Method, system and equipment for processing data missing and medium

Abstract

The invention relates to the technical field of data processing, and discloses a method, system, equipment and medium for processing data missing, and the method comprises the steps: selecting a plurality of diagnosis systems capable of measuring a physical quantity to be measured according to a preset strategy, obtaining the historical diagnosis data of a fault system and a non-fault system for the physical quantity to be measured, taking the historical diagnosis data of the non-fault system and the historical diagnosis data of the fault system, which are subjected to time sequence unification, as a plurality of intermediate diagnosis data, taking the plurality of intermediate diagnosis data as a data set, dividing the data set into a training set, a verification set and a test set according to a certain proportion, and calculating an initial learning model; and taking the diagnosis data corresponding to the fault system in the training set as the output of the initial learning model, and taking the diagnosis data of all non-fault systems as the input of the initial learning model. The data incomplete technology provided by the invention has better accuracy.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, system, device and medium for processing missing data. Background Art

[0002] In the process of building a big data platform, when key data is missing, appropriate methods should be used to handle these missing values ​​in the data preprocessing stage to ensure the effectiveness and accuracy of subsequent analysis, modeling or decision support. The processing strategy for key data missing usually includes the following aspects:

[0003] Data completion:

[0004] ① Mean filling: For numerical features, if there are few missing values, the missing values ​​can be replaced by the average value of the non-missing values ​​of the feature. However, this method may conceal the true distribution of the data, especially when the data is skewed or has extreme values.

[0005] ②Median filling: Similar to mean filling, it is suitable for cases where the data distribution is skewed, and the missing values ​​are replaced by the median of the non-missing values ​​of the feature.

[0006] ③ Mode filling: For categorical variables, the most frequently occurring category value can be used to fill in missing values.

[0007] Interpolation method:

[0008] ① Linear interpolation: Applicable to ordered data. In time series or other continuous scales, it estimates missing values ​​by fitting a straight line through known adjacent data points.

[0009] ② Polynomial interpolation: For data with more complex trends, high-order polynomials can be used for interpolation.

[0010] ③Spline interpolation: In some cases, the spline function can provide smoother interpolation results, especially suitable for processing irregularly distributed data points.

[0011] Delete or ignore

[0012] As for the data completion scheme, in nuclear fusion experiments, it often happens that in a certain experiment, the measuring instrument does not collect any signal from the beginning to the end. At this time, if the ordinary data completion scheme is used, it is difficult to find relatively accurate values ​​to fill in due to the lack of reference data; similarly, the interpolation method cannot be used when a large amount of data is missing; if it is completely deleted, then this channel of signal data will be completely missing. Sometimes, many algorithms rely on this data, which will cause the downstream algorithms to fail to operate normally. Summary of the invention

[0013] The present invention provides a method, system, device and medium for processing missing data, which mainly solves the technical solution of how to complete the missing data in a large scientific device or a complex system when an instrument measuring a certain signal does not collect signal data or only collects very little signal data.

[0014] The present invention is achieved through the following technical solutions:

[0015] A method for dealing with missing data, comprising:

[0016] S1, setting one of the diagnostic systems measuring a certain physical quantity as a faulty system, setting the physical quantity as a physical quantity to be measured, selecting a plurality of diagnostic systems capable of measuring the physical quantity to be measured according to a preset strategy, and the plurality of diagnostic systems as non-faulty systems;

[0017] S2, obtaining historical diagnostic data of the faulty system and the non-faulty system for the physical quantity to be measured, and unifying the time sequence of the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system;

[0018] S3, treating the historical diagnostic data of the non-faulty system and the historical diagnostic data of the faulty system that have been time-sequence unified as a plurality of intermediate diagnostic data, wherein each of the intermediate diagnostic data includes diagnostic data collected at the same sampling time for each of the non-faulty system and the faulty system;

[0019] S4, treating the plurality of intermediate diagnostic data as a data set and dividing the data set into a training set, a validation set and a test set according to a certain ratio;

[0020] S5, designing an initial learning model, taking the diagnostic data corresponding to the faulty system in the training set as the output of the initial learning model, taking the diagnostic data of all non-faulty systems as the input of the initial learning model, training the initial learning model to obtain an ultimate learning model corresponding to the faulty system, and verifying the ultimate learning model through a validation set, and at the same time, testing the ultimate learning model through a test set;

[0021] S6. When a diagnostic system of the physical quantity to be measured fails, current diagnostic data of several diagnostic systems of the physical quantity to be measured except the failed system are obtained, and the current diagnostic data are input into the ultimate learning model corresponding to the failed system to obtain the diagnostic data of the diagnostic system after the failure.

