A Transformer Vibration Feature Extraction Method Based on Multi-Measuring Point Collaboration and Time-Frequency Matrix Reconstruction

Abstract

This invention discloses a method for extracting transformer vibration features based on multi-point collaborative sampling and time-frequency matrix reconstruction, comprising the following steps: S1. Establishing a finite element simulation model based on the transformer structure and acquiring original vibration signals from multiple locations on the surface of the tank; S2. Performing normalization preprocessing on the original vibration signals and dividing them into multiple data samples through sliding window sampling; S3. Multiplying each data sample by a window function, calculating the Fourier transform of each window, and constructing the short-time Fourier transform time-frequency matrix of the vibration signal by spectral splicing; S4. Based on the short-time Fourier transform time-frequency matrix, calculating the instantaneous frequency after signal compression, and mapping pseudo-frequency components within a set width to the instantaneous frequency according to the mapping relationship, thereby obtaining a high-resolution time-frequency matrix reconstructed by synchronous Fourier compression transform. This invention has the advantages of high feature extraction accuracy and high reliability.

Description

Technical Field

[0001] This invention mainly relates to the field of vibration detection technology for oil-immersed transformer enclosures, specifically to a method for extracting transformer vibration features based on multi-measurement point collaboration and time-frequency matrix reconstruction. Background Technology

[0002] Vibration signals from the tank of an oil-immersed transformer are crucial information carriers reflecting the mechanical state of key components such as the windings and core. Accurate and early extraction of fault characteristics has long been a core technological bottleneck in the field of intelligent diagnostics for power equipment. This challenge stems primarily from two inherent difficulties: First, transformers are subjected to complex coupled excitations of electromagnetic force, magnetostriction and mechanical stress during operation. Their vibration signals exhibit typical non-stationarity (time-varying statistical characteristics) and nonlinear dynamic behavior. Traditional time-domain or frequency-domain analysis methods based on the stationarity assumption are difficult to fully characterize their dynamic evolution mechanism. Secondly, the characteristic components caused by early mechanical faults (such as slight deformation of the winding or slight loosening of the iron core) are extremely weak and are often submerged in strong background noise (including electromagnetic noise, load fluctuations and environmental interference) with a signal-to-noise ratio of less than -10dB, causing conventional detection methods to fail.

[0003] Although time-frequency analysis technology can jointly analyze the time-frequency structure of signals and provide a new approach for condition monitoring, under such poor signal-to-noise ratio conditions, existing time-frequency analysis methods still generally face problems such as characteristic energy diffusion, insufficient time-frequency resolution, and poor noise robustness, making it difficult to achieve reliable identification and early warning of early faults.

[0004] Therefore, developing a novel analytical method that can effectively extract weak non-stationary fault characteristics from transformer vibration signals under strong noise background is of great theoretical significance and engineering urgency. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention provides a transformer vibration feature extraction method based on multi-measurement point collaboration and time-frequency matrix reconstruction that offers accurate and reliable feature extraction.

[0006] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows: A method for extracting transformer vibration features based on multi-measurement point collaboration and time-frequency matrix reconstruction includes the following steps: S1. Establish a finite element simulation model based on the actual transformer structure and obtain the original vibration signals at multiple locations on the surface of the oil tank; S2. The original vibration signal collected in step S1 is normalized and preprocessed, and then divided into multiple data samples by sliding window sampling; S3. Multiply each data sample obtained in step S2 by the window function, calculate the Fourier transform of each window respectively, and construct the short-time Fourier transform time-frequency matrix of the vibration signal by splicing the spectrum. S4. Based on the short-time Fourier transform time-frequency matrix obtained in step S3, calculate the instantaneous frequency of the compressed signal, and map the pseudo-frequency components within a set width to the instantaneous frequency according to the mapping relationship to obtain the high-resolution time-frequency matrix reconstructed by the Fourier synchronous compression transform.

