Transformer fault real-time diagnosis method

Abstract

According to the transformer fault real-time diagnosis method provided by the invention, a delay estimation method based on wavelet denoising and secondary cross-correlation is used to eliminate the time difference between each path of sound signals, so that the time synchronization of each path of sound signals is realized, the accuracy of voiceprint feature extraction is improved, and meanwhile, the environmental noise can be effectively inhibited. According to the method, each path of sound signal is independently processed in the stages of pre-emphasis, time synchronization, MFCC feature extraction and quantum coding, specific fault information of each path of sound signal is reserved, and the flexibility of quantum feature enhancement is improved. According to the method, on the basis of MFCC feature extraction, a quantum feature coding technology is further used, the MFCC feature of each channel is mapped to a quantum state and parameterized entanglement transformation is performed, so that the expression ability of fault features is effectively enhanced, the noise sensitivity of the MFCC features is reduced, the nonlinear correlation between the features in the channels and the features between the channels is fully mined, and the fault detection accuracy is improved. The accuracy of a fault diagnosis result can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer monitoring, and particularly to a method for real-time diagnosis of transformer faults. Background Art

[0002] When a transformer is in operation, noise is generated due to the vibration of the iron core and windings. The transformer voiceprint signal contains a large amount of information characteristics that can reflect the operating state of the equipment. Some existing transformer voiceprint monitoring technologies have defects. Using multiple microphone sensors in the acoustic signal acquisition link can obtain more comprehensive voiceprint characteristics, capture the correlation between acoustic signals at different positions, and enhance the ability to resist environmental noise interference. However, factors such as the different distances between the sound source and each microphone sensor, system hardware delay, and insufficient clock synchronization accuracy will all cause the acoustic signals of each channel to be asynchronous in time. The existing transformer voiceprint monitoring technology design ignores the time difference between the acoustic signals of each channel, resulting in inaccurate voiceprint feature extraction and weakened environmental noise suppression ability.

[0003] In addition, the existing transformer voiceprint monitoring technology based on MFCC (Mel Frequency Cepstral Coefficients) feature extraction also has problems such as high noise sensitivity and lack of non-linear correlation between features, resulting in inaccurate fault diagnosis results and a high false alarm rate. Summary of the Invention

[0004] In view of this, the present invention provides a method for real-time diagnosis of transformer faults to improve the accuracy of fault diagnosis results.

[0005] A method for real-time diagnosis of transformer faults includes: Step S1, performing real-time multi-channel voiceprint acquisition on the transformer using a plurality of microphone sensors, and performing pre-emphasis processing on each channel of sound signal collected by the microphone sensors to obtain the pre-emphasized sound signal; Step S2, using the first channel of sound signal in the pre-emphasized sound signal as a reference signal, and respectively estimating the time delay estimation values between each channel of sound signal except the first channel of sound signal and the first channel of sound signal using a delay estimation method based on wavelet denoising and quadratic cross-correlation, and obtaining the synchronized sound signals of each channel based on the time delay estimation values; Step S3, respectively performing MFCC feature extraction on the synchronized sound signals of each channel to obtain an MFCC feature matrix, and each row of the MFCC feature matrix stores the MFCC features of each dimension of one channel of sound signal; Step S4: Independently normalize the MFCC features of each dimension of each sound signal in the MFCC feature matrix, initialize the MFCC features of each dimension of each normalized sound signal, map them to quantum states, allocate a number of qubits for the MFCC features of each dimension of each sound signal, construct a cross-layer quantum entanglement circuit, and then input each qubit into the cross-layer quantum entanglement circuit to obtain the final quantum state. Calculate the expectation value of the Pauli-Z operator of each qubit in the final quantum state, and thus obtain the features after quantum feature enhancement. Step S5: Input the features after quantum feature enhancement into the CNN-LSTM hybrid network and train it. Use the trained CNN-LSTM hybrid network to diagnose the faults of the transformer.

