Magnet quenching detection method based on attention mechanism feature fusion

Abstract

The invention relates to the technical field of magnet quenching detection, and discloses a magnet quenching detection method based on attention mechanism feature fusion, which comprises the following steps: preprocessing existing time sequence data and to-be-detected time sequence data of a superconducting magnet to obtain an existing data signal frame and a to-be-detected data signal frame; extracting time domain features and frequency domain features from the existing data signal frame and the data signal frame to be detected; performing convolutional feature extraction on the existing data signal frame and the data signal frame to be detected by using a convolutional neural network; for an existing data signal frame, splicing the extracted time domain features and frequency domain features with the convolution features, extracting attention weights of the spliced features by using an attention network, and obtaining fusion features by using the extracted attention weights; using a deep neural network as a classifier to carry out quench detection on the fused features to obtain a detection result; training the detection model, and determining an optimal model; and obtaining a detection result of the to-be-detected data signal frame according to the optimal model. Therefore, magnet quenching detection is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of magnet quench detection, and in particular to a magnet quench detection method based on attention mechanism feature fusion. Background Art

[0002] Superconducting magnetic levitation technology has become an important branch of modern rail transportation. It uses the strong magnetic field formed by superconducting materials in the superconducting state to interact with the ground magnetic field to suspend and drive the train. In this way, the train can be driven at high speed with very little loss of electrical energy, which is one of the efficient ways to replace traditional internal combustion engines and motor drives.

[0003] Superconducting magnets are one of the core components of the power module of superconducting maglev trains. In order for the maglev train to work normally, the superconducting magnets must be stable in the superconducting state. The maintenance of the superconducting state is related to multiple physical fields such as electricity, force, heat, and magnetism. Larger fluctuations will cause the superconducting magnet to lose its superconducting state, that is, the superconducting magnet will quench.

[0004] The current methods for superconducting magnet quench detection mainly include direct detection method and sequence analysis method. The direct detection method mainly monitors certain parameters (such as voltage, temperature or magnetic field) to determine whether a quench event has occurred. The sequence analysis method mainly uses superconducting magnet time series data, such as voltage signals, temperature information, magnetic field strength signals, etc., to achieve quench detection by analyzing the correlation and changes in components in the signal.

[0005] However, the direct detection method is susceptible to noise and electromagnetic interference, and is prone to false detection in the complex multi-field environment of superconducting magnets. At the same time, the direct detection method mainly relies on the threshold to determine the quenching of the superconducting magnet. Due to the differences in the structure of the superconducting magnet and the installation position of the sensor, the threshold setting is different, resulting in hysteresis in the quenching judgment, which affects the safe operation of the equipment. The sequence analysis method is highly dependent on the quantity and quality of historical data. In many actual engineering applications, in order to ensure the stability of the system, the sensors of the equipment are usually arranged redundantly, and the large amount of data collected by multiple sensors contains a lot of irrelevant information, which affects the detection accuracy of the sequence analysis method. Summary of the invention

[0006] The present invention provides a magnet quench detection method based on attention mechanism feature fusion, which can solve the technical problems in the prior art.

[0007] The present invention provides a magnetic quench detection method based on attention mechanism feature fusion, wherein the method comprises:

[0008] Preprocessing the existing time series data and the time series data to be tested of the superconducting magnet to obtain an existing data signal frame and a data signal frame to be tested;

[0009] Extracting time domain features and frequency domain features from existing data signal frames and data signal frames to be tested;

[0010] Using convolutional neural network to extract convolutional features of existing data signal frames and data signal frames to be tested;

[0011] For existing data signal frames, the extracted time domain features and frequency domain features are spliced ​​with the convolution features, the attention weights of the spliced ​​features are extracted using the attention network, and the fusion features are obtained using the extracted attention weights;

[0012] Use deep neural network as classifier to perform quench detection on the fused features to obtain the detection results;

[0013] The training data set is obtained by using the detection results and the cross entropy loss function. The training data set includes a training set and a validation set.

