GIS partial discharge fault-oriented sf6 decomposition product multi-component spectrum detection method and system

Abstract

The present application belongs to the technical field of fault diagnosis, and relates to a GIS partial discharge fault-oriented SF6 decomposition product multi-component spectrum detection method and system. The method comprises the following steps: firstly, obtaining initial spectrum data of SF6 decomposition mixed gas in a GIS three-phase gas chamber, and obtaining corrected spectrum data through support vector machine baseline correction; obtaining independent spectrum data through independent component analysis algorithm blind source separation; obtaining enhanced spectrum data through baseline drift calculation, partial least squares concentration inversion reconstructed spectrum and residual determination; extracting absorption peak characteristic value and completing matching verification to determine the actual concentration of each characteristic gas; obtaining the screening concentration of the target characteristic gas through diagnostic contribution degree threshold screening, inputting the random forest model to calculate the diagnostic weight data; finally, combining the weight and the concentration to calculate the insulation life decay amount, and outputting the GIS equipment fault state judgment result. The present method can adapt to complex operation conditions of GIS on site, can eliminate multi-component spectrum cross interference, and can realize accurate quantification.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fault diagnosis, and particularly relates to a GIS partial discharge fault-oriented SF6 decomposition product multi-component spectrum detection method and system. BACKGROUND

[0002] Gas insulated metal enclosed switchgear (GIS) has become the core equipment in high-voltage and extra-high-voltage power transmission and transformation systems due to its small footprint, high operation reliability and long maintenance cycle, and its insulation operation state is directly related to the safe and stable operation of the power grid. Sulfur hexafluoride (SF6) is widely used as an insulating medium for GIS equipment due to its excellent insulation and arc extinguishing performance. During long-term operation of the equipment, partial discharge caused by internal insulation defects can lead to decomposition of SF6 gas, generating SO2, H2S, CO, HF and other characteristic decomposition products. The type, concentration level and change trend of such products have a clear corresponding relationship with the type and severity of internal insulation faults of the equipment. Therefore, decomposition component analysis is a mainstream technical means for evaluating the insulation state of GIS equipment and identifying latent faults.

[0003] Currently, the mainstream detection methods for SF6 decomposition products can be divided into offline laboratory detection and online monitoring. Offline detection is represented by gas chromatography, gas detection tube method and electrochemical sensor method. Gas chromatography has high detection accuracy, but requires on-site sampling and sending for detection, which has a long detection cycle, complex operation process and cannot realize continuous real-time monitoring. In addition, secondary errors can be easily introduced during sampling and transportation. The detection tube method and electrochemical sensor method are simple to operate, but have single detection components, serious cross-interference and insufficient long-term operation stability, which cannot meet the needs of high-precision fault diagnosis.

[0004] To address the shortcomings of offline detection, online detection technologies based on spectral principles have been extensively studied, including tunable diode laser absorption spectroscopy, Fourier transform infrared spectroscopy, and photoacoustic spectroscopy. However, existing spectral detection schemes still have many limitations in practical engineering applications: First, single spectral techniques cannot simultaneously cover multiple components and detect trace amounts, failing to fully cover the quantitative detection needs of various SF6 characteristic decomposition products; Second, the complex electromagnetic environment, temperature, and pressure fluctuations in GIS sites easily cause spectral baseline drift and noise interference, and conventional baseline correction methods have poor adaptability to nonlinear drift, directly affecting the accuracy of concentration quantification; Third, the infrared absorption peaks of multi-component decomposition products have a large number of overlapping regions, making it difficult to eliminate cross-interference caused by spectral aliasing, and the trace characteristic signals generated by early latent faults are easily masked by background gases and noise, resulting in a high rate of missed detection of early faults; Fourth, existing fault diagnosis models mostly use fixed-weight component ratio criteria, which cannot adaptively adjust the diagnostic weights of different characteristic gases according to measured data, making the diagnostic results susceptible to interference components, and most can only achieve qualitative judgment of the presence or absence of faults, making it difficult to quantitatively assess the trend of equipment insulation degradation and the degree of lifespan decline. Summary of the Invention

[0005] Based on this, it is necessary to provide a method and system for multi-component spectral detection of SF6 decomposition products for GIS partial discharge faults, which can effectively suppress spectral baseline drift and noise interference under complex on-site working conditions, achieve accurate unmixing and high-sensitivity quantitative detection of multi-component SF6 decomposition products, and quantitatively assess the insulation degradation trend and lifespan attenuation of equipment.

[0006] Firstly, this application provides a multi-component spectroscopic detection method for SF6 decomposition products in GIS partial discharge faults, including:

[0007] Initial spectral data of SF6 decomposition mixture in the three-phase gas chamber of a gas-insulated metal-enclosed switchgear are obtained. Baseline correction is performed on the initial spectral data using a support vector machine to obtain corrected spectral data.

[0008] Based on the corrected spectral data, the independent component analysis algorithm is used to perform blind source separation, and the independent spectral data corresponding to each partial discharge characteristic gas are obtained.

[0009] Baseline drift is calculated based on the absorption peak characteristics of independent spectral data. The spectrum is then reconstructed using partial least squares concentration inversion, and the spectral residuals are calculated. For reconstructed spectra with excessive spectral residuals, signal enhancement is performed to obtain enhanced spectral data.