[0006] According to the transformer fault real-time diagnosis method provided by the present invention, the following beneficial effects are achieved: (1) The present invention uses the delay estimation method based on wavelet denoising and quadratic cross-correlation to eliminate the time difference between each path of sound signals, thereby realizing the time synchronization of each path of sound signals, significantly improving the accuracy of voiceprint feature extraction, and effectively suppressing environmental noise at the same time.

[0007] (2) The present invention independently processes each path of sound signals in the pre-emphasis, time synchronization, MFCC feature extraction, and quantum coding stages, retains the specific fault information of each path of sound signals, improves the flexibility of quantum feature enhancement, and can provide better input features for the CNN-LSTM hybrid network.

[0008] (3) On the basis of MFCC feature extraction, the present invention further uses quantum feature coding technology. By mapping the MFCC features of each channel to quantum states and performing parametric entanglement transformation, the expression ability of fault features is effectively enhanced, the noise sensitivity of MFCC features is reduced, the non-linear correlation between intra-channel features and inter-channel features is fully explored, and the problems of high noise sensitivity and lack of non-linear correlation between features existing in traditional methods are solved, which can improve the accuracy of fault diagnosis results and reduce the diagnostic false alarm rate. Description of the Drawings

[0009] Figure 1 It is a flowchart of the transformer fault real-time diagnosis method provided by the embodiment of the present invention. Detailed Embodiments

[0010] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The embodiments described below with reference to the drawings are exemplary and are intended to explain the embodiments of the present invention, and should not be construed as a limitation of the present invention.

[0011] Please refer to Figure 1 , an embodiment of the present invention provides a method for real-time diagnosis of transformer faults, including steps S1 to S5: Step S1, use a plurality of microphone sensors to perform real-time multi-channel voiceprint acquisition on the transformer, and perform pre-emphasis processing on each path of voice signal collected by the microphone sensors to obtain the pre-emphasized voice signal.

[0012] In this embodiment, 8 microphone sensors are installed around the transformer, so the total number of channels is 8. The total number of channels is the total number of signal paths, and the voice signals in the working state of the transformer are collected in real-time and multiplexed. During the propagation of the voice signal, the high-frequency components attenuate faster than the low-frequency components. Pre-emphasis can enhance the high-frequency components, making the consonant details in the voice signal more obvious. Transmit each path of signal to a high-pass filter for pre-emphasis processing. The expressions of each path of voice signal after pre-emphasis are as follows: ; ; Among them, represents the first path of voice signal after pre-emphasis processing, represents the th path of voice signal after pre-emphasis processing, , represents the total number of channels, is the first path of source signal, is the first path of noise signal, is the th path of source signal without time-domain correction, is the th path of noise signal, is time, the th path of voice signal and the time delay between the first path of voice signal. and are theoretically uncorrelated Gaussian white noises.

[0013] Step S2, use the first path of voice signal in the pre-emphasized voice signal as a reference signal, and use the delay estimation method based on wavelet denoising and quadratic cross-correlation to estimate the time delay estimation values between each path of voice signal except the first path of voice signal and the first path of voice signal respectively, and obtain the synchronized voice signals of each path based on the time delay estimation values.

[0014] Among them, step S2 specifically includes: Step S2.1, solve the autocorrelation function of the first path of voice signal, the first path of voice signal and the The cross-correlation function of the first path of sound signal is expressed as follows: ; where, represents the autocorrelation function of the first path of sound signal, is 's autocorrelation function, is and 's cross-correlation function, is and 's cross-correlation function, is 's autocorrelation function, is the time delay; ; where, represents the cross-correlation function of the first path of sound signal and the th path of sound signal, is and 's cross-correlation function, is and 's cross-correlation function, is and 's cross-correlation function, is and 's cross-correlation function.

