Adaptive noise reduction method and system for distribution transformer

Abstract

The application provides an adaptive noise reduction method and system of a distribution transformer, and belongs to the technical field of noise reduction. The method comprises: comparing and analyzing an original operation voiceprint signal with a dynamic reference voiceprint signal, evaluating a noise suppression coefficient under a current working condition, and constructing an adaptive filtering standard according to the noise suppression coefficient; predicting an evolution direction of a noise spectrum of the distribution transformer based on a change trend of aligned real-time load time sequence data, and preloading candidate filtering parameters adapted to the evolution direction; if it is detected that the current noise suppression coefficient in the adaptive filtering standard is less than a preset suppression threshold, updating the candidate filtering parameters in combination with the dynamic reference voiceprint signal to obtain target filtering parameters; and filtering the original operation voiceprint signal by using the target filtering parameters. The application can realize adaptive suppression of noise of the distribution transformer under the condition of dynamic change of a load.

Description

Technical Field

[0001] This application belongs to the field of noise reduction technology, and more specifically, relates to an adaptive noise reduction method and system for distribution transformers. Background Technology

[0002] As a crucial hub in power systems, the operating noise of distribution transformers is a key information carrier for assessing their health status. Currently, transformer condition monitoring and noise reduction technology based on acoustic signature signals has become a research hotspot. Existing noise reduction methods typically employ filters with fixed parameters to filter transformer operating acoustic signatures, or implement noise suppression after constructing a static reference signal based on historical acoustic signature data. However, the load of distribution transformers exhibits significant temporal dynamics and random fluctuations. Under different load conditions, the magnetic flux density of the transformer core and the electrodynamic force of the windings change accordingly, leading to complex nonlinear drifts in its noise spectrum structure (including energy distribution in each frequency band and spectral centroid). Static filter parameters or fixed reference signals are difficult to track the dynamic migration of the acoustic signature spectrum caused by load changes, resulting in a significant degradation of the noise reduction effect with load fluctuations in actual operation, severely impacting the reliability of subsequent equipment condition assessments based on acoustic signature data. Therefore, how to achieve adaptive noise suppression of distribution transformers under dynamic load conditions is a pressing technical problem to be solved in this field. Summary of the Invention

[0003] The purpose of this application is to provide an adaptive noise reduction method and system for distribution transformers that can adaptively suppress noise under dynamic load conditions. To achieve the above objective, the technical solution provided by this application is as follows: Firstly, an adaptive noise reduction method for distribution transformers is provided, including: A joint feature library of historical acoustic signatures and historical loads of distribution transformers is constructed. Based on the joint feature library, the noise components of distribution transformers under different load conditions are stripped to generate initial reference acoustic signature signals that correspond one-to-one with each load condition. The original operating acoustic fingerprint signal and real-time load time sequence data of the distribution transformer are collected synchronously. The real-time load time sequence data and the original operating acoustic fingerprint signal are spatiotemporally aligned. Based on the aligned real-time load time sequence data, the initial reference acoustic fingerprint signal that matches the current load condition is dynamically corrected to obtain the dynamic reference acoustic fingerprint signal. The original operating acoustic signature signal is compared and analyzed with the dynamic reference acoustic signature signal. The noise suppression coefficient under the current operating condition is evaluated based on the comparison results, and an adaptive filtering standard is constructed based on the noise suppression coefficient. Based on the changing trend of the aligned real-time load time series data, the evolution direction of the noise spectrum of the distribution transformer is predicted, and candidate filter parameters adapted to the evolution direction are preloaded according to the evolution direction. If the current noise suppression coefficient in the adaptive filtering standard is detected to be less than the preset suppression threshold or deviates from the reference suppression range, the candidate filtering parameters are updated in conjunction with the dynamic reference acoustic signal to obtain the target filtering parameters. The original operating acoustic signal is filtered using target filtering parameters to achieve adaptive noise reduction for the distribution transformer.

[0004] Secondly, an adaptive noise reduction system for distribution transformers is provided, comprising: The feature library construction module is used to build a joint feature library of historical acoustic signatures and historical loads of distribution transformers. Based on the joint feature library, the noise components of distribution transformers under different load conditions are stripped to generate initial reference acoustic signature signals that correspond one-to-one with each load condition. The alignment correction module is used to synchronously acquire the original operating acoustic fingerprint signal and real-time load timing data of the distribution transformer, align the real-time load timing data with the original operating acoustic fingerprint signal in time and space, and dynamically correct the initial reference acoustic fingerprint signal that matches the current load condition based on the aligned real-time load timing data to obtain the dynamic reference acoustic fingerprint signal. The comparison and evaluation module is used to compare and analyze the original operating acoustic fingerprint signal with the dynamic reference acoustic fingerprint signal, evaluate the noise suppression coefficient under the current operating condition based on the comparison results, and construct an adaptive filtering standard based on the noise suppression coefficient. The spectrum prediction and preloading module is used to predict the evolution direction of the noise spectrum of the distribution transformer based on the changing trend of the aligned real-time load time series data, and preload candidate filter parameters that are adapted to the evolution direction according to the evolution direction. The parameter iteration update module is used to update the candidate filter parameters in conjunction with the dynamic reference acoustic signal if the current noise suppression coefficient in the adaptive filtering standard is less than the preset suppression threshold or deviates from the reference suppression range, so as to obtain the target filter parameters. The noise reduction and filtering module is used to filter the original operating acoustic signal using target filtering parameters to achieve adaptive noise reduction for the distribution transformer.

[0005] Thirdly, embodiments of this application also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement an adaptive noise reduction method for a distribution transformer provided in any possible implementation of the first aspect.

[0006] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements an adaptive noise reduction method for a distribution transformer provided by any possible implementation of the first aspect.

[0007] The beneficial effects of the technical solution provided in this application are as follows: This application provides an adaptive noise reduction method and system for distribution transformers, which, compared with related technologies, offers the following advantages: This application first generates a clean initial reference voiceprint signal for each operating condition by using a joint feature library of historical voiceprints and historical loads and stripping noise components, thus eliminating the influence of load differences on the reference signal itself. Then, this application simultaneously collects real-time running voiceprint and load data and performs spatiotemporal alignment, ensuring strict consistency between the two types of data in terms of time reference and spatial location. Subsequently, the initial reference is dynamically corrected based on the aligned real-time load data, allowing the reference voiceprint signal to update in real time with load changes, avoiding the distortion and mismatch problem of a fixed reference when the load changes. Based on this, this application compares and analyzes the original voiceprint with the dynamic reference, quantitatively assesses the current noise deviation, and constructs an adaptive filtering standard, providing a clear and controllable target basis for subsequent filter parameter adjustments. In the parameter optimization stage, this application uses the load time-series change trend to predict the direction of noise spectrum evolution and preloads matching candidate filter parameters, enabling them to be quickly activated and updated in conjunction with the dynamic reference when the suppression coefficient is abnormal, greatly shortening the response delay. Finally, the above mechanisms work together to form a closed-loop control that coordinates load prediction and pre-adjustment with real-time voiceprint correction, ensuring that the noise reduction effect is unaffected by load fluctuations. Therefore, the embodiments of this application can achieve adaptive suppression of distribution transformer noise under dynamic load changes. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0009] Figure 1 A flowchart illustrating an adaptive noise reduction method for a distribution transformer provided in an embodiment of this application; Figure 2 A structural block diagram of an adaptive noise reduction system for a distribution transformer provided in an embodiment of this application; Figure 3 A schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0010] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0011] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.” When describing multiple (two or more) items, if the relationship between the multiple items is not explicitly defined, the multiple items can refer to one, several or all of the multiple items. For example, the description of "parameter A includes A1, A2, A3" can be implemented as parameter A includes A1 or A2 or A3, or it can be implemented as parameter A includes at least two of the three items A1, A2 and A3.