[0010] Based on the enhanced spectral data, the absorption peak characteristic values ​​of each partial discharge characteristic gas are extracted, and it is determined whether the absorption peak characteristic values ​​meet the preset characteristic matching conditions. If they do, the actual concentration values ​​of each partial discharge characteristic gas are determined.

[0011] Obtain the actual concentration value, calculate the diagnostic contribution value corresponding to each partial discharge characteristic gas, and if the diagnostic contribution value is less than the preset contribution threshold, remove the corresponding partial discharge characteristic gas to obtain the screening concentration value of the target partial discharge characteristic gas.

[0012] The selected concentration values ​​are input into the random forest model to calculate the diagnostic weight values ​​of the partial discharge characteristic gases of each target, thus obtaining the diagnostic weight data.

[0013] Based on the diagnostic weight data and the screening concentration values, the insulation life attenuation of partial discharge is calculated, and the fault status judgment result of the gas-insulated metal-enclosed switchgear is obtained.

[0014] In one embodiment, blind source separation is performed using an independent component analysis algorithm based on the corrected spectral data to obtain independent spectral data corresponding to each partial discharge characteristic gas, including:

[0015] An initial spectral matrix is ​​constructed based on the corrected spectral data.

[0016] The initial spectral matrix is ​​iteratively calculated using the independent component analysis algorithm to obtain the target separation matrix.

[0017] Blind source separation is performed on the initial spectral matrix based on the target separation matrix to obtain candidate independent components.

[0018] Matching and judgment are performed by calculating the Pearson correlation coefficient between each candidate independent component and the target absorption feature corresponding to each partial discharge characteristic gas in the absorption feature library:

[0019]

[0020] in, Indicates the first The candidate independent component and the first The correlation coefficient of the absorption characteristics of each target Indicates the first The candidate independent components in the th... The value at each wavelength point Indicates the first The average of the candidate independent components, Indicates the first The target absorption feature in the first The value at each wavelength point Indicates the first The average value of the absorption characteristics of each target.

[0021] The absorption feature library is a pre-built database that stores the target absorption spectral features corresponding to each partial discharge characteristic gas.

[0022] If a candidate independent component matches any target absorption feature, then the candidate independent component is determined to be the independent spectral data corresponding to the partial discharge characteristic gas.

[0023] In one embodiment, an independent component analysis algorithm is used to perform iterative calculations on the initial spectral matrix to obtain the target separation matrix, including:

[0024] The initial spectral matrix is ​​centered to obtain the centered spectral matrix:

[0025]

[0026] in, Represents a vector consisting entirely of 1s. Represents the initial spectral matrix. This indicates the total number of spectral wavelength sampling points. Indicates the number of consecutive samples. Indicates the first The corrected spectral vector obtained from the second sampling.

[0027] Perform whitening on the centered spectral matrix to obtain the whitened spectral matrix:

[0028]

[0029] in, express eigenvector matrix, express The eigenvalue diagonal matrix.

[0030] The FastICA iterative algorithm is used to solve for the separation vector in the whitening spectral matrix to obtain the target separation matrix:

[0031]

[0032]

[0033] in, Indicates the first The separation vector of the next iteration Represents a nonlinear function. , Represents the mathematical expectation operation, iterating to... When the time stops, the resulting separation vectors constitute the target separation matrix. , This indicates the number of gas types characteristic of partial discharge.

[0034] In one embodiment, the enhanced spectral data is generated through the following process:

[0035] Extract the spectral absorption peak characteristics corresponding to independent spectral data.

[0036] The characteristics of spectral absorption peaks include the absorption peak position, absorption peak height, absorption peak area, absorption peak half width at half maximum (FWHM), and absorption peak shape coefficient.

[0037] The baseline drift is calculated based on the baseline regions on both sides of the spectral absorption peak characteristics.

[0038] The formula for calculating baseline drift is:

[0039]

[0040] in, Indicates the baseline drift amount. This indicates the starting wavelength of the baseline region corresponding to the spectral absorption peak characteristics. Indicates the termination wavelength of the baseline region. Indicates independent spectral data in The absorbance value at that point Indicates independent spectral data in The absorbance value at that location.

[0041] If the baseline drift is greater than the preset drift threshold, the partial least squares algorithm is used to perform concentration inversion calculation on the independent spectral data to obtain the concentration inversion matrix.

[0042] The reconstructed spectrum is generated based on the concentration inversion matrix.

[0043] The spectral residuals between the reconstructed spectrum and the independent spectral data were calculated.

[0044]

[0045] in, Indicates spectral residual, Represents the first independent spectral data matrix. The first sample absorbance values ​​at each wavelength point Represents the first element in the reconstructed spectral matrix. The first sample Absorbance values ​​at each wavelength point.

[0046] If the spectral residual is greater than the preset residual threshold, signal enhancement processing is performed on the reconstructed spectrum to obtain enhanced spectral data.

[0047] In one embodiment, the expression for the diagnostic contribution value corresponding to each partial discharge characteristic gas is:

[0048]

[0049] in, Indicates the first The diagnostic contribution values ​​corresponding to the characteristic gases of partial discharge. Indicates the first The actual concentration values ​​of the characteristic gases of partial discharge. Indicates the first Preset critical fault concentration values ​​for a type of partial discharge characteristic gas. This represents the total number of types of gases characteristic of partial discharge.