[0015] Step S2.2, use wavelet transform to decompose and denoise the autocorrelation function of the first path of sound signal and the cross-correlation function of the first path of sound signal and the th path of sound signal respectively, to obtain the decomposed and denoised autocorrelation function and cross-correlation function; Since the noise terms and are not actually ideal Gaussian white noise, the noise correlation terms , and are not zero. Using the direct correlation method will result in a large error in the time delay estimation value at low signal-to-noise ratio. Therefore, use wavelet transform to decompose and denoise the autocorrelation function and cross-correlation function respectively, and the expressions are as follows: ; ; ; ; where, is the autocorrelation function of the first audio signal after wavelet denoising, is the time shift factor, is the approximation coefficient of, is the scaling function of the low frequency band, is the scale factor, is the scaling function, is the detail coefficient of, is the wavelet function of the high frequency band, is the wavelet function, is the cross-correlation function between the first audio signal after wavelet denoising and the th audio signal, is the approximation coefficient of, is the detail coefficient of.

[0016] By setting appropriate thresholds, set to zero the wavelet coefficients in , , , that have a strong correlation with noise, so as to achieve the purpose of denoising.

[0017] Step S2.3, calculate the quadratic cross-correlation function of the autocorrelation function and the cross-correlation function after decomposition and denoising to obtain the quadratic cross-correlation function between the first audio signal and the th audio signal; For and calculate the quadratic cross-correlation function to further suppress the residual noise and obtain the quadratic cross-correlation function between the first signal and the th signal: ; Among them, is the quadratic cross-correlation function between the first audio signal and the th audio signal, is the expectation operation; is the time-shifted quadratic cross-correlation function between the first audio signal and the th audio signal, which is the result obtained after delaying by on the time axis. By introducing explore the correlation at different delays to improve the accuracy of time delay estimation; Step S2.4, perform peak search on the quadratic cross-correlation function between the first audio signal and the th audio signal, and for the first audio signal and the The time delay corresponding to the peak of the second-order cross-correlation function of the first-channel sound signal is used as the time delay estimation value of the first-channel sound signal and the -channel sound signal; Obtain and perform peak search on it. Take the time delay corresponding to the peak value as the time delay estimation value of the first-channel signal and the -channel signal: ; wherein, represents taking the maximum value.

[0018] Step S2.5, substitute the time delay estimation values of the first-channel sound signal and the -channel sound signal into the expressions of the pre-emphasized sound signals of each channel, and perform time-domain correction on the sound signals of each channel except the first-channel sound signal, so that the sound signals of each channel except the first-channel sound signal are aligned with the first-channel sound signal in the time domain, realizing the time synchronization of the sound signals of each channel, and thus obtaining the synchronized sound signals of each channel. The expressions are as follows: ; wherein, is the -channel sound signal after time-domain correction, is the -channel sound signal after time-domain correction, is the -channel source signal after time-domain correction.

[0019] Step S3, perform MFCC feature extraction on the synchronized sound signals of each channel respectively to obtain an MFCC feature matrix. Each row of the MFCC feature matrix stores the MFCC features of each dimension of a sound signal of one channel.

[0020] Step S3 specifically includes: Step S3.1, send the obtained after synchronization into a data acquisition card for analog-to-digital conversion. In this embodiment, the sampling rate ≥ 16 kHz and the resolution is 16 bit; Step S3.2, perform frame segmentation on the multiplexed sound signals after analog-to-digital conversion; Step S3.3, perform windowing on the multiplexed sound signals after frame segmentation to obtain windowed frame signals , is the frame index, is the sampling point index within the time domain frame. In this embodiment, the frame length is set to 25 ms and the frame shift is 15 ms. The window function uses a Hamming window; Step S3.4, for the windowed frame signal perform a fast Fourier transform to obtain the frequency spectrum , where is the index of the frequency component in the frequency domain; Step S3.5, take the square of the frequency spectrum as the energy spectrum and input it into the Mel filter bank to obtain the logarithmic energy spectrum output by the Mel filter bank of the -th sound signal; Step S3.6, perform a discrete cosine transform on the logarithmic energy spectrum output by the Mel filter bank of the -th sound signal to obtain the MFCC features of the -th sound signal from dimension 0 to dimension , and take the logarithm of the time-domain energy of the frame signal as the -th dimension MFCC feature of the -th sound signal, thus constituting the MFCC feature vector of the -th sound signal with a dimension of , and further constituting the MFCC feature matrix , where is a set of real numbers, is the total number of dimensions of the MFCC features. In this embodiment, takes 13; in the feature vector , the -th dimension MFCC feature of the -th sound signal is . , , is a set of real numbers, is the total number of dimensions of the MFCC features. In this embodiment, takes 13; the -th dimension MFCC feature of the -th sound signal in the feature vector is .