[0012] It is understood that in the embodiments of this application, data such as user information are involved. When the embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with relevant laws, regulations and standards.

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0014] This application provides an adaptive noise reduction method for distribution transformers, which can be executed by electronic devices, such as... Figure 1 As shown, the method may include: S101: Construct a joint feature library of historical acoustic signatures and historical loads of distribution transformers. Based on the joint feature library, strip the noise components of distribution transformers under different load conditions to generate initial reference acoustic signature signals that correspond one-to-one with each load condition.

[0015] In this embodiment, the operating acoustic signals and corresponding load data of the distribution transformer under multiple load conditions within a preset historical time period are collected to form historical operating acoustic signals and historical load data, thus forming a joint feature library of historical acoustic signals and historical loads.

[0016] In this embodiment, noise components of distribution transformers under different load conditions are stripped based on a joint feature library to generate initial reference acoustic signature signals corresponding one-to-one with each load condition, including: The historical operating voiceprint signals under various load conditions in the joint feature library are preprocessed by framing and windowing to obtain the time-domain voiceprint signals of each frame. Short-time Fourier transform is performed on the time-domain acoustic signature signal of each frame to obtain the frequency domain amplitude spectrum of each frame. Inter-frame similarity is calculated based on the frequency domain amplitude spectrum of each frame to obtain the amplitude spectrum similarity matrix; Periodic feature analysis is performed on the amplitude spectrum similarity matrix to obtain the repetition period in the voiceprint signal; Modeling based on repetition cycles yields a model of repetitive segments. The frequency domain amplitude spectrum of each frame is compared with the repeating segment model to extract the repeating pattern components in the historical voiceprint signal. A soft time-frequency mask is generated based on repetitive pattern components, and an adaptive thresholding process is applied to the soft time-frequency mask to obtain a working condition noise separation mask. Using a working condition noise separation mask, the steady-state noise component of the historical operating acoustic signal is separated to obtain the load-related steady-state noise component. The steady-state noise component is reconstructed in the time domain to obtain the initial reference acoustic signature signal corresponding to each load condition.

[0017] In this embodiment, based on the historical acoustic signature signal collected at a preset sampling rate, the historical acoustic signature signal is divided into multiple overlapping short-time analysis frames according to a preset frame length. The overlap between frames is controlled by a preset frame shift. The selection principles for the preset frame length and preset frame shift are: the preset frame length should contain at least two power frequency cycles to ensure low-frequency resolution, and the preset frame shift should maintain a certain overlap between adjacent frames to reduce information loss while ensuring time resolution. After framing, the time-domain acoustic signature signal of each frame is multiplied point-by-point with a preset window function. The preset window function is used to gradually attenuate the data at both ends of the frame.

[0018] Perform a discrete Fourier transform on each frame in sequence, calculate the complex spectrum data of each frame at each frequency point, take the modulus value of the complex spectrum data at each frequency point, and obtain the frequency domain amplitude spectrum of each frame.

[0019] Starting with the first and second frames, reference frames and target frames are selected sequentially according to their frame indices, and the similarity between the frequency domain amplitude spectra of each pair of frames is calculated. This similarity can be characterized by the cosine of the angle between the two amplitude spectrum vectors, with a value ranging from 0 to 1. A value close to 1 indicates that the spectral structures of the two frames are highly similar, while a value close to 0 indicates significant differences. The above calculation is performed on all frame pairs, and the results are stored as an amplitude spectrum similarity matrix, where each element of the matrix represents the similarity between the two frames at the corresponding frame index.

[0020] When performing periodic feature analysis on the amplitude spectrum similarity matrix, a search is conducted along the diagonal of the matrix to identify periodic high-similarity paths. Specifically, autocorrelation analysis is performed on each row or column of the matrix to find the delay that causes the autocorrelation coefficient to reach a local maximum; this delay is the candidate repetition period. By statistically analyzing the candidate repetition periods of multiple rows or columns, the candidate value with the highest frequency is selected as the final determined repetition period.

[0021] The frames in the historical voiceprint signal are grouped according to the repetition period. Frames with the same period index are grouped together. The frequency domain amplitude spectrum of all frames in the group is statistically processed on a frequency-by-frequency basis to obtain the statistical model of the frequency domain amplitude spectrum within the repetition period. This statistical model is the repetition segment model.

[0022] Based on the period index of each frame, the similarity between the frequency domain amplitude spectrum of the frame and the repeating segment model corresponding to the period index is calculated. If the similarity is higher than the preset matching threshold, the frame is determined to have a repeating pattern component, and the position and amplitude information of the frame in the time-frequency representation are recorded.

[0023] For time-frequency points that are determined to have repetitive pattern components, their mask values ​​are set to values ​​close to 1. For time-frequency points that are not determined to have repetitive pattern components, their mask values ​​are set to values ​​close to 0, thereby generating a soft time-frequency mask matrix with values ​​ranging from 0 to 1.

[0024] Based on the numerical distribution of each element in the soft time-frequency mask matrix, an adaptive threshold selection algorithm is used to determine the separation threshold. Elements in the soft time-frequency mask that are greater than or equal to the threshold are set to the first preset value, and elements that are less than the threshold are set to the second preset value, thus obtaining the operating condition noise separation mask.

[0025] The frequency domain representation of the historical operating acoustic signal is multiplied with the operating noise separation mask at each time point. The spectral data corresponding to the time-frequency point with the mask value of the first preset value is retained, and the spectral data corresponding to the time-frequency point with the mask value of the second preset value is set to zero, so as to obtain the frequency domain representation of the steady-state noise component related to the load.

[0026] An inverse Fourier transform is performed on the frequency domain representation of the steady-state noise component, converting the frequency domain data of each frame back to the time domain segment frame by frame. Then, adjacent time domain segments are overlapped and added according to the frame shift at the time of framing to obtain the steady-state noise component waveform in the continuous time domain, which is the initial reference acoustic signal.

[0027] This embodiment can automatically extract the steady-state noise components related to the load by extracting periodic features from the historical acoustic signature signals under various load conditions and using adaptive masking separation. It does not require manual labeling or preset noise templates, and has high separation accuracy and strong adaptability, providing an initial benchmark that accurately corresponds to each load condition for subsequent dynamic benchmark correction.

[0028] S102: Synchronously acquire the original operating acoustic fingerprint signal and real-time load timing data of the distribution transformer, align the real-time load timing data with the original operating acoustic fingerprint signal in time and space, and dynamically correct the initial reference acoustic fingerprint signal that matches the current load condition based on the aligned real-time load timing data to obtain the dynamic reference acoustic fingerprint signal.

[0029] In this embodiment, the original operating acoustic signature signal is acquired by the acoustic signature acquisition device at a preset sampling rate, and the real-time load timing data is synchronously acquired by the load monitoring device. The spatiotemporal alignment process includes two dimensions: time alignment and spatial alignment. Specifically, time alignment involves interpolating and resampling the real-time load timing data using the sampling time sequence of the original operating acoustic signature signal as the target time axis, ensuring that the aligned real-time load timing data has the same time resolution as the original operating acoustic signature signal. Spatial alignment involves normalizing the spatial coordinates of the acoustic signature acquisition device and the load monitoring device to a preset standard reference coordinate system, calculating the time delay compensation based on the spatial distance difference between the sound source and the two devices, and using this time delay compensation to correct the time delay of the original operating acoustic signature signal, ensuring that the original operating acoustic signature signal and the real-time load timing data are normalized to the same reference point in the spatial dimension. After spatiotemporal alignment processing, the aligned original operating acoustic signature signal and the aligned real-time load timing data are obtained.

