A dynamic denoising method for equipment historical data time domain numerical purification

Abstract

The application relates to the technical field of data processing, and discloses a dynamic denoising method for time domain numerical purification of equipment history data, which comprises the following steps: obtaining original equipment history time domain data, then pre-processing the obtained data to output standardized equipment history time domain data; performing time domain and frequency domain joint analysis based on the standardized equipment history time domain data, and performing scene division; then based on the scene division result, matching a denoising algorithm combination in a targeted manner; after denoising is completed, using a weighting algorithm to calculate a single-group algorithm comprehensive score, and determining a denoising effect grade according to the single-group algorithm comprehensive score result; finally, screening an optimal denoising algorithm with the highest comprehensive score, and finally outputting the data after denoising according to the optimal denoising algorithm. The method completely solves the technical problems that a traditional single algorithm is difficult to cope with complex noise coupling and is prone to causing effective information loss.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a dynamic denoising method for time-domain numerical purification of historical data from equipment. Background Technology

[0002] In fields such as power system automation and industrial equipment monitoring, historical time-domain data of equipment (such as voltage, current, power, vibration signals, etc.) is the core basis for achieving condition assessment, fault diagnosis, and optimized scheduling. However, the data is inevitably subject to multi-source interference during the acquisition, transmission, and storage process, resulting in the following key pain points in existing technologies:

[0003] The noise types are complex and coupled: in real-world scenarios, there are simultaneously strong DC offset baseline drift (sensor temperature drift, zero-point drift), high-frequency harmonic narrowband interference (power grid harmonics, electromagnetic interference) and Gaussian random noise (thermal noise of electronic devices). Existing methods mostly use a single denoising algorithm, which is difficult to achieve the suppression effect of multiple types of noise.

[0004] Poor algorithm adaptability, which can easily lead to information loss: Traditional fixed algorithms often adopt a "one-size-fits-all" strategy, which lacks specificity for different noise scenarios. They either excessively smooth out effective features, such as transient signals in power systems, or leave residual interference due to incomplete noise removal, which cannot meet the needs of high-precision data analysis.

[0005] Therefore, this application proposes a dynamic denoising method for time-domain numerical purification of historical data of equipment to solve the above problems. Summary of the Invention

[0006] Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a dynamic denoising method for time-domain numerical purification of historical data of equipment. It achieves a successful balance between denoising effect and signal fidelity by accurately classifying multiple types of noise and implementing differentiated suppression strategies. Combined with quantitative evaluation and optimal algorithm closed-loop screening mechanism, it completely solves the technical problem that traditional single algorithms are unable to cope with complex noise coupling and are prone to loss of effective information.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: a dynamic denoising method for time-domain numerical purification of historical data of equipment, comprising the following steps:

[0010] Step 1: Obtain the raw historical time-domain data of the equipment, and then preprocess the acquired data to output standardized historical time-domain data of the equipment.

[0011] Step 2: Perform joint time-domain and frequency-domain analysis based on the historical time-domain data of standardized equipment, and divide the scenarios accordingly;

[0012] Step 3: Based on the scene segmentation results, select appropriate combinations of denoising algorithms;

[0013] Step 4: After denoising is completed, calculate the comprehensive score of each algorithm group, determine the denoising effect level based on the comprehensive score of each algorithm group, and finally select the best denoising algorithm with the highest comprehensive score.

[0014] Step 5: Output the data after denoising using the optimal denoising algorithm.

[0015] Preferably, in step one, the preprocessing includes:

[0016] First, the acquired raw equipment historical time-domain data is formatted and missing data is filled using linear interpolation. Then, abnormal data is removed using the 3σ criterion combined with the quartile method, and the processed data is normalized to [0,1]. Finally, standardized equipment historical time-domain data is output.

[0017] Preferably, the specific steps of step two are as follows:

[0018] S1. Calculate the DC component amplitude based on standardized equipment historical time-domain data. Baseline drift slope Determine baseline characteristics;

[0019] S2. Calculate the noise power spectral density based on standardized equipment historical time-domain data. High-frequency component ratio Determine the type of noise;

[0020] S3. Calculate the first-order difference variance in the time domain based on standardized equipment historical time-domain data. Determine the smoothness of the data;

[0021] S4. Based on the calculation results, the historical time-domain data of the standardized equipment is divided into scenarios.

