Raman spectrum denoising method and device applied to transformer oil, equipment and medium

Abstract

The invention relates to the field of transformer oil detection, in particular to a Raman spectrum denoising method, device and equipment applied to transformer oil and a medium. Raman spectrum, oil temperature and vibration amplitude of transformer oil are synchronously acquired; correcting the baseline of the Raman spectrum according to the corresponding relation between the oil temperature and the preset temperature and the baseline intensity to obtain a first de-noised spectrum without the temperature noise; performing median filtering on the Raman spectrum according to the vibration amplitude to obtain a second de-noised spectrum without vibration noise; performing wavelet transform denoising on the second denoising spectrum to obtain a third denoising spectrum without random noise; and correcting the third de-noised spectrum according to the peak position of the internal standard substance peak in the third de-noised spectrum and the standard peak position of the internal standard substance peak under the standard condition to obtain a de-noised target Raman spectrum. Therefore, targeted treatment of temperature noise, vibration noise and random noise is realized, and the detection reliability of the Raman spectrum of the transformer oil in a complex environment is improved.

Description

Technical Field

[0001] This invention relates to the field of transformer oil testing, specifically to a Raman spectroscopy noise reduction method, apparatus, equipment, and medium for transformer oil. Background Technology

[0002] Transformer oil plays multiple crucial roles in power transformers, including insulation, cooling, and protection. Its chemical stability and insulation performance directly affect the transformer's operational reliability. Over time, transformer oil gradually ages due to thermal, electrical, and mechanical stresses, as well as material decomposition. For example, aging insulation paper produces characteristic compounds such as furfural, and the oil itself oxidizes and produces deposits. These changes reduce the oil's insulation strength and may lead to faults. Therefore, regular testing of transformer oil, analyzing its physicochemical indicators (such as furfural content and dissolved gases) to assess the transformer's insulation condition, diagnose potential faults, and provide timely warnings, is essential for ensuring the stable and safe operation of the entire power system.

[0003] Raman spectroscopy, as a rapid, non-destructive analytical method that can provide molecular fingerprint information, has shown great potential in the field of transformer oil testing. Its basic principle is that when a laser irradiates an oil sample, the molecules in the oil undergo Raman scattering. By analyzing the change in the frequency of the scattered light relative to the incident light frequency (i.e., the Raman shift), information on molecular vibration and rotation can be obtained, thereby enabling qualitative and quantitative analysis of the chemical components in the oil.

[0004] In transformer oil condition monitoring, to achieve more timely and accurate detection, on-site testing has gradually replaced traditional laboratory sampling and analysis in recent years. However, this shift in testing methods has exposed Raman spectroscopy to more complex environmental interference challenges, affecting the sensitivity and reliability of the detection. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, equipment, and medium for Raman spectroscopy noise reduction applied to transformer oil, thereby solving the problems of the prior art.

[0006] This invention is achieved through the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a Raman spectral noise reduction method for transformer oil, comprising:

[0008] Simultaneously collect the Raman spectrum, oil temperature, and vibration amplitude of the transformer oil;

[0009] Based on the oil temperature and the preset relationship between temperature and baseline intensity, the baseline of the Raman spectrum is corrected to obtain a first denoised spectrum with temperature noise removed.

[0010] The Raman spectrum is subjected to median filtering based on the vibration amplitude to obtain a second denoised spectrum with vibration noise removed.

[0011] The second denoised spectrum is subjected to wavelet transform denoising to obtain the third denoised spectrum with random noise removed;

[0012] Based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions, the third denoised spectrum is corrected to obtain the target Raman spectrum after noise removal.

[0013] Preferably, the step of correcting the baseline of the Raman spectrum based on the oil temperature and a preset correspondence between temperature and baseline intensity to obtain a first denoised spectrum with temperature noise removed includes:

[0014] Based on the oil temperature and the preset correspondence between temperature and baseline intensity, the target baseline intensity corresponding to the oil temperature is obtained;

[0015] The Raman spectrum is fitted based on the target baseline intensity and an asymmetric least squares smoothing algorithm to obtain an estimated baseline that matches the oil temperature;

[0016] The first denoised spectrum, after removing temperature noise, is obtained based on the difference between the Raman spectrum and the estimated baseline.

[0017] Preferably, the step of fitting the Raman spectrum based on the target baseline intensity and an asymmetric least squares smoothing algorithm to obtain an estimated baseline matching the oil temperature includes:

[0018] Baseline initialization is performed based on the target baseline intensity to obtain an initial baseline plane;

[0019] Based on the initial baseline plane, a weight matrix is ​​constructed to obtain an asymmetric weight distribution;

[0020] The intermediate estimated baseline is obtained by performing least squares solution based on the asymmetric weight distribution and smoothing constraint parameters.

[0021] Based on the intermediate estimated baseline, an iterative judgment is made. If convergence is not achieved, the process returns to the weight matrix construction step to continue optimization.

[0022] Based on the final converged result, a baseline is output to obtain an estimated baseline that matches the oil temperature.

[0023] Preferably, the correspondence between the temperature and the baseline intensity is obtained in the following way:

[0024] The preset temperature control environment is heated and kept constant according to the preset temperature gradient to obtain multiple steady-state temperature points.

[0025] Raman spectra of a preset standard transformer oil were collected under a temperature-controlled environment corresponding to each steady-state temperature point to obtain an initial Raman spectrum set corresponding to each steady-state temperature point.

[0026] For each initial Raman spectrum in each initial Raman spectrum set, a specific wavenumber range without Raman characteristic peaks is determined from the initial Raman spectrum;

[0027] The initial baseline intensity is obtained by performing numerical statistics on the spectral signal intensity within the specific wavenumber interval.