[0030] In this embodiment, based on the aligned real-time load timing data, the initial reference acoustic signature signal matching the current load condition is dynamically corrected to obtain a dynamic reference acoustic signature signal, including: The load deviation is calculated based on the aligned real-time load timing data and the historical load timing data under the reference load condition corresponding to the initial reference acoustic signature signal. The load deviation is normalized to obtain the normalized load deviation value; Based on the normalized load deviation value, the preset amplitude correction function is called to determine the voiceprint amplitude correction coefficient. Based on the normalized load deviation value, the preset frequency correction function is called to determine the voiceprint frequency correction coefficient; The amplitude data of each frequency band of the initial reference acoustic signature signal are corrected based on the acoustic signature amplitude correction coefficient to obtain the frequency domain acoustic signature data after amplitude correction. Based on the frequency correction coefficient, the center frequency data of each frequency band of the frequency domain acoustic fingerprint data after amplitude correction are processed to perform frequency offset correction to obtain the corrected frequency domain acoustic fingerprint data. The corrected frequency domain acoustic data is subjected to inverse Fourier transform to obtain the dynamic reference acoustic signal.

[0031] In this embodiment, it is necessary to first obtain the historical load time-series data under the reference load condition corresponding to the initial reference acoustic signature signal. This historical load time-series data consists of historical load data records corresponding to the initial reference acoustic signature signal in the joint feature library. Both the aligned real-time load time-series data and the historical load time-series data under the reference load condition corresponding to the initial reference acoustic signature signal are time-series data, and the two have a one-to-one corresponding sampling time relationship in the time dimension. Using the same sampling time as the reference, the load value at each sampling time in the aligned real-time load time-series data is calculated by subtracting the load value at the corresponding sampling time in the historical load time-series data to obtain the instantaneous load deviation value at each sampling time. The instantaneous load deviation values ​​at each sampling time are then arithmetically averaged to obtain a comprehensive load deviation as a representative deviation value for the current operating condition. The purpose of this averaging process is to eliminate the uncertainty caused by single-point data fluctuations.

[0032] The load deviation is normalized by dividing the rated load value of the distribution transformer by the rated load value. This normalized load deviation value typically ranges from -1 to +1, where a negative value indicates the current load is lower than the benchmark load, and a positive value indicates the current load is higher than the benchmark load. Other benchmark values ​​can also be used for normalization, such as the average or maximum value of historical load time-series data under the benchmark load condition, as long as the benchmark value allows for comparability of load deviations between different operating conditions.

[0033] When calling the preset amplitude correction function and frequency correction function, the normalized load deviation value is used as the function input variable, and substituted into the amplitude correction function and frequency correction function for calculation. The amplitude correction function and frequency correction function are functions that are determined in advance through experimental calibration or fitting of historical data. For example, in a laboratory environment or in historical field operation data, acoustic fingerprint data of the same distribution transformer under multiple different load conditions are obtained. The curves of the gain coefficient of the acoustic fingerprint amplitude relative to the reference condition under each condition and the curves of the center frequency offset of each frequency band with the load are statistically analyzed. The above curves are fitted into continuous functions, thus obtaining the amplitude correction function and frequency correction function. The output value of the amplitude correction function is the acoustic fingerprint amplitude correction coefficient, and the output value of the frequency correction function is the acoustic fingerprint frequency correction coefficient.

[0034] The initial reference acoustic signature signal has been stored in the system as frequency domain data through short-time Fourier transform. This frequency domain data includes amplitude and phase information for each frequency band. Amplitude correction processing only applies to the amplitude information and does not change the phase information. Specifically, the amplitude data of each frequency band is multiplied by the acoustic signature amplitude correction coefficient to obtain the amplitude-corrected amplitude of each frequency band. If the acoustic signature amplitude correction coefficient is greater than one, the overall amplitude of each frequency band is amplified; if it is less than one, the overall amplitude of each frequency band is reduced. This multiplicative correction method maintains the relative amplitude ratio between frequency bands, which conforms to the physical characteristic of the overall gain change caused by transformer load variations.

[0035] The center frequency data of each frequency band is added to the acoustic frequency correction coefficient to obtain the center frequency of each band after frequency offset correction. If the acoustic frequency correction coefficient is positive, the center frequency of each band shifts towards higher frequencies; if it is negative, it shifts towards lower frequencies. This additive correction method reflects the physical characteristic that load changes cause an overall drift in harmonic frequencies. After frequency offset correction, the amplitude data of each frequency band remains unchanged; only the band position shifts. The frequency offset correction process can be further refined: different frequency bands use different frequency correction coefficients. For example, the offset patterns of the fundamental frequency band and the octave bands may differ. In this case, the frequency correction function outputs a coefficient vector, and each frequency band is independently offset corrected using its corresponding coefficient.

[0036] The order of amplitude correction and frequency offset correction can be interchanged. In this embodiment, amplitude correction is performed first, followed by frequency offset correction, but performing frequency offset correction first and then amplitude correction can achieve the same effect.

[0037] The corrected frequency domain acoustic signature data includes the corrected amplitude information for each frequency band, the original phase information, and the corrected center frequency information for each frequency band. Using this corrected frequency domain acoustic signature data as input, an inverse Fourier transform operation is performed to convert it from a frequency domain representation to a time domain waveform. The resulting time domain waveform is the dynamic reference acoustic signature signal. This dynamic reference acoustic signature signal has been synchronously corrected in both amplitude and frequency dimensions based on the current real-time load, accurately reflecting the reference acoustic signature characteristics under the current load conditions.

[0038] This embodiment uses amplitude and frequency dual-dimensional correction driven by real-time load deviation to enable the initial reference acoustic signature signal to be updated in real time with the actual load changes. This fundamentally avoids the problem of mismatch when the fixed reference changes with the load, and provides a reference benchmark that is highly matched with the current operating conditions for the accurate evaluation of the subsequent noise suppression coefficient.

[0039] S103: Compare and analyze the original operating acoustic signal with the dynamic reference acoustic signal, evaluate the noise suppression coefficient under the current operating condition based on the comparison results, and construct an adaptive filtering standard based on the noise suppression coefficient.

[0040] In this embodiment, the original operating acoustic signature signal is compared and analyzed with the dynamic reference acoustic signature signal. Based on the comparison results, the noise suppression coefficient under the current operating condition is evaluated, and an adaptive filtering standard is constructed based on the noise suppression coefficient, including: Perform a short-time Fourier transform on the original voiceprint signal to obtain the original voiceprint frequency domain data; Short-time Fourier transform is performed on the dynamic reference acoustic signature signal to obtain the reference acoustic signature frequency domain data; The difference calculation is performed on the amplitude of the same frequency point between the original acoustic signature frequency domain data and the reference acoustic signature frequency domain data to obtain the amplitude difference data of each frequency point; Based on the absolute value of the amplitude difference data at each frequency point, the variance of the absolute value of the amplitude difference data is calculated to obtain the spectrum deviation index. The noise suppression coefficient for the current operating condition is obtained by mapping and converting the spectrum deviation index. The desired attenuation in the adaptive filtering standard is determined based on the difference between the noise suppression coefficient and the preset ideal suppression coefficient. The desired attenuation is smoothed by a smoothing filter to obtain the smoothed desired attenuation, and the smoothed desired attenuation is used as the adaptive filtering standard.