[0022] As an optimization, the preset strategy is specifically:

[0023] A1. Based on the spatial position, select a number of first M first diagnostic systems that are closest to the faulty system and measure the physical quantity to be measured, where M is a positive integer;

[0024] A2, selecting the first N second diagnostic systems with the highest sampling accuracy for measuring the physical quantity to be measured, where N is a positive integer;

[0025] A3, selecting the intersection of M first diagnostic systems and N second diagnostic systems to obtain a third diagnostic system, and determining whether the number of the third diagnostic systems is greater than a lower threshold, if so, jump to A4, otherwise, increase M and N, and jump to A1;

[0026] A4, determining whether the number of the third diagnostic systems is greater than an upper threshold value, if so, jumping to A5, otherwise, treating a number of the third diagnostic systems as non-faulty systems;

[0027] A5. Select the first V third diagnostic systems closest to the faulty system, where V is the upper threshold value. If there are at least two first V third diagnostic systems closest to the faulty system, select the third diagnostic system with the highest accuracy from the at least two first V third diagnostic systems as the first V third diagnostic system.

[0028] As an optimization, in S2, the specific process of unifying the time sequence of the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system is:

[0029] Determine whether the sampling frequency of the non-faulty system is higher than the sampling frequency of the faulty system. If so, downsample the historical diagnostic data corresponding to the non-faulty system according to the sampling frequency of the faulty system. Otherwise, complete the historical diagnostic data corresponding to the non-faulty system by interpolation.

[0030] As an optimization, the initial learning model is a machine learning model.

[0031] As an optimization, the initial learning model is a deep learning model.

[0032] As an optimization, when the historical diagnostic data corresponding to the non-faulty system is supplemented by an interpolation method, the accuracy of the supplemented diagnostic data is the same as the sampling accuracy of the non-faulty system.

[0033] As an optimization, the upper threshold and the lower threshold are equal; or the upper threshold is higher than the lower threshold.

[0034] The present invention also discloses a system for processing missing data, comprising:

[0035] A setting module, used for setting one of the diagnostic systems measuring a certain physical quantity as a faulty system, setting the physical quantity as a physical quantity to be measured, selecting a number of diagnostic systems capable of measuring the physical quantity to be measured according to a preset strategy, and setting the number of diagnostic systems as non-faulty systems;

[0036] An acquisition module, used for acquiring historical diagnostic data of the faulty system and the non-faulty system for the physical quantity to be measured, and performing time sequence unification on the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system;

[0037] A statistical module, used for treating the historical diagnostic data of the non-faulty system and the historical diagnostic data of the faulty system which are time-sequence unified as a plurality of intermediate diagnostic data, wherein each of the intermediate diagnostic data includes diagnostic data collected at the same sampling time for each of the non-faulty system and the faulty system;

[0038] an allocation module, which regards the plurality of intermediate diagnostic data as a data set and divides the data set into a training set, a validation set and a test set according to a certain ratio;

[0039] A model training module is used to design an initial learning model, use the diagnostic data corresponding to the faulty system in the training set as the output of the initial learning model, use the diagnostic data of all non-faulty systems as the input of the initial learning model, train the initial learning model to obtain an ultimate learning model corresponding to the faulty system, and verify the ultimate learning model through a verification set, and at the same time, test the ultimate learning model through a test set;

[0040] The input module is used to obtain the current diagnostic data of several diagnostic systems of the physical quantity to be measured except the faulty system when a diagnostic system of the physical quantity to be measured fails, and input the current diagnostic data into the ultimate learning model corresponding to the faulty system to obtain the diagnostic data of the diagnostic system after the failure.