[0050] In one embodiment, the selected concentration values ​​are input into a random forest model to calculate the diagnostic weight values ​​for each target partial discharge characteristic gas, resulting in diagnostic weight data, including:

[0051] The selected concentration values ​​are input into a pre-trained random forest model. When performing feature splitting at each node of all decision trees in the random forest model, the decrease in Gini index corresponding to each target partial discharge characteristic gas is calculated:

[0052]

[0053] in, Indicates the first The characteristic gas of the target partial discharge in the first The decrease in Gini index corresponding to feature splitting of each decision tree node. Indicates the first before the split The Gini index of each node, , This indicates the total number of partial discharge fault types. Indicates the first Among the nodes belonging to the node The proportion of samples with this type of fault to the total number of samples in the node. Indicates the first before the split The total number of samples at each node. This represents the number of samples in the left child node after the split. The Gini index of the left child node after the split. This represents the number of samples in the right child node after the split. This represents the Gini index of the right child node after the split.

[0054] The gas correlation degree corresponding to the partial discharge characteristic gas of each target is obtained by averaging the decrease of the Gini index of all nodes in all decision trees corresponding to the partial discharge characteristic gas of each target.

[0055] Determine whether the correlation degree of each gas is greater than the preset correlation degree threshold.

[0056] If the gas correlation degree is greater than the preset correlation degree threshold, the gas correlation degree is normalized to obtain the diagnostic weight value corresponding to the partial discharge characteristic gas of each target.

[0057] Diagnostic weight data is generated based on the values ​​of each diagnostic weight.

[0058] In one embodiment, based on diagnostic weight data and screening concentration values, the insulation life attenuation of partial discharge is calculated to obtain the fault status determination result of the gas-insulated metal-enclosed switchgear, including:

[0059] A partial discharge feature vector is constructed by weighting the diagnostic weight data and the selected concentration values.

[0060] The partial discharge feature vectors are input into a pre-trained support vector machine classification model to obtain the discharge pattern classification results.

[0061] By combining the classification results of multiple consecutive discharge modes with the corresponding actual concentration values, degradation trend indicators were extracted.

[0062] Determine whether the degradation trend indicator is greater than the preset degradation trend threshold.

[0063] If the degradation trend index is greater than the preset degradation trend threshold, the insulation life attenuation of the gas-insulated metal-enclosed switchgear is calculated based on the degradation trend index.

[0064] Based on the insulation life attenuation, the fault status determination result of the gas-insulated metal-enclosed switchgear is generated.

[0065] The fault status determination results include discharge mode classification results, insulation life attenuation, fault severity level, and warning level.

[0066] Secondly, this application also provides a multi-component spectroscopic detection system for SF6 decomposition products in GIS partial discharge faults, the system comprising:

[0067] The spectral preprocessing module is used to acquire the initial spectral data of the SF6 decomposition mixture in the three-phase gas chamber of the gas-insulated metal-enclosed switchgear. The initial spectral data is then subjected to baseline correction processing using a support vector machine to obtain corrected spectral data.

[0068] The spectral unmixing enhancement module is used to perform blind source separation based on the corrected spectral data using the independent component analysis algorithm to obtain independent spectral data corresponding to each partial discharge characteristic gas; it is also used to calculate the baseline drift based on the absorption peak characteristics of the independent spectral data, reconstruct the spectrum through partial least squares concentration inversion and calculate the spectral residual, and perform signal enhancement on the reconstructed spectrum with excessive spectral residual to obtain enhanced spectral data.

[0069] The concentration quantitative screening module is used to extract the absorption peak characteristic values ​​of each partial discharge characteristic gas based on the enhanced spectral data, determine whether the absorption peak characteristic values ​​meet the preset characteristic matching conditions, and if so, determine the actual concentration value of each partial discharge characteristic gas; it is also used to obtain the actual concentration value, calculate the diagnostic contribution value corresponding to each partial discharge characteristic gas, and if the diagnostic contribution value is less than the preset contribution threshold, remove the corresponding partial discharge characteristic gas to obtain the screening concentration value of the target partial discharge characteristic gas.

[0070] The fault diagnosis and assessment module is used to input the screened concentration values ​​into the random forest model, calculate the diagnostic weight values ​​of the characteristic gases of partial discharge for each target, and obtain diagnostic weight data; it is also used to calculate the insulation life decay of partial discharge based on the diagnostic weight data and the screened concentration values, and obtain the fault status judgment result of the gas-insulated metal-enclosed switchgear.