[0041] In this embodiment, the original running acoustic signature signal is the time-domain acoustic signature signal obtained in the synchronous acquisition step. The original running acoustic signature signal is divided into frames with a preset frame length, and the frames are overlapped by a preset frame shift. A preset window function is applied to each frame for windowing processing, followed by a Discrete Fourier Transform to obtain the complex spectrum data of each frame at each frequency point. The amplitude of the complex spectrum at each frequency point is taken to obtain the original acoustic signature frequency domain data. This original acoustic signature frequency domain data retains the amplitude and phase information of each frequency point. The amplitude information is used for subsequent difference calculations, while the phase information can be temporarily ignored in subsequent processing. When performing a short-time Fourier Transform on the dynamic reference acoustic signature signal, the same frame length, frame shift, and window function as the original running acoustic signature signal are used to ensure that the two signals maintain consistent time-frequency resolution, allowing each frequency point to correspond accurately.

[0042] Based on the frequency index, the amplitude of the original acoustic signature frequency domain data at each frequency point is subtracted from the amplitude of the reference acoustic signature frequency domain data at the same frequency point to obtain the amplitude difference data for each frequency point. This amplitude difference data is a one-dimensional sequence with a sequence length equal to the total number of frequency points. Each element represents the difference between the original amplitude and the reference amplitude at the corresponding frequency point. The positive or negative sign of the amplitude difference data indicates whether the original amplitude is higher or lower than the reference amplitude, and the absolute value of the amplitude difference data indicates the degree of deviation.

[0043] First, take the absolute value of each element in the amplitude difference data sequence to obtain the amplitude difference absolute value sequence. Then, calculate the average of this amplitude difference absolute value sequence, and then calculate the average of the squares of the differences between each element and the average value. This average is the variance of the amplitude difference data absolute values. The magnitude of this variance is positively correlated with the degree of spectral deviation; the larger the variance, the more dispersed the amplitude differences at each frequency point, and the more serious the overall spectral structure deviates from the reference.

[0044] The noise suppression coefficient is calculated using a pre-defined mapping function, with the spectral deviation index as input. This mapping function can be, for example, a monotonically increasing or monotonically decreasing function. When the definition of the noise suppression coefficient is positively correlated with the spectral deviation index, a larger spectral deviation index results in a larger noise suppression coefficient, indicating a greater need for noise suppression. Conversely, when the definition of the noise suppression coefficient is negatively correlated with the spectral deviation index, the opposite mapping relationship is used. The parameters of this mapping function can be pre-calibrated according to the application scenario during system initialization.

[0045] The preset ideal suppression coefficient is a target value pre-stored by the system. Subtracting the noise suppression coefficient from the preset ideal suppression coefficient results in a negative difference if the noise suppression coefficient is less than the preset ideal suppression coefficient, indicating insufficient suppression and requiring an increase in attenuation. Conversely, a positive difference indicates excessive suppression and requires a decrease in attenuation. The absolute value of this difference, or its sign, determines the direction of attenuation increase or decrease, yielding the desired attenuation amount.

[0046] The expected attenuation calculated for the current frame is weighted and summed with the smoothed expected attenuation for the previous frame to obtain the smoothed expected attenuation for the current frame. This processing method is a first-order low-pass smoothing filter. The value of the smoothing coefficient in the smoothing filter process determines the degree of smoothing. A smoothing coefficient close to 1 results in a significant smoothing effect but a slower response, while a smoothing coefficient close to 0 results in a faster response but a weaker smoothing effect. An appropriate smoothing coefficient can be selected based on the actual application scenario. The smoothed expected attenuation is used as an adaptive filtering standard and is used as the basis for updating candidate filter parameters in subsequent steps.

[0047] This embodiment combines frequency domain difference calculation and variance statistics to obtain a more robust spectral deviation metric than single-frequency point evaluation or simple average evaluation, effectively suppressing the interference of random fluctuations on the evaluation results. Through smoothing filtering of the expected attenuation, the adaptive filtering standard changes smoothly, avoiding frequent oscillations of the filtering parameters under steady-state conditions, thus improving the stability of the system and the consistency of the noise reduction effect.

[0048] S104: Based on the changing trend of the real-time load time series data after alignment, predict the evolution direction of the noise spectrum of the distribution transformer, and preload candidate filter parameters that are adapted to the evolution direction according to the evolution direction.

[0049] In this embodiment, based on the changing trend of the aligned real-time load time-series data, the evolution direction of the noise spectrum of the distribution transformer is predicted, including: Obtain aligned real-time load time series data within a preset sliding time window to obtain window load time series data; Linear fitting is performed on the window load time series data to obtain the load time series fitting line; The load time series fitting line is compared with the load value at the same sampling time in the window load time series data to obtain the fitting residual value at each sampling time. The load time series fitting residual sequence is generated based on the fitting residual value at each sampling time. The continuous load offset characteristic is obtained by summing up the residual values ​​in the load time series fitting residual sequence. Based on the numerical range of the continuous load offset characteristic, the dynamic evolution trend of the load is determined; Query the spectrum change pattern data corresponding to the dynamic evolution trend of the load from the joint feature library, and use the spectrum change pattern data as the evolution direction of the noise spectrum.

[0050] In this embodiment, the length of the preset sliding time window is pre-set, which determines the duration of the load data participating in the trend analysis. The window length is configurable, and its setting must balance the lead time for prediction and the ability to resist interference. If the window is too short, there will be insufficient samples participating in the analysis, the statistical significance of the trend determination will be reduced, and random fluctuations will easily be misjudged as trends; if the window is too long, the sensitivity of the trend determination to recent changes will decrease, and the timeliness of the prediction will be insufficient. The preset sliding time window slides forward along the time axis with a preset step size. After each slide, the aligned real-time load time series data of the most recent period before the current time in the window is retrieved to obtain the window load time series data.

[0051] Using the sampling time as the independent variable and the load value as the dependent variable, a linear regression is performed on the data points within the window using the least squares criterion. The line parameters that minimize the sum of the squared vertical distances from each data point to the fitted line are determined. These parameters include the slope and the intercept. The line determined by these parameters is the load time-series fitted line. The slope of this fitted line represents the average rate of change of the load within the window period. A positive slope indicates an overall upward trend in the load, a negative slope indicates an overall downward trend, and a slope close to zero indicates a stable overall load.

[0052] Using the sampling time as an index, the actual load value at each sampling time in the window load time series data is subtracted from the corresponding fitted value on the load time series fitting line, yielding the fitting residual value for that sampling time. This difference reflects the direction and degree of deviation of the actual load from the fitted line. The fitting residual values ​​at each sampling time are arranged in chronological order to generate a load time series fitting residual sequence.

[0053] Starting from the first sampling time within the window, the fitted residual values ​​at each sampling time are summed up to obtain the continuous load offset characteristic. This characteristic reflects the cumulative degree of unidirectional load offset relative to the fitted line within the window period. When the load continuously deviates from the fitted line in the same direction, the residual values ​​at each sampling time have the same sign, and the absolute value of the cumulative sum gradually increases; when the load fluctuates randomly around the fitted line, the positive and negative residuals cancel each other out, and the absolute value of the cumulative sum remains small.