[0028] Numerical statistics are performed on all initial baseline intensities corresponding to the initial Raman spectrum set to obtain the baseline intensity corresponding to each steady-state temperature point.

[0029] Curve fitting is performed based on the baseline intensity corresponding to each steady-state temperature point to obtain the correspondence between temperature and baseline intensity.

[0030] Preferably, the step of performing median filtering on the Raman spectrum based on the vibration amplitude to obtain a second denoised spectrum with vibration noise removed includes:

[0031] The vibration intensity is assessed based on the vibration amplitude to obtain the vibration intensity level.

[0032] A dynamic window is obtained by mapping the filtering window based on the vibration intensity level.

[0033] The first denoised spectrum is calculated by sliding window midpoint based on the dynamic window to obtain the filtered signal sequence.

[0034] Boundary data processing is performed based on the filtered signal sequence to obtain a second denoised spectrum with vibration noise removed.

[0035] Preferably, the step of performing wavelet transform denoising on the second denoised spectrum to obtain a third denoised spectrum with random noise removed includes:

[0036] Based on the oil temperature and the preset temperature-threshold correspondence, wavelet threshold calculation is performed to obtain an adaptive threshold;

[0037] The second denoised spectrum is decomposed into multiple scales based on the wavelet basis functions to obtain the wavelet coefficients of each layer.

[0038] The wavelet coefficients of each layer are subjected to soft thresholding based on the adaptive threshold to obtain the denoised wavelet coefficients of each layer.

[0039] The signal is reconstructed based on the denoised wavelet coefficients of each layer and the wavelet basis functions to obtain the initial denoised spectrum.

[0040] The initial denoised spectrum is post-processed and optimized according to the spectral smoothness requirements to obtain a third denoised spectrum with random noise removed.

[0041] Preferably, the step of correcting the third denoised spectrum based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions to obtain the target Raman spectrum after noise removal includes:

[0042] The peak position of the internal standard is obtained by identifying the peak of the internal standard based on the third denoised spectrum.

[0043] The offset is calculated based on the current internal standard peak position and the standard peak position to obtain the peak position drift.

[0044] The spectral data points are shifted according to the peak position shift amount and the preset peak position and wavelength correction relationship to obtain the peak position corrected spectrum;

[0045] The intensity compensation factor is calculated based on the peak position drift and the preset peak position and intensity correction relationship to obtain the intensity correction coefficient;

[0046] The peak-position corrected spectrum is adjusted globally based on the intensity correction coefficient to obtain the target Raman spectrum after noise removal.

[0047] Secondly, embodiments of the present invention provide a Raman spectroscopy noise reduction device for transformer oil, comprising:

[0048] The acquisition module is used to simultaneously acquire the Raman spectrum, oil temperature, and vibration amplitude of transformer oil;

[0049] The first noise reduction module is used to correct the baseline of the Raman spectrum according to the oil temperature and the preset correspondence between temperature and baseline intensity, so as to obtain a first noise reduction spectrum with temperature noise removed.

[0050] The second noise reduction module is used to perform median filtering on the Raman spectrum based on the vibration amplitude to obtain a second noise-reduced spectrum with vibration noise removed.

[0051] The third denoising module is used to perform wavelet transform denoising on the second denoised spectrum to obtain a third denoised spectrum with random noise removed.

[0052] The correction module is used to correct the third denoised spectrum based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions, so as to obtain the target Raman spectrum after noise removal.

[0053] Thirdly, embodiments of the present invention provide an electronic device, including: at least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method of the first aspect described above.

[0054] Fourthly, embodiments of the present invention provide a storage medium storing computer program instructions, which, when executed by a processor, implement the method of the first aspect described above.

[0055] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0056] This method achieves quantitative perception of environmental interference factors by simultaneously acquiring oil temperature and vibration amplitude data. Baseline correction based on the temperature-baseline intensity correspondence effectively suppresses fluorescence background interference caused by temperature changes, solving the problem of insufficient adaptability of traditional fixed-parameter baseline correction methods in oil temperature fluctuation scenarios. Through vibration amplitude adaptive median filtering, this method can dynamically adjust the filtering parameters for different vibration intensities, eliminating vibration-induced impulse noise while better preserving the morphological characteristics of Raman characteristic peaks. Subsequent wavelet transform denoising further eliminates random noise components in the spectrum, improving the signal-to-noise ratio. Finally, through the peak position correction mechanism of the internal standard peak, the systematic spectral peak shift caused by the coupling effect of environmental factors is eliminated, ensuring the accuracy of qualitative identification and quantitative analysis. This staged, multi-level denoising strategy enables the method to effectively cope with complex environmental interference in transformer field testing, providing a more reliable spectral data foundation for transformer oil condition assessment. Compared with traditional single denoising methods, this invention achieves targeted processing of temperature noise, vibration noise and random noise through the synergistic effect of environmental parameters and denoising algorithms. While maintaining the integrity of Raman features, it significantly improves the detection reliability of transformer oil Raman spectra in complex environments. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0058] Figure 1 This is a schematic flowchart of the Raman spectroscopy noise reduction method for transformer oil provided by the present invention;

[0059] Figure 2A schematic diagram of the Raman spectroscopy noise reduction device for transformer oil provided by the present invention;

[0060] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0062] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0063] It should be noted that all actions involving the acquisition of signals, information, or data in this invention are carried out in compliance with the relevant data protection laws and regulations of the locality and with authorization from the owner of the relevant device.

