Transformer vibration anomaly detection method based on millimeter wave radar

Abstract

The invention relates to the technical field of transformer detection, in particular to a millimeter wave radar-based transformer vibration anomaly detection method, which comprises the following steps of: acquiring an original reflection signal on the surface of a transformer, performing time domain and frequency domain analysis on the original reflection signal, and constructing a vibration feature vector; based on the constructed vibration feature vector, unsupervised learning is carried out on the vibration mode of the normal transformer through a deep self-encoding network, and potential features of the normal vibration mode are automatically learned; converting the new reflected signal into a new vibration feature vector, reconstructing the new vibration feature vector by using a deep self-encoding network, calculating a reconstruction error, and if the reconstruction error exceeds a predetermined threshold, judging that the vibration mode is an abnormal vibration mode; a self-adaptive threshold strategy is adopted, and a preset threshold is automatically adjusted so as to adapt to the change of the operation states of different transformer devices along with time. According to the invention, a self-adaptive threshold strategy is introduced, so that running characteristics of different transformer devices and fluctuation of an external environment can be adapted in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer detection, and in particular to a transformer vibration anomaly detection method based on millimeter wave radar. Background Art

[0002] With the continuous development of intelligent equipment in power systems, transformers, as important equipment in power transmission and distribution systems, have a direct impact on the stability and safety of power supply. Transformers are subject to varying degrees of vibration during long-term operation, and abnormal changes in vibration are often early signals of transformer failure. Therefore, timely and accurate detection of transformer vibration anomalies is of great significance. Traditional transformer vibration anomaly detection methods usually rely on sensor monitoring or analysis based on mechanical structure. These methods have certain limitations and are difficult to fully and accurately reflect the true state of equipment operation.

[0003] At present, research in the field of transformer vibration anomaly detection is gradually developing from traditional mechanical sensor detection to more intelligent technology. Among them, vibration monitoring technology based on millimeter wave radar has become a hot topic in recent years because it can penetrate the external structure of the transformer and monitor vibration signals without contact. Millimeter wave radar can obtain vibration information on the transformer surface in real time by emitting millimeter waves and receiving their reflected signals. However, the reflected signals received by millimeter wave radar contain a lot of noise and interference information, which makes it a challenge to extract accurate vibration features from them.

[0004] Traditional vibration anomaly detection methods usually rely on manually set thresholds or simple statistical methods, such as judgments based on features such as mean and standard deviation. This method not only relies on manual experience, but is also prone to misjudgment or missed judgments when facing complex and dynamic working environments. In particular, when factors such as the load, temperature, and environment of the transformer change, simple threshold settings are often difficult to cope with. Summary of the invention

[0005] The invention provides a transformer vibration anomaly detection method based on millimeter wave radar.

[0006] The transformer vibration abnormality detection method based on millimeter wave radar includes the following steps: S1, uses millimeter wave radar to transmit millimeter waves to the external structure of the transformer, obtains the original reflection signal on the transformer surface, performs time domain and frequency domain analysis on the original reflection signal, and constructs a vibration feature vector; S2, training and feature learning of deep autoencoder network: based on the constructed vibration feature vector, unsupervised learning of normal transformer vibration mode is performed through deep autoencoder network, the deep autoencoder network is composed of encoder and decoder, the encoder compresses the input vibration feature vector into a low-dimensional feature representation, and the decoder restores the original vibration feature vector from the low-dimensional feature, and automatically learns the potential features of the normal vibration mode through training network; S3, detection of abnormal vibration patterns: after training is completed, new reflection signals are collected again, and the new reflection signals are converted into new vibration feature vectors by the method of S1. The new vibration feature vectors are reconstructed using a deep autoencoder network, and the reconstruction error is calculated. If the reconstruction error exceeds a predetermined threshold, it is judged as an abnormal vibration pattern, and the abnormal patterns that are different from the normal vibration patterns are adaptively identified, including changes in frequency, amplitude, and period. S4, adopts an adaptive threshold strategy to automatically adjust the predetermined threshold to adapt to the changes in the operating status of different transformer equipment over time.