[0071] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0072] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0073] The aforementioned method, system, computer equipment, and storage medium for multi-component spectral detection of SF6 decomposition products in GIS partial discharge faults first acquire initial spectral data of the SF6 decomposition mixture in the three-phase gas chamber of a gas-insulated metal-enclosed switchgear. Baseline correction is then performed on the initial spectral data using a support vector machine to suppress baseline drift caused by environmental interference, resulting in corrected spectral data. Using this corrected spectral data as input, an independent component analysis algorithm is employed to perform blind source separation, decoupling the overlapping absorption spectra of the multi-component gases to obtain independent spectral data corresponding to each partial discharge characteristic gas. Absorption peak features are extracted from the independent spectral data, and baseline drift is calculated. Concentration inversion is performed using a partial least squares algorithm to generate a reconstructed spectrum. The spectral residual between the reconstructed spectrum and the independent spectral data is calculated. For reconstructed spectra where the spectral residual exceeds a preset threshold, signal enhancement processing is performed. The process involves: obtaining enhanced spectral data to improve the signal-to-noise ratio of weak fault characteristic signals; extracting the absorption peak feature values ​​of each partial discharge characteristic gas based on the enhanced spectral data, verifying them against preset feature matching conditions, and determining the actual concentration values ​​of each partial discharge characteristic gas after successful matching; calculating the corresponding diagnostic contribution value based on the actual concentration value of each characteristic gas, comparing the diagnostic contribution value with a preset contribution threshold, and eliminating characteristic gases with contribution values ​​below the threshold to obtain the screening concentration values ​​of the target partial discharge characteristic gases; inputting the screening concentration values ​​into a random forest model to calculate the diagnostic weight values ​​of each target partial discharge characteristic gas, generating diagnostic weight data; finally, weighting the screening concentration values ​​with the diagnostic weight data to calculate the insulation life attenuation corresponding to partial discharge, and obtaining the fault state judgment result of the gas-insulated metal-enclosed switchgear.

[0074] This method is adaptable to complex operating conditions in GIS environments. It improves spectral data quality through two-stage processing of baseline correction and signal enhancement, reducing the impact of environmental interference on quantitative results. The blind source separation mechanism of independent component analysis eliminates spectral cross-interference of multi-component decomposition products, enabling simultaneous and accurate quantification of multiple characteristic decomposition products. A mechanism combining contribution-based initial screening and adaptive weight allocation using random forests removes weakly correlated interfering components and optimizes diagnostic weight allocation, improving the accuracy and anti-interference capability of fault diagnosis results. Simultaneously, it can quantify the degree of insulation life decay and multi-dimensional fault status information, providing data support for insulation status assessment and condition-based maintenance decisions for GIS equipment. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 A flowchart of a multi-component spectral detection method for SF6 decomposition products in GIS partial discharge faults provided in an embodiment of the present invention;

[0077] Figure 2 This is a structural block diagram of a multi-component spectral detection system for SF6 decomposition products in GIS partial discharge faults, provided in an embodiment of the present invention. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0079] In one embodiment, such as Figure 1 As shown, this application provides a multi-component spectroscopic detection method for SF6 decomposition products in GIS partial discharge faults, which may include the following steps:

[0080] Step S101: Obtain the initial spectral data of the SF6 decomposition mixture in the three-phase gas chamber of the gas-insulated metal-enclosed switchgear, and perform baseline correction processing on the initial spectral data using a support vector machine to obtain the corrected spectral data.

[0081] Specifically, the initial spectral data of the SF6 decomposition mixture in the three-phase gas chamber of the gas-insulated metal-enclosed switchgear is obtained. The transmission spectrum of the SF6 decomposition mixture inside the chamber is acquired by a spectral detection component deployed in the chamber, resulting in a continuous absorbance spectral signal covering the target detection band, i.e., the initial spectral data. A support vector machine is used to perform baseline correction processing on the initial spectral data. The nonlinear regression fitting capability of the support vector machine is used to fit the nonlinear baseline drift component introduced by ambient temperature fluctuations, pressure changes, and electromagnetic interference in the spectral signal. The fitted baseline component is subtracted from the initial spectral data to obtain the corrected spectral data with baseline interference removed.

[0082] Step S102: Based on the corrected spectral data, blind source separation is performed using the independent component analysis algorithm to obtain the independent spectral data corresponding to each partial discharge characteristic gas.

[0083] Furthermore, the corrected spectral data is used as input, and the independent component analysis algorithm is employed to perform blind source separation processing. Independent component analysis is a blind signal separation method that can decouple the overlapping absorption signals of different gases from the observed spectrum of multi-component mixtures under the condition that the complete prior information of the source signals of each component is unknown. By solving the optimal separation matrix through matrix iteration, the mixed spectrum is decomposed into multiple statistically independent single-component spectral components, and finally the independent spectral data corresponding to each partial discharge characteristic gas are obtained, thus eliminating the spectral cross-interference between multi-component gases from the root.

[0084] Step S103: Calculate the baseline drift based on the absorption peak characteristics of the independent spectral data, reconstruct the spectrum using partial least squares concentration inversion and calculate the spectral residuals, and perform signal enhancement on the reconstructed spectra with excessive spectral residuals to obtain enhanced spectral data.

[0085] Based on independent spectral data, the spectral absorption peak characteristics of the corresponding gas are extracted. The baseline drift is calculated according to the spectral values ​​within the preset baseline interval on both sides of the absorption peak to evaluate the baseline deviation level remaining after blind source separation. The partial least squares algorithm is used to perform concentration inversion calculation on the independent spectral data to obtain the concentration inversion matrix of each component. Based on the concentration inversion results, the standard reconstructed spectrum of the corresponding component is generated. The spectral residual between the reconstructed spectrum and the original independent spectral data is calculated in the entire band range to characterize the degree of residual noise and interference in the signal. If the spectral residual is greater than the preset residual threshold, it indicates that the trace fault characteristic signal is masked by noise. In this case, signal enhancement processing is performed on the reconstructed spectrum to improve the signal-to-noise ratio of the trace characteristic absorption signal, and finally, enhanced spectral data is obtained.