[0022] Preferably, the DC component amplitude The calculation formula is:

[0023]

[0024] in, Represents standardized historical time-domain data points of the equipment; Represents the total number of data points;

[0025] The baseline drift slope The calculation formula is:

[0026]

[0027] in, Representing the The time for each data point;

[0028] The DC component amplitude and baseline drift slope are compared with the DC component amplitude threshold and baseline drift slope threshold, respectively:

[0029] When the DC component amplitude is greater than the DC component amplitude threshold, a DC offset is determined to exist.

[0030] When the absolute value of the baseline drift slope is greater than the baseline drift slope threshold, baseline drift is determined to exist.

[0031] When the DC component amplitude and baseline drift slope do not exceed the DC component amplitude threshold and baseline drift slope threshold respectively, the baseline is determined to be normal.

[0032] When the DC component amplitude and baseline drift slope both exceed the DC component amplitude threshold and baseline drift slope threshold, a combined DC offset and baseline drift anomaly is determined.

[0033] The noise power spectral density The calculation formula is:

[0034]

[0035] in, Represents the sampling frequency; Representing the One frequency point; Represents standardized data Discrete Fourier Transform, ;

[0036] The proportion of high-frequency components The calculation formula is:

[0037]

[0038] in, Represents the sum of power in the high-frequency band; Represents the sum of the total power across the entire frequency band; Represents a preset high-frequency range;

[0039] The peak values ​​in the noise power spectral density are extracted and compared with the noise power spectral density threshold. At the same time, the proportion of high-frequency components is compared with the high-frequency component proportion threshold.

[0040] When the peak value is greater than the noise power spectral density threshold and the proportion of high-frequency components is greater than the high-frequency component proportion threshold, it is determined to be high-frequency harmonic narrowband interference.

[0041] When the peak value is less than or equal to the noise power spectral density threshold and the proportion of high-frequency components is less than or equal to the proportion of high-frequency components threshold, it is determined to be Gaussian random noise.

[0042] When the peak value is greater than the noise power spectral density threshold and the proportion of high-frequency components is less than or equal to the proportion of high-frequency components threshold, it is determined to be high-frequency broadband noise.

[0043] When the peak value is less than or equal to the noise power spectral density threshold and the proportion of high-frequency components is greater than the proportion of high-frequency components threshold, it is determined to be high-frequency random noise.

[0044] The time-domain first-order difference variance The calculation formula is:

[0045]

[0046] in, Represents the first-order difference. ; Represents the mean of the differences. ;

[0047] The first-order difference variance in the time domain is compared with a smoothness threshold. If the first-order difference variance in the time domain is less than or equal to the smoothness threshold, the data is judged to be smooth; if the first-order difference variance in the time domain is greater than the smoothness threshold, the data is judged to be unsmooth.

[0048] Preferably, the step of dividing the standardized equipment historical time-domain data into scenarios includes:

[0049] Retrieve the judgment results of S1~S3, and classify the data that meet either the condition of DC offset or baseline drift into strong DC offset baseline drift scenarios.

[0050] Data identified as high-frequency harmonic narrowband interference is classified into high-frequency harmonic narrowband interference scenarios.

[0051] Data identified as Gaussian random noise or high-frequency broadband noise will be uniformly classified into the Gaussian random noise scene.

[0052] After the scene is divided, the classification results are bound to the corresponding data.

[0053] Preferably, the targeted matching denoising algorithm combination in step three includes:

[0054] Based on a scenario with strong DC offset baseline drift:

[0055] The baseline correction algorithm is executed first to eliminate DC components and drift trends. Then, based on the data smoothness determination results, a sliding window averaging method is selected for secondary optimization. This is based on a high-frequency harmonic narrowband interference scenario.

[0056] The wavelet soft thresholding denoising algorithm is executed first to suppress high-frequency spikes and narrowband interference, and then the Savitzky-Gory smoothing algorithm is used to optimize the smoothness of the denoised data.