[0007] Optionally, performing time domain and frequency domain analysis on the original reflection signal in S1 to construct the vibration feature vector specifically includes: S11, time domain analysis: perform peak detection on the original reflected signal and analyze the amplitude, periodic changes, and instantaneous frequency characteristics of the signal, including: Peak detection is performed through the maximum and minimum values ​​of the signal to calculate the amplitude change of the signal ; Use the signal's zero crossing points and periodic peaks to analyze periodic changes in vibration ; Calculate the instantaneous frequency of a signal , reflecting the frequency change of vibration; S12, frequency domain analysis: Convert the original reflection signal to the frequency domain through Fourier transform, analyze the spectrum characteristics of the signal, and extract the frequency components related to the vibration mode, including: Use fast Fourier transform to perform frequency domain analysis on the reflected signal to obtain the signal spectrum and extract the main frequency components in the signal spectrum ; Analyze the frequency band distribution in the spectrum and calculate the frequency bandwidth of the vibration signal ; S13, combines the time domain features and frequency domain features (amplitude, period, frequency) into a vibration feature vector , the vibration eigenvector It is expressed as: ; in, is the amplitude, It is cyclical. is the main frequency, is the frequency bandwidth, It's time The instantaneous frequency.

[0008] Optionally, the S2 specifically includes: S21, based on the constructed vibration eigenvector , unsupervised learning of normal transformer vibration modes is performed through a deep autoencoder network, the deep autoencoder network comprising an encoder and a decoder, wherein: Encoder part: The encoder receives the input vibration feature vector , which is compressed through a multi-layer neural network to obtain a low-dimensional feature representation , the encoder gradually reduces the dimension of the input data through several hidden layers to obtain a low-dimensional feature representation , this low-dimensional feature representation can capture the key features of normal vibration modes: ; Decoder part: The decoder represents low-dimensional features Restore the original vibration feature vector through the inverse neural network structure , that is, the input feature vector is gradually reconstructed through back propagation, the goal is to minimize the reconstruction error so that the reconstruction result is as close as possible to the input vibration feature vector: ; S22, training process: Through unsupervised learning, a large number of normal transformer vibration feature vectors are used for training, and the parameters of the encoder and decoder are optimized so that the encoder can effectively compress the normal vibration feature vector into a low-dimensional space and accurately reconstruct the original input through the decoder. During the training process, the reconstruction error function is minimized. To optimize the network parameters, the reconstruction error is measured using the mean square error loss function: ;in, Indicates The vibration feature vector of training samples, Indicates The vibration feature vector obtained by reconstructing the training samples through the autoencoder network, L represents the reconstruction error of the network, which represents the difference between the input and reconstructed signals.

[0009] Optionally, the encoder uses a fully connected layer structure to gradually reduce the dimension of the data; Assume that the input vibration eigenvector is a -dimensional vector , the encoder's task is to compress it into a -dimensional low-dimensional feature representation ,in ; The encoder process is implemented through the following fully connected layers: ,in, is the weight matrix of the first layer, with a size of , used to convert input data from -Dimensions are mapped to -dimension, is the bias term, with a size of -dimension, is the activation function, using ReLU or Sigmoid activation function, It is a low-dimensional feature representation.

[0010] Optionally, the decoder is implemented by a reverse fully connected layer to convert the low-dimensional features Gradually map back to the high-dimensional space to restore the original vibration characteristics: ,in, is the weight matrix of the first layer of the decoder, with a size of , used to map low-dimensional features back to high-dimensional space, is the bias term, with a size of dimension, is the activation function.

[0011] Optionally, in S3, a new reflection signal is collected again The signal is the transformer vibration data to be detected. The time domain and frequency domain analysis methods in S1 are used to analyze the new reflected signal. Perform time domain and frequency domain feature extraction to construct a new vibration feature vector , use the trained deep autoencoder network to train the new vibration feature vector Reconstruct and obtain the reconstructed vibration eigenvector , the decoder recovers the original high-dimensional feature vector from the low-dimensional feature space, and the error is calculated by the mean square error: , the calculated reconstruction error is compared with the predetermined threshold For comparison, if Exceeding a predetermined threshold , it is judged as an abnormal vibration mode.

[0012] Optionally, the predetermined threshold in S3 is an initial threshold , set based on historical data and statistical analysis, initial threshold Set to a multiple of the reconstruction error: ,in, is the mean value of the reconstruction error of the normal vibration mode, is the standard deviation of the reconstruction error of the normal vibration mode, Is a constant multiple used to adjust the sensitivity of the threshold.