[0086] Step S104: Based on the enhanced spectral data, extract the absorption peak characteristic values ​​of each partial discharge characteristic gas, determine whether the absorption peak characteristic values ​​meet the preset characteristic matching conditions, and if so, determine the actual concentration values ​​of each partial discharge characteristic gas.

[0087] Based on enhanced spectral data, characteristic values ​​of the absorption peaks corresponding to each partial discharge characteristic gas are extracted, including quantitative parameters such as absorption peak position, peak height, peak area, and full width at half maximum (FWHM). The extracted absorption peak characteristic values ​​are verified against preset characteristic matching conditions, which are pre-calibrated parameter threshold ranges for the standard absorption peaks of each characteristic gas, used to eliminate noise spurious peaks and interference peaks. When the absorption peak characteristic value meets the characteristic matching conditions, the absorption peak is determined to be an effective absorption peak of the target characteristic gas. Based on the Lambert-Beer law and the pre-calibrated concentration-absorbance calibration curve, the actual concentration values ​​of each partial discharge characteristic gas are calculated and determined.

[0088] Step S105: Obtain the actual concentration value, calculate the diagnostic contribution value corresponding to each partial discharge characteristic gas, and if the diagnostic contribution value is less than the preset contribution threshold, remove the corresponding partial discharge characteristic gas to obtain the screening concentration value of the target partial discharge characteristic gas.

[0089] The actual concentration values ​​of each partial discharge characteristic gas are obtained, and the diagnostic contribution value corresponding to each characteristic gas is calculated. The diagnostic contribution value is used to characterize the correlation between a single characteristic gas and the partial discharge fault. It is calculated as the ratio of the actual concentration of a single gas to the corresponding fault critical concentration, and then normalized by the sum of the ratios of all characteristic gases. The diagnostic contribution value of each characteristic gas is compared with a preset contribution threshold. If the diagnostic contribution value is less than the preset contribution threshold, it indicates that the gas has a weak correlation with the current partial discharge fault and belongs to the interference component. In this case, the corresponding partial discharge characteristic gas is removed, and the component with a strong correlation is retained as the target partial discharge characteristic gas, and the corresponding screening concentration value is output.

[0090] Step S106: Input the selected concentration values ​​into the random forest model, calculate the diagnostic weight values ​​of the partial discharge characteristic gases of each target, and obtain the diagnostic weight data.

[0091] The concentration values ​​of the target partial discharge characteristic gases are input into a pre-trained random forest model to calculate the diagnostic weight values ​​of each target partial discharge characteristic gas. The random forest model quantifies the importance of fault features by splitting nodes of multiple decision trees. Based on the decrease in Gini index before and after node splitting, the average importance of each characteristic gas in all decision trees is calculated to obtain the fault correlation degree of each gas. After threshold screening and normalization, the diagnostic weight values ​​corresponding to each target characteristic gas are obtained. All diagnostic weight values ​​are integrated to form diagnostic weight data, realizing the adaptive allocation of fault diagnosis weights.

[0092] Step S107: Based on the diagnostic weight data and the screening concentration values, calculate the insulation life attenuation of partial discharge to obtain the fault status judgment result of the gas-insulated metal-enclosed switchgear.

[0093] Based on diagnostic weight data and screened concentration values, the concentration data of each target characteristic gas are weighted and fused to construct a partial discharge fault feature vector. Combining the discharge mode classification model with continuous time-series historical detection data, the equipment insulation degradation trend index is calculated, and the insulation life attenuation corresponding to partial discharge is further quantitatively calculated. Finally, the fault status judgment result of gas-insulated metal-enclosed switchgear, which includes discharge mode classification results, fault severity level, insulation life attenuation, and warning level, is generated, providing a quantitative basis for equipment insulation status assessment and operation and maintenance decisions.

[0094] The aforementioned multi-component spectral detection method for SF6 decomposition products in GIS partial discharge faults first acquires initial spectral data of the SF6 decomposition mixture in the GIS three-phase gas chamber, and obtains corrected spectral data through support vector machine baseline correction. Then, through blind source separation using independent component analysis (ICM) algorithm, independent spectral data of each partial discharge characteristic gas are obtained. After baseline drift calculation, partial least squares concentration inversion to reconstruct the spectrum, and residual judgment, signal enhancement is performed on spectra with excessive residuals to obtain enhanced spectral data. Absorption peak feature values ​​are extracted and matched to determine the actual concentration of each characteristic gas. The target characteristic gas concentration is obtained through diagnostic contribution threshold screening, and input into a random forest model to calculate diagnostic weight data. Finally, the insulation life attenuation is calculated by combining the weights and concentrations, and the GIS equipment fault status judgment result is output. This method is adaptable to complex operating conditions in GIS environments. Baseline correction and signal enhancement improve the anti-interference capability of spectral data, eliminating cross-interference of multi-component spectra and achieving accurate quantification. Adaptive feature screening and weight allocation improve diagnostic accuracy, while quantitatively assessing the degree of insulation degradation, providing data support for condition-based maintenance of GIS equipment.