[0057] Based on Gaussian random noise scenario:

[0058] The optimal single algorithm is dynamically selected based on the data volume and the first-order difference variance. When the data volume is less than the preset algorithm switching threshold, the sliding window averaging algorithm is used first to filter out random noise. When the data volume is greater than the preset algorithm switching threshold, the Savitsky-Gorye smoothing algorithm is selected.

[0059] Preferably, in step four, the formula for calculating the comprehensive score of a single algorithm group is:

[0060]

[0061] in, This represents the overall score of a single algorithm group; The normalized score representing the overall energy decay ratio; Normalized score representing high-frequency noise power attenuation; Normalized score representing the rate of change of spectral flatness; The normalized score representing the temporal smoothness ratio; The normalized score representing the rate of change of the first-order autocorrelation coefficient; , , , , Represents weight, and .

[0062] Preferably, the formula for calculating the overall energy attenuation ratio is:

[0063]

[0064] in, Represents the overall energy decay ratio; Represents standardized data; This represents the data after denoising;

[0065] The formula for calculating high-frequency noise power attenuation is:

[0066]

[0067] in, This represents high-frequency noise power attenuation; The total power of the high-frequency band representing standardized data; This represents the total power of the high-frequency band of the data after noise reduction;

[0068] The formula for calculating the rate of change of spectral flatness is:

[0069]

[0070] in, Represents the rate of change of spectral flatness; This represents the spectral flatness of the denoised data. The spectral flatness of the standardized data;

[0071] The formula for calculating the temporal smoothness ratio is:

[0072]

[0073] in, Represents the time-domain smoothness ratio; This represents the first-order difference variance of the denoised data; This represents the first-order difference variance of the standardized data;

[0074] The formula for calculating the rate of change of the first-order autocorrelation coefficient is:

[0075]

[0076] in, This represents the rate of change of the first-order autocorrelation coefficient; Represents standardized data One data point; Represents standardized data One data point; Represents the covariance function; Represents the standard deviation function; Represents the data after denoising. One data point; Represents the data after denoising. Data points.

[0077] Preferably, the comprehensive score of the single algorithm is compared with the comprehensive score threshold of the single algorithm. If the comprehensive score of the single algorithm exceeds the comprehensive score threshold of the single algorithm, it means that the noise suppression effect is qualified; otherwise, it is judged as unqualified.

[0078] Iterate through all algorithms in the targeted matching denoising algorithm combination, compare the comprehensive scores of each group of algorithms, and select the algorithm with the highest comprehensive score as the optimal denoising algorithm for the current data.

[0079] If the combined score of all algorithms is deemed unqualified, an alarm will be triggered indicating abnormal output data.

[0080] Preferably, in step five, the output data retains the original timestamp, and metadata is added to the denoised data to record the denoising algorithm used, evaluation level, and processing time.

[0081] Compared with the prior art, the present invention provides a dynamic denoising method for time-domain numerical purification of historical data of equipment, which has the following beneficial effects:

[0082] 1. This invention achieves accurate classification and targeted suppression of multiple types of noise: By constructing a noise scene division mechanism, it accurately identifies three types of scenes: strong DC offset baseline drift, high-frequency harmonic narrowband interference, and Gaussian random noise, and matches the optimal combination of denoising algorithms for different scenes, solving the problem that a single algorithm cannot adapt to complex noise.

[0083] 2. This invention achieves a balance between noise suppression and effective signal preservation: through a dynamic scheduling serial combination algorithm, it thoroughly filters out target noise while preserving key features of equipment operation to the greatest extent, thereby improving the fidelity of data purification.

[0084] 3. This invention establishes a quantitative evaluation system to achieve closed-loop selection of the optimal algorithm: calculate the comprehensive score of a single group of algorithms, determine the level of denoising effect based on the comprehensive score of a single group of algorithms, and finally select the optimal denoising algorithm with the highest comprehensive score. This objectively determines the level of denoising effect, solves the problems of denoising effect not being quantifiable and algorithm selection relying on experience, and ensures the stability and optimality of denoising quality.