[0064] Example 1

[0065] Please see Figure 1 This invention provides a Raman spectroscopy noise reduction method for transformer oil, comprising:

[0066] S1. Synchronously acquire the Raman spectrum, oil temperature, and vibration amplitude of transformer oil;

[0067] This step aims to simultaneously acquire raw spectral data and key environmental parameters. Raman spectroscopy is a spectrum obtained by collecting the inelastic scattered light signal generated after laser interaction with transformer oil molecules and unfolding it according to wavelength or wavenumber. It can reflect molecular vibrational or rotational energy level information, thus being used for qualitative or quantitative analysis of components such as dissolved gases and aging products in the oil. However, in the actual operating environment of transformers, oil temperature fluctuations can cause baseline drift and peak distortion in the Raman spectrum, while mechanical vibrations transmitted from the transformer itself and the external environment can cause optical path alignment deviations, manifested as random fluctuations in signal intensity and sudden spike noise. Therefore, while acquiring Raman spectra, oil temperature and vibration amplitude are recorded by integrating temperature and vibration sensors, providing a data foundation for subsequent noise separation and correction with clear physical directionality.

[0068] S2. Based on the oil temperature and the preset correspondence between temperature and baseline intensity, the baseline of the Raman spectrum is corrected to obtain a first denoised spectrum with temperature noise removed.

[0069] The core of this step lies in eliminating baseline interference such as fluorescence background caused by temperature changes. Baseline intensity specifically refers to the low-frequency background signal intensity in the Raman spectrum that does not change with Raman shift. It is typically contributed by fluorescence generated by impurities and aging products in transformer oil under laser excitation, as well as blackbody radiation, and its intensity is positively correlated with oil temperature. The preset temperature-baseline intensity correspondence is a quantitative model or lookup table established by measuring the average or median signal intensity of transformer oil in the wavenumber range without characteristic peaks at different temperatures under controlled laboratory conditions, through linear or polynomial fitting. During implementation, the system queries this relationship based on the real-time acquired oil temperature to obtain the target baseline intensity. Guided by this, an asymmetric least squares smoothing algorithm is used to fit the original Raman spectrum. This algorithm iteratively solves for a smooth estimated baseline that matches the current oil temperature by assigning high weights to data points below the current estimated baseline and low weights to Raman characteristic peak data points above it using an asymmetric weighting strategy. Finally, the original spectrum is subtracted from this estimated baseline, resulting in a first denoised spectrum that removes temperature-correlated background noise, providing a cleaner signal platform for subsequent processing.

[0070] In some embodiments, the correspondence between the temperature and the baseline intensity is obtained in the following way:

[0071] The preset temperature control environment is heated and kept constant according to the preset temperature gradient to obtain multiple steady-state temperature points.

[0072] Raman spectra of a preset standard transformer oil were collected under a temperature-controlled environment corresponding to each steady-state temperature point to obtain an initial Raman spectrum set corresponding to each steady-state temperature point.

[0073] For each initial Raman spectrum in each initial Raman spectrum set, a specific wavenumber range without Raman characteristic peaks is determined from the initial Raman spectrum;

[0074] The initial baseline intensity is obtained by performing numerical statistics on the spectral signal intensity within the specific wavenumber interval.

[0075] Numerical statistics are performed on all initial baseline intensities corresponding to the initial Raman spectrum set to obtain the baseline intensity corresponding to each steady-state temperature point.

[0076] Curve fitting is performed based on the baseline intensity corresponding to each steady-state temperature point to obtain the correspondence between temperature and baseline intensity.

[0077] In this context, temperature gradient refers to a series of temperature values ​​divided at certain intervals from the initial temperature to the final temperature; temperature-controlled environment refers to the experimental apparatus capable of precisely controlling the sample temperature; steady-state temperature point refers to the state in which the sample temperature reaches a set value and remains stable; standard transformer oil refers to a reference oil sample with a clearly defined and stable chemical composition; initial Raman spectrum set refers to the collection of multiple raw spectral data acquired at a specific temperature; specific wavenumber interval refers to the background region in the spectrum that does not contain Raman characteristic peaks, usually selected in a band with a high Raman shift and no characteristic peaks; curve fitting refers to using mathematical methods to find a curve expression or functional relationship that best reflects the change between temperature and baseline intensity. In practice, the standard transformer oil sample is first systematically calibrated in a controlled laboratory environment: precise temperature control is performed according to a preset temperature gradient, and Raman spectral datasets are collected at different steady-state temperature points; for each spectrum, a specific wavenumber interval without Raman characteristic peaks is selected, and the statistical value of its signal intensity is calculated as the baseline intensity of that spectrum; then, multiple baseline intensity values ​​at the same temperature are integrated and statistically analyzed to obtain the representative baseline intensity at that temperature point; finally, a continuous correspondence model between temperature and baseline intensity is established through curve fitting (such as linear regression or polynomial fitting). This calibration process ensures the accuracy of baseline intensity measurement and the reliability of the temperature correlation model, enabling the determination of the corresponding target baseline intensity based on the real-time measured oil temperature in field testing, providing accurate amplitude guidance for subsequent asymmetric least squares smoothing algorithms. This solves the problem that traditional fixed-parameter baseline correction methods are difficult to adapt to dynamic changes in oil temperature, improving the reliability of transformer oil Raman spectroscopy detection in complex environments.

[0078] In some embodiments, S2, based on the oil temperature and a preset correspondence between temperature and baseline intensity, the baseline of the Raman spectrum is corrected to obtain a first denoised spectrum with temperature noise removed, including:

[0079] S21. Based on the oil temperature and the preset correspondence between temperature and baseline intensity, obtain the target baseline intensity corresponding to the oil temperature;

[0080] The relationship between temperature and baseline intensity is established through a quantitative model calibrated in the laboratory. This model records the signal intensity response of transformer oil in the region without characteristic peaks at different temperatures. The target baseline intensity is the baseline signal level estimated at the current oil temperature. This relationship is established because baseline interference such as fluorescence background in transformer oil has a clear physical correlation with temperature, and directly using a fixed baseline subtraction method is difficult to adapt to dynamic changes in oil temperature. During implementation, the system queries this preset relationship to convert the real-time oil temperature into a specific expected baseline intensity value. This approach allows the baseline correction process to respond to changes in oil temperature, providing an accurate amplitude reference for subsequent adaptive baseline fitting.