[0095] In one embodiment, blind source separation is performed using an independent component analysis algorithm based on the corrected spectral data to obtain independent spectral data corresponding to each partial discharge characteristic gas. This may include the following steps:

[0096] Step S201: Construct an initial spectral matrix based on the corrected spectral data.

[0097] For example, the calibration spectral data obtained from multiple consecutive acquisitions are acquired and arranged in a two-dimensional manner with wavelength sampling points as the row dimension and sampling time sequence as the column dimension to generate an initial spectral matrix; where the size of the row dimension is equal to the total number of wavelength sampling points in the target detection band, the size of the column dimension is equal to the total number of consecutive samplings, and each column vector corresponds to the full-band calibration spectral absorbance value sequence obtained from one sampling.

[0098] Step S202: The initial spectral matrix is ​​iteratively calculated using the independent component analysis algorithm to obtain the target separation matrix.

[0099] Using the initial spectral matrix as input, a fast independent component analysis (CFI) algorithm is employed for iterative optimization. By updating the separation vectors round by round, the statistical independence between the output components obtained through the separation vector mapping is continuously improved. The iteration process terminates when the difference between two adjacent separation vectors is less than a preset convergence threshold. After convergence, all converged separation vectors are combined row by row to obtain the target separation matrix. The target separation matrix is ​​used to establish a linear mapping relationship between the mixed observation spectrum and the independent source signals.

[0100] Step S203: Perform blind source separation on the initial spectral matrix according to the target separation matrix to obtain candidate independent components.

[0101] By performing a linear transformation operation on the initial spectral matrix using the target separation matrix, the aliased multi-component absorption spectral signals are decoupled, resulting in several statistically independent one-dimensional signal components, i.e., candidate independent components. Each candidate independent component corresponds to a potential independent signal source. Since this stage only completes signal separation based on statistical independence and the correspondence between the components and the actual partial discharge characteristic gases has not yet been verified, they are defined as candidate components.

[0102] Step S204: Matching and judgment are performed by calculating the Pearson correlation coefficient between each candidate independent component and the target absorption feature corresponding to each partial discharge characteristic gas in the absorption feature library.

[0103]

[0104] in, Indicates the first The candidate independent component and the first The correlation coefficient of the absorption characteristics of each target Indicates the first The candidate independent components in the th... The value at each wavelength point Indicates the first The average of the candidate independent components, Indicates the first The target absorption feature in the first The value at each wavelength point Indicates the first The average value of the absorption characteristics of each target.

[0105] Optionally, the absorption feature library is a pre-built database that stores the target absorption spectral features corresponding to each partial discharge characteristic gas.

[0106] Step S205: If the candidate independent component matches any target absorption feature, then the candidate independent component is determined to be the independent spectral data corresponding to the partial discharge characteristic gas.

[0107] Specifically, an initial spectral matrix is ​​constructed based on the calibrated spectral data. An independent component analysis (ICA) algorithm is then used to iteratively calculate the initial spectral matrix to obtain the target separation matrix. Blind source separation is then performed on the initial spectral matrix using the target separation matrix to obtain several candidate independent components, thus completing the initial decoupling of the multi-component aliased absorption spectra. A pre-constructed absorption feature library is introduced, which is a database storing the target absorption spectral features corresponding to each partial discharge characteristic gas. Matching is performed by calculating the Pearson correlation coefficient between each candidate independent component and each target absorption feature in the absorption feature library. If a candidate independent component matches any target absorption feature, then that candidate independent component is determined to be the independent spectral data of the corresponding partial discharge characteristic gas.

[0108] This embodiment utilizes the blind source separation capability of independent component analysis to decouple the overlapping absorption spectra of multi-component gases without requiring complete prior information, thereby reducing the impact of spectral cross-interference on quantitative results from the source. Simultaneously, by combining the absorption feature library with the Pearson correlation coefficient matching and verification mechanism, it can effectively screen out noise and spurious components generated during the separation process, ensuring that the output independent spectral data corresponds to the real partial discharge characteristic gas components.

[0109] In one embodiment, using an independent component analysis algorithm to perform iterative calculations on the initial spectral matrix to obtain the target separation matrix may include the following steps:

[0110] Step S301: Perform a centering process on the initial spectral matrix to obtain a centered spectral matrix:

[0111]

[0112] in, Represents a vector consisting entirely of 1s. Represents the initial spectral matrix. This indicates the total number of spectral wavelength sampling points. Indicates the number of consecutive samples. Indicates the first The corrected spectral vector obtained from the second sampling.

[0113] Step S302: Perform whitening processing on the centered spectral matrix to obtain a whitened spectral matrix:

[0114]

[0115] in, express eigenvector matrix, express The eigenvalue diagonal matrix.

[0116] Step S303: The FastICA iterative algorithm is used to solve for the separation vector in the whitening spectral matrix to obtain the target separation matrix.

[0117]

[0118]

[0119] in, Indicates the first The separation vector of the next iteration Represents a nonlinear function. , Represents the mathematical expectation operation, iterating to... When the time stops, the resulting separation vectors constitute the target separation matrix. , This indicates the number of gas types characteristic of partial discharge.