[0014] The detection model is trained according to the training set, and the detection model includes a convolutional neural network, an attention network, and a deep neural network to obtain a trained detection model. The trained detection model is verified according to the verification set, and the trained detection model with the highest accuracy score of the verification set is determined as the optimal model;

[0015] According to the optimal model and the time domain characteristics and frequency domain characteristics of the data signal frame to be tested, convolution feature extraction, fusion feature acquisition and quench detection are performed on the data signal frame to be tested in sequence to obtain the detection result of the data signal frame to be tested.

[0016] Preferably, preprocessing the existing time series data and the time series data to be measured of the superconducting magnet includes:

[0017] The existing time series data and the time series data to be measured of the superconducting magnet are framed and windowed.

[0018] Preferably, the existing time series data of the superconducting magnet is framed and windowed by the following formula:

[0019] x i =w i ·X,

[0020] Among them, X represents the existing time series data, w i represents the i-th Hanning frame window, · represents the vector dot multiplication operator, x i Indicates that there is a data signal frame.

[0021] Preferably, the time domain features include kurtosis features and root mean square features, and the kurtosis features and root mean square features extracted from the existing data signal frame are respectively:

[0022]

[0023]

[0024] Among them, μ k and σ k Indicates that there is data signal frame x i The corresponding mean and standard deviation, E represents the expectation operator, Kurt i Indicates that there is data signal frame x i The kurtosis eigenvalue of T represents the existing data signal frame x i Length, RMS i Indicates that there is data signal frame x i is the RMS eigenvalue of , and t represents time.

[0025] Preferably, the frequency domain features include center frequency and power spectrum. The center frequency and power spectrum extracted from the existing data signal frame are:

[0026]

[0027] P(f i )=|X(k) i | 2 ,

[0028] Among them, f i represents frequency, P(f i ) represents the frequency f i Existing data signal frame x at i The power spectrum, centfreq i Indicates that there is data signal frame x i The center frequency of, k represents the frequency index range, X(k) i Indicates that there is data signal frame x i The frequency domain signal in the frequency index range k.

[0029] Preferably, the convolution feature extraction of the existing data signal frame is performed by the following formula:

[0030] F ci =Pool 2 (C 3×3 (Pool 1 (C 5×5 (X i )))),

[0031] Among them, C m×m represents a convolutional layer with a kernel size of m, Pool represents a pooling layer, and X i Indicates that there is data signal frame x i The frequency domain signal, F ci For the existing data signal frame x i The convolution features.

[0032] Preferably, the extracted time-domain features and frequency-domain features are concatenated with the convolutional features by the following formula:

[0033] Fea = [Ftf i , F ci ,

[0034] Ftf i = [Kurt i , RMS i , centfreq i , P(f i )],

[0035] where Fea is the feature after concatenation of the existing data signal frames.

[0036] Preferably, the attention weights of the concatenated features are extracted and the fused features are obtained by the following formula:

[0037] aWeight = aNet(Fea),

[0038] F fusion = Fea · aWeight,

[0039] where aWeight represents the attention weight of the feature after concatenation of the existing data signal frames, aNet represents the attention network, and F fusion represents the fused feature of the existing data signal frames.

[0040] Preferably, the quench detection is performed on the fused features by the following formula:

[0041] Y = deepnet(F fusion ),

[0042] where Y represents the detection result of the deep network for the existing data signal frames, and deepnet represents the deep neural network.

[0043] Preferably, the cross-entropy loss function is:

[0044]

[0045] where L represents the cross-entropy loss function, N is the number of sample points, y i represents the magnet state of the i-th existing data signal frame, represents the predicted state of the i-th existing data signal frame.