[0041] The present invention also discloses an electronic device, comprising at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method for processing data missing as described above.

[0042] The present invention also discloses a storage medium storing a computer program, wherein the computer program implements the aforementioned method for processing missing data when executed by a processor.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] The present invention uses strongly correlated physical quantities with different sampling accuracies and spatial positions, and can complete data with each other through machine learning / deep learning algorithms. In addition, when a large amount of data is missing, the data fitting results can still obtain good accuracy. This data completion technology is particularly important for high-reliability systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0046] Figure 1 A flowchart of a method for processing missing data according to the present invention;

[0047] Figure 2 for Figure 1 Another sequence diagram of . DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0049] For a large system, the measurement of key physical quantities is very complicated. Usually, different technical solutions are used to obtain results of different accuracy or different positions. For example, the measurement of plasma density in the Tokamak experimental device can adopt more than a dozen diagnostic solutions: carbon dioxide interferometer, far-infrared laser interferometer, Thomson scattering system, mid-plane probe system, microwave frequency modulation reflection system, etc. After the raw data of each system is acquired, complex calculations and processing are required to obtain the results that researchers want. Once some subsystems fail in these diagnostic systems, some acquisition channels will not be able to collect data. This is common in high-precision physical experiments. Once there is no data in the acquisition channel, researchers cannot calculate the physical indicators that depend on the data.

[0050] At this time, the technical solution of the present invention is: using diagnostic data of different positions and different accuracies in the same experiment, and using machine learning / deep learning algorithms to fit the missing data to complete the data. The detailed process is as follows.

[0051] This embodiment 1 provides a method for processing missing data, such as Figure 1-2 As shown, including:

[0052] S1, setting one of the diagnostic systems measuring a certain physical quantity as a faulty system, setting the physical quantity as a physical quantity to be measured, selecting a plurality of diagnostic systems capable of measuring the physical quantity to be measured according to a preset strategy, and the plurality of diagnostic systems as non-faulty systems;

[0053] Assume that a certain diagnostic device A is damaged due to long-term use, and after the experiment is completed, it is found that the data is missing. In other words, the diagnostic device A is the fault system mentioned in the present invention.

[0054] After the faulty system is determined, it is necessary to select other diagnostic systems that measure the same physical quantity as the faulty system according to a preset strategy. In this embodiment, the preset strategy is specifically:

[0055] A1. Based on the spatial position, select a number of first M first diagnostic systems that are closest to the faulty system and measure the physical quantity to be measured, where M is a positive integer;

[0056] First, according to the spatial position and the distance to the fault system from near to far, the adjacent diagnostic systems are checked, and M diagnostic systems that measure the same physical quantity are selected from these diagnostic systems, and the selected ones are made the first diagnostic system.

[0057] A2, selecting the first N second diagnostic systems with the highest sampling accuracy for measuring the physical quantity to be measured, where N is a positive integer;

[0058] Then, according to the sampling accuracy, from high to low, a diagnostic system for measuring the same physical quantity is selected, and the selected one is made the second diagnostic system.

[0059] In the Tokamak experiment, the diagnostic equipment (diagnostic system) is numerous, high-precision, but relatively fragile, often fails, and lacks a reliable means of obtaining usable experimental data. In the diagnosis of plasma density, similar technical solutions can be applied, as shown below:

[0060] There are 9 diagnostic systems for plasma density measurement, namely, far-infrared laser interferometer, carbon dioxide interferometer, core Thomson scattering system, divertor Thomson scattering system, beam emission spectrum diagnosis, microwave Doppler reflection system, mid-plane probe system, target plate probe system, microwave frequency modulation reflection system. These systems are distributed in different positions of the whole device, and the diagnostic physical principles are different, and the adopted frequencies and precisions are also quite different.