[0085] 4. This invention improves the efficiency of project implementation and data traceability: By automatically identifying scenarios, it reduces the cost of manual intervention and adds metadata to the purified data, realizing full-chain traceability of data purification and providing reliable data support for subsequent equipment status analysis, fault diagnosis and algorithm iteration. Attached Figure Description

[0086] Figure 1 This is a diagram illustrating the steps of the method of the present invention. Detailed Implementation

[0087] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It is worth noting that this application also relates to prior art. Since prior art is well known to those skilled in the art, it will not be described in detail in this application.

[0088] Please see Figure 1A dynamic denoising method for time-domain numerical purification of historical device data includes the following steps:

[0089] Step 1: Obtain the original historical time domain data of the equipment, then format the obtained original historical time domain data of the equipment, and use the linear interpolation filling method to fill the missing data. Then, use the 3σ criterion combined with the quartile method to remove abnormal data, and normalize the processed data to [0,1]. Finally, the standardized historical time domain data of the equipment is output.

[0090] Step 1 constructs a standardized data base through multiple data preprocessing steps. Among them, linear interpolation ensures the continuity and integrity of time-series data, avoiding missing values ​​that could cause subsequent algorithms to break down. The anomaly removal mechanism effectively filters out random outliers and systematic biases while preserving the true operating conditions of the equipment to the greatest extent. Normalization eliminates the influence of differences in units and amplitudes. The final output standardized data not only provides a high-quality and flawless input foundation for the accurate segmentation of subsequent noise scenes, but also ensures that various denoising algorithms have consistent input characteristics under a unified scale, significantly improving the accuracy of subsequent noise identification and the adaptability and robustness of denoising algorithms.

[0091] Step 2: Perform joint time-domain and frequency-domain analysis based on standardized equipment historical time-domain data, and classify scenarios accordingly.

[0092] S1. Calculate the DC component amplitude based on standardized equipment historical time-domain data. Baseline drift slope Determine baseline characteristics and accurately quantify the static bias and dynamic drift characteristics of the signal from the perspective of time-domain trends;

[0093] DC component amplitude The calculation formula is:

[0094]

[0095] in, Represents standardized historical time-domain data points of the equipment; Represents the total number of data points;

[0096] Baseline drift slope The calculation formula is:

[0097]

[0098] in, Representing the The time for each data point;

[0099] The DC component amplitude and baseline drift slope are compared with the DC component amplitude threshold and baseline drift slope threshold, respectively:

[0100] When the DC component amplitude is greater than the DC component amplitude threshold, a DC offset is determined to exist.

[0101] When the absolute value of the baseline drift slope is greater than the baseline drift slope threshold, baseline drift is determined to exist.

[0102] When the DC component amplitude and baseline drift slope do not exceed the DC component amplitude threshold and baseline drift slope threshold respectively, the baseline is determined to be normal.

[0103] When the DC component amplitude and baseline drift slope both exceed the DC component amplitude threshold and baseline drift slope threshold, a combined DC offset and baseline drift anomaly is determined.

[0104] By decoupling the baseline anomaly into the DC component amplitude of a fixed offset and the slowly changing trend of baseline drift slope Two quantifiable metrics can accurately identify strong DC offset and baseline drift scenarios, providing a clear basis for prioritizing the execution of baseline correction algorithms and avoiding the failure of subsequent high-frequency denoising algorithms or the generation of artifacts due to the failure to identify baseline problems.

[0105] S2. Calculate the noise power spectral density based on standardized equipment historical time-domain data. High-frequency component ratio To determine the type of noise, distinguish between the frequency characteristics and energy concentration of the noise from the perspective of frequency domain energy distribution;

[0106] Noise power spectral density The calculation formula is:

[0107]

[0108] in, Represents the sampling frequency; Representing the One frequency point; Represents standardized data Discrete Fourier Transform, ;

[0109] High-frequency component proportion The calculation formula is:

[0110]

[0111] in, Represents the sum of power in the high-frequency band; Represents the sum of the total power across the entire frequency band; Represents a preset high-frequency range;

[0112] The peak values ​​in the noise power spectral density are extracted and compared with the noise power spectral density threshold. At the same time, the proportion of high-frequency components is compared with the high-frequency component proportion threshold.