[0046] Through the above technical scheme, the time domain, frequency domain and convolutional neural network feature expressions of multiple sensors of the superconducting magnet can be extracted and feature fusion can be performed to reduce the high false detection rate and missed detection rate caused by parameter fluctuations when the direct detection method relies on threshold judgment. At the same time, feature fusion based on the attention mechanism can extract attention weights, highlight important information, and reduce interference from irrelevant information, which is conducive to improving the detection accuracy of superconducting magnet quench detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The included drawings are used to provide a further understanding of the embodiments of the present invention, which constitute a part of the specification, are used to illustrate the embodiments of the present invention, and together with the text description, explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 A flowchart of a magnet quench detection method based on attention mechanism feature fusion according to an embodiment of the present invention is shown;

[0049] Figure 2 A schematic diagram showing accuracy curves of training and verification processes according to an embodiment of the present invention is shown;

[0050] Figure 3 FIG. 4 shows a confusion matrix of quench detection results according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] It should be noted that, in the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0053] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0054] Figure 1 A flowchart of a magnet quench detection method based on attention mechanism feature fusion according to an embodiment of the present invention is shown.

[0055] like Figure 1 As shown, an embodiment of the present invention provides a magnet quench detection method based on attention mechanism feature fusion, wherein the method includes:

[0056] Preprocessing the existing time series data and the time series data to be tested (vibration and voltage time series data, original time series data acquired by sensors) of the superconducting magnet to obtain an existing data signal frame and a data signal frame to be tested;

[0057] Extracting time domain features and frequency domain features from existing data signal frames and data signal frames to be tested;

[0058] Using convolutional neural network to extract convolutional features of existing data signal frames and data signal frames to be tested;

[0059] For existing data signal frames, the extracted time domain features and frequency domain features are spliced ​​with the convolution features, the attention weights of the spliced ​​features are extracted using the attention network, and the fusion features are obtained using the extracted attention weights;

[0060] Use the deep neural network as a classifier to perform quench detection on the fused features to obtain the detection result (classification result);

[0061] The training data set is obtained by using the detection results and the cross entropy loss function. The training data set includes a training set and a validation set.

[0062] For example, the operating state of the magnet (ie, the detection result, 0 indicates no quench, 1 indicates quench) may be represented by a set of labels.

[0063] The detection model is trained according to the training set. The detection model includes a convolutional neural network, an attention network, and a deep neural network. The trained detection model is obtained. The trained detection model is verified according to the verification set. The trained detection model with the highest accuracy score of the verification set is determined as the optimal model (network iterative training).

[0064] That is, the detection model is first trained on the training set, and then the performance of the model after each round of training is evaluated on the validation set. After the training process is completed, the model with the highest accuracy score on the validation set is selected as the final model.

[0065] According to the optimal model and the time domain characteristics and frequency domain characteristics of the data signal frame to be tested, convolution feature extraction, fusion feature acquisition and quench detection are performed on the data signal frame to be tested in sequence to obtain the detection result of the data signal frame to be tested.

[0066] That is, the final model is used to perform convolutional neural network feature extraction, attention mechanism feature fusion and deep learning classification network to obtain the test results of the samples to be tested, thereby realizing the quench detection of superconducting magnets.

[0067] Through the above technical scheme, the time domain, frequency domain and convolutional neural network feature expressions of multiple sensors of the superconducting magnet can be extracted and feature fusion can be performed to reduce the high false detection rate and missed detection rate caused by parameter fluctuations when the direct detection method relies on threshold judgment. At the same time, feature fusion based on the attention mechanism can extract attention weights, highlight important information, and reduce interference from irrelevant information, which is conducive to improving the detection accuracy of superconducting magnet quench detection.

[0068] In the present invention, the preprocessing and time-frequency domain feature extraction of the existing time series data and the time series data to be tested (the existing samples and the samples to be tested) are the same (ie, the method is shared), and the existing time series data is used as an example for explanation below.

[0069] According to an embodiment of the present invention, preprocessing existing time series data and time series data to be tested of a superconducting magnet includes:

[0070] The existing time series data and the time series data to be measured of the superconducting magnet are framed and windowed.