[0021] Step S4, independently normalize the MFCC features of each dimension of each sound signal in the MFCC feature matrix, initialize the MFCC features of each dimension of each normalized sound signal, map them to quantum states, allocate a number of qubits to the MFCC features of each dimension of each sound signal, construct a cross-layer quantum entanglement circuit, and then input each qubit into the cross-layer quantum entanglement circuit to obtain the final quantum state, calculate the expectation value of the Pauli-Z operator of each qubit in the final quantum state, and further obtain the features after quantum feature enhancement.

[0022] The quantum-inspired feature encoding technology effectively enhances the expression ability of fault features by mapping the MFCC features of each channel to quantum states and performing parametric entanglement transformation, and solves the problems of noise sensitivity and lack of non-linear correlation between features existing in traditional methods such as MFCC.

[0023] Step S4 specifically includes: ​​​Step S4.1: For each path of voice signal in the MFCC feature matrix, perform percentile rank normalization on each of the MFCC features independently, and then take the inverse function of the standard normal distribution for the result to convert the uniform distribution into a Gaussian distribution, so as to eliminate the dimensional differences of different MFCC dimensions: ; Among them, is the -th dimensional MFCC feature of the -th path of voice signal after normalization, is the inverse function of the standard normal distribution, is the sorting position of in ascending order of numerical values among all training samples, and is the total number of training samples. In this embodiment, the total number of training samples is expanded to 10,000 through data augmentation to cover various fault types and normal working states.

[0024] Step S4.2: Apply a rotation gate to the normalized feature to initialize the quantum state, map to the coherent superposition state of qubits, and allocate qubits for each dimensional MFCC feature to obtain the initialized qubits: ; Among them, is the quantum state of the -th qubit of the -th dimensional MFCC feature of the -th path of voice signal at the 0-th layer of the quantum circuit. The quantum state of the qubit at the 0-th layer is the initialized qubit; is the total number of qubits allocated for each dimensional MFCC feature. In this embodiment, = 5; is the rotation gate around the Y axis, represents mapping from to through a fixed scaling factor so as to cover the complete superposition state of the qubit; is the rotation gate rotating around the Z axis, introducing the phase degree of freedom, so that the quantum state can represent more complex feature correlations in the complex space; is the -th dimensional MFCC feature of the -th path of voice signal, and The phase parameter of the 0th layer of qubits. The phase parameter of the 0th layer is the initialization phase parameter, which is used to dynamically adjust the phase characteristics of the quantum state. It is initialized according to the following rules to avoid all quantum states collapsing to the same direction: ; Step S4.3, construct an interlayer entangled quantum circuit with a total number of layers of ; Step S4.4, input the initialized qubits into the interlayer entangled quantum circuit to obtain the final quantum state: ; where, is the final quantum state, is the product symbol; is the initial quantum state of the quantum circuit, which is the initial state of all qubits before any operation after initialization, , is the tensor product symbol; is the entangled controlled-NOT gate within the th layer, is the interlayer entangled controlled-NOT gate; is the th -dimensional MFCC feature of the th qubit of the th layer, which is used to dynamically adjust the amplitude of the quantum state (through backpropagation training); is the th -dimensional MFCC feature of the th qubit of the th layer, Step S4.5, calculate the expectation value of the Pauli-Z operator for each qubit, and generate a total theoretical dimension of the feature of , which is compressed to dimensions through principal component analysis to adapt to the input of the downstream network. The feature after quantum feature enhancement is: ; ; where, is the principal component analysis; is the union symbol, which concatenates the expectation values of each channel, dimension, and qubit into a high-dimensional vector; is the Pauli-Z measurement operator of the th -dimensional MFCC feature of the th qubit of the For the final quantum state The expected value under the action of the operator ; is the 2×2 identity matrix.