[0113] When the peak value is greater than the noise power spectral density threshold and the proportion of high-frequency components is greater than the high-frequency component proportion threshold, it is determined to be high-frequency harmonic narrowband interference.

[0114] When the peak value is less than or equal to the noise power spectral density threshold and the proportion of high-frequency components is less than or equal to the proportion of high-frequency components threshold, it is determined to be Gaussian random noise.

[0115] When the peak value is greater than the noise power spectral density threshold and the proportion of high-frequency components is less than or equal to the proportion of high-frequency components threshold, it is determined to be high-frequency broadband noise.

[0116] When the peak value is less than or equal to the noise power spectral density threshold and the proportion of high-frequency components is greater than the proportion of high-frequency components threshold, it is determined to be high-frequency random noise.

[0117] Using noise power spectral density It can accurately locate the characteristic frequencies of narrowband interference, combined with the proportion of high-frequency components. The overall high-frequency interference intensity is quantified, thereby effectively distinguishing between high-frequency harmonic narrowband interference and Gaussian random noise. This solves the problem that single time-domain analysis cannot identify the source of noise and ensures the targeted selection of denoising algorithms.

[0118] S3. Calculate the first-order difference variance in the time domain based on standardized equipment historical time-domain data. To determine the smoothness of data, from the perspective of time-domain fluctuations, quantify the overall roughness and degree of abrupt changes in the signal;

[0119] First-order difference variance in the time domain The calculation formula is:

[0120]

[0121] in, Represents the first-order difference. ; Represents the mean of the differences. ;

[0122] The first-order difference variance in the time domain is compared with a smoothness threshold. If the first-order difference variance in the time domain is less than or equal to the smoothness threshold, the data is judged to be smooth; if the first-order difference variance in the time domain is greater than the smoothness threshold, the data is judged to be unsmooth.

[0123] First-order difference variance in the time domain It can intuitively reflect the degree of data contamination by random noise, providing a direct basis for the classification of Gaussian random noise scenarios, and also providing a data-driven reference standard for the adaptive adjustment of subsequent algorithm parameters, thus realizing a quantitative evaluation of the smoothness.

[0124] S4. Based on the calculation results, the historical time-domain data of standardized equipment is divided into scenarios, and a multi-condition scenario classification mechanism with clear priorities and logical mutual exclusion is constructed.

[0125] Retrieve the judgment results of S1~S3, and classify the data that meet either the condition of DC offset or baseline drift into strong DC offset baseline drift scenarios.

[0126] Data identified as high-frequency harmonic narrowband interference is classified into high-frequency harmonic narrowband interference scenarios.

[0127] Data identified as Gaussian random noise or high-frequency broadband noise will be uniformly classified into the Gaussian random noise scene.

[0128] After the scene is divided, the classification results are bound to the corresponding data;

[0129] Scene segmentation ensures the highest priority processing of signal underlying trend anomalies, avoids the obscuring of other noise features due to baseline issues, and decouples complex noise coupling problems into three standard processable scenarios. This solves the pain point of poor adaptability of traditional one-size-fits-all denoising algorithms, enabling the system to directly index and execute the optimal combination of denoising algorithms based on labels. It achieves seamless connection from noise analysis to algorithm execution, significantly improving the automation level and decision-making efficiency of the denoising process.

[0130] Step 3: Based on the scene segmentation results, targeted matching of denoising algorithm combinations is performed, which completely solves the technical pain point that traditional single algorithms cannot simultaneously suppress multiple types of noise and preserve effective signals.

[0131] Based on a scenario with strong DC offset baseline drift:

[0132] The baseline correction algorithm is executed first to eliminate DC components and drift trends. Then, based on the data smoothness judgment results, the sliding window averaging is selected for secondary optimization, which not only eliminates the underlying trend anomalies, but also achieves mild smoothing optimization.

[0133] Based on high-frequency harmonic narrowband interference scenarios:

[0134] The wavelet soft thresholding denoising algorithm is executed first to suppress high-frequency spikes and narrowband interference. Then, the Savitzky-Gory smoothing algorithm is used to optimize the smoothness of the denoised data. While accurately filtering out high-frequency spikes and narrowband interference, waveform details are repaired and smoothness is improved.