[0007] Preferably, the specific process of step S3 is as follows: S301. Multiply each data sample by the window function point by point, perform a fast Fourier transform on the windowed signal segment, and obtain the spectrum of the time segment; S302. Slide the window along the time axis and repeat step S301 to arrange the spectra corresponding to all time segments in chronological order, thus forming the short-time Fourier transform time-frequency matrix of the sample. Its expression is:

[0008] in, This is the original vibration signal; For window functions, For frequency, For time parameters, For integration variables, It is the imaginary unit.

[0009] Preferably, the window function length is dynamically adjusted according to the instantaneous frequency change rate of the data sample. When the signal changes drastically, a short window is used to maintain time resolution, and when the change is gradual, a long window is used to improve frequency resolution.

[0010] Preferably, the specific process of dynamically adjusting the window function length based on the instantaneous frequency change rate of the data samples is as follows: The instantaneous frequency change rate df(t) / dt of the vibration signal is calculated in real time, and the phase derivative of the analytic signal is extracted by Hilbert transform. Establish the mapping relationship between window length L and instantaneous frequency change rate df(t) / dt: L(t)=Lmin+(Lmax-Lmin)×exp(-α×|df(t) / dt|), where α is the adjustment factor, and Lmin and Lmax are the preset minimum and maximum window lengths, respectively.

[0011] Preferably, the window function is a Hamming window, a Hanning window, or a Kaiser window.

[0012] Preferably, the specific process of step S4 is as follows: S401. Results of the short-time Fourier transform Differentiate the phase of each frequency component along the time direction to extract the instantaneous frequency of the compressed signal. :

[0013] in, Indicates the instantaneous frequency; arg represents the argument operation. S402. Establish the mapping relationship from the STFT frequency coordinate f to the actual instantaneous frequency coordinate; for each time point t, The energy value at frequency f is redistributed to the instantaneous frequency. At the corresponding new frequency η, the one-dimensional vibration signal is converted into a high-resolution time-frequency image using Fourier synchronous compression transform. The specific formula is as follows:

[0014] in, This is the time-frequency representation after synchronous compression transformation. For the Dirac function, This represents the compressed frequency variable.

[0015] Preferably, in step S2, the normalization preprocessing adopts the Z-Score normalization method or the Max-min normalization method.

[0016] The present invention also discloses a computer program product, comprising a computer program that, when executed by a processor, performs the steps of the method described above.

[0017] The present invention further discloses a computer-readable storage medium having a computer program stored thereon, the computer program executing the steps of the method described above when run by a processor.

[0018] The present invention also discloses a transformer vibration feature extraction system based on multi-measurement point collaboration and time-frequency matrix reconstruction, including a memory and a processor connected to each other. The memory stores a computer program, and the computer program executes the steps of the method described above when run by the processor.

[0019] Compared with the prior art, the advantages of the present invention are as follows: This invention can accurately capture weak characteristic signals caused by early mechanical defects such as winding deformation and core loosening under conditions of strong electromagnetic interference and complex mechanical noise. Starting from the physical mechanism of vibration generation, the vibration spectrum of a healthy transformer strictly follows the excitation law of magnetostriction effect and winding electromagnetic force; however, in the early stages of a fault, the spectral energy distribution will exhibit three typical evolution trends: 1. Energy redistribution of characteristic harmonic clusters; 2. Abnormal enhancement of sideband modulation structures; 3. Non-stationary transient fluctuations in vibration amplitude. These microscopic features constitute a unique "fingerprint" for identifying mechanical faults.

[0020] This method employs a time-frequency joint analysis technique by constructing a time-frequency joint analysis system. The two-dimensional energy distribution matrix, while maintaining the ability to locate transient impact events in time, achieves precise focusing on the modulation frequency band structure, thereby transforming early fault characteristics that were previously masked by noise into quantifiable and traceable monitoring indicators. This technological advancement not only provides a reliable basis for the early diagnosis and warning of transformer mechanical faults, but more importantly, the extracted time-frequency characteristics have clear physical mechanism support, providing an interpretable and verifiable decision-making basis for subsequent condition assessment and the formulation of predictive maintenance strategies.