[0120] Specifically, the initial spectral matrix is ​​sequentially centered, whitened, and iteratively optimized to solve for the target separation matrix. First, the initial spectral matrix is ​​centered by eliminating the mean offset component of the spectral data using a corresponding formula, resulting in a centered spectral matrix. This effectively removes computational interference caused by fixed offsets in the spectral data. The initial spectral matrix is ​​constructed using spectral wavelength sampling points and the number of consecutive samplings as dimensions. Each vector corresponds to the corrected spectral data from a single sampling, and vectors containing all ones are used for matrix mean normalization. Next, the centered spectral matrix is ​​whitened by solving for the eigenvector matrix and eigenvalue diagonal matrix of the covariance matrix of the centered spectral matrix. A matrix transformation is then performed using a corresponding formula to obtain a whitened spectral matrix with normalized variance and no component correlation. This eliminates the linear correlation between dimensions of the original spectral data, simplifying the subsequent iterative solution for independent components. Finally, the FastICA iterative algorithm is used to solve for the separation vector in the whitened spectral matrix. The tanh function is selected as the nonlinear function for iterative calculation. The iterative separation vector is continuously updated, and the updated separation vector is normalized, ensuring that the difference between two adjacent iterations is less than 1. As a convergence termination condition, after the iteration is completed, all effective separation vectors are combined to construct a target separation matrix that is adapted to the number of types of partial discharge characteristic gases.

[0121] This embodiment effectively solves the problems of slow convergence speed, low iterative accuracy, and unstable separation effect of traditional independent component analysis algorithms. Matrix centering and whitening preprocessing can eliminate redundant offset errors and linear correlations in spectral data, significantly reducing the computational complexity of iterative operations and improving the overall computational efficiency of the algorithm. The high-precision iterative convergence threshold setting ensures the accuracy of the separation vector solution, enabling the final target separation matrix to have good signal decoupling capability. It can accurately adapt to the separation scenario of aliased spectra of multi-component SF6 decomposition products, effectively ensuring the accuracy and stability of subsequent multi-component independent spectral data extraction, and providing a reliable algorithmic foundation for subsequent spectral matching, gas concentration quantification, and fault diagnosis.

[0122] In one embodiment, enhanced spectral data can be generated through the following process:

[0123] Step S401: Extract the spectral absorption peak characteristics corresponding to the independent spectral data.

[0124] Optionally, the spectral absorption peak characteristics include absorption peak position, absorption peak height, absorption peak area, absorption peak half width at half maximum (FWHM), and absorption peak shape coefficient.

[0125] Step S402: Calculate the baseline drift based on the baseline regions on both sides of the spectral absorption peak characteristics.

[0126] The formula for calculating baseline drift is:

[0127]

[0128] in, Indicates the baseline drift amount. This indicates the starting wavelength of the baseline region corresponding to the spectral absorption peak characteristics. Indicates the termination wavelength of the baseline region. Indicates independent spectral data in The absorbance value at that point Indicates independent spectral data in The absorbance value at that location.

[0129] Step S403: If the baseline drift is greater than the preset drift threshold, the partial least squares algorithm is used to perform concentration inversion calculation on the independent spectral data to obtain the concentration inversion matrix.

[0130] The calculated baseline drift is compared with a preset drift threshold. When the baseline drift exceeds the threshold, it is determined that the corresponding independent spectral data is highly affected by baseline drift interference and quantitative correction is required. Using the full-band absorbance sequence of the independent spectral data as input, a partial least squares algorithm is used to perform multivariate regression. By extracting the principal components of the spectral and concentration variables and maximizing their covariance, the regression coefficients are solved to establish a quantitative mapping relationship between absorbance values ​​and gas concentrations, generating a concentration inversion matrix. This matrix can realize the rapid conversion of single-component independent spectra to corresponding characteristic gas concentrations.

[0131] Step S404: Generate the reconstructed spectrum based on the concentration inversion matrix.

[0132] Based on the concentration inversion matrix operation, the inversion concentration value of the corresponding characteristic gas is obtained. Combined with the pre-calibrated standard absorption spectrum parameters of the characteristic gas, a theoretical absorption spectrum without baseline interference and noise at the corresponding concentration is generated through linear mapping operation, which is the reconstructed spectrum. The reconstructed spectrum characterizes the absorbance distribution law of the target component under ideal detection conditions and can be used as a reference signal.

[0133] Step S405: Calculate the spectral residual between the reconstructed spectrum and the independent spectral data.

[0134]

[0135] in, Indicates spectral residual, Represents the first independent spectral data matrix. The first sample absorbance values ​​at each wavelength point Represents the first element in the reconstructed spectral matrix. The first sample Absorbance values ​​at each wavelength point.

[0136] Step S406: If the spectral residual is greater than the preset residual threshold, then perform signal enhancement processing on the reconstructed spectrum to obtain enhanced spectral data.

[0137] The calculated spectral residual is compared with a preset residual threshold. When the spectral residual exceeds the threshold, it is determined that the signal-to-noise ratio of the original independent spectrum is too low and the trace feature absorption signal is easily masked by background noise. Using the reconstructed spectrum as a feature reference template, signal enhancement processing is performed on the effective absorption band of the original independent spectrum. By suppressing background noise components and amplifying the amplitude of feature absorption signals, the spectral signal-to-noise ratio is improved. After processing, the enhanced spectral data is output.