[0081] S22. Fit the Raman spectrum according to the target baseline intensity and the asymmetric least squares smoothing algorithm to obtain an estimated baseline that matches the oil temperature;

[0082] The asymmetric least squares smoothing algorithm is a mathematical method that separates the baseline and characteristic peaks through asymmetric weight allocation. It assigns high weights to data points below the current estimated baseline and low weights to data points above it. This method is used because the baseline in Raman spectroscopy is a low-frequency, slowly varying signal, while the characteristic peaks are sharp, high-frequency signals. A fitting technique is needed to automatically identify and protect the characteristic peaks from baseline absorption. In practice, the target baseline intensity is used as an initial guide, and the optimal smoothing curve is solved iteratively. In each iteration, the weights of the data points are dynamically adjusted based on their position relative to the current estimated baseline. The resulting estimated baseline closely matches the real background without obscuring the Raman characteristic peaks, thus achieving accurate reconstruction of the baseline shape.

[0083] In some embodiments, S22, fitting the Raman spectrum based on the target baseline intensity and an asymmetric least squares smoothing algorithm to obtain an estimated baseline matching the oil temperature includes:

[0084] S221. Initialize the baseline based on the target baseline intensity to obtain the initial baseline plane;

[0085] Baseline initialization refers to the process of constructing an initial baseline sequence with the same number of points as the original spectral data points. Each point in this sequence has a value equal to the target baseline intensity, forming a flat initial baseline plane. Since transformer oil has specific background fluorescence intensity levels at different temperatures, using this as the starting point for iteration allows the baseline fitting process to quickly approach the true background state. In practice, the system generates a horizontal baseline sequence of equal height based on the target baseline intensity value determined by the real-time oil temperature. This approach provides a clear physical starting point for subsequent iterative calculations, avoiding convergence instability issues that may arise from random initialization, and providing an accurate reference benchmark for subsequent weight allocation.

[0086] S222. Construct a weight matrix based on the initial baseline plane to obtain an asymmetric weight distribution;

[0087] The weight matrix construction involves creating a diagonal matrix where the diagonal elements represent the relative importance of each data point in the fitting process. Asymmetric weight distribution specifically refers to a distribution pattern that assigns greater weight to data points below the current baseline and less weight to data points above it. The reason for using asymmetric weight allocation is that data points below the baseline in the Raman spectrum are more likely to represent the true background signal, while data points above the baseline often contain Raman characteristic peak information. In practice, the system compares the signal intensity of each data point with the corresponding value at the initial baseline plane, calculating the corresponding weight value based on their relative position. This allocation mechanism ensures that the baseline fitting process preferentially fits the true background signal region while avoiding interference from Raman characteristic peaks on the baseline shape.

[0088] S223. Solve the intermediate estimated baseline by least squares based on the asymmetric weight distribution and smoothing constraint parameters.

[0089] The least squares approach involves finding the optimal baseline estimate by minimizing the sum of the weighted sum of squared residuals and the smoothing constraint term. The smoothing constraint parameter is a regularization parameter that controls the smoothness of the baseline, balancing goodness of fit and baseline smoothness. The weighted least squares method is used to maintain the physical plausibility of the baseline while accurately fitting the background signal; that is, the baseline should be a smooth, continuous low-frequency signal. In implementation, the system uses an asymmetric weight distribution as the weighting matrix, combines it with the smoothing constraint parameter to construct an optimization problem, and obtains a new baseline estimate by solving a system of linear equations. This calculation process effectively resists the interference of Raman peaks while preserving the low-frequency characteristics of the baseline, producing a more reasonable intermediate baseline estimate.

[0090] S224. Perform iterative judgment based on the intermediate estimated baseline. If convergence is not achieved, return to the weight matrix construction step to continue optimization.

[0091] The iterative judgment step refers to the process of determining whether the algorithm should terminate by comparing the difference between the baseline estimates obtained from two adjacent iterations to see if it is less than a preset threshold. Since baseline fitting needs to reach a stable equilibrium, it is necessary to avoid both premature termination leading to underfitting and excessive iteration resulting in wasted computational resources. In practice, the system calculates the root mean square error between the current intermediate baseline estimate and the previous baseline estimate. When this error value is less than a set threshold, convergence is determined; otherwise, the current intermediate baseline estimate is returned as the new baseline plane to the weight matrix construction step. This iterative mechanism ensures that the baseline estimate can be gradually optimized to a stable state.

[0092] S225. Based on the final converged result, output the baseline to obtain the estimated baseline that matches the oil temperature.

[0093] Baseline output refers to the process of providing the final baseline estimate after iterative convergence as the algorithm result to subsequent processing steps. The baseline estimate, after sufficient iterative optimization, has reached the best-fit state and can accurately reflect the background characteristics at the current temperature. During implementation, the system outputs the intermediate estimated baseline that meets the convergence condition as the final result. This baseline, optimized through multiple rounds, maintains physical consistency with the target baseline intensity and adapts to local spectral characteristics through an asymmetric weighting mechanism, ultimately forming an estimated baseline that matches the oil temperature and closely conforms to the true spectral background.

[0094] S23. Based on the difference between the Raman spectrum and the estimated baseline, a first denoised spectrum with temperature noise removed is obtained.