[0071] That is, preprocessing may include framing and windowing. Framing is used to divide the original time series data into frame segments suitable for processing to ensure the time accuracy of the detection results. Frame overlap ensures the continuity between signal frames. Windowing can alleviate the frequency leakage problem generated by the signal during the framing process. In the present invention, 0.1s is used as the framing time length, and the frame overlap is half of the frame length, that is, 0.05s (overlapping with the adjacent two framing windows by 0.05s respectively). The window function can choose to use the Hanning window, and the partial value of the window function outside 0.1s is 0.

[0072] According to an embodiment of the present invention, the existing time series data of the superconducting magnet is framed and windowed by the following formula:

[0073] x i =w i ·X,

[0074] Among them, X represents the existing time series data, w i represents the i-th Hanning framing window, · represents the vector dot multiplication operator, x i Indicates that there is a data signal frame.

[0075] According to an embodiment of the present invention, the time domain features include kurtosis features and root mean square features. The kurtosis features and root mean square features extracted from the existing data signal frame are respectively:

[0076]

[0077]

[0078] Among them, μ k and σ k Indicates that there is data signal frame x i The corresponding mean and standard deviation, E represents the expectation operator, Kurt i Indicates that there is data signal frame x i The kurtosis eigenvalue of T represents the existing data signal frame x i Length, RMS i Indicates that there is data signal frame x i is the RMS eigenvalue of , and t represents time.

[0079] According to an embodiment of the present invention, the frequency domain features include a center frequency and a power spectrum. The center frequency and power spectrum extracted from the existing data signal frame are:

[0080]

[0081] P(f i )=|X(k) i | 2 ,

[0082] Among them, f i represents frequency, P(f i ) represents the frequency f i Existing data signal frame x at i The power spectrum, centfreq i Indicates that there is data signal frame x i The center frequency of, k represents the frequency index range, X(k) i Indicates that there is data signal frame x iFrequency-domain signal within the frequency index range k.

[0083] In the present invention, the frequency-domain signal X(k) i can be obtained through Fourier transform (i.e., transferring the existing signal frame to the frequency domain):

[0084]

[0085] where N represents the number of sample points, and x(n) i represents the signal frame of the nth sample.

[0086] According to an embodiment of the present invention, the convolution feature extraction of the existing data signal frame is performed through the following formula:

[0087] F ci = Pool 2 (C 3×3 (Pool 1 (C 5×5 (X i ))))

[0088] where C m×m represents the convolutional layer with a kernel size of m, Pool represents the pooling layer (the first pooling layer and the second pooling layer), X i represents the frequency-domain signal of the existing data signal frame x i , and F ci is the convolution feature of the existing data signal frame x i .

[0089] That is, feature extraction can be achieved by using two convolutional-pooling layers, and m takes values 3 and 5 respectively in the above formula.

[0090] According to an embodiment of the present invention, the extracted time-domain features and frequency-domain features are concatenated with the convolution features through the following formula:

[0091] Fea = [Ftf i , F ci ,

[0092] where Ftf i = [Kurt i , RMS i , centfreq i , P(f i )],

[0093] where Fea is the concatenated feature of the existing data signal frame.

[0094] According to an embodiment of the present invention, the attention weights of the concatenated features are extracted and the fused feature is obtained through the following formula:

[0095] aWeight=aNet(Fea),

[0096] F fusion =Fea·aWeight,

[0097] Among them, aWeight represents the attention weight of the features after splicing the existing data signal frame, aNet represents the attention network, and F fusion Indicates the fusion features of existing data signal frames.

[0098] In the present invention, the structure of the attention network aNet is:

[0099] data out =Sigmoid(FC(GlobalPooling(data in ))),

[0100] data in 、data out They represent the input and output of the attention network aNet respectively, GlobalPooling represents the global pooling layer, FC represents the fully connected layer, and Sigmoid represents the sigmoid activation function.