[0061] In step A1, for example, if the faulty system is a far-infrared laser interferometer, then based on the spatial position, the first M diagnostic systems closest to the far-infrared laser interferometer are found as the first diagnostic system from the eight non-faulty systems (carbon dioxide interferometer, core Thomson scattering system, divertor Thomson scattering system, beam emission spectrum diagnosis, microwave Doppler reflection system, mid-plane probe system, target plate probe system, and microwave frequency modulation reflection system) except the faulty system.

[0062] In step A2, based on the sampling accuracy, the first N diagnostic systems with the highest sampling accuracy are found from the 8 non-faulty systems (carbon dioxide interferometer, core Thomson scattering system, divertor Thomson scattering system, beam emission spectrum diagnosis, microwave Doppler reflection system, mid-plane probe system, target plate probe system, microwave frequency modulation reflection system) as the second diagnostic system.

[0063] A3, selecting the intersection of M first diagnostic systems and N second diagnostic systems to obtain a third diagnostic system, and determining whether the number of the third diagnostic systems is greater than a lower threshold, if so, jump to A4, otherwise, increase M and N, and jump to A1;

[0064] In order to have enough data to train the initial learning model, so that the training of the initial learning model can be more accurate, a lower threshold needs to be set to ensure the accuracy of the training of the initial learning model. However, in order to prevent the initial learning model training from taking too long due to too much data, an upper threshold needs to be set.

[0065] In this embodiment, the upper threshold and the lower threshold may be equal or unequal.

[0066] For example, in this embodiment, the upper threshold and the lower threshold are equal to 4.

[0067] For example, the first diagnostic system includes a carbon dioxide interferometer, a core Thomson scattering system, a divertor Thomson scattering system, a beam emission spectrum diagnosis, and a microwave Doppler reflection system; the second diagnostic system includes a beam emission spectrum diagnosis, a microwave Doppler reflection system, a mid-plane probe system, a target plate probe system, and a microwave frequency modulation reflection system. Then, the third diagnostic system obtained by the intersection is a microwave Doppler reflection system. Since the number of the third diagnostic systems is 1 at this time, which is less than the lower limit of the threshold, it is necessary to increase the values ​​of M and N so that the intersection of the first diagnostic system and the second diagnostic system becomes more.

[0068] A4, determining whether the number of the third diagnostic systems is greater than an upper threshold value, if so, jumping to A5, otherwise, treating a number of the third diagnostic systems as non-faulty systems;

[0069] If the intersection of the first diagnostic system and the second diagnostic system increases after increasing the values ​​of M and N, resulting in the number of third diagnostic systems being greater than the upper threshold, then it is necessary to select third diagnostic systems between the upper threshold and the lower threshold from these third diagnostic systems. Specifically, refer to step A5. If the number of third diagnostic systems in the intersection of the first diagnostic system and the second diagnostic system is just between the upper threshold and the lower threshold after increasing the values ​​of M and N, then these third diagnostic systems can be selected.

[0070] A5. Select the first V third diagnostic systems closest to the faulty system, where V is the upper threshold value. If there are at least two first V third diagnostic systems closest to the faulty system, select the third diagnostic system with the highest accuracy from the at least two first V third diagnostic systems as the first V third diagnostic system.

[0071] The third diagnostic system that is closest in spatial position is selected first. If there are two third diagnostic systems ranked in Vth place, the accuracy of the two systems is compared.

[0072] The third diagnostic system is the non-faulty system. Sampling accuracy refers to the amount of sampled data per unit time. For example, if 100 data are sampled in one second and 1000 data are sampled in one second, the accuracy of sampling 1000 data in one second is higher than that of sampling 100 data in one second.

[0073] The data collected by all diagnostic systems that measure this physical quantity before the fault occurs are obtained (including normal data collected before the device A that fails) , that is, step S2.

[0074] S2, obtaining historical diagnostic data of the faulty system and the non-faulty system for the physical quantity to be measured, and unifying the time sequence of the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system;

[0075] In S2, the specific process of unifying the time sequence of the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system is as follows:

[0076] Determine whether the sampling frequency of the non-faulty system is higher than the sampling frequency of the faulty system. If so, downsample the historical diagnostic data corresponding to the non-faulty system according to the sampling frequency of the faulty system. Otherwise, supplement the historical diagnostic data corresponding to the non-faulty system by interpolation. When the historical diagnostic data corresponding to the non-faulty system is supplemented by interpolation, the accuracy of the supplemented diagnostic data is the same as the sampling accuracy of the non-faulty system.

