fig.10 The electronic device 70 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0195] like Fig.10 As shown, the electronic device 70 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 70 may include, but are not limited to: at least one processor 71, at least one memory 72, and a bus 73 connecting different system components (including the memory 72 and the processor 71).

[0196] The bus 73 includes a data bus, an address bus, and a control bus.

[0197] The memory 72 may include a volatile memory, such as a random access memory (RAM) 721 and / or a cache memory 722 , and may further include a read only memory (ROM) 723 .

[0198] The memory 72 may also include a program tool 725 (or utility) having a set (at least one) of program modules 724, such program modules 724 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the implementation of a network environment.

[0199] The processor 71 executes various functional applications and data processing by running the computer program stored in the memory 72, such as the method for constructing a missing data repair model or the missing data repair method of any of the above embodiments.

[0200] The electronic device 70 may also communicate with one or more external devices 74. Such communication may be performed via an input / output (I / O) interface 75. Furthermore, the model generating electronic device 70 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 76. Fig.10 As shown, the network adapter 76 communicates with other modules of the electronic device 70 via the bus 73. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 70, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0201] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0202] Example 6

[0203] The embodiments of the present disclosure also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a missing data repair model or the missing data repair method of any of the above embodiments.

[0204] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0205] In a possible implementation, the present disclosure can also be implemented in the form of a program product, which includes program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps in the method for constructing a missing data repair model in any of the above-mentioned embodiments, or the steps in the missing data repair method.

[0206] Among them, the program code for executing the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as a separate software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0207] Although the specific embodiments of the present disclosure are described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present disclosure, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A method for constructing a missing data repair model, characterized in that: The missing data repair model includes a generator and a discriminator; the construction method includes: Obtaining an original time series data set within a preset time period as an initial time series data set; Performing a data deletion operation on the initial time series data set to obtain a sample missing data set; Performing a preprocessing operation on the sample missing data set to obtain a sample data set to be repaired, and performing the preprocessing operation on the initial time series data set to obtain a sample real data set; Inputting the sample data set to be repaired into the generator to output a sample repair data set corresponding to the sample data set to be repaired; Inputting the sample data set to be repaired and the sample real data set into the discriminator to output a discrimination result; Feeding the discrimination result back to the generator, and performing adversarial training on the generator and the discriminator to construct the missing data repair model; The discrimination result is used to characterize the closeness between the sample repair data set and the sample real data set, and the missing data repair model is used to repair the actual repair data set corresponding to the input actual missing data set.

2. The construction method according to claim 1, characterized in that: The preprocessing operation includes a grayscale transformation algorithm operation; The sample real data set and the sample missing data set are both image data sets.

3. The construction method according to claim 2, characterized in that: The generator includes a first convolutional layer, a first pooling layer, a first fully connected layer, a first recurrent layer, and a second fully connected layer connected in sequence; The first convolutional layer is used to extract the spatial features of the sample data set to be repaired to obtain first spatial feature data; The first pooling layer is used to perform a downsampling operation on the first spatial feature data to obtain second spatial feature data; The first fully connected layer is used to perform a feature fusion operation on the second spatial feature data to obtain first fused data; The first circulation layer is used to extract the time series characteristics of the sample data set to be repaired to obtain first time characteristic data; The second fully connected layer is used to obtain the sample repair data set based on the first fusion data and the first time feature data mapping.

4. The construction method according to claim 3, characterized in that: The discriminator includes a second convolutional layer, a second pooling layer, a third fully connected layer, a second recurrent layer and a fourth fully connected layer which are connected in sequence; The second convolutional layer is used to extract the spatial features of the sample repair data set to obtain third spatial feature data, and is used to extract the spatial features of the sample real data set to obtain fourth spatial feature data; The second pooling layer is used to perform a downsampling operation on the third spatial data to obtain fifth spatial feature data, and is used to perform a downsampling operation on the fourth spatial feature data to obtain sixth spatial feature data; The third fully connected layer is used to perform a feature fusion operation on the fifth spatial feature data to obtain second fused data, and is used to perform a feature fusion operation on the sixth spatial feature data to obtain third fused data; The second circulation layer is used to extract the time series features of the sample repair data set to obtain second time feature data, and is used to extract the time series features of the sample real data set to obtain third time feature data; The fourth fully connected layer is used to determine the degree of closeness between the sample repair data set and the sample real data set based on the second fusion data, the third fusion data, the second time feature data and the third time feature data.

5. The construction method according to any one of claims 1 to 4, characterized in that: The initial time series data set includes at least one of the core outlet temperature, the cold pipe section temperature, and the hot pipe section temperature; The initial time series data set includes historical operation data of the nuclear power plant actually collected; And / or, the initial time series data set includes simulation operation data of the nuclear power plant generated based on simulation software.

6. A missing data repair method, characterized in that: The missing data repair method is implemented based on a missing data repair model constructed by the missing data repair model construction method according to any one of claims 1 to 5; The missing data repair method comprises: Get the actual time series data set; Performing a preprocessing operation on the actual time series data set to obtain an actual missing data set; The actual missing data set is input into the generator of the missing data repair model to generate an actual repair data set corresponding to the actual missing data set.

7. A system for constructing a missing data repair model, characterized in that: The missing data repair model includes a generator and a discriminator; the construction system includes: An initial data acquisition module is used to acquire an original time series data set within a preset time period as an initial time series data set; A sample missing data acquisition module is used to perform a data deletion operation on the initial time series data set to obtain a sample missing data set; A sample preprocessing module is used to perform a preprocessing operation on the sample missing data set to obtain a sample data set to be repaired, and to perform the preprocessing operation on the initial time series data set to obtain a sample real data set; A generator module, used for inputting the sample data set to be repaired into the generator to output sample repair data corresponding to the sample data set to be repaired; A discriminator module, used for inputting the sample repair data and the sample real data set into the discriminator to output a discrimination result; An adversarial training module, used for feeding back the discrimination result to the generator, performing adversarial training on the generator and the discriminator to construct the missing data repair model; The discrimination result is used to characterize the closeness between the sample repair data set and the sample real data set, and the missing data repair model is used to repair the actual repair data set corresponding to the input actual missing data set.

8. A missing data repair system, characterized in that: The missing data repair system is implemented based on the missing data repair model constructed by the missing data repair model construction system according to claim 7; The missing data repair system comprises: An actual missing data acquisition module is used to acquire an actual missing data set; An actual preprocessing module, used to perform a preprocessing operation on the actual missing data set to obtain an actual data set to be repaired; The data generation module is used to input the actual data set to be repaired into the generator of the missing data repair model to generate an actual repair data set corresponding to the actual data set to be repaired.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, the processor implements the method for constructing a missing data repair model described in any one of claims 1 to 5, or implements the missing data repair method described in claim 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the method for constructing a missing data repair model according to any one of claims 1 to 5, or implements the missing data repair method according to claim 6.

Images