[0025] Among them, in this embodiment, when constructing a cross-layer entangled quantum circuit with a total number of layers of The total number of layers of the quantum circuit is determined by 5-fold cross-validation, and the value ranges from 2 to 5. In this example, it is preferably 3 layers. The construction process of the cross-layer entangled quantum circuit includes three steps: intra-layer entanglement, cross-layer entanglement, and rotation gate operation. Specifically, step S4.3 specifically includes: Step S4.3.1, intra-layer entanglement, applying a chained controlled-NOT gate between the qubit of each-dimensional MFCC feature within the same layer: ; where CNOT represents the controlled-NOT gate, is the position of the th qubit of the th-dimensional MFCC feature of the th voice signal in the th layer, is the position of the th qubit of the th-dimensional MFCC feature of the th voice signal in the th layer; Step S4.3.2, since the first qubit of each layer transmits the main information of the feature, cross-layer entanglement is applied only between the first qubits of the th layer and the th layer, reducing the amount of calculation while retaining cross-layer information transmission: ; where, is the position of the first qubit of the th qubit of the th-dimensional MFCC feature of the th voice signal in the th layer, is the position of the first qubit of the th qubit of the th-dimensional MFCC feature of the Step S4.3.3, applying a trainable rotation gate to each layer: ; where is the th qubit of the The th qubit of the -dimensional MFCC feature at the th -th -dimensional MFCC feature of the th qubit at the th layer.

[0026] Step S5: Input the features enhanced by quantum feature enhancement into the CNN-LSTM hybrid network for training, and use the trained CNN-LSTM hybrid network to diagnose the faults of the transformer.

[0027] In this embodiment, the CNN-LSTM hybrid network includes: an input layer, a first convolutional layer (64 3×3 convolutional kernels, stride = 1), a max pooling layer (pooling window = 2×2, stride = 2), a second convolutional layer (128 3×3 convolutional kernels, stride = 1), a time series reshaping layer (converting the convolutional output into a time step series), a bidirectional LSTM layer, an attention layer, a fully connected layer (128 neurons, activation function is ReLU), and a dual-task output layer.

[0028] During the model training process, the Adam optimizer is adopted, and hyperparameters such as the learning rate, batch size, loss weight, number of LSTM cells, and Dropout ratio of the fully connected layer are optimized through grid search and cross-validation.

[0029] In summary, according to the transformer fault real-time diagnosis method of the above embodiment, the following beneficial effects are achieved: (1) The present invention uses the delay estimation method based on wavelet denoising and quadratic cross-correlation to eliminate the time difference between each path of sound signals, thereby realizing the time synchronization of each path of sound signals, significantly improving the accuracy of voiceprint feature extraction, and effectively suppressing environmental noise at the same time.

[0030] (2) In the pre-emphasis, time synchronization, MFCC feature extraction, and quantum coding stages, the present invention independently processes each path of sound signals, retains the specific fault information of each path of sound signals, improves the flexibility of quantum feature enhancement, and can provide better input features for the CNN-LSTM hybrid network.

[0031] (3)Based on the MFCC feature extraction, the present invention further uses the quantum feature encoding technology. By mapping the MFCC features of each channel to the quantum state and performing parametric entanglement transformation, the expression ability of fault features is effectively enhanced, the noise sensitivity of MFCC features is reduced, the non-linear correlation of features within and between channels is fully explored, the problems of high noise sensitivity and lack of non-linear correlation between features existing in traditional methods are solved, the accuracy of fault diagnosis results can be improved, and the diagnostic false alarm rate can be reduced.