[0021] The method of this invention integrates signal processing techniques such as short-time Fourier transform and sliding time-domain synchronous averaging to construct a complete vibration signal feature extraction system. It can effectively overcome the limitations of traditional methods in complex working conditions and significantly improve the accuracy and computational efficiency of feature extraction for non-stationary, multi-component vibration signals. It is particularly suitable for the accurate analysis and state identification of vibration signals during the operation of power transformers.

[0022] This invention constructs a feature extraction system integrating multiple technologies, innovatively combining short-time Fourier transform and sliding time-domain synchronous averaging to form a complementary and synergistic processing flow. This invention enables effective analysis of non-stationary, multi-component signals. This fusion method is particularly effective for the non-stationarity and multi-component characteristics of vibration signals during transformer operation, enabling more precise separation and extraction of essential features hidden under complex operating conditions. This invention balances the accuracy of feature extraction with computational efficiency. Through the synergistic effect of the technologies, while improving the anti-interference capability and accuracy of feature extraction, it also ensures the computational efficiency of the entire analysis process, meeting the practical needs of online monitoring. These innovations work together to ultimately provide more reliable technical support for the accurate monitoring and fault diagnosis of transformer mechanical conditions.

[0023] This invention significantly improves the ability to identify fault features in the time-frequency domain through an innovative signal processing flow. Specifically, normalization processing effectively eliminates the dimensional differences of multi-source data, achieving standardized expression of feature parameters; the time-frequency analysis method based on short-time Fourier transform enhances the time-frequency joint analysis capability for non-stationary signals and improves signal-to-noise separation performance; and the introduction of synchronous compression transform technology, through a frequency domain energy refocusing mechanism, improves the identification accuracy of instantaneous frequency features by an order of magnitude.

[0024] At the engineering application level, the method of this invention demonstrates good universality and scalability. It can not only accurately identify typical mechanical faults in oil-immersed transformers, but also be extended to multiple industrial monitoring scenarios such as acoustic testing and vibration monitoring. This technological breakthrough provides key support for the construction of intelligent equipment condition monitoring systems, effectively reducing operation and maintenance costs while improving equipment operational reliability, and possesses significant engineering application value. Attached Figure Description

[0025] Figure 1 This is a flowchart of an embodiment of the transformer vibration feature extraction method of the present invention.

[0026] Figure 2 This is an image of the original vibration signal extracted from the surface of the oil tank of the oil-immersed transformer according to the present invention.

[0027] Figure 3 This is a diagram of the signal convolution method of the present invention through multi-channel input.

[0028] Figure 4 This is a schematic diagram of overlapping sampling in this invention.

[0029] Figure 5 This is an example diagram of a time-domain signal in this invention.

[0030] Figure 6 These are the FSST time-frequency diagrams corresponding to different window functions in this invention. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown in the figure, the transformer vibration feature extraction method based on multi-measurement point collaboration and time-frequency matrix reconstruction provided by the present invention includes the following steps: S1. A finite element simulation model is established based on the actual transformer structure, and multiple sensors are arranged on the surface of the oil tank to collect the original vibration signals; the original vibration signals are represented in the form of a one-dimensional array, denoted as . s ( t ),in t This is a time parameter.

[0033] S2. Normalize and preprocess the acquired raw vibration signals and divide them into multiple data samples; Specifically, the original vibration signal is normalized using the Z-Score normalization method or the Max-min normalization method; then, the normalized vibration signal is divided into multiple data samples using the sliding window sampling method.

[0034] S3. Multiply each data sample by the window function, calculate the Fourier transform of each window, and construct the short-time Fourier transform time-frequency matrix of the vibration signal by splicing the spectrum. Specifically, data samples s ( t ) and window function g ( t Multiply by the given values, calculate the Fourier transform of each window, and obtain the short-time Fourier transform matrix by splicing the spectra. :

[0035] in, This is the result of the short-time Fourier transform. g ( t ) is a window function.

[0036] The Fourier transform formula is: ; In the formula, The signal after Fourier transform. f For frequency, i It is the imaginary unit.