[0135] Based on Gaussian random noise scenario:

[0136] The optimal single algorithm is dynamically selected based on the data volume and the first-order difference variance. When the data volume is less than the preset algorithm switching threshold, the sliding window averaging algorithm is used first to filter out random noise. When the data volume is greater than the preset algorithm switching threshold, the Savitzky-Gory smoothing algorithm is selected. This achieves a balance between computational efficiency and data fidelity while ensuring the denoising effect.

[0137] Step 3's overall strategy, through scenario-based precise matching and adaptive decision-making, maximizes the suppression of target noise while minimizing the damage to key characteristics of equipment operation, significantly improving the targeting, effectiveness, and engineering execution efficiency of noise reduction, and providing high-quality assurance for subsequent data applications.

[0138] Step 4: After denoising is completed, calculate the comprehensive score of each algorithm group, determine the denoising effect level based on the comprehensive score of each algorithm group, and finally select the optimal denoising algorithm with the highest comprehensive score. This upgrades the evaluation of denoising effect from subjective qualitative judgment to objective quantitative decision-making, solving the key pain points of traditional methods where denoising quality cannot be quantified and the optimal algorithm depends on human experience for selection.

[0139] The formula for calculating the overall score of a single algorithm is:

[0140]

[0141] in, This represents the overall score of a single algorithm group; The normalized score representing the overall energy decay ratio; Normalized score representing high-frequency noise power attenuation; Normalized score representing the rate of change of spectral flatness; The normalized score representing the temporal smoothness ratio; The normalized score representing the rate of change of the first-order autocorrelation coefficient; , , , , Represents weight, and ;

[0142] The formula for calculating the overall energy decay ratio is:

[0143]

[0144] in, Represents the overall energy decay ratio; Represents standardized data; This represents the data after denoising;

[0145] The formula for calculating high-frequency noise power attenuation is:

[0146]

[0147] in, This represents high-frequency noise power attenuation; The total power of the high-frequency band representing standardized data; This represents the total power of the high-frequency band of the data after noise reduction;

[0148] The formula for calculating the rate of change of spectral flatness is:

[0149]

[0150] in, Represents the rate of change of spectral flatness; This represents the spectral flatness of the denoised data. The spectral flatness of the standardized data;

[0151] The formula for calculating the temporal smoothness ratio is:

[0152]

[0153] in, Represents the time-domain smoothness ratio; This represents the first-order difference variance of the denoised data; This represents the first-order difference variance of the standardized data;

[0154] The formula for calculating the rate of change of the first-order autocorrelation coefficient is:

[0155]

[0156] in, This represents the rate of change of the first-order autocorrelation coefficient; Represents standardized data One data point; Represents standardized data One data point; Represents the covariance function; Represents the standard deviation function; Represents the data after denoising. One data point; Represents the data after denoising. One data point;

[0157] The overall score of a single algorithm is compared with the threshold for the overall score of a single algorithm. If the overall score of a single algorithm exceeds the threshold for the overall score of a single algorithm, it means that the noise suppression effect is qualified; otherwise, it is deemed unqualified.

[0158] Iterate through all algorithms in the targeted matching denoising algorithm combination, compare the comprehensive scores of each group of algorithms, and select the algorithm with the highest comprehensive score as the optimal denoising algorithm for the current data.

[0159] If the combined score of all algorithms is deemed unqualified, an alarm will be triggered indicating abnormal output data.

[0160] Step four calculates the comprehensive score of a single algorithm group and classifies the effect level. This can accurately quantify the balance between noise suppression and effective signal preservation of each algorithm combination, clearly define the quality boundary of data purification, and then select the best algorithm with the highest comprehensive score. This not only ensures that the noise reduction effect of the output data reaches the theoretical optimal level in the current scenario, but also significantly improves the intelligence level of the process by replacing manual intervention with automated decision-making. This effectively ensures the stability, reliability and optimality of the data purification results.