[0077] Since the data sampling frequencies of different systems are different, it is necessary to complete the data according to the data sampling frequency of device A. Generally, for sampling frequencies lower than the target data, the present invention adopts the interpolation method; for sampling frequencies higher than the target data, the present invention directly downsamples the original diagnostic data of the non-faulty system.

[0078] Note: The processed data comes from different acquisition systems and corresponds to the distribution of the same physical quantity at different locations in space. Although it is related to the target data, this connection is still difficult to accurately express through precise mathematical or physical formulas.

[0079] Since for the same complex system, these measured physical quantities generally do not experience sudden changes in magnitude, this creates the prerequisite for fitting missing data through machine learning / deep learning algorithms.

[0080] Therefore, the process is as follows:

[0081] S3, treating the historical diagnostic data of the non-faulty system and the historical diagnostic data of the faulty system that have been time-sequence unified as a plurality of intermediate diagnostic data, wherein each of the intermediate diagnostic data includes diagnostic data collected at the same sampling time for each of the non-faulty system and the faulty system;

[0082] S4, treating the plurality of intermediate diagnostic data as a data set and dividing the data set into a training set, a validation set and a test set according to a certain ratio;

[0083] The intermediate diagnostic data that has undergone the aforementioned data preprocessing stage is randomly sampled using an algorithm and processed into a training set, a validation set, and a test set respectively.

[0084] S5. Design an initial learning model, use the diagnostic data corresponding to the faulty system in the training set as the output of the initial learning model, use the diagnostic data of all non-faulty systems as the input of the initial learning model, train the initial learning model to obtain the ultimate learning model corresponding to the faulty system, and verify the ultimate learning model through a validation set. At the same time, test the ultimate learning model through a test set. The initial learning model is a machine learning model or a deep learning model.

[0085] In the training set data, the historical data collected by diagnostic device A is used as the dependent variable, that is, the output of the initial learning model, and the data collected by other devices (non-faulty systems) is used as the independent variable, that is, the input of the initial learning model. Then, after selecting the appropriate machine learning / deep learning algorithm (model) and loss function, the data of diagnostic device A is fitted.

[0086] After several iterations, the loss function result is calculated, and when the set threshold is met, the iteration will stop. At this time, the weight matrix corresponding to the algorithm is the learning model we want to obtain, that is, the ultimate learning model.

[0087] Each diagnostic system (diagnostic equipment) is designed to carry out the above process, and an ultimate learning model can be obtained accordingly.

[0088] S6. When a diagnostic system of the physical quantity to be measured fails, current diagnostic data of several diagnostic systems of the physical quantity to be measured except the failed system are obtained, and the current diagnostic data are input into the ultimate learning model corresponding to the failed system to obtain the diagnostic data of the diagnostic system after the failure.

[0089] When the above fault occurs, in the same experiment, the data collected by all the non-faulty systems measuring this physical quantity (device A has no data) are obtained, and the data collected by other diagnostic devices except diagnostic device A are preprocessed according to steps S1-S2. Then, the corresponding ultimate learning model is called, and the data of normal devices is used as input to solve all the time series data of fault diagnosis device A in this experiment, so as to complete the data completion.

[0090] There are 9 diagnostic systems for plasma density measurement, namely, far-infrared laser interferometer, carbon dioxide interferometer, core Thomson scattering system, divertor Thomson scattering system, beam emission spectrum diagnosis, microwave Doppler reflection system, mid-plane probe system, target plate probe system, microwave frequency modulation reflection system. These systems are distributed in different positions of the whole device, and the diagnostic physical principles are different, and the adopted frequencies and precisions are also quite different.