[0037] S4. Calculate the instantaneous frequency of the compressed data sample, establish a frequency domain mapping relationship, map the pseudo-frequency components within a set width to the instantaneous frequency, and obtain the time-frequency matrix after Fourier synchronous compression transformation.

[0038] Specifically, regarding the short-time Fourier transform results By taking the derivative, we obtain the instantaneous frequency of the compressed signal. :

[0039] in, This represents the instantaneous frequency, and arg represents the argument operation.

[0040] Then based on the mapping relationship The pseudo-frequency components within a set width are mapped to instantaneous frequencies, and the one-dimensional vibration signal is converted into a time-frequency image using a Fourier synchronous compression transform:

[0041] in, This is the time-frequency representation after synchronous compression transformation. For the Dirac function, This represents the compressed frequency variable.

[0042] This invention can accurately capture weak characteristic signals caused by early mechanical defects such as winding deformation and core loosening under conditions of strong electromagnetic interference and complex mechanical noise. Starting from the physical mechanism of vibration generation, the vibration spectrum of a healthy transformer strictly follows the excitation law of magnetostriction effect and winding electromagnetic force; however, in the early stages of a fault, the spectral energy distribution will exhibit three typical evolution trends: 1. Energy redistribution of characteristic harmonic clusters; 2. Abnormal enhancement of sideband modulation structures; 3. Non-stationary transient fluctuations in vibration amplitude. These microscopic features constitute a unique "fingerprint" for identifying mechanical faults.

[0043] This method employs a time-frequency joint analysis technique by constructing a time-frequency joint analysis system. The two-dimensional energy distribution matrix, while maintaining the ability to locate transient impact events in time, achieves precise focusing on the modulation frequency band structure, thereby transforming early fault characteristics that were previously masked by noise into quantifiable and traceable monitoring indicators. This technological advancement not only provides a reliable basis for the early diagnosis and warning of transformer mechanical faults, but more importantly, the extracted time-frequency characteristics have clear physical mechanism support, providing an interpretable and verifiable decision-making basis for subsequent condition assessment and the formulation of predictive maintenance strategies.

[0044] The method of this invention integrates signal processing techniques such as short-time Fourier transform and sliding time-domain synchronous averaging to construct a complete vibration signal feature extraction system. It can effectively overcome the limitations of traditional methods in complex working conditions and significantly improve the accuracy and computational efficiency of feature extraction for non-stationary, multi-component vibration signals. It is particularly suitable for the accurate analysis and state identification of vibration signals during the operation of power transformers.

[0045] This invention constructs a feature extraction system integrating multiple technologies, innovatively combining short-time Fourier transform and sliding time-domain synchronous averaging to form a complementary and synergistic processing flow. This invention enables effective analysis of non-stationary, multi-component signals. This fusion method is particularly effective for the non-stationarity and multi-component characteristics of vibration signals during transformer operation, enabling more precise separation and extraction of essential features hidden under complex operating conditions. This invention balances the accuracy of feature extraction with computational efficiency. Through the synergistic effect of the technologies, while improving the anti-interference capability and accuracy of feature extraction, it also ensures the computational efficiency of the entire analysis process, meeting the practical needs of online monitoring. These innovations work together to ultimately provide more reliable technical support for the accurate monitoring and fault diagnosis of transformer mechanical conditions.

[0046] This invention significantly improves the ability to identify fault features in the time-frequency domain through an innovative signal processing flow. Specifically, normalization processing effectively eliminates the dimensional differences of multi-source data, achieving standardized expression of feature parameters; the time-frequency analysis method based on short-time Fourier transform enhances the time-frequency joint analysis capability for non-stationary signals and improves signal-to-noise separation performance; and the introduction of synchronous compression transform technology, through a frequency domain energy refocusing mechanism, improves the identification accuracy of instantaneous frequency features by an order of magnitude.

[0047] At the engineering application level, the method of this invention demonstrates good universality and scalability. It can not only accurately identify typical mechanical faults in oil-immersed transformers, but also be extended to multiple industrial monitoring scenarios such as acoustic testing and vibration monitoring. This technological breakthrough provides key support for the construction of intelligent equipment condition monitoring systems, effectively reducing operation and maintenance costs while improving equipment operational reliability, and possesses significant engineering application value.