[0054] The judgment threshold range is based on a preset range. This range includes a first judgment threshold and a second judgment threshold, where the first judgment threshold is positive and the second judgment threshold is negative. When the continuous load offset characteristic is greater than the first judgment threshold, the dynamic trend of the load is judged to be an upward trend; when the continuous load offset characteristic is less than the second judgment threshold, the dynamic trend of the load is judged to be a downward trend; when the continuous load offset characteristic is greater than or equal to the second judgment threshold and less than or equal to the first judgment threshold, the dynamic trend of the load is judged to be a stable trend.

[0055] The joint feature library pre-stores the correspondence between various trend types and spectral change pattern data. Using the type of load dynamic evolution trend as the query index, records matching the trend type are retrieved from the joint feature library, and the spectral change pattern data in the records are extracted and used as the evolution direction of the currently predicted noise spectrum.

[0056] This embodiment uses linear fitting and residual accumulation analysis within a sliding window to reliably extract the true evolution trend of the load from load data with random fluctuations, effectively suppressing the interference of instantaneous fluctuations on trend determination. By mapping the load evolution trend to the direction of spectral evolution, the filtering parameters can be preloaded before the actual change of the acoustic spectrum, shortening the system response time and improving the noise reduction effect in scenarios with rapid load changes.

[0057] S105: If the current noise suppression coefficient in the adaptive filtering standard is detected to be less than the preset suppression threshold or deviates from the reference suppression range, the candidate filtering parameters are updated in conjunction with the dynamic reference acoustic signal to obtain the target filtering parameters.

[0058] In this embodiment, the candidate filter parameters are updated by combining the dynamic reference acoustic signal to obtain the target filter parameters, including: The original voiceprint signal is divided into frequency bands to obtain the original voiceprint sub-band signals of each frequency band. The dynamic reference acoustic signature signal is divided into frequency bands to obtain the reference acoustic signature sub-band signal for each frequency band. The difference between the original acoustic pattern subband signal and the reference acoustic pattern subband signal in the same frequency band is calculated to obtain the error subband signal of that frequency band. The gradient vector of the frequency band is obtained by performing gradient calculation on the error subband signal and the original acoustic pattern subband signal. Based on the gradient vector and the preset step size factor, the filter coefficients corresponding to the frequency band in the candidate filter parameters are iteratively updated. When the preset iteration termination condition is met, the updated filter coefficients of the frequency band are obtained. The target filter parameters are generated based on the updated filter coefficients for each frequency band.

[0059] In this embodiment, when the current noise suppression coefficient in the adaptive filtering standard is detected to be less than the preset suppression threshold or deviates from the reference suppression range, the parameter update process is triggered. Frequency band division can be implemented using a filter group approach, that is, using a group of bandpass filters to decompose the signal into subbands. The passband range of each bandpass filter covers a preset sub-frequency band interval, and adjacent sub-frequency bands are separated by a preset guard interval to prevent inter-band crosstalk. The original running acoustic signature signal is processed in this way to obtain the original acoustic signature sub-band signal for each frequency band; the dynamic reference acoustic signature signal is processed in the same way to obtain the reference acoustic signature sub-band signal for each frequency band. It should be noted that frequency band division can also be implemented using a time-frequency transformation approach, that is, after a short-time Fourier transform, adjacent frequency points are merged into multiple frequency bands according to frequency intervals, and the sub-band signal of each frequency band is composed of the frequency domain data of all frequency points within that frequency band.

[0060] Based on the sampling time, the amplitude of the original acoustic pattern subband signal at each sampling time is subtracted point by point from the amplitude of the reference acoustic pattern subband signal at the same sampling time to obtain the error subband signal of that frequency band.

[0061] The instantaneous gradient estimate is obtained by multiplying the sampled value of the error subband signal at the current moment with the sampled value of the original acoustic waveform subband signal at the current moment. The sign and magnitude of this instantaneous gradient estimate reflect the direction and magnitude of the filter coefficient adjustment. Time smoothing is then applied to the instantaneous gradient estimate by weighted summing the current instantaneous gradient estimate with the smoothed gradient estimate from the previous moment. This smoothing process helps suppress random fluctuations in the gradient estimation.

[0062] The filter coefficients corresponding to the frequency band in the candidate filter parameters are used as the initial values ​​for iteration. During each iteration, the current filter coefficient is subtracted from the product of the step size factor and the smoothed gradient estimate to obtain the updated filter coefficients. This update rule adjusts the filter coefficients in the opposite direction of the gradient vector, thus indicating the direction in which the signal energy of the error subband decreases.

[0063] The preset iteration termination conditions are: the number of iterations reaches the preset maximum number of iterations, or the absolute value of the difference between the filter coefficients obtained from two adjacent iterations is less than the preset convergence threshold. During the iteration update process, when either of the above two conditions is met, the iteration stops, and the filter coefficients at the current moment are used as the updated filter coefficients for that frequency band.

[0064] After each frequency band independently completes the above iterative update, the updated filter coefficients of each frequency band are arranged and combined according to the frequency band division order to obtain the target filter parameters covering the entire frequency band.

[0065] This embodiment updates the filter coefficients of each frequency band independently by frequency band, enabling each frequency band to independently adapt to the noise spectrum distribution characteristics under the current operating conditions. This avoids the excessive dominance of the high-energy frequency band on the update direction in the joint update of the entire frequency band. By using the candidate filter parameters as the starting point of the iteration, the time required for iteration convergence is significantly shortened, allowing the filter parameters to respond quickly after changes in operating conditions. The combination of candidate filter parameter preloading and current step iteration update realizes adaptive parameter optimization that coordinates load prediction and pre-adjustment with real-time soundprint correction.

[0066] S106: The original operating acoustic signal is filtered using the target filtering parameters to achieve adaptive noise reduction of the distribution transformer.

[0067] In this embodiment, the original running voiceprint signal is filtered using target filtering parameters, including: The original voiceprint signal is split into multiple frequency bands to obtain multiple original voiceprint sub-band signals. Based on the target filtering parameters, adaptive weighted filtering is performed on each original acoustic subband signal to obtain multi-channel noise-reduced subband signals. Frequency band fusion and reconstruction are performed on the multi-channel denoised sub-band signals to obtain a global denoised ripple signal; The global noise reduction ripple signal is used as the target noise reduction output signal.

[0068] In this embodiment, the same frequency band division method as the frequency band division step is used to ensure that the frequency band division method of the original voiceprint sub-band signal is consistent with the frequency band division method in the parameter update step. Frequency band splitting can be implemented using a filter bank approach, that is, using a set of bandpass filters to decompose the original running voiceprint signal into sub-bands, with the passband range of each bandpass filter consistent with the frequency range of each band in the target filtering parameters. Frequency band splitting can also be implemented using a time-frequency transformation approach, that is, after transforming the signal to the frequency domain through a short-time Fourier transform, it is divided into multiple sub-bands according to frequency intervals, and then each sub-band is converted back to the time domain by performing an inverse Fourier transform, resulting in multiple original voiceprint sub-band signals. The sampling rate of each original voiceprint sub-band signal is consistent with the original running voiceprint signal, and each sub-band signal is a time-domain waveform.

[0069] The target filtering parameters include filter coefficients corresponding to each frequency band. These coefficients can be represented as tap coefficients of a finite impulse response (FIR) filter or recursive coefficients of an infinite impulse response (IOR) filter. Digital filters for each frequency band are constructed using these filter coefficients. The original acoustic signature sub-band signals are then input into their respective digital filters for filtering. The output of each digital filter is the noise-reduced sub-band signal. Filtering is performed independently for each frequency band, and the order and coefficients of each filter are different, each tailored to the noise suppression requirements of that frequency band.