[0101] According to an embodiment of the present invention, quench detection is performed on the fused features by the following formula:

[0102] Y=deepnet(F fusion ),

[0103] Among them, Y represents the detection result of the deep network for the existing data signal frame, and deepnet represents the deep neural network.

[0104] According to one embodiment of the present invention, the cross entropy loss function is:

[0105]

[0106] Among them, L represents the cross entropy loss function, N is the number of sample points, and y i represents the magnet state of the i-th existing data signal frame, Represents the predicted state of the i-th existing data signal frame.

[0107] Taking the vibration and voltage data of a superconducting magnet test platform as an example, the data is first divided into a training set and a validation set. Figure 1 The model is trained using the steps on the left side of the figure. The validation set data is then used to evaluate the model effect of this round of training. The above steps are repeated until the model converges. The accuracy curve of the training and validation process is shown in the figure below. Figure 2After determining the optimal model, the optimal model is used to perform quench detection on the test data, and the confusion matrix of the quench detection result is as follows: Figure 3 The results show that the magnet quench detection method based on attention mechanism feature fusion proposed in the present invention can effectively distinguish the normal and quench conditions of superconducting magnets. In the data obtained from the actual test platform, the quench detection accuracy is higher than 96%, which can effectively detect the working state of superconducting magnets and meet the demand for high-speed detection of superconducting magnets in actual situations.

[0108] It can be seen from the above embodiments that the method described in the present invention performs time domain and frequency domain feature extraction and deep learning-based convolutional neural network feature extraction for the original signals collected by multiple vibration sensors and voltage sensors of the superconducting magnet, and uses the attention mechanism to realize feature fusion, highlight valuable information, and finally use the deep neural network for classification. Using the detection method disclosed in the present invention, it is possible to ensure accurate quench detection when the amount of data collected by multiple sensors is large. Compared with the traditional direct detection method, it effectively limits the misjudgment and missed judgment problems caused by noise and electromagnetic interference. At the same time, compared with the sequence analysis method, the weight of important information is highlighted through feature extraction and feature fusion, and the interference of irrelevant information is reduced, which meets the actual high-precision requirements for superconducting magnet quench detection.

[0109] In the description of the present invention, it is necessary to understand that the directions or positional relationships indicated by directional words such as "front, back, up, down, left, right", "lateral, vertical, perpendicular, horizontal" and "top, bottom" are usually based on the directions or positional relationships shown in the drawings. They are only for the convenience of describing the present invention and simplifying the description. Unless otherwise specified, these directional words do not indicate or imply that the devices or elements referred to must have a specific direction or be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the scope of protection of the present invention. The directional words "inside and outside" refer to the inside and outside relative to the contours of each component itself.

[0110] For ease of description, spatial relative terms, such as "above", "over", "on the upper surface", "upper", etc., may be used herein to describe the spatial positional relationship of one device or feature to other devices or features as shown in the figures. It should be understood that the spatial relative terms are intended to encompass different orientations in use or operation in addition to the orientation depicted in the figures. For example, if the device in the figures is inverted, the device described as "above" or "over" other devices or structures will then be positioned "below" or "under" the other devices or structures. Thus, the exemplary term "above" can include both the orientations of "above" and "below". The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the corresponding explanations for the spatial relative descriptions used herein will be made accordingly.

[0111] In addition, it should be noted that the use of terms such as "first", "second", etc. to limit components is only for the convenience of differentiating the corresponding components. Without additional statements, the above terms have no special meanings, and thus should not be construed as limiting the protection scope of the present invention.