[0048] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0049] The core technical solution of this invention adopts a multi-level signal processing architecture, and achieves accurate feature extraction of vibration signals through the following three key steps: First, a data standardization system is established based on normalization processing to eliminate the dimensional differences of multi-source feature parameters and ensure the consistency of the benchmark for subsequent analysis. Secondly, a time-frequency joint analysis framework is constructed using short-time Fourier transform to achieve time-frequency domain collaborative analysis of non-stationary signals, effectively separating fault features from background noise; Finally, synchronous compression transform is used to implement frequency domain energy refocusing, breaking through the frequency resolution limitations of traditional methods.

[0050] This progressive processing strategy, through the complementary advantages of each stage, forms a complete signal purification and feature enhancement technology chain. It can not only accurately extract fault features from mechanical vibration signals, but also be widely used in signal analysis scenarios in multiple fields such as acoustic detection and industrial monitoring.

[0051] See appendix Figure 1 The transformer vibration feature extraction method based on multi-measurement point collaboration and time-frequency matrix reconstruction provided by this invention includes the following steps: S1. Establish a finite element simulation model based on actual time, and use sensors to acquire the original vibration signal of the surface of the oil-immersed transformer tank. Specifically, this includes the following steps: First, based on the actual structural drawings of the target transformer, a three-dimensional electromagnetic-mechanical coupling simulation model is established using finite element analysis software such as ANSYS or COMSOL. Through simulation analysis of the transformer's surface vibration response under rated load and typical fault conditions (such as slight winding deformation and localized core loosening), the vibration-sensitive areas and key modes on the tank surface are determined.

[0052] A signal acquisition system is constructed, including facilities such as a signal generator, signal sensors, and a signal acquisition unit. Parameters such as the excitation method, excitation amplitude, excitation time, and excitation waveform of the signal generator are set, and parameters such as the sampling frequency, sampling time, and sampling length of the signal acquisition unit are determined. Specifically, based on simulation results, multiple (e.g., 8) piezoelectric accelerometers are optimally deployed at key locations on the surface of the oil tank (such as the middle of the tank wall, the area corresponding to the winding, and the area corresponding to the iron core) to form a distributed monitoring array. Figure 2 As shown. The sensor model selected has a frequency response range of 0.5Hz-5kHz and a sensitivity of 100mV / g to ensure coverage of the main mechanical vibration frequency band of the transformer.

[0053] After the experimental setup was constructed and confirmed to be correct, the signal acquisition system was connected to the experimental components to carry out the experiment. The signal acquisition system was used to collect the vibration signals of each sensor of the transformer under stable operating conditions, which were recorded as the original signal s(t).

[0054] The collected vibration signals are examined, and vibration signals with obvious deviations (such as strong noise interference) are removed. The phase of each vibration signal is aligned and the time domain is averaged to generate the fused original vibration signal.

[0055] Figure 3 This invention demonstrates the collaborative processing and feature enhancement workflow for multi-channel vibration signals. For example... Figure 3 As shown, the raw vibration signals synchronously collected from multiple sensors located at different positions on the surface of the oil tank are first phase aligned and time synchronized to form parallel multi-channel inputs. Subsequently, the signals from each channel are fed into a specially designed convolution processing module. This module extracts and fuses local features of the signals through convolution kernels, effectively enhancing the energy of fault-related components and suppressing inconsistencies between channels. The processed multi-channel data is further fused into a high-quality enhanced signal, which serves as the input for subsequent time-frequency analysis.

[0056] Figure 3 This intuitively demonstrates how the present invention improves the signal-to-noise ratio and feature saliency of data from both spatial and signal dimensions through the synergy and convolution operation of multi-measurement point information, laying a key foundation for subsequent high-precision time-frequency analysis.