[0013] Optionally, the reconstruction error of the vibration feature vector may change over time, and the adaptive threshold strategy in S4 specifically includes: Calculate the new reconstruction error mean and standard deviation based on the sliding window in each period of time, and set a size of The sliding window in which the reconstruction error of the vibration feature is calculated in real time is: ; ; in, Indicates at a point in time At this moment, based on the mean of all reconstruction errors in the sliding window, this value reflects the average level of reconstruction errors in the current window. is the size of the sliding window, Indicated in The corresponding reconstruction error value is is the index of the current moment, indicating the current time point or data position when calculating the mean; Indicates at a point in time At this moment, the fluctuation range of the reconstruction error in the current window is measured based on the standard deviation of all reconstruction errors in the sliding window; The scheduled threshold is updated to: ,in is an adjustable constant used to dynamically control the sensitivity of the threshold, is the adaptive threshold at time t.

[0014] Optionally, a comprehensive adjustment mechanism is also included, wherein the comprehensive adjustment mechanism is used to ensure that the initial threshold With the updated threshold Smooth combination: When the system just starts running (with less data), the initial threshold is used as the basis. As more real-time data accumulates, the threshold calculated by the sliding window gradually becomes dominant, and the detection sensitivity is adjusted dynamically. The comprehensive adjustment mechanism introduces a smoothing factor. During the adjustment, the initial threshold and the threshold updated by the sliding window are weighted averaged to avoid sudden changes or over-reliance on a certain threshold: ,in, is the smoothing factor, and , controls the balance between the initial threshold and the sliding window calculation.

[0015] Beneficial effects of the present invention: The present invention is a transformer vibration anomaly detection method based on millimeter-wave radar. Vibration feature learning and reconstruction error calculation are performed through a deep autoencoder network, so that abnormal vibration modes different from normal vibration modes can be identified. Compared with the traditional method, there is no need for complicated signal processing procedures and manual setting of thresholds. Instead, the vibration features are automatically learned and the thresholds are adaptively adjusted according to real-time data, which significantly improves the accuracy of anomaly detection. In particular, under complex operating conditions such as transformer load changes and fluctuations, the system can respond flexibly and avoid false alarms and missed alarms.

[0016] The present invention uses a deep autoencoder network to perform unsupervised learning on normal vibration patterns, and reconstructs new vibration feature vectors based on the learned low-dimensional feature representation, which can effectively capture the potential laws and slight changes in transformer vibration. It combines deep learning technology and uses encoder and decoder structures to automatically extract and compress key information of input vibration features, thereby greatly improving the system's ability to recognize complex vibration patterns. Compared with traditional technologies based on manual feature extraction, the deep autoencoder network can not only learn the intrinsic characteristics of vibration data more comprehensively, but also adapt to subtle changes in the transformer state.

[0017] The present invention introduces an adaptive threshold strategy, which automatically adjusts the detection threshold by dynamically calculating the reconstruction error mean and standard deviation based on a sliding window, combined with the operating status of the transformer and changes in the external environment. The adaptive threshold mechanism can adapt to the operating characteristics of different transformer equipment and fluctuations in the external environment in real time, effectively cope with the impact of factors such as load changes and temperature fluctuations on the vibration mode, ensure stability and efficiency under different working conditions, and avoid the problem of over-sensitivity or insensitivity caused by fixed thresholds by flexibly adjusting the thresholds. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A schematic diagram of a flow chart of an abnormality detection method according to an embodiment of the present invention; Figure 2 Schematic diagram of threshold update according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0021] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0022] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0023] like Figure 1-Figure 2 As shown, the transformer vibration abnormality detection method based on millimeter wave radar includes the following steps: S1, uses millimeter wave radar to transmit millimeter waves to the external structure of the transformer, obtains the original reflection signal on the transformer surface, performs time domain and frequency domain analysis on the original reflection signal, and constructs a vibration feature vector; S2, training and feature learning of deep autoencoder network: Based on the constructed vibration feature vector, the normal transformer vibration mode is unsupervisedly learned through the deep autoencoder network. The deep autoencoder network consists of an encoder and a decoder. The encoder compresses the input vibration feature vector into a low-dimensional feature representation, and the decoder restores the original vibration feature vector from the low-dimensional feature. By training the network, the potential features of the normal vibration mode are automatically learned; S3, detection of abnormal vibration patterns: after training is completed, new reflection signals are collected again, and the new reflection signals are converted into new vibration feature vectors by the method of S1. The new vibration feature vectors are reconstructed using a deep autoencoder network, and the reconstruction error is calculated. If the reconstruction error exceeds a predetermined threshold, it is judged as an abnormal vibration pattern, and the abnormal patterns that are different from the normal vibration patterns are adaptively identified, including changes in frequency, amplitude, and period. S4, as the transformer operating environment and status change, the predetermined threshold needs to be dynamically adjusted. An adaptive threshold strategy is adopted to automatically adjust the predetermined threshold to adapt to the changes in the operating status of different transformer equipment over time.