[0095] The purpose of this interpolation operation is to remove the estimated background signal from the original spectrum while retaining effective Raman feature information. Temperature-related noise mainly manifests as background interference superimposed on the Raman signal. By subtracting the estimated baseline from the original spectrum, these two types of signals can be effectively separated. In practice, this operation directly removes most of the low-frequency background caused by temperature effects, allowing previously submerged weak Raman features to emerge, while maintaining the original shape and relative intensity of the feature peaks, thus providing a higher-quality spectral data foundation for subsequent processing steps.

[0096] S3. Perform median filtering on the Raman spectrum based on the vibration amplitude to obtain a second denoised spectrum with vibration noise removed;

[0097] This step aims to suppress impulse noise introduced by mechanical vibration. Median filtering is a nonlinear signal processing technique that replaces the value of a point in a signal sequence with the median of all points in its neighborhood, which is particularly effective in removing isolated spikes. Vibration noise is caused by instantaneous slight changes in the relative position of the optical probe and the oil sample due to mechanical vibration, which manifests as abrupt, drastically varying spikes in the spectrum. To achieve adaptive filtering, a mapping relationship between the vibration amplitude level and the median filter window size needs to be established in advance. For example, a larger window is automatically selected when the vibration amplitude is large to enhance the denoising capability. In practice, the system determines the window size based on the real-time vibration amplitude, and then uses this window to slide through the first denoised spectrum. Within each window, the covered spectral intensity data is sorted, and the median is taken as the new output of the window center point. Through this process, sudden spike noise caused by vibration can be effectively smoothed out, while the edge features of Raman spectral peaks are well preserved, resulting in a smoother second denoised spectrum.

[0098] In some embodiments, S3, median filtering of the Raman spectrum based on the vibration amplitude to obtain a second denoised spectrum with vibration noise removed, includes:

[0099] S31. Evaluate the vibration intensity based on the vibration amplitude to obtain the vibration intensity level;

[0100] Vibration intensity assessment refers to the process of feature extraction and classification of the raw signals acquired by vibration sensors. Vibration amplitude is usually expressed as acceleration values ​​(unit: g), and vibration intensity levels are discrete classifications (e.g., low, medium, and high) based on preset thresholds. The consideration for intensity level classification is that vibrations of different intensities cause different interference characteristics to the spectrum, requiring differentiated processing strategies. In practice, the system calculates the root mean square value of the real-time acquired vibration signal as an intensity index and compares it with a preset threshold range to determine the level. This classification process provides an accurate basis for subsequent adaptive filtering to determine the operating condition.

[0101] S32. Map the filter window according to the vibration intensity level to obtain a dynamic window;

[0102] In this system, the filter window mapping refers to establishing a correspondence between vibration intensity levels and the size of the median filter window, while the dynamic window refers to the filter window size adaptively determined based on real-time vibration conditions. A fixed filter window cannot simultaneously meet the denoising requirements under different vibration intensities: a window that is too small cannot effectively suppress strong vibration noise, while a window that is too large may lead to signal distortion. During implementation, the system uses predefined mapping relationships (e.g., low vibration corresponds to window size 3, medium vibration to 5, and high vibration to 7) to convert vibration intensity levels into specific window size parameters. This design allows the filtering process to automatically adjust the processing intensity according to the vibration intensity.

[0103] S33. Calculate the sliding window midpoint of the first denoised spectrum according to the dynamic window to obtain the filtered signal sequence;

[0104] The sliding window median calculation refers to the process of sequentially truncating subsequences along the spectral data sequence and calculating their medians using a dynamic window as its width. Vibration noise typically manifests as isolated spikes in the spectrum, and median filtering effectively suppresses this type of impulse noise while preserving signal edge characteristics. In practice, the system uses a dynamic window as the width of the sliding window, sequentially sorting the values ​​in the neighborhood of each data point in the first denoised spectrum and taking the median. This process effectively eliminates sudden spike interference caused by vibration while preserving the morphological information of Raman characteristic peaks.

[0105] S34. Perform boundary data processing based on the filtered signal sequence to obtain a second denoised spectrum with vibration noise removed.

[0106] Boundary data processing refers to special processing of the spectral regions at both ends that cannot be fully covered by the sliding window, in order to maintain the integrity of the entire spectrum. This addresses the problem that standard sliding windows cannot be centered at the beginning and end of the data sequence, leading to missing boundary points. In practice, methods such as mirror expansion, constant padding, or local window reduction can be used to specially process the boundary regions, ensuring that the entire spectrum receives appropriate filtering. The second denoised spectrum obtained after this step effectively suppresses impulse noise caused by vibration while maintaining the integrity and continuity of the spectral data, providing further improved spectral data for subsequent processing steps.

[0107] S4. Perform wavelet transform denoising on the second denoised spectrum to obtain the third denoised spectrum with random noise removed.

[0108] This step focuses on filtering out residual random noise in the spectrum, such as detector thermal noise and shot noise. Wavelet transform denoising uses wavelet basis functions to decompose the signal into subbands of different scales and frequencies. The effective signal in the Raman spectrum is usually represented by a peak at a specific Raman shift, corresponding to larger coefficients at certain scales after wavelet decomposition, while random noise is widely distributed across all scales and has smaller coefficients. In practice, firstly, a wavelet basis function matching the shape of the Raman peak is selected to decompose the second denoised spectrum into approximate coefficients and detail coefficients at different scales. Then, a threshold is set according to the statistical characteristics of the noise, and the detail coefficients of each layer obtained from the decomposition are processed. For example, a soft thresholding function is used to set coefficients below the threshold to zero, while coefficients above the threshold are shrunk. This process can effectively suppress the wavelet coefficients corresponding to noise. Finally, the processed coefficients are used to reconstruct the signal using inverse wavelet transform. Through the multi-resolution analysis characteristics of wavelet transform, the level of random noise can be reduced while the detail information of Raman characteristic peaks can be well preserved, thus obtaining a third denoised spectrum with a further improved signal-to-noise ratio.