[0057] This invention achieves optimized measurement points and collaborative data acquisition: Based on the refined finite element model and numerical simulation of the transformer, a distributed sensor array is optimally deployed on the surface of the oil tank, enabling the synchronous acquisition of dynamic response signals from multiple locations from key network nodes, laying a high-quality data foundation for subsequent analysis.

[0058] S2. The vibration signal obtained in step S1 is normalized and divided into multiple groups of data samples, such as... Figure 5 As shown.

[0059] S201. The original vibration signal of the transformer tank surface is collected using a sensor, and the collected original vibration signal is represented in the form of a one-dimensional array, wherein... t For time parameters; S202. The original vibration signal matrix is ​​normalized to eliminate the influence of dimensions. The specific method is as follows: Method 1) Using Z The Score normalization method processes the fused vibration signal and transforms the vibration signal into a standard normal distribution. Method 2) Using Max The min normalization method processes the fused vibration signal, mapping the vibration signal to the range [0,1].

[0060] S203. Set the parameters for sliding window sampling: window size and translation step size. Sliding window sampling is performed on the preprocessed vibration signal matrix to obtain multiple data samples. The length of a single data sample is equal to the sampling window size. Adjusting the translation step size controls the number of data samples.

[0061] Figure 4 A schematic diagram illustrating the sliding window overlap sampling strategy employed in this invention is shown, visually demonstrating how a continuous vibration signal sequence is divided into a series of data samples for subsequent analysis. Figure 4 As shown, a fixed-length sampling window slides sequentially along the time axis with a specific translation step size. Each signal segment covered by the window constitutes an independent data sample. This overlapping method ensures that each sample contains enough information for stable frequency domain analysis, and avoids the loss of important transient features caused by window boundary cutting through partial overlap between samples. Thus, while maximizing data utilization, it ensures the effective preservation of signal temporal continuity in segmented processing.

[0062] S3. For each data sample, perform a short-time Fourier transform (STFT) to construct the initial time-frequency matrix; Each data sample s ( t ) and window function ( tMultiply the signals point by point, perform a Fast Fourier Transform (FFT) on the windowed signal segments to obtain the spectrum of that time segment. Slide the window along the time axis and repeat the above process, arranging the spectra corresponding to all time segments in chronological order to form the short-time Fourier transform time-frequency matrix of the sample. Its expression is:

[0063] in, This represents the result of the short-time Fourier transform. t ) is a window function. For frequency, For time parameters, It is the imaginary unit.

[0064] This matrix visually illustrates the distribution of signal energy over time and frequency in the form of a two-dimensional image.

[0065] Where the window function ( t You can choose from Hanming windows, Kaiser windows, or Hanning windows, corresponding to... Figure 6 The specific values ​​(a), (b), and (c) in the text depend on the actual situation.

[0066] An adaptive window function adjustment mechanism is employed: based on s ( t The window function length is dynamically adjusted based on the instantaneous frequency change rate. A short window is used when the signal changes drastically to maintain time resolution, while a long window is used when the change is gradual to improve frequency resolution. The specific process is as follows: Real-time signal calculation s ( t The instantaneous frequency change rate df(t) / dt of the analytic signal is obtained by extracting the phase derivative of the analytic signal through Hilbert transform; Establish the mapping relationship between window length L and instantaneous frequency change rate df(t) / dt: L(t)=Lmin+(Lmax-Lmin)×exp(-α×|df(t) / dt|), where α is the adjustment factor, and Lmin and Lmax are the preset minimum and maximum window lengths, respectively; When the change in adjacent window length exceeds 20%, a dual-window transition smoothing algorithm is used to avoid spectral abrupt changes.

[0067] The dual-window transition smoothing algorithm is a transition strategy designed to avoid spectral jumps or breaks in time-frequency ridges caused by sudden changes in window length when adaptively adjusting the short-time Fourier transform window length. Its core principle is as follows: when the algorithm determines that a switch from the current window length L1 to a new window length L2 is needed, it does not switch immediately. Instead, within a brief overlapping transition interval, it simultaneously uses two window functions of different lengths, L1 and L2, to process the same signal segment, and then performs a linear weighted fusion of the two resulting spectra. As processing progresses, the weights smoothly shift from relying entirely on the old window spectrum to relying entirely on the new window spectrum. This method effectively maintains the continuity and consistency of the time-frequency matrix at the window length change points, avoiding the introduction of spurious transient frequency components, thereby ensuring the accuracy of subsequent time-frequency analysis.