[0032] The above embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A real-time transformer fault diagnosis method, characterized in that: include: Step S1, using a plurality of microphone sensors to perform real-time multi-channel voiceprint collection on the transformer, and performing pre-emphasis processing on each sound signal collected by the microphone sensor to obtain a pre-emphasized sound signal; Step S2, using the first sound signal in the sound signal after pre-emphasis processing as a reference signal, using a delay estimation method based on wavelet denoising and quadratic cross-correlation to respectively estimate the time delay estimation values ​​between each sound signal except the first sound signal and the first sound signal, and obtaining the synchronized sound signals based on the time delay estimation values; Step S3, performing MFCC feature extraction on each of the synchronized sound signals to obtain an MFCC feature matrix, wherein each row of the MFCC feature matrix stores the MFCC features of each dimension of each sound signal; Step S4, respectively normalize the MFCC features of each dimension of each sound signal in the MFCC feature matrix independently, initialize the MFCC features of each dimension of each sound signal after normalization, and map them to the quantum state, allocate a number of quantum bits to the MFCC features of each dimension of each sound signal, construct a cross-layer quantum entanglement circuit, and then input each quantum bit into the cross-layer quantum entanglement circuit to obtain the final quantum state, calculate the Pauli-Z operator expectation value of each quantum bit in the final quantum state, and then obtain the features after quantum feature enhancement; Step S5, inputting the features enhanced by quantum features into the CNN-LSTM hybrid network and training it, and using the trained CNN-LSTM hybrid network to perform fault diagnosis on the transformer.

2. The real-time transformer fault diagnosis method according to claim 1, characterized in that: Step S1 satisfies the following formula: ； ； in, Represents the first sound signal after pre-emphasis processing. Indicates the first Road sound signal, , Indicates the total number of channels, is the first source signal, is the first noise signal, For the The source signal is not time-domain corrected. For the Road noise signal, For time, No. The time delay between the first sound signal and the second sound signal.

3. The transformer fault real-time diagnosis method according to claim 2, characterized in that: Step S2 specifically includes: Step S2.1, solving the autocorrelation function of the first sound signal, the first sound signal and the second sound signal other than the first sound signal Cross-correlation function of the road sound signal; Step S2.2, using wavelet transform to respectively analyze the autocorrelation function of the first sound signal, the first sound signal and the Decomposing and denoising the cross-correlation function of the road sound signal to obtain the autocorrelation function and cross-correlation function after decomposition and denoising; Step S2.3, calculate the secondary cross-correlation function of the autocorrelation function and the cross-correlation function after decomposition and denoising, and obtain the first sound signal and the second sound signal. The quadratic cross-correlation function of the road sound signal; Step S2.4, the first sound signal and the The peak value of the second cross-correlation function of the first sound signal and the The time delay corresponding to the peak value of the secondary cross-correlation function of the first sound signal and the second sound signal is taken as the time delay corresponding to the peak value of the second sound signal. time delay estimate of the road sound signal; Step S2.5, the first sound signal and the The time delay estimation value of each sound signal is substituted into the expression of each sound signal after pre-emphasis, and time domain correction is performed on each sound signal except the first sound signal, so that each sound signal except the first sound signal is aligned with the first sound signal in the time domain, and time synchronization of each sound signal is achieved, thereby obtaining each synchronized sound signal.

4. The transformer fault real-time diagnosis method according to claim 3 is characterized in that: Step S2.2 satisfies the following equation: ； ； ； ； in, is the autocorrelation function of the first sound signal after wavelet denoising, is the time shift factor, for The approximate coefficient of is the scaling function of the low frequency band, is the scale factor, is the scale function, for The detail factor, is the wavelet function of the high frequency band, is the wavelet function, is the first sound signal after wavelet denoising and the The cross-correlation function of the sound signal of the road, for The approximate coefficient of for The detail factor.