[0109] In some embodiments, S4, wavelet transform is performed on the second denoised spectrum to obtain a third denoised spectrum with random noise removed, including:

[0110] S41. Based on the oil temperature and the preset temperature-threshold correspondence, wavelet threshold calculation is performed to obtain an adaptive threshold;

[0111] The wavelet threshold is the critical value used to distinguish between signal and noise in wavelet denoising, and its magnitude directly affects the denoising effect. The preset temperature-threshold correspondence is a mathematical model established through experimental calibration, reflecting the variation of noise levels at different temperatures. This step addresses the problem that temperature changes in transformer oil can alter the intensity of random noise in the Raman spectrum, making it difficult to achieve ideal denoising results under different operating conditions with a fixed threshold. The system calculates the threshold parameter matching the current temperature based on the preset relationship model obtained from the real-time oil temperature query. This adaptive mechanism allows the threshold setting to automatically adjust with temperature changes, providing a suitable discrimination standard for subsequent wavelet denoising.

[0112] S42. Perform multi-scale decomposition on the second denoised spectrum according to the wavelet basis function to obtain the wavelet coefficients of each layer;

[0113] In this process, wavelet basis functions are the mother wavelets used for signal decomposition, such as Daubechies wavelets or Symlets wavelets. Multi-scale decomposition refers to the process of decomposing the signal into multiple scale spaces according to different frequency bandwidths. Wavelet coefficients include approximation coefficients and detail coefficients, reflecting the low-frequency profile and high-frequency details of the signal, respectively. This step addresses the issue that effective signals and random noise in Raman spectra exhibit different characteristics at different scales; effective signals are typically concentrated at specific scales and have large coefficients, while noise is widely distributed and has small coefficients. During implementation, wavelet basis functions matching the Raman peak shape characteristics are selected, and the second denoised spectrum is decomposed into a specified number of layers to obtain wavelet coefficients arranged from low frequency to high frequency. This decomposition provides a multi-resolution analytical basis for subsequent thresholding processing.

[0114] S43. Perform soft thresholding on the wavelet coefficients of each layer according to the adaptive threshold to obtain the denoised wavelet coefficients of each layer.

[0115] Soft thresholding is a nonlinear filtering method characterized by setting wavelet coefficients with absolute values ​​less than a threshold to zero and shrinking coefficients with values ​​greater than the threshold. This method maintains the overall continuity of the signal and avoids artificial oscillations in the reconstructed signal. In practice, a calculated adaptive threshold is applied to each level of detail coefficients, and zeroing or shrinking is performed based on the relationship between the absolute value of the coefficient and the threshold, while approximate coefficients are usually retained unchanged. This process effectively suppresses wavelet coefficients representing noise while preserving the main coefficients representing the effective signal, thus achieving signal-noise separation.

[0116] S44. Reconstruct the signal based on the denoised wavelet coefficients of each layer and the wavelet basis functions to obtain the initial denoised spectrum;

[0117] Signal reconstruction refers to the process of reconstructing a time-domain signal using processed wavelet coefficients through inverse wavelet transform. The consideration for signal reconstruction is the need to convert the thresholded frequency-domain coefficients back to their original time-domain representation for subsequent analysis and application. In practice, the retained approximation coefficients and the processed detail coefficients at each level are used as input, and the same wavelet basis functions as those used in decomposition are employed for inverse wavelet transform. The resulting initial denoised spectrum has removed most of the random noise, but slight distortion or residual noise may remain in some regions.

[0118] S45. The initial denoised spectrum is post-processed and optimized according to the spectral smoothness requirements to obtain a third denoised spectrum with random noise removed.

[0119] Post-processing optimization refers to fine-tuning the initially denoised spectrum, including slight smoothing or local correction. Since wavelet thresholding may introduce slight Gibbs phenomenon or local distortion while removing noise, further optimization is needed to improve spectral quality. In practice, depending on the application's requirements for spectral smoothness, low-pass filtering or moving average methods can be used to moderately process the initial denoised spectrum, eliminating potential local oscillations while preserving the main morphological characteristics of Raman peaks. The third denoised spectrum obtained after this step further improves spectral smoothness and signal-to-noise ratio while retaining the effective Raman signal.

[0120] S5. Based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions, the third denoised spectrum is corrected to obtain the target Raman spectrum after noise removal.

[0121] This step supplements and refines the aforementioned denoising process, aiming to eliminate systematic peak shifts that may be caused by the coupling effect of environmental factors, ensuring the accuracy of qualitative analysis and quantitative models. The internal standard peak refers to a characteristic peak of a stable component in transformer oil whose Raman characteristic peaks are not significantly affected by oil aging or fault products, such as certain CH stretching vibration peaks in the transformer oil matrix. Peak position refers to the Raman shift value corresponding to this characteristic peak in the Raman spectrum. Standard conditions typically refer to a laboratory-controlled baseline state where environmental interference is minimized; the standard peak position is the accepted or calibrated value of the Raman shift measured under this state for the internal standard peak. During correction, the current peak position of the internal standard peak is first accurately identified and located in the third denoised spectrum, and its drift from the standard peak position is calculated. This drift comprehensively reflects the overall influence of environmental factors such as temperature and pressure on the spectral system. Subsequently, based on this drift, a comprehensive Raman shift translation correction is performed on all data points of the third denoised spectrum. Further compensation for spectral intensity can be made based on a preset peak position-intensity correction relationship. After this step, the target Raman spectrum is obtained, and the peak positions are accurately calibrated, thus laying a reliable foundation for subsequent substance identification based on fixed peak positions and precise quantitative analysis based on peak area or peak height.