[0068] This step involves the construction and enhancement of the time-frequency matrix. By proposing an improved multi-point signal joint spectrum transformation algorithm and combining it with an adaptive window function and a segmented processing strategy, an enhanced time-frequency distribution matrix that can accurately reflect the local time-frequency characteristics of the signal is constructed.

[0069] S4. Based on the short-time Fourier transform results, calculate the instantaneous frequency of the compressed signal, establish a frequency domain mapping relationship, and map the pseudo-frequency components within a set width to the instantaneous frequency, thereby obtaining the time-frequency matrix reconstructed by the Fourier Synchronous Compression Transform (FSST). The specific steps are as follows: S401. Differentiate the phase of each frequency component along the time direction from the short-time Fourier transform result to extract the instantaneous frequency of the compressed signal. :

[0070] in, This represents the instantaneous frequency, and arg represents the argument operation.

[0071] S402. Establish the mapping relationship from the STFT frequency coordinate f to the actual instantaneous frequency coordinate. For each time point t, The energy value at frequency f is redistributed to the instantaneous frequency. At the corresponding new frequency η, the one-dimensional vibration signal is converted into a high-resolution time-frequency image using Fourier synchronous compression transform. The specific formula is as follows:

[0072] in, This is the time-frequency representation after synchronous compression transformation. For the Dirac function, This represents the compressed frequency variable.

[0073] By introducing signal compression sensing and frequency mapping technology, the time-frequency matrix is ​​reduced in dimension and its features are purified. Finally, the frequency domain features are accurately mapped to the time-frequency plane to generate a high-resolution time-frequency distribution map, thereby clearly revealing the transient characteristics of the vibration signal.

[0074] Finally, from the reconstructed high-resolution time-frequency matrix The time-frequency characteristic parameters characterizing the mechanical state of the transformer are extracted and output for fault diagnosis or condition assessment. These time-frequency characteristic parameters include: the energy distribution of characteristic harmonic clusters, the intensity of the sideband modulation structure, and the non-stationary transient fluctuation index of the vibration amplitude.

[0075] In summary, the vibration time-frequency feature extraction method for oil-immersed transformers provided by this invention significantly improves the effective separation capability of fault features by optimizing the signal processing flow. The extracted time-frequency features can be used as key inputs to support intelligent diagnostic models such as neural networks, thereby enhancing the sensitivity and accuracy of fault detection. This method provides reliable technical support for equipment condition monitoring and maintenance management, helps to identify potential faults early, reduces the risk of unplanned downtime, and ensures the safe and stable operation of power equipment.

[0076] The present invention also discloses a computer program product, comprising a computer program that, when executed by a processor, performs the steps of the method described above.

[0077] The present invention further discloses a computer-readable storage medium having a computer program stored thereon, the computer program executing the steps of the method described above when run by a processor.

[0078] The present invention also discloses a transformer vibration feature extraction system based on multi-measurement point collaboration and time-frequency matrix reconstruction, including a memory and a processor connected to each other. The memory stores a computer program, and the computer program executes the steps of the method described above when run by the processor.

[0079] The products, media, and systems of the present invention, corresponding to the methods described above, also possess the advantages described above.