[0161] Step 5: Output the data after denoising using the optimal denoising algorithm. The output data retains the original timestamp, and metadata is added to the denoised data to record the denoising algorithm used, evaluation level, and processing time. This constructs a complete data traceability link from raw data acquisition to final denoising optimization, solving the pain points of black box data processing and untraceable results in industrial scenarios. Ultimately, it achieves standardization, traceability, and auditability of data purification results, significantly improving the management level of data assets and the decision reliability of downstream applications.

[0162] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic denoising method for time-domain numerical purification of historical data of equipment, characterized in that, Includes the following steps: Step 1: Obtain the raw historical time-domain data of the equipment, and then preprocess the acquired data to output standardized historical time-domain data of the equipment. Step 2: Perform joint time-domain and frequency-domain analysis based on the historical time-domain data of standardized equipment, and divide the scenarios accordingly; Step 3: Based on the scene segmentation results, select appropriate combinations of denoising algorithms; Step 4: After denoising is completed, calculate the comprehensive score of each algorithm group, determine the denoising effect level based on the comprehensive score of each algorithm group, and finally select the best denoising algorithm with the highest comprehensive score. Step 5: Output the data after denoising using the optimal denoising algorithm.

2. The dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 1, characterized in that, In step one, the preprocessing includes: First, the acquired raw equipment historical time-domain data is formatted and missing data is filled using linear interpolation. Then, abnormal data is removed using the 3σ criterion combined with the quartile method, and the processed data is normalized to [0,1]. Finally, standardized equipment historical time-domain data is output.

3. The dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 2, characterized in that, The specific steps for step two are as follows: S1. Calculate the DC component amplitude based on standardized equipment historical time-domain data. Baseline drift slope Determine baseline characteristics; S2. Calculate the noise power spectral density based on standardized equipment historical time-domain data. High-frequency component ratio Determine the type of noise; S3. Calculate the first-order difference variance in the time domain based on standardized equipment historical time-domain data. Determine the smoothness of the data; S4. Based on the calculation results, the historical time-domain data of the standardized equipment is divided into scenarios.

4. The dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 3, characterized in that, The DC component amplitude The calculation formula is: in, Represents standardized historical time-domain data points of the equipment; Represents the total number of data points; The baseline drift slope The calculation formula is: in, Representing the The time for each data point; The DC component amplitude and baseline drift slope are compared with the DC component amplitude threshold and baseline drift slope threshold, respectively: When the DC component amplitude is greater than the DC component amplitude threshold, a DC offset is determined to exist. When the absolute value of the baseline drift slope is greater than the baseline drift slope threshold, baseline drift is determined to exist. When the DC component amplitude and baseline drift slope do not exceed the DC component amplitude threshold and baseline drift slope threshold respectively, the baseline is determined to be normal. When the DC component amplitude and baseline drift slope both exceed the DC component amplitude threshold and baseline drift slope threshold, a combined DC offset and baseline drift anomaly is determined. The noise power spectral density The calculation formula is: in, Represents the sampling frequency; Representing the One frequency point; Represents standardized data Discrete Fourier Transform, ; The proportion of high-frequency components The calculation formula is: in, Represents the sum of power in the high-frequency band; Represents the sum of the total power across the entire frequency band; Represents a preset high-frequency range; The peak values ​​in the noise power spectral density are extracted and compared with the noise power spectral density threshold. At the same time, the proportion of high-frequency components is compared with the high-frequency component proportion threshold. When the peak value is greater than the noise power spectral density threshold and the proportion of high-frequency components is greater than the high-frequency component proportion threshold, it is determined to be high-frequency harmonic narrowband interference. When the peak value is less than or equal to the noise power spectral density threshold and the proportion of high-frequency components is less than or equal to the proportion of high-frequency components threshold, it is determined to be Gaussian random noise. When the peak value is greater than the noise power spectral density threshold and the proportion of high-frequency components is less than or equal to the proportion of high-frequency components threshold, it is determined to be high-frequency broadband noise. When the peak value is less than or equal to the noise power spectral density threshold and the proportion of high-frequency components is greater than the proportion of high-frequency components threshold, it is determined to be high-frequency random noise. The time-domain first-order difference variance The calculation formula is: in, Represents the first-order difference. ; Represents the mean of the differences. ; The first-order difference variance in the time domain is compared with a smoothness threshold. If the first-order difference variance in the time domain is less than or equal to the smoothness threshold, the data is judged to be smooth; if the first-order difference variance in the time domain is greater than the smoothness threshold, the data is judged to be unsmooth.