[0070] The denoised sub-band signals are superimposed and synthesized in ascending frequency order to obtain a full-band time-domain acoustic signature signal. Band fusion reconstruction and band splitting are inverse operations. If band splitting is implemented using a filter bank, then band fusion reconstruction uses the corresponding synthesis filter bank. If band splitting is implemented using time-frequency transformation, then band fusion reconstruction transforms each time-domain sub-band signal back to the frequency domain, concatenates them in frequency order, and then performs an inverse Fourier transform to obtain the full-band time-domain signal. During fusion reconstruction, gain compensation and time delay compensation are performed on each sub-band signal to eliminate amplitude and phase distortions introduced by the filter bank, ensuring the amplitude accuracy and waveform fidelity of the reconstructed signal.

[0071] Using the global noise reduction ripple signal as the target noise reduction output signal, this signal is the final output ripple signal after being processed through a complete adaptive noise reduction process. It can be used for subsequent applications such as power distribution transformer operation status monitoring, fault diagnosis, and health assessment.

[0072] This embodiment uses frequency band splitting and independent weighted filtering to enable each frequency band to use its own optimal filtering parameters for precise noise reduction, avoiding insufficient or excessive suppression caused by uniform processing of the entire frequency band; through frequency band fusion reconstruction, the integrity and availability of the output signal are guaranteed, providing a reliable acoustic data foundation for subsequent equipment status assessment.

[0073] As can be seen from the above, this embodiment first generates a pure initial reference voiceprint signal under each operating condition by using a joint feature library of historical voiceprints and historical loads and stripping noise components, thus eliminating the influence of load differences on the reference signal itself. Then, this embodiment synchronously collects real-time running voiceprint and load data and performs spatiotemporal alignment, ensuring strict consistency between the two types of data in terms of time reference and spatial location. Subsequently, the initial reference is dynamically corrected based on the aligned real-time load data, enabling the reference voiceprint signal to update in real time with load changes, avoiding the distortion and mismatch problem of a fixed reference when the load changes. Based on this, this embodiment compares and analyzes the original voiceprint with the dynamic reference, quantitatively assesses the current noise deviation, and constructs an adaptive filtering standard, providing a clear and controllable target basis for subsequent filter parameter adjustments. In the parameter optimization stage, this embodiment uses the load time-series change trend to predict the direction of noise spectrum evolution, pre-loads matching candidate filter parameters, and enables them to be quickly activated and updated in conjunction with the dynamic reference when the suppression coefficient is abnormal, greatly shortening the response delay. Ultimately, the aforementioned mechanisms work together to form a closed-loop control system that combines load prediction and pre-adjustment with real-time acoustic signature correction, ensuring that the noise reduction effect is unaffected by load fluctuations. Therefore, the embodiments of this application can achieve adaptive suppression of distribution transformer noise under dynamic load conditions.

[0074] Based on the same principle as the adaptive noise reduction method for distribution transformers provided in the embodiments of this application, the embodiments of this application also provide an adaptive noise reduction system for distribution transformers, such as... Figure 2 As shown, the adaptive noise reduction system 20 for a distribution transformer may specifically include: a feature library construction module 21, an alignment correction module 22, a comparison and evaluation module 23, a spectrum prediction and preloading module 24, a parameter iterative update module 25, and a noise reduction filtering module 26.

[0075] Among them, the feature library construction module 21 is used to construct a joint feature library of historical acoustic text and historical load of distribution transformers. Based on the joint feature library, the noise components of distribution transformers under different load conditions are stripped to generate an initial reference acoustic text signal that corresponds one-to-one with each load condition. The alignment correction module 22 is used to synchronously collect the original operating acoustic fingerprint signal and real-time load timing data of the distribution transformer, align the real-time load timing data with the original operating acoustic fingerprint signal in time and space, and dynamically correct the initial reference acoustic fingerprint signal that matches the current load condition based on the aligned real-time load timing data to obtain the dynamic reference acoustic fingerprint signal. The comparison and evaluation module 23 is used to compare and analyze the original operating acoustic signal with the dynamic reference acoustic signal, evaluate the noise suppression coefficient under the current operating condition based on the comparison results, and construct an adaptive filtering standard based on the noise suppression coefficient. The spectrum prediction and preloading module 24 is used to predict the evolution direction of the noise spectrum of the distribution transformer based on the changing trend of the aligned real-time load time series data, and preload candidate filter parameters that are adapted to the evolution direction according to the evolution direction. The parameter iteration update module 25 is used to update the candidate filter parameters in conjunction with the dynamic reference acoustic signal if the current noise suppression coefficient in the adaptive filtering standard is less than the preset suppression threshold or deviates from the reference suppression range, so as to obtain the target filter parameters. The noise reduction filtering module 26 is used to filter the original operating acoustic signal using target filtering parameters to achieve adaptive noise reduction of the distribution transformer.

[0076] In one embodiment of this application, the feature library construction module 21 is specifically used for: The historical operating voiceprint signals under various load conditions in the joint feature library are preprocessed by framing and windowing to obtain the time-domain voiceprint signals of each frame. Short-time Fourier transform is performed on the time-domain acoustic signature signal of each frame to obtain the frequency domain amplitude spectrum of each frame. Inter-frame similarity is calculated based on the frequency domain amplitude spectrum of each frame to obtain the amplitude spectrum similarity matrix; Periodic feature analysis is performed on the amplitude spectrum similarity matrix to obtain the repetition period in the voiceprint signal; Modeling based on repetition cycles yields a model of repetitive segments. The frequency domain amplitude spectrum of each frame is compared with the repeating segment model to extract the repeating pattern components in the historical voiceprint signal. A soft time-frequency mask is generated based on repetitive pattern components, and an adaptive thresholding process is applied to the soft time-frequency mask to obtain a working condition noise separation mask. Using a working condition noise separation mask, the steady-state noise component of the historical operating acoustic signal is separated to obtain the load-related steady-state noise component. The steady-state noise component is reconstructed in the time domain to obtain the initial reference acoustic signature signal corresponding to each load condition.

[0077] In one embodiment of this application, the alignment correction module 22 is specifically used for: The load deviation is calculated based on the aligned real-time load timing data and the historical load timing data under the reference load condition corresponding to the initial reference acoustic signature signal. The load deviation is normalized to obtain the normalized load deviation value; Based on the normalized load deviation value, the preset amplitude correction function is called to determine the voiceprint amplitude correction coefficient. Based on the normalized load deviation value, the preset frequency correction function is called to determine the voiceprint frequency correction coefficient; The amplitude data of each frequency band of the initial reference acoustic signature signal are corrected based on the acoustic signature amplitude correction coefficient to obtain the frequency domain acoustic signature data after amplitude correction. Based on the frequency correction coefficient, the center frequency data of each frequency band of the frequency domain acoustic fingerprint data after amplitude correction are processed to perform frequency offset correction to obtain the corrected frequency domain acoustic fingerprint data. The corrected frequency domain acoustic data is subjected to inverse Fourier transform to obtain the dynamic reference acoustic signal.