[0024] In S1, the original reflection signal is analyzed in the time domain and frequency domain to construct the vibration feature vector, which specifically includes: S11, time domain analysis: perform peak detection on the original reflected signal and analyze the amplitude, periodic changes, and instantaneous frequency characteristics of the signal, including: Peak detection is performed through the maximum and minimum values ​​of the signal to calculate the amplitude change of the signal ; Use the signal's zero crossing points and periodic peaks to analyze periodic changes in vibration ; Calculate the instantaneous frequency of a signal , reflecting the frequency change of vibration; S12, frequency domain analysis: Convert the original reflection signal to the frequency domain through Fourier transform, analyze the spectrum characteristics of the signal, and extract the frequency components related to the vibration mode, including: Use fast Fourier transform to perform frequency domain analysis on the reflected signal to obtain the signal spectrum and extract the main frequency components in the signal spectrum ; Analyze the frequency band distribution in the spectrum and calculate the frequency bandwidth of the vibration signal ; S13, combines the time domain features and frequency domain features (amplitude, period, frequency) into a vibration feature vector , the vibration eigenvector It is expressed as: ; in, is the amplitude, It is cyclical. is the main frequency, is the frequency bandwidth, It's time The instantaneous frequency.

[0025] As input data for subsequent deep autoencoder network training.

[0026] 1. Time domain analysis part.

[0027] 1.1, Amplitude calculation: define the reflection signal is the original reflection signal that changes with time. After peak detection, take the maximum value of the signal and minimum value , then the amplitude Defined as: ,in, For signal The maximum value of For signal The minimum value of 1.2, Periodicity analysis and period calculation: Periodicity analysis is based on the zero crossing point of the signal, which is the turning point of the signal from positive to negative or from negative to positive; Period calculation formula: Define signal The zero crossing point is , the zero crossing point indicates the time point when the signal crosses the zero axis, and the time difference between consecutive zero crossing points is calculated: ,in, is the period of the signal, which represents the average time difference between two consecutive zero crossing points. The time of the 𝑖th zero crossing.

[0028] 1.3, Instantaneous frequency calculation: The instantaneous frequency reflects the frequency change of the signal at a certain moment, and is usually calculated by Hilbert transform. Using the Hilbert transform Get the parsed signal of the signal: ; Analyze the instantaneous frequency of a signal Obtained by phase derivative: ;in, For signal The Hilbert transform of is a complex signal, consisting of the original signal and its Hilbert transform. For signal The phase of It is the instantaneous frequency, which indicates the frequency of the signal at a certain moment.

[0029] 2. Frequency domain analysis part.

[0030] 2.1, Fourier transform: The original reflected signal is transformed into Convert to frequency domain signal , analyze the frequency content of the signal: ,in, It's a signal The frequency domain representation of , which indicates the intensity of different frequency components, is the Fourier transform operator, It's the frequency.

[0031] 2.2, Spectrum analysis and frequency component extraction: Spectrum analysis is used to extract the main frequency components in the signal, especially the fundamental frequency and harmonics of vibration; Calculating the spectrum The frequency corresponding to the maximum value of is: ,in, is the dominant frequency of the signal, indicating the fundamental frequency of the vibration, is the signal at frequency The amplitude at .