5. A dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 3, characterized in that, The process of dividing standardized equipment historical time-domain data into scenarios includes: Retrieve the judgment results of S1~S3, and classify the data that meet either the condition of DC offset or baseline drift into strong DC offset baseline drift scenarios. Data identified as high-frequency harmonic narrowband interference is classified into high-frequency harmonic narrowband interference scenarios. Data identified as Gaussian random noise or high-frequency broadband noise will be uniformly classified into the Gaussian random noise scene. After the scene is divided, the classification results are bound to the corresponding data.

6. A dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 5, characterized in that, The third step, the targeted matching denoising algorithm combination, includes: Based on a scenario with strong DC offset baseline drift: The baseline correction algorithm is executed first to eliminate DC components and drift trends. Then, based on the data smoothness determination results, a sliding window averaging method is selected for secondary optimization. This is based on a high-frequency harmonic narrowband interference scenario. The wavelet soft thresholding denoising algorithm is executed first to suppress high-frequency spikes and narrowband interference, and then the Savitzky-Gory smoothing algorithm is used to optimize the smoothness of the denoised data. Based on Gaussian random noise scenario: The optimal single algorithm is dynamically selected based on the data volume and the first-order difference variance. When the data volume is less than the preset algorithm switching threshold, the sliding window averaging algorithm is used first to filter out random noise. When the data volume is greater than the preset algorithm switching threshold, the Savitsky-Gorye smoothing algorithm is selected.

7. A dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 6, characterized in that, In step four, the formula for calculating the comprehensive score of a single algorithm group is as follows: in, This represents the overall score of a single algorithm group; The normalized score representing the overall energy decay ratio; Normalized score representing high-frequency noise power attenuation; Normalized score representing the rate of change of spectral flatness; The normalized score representing the temporal smoothness ratio; The normalized score representing the rate of change of the first-order autocorrelation coefficient; , , , , Represents weight, and .

8. A dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 7, characterized in that, The formula for calculating the overall energy attenuation ratio is: in, Represents the overall energy decay ratio; Represents standardized data; This represents the data after denoising; The formula for calculating high-frequency noise power attenuation is: in, This represents high-frequency noise power attenuation; The total power of the high-frequency band representing standardized data; This represents the total power of the high-frequency band of the data after noise reduction; The formula for calculating the rate of change of spectral flatness is: in, Represents the rate of change of spectral flatness; This represents the spectral flatness of the denoised data. The spectral flatness of the standardized data; The formula for calculating the temporal smoothness ratio is: in, Represents the time-domain smoothness ratio; This represents the first-order difference variance of the denoised data; This represents the first-order difference variance of the standardized data; The formula for calculating the rate of change of the first-order autocorrelation coefficient is: in, This represents the rate of change of the first-order autocorrelation coefficient; Represents standardized data One data point; Represents standardized data One data point; Represents the covariance function; Represents the standard deviation function; Represents the data after denoising. One data point; Represents the data after denoising. Data points.

9. A dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 7, characterized in that, The overall score of the single algorithm is compared with the threshold of the overall score of the single algorithm. If the overall score of the single algorithm exceeds the threshold of the overall score of the single algorithm, it means that the noise suppression effect is qualified; otherwise, it is judged as unqualified. Iterate through all algorithms in the targeted matching denoising algorithm combination, compare the comprehensive scores of each group of algorithms, and select the algorithm with the highest comprehensive score as the optimal denoising algorithm for the current data. If the combined score of all algorithms is deemed unqualified, an alarm will be triggered indicating abnormal output data.

10. A dynamic denoising method for time-domain numerical purification of historical equipment data according to claim 9, characterized in that, In step five, the output data retains the original timestamp, and metadata is added to the denoised data to record the denoising algorithm used, evaluation level, and processing time.

Images