[0078] In one embodiment of this application, the comparison and evaluation module 23 is specifically used for: Perform a short-time Fourier transform on the original voiceprint signal to obtain the original voiceprint frequency domain data; Short-time Fourier transform is performed on the dynamic reference acoustic signature signal to obtain the reference acoustic signature frequency domain data; The difference calculation is performed on the amplitude of the same frequency point between the original acoustic signature frequency domain data and the reference acoustic signature frequency domain data to obtain the amplitude difference data of each frequency point; Based on the absolute value of the amplitude difference data at each frequency point, the variance of the absolute value of the amplitude difference data is calculated to obtain the spectrum deviation index. The noise suppression coefficient for the current operating condition is obtained by mapping and converting the spectrum deviation index. The desired attenuation in the adaptive filtering standard is determined based on the difference between the noise suppression coefficient and the preset ideal suppression coefficient. The desired attenuation is smoothed by a smoothing filter to obtain the smoothed desired attenuation, and the smoothed desired attenuation is used as the adaptive filtering standard.

[0079] In one embodiment of this application, the spectrum prediction preloading module 24 is specifically used for: Obtain aligned real-time load time series data within a preset sliding time window to obtain window load time series data; Linear fitting is performed on the window load time series data to obtain the load time series fitting line; The load time series fitting line is compared with the load value at the same sampling time in the window load time series data to obtain the fitting residual value at each sampling time. The load time series fitting residual sequence is generated based on the fitting residual value at each sampling time. The continuous load offset characteristic is obtained by summing up the residual values ​​in the load time series fitting residual sequence. Based on the numerical range of the continuous load offset characteristic, the dynamic evolution trend of the load is determined; Query the spectrum change pattern data corresponding to the dynamic evolution trend of the load from the joint feature library, and use the spectrum change pattern data as the evolution direction of the noise spectrum.

[0080] In one embodiment of this application, the parameter iteration update module 25 is specifically used for: The original voiceprint signal is divided into frequency bands to obtain the original voiceprint sub-band signals of each frequency band. The dynamic reference acoustic signature signal is divided into frequency bands to obtain the reference acoustic signature sub-band signal for each frequency band. The difference between the original acoustic pattern subband signal and the reference acoustic pattern subband signal in the same frequency band is calculated to obtain the error subband signal of that frequency band. The gradient vector of the frequency band is obtained by performing gradient calculation on the error subband signal and the original acoustic pattern subband signal. Based on the gradient vector and the preset step size factor, the filter coefficients corresponding to the frequency band in the candidate filter parameters are iteratively updated. When the preset iteration termination condition is met, the updated filter coefficients of the frequency band are obtained. The target filter parameters are generated based on the updated filter coefficients for each frequency band.

[0081] In one embodiment of this application, the noise reduction filtering module 26 is specifically used for: The original voiceprint signal is split into multiple frequency bands to obtain multiple original voiceprint sub-band signals. Based on the target filtering parameters, adaptive weighted filtering is performed on each original acoustic subband signal to obtain multi-channel noise-reduced subband signals. Frequency band fusion and reconstruction are performed on the multi-channel denoised sub-band signals to obtain a global denoised ripple signal; The global noise reduction ripple signal is used as the target noise reduction output signal.

[0082] The apparatus in this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0083] Figure 3 A schematic diagram of the structure of an electronic device to which this application embodiment applies is shown, such as... Figure 3 As shown, the electronic device can be used to implement the methods provided in any embodiment of this application.

[0084] like Figure 3 As shown, the electronic device 300 may primarily include at least one processor 301. Figure 3 The diagram shows components such as a memory 302, a communication module 303, and an input / output interface 304. Optionally, these components can be connected and communicate with each other via a bus 305. It should be noted that... Figure 3 The structure of the electronic device 300 shown is merely illustrative and does not constitute a limitation on the electronic devices to which the methods provided in the embodiments of this application are applicable.

[0085] The memory 302 can be used to store operating systems and applications, etc. The applications can include computer programs that implement the methods shown in the embodiments of this application when invoked by the processor 301, and can also include programs for implementing other functions or services. The memory 302 can be ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices that can store information and computer programs, or it can be EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.

[0086] Processor 301 is connected to memory 302 via bus 305 and implements corresponding functions by calling the application programs stored in memory 302. Processor 301 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0087] Electronic device 300 can connect to a network via communication module 303 (which may include, but is not limited to, components such as a network interface) to communicate with other devices (such as user terminals or servers) through the network and achieve data interaction, such as sending data to or receiving data from other devices. Communication module 303 may include wired network interfaces and / or wireless network interfaces, meaning the communication module may include at least one of wired or wireless communication modules.

[0088] The electronic device 300 can connect to necessary input / output devices, such as a keyboard and display device, via the input / output interface 304. The electronic device 300 itself may have a display device, and other external display devices can also be connected via the input / output interface 304. Optionally, a storage device, such as a hard drive, can also be connected via the input / output interface 304 to store data from the electronic device 300, retrieve data from the storage device, or store data from the storage device in the memory 302. It is understood that the input / output interface 304 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 304 can be a component of the electronic device 300 or an external device connected to the electronic device 300 when needed.

[0089] The bus 305 used to connect the components may include a path for transmitting information between the components. The bus 305 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Depending on its function, the bus 305 may be divided into an address bus, a data bus, a control bus, etc.

[0090] Optionally, for the solution provided in the embodiments of this application, the memory 302 can be used to store a computer program that executes the solution of this application, and the processor 301 runs the computer program. When the processor 301 runs the computer program, it implements the operation of the method or apparatus provided in the embodiments of this application.

[0091] Based on the same principle as the method provided in the embodiments of this application, the embodiments of this application provide a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it can implement the corresponding content of the aforementioned method embodiments.

[0092] It should be noted that the terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the figures or text.

[0093] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0094] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0095] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. An adaptive noise reduction method for distribution transformers, characterized in that, include: A joint feature library of historical acoustic signatures and historical loads of distribution transformers is constructed. Based on the joint feature library, the noise components of distribution transformers under different load conditions are stripped to generate initial reference acoustic signature signals that correspond one-to-one with each load condition. The original operating acoustic fingerprint signal and real-time load time sequence data of the distribution transformer are collected synchronously. The real-time load time sequence data and the original operating acoustic fingerprint signal are spatiotemporally aligned. Based on the aligned real-time load time sequence data, the initial reference acoustic fingerprint signal that matches the current load condition is dynamically corrected to obtain the dynamic reference acoustic fingerprint signal. The original operating acoustic signature signal is compared and analyzed with the dynamic reference acoustic signature signal. The noise suppression coefficient under the current operating condition is evaluated based on the comparison results, and an adaptive filtering standard is constructed based on the noise suppression coefficient. Based on the changing trend of the aligned real-time load time-series data, the evolution direction of the noise spectrum of the distribution transformer is predicted, and candidate filter parameters adapted to the evolution direction are preloaded according to the evolution direction. If the current noise suppression coefficient in the adaptive filtering standard is detected to be less than the preset suppression threshold or deviates from the reference suppression range, the candidate filtering parameters are updated in conjunction with the dynamic reference acoustic signal to obtain the target filtering parameters. The original operating acoustic signal is filtered using the target filtering parameters to achieve adaptive noise reduction for the distribution transformer.