5. The transformer fault real-time diagnosis method according to claim 4, characterized in that: Step S2.3 satisfies the following equation: ； in, For the first sound signal and the The quadratic cross-correlation function of the sound signal of the road, is the expectation operation, For the first sound signal and the The time-shifted quadratic cross-correlation function of the sound signal on the road.

6. The transformer fault real-time diagnosis method according to claim 5, characterized in that: Step S2.5 satisfies the following equation: ； in, After time domain correction Road sound signal, After time domain correction Road sound signal, After time domain correction Road source signal.

7. The transformer fault real-time diagnosis method according to claim 6, characterized in that: Step S4 specifically includes: Step S4.1, respectively analyze the characteristics of each sound signal in the MFCC feature matrix Each MFCC feature is independently quantile-rank normalized, and then the inverse function of the standard normal distribution is taken for the result to convert the uniform distribution into a Gaussian distribution. The expression is: ； in, After normalization, The first sound signal dimensional MFCC features, is the inverse function of the standard normal distribution, for The sorting position of all training samples in ascending order of numerical value, is the total number of training samples; Step S4.2, normalized features Apply a revolving door to initialize the quantum state, Mapped to the coherent superposition state of quantum bits, assign quantum bits to each dimensional MFCC feature, and obtain the initialized quantum bits, the expression is: ； in, For the The first sound signal The first dimension of MFCC features The quantum state of a quantum bit at the 0th layer of the quantum circuit is the initialization quantum bit. is a revolving door around the Y axis, Represented by a fixed scaling factor Will from Map to ; It is a revolving door that rotates around the Z axis; For the The first sound signal The first dimension of MFCC features The phase parameter of the 0th layer of qubits, ; Step S4.3, construct the total number of layers Cross-layer entangled quantum circuits; Step S4.4, input the initialized quantum bit into the cross-layer entangled quantum circuit to obtain the final quantum state, which is expressed as: ； in, is the final quantum state, is the multiplication symbol; is the initial quantum state of the quantum circuit, which is the initial state of all quantum bits before any operation after initialization. , is the tensor product symbol; For the Intra-layer entangled controlled NOT gate, It is a cross-layer entangled controlled NOT gate; For the The first sound signal The first dimension of MFCC features The quantum bit Angle parameters of the layer; For the The first sound signal The first dimension of MFCC features The quantum bit The phase parameters of the layer, The total number of qubits allocated to each dimension of MFCC features; Step S4.5, calculate the expected value of the Pauli-Z operator for each qubit, and the total number of theoretical dimensions of the generated features is , compressed to Dimension, characteristics after quantum feature enhancement for: ； ； in, It is principal component analysis; For the union symbol, the expected values ​​of each channel, dimension, and qubit are concatenated into a high-dimensional vector; For the The first sound signal The first dimension of MFCC features The Pauli-Z measurement operator for qubits, The final quantum state In the operator Expected value under action; is a 2×2 identity matrix.

8. The transformer fault real-time diagnosis method according to claim 7, characterized in that: Step S4.3 specifically includes: Step S4.3.1, intra-layer entanglement, for each dimension of MFCC features in the same layer A chain controlled NOT gate is applied between the quantum bits, and the expression is: ； Among them, CNOT represents a controlled NOT gate, For the The first sound signal The first dimension of MFCC features The quantum bit in The location of the layer, For the The first sound signal The first dimension of MFCC features The quantum bit in The location of the layer; Step S4.3.2, cross-layer entanglement, in Layer and The cross-controlled NOT gate is applied between the first quantum bits of the layer, and the expression is: ； in, For the The first sound signal The first quantum bit of the MFCC feature is The location of the layer, For the The first sound signal The first quantum bit of the MFCC feature is The location of the layer; Step S4.3.3, apply a trainable revolving door, the expression is: ； in For the The first sound signal The first dimension of MFCC features The quantum bit in The quantum state of the layer, For the The first sound signal The first dimension of MFCC features The quantum bit in The quantum state of the layer.

Images