[0032] 2.3, Bandwidth and Amplitude Analysis: Bandwidth It is expressed by the half-peak width in the spectrum, reflecting the frequency distribution of the signal. The bandwidth is calculated as: ,in, is the highest frequency in the signal spectrum, is the lowest frequency in the signal spectrum.

[0033] S2 specifically includes: S21, based on the constructed vibration eigenvector , unsupervised learning of normal transformer vibration patterns is performed through a deep autoencoder network, which includes an encoder and a decoder, where: Encoder part: The encoder receives the input vibration feature vector , which is compressed through a multi-layer neural network to obtain a low-dimensional feature representation , the encoder gradually reduces the dimension of the input data through several hidden layers to obtain a low-dimensional feature representation , this low-dimensional feature representation can capture the key features of normal vibration modes: ; Decoder part: The decoder represents low-dimensional features Restore the original vibration feature vector through the inverse neural network structure , that is, the input feature vector is gradually reconstructed through back propagation, the goal is to minimize the reconstruction error so that the reconstruction result is as close as possible to the input vibration feature vector: ; S22, training process: Through unsupervised learning, a large number of normal transformer vibration feature vectors are used for training, and the parameters of the encoder and decoder are optimized so that the encoder can effectively compress the normal vibration feature vector into a low-dimensional space and accurately reconstruct the original input through the decoder. During the training process, the reconstruction error function is minimized. To optimize the network parameters, the reconstruction error is measured using the mean square error loss function: ;in, Indicates The vibration feature vector of training samples, Indicates The vibration feature vector obtained by reconstructing the training samples through the autoencoder network, L represents the reconstruction error of the network, which represents the difference between the input and reconstructed signals.

[0034] The encoder uses a fully connected layer structure to gradually reduce the dimension of the data; Assume that the input vibration eigenvector is a -dimensional vector , the encoder's task is to compress it into a -dimensional low-dimensional feature representation ,in ; The encoder process is implemented through the following fully connected layers: ,in, is the weight matrix of the first layer, with a size of , used to convert input data from -Dimensions are mapped to -dimension, is the bias term, with a size of -dimension, is the activation function, using ReLU or Sigmoid activation function, It is a low-dimensional feature representation; The dimension of the data can be further reduced through multiple layers of fully connected layers: ,here and is the weight matrix and bias term of the second layer, which further maps the data.

[0035] The decoder is implemented by a reverse fully connected layer, which transforms the low-dimensional features Gradually map back to the high-dimensional space to restore the original vibration characteristics: ,in, is the weight matrix of the first layer of the decoder, with a size of , used to map low-dimensional features back to high-dimensional space, is the bias term, with a size of dimension, is the activation function. Similarly, the decoder can also use multiple layers of reverse fully connected layers: ,here and are the second layer weights and bias terms of the decoder.

[0036] In S3, a new reflection signal is collected again The signal is the transformer vibration data to be detected. The time domain and frequency domain analysis methods in S1 are used to analyze the new reflected signal. Extract time domain and frequency domain features and construct new vibration feature vectors , use the trained deep autoencoder network to train the new vibration feature vector Reconstruct and obtain the reconstructed vibration eigenvector , the decoder recovers the original high-dimensional feature vector from the low-dimensional feature space, and the error is calculated by the mean square error: , the calculated reconstruction error is compared with the predetermined threshold For comparison, if Exceeding a predetermined threshold , it is judged as an abnormal vibration mode.

[0037] The predetermined threshold in S3 is the initial threshold , set based on historical data and statistical analysis, initial threshold Set to a multiple of the reconstruction error: ,in, is the mean value of the reconstruction error of the normal vibration mode, is the standard deviation of the reconstruction error of the normal vibration mode, Is a constant multiple used to adjust the sensitivity of the threshold.

[0038] The reconstruction error of the vibration feature vector will change over time. The adaptive threshold strategy in S4 specifically includes: Calculate the new reconstruction error mean and standard deviation based on the sliding window in each period of time, and set a size of The sliding window in which the reconstruction error of the vibration feature is calculated in real time is: ; ; in, Indicates at a point in time At this moment, based on the mean of all reconstruction errors in the sliding window, this value reflects the average level of reconstruction errors in the current window. is the size of the sliding window, Indicated in The corresponding reconstruction error value is is the index of the current moment, indicating the current time point or data position when calculating the mean; Indicates at a point in time At this moment, the fluctuation range of the reconstruction error in the current window is measured based on the standard deviation of all reconstruction errors in the sliding window; The scheduled threshold is updated to: ,in is an adjustable constant used to dynamically control the sensitivity of the threshold, is the adaptive threshold at time t.