[0122] In some embodiments, S5, based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions, the third denoised spectrum is corrected to obtain the target Raman spectrum after noise removal, including:

[0123] S51. Based on the third denoised spectrum, identify the internal standard peak to obtain the current internal standard peak position;

[0124] The internal standard peak refers to a characteristic Raman peak that is stable in transformer oil and is not significantly affected by oil aging or fault products, such as the CH2 bending vibration peak near 1440 cm⁻¹. The current internal standard peak position refers to the Raman shift value of this characteristic peak actually measured in the third denoised spectrum. The purpose of this step is to find a reliable spectral reference standard to assess the systematic shift caused by environmental factors. In practice, by finding local maxima within the expected wavenumber range of the third denoised spectrum and using curve fitting methods to accurately locate the peak position coordinates, this step provides an accurate measurement benchmark for subsequent peak drift correction.

[0125] S52. Calculate the offset based on the current internal standard peak position and the standard peak position to obtain the peak position drift.

[0126] The standard peak position refers to the reference value of the internal standard peak position measured under standard laboratory conditions, while the peak position shift refers to the difference in Raman shift between the current measured value and the standard value. This value comprehensively reflects the overall influence of environmental factors such as temperature and pressure on the spectral system. During implementation, the current internal standard peak position is subtracted from the pre-stored standard peak position; the difference obtained is the peak position shift. This quantitative indicator provides an accurate adjustment basis for subsequent spectral calibration.

[0127] S53. Based on the peak position shift amount and the preset peak position and wavelength correction relationship, the spectral data points are shifted to obtain the peak position corrected spectrum.

[0128] The peak position and wavelength correction relationship refers to the established correspondence between the peak position shift and the overall spectral shift, typically expressed as a linear function. Spectral peak shifts caused by environmental factors usually manifest as a systematic shift of the entire spectrum along the wavenumber axis. Based on the calculated peak position shift and the preset correction relationship, the wavenumber values ​​that need adjustment are determined, and all data points in the entire spectrum are correspondingly shifted and compensated along the wavenumber axis. This operation effectively eliminates the systematic wavenumber shift caused by environmental factors.

[0129] S54. Calculate the intensity compensation factor based on the peak position drift and the preset peak position and intensity correction relationship to obtain the intensity correction coefficient;

[0130] The peak position and intensity correction relationship is established experimentally as a correlation model between peak position shift and signal intensity change. The intensity correction coefficient is a multiplier used to adjust the spectral signal intensity. Environmental factors can cause not only peak position shift but also changes in Raman signal intensity, affecting the accuracy of quantitative analysis. Based on the measured peak position shift, a preset correction relationship is consulted to calculate the corresponding intensity correction coefficient, which reflects the degree of signal intensity attenuation or enhancement under the current environmental conditions.

[0131] S55. Perform global intensity adjustment on the peak-position corrected spectrum according to the intensity correction coefficient to obtain the target Raman spectrum after noise removal.

[0132] Global intensity adjustment refers to the process of multiplying the intensity values ​​of all data points in the spectrum by an intensity correction factor. This is necessary to recover signal intensity variations caused by environmental factors and ensure the comparability of spectra acquired under different conditions. Specifically, the intensity value of each data point in the peak-position corrected spectrum is multiplied by the calculated intensity correction factor. This operation compensates for the influence of environmental factors on signal intensity. The target Raman spectrum obtained after this step corrects both the wavenumber axis shift and the intensity variation, providing accurate and reliable spectral data for subsequent qualitative identification and quantitative analysis.

[0133] Example 2

[0134] Please see Figure 2 This invention provides a Raman spectroscopy noise reduction device for transformer oil, comprising:

[0135] Acquisition module 201 is used to synchronously acquire the Raman spectrum, oil temperature, and vibration amplitude of transformer oil;

[0136] The first noise reduction module 202 is used to correct the baseline of the Raman spectrum according to the oil temperature and the preset correspondence between temperature and baseline intensity, so as to obtain a first noise reduction spectrum with temperature noise removed.

[0137] The second noise reduction module 203 is used to perform median filtering on the Raman spectrum according to the vibration amplitude to obtain a second noise-reduced spectrum with vibration noise removed.

[0138] The third denoising module 204 is used to perform wavelet transform denoising on the second denoised spectrum to obtain a third denoised spectrum with random noise removed.

[0139] The correction module 205 is used to correct the third denoised spectrum according to the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions, so as to obtain the target Raman spectrum after noise removal.

[0140] It should be noted that each module and unit in the Raman spectroscopy denoising device for transformer oil in this embodiment corresponds one-to-one with each step in the Raman spectroscopy denoising method for transformer oil in the aforementioned embodiment. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned Raman spectroscopy denoising method for transformer oil, and will not be repeated here.

[0141] Example 3

[0142] Please see Figure 3 This embodiment provides an electronic device, including at least one processor 301 and a memory 302. Optionally, the device further includes a communication component 303. The processor 301, memory 302, and communication component 303 are connected via a bus 304.

[0143] In a specific implementation, at least one processor 301 executes computer execution instructions stored in memory 302, causing at least one processor 301 to perform the above-described method.

[0144] The specific implementation process of processor 301 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0145] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0146] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0147] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0148] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0149] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0150] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0151] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0152] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0153] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0154] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0155] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0156] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0157] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A Raman spectroscopy noise reduction method for transformer oil, characterized in that, include: Simultaneously collect the Raman spectrum, oil temperature, and vibration amplitude of the transformer oil; Based on the oil temperature and the preset relationship between temperature and baseline intensity, the baseline of the Raman spectrum is corrected to obtain a first denoised spectrum with temperature noise removed. The Raman spectrum is subjected to median filtering based on the vibration amplitude to obtain a second denoised spectrum with vibration noise removed. The second denoised spectrum is subjected to wavelet transform denoising to obtain the third denoised spectrum with random noise removed; Based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions, the third denoised spectrum is corrected to obtain the target Raman spectrum after noise removal.