[0091] During the plasma discharge process, the rupture prediction algorithm often needs to rely on a certain plasma density signal, and because plasma rupture means the uncontrolled abnormal release of energy, it is generally more dangerous. Therefore, the reliability of rupture prediction is very important. Once data missing is found, modifying the code and retraining may affect the performance of the model, and there is uncertainty in the results. Therefore, it is generally necessary to complete the missing data.

[0092] In the method of the present invention, by utilizing the density measurement results of the nine diagnostic systems in history, a corresponding neural network model (ultimate learning model) can be trained for each of the nine diagnostic systems. For example, when one of the acquisition channels of the carbon dioxide interferometer system fails, resulting in the carbon dioxide interferometer system having no diagnostic data, the missing data can be successfully supplemented by the ultimate learning model of the carbon dioxide interferometer system.

[0093] Embodiment 2 also discloses a system for processing missing data, including:

[0094] A setting module, used for setting one of the diagnostic systems measuring a certain physical quantity as a faulty system, setting the physical quantity as a physical quantity to be measured, selecting a number of diagnostic systems capable of measuring the physical quantity to be measured according to a preset strategy, and setting the number of diagnostic systems as non-faulty systems;

[0095] An acquisition module, used for acquiring historical diagnostic data of the faulty system and the non-faulty system for the physical quantity to be measured, and performing time sequence unification on the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system;

[0096] A statistical module, used for treating the historical diagnostic data of the non-faulty system and the historical diagnostic data of the faulty system which are time-sequence unified as a plurality of intermediate diagnostic data, wherein each of the intermediate diagnostic data includes diagnostic data collected at the same sampling time for each of the non-faulty system and the faulty system;

[0097] an allocation module, which regards the plurality of intermediate diagnostic data as a data set and divides the data set into a training set, a validation set and a test set according to a certain ratio;

[0098] A model training module is used to design an initial learning model, use the diagnostic data corresponding to the faulty system in the training set as the output of the initial learning model, use the diagnostic data of all non-faulty systems as the input of the initial learning model, train the initial learning model to obtain an ultimate learning model corresponding to the faulty system, and verify the ultimate learning model through a verification set, and at the same time, test the ultimate learning model through a test set;

[0099] The input module is used to obtain the current diagnostic data of several diagnostic systems of the physical quantity to be measured except the faulty system when a diagnostic system of the physical quantity to be measured fails, and input the current diagnostic data into the ultimate learning model corresponding to the faulty system to obtain the diagnostic data of the diagnostic system after the failure.

[0100] Example 3 also discloses an electronic device, comprising at least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method for processing data missing as described in Example 1.

[0101] Embodiment 4 further discloses a storage medium storing a computer program, which, when executed by a processor, implements a method for processing missing data in embodiment 1.

[0102] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for processing missing data, characterized in that: include: S1, setting one of the diagnostic systems measuring a certain physical quantity as a faulty system, setting the physical quantity as a physical quantity to be measured, selecting a plurality of diagnostic systems capable of measuring the physical quantity to be measured according to a preset strategy, and the plurality of diagnostic systems as non-faulty systems; S2, obtaining historical diagnostic data of the faulty system and the non-faulty system for the physical quantity to be measured, and unifying the time sequence of the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system; S3, treating the historical diagnostic data of the non-faulty system and the historical diagnostic data of the faulty system that have been time-sequence unified as a plurality of intermediate diagnostic data, wherein each of the intermediate diagnostic data includes diagnostic data collected at the same sampling time for each of the non-faulty system and the faulty system; S4, treating the plurality of intermediate diagnostic data as a data set and dividing the data set into a training set, a validation set and a test set according to a certain ratio; S5, designing an initial learning model, taking the diagnostic data corresponding to the faulty system in the training set as the output of the initial learning model, taking the diagnostic data of all non-faulty systems as the input of the initial learning model, training the initial learning model to obtain an ultimate learning model corresponding to the faulty system, and verifying the ultimate learning model through a validation set, and at the same time, testing the ultimate learning model through a test set; S6. When a diagnostic system of the physical quantity to be measured fails, current diagnostic data of several diagnostic systems of the physical quantity to be measured except the failed system are obtained, and the current diagnostic data are input into the ultimate learning model corresponding to the failed system to obtain the diagnostic data of the diagnostic system after the failure.