[0039] It also includes a comprehensive adjustment mechanism to ensure that the initial threshold With the updated threshold Smooth combination: When the system just starts running (with less data), the initial threshold is used as the basis. As more real-time data accumulates, the threshold calculated by the sliding window gradually becomes dominant, and the detection sensitivity is adjusted dynamically. The comprehensive adjustment mechanism introduces a smoothing factor. During the adjustment, the initial threshold and the threshold updated by the sliding window are weighted averaged to avoid sudden changes or over-reliance on a certain threshold: ,in, is the smoothing factor, and , controls the balance between the initial threshold and the sliding window calculation.

[0040] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0041] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A transformer vibration anomaly detection method based on millimeter wave radar, characterized in that: The following steps are involved: S1, uses millimeter wave radar to transmit millimeter waves to the external structure of the transformer, obtains the original reflection signal on the transformer surface, performs time domain and frequency domain analysis on the original reflection signal, and constructs a vibration feature vector; S2, training and feature learning of deep autoencoder network: based on the constructed vibration feature vector, unsupervised learning of normal transformer vibration mode is performed through deep autoencoder network, the deep autoencoder network is composed of encoder and decoder, the encoder compresses the input vibration feature vector into a low-dimensional feature representation, and the decoder restores the original vibration feature vector from the low-dimensional feature, and automatically learns the potential features of the normal vibration mode through training network; S3, detection of abnormal vibration patterns: after training is completed, new reflection signals are collected again, and the new reflection signals are converted into new vibration feature vectors by the method of S1. The new vibration feature vectors are reconstructed using a deep autoencoder network, and the reconstruction error is calculated. If the reconstruction error exceeds a predetermined threshold, it is judged as an abnormal vibration pattern, and the abnormal patterns that are different from the normal vibration patterns are adaptively identified, including changes in frequency, amplitude, and period. S4, adopts an adaptive threshold strategy to automatically adjust the predetermined threshold to adapt to the changes in the operating status of different transformer equipment over time.

2. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 1 is characterized in that: In S1, the original reflection signal is analyzed in the time domain and the frequency domain to construct the vibration feature vector, which specifically includes: S11, time domain analysis: perform peak detection on the original reflected signal and analyze the amplitude, periodic changes, and instantaneous frequency characteristics of the signal, including: Peak detection is performed through the maximum and minimum values ​​of the signal to calculate the amplitude change of the signal ; Use the signal's zero crossing points and periodic peaks to analyze periodic changes in vibration ; Calculate the instantaneous frequency of a signal , reflecting the frequency change of vibration; S12, frequency domain analysis: Convert the original reflection signal to the frequency domain through Fourier transform, analyze the spectrum characteristics of the signal, and extract the frequency components related to the vibration mode, including: Use fast Fourier transform to perform frequency domain analysis on the reflected signal to obtain the signal spectrum and extract the main frequency components in the signal spectrum ; Analyze the frequency band distribution in the spectrum and calculate the frequency bandwidth of the vibration signal ; S13, combining the time domain features and frequency domain features into a vibration feature vector , the vibration eigenvector It is expressed as: ； in, is the amplitude, It is cyclical. is the main frequency, is the frequency bandwidth, It's time The instantaneous frequency.

3. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 2 is characterized in that: The S2 specifically includes: S21, based on the constructed vibration eigenvector , unsupervised learning of normal transformer vibration modes is performed through a deep autoencoder network, the deep autoencoder network comprising an encoder and a decoder, wherein: Encoder part: The encoder receives the input vibration feature vector , which is compressed through a multi-layer neural network to obtain a low-dimensional feature representation , the encoder gradually reduces the dimension of the input data through several hidden layers to obtain a low-dimensional feature representation , this low-dimensional feature representation can capture the key features of normal vibration modes; Decoder part: The decoder represents low-dimensional features Restore the original vibration feature vector through the inverse neural network structure , that is, the input feature vector is gradually reconstructed through back propagation, the goal is to minimize the reconstruction error so that the reconstruction result is close to the input vibration feature vector: ; S22, training process: Through unsupervised learning, the vibration feature vector of the normal transformer is used for training, and the parameters of the encoder and decoder are optimized so that the encoder compresses the normal vibration feature vector into a low-dimensional space and accurately reconstructs the original input through the decoder. During the training process, the reconstruction error function is minimized. To optimize the network parameters, the reconstruction error is measured using the mean square error loss function: ;in, Indicates The vibration feature vector of training samples, Indicates The vibration feature vector obtained by reconstructing the training samples through the autoencoder network, L represents the reconstruction error of the network, which represents the difference between the input and reconstructed signals.

4. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 3 is characterized in that: The encoder uses a fully connected layer structure to gradually reduce the dimension of the data; Assume that the input vibration eigenvector is a -dimensional vector , the encoder's task is to compress it into a -dimensional low-dimensional feature representation ,in ; The encoder process is implemented through the following fully connected layers: ,in, is the weight matrix of the first layer, with a size of , used to convert input data from -Dimensions are mapped to -dimension, is the bias term, with a size of -dimension, is the activation function, It is a low-dimensional feature representation.

5. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 4 is characterized in that: The decoder is implemented by a reverse fully connected layer, converting the low-dimensional features Gradually map back to the high-dimensional space to restore the original vibration characteristics: ,in, is the weight matrix of the first layer of the decoder, with a size of , used to map low-dimensional features back to high-dimensional space, is the bias term, with a size of dimension, is the activation function.

6. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 5 is characterized in that: In S3, a new reflection signal is collected again. The signal is the transformer vibration data to be detected. The time domain and frequency domain analysis methods in S1 are used to analyze the new reflected signal. Extract time domain and frequency domain features and construct new vibration feature vectors , use the trained deep autoencoder network to train the new vibration feature vector Reconstruct and obtain the reconstructed vibration eigenvector , the decoder recovers the original high-dimensional feature vector from the low-dimensional feature space, and the error is calculated by the mean square error: , the calculated reconstruction error is compared with the predetermined threshold For comparison, if Exceeding a predetermined threshold , it is judged as an abnormal vibration mode.

7. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 6 is characterized in that: The predetermined threshold in S3 is the initial threshold , set based on historical data and statistical analysis, initial threshold Set to a multiple of the reconstruction error: ,in, is the mean value of the reconstruction error of the normal vibration mode, is the standard deviation of the reconstruction error of the normal vibration mode, Is a constant multiple used to adjust the sensitivity of the threshold.

8. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 7 is characterized in that: The reconstruction error of the vibration feature vector will change over time. The adaptive threshold strategy in S4 specifically includes: Calculate the new reconstruction error mean and standard deviation based on the sliding window in each period of time, and set a size of The sliding window in which the reconstruction error of the vibration feature is calculated in real time is: ； ； in, Indicates at a point in time At this moment, based on the mean of all reconstruction errors in the sliding window, this value reflects the average level of reconstruction errors in the current window. is the size of the sliding window, Indicated in The corresponding reconstruction error value is is the index of the current moment, indicating the current time point or data position when calculating the mean; Indicates at a point in time At this moment, the fluctuation range of the reconstruction error in the current window is measured based on the standard deviation of all reconstruction errors in the sliding window; The scheduled threshold is updated to: ,in is an adjustable constant used to dynamically control the sensitivity of the threshold, is the adaptive threshold at time t.

9. The transformer vibration anomaly detection method based on millimeter wave radar according to claim 8, characterized in that: Also included is a comprehensive adjustment mechanism for ensuring that the initial threshold With the updated threshold Smooth combination: When the system just starts running, the initial threshold is used as the basis. As more real-time data accumulates, the threshold calculated by the sliding window gradually becomes dominant, and the detection sensitivity is adjusted dynamically. The comprehensive adjustment mechanism introduces a smoothing factor. During the adjustment, the initial threshold and the threshold updated by the sliding window are weighted averaged to avoid sudden changes or over-reliance on a certain threshold: ,in, is the smoothing factor, and , controls the balance between the initial threshold and the sliding window calculation.

Images