2. The method according to claim 1, characterized in that, The step of correcting the baseline of the Raman spectrum based on the oil temperature and a preset relationship between temperature and baseline intensity to obtain a first denoised spectrum with temperature noise removed includes: Based on the oil temperature and the preset correspondence between temperature and baseline intensity, the target baseline intensity corresponding to the oil temperature is obtained; The Raman spectrum is fitted based on the target baseline intensity and an asymmetric least squares smoothing algorithm to obtain an estimated baseline that matches the oil temperature; The first denoised spectrum, after removing temperature noise, is obtained based on the difference between the Raman spectrum and the estimated baseline.

3. The method according to claim 2, characterized in that, The step of fitting the Raman spectrum based on the target baseline intensity and an asymmetric least squares smoothing algorithm to obtain an estimated baseline matching the oil temperature includes: Baseline initialization is performed based on the target baseline intensity to obtain an initial baseline plane; Based on the initial baseline plane, a weight matrix is ​​constructed to obtain an asymmetric weight distribution; The intermediate estimated baseline is obtained by performing least squares solution based on the asymmetric weight distribution and smoothing constraint parameters. Based on the intermediate estimated baseline, an iterative judgment is made. If convergence is not achieved, the process returns to the weight matrix construction step to continue optimization. Based on the final converged result, a baseline is output to obtain an estimated baseline that matches the oil temperature.

4. The method according to claim 1, characterized in that, The correspondence between temperature and baseline intensity is obtained in the following way: The preset temperature control environment is heated and kept constant according to the preset temperature gradient to obtain multiple steady-state temperature points. Raman spectra of a preset standard transformer oil were collected under a temperature-controlled environment corresponding to each steady-state temperature point to obtain an initial Raman spectrum set corresponding to each steady-state temperature point. For each initial Raman spectrum in each initial Raman spectrum set, a specific wavenumber range without Raman characteristic peaks is determined from the initial Raman spectrum; The initial baseline intensity is obtained by performing numerical statistics on the spectral signal intensity within the specific wavenumber interval. Numerical statistics are performed on all initial baseline intensities corresponding to the initial Raman spectrum set to obtain the baseline intensity corresponding to each steady-state temperature point. Curve fitting is performed based on the baseline intensity corresponding to each steady-state temperature point to obtain the correspondence between temperature and baseline intensity.

5. The method according to claim 1, characterized in that, The step of performing median filtering on the Raman spectrum based on the vibration amplitude to obtain a second denoised spectrum with vibration noise removed includes: The vibration intensity is assessed based on the vibration amplitude to obtain the vibration intensity level. A dynamic window is obtained by mapping the filtering window based on the vibration intensity level. The first denoised spectrum is calculated by sliding window midpoint based on the dynamic window to obtain the filtered signal sequence. Boundary data processing is performed based on the filtered signal sequence to obtain a second denoised spectrum with vibration noise removed.

6. The method according to claim 1, characterized in that, The step of performing wavelet transform denoising on the second denoised spectrum to obtain a third denoised spectrum with random noise removed includes: Based on the oil temperature and the preset temperature-threshold correspondence, wavelet threshold calculation is performed to obtain an adaptive threshold; The second denoised spectrum is decomposed into multiple scales based on the wavelet basis functions to obtain the wavelet coefficients of each layer. The wavelet coefficients of each layer are subjected to soft thresholding based on the adaptive threshold to obtain the denoised wavelet coefficients of each layer. The signal is reconstructed based on the denoised wavelet coefficients of each layer and the wavelet basis functions to obtain the initial denoised spectrum. The initial denoised spectrum is post-processed and optimized according to the spectral smoothness requirements to obtain a third denoised spectrum with random noise removed.

7. The method according to any one of claims 1-6, characterized in that, The step of correcting the third denoised spectrum based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions to obtain the target Raman spectrum after noise removal includes: The peak position of the internal standard is obtained by identifying the peak of the internal standard based on the third denoised spectrum. The offset is calculated based on the current internal standard peak position and the standard peak position to obtain the peak position drift. The spectral data points are shifted according to the peak position shift amount and the preset peak position and wavelength correction relationship to obtain the peak position corrected spectrum; The intensity compensation factor is calculated based on the peak position drift and the preset peak position and intensity correction relationship to obtain the intensity correction coefficient; The peak-position corrected spectrum is adjusted globally based on the intensity correction coefficient to obtain the target Raman spectrum after noise removal.

8. A Raman spectroscopy noise reduction device for transformer oil, characterized in that, include: The acquisition module is used to simultaneously acquire the Raman spectrum, oil temperature, and vibration amplitude of transformer oil; The first noise reduction module is used to correct the baseline of the Raman spectrum according to the oil temperature and the preset correspondence between temperature and baseline intensity, so as to obtain a first noise reduction spectrum with temperature noise removed. The second noise reduction module is used to perform median filtering on the Raman spectrum based on the vibration amplitude to obtain a second noise-reduced spectrum with vibration noise removed. The third denoising module is used to perform wavelet transform denoising on the second denoised spectrum to obtain a third denoised spectrum with random noise removed. The correction module is used to correct the third denoised spectrum based on the peak position of the internal standard peak in the third denoised spectrum and the standard peak position of the internal standard peak under standard conditions, so as to obtain the target Raman spectrum after noise removal.

9. An electronic device, characterized in that, include: At least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, The method as described in any one of claims 1-7 is implemented when the computer program instructions are executed by the processor.

Images