[0080] The present invention can implement all or part of the processes in the methods of the above embodiments, or it can be implemented by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium includes: any entity or device capable of carrying computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. The memory is used to store computer programs and / or modules. The processor implements various functions by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory. The memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0081] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for extracting transformer vibration features based on multi-measurement point collaboration and time-frequency matrix reconstruction, characterized in that, Including the following steps: S1. Establish a finite element simulation model based on the actual transformer structure and obtain the original vibration signals at multiple locations on the surface of the oil tank; S2. The original vibration signal collected in step S1 is normalized and preprocessed, and then divided into multiple data samples by sliding window sampling; S3. Multiply each data sample obtained in step S2 by the window function, calculate the Fourier transform of each window respectively, and construct the short-time Fourier transform time-frequency matrix of the vibration signal by splicing the spectrum. S4. Based on the short-time Fourier transform time-frequency matrix obtained in step S3, calculate the instantaneous frequency of the compressed signal, and map the pseudo-frequency components within a set width to the instantaneous frequency according to the mapping relationship to obtain the high-resolution time-frequency matrix reconstructed by the Fourier synchronous compression transform.

2. The transformer vibration feature extraction method based on multi-measurement point collaboration and time-frequency matrix reconstruction according to claim 1, characterized in that, The specific process of step S3 is as follows: S301. Multiply each data sample by the window function point by point, perform a fast Fourier transform on the windowed signal segment, and obtain the spectrum of the time segment; S302. Slide the window along the time axis and repeat step S301 to arrange the spectra corresponding to all time segments in chronological order, thus forming the short-time Fourier transform time-frequency matrix of the sample. Its expression is: in, This is the original vibration signal; For window functions, For frequency, For time parameters, For integration variables, It is the imaginary unit.

3. The transformer vibration feature extraction method based on multi-measurement point collaboration and time-frequency matrix reconstruction according to claim 2, characterized in that, The window function length is dynamically adjusted based on the instantaneous frequency change rate of the data sample. When the signal changes drastically, a short window is used to maintain time resolution, while when the change is gradual, a long window is used to improve frequency resolution.

4. The transformer vibration feature extraction method based on multi-measurement point collaboration and time-frequency matrix reconstruction according to claim 3, characterized in that, The specific process of dynamically adjusting the window function length based on the instantaneous rate of change of the data sample is as follows: The instantaneous frequency change rate df(t) / dt of the vibration signal is calculated in real time, and the phase derivative of the analytic signal is extracted by Hilbert transform. Establish the mapping relationship between window length L and instantaneous frequency change rate df(t) / dt: L(t)=Lmin+(Lmax-Lmin)×exp(-α×|df(t) / dt|), where α is the adjustment factor, and Lmin and Lmax are the preset minimum and maximum window lengths, respectively.

5. The method for extracting transformer vibration features based on multi-measurement point collaboration and time-frequency matrix reconstruction according to claim 2, 3, or 4, characterized in that, The window function is a Hamming window, a Hanning window, or a Kaiser window.

6. The method for extracting transformer vibration features based on multi-measurement point collaboration and time-frequency matrix reconstruction according to any one of claims 1-4, characterized in that, The specific process of step S4 is as follows: S401. Results of the short-time Fourier transform Differentiate the phase of each frequency component along the time direction to extract the instantaneous frequency of the compressed signal. : in, Indicates the instantaneous frequency; arg represents the argument operation. S402. Establish the mapping relationship from the STFT frequency coordinate f to the actual instantaneous frequency coordinate; for each time point t, The energy value at frequency f is redistributed to the instantaneous frequency. At the corresponding new frequency η, the one-dimensional vibration signal is converted into a high-resolution time-frequency image using Fourier synchronous compression transform. The specific formula is as follows: in, This is the time-frequency representation after synchronous compression transformation. For the Dirac function, This represents the compressed frequency variable.

7. The method for extracting transformer vibration features based on multi-measurement point collaboration and time-frequency matrix reconstruction according to any one of claims 1-4, characterized in that, In step S2, the normalization preprocessing adopts the Z-Score normalization method or the Max-min normalization method.

8. A computer program product, comprising a computer program, characterized in that, The computer program is executed by the processor to perform the steps of the method as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program, when run by a processor, performs the steps of the method as described in any one of claims 1-7.

10. A transformer vibration feature extraction system based on multi-measurement point collaboration and time-frequency matrix reconstruction, comprising an interconnected memory and a processor, wherein the memory stores a computer program, characterized in that, The computer program, when run by a processor, performs the steps of the method as described in any one of claims 1-7.

Images