[0112] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A magnet quench detection method based on attention mechanism feature fusion, It is characterized in that The method includes: Preprocessing the existing time series data and the time series data to be tested of the superconducting magnet to obtain an existing data signal frame and a data signal frame to be tested; Extracting time domain features and frequency domain features from existing data signal frames and data signal frames to be tested; Using convolutional neural network to extract convolutional features of existing data signal frames and data signal frames to be tested; For existing data signal frames, the extracted time domain features and frequency domain features are spliced ​​with the convolution features, the attention weights of the spliced ​​features are extracted using the attention network, and the fusion features are obtained using the extracted attention weights; Use deep neural network as classifier to perform quench detection on the fused features to obtain the detection results; The training data set is obtained by using the detection results and the cross entropy loss function. The training data set includes a training set and a validation set. The detection model is trained according to the training set, and the detection model includes a convolutional neural network, an attention network, and a deep neural network to obtain a trained detection model. The trained detection model is verified according to the verification set, and the trained detection model with the highest accuracy score of the verification set is determined as the optimal model; According to the optimal model and the time domain characteristics and frequency domain characteristics of the data signal frame to be tested, convolution feature extraction, fusion feature acquisition and quench detection are performed on the data signal frame to be tested in sequence to obtain the detection result of the data signal frame to be tested.

2. The method according to claim 1, It is characterized in that Preprocessing the existing time series data and the time series data to be measured of the superconducting magnet includes: The existing time series data and the time series data to be measured of the superconducting magnet are framed and windowed.

3. The method according to claim 2, It is characterized in that The existing time series data of the superconducting magnet is framed and windowed by the following formula: x i =w i ·X, Among them, X represents the existing time series data, w i represents the i-th Hanning frame window, · represents the vector dot multiplication operator, x i Indicates that there is a data signal frame.

4. The method according to claim 3, It is characterized in that The time domain features include kurtosis features and root mean square features. The kurtosis features and root mean square features extracted from the existing data signal frame are: Among them, μ k and σ k Indicates that there is data signal frame x i The corresponding mean and standard deviation, E represents the expectation operator, Kurt i Indicates that there is data signal frame x i The kurtosis eigenvalue of T represents the existing data signal frame x i Length, RMS i Indicates that there is data signal frame x i is the RMS eigenvalue of , and t represents time.

5. The method according to claim 4, It is characterized in that The frequency domain features include center frequency and power spectrum. The center frequency and power spectrum extracted from the existing data signal frame are: P(f i )=|X(k) i | 2 , Among them, f i represents frequency, P(f i ) represents the frequency f i Existing data signal frame x at i The power spectrum, centfreq i Indicates that there is data signal frame x i The center frequency of, k represents the frequency index range, X(k) i Indicates that there is data signal frame x i The frequency domain signal in the frequency index range k.

6. The method according to claim 5, It is characterized in that The convolution feature extraction of the existing data signal frame is performed by the following formula: F ci =Pool 2 (C 3×3 (Pool 1 (C 5×5 (X i )))), Among them, C m×m represents a convolutional layer with a kernel size of m, Pool represents a pooling layer, and Xi represents an existing data signal frame x i The frequency domain signal, F ci For the existing data signal frame x i Convolutional features.

7. The method according to claim 6, It is characterized in that The extracted time domain features and frequency domain features are concatenated with the convolution features by the following formula: Fea=[Ftf i ,F ci ], Ftf i =[Kurt i ,RMS i ,centfreq i ,P(f i )], Among them, Fea is the feature after splicing of existing data signal frames.

8. The method according to claim 7, It is characterized in that The attention weights of the concatenated features are extracted and the fused features are obtained by the following formula: aWeight=aNet(Fea), F fusion =Fea·aWeight, Among them, aWeight represents the attention weight of the features after splicing the existing data signal frame, aNet represents the attention network, and F fusion Indicates the fusion features of existing data signal frames.

9. The method according to claim 8, It is characterized in that The quench detection is performed on the fused features using the following formula: Y=deepnet(F fusion ), Among them, Y represents the detection result of the deep network for the existing data signal frame, and deepnet represents the deep neural network.

10. The method according to claim 9, It is characterized in that The cross entropy loss function is: Among them, L represents the cross entropy loss function, N is the number of sample points, and y i represents the magnet state of the i-th existing data signal frame, Represents the predicted state of the i-th existing data signal frame.

Images