2. A method for processing missing data according to claim 1, characterized in that: The preset strategy is specifically: A1. Based on the spatial position, select a number of first M first diagnostic systems that are closest to the faulty system and measure the physical quantity to be measured, where M is a positive integer; A2, selecting the first N second diagnostic systems with the highest sampling accuracy for measuring the physical quantity to be measured, where N is a positive integer; A3, selecting the intersection of M first diagnostic systems and N second diagnostic systems to obtain a third diagnostic system, and determining whether the number of the third diagnostic systems is greater than a lower threshold, if so, jump to A4, otherwise, increase M and N, and jump to A1; A4, determining whether the number of the third diagnostic systems is greater than an upper threshold value, if so, jumping to A5, otherwise, treating a number of the third diagnostic systems as non-faulty systems; A5. Select the first V third diagnostic systems closest to the faulty system, where V is the upper threshold value. If there are at least two first V third diagnostic systems closest to the faulty system, select the third diagnostic system with the highest accuracy from the at least two first V third diagnostic systems as the first V third diagnostic system.

3. A method for processing missing data according to claim 1, characterized in that: In S2, the specific process of unifying the time sequence of the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system is as follows: Determine whether the sampling frequency of the non-faulty system is higher than the sampling frequency of the faulty system. If so, downsample the historical diagnostic data corresponding to the non-faulty system according to the sampling frequency of the faulty system. Otherwise, complete the historical diagnostic data corresponding to the non-faulty system by interpolation.

4. The method for processing missing data according to claim 1, characterized in that: The initial learning model is a machine learning model.

5. The method for processing missing data according to claim 1, characterized in that: The initial learning model is a deep learning model.

6. A method for processing missing data according to claim 3, characterized in that: When the historical diagnostic data corresponding to the non-faulty system is supplemented by the interpolation method, the accuracy of the supplemented diagnostic data is the same as the sampling accuracy of the non-faulty system.

7. A method for processing missing data according to claim 2, characterized in that: The upper threshold and the lower threshold are equal; or the upper threshold is higher than the lower threshold.

8. A system for processing missing data, characterized in that: include: A setting module, used for setting one of the diagnostic systems measuring a certain physical quantity as a faulty system, setting the physical quantity as a physical quantity to be measured, selecting a number of diagnostic systems capable of measuring the physical quantity to be measured according to a preset strategy, and setting the number of diagnostic systems as non-faulty systems; An acquisition module, used for acquiring historical diagnostic data of the faulty system and the non-faulty system for the physical quantity to be measured, and performing time sequence unification on the historical diagnostic data of the non-faulty system based on the sampling frequency of the faulty system; A statistical module, used for treating the historical diagnostic data of the non-faulty system and the historical diagnostic data of the faulty system which are time-sequence unified as a plurality of intermediate diagnostic data, wherein each of the intermediate diagnostic data includes diagnostic data collected at the same sampling time for each of the non-faulty system and the faulty system; an allocation module, which regards the plurality of intermediate diagnostic data as a data set and divides the data set into a training set, a validation set and a test set according to a certain ratio; A model training module is used to design an initial learning model, use the diagnostic data corresponding to the faulty system in the training set as the output of the initial learning model, use the diagnostic data of all non-faulty systems as the input of the initial learning model, train the initial learning model to obtain an ultimate learning model corresponding to the faulty system, and verify the ultimate learning model through a verification set, and at the same time, test the ultimate learning model through a test set; The input module is used to obtain the current diagnostic data of several diagnostic systems of the physical quantity to be measured except the faulty system when a diagnostic system of the physical quantity to be measured fails, and input the current diagnostic data into the ultimate learning model corresponding to the faulty system to obtain the diagnostic data of the diagnostic system after the failure.

9. An electronic device, characterized in that: It includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method for processing data missing as described in any one of claims 1 to 7.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a method for processing missing data according to any one of claims 1 to 8 is implemented.

Images