2. The adaptive noise reduction method for a distribution transformer as described in claim 1, characterized in that, The step of stripping noise components from distribution transformers under different load conditions based on the joint feature library to generate initial reference acoustic signature signals corresponding one-to-one with each load condition includes: The historical operating voiceprint signals under various load conditions in the joint feature library are preprocessed by framing and windowing to obtain the time-domain voiceprint signals of each frame. Short-time Fourier transform is performed on the time-domain acoustic signature signal of each frame to obtain the frequency domain amplitude spectrum of each frame. Inter-frame similarity is calculated based on the frequency domain amplitude spectrum of each frame to obtain an amplitude spectrum similarity matrix; Periodic feature analysis is performed on the amplitude spectrum similarity matrix to obtain the repetition period in the voiceprint signal; Based on the repetition cycle, a repetitive segment model is obtained; The frequency domain amplitude spectrum of each frame is compared with the repeating segment model to extract the repeating pattern components in the historical voiceprint signal. A soft time-frequency mask is generated based on the repetitive pattern components, and an adaptive thresholding process is applied to the soft time-frequency mask to obtain a working condition noise separation mask. Using the aforementioned operating condition noise separation mask, the historical operating acoustic signal is processed to separate the steady-state noise components, thereby obtaining the load-related steady-state noise components. The steady-state noise component is reconstructed in the time domain to obtain the initial reference acoustic signature signal corresponding to each load condition.

3. The adaptive noise reduction method for a distribution transformer as described in claim 1, characterized in that, The step of dynamically correcting the initial reference acoustic signature signal that matches the current load condition based on the aligned real-time load timing data to obtain the dynamic reference acoustic signature signal includes: Based on the aligned real-time load timing data and the historical load timing data under the reference load condition corresponding to the initial reference acoustic signature signal, the load deviation is calculated. The load deviation is normalized to obtain a normalized load deviation value. Based on the normalized load deviation value, a preset amplitude correction function is invoked to determine the voiceprint amplitude correction coefficient. Based on the normalized load deviation value, a preset frequency correction function is invoked to determine the voiceprint frequency correction coefficient. Based on the aforementioned acoustic signature amplitude correction coefficient, amplitude correction processing is performed on the amplitude data of each frequency band of the initial reference acoustic signature signal to obtain amplitude-corrected frequency domain acoustic signature data. Based on the aforementioned acoustic frequency correction coefficient, the center frequency data of each frequency band of the amplitude-corrected frequency domain acoustic data are subjected to frequency offset correction processing to obtain the corrected frequency domain acoustic data. The corrected frequency domain acoustic data is subjected to inverse Fourier transform to obtain the dynamic reference acoustic signal.

4. The adaptive noise reduction method for a distribution transformer as described in claim 1, characterized in that, The step of comparing and analyzing the original operating acoustic signature signal with the dynamic reference acoustic signature signal, evaluating the noise suppression coefficient under the current operating condition based on the comparison results, and constructing an adaptive filtering standard based on the noise suppression coefficient includes: The original operating voiceprint signal is subjected to a short-time Fourier transform to obtain the original voiceprint frequency domain data; The dynamic reference acoustic signature signal is subjected to a short-time Fourier transform to obtain the reference acoustic signature frequency domain data. The difference calculation is performed on the amplitude of the same frequency point between the original acoustic signature frequency domain data and the reference acoustic signature frequency domain data to obtain the amplitude difference data of each frequency point; Based on the absolute value of the amplitude difference data at each frequency point, the variance of the absolute value of the amplitude difference data is calculated to obtain the spectrum deviation index. The frequency deviation index is mapped and converted to obtain the noise suppression coefficient for the current operating condition. Based on the difference between the noise suppression coefficient and the preset ideal suppression coefficient, the expected attenuation in the adaptive filtering standard is determined; The desired attenuation is smoothed by a smoothing filter to obtain a smoothed desired attenuation, and the smoothed desired attenuation is used as the adaptive filtering standard.

5. The adaptive noise reduction method for a distribution transformer as described in claim 1, characterized in that, The prediction of the evolution direction of the noise spectrum of the distribution transformer based on the changing trend of the aligned real-time load time-series data includes: Obtain aligned real-time load time series data within a preset sliding time window to obtain window load time series data; The window load time series data is subjected to linear fitting to obtain a load time series fitting line; The load time series fitting line is compared with the load value at the same sampling time in the window load time series data to obtain the fitting residual value at each sampling time. The load time series fitting residual sequence is generated based on the fitting residual value at each sampling time. The load continuous offset feature is obtained by summing up the residual values ​​in the load time series fitting residual sequence. Based on the numerical range of the continuous load offset characteristic, the dynamic evolution trend of the load is determined; The spectrum change pattern data corresponding to the dynamic evolution trend of the load is queried from the joint feature library, and the spectrum change pattern data is used as the evolution direction of the noise spectrum.

6. The adaptive noise reduction method for a distribution transformer as described in claim 1, characterized in that, The process of updating the candidate filter parameters by combining the dynamic reference acoustic signal to obtain the target filter parameters includes: The original operating voiceprint signal is divided into frequency bands to obtain the original voiceprint sub-band signals of each frequency band. The dynamic reference acoustic signature signal is divided into frequency bands to obtain the reference acoustic signature sub-band signal for each frequency band. The difference between the original acoustic pattern subband signal and the reference acoustic pattern subband signal in the same frequency band is calculated to obtain the error subband signal of that frequency band. The gradient vector of the frequency band is obtained by performing gradient calculation on the error subband signal and the original acoustic pattern subband signal. Based on the gradient vector and the preset step size factor, the filter coefficients corresponding to the frequency band in the candidate filter parameters are iteratively updated. When the preset iteration termination condition is met, the updated filter coefficients of the frequency band are obtained. The target filter parameters are generated based on the updated filter coefficients for each frequency band.

7. The adaptive noise reduction method for a distribution transformer as described in claim 1, characterized in that, The filtering of the original running voiceprint signal using the target filtering parameters includes: The original operating voiceprint signal is split into multiple frequency bands to obtain multiple original voiceprint sub-band signals; Based on the target filtering parameters, adaptive weighted filtering is performed on each of the original acoustic subband signals to obtain multi-channel noise-reduced subband signals. The frequency band fusion and reconstruction of the multiple denoised sub-band signals are performed to obtain the global denoised ripple signal. The global noise reduction ripple signal is used as the target noise reduction output signal.

8. An adaptive noise reduction system for a distribution transformer, characterized in that, include: The feature library construction module is used to construct a joint feature library of historical acoustic signatures and historical loads of distribution transformers. Based on the joint feature library, the noise components of distribution transformers under different load conditions are stripped to generate an initial reference acoustic signature signal that corresponds one-to-one with each load condition. The alignment correction module is used to synchronously acquire the original operating acoustic fingerprint signal and real-time load timing data of the distribution transformer, align the real-time load timing data with the original operating acoustic fingerprint signal in time and space, and dynamically correct the initial reference acoustic fingerprint signal that matches the current load condition based on the aligned real-time load timing data to obtain a dynamic reference acoustic fingerprint signal. The comparison and evaluation module is used to compare and analyze the original operating voiceprint signal with the dynamic reference voiceprint signal, evaluate the noise suppression coefficient under the current operating condition based on the comparison results, and construct an adaptive filtering standard based on the noise suppression coefficient. The spectrum prediction and preloading module is used to predict the evolution direction of the noise spectrum of the distribution transformer based on the changing trend of the aligned real-time load time series data, and preload candidate filter parameters that are adapted to the evolution direction according to the evolution direction. The parameter iteration update module is used to update the candidate filter parameters in conjunction with the dynamic reference acoustic signal if the current noise suppression coefficient in the adaptive filtering standard is detected to be less than the preset suppression threshold or deviates from the reference suppression range, so as to obtain the target filter parameters. The noise reduction filtering module is used to filter the original operating acoustic signal using the target filtering parameters to achieve adaptive noise reduction of the distribution transformer.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes an adaptive noise reduction method for a distribution transformer according to any one of claims 1 to 7 when running the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements an adaptive noise reduction method for a distribution transformer as described in any one of claims 1 to 7.

Images