[0138] Specifically, the spectral absorption peak features corresponding to the independent spectral data are extracted. These features include the absorption peak position, peak height, peak area, full width at half maximum (FWHM), and peak shape coefficient. Baseline drift is calculated based on the baseline regions on both sides of the absorption peak features. If the baseline drift exceeds a preset drift threshold, a partial least squares algorithm is used to perform concentration inversion calculations on the independent spectral data to obtain a concentration inversion matrix. Reconstructed spectra of the corresponding components are then generated based on this matrix. The spectral residual between the reconstructed spectrum and the independent spectral data is calculated. If the spectral residual exceeds a preset residual threshold, signal enhancement processing is performed on the reconstructed spectrum to obtain enhanced spectral data.

[0139] This embodiment can screen spectral data with baseline interference by baseline drift amount before performing concentration inversion, reducing invalid calculations and improving overall processing efficiency; based on partial least squares concentration inversion and spectral residual calculation, it can quantify the noise level of spectral signals and accurately identify weak fault features masked by noise; the on-demand signal enhancement processing only performs feature enhancement on low signal-to-noise ratio spectra with excessive residuals, which can improve the signal intensity of trace decomposition products while avoiding over-correction of high-quality spectra, ensuring the accuracy of subsequent concentration quantification results and providing reliable data support for fault diagnosis.

[0140] In one embodiment, the expression for the diagnostic contribution value corresponding to each partial discharge characteristic gas is:

[0141]

[0142] in, Indicates the first The diagnostic contribution values ​​corresponding to the characteristic gases of partial discharge. Indicates the first The actual concentration values ​​of the characteristic gases of partial discharge. Indicates the first Preset critical fault concentration values ​​for a type of partial discharge characteristic gas. This represents the total number of types of gases characteristic of partial discharge.

[0143] The diagnostic contribution calculation method in this embodiment eliminates the differences in concentration magnitude and dimensions between different characteristic gases, and can objectively quantify the correlation between each component and the partial discharge fault, providing a clear quantitative basis for subsequent characteristic gas screening. Through this contribution index, interfering components with weak correlation to the current fault can be accurately identified and eliminated, simplifying the input feature dimensions of the subsequent diagnostic model, reducing the interference of invalid components on the diagnostic results, and adapting to the dynamic changes in the contribution of each characteristic component under different fault types, thereby improving the accuracy and scenario adaptability of the fault diagnosis results.

[0144] In one embodiment, inputting the selected concentration values ​​into a random forest model to calculate the diagnostic weight values ​​for each target partial discharge characteristic gas, thereby obtaining diagnostic weight data, may include the following steps:

[0145] Step S501: Input the selected concentration values ​​into the pre-trained random forest model. When performing feature splitting at each node of all decision trees in the random forest model, calculate the Gini index decrease corresponding to each target partial discharge characteristic gas.

[0146]

[0147] in, Indicates the first The characteristic gas of the target partial discharge in the first The decrease in Gini index corresponding to feature splitting of each decision tree node. Indicates the first before the split The Gini index of each node, , This indicates the total number of partial discharge fault types. Indicates the first Among the nodes belonging to the node The proportion of samples with this type of fault to the total number of samples in the node. Indicates the first before the split The total number of samples at each node. This represents the number of samples in the left child node after the split. The Gini index of the left child node after the split. This represents the number of samples in the right child node after the split. This represents the Gini index of the right child node after the split.

[0148] Indicatively, the concentration values ​​of the target partial discharge characteristic gases obtained by diagnostic contribution screening are used as input and imported into a random forest model pre-trained using GIS partial discharge historical fault samples. The model consists of multiple independently constructed decision trees. During the node splitting process, each decision tree classifies and purifies fault samples by introducing gas concentration features. The difference in the Gini index of the node before and after feature splitting is the decrease in the Gini index of the corresponding gas feature, which is used to quantify the information gain and contribution of the gas feature to the classification of fault samples.

[0149] Step S502: Take the average value of the Gini index decrease of all nodes in all decision trees corresponding to the partial discharge characteristic gases of each target to obtain the gas correlation degree corresponding to the partial discharge characteristic gases of each target.

[0150] Furthermore, since the construction process of a single decision tree involves randomness in sample sampling and feature selection, the decrease in the Gini index of a single node cannot stably characterize the overall fault correlation of the gas. Therefore, we traverse all split nodes of all decision trees in the random forest and perform an arithmetic average of the decrease in the Gini index of all gases corresponding to a single target partial discharge characteristic gas. This eliminates the random fluctuation bias of a single node or a single tree and obtains a quantitative index that can stably and objectively reflect the overall correlation between the gas and the partial discharge fault, namely, the gas correlation degree.

[0151] Step S503: Determine whether the correlation degree of each gas is greater than the preset correlation degree threshold.

[0152] The gas correlation degree corresponding to each target partial discharge characteristic gas is compared with the pre-calibrated correlation degree threshold one by one to complete the secondary screening of fault diagnosis features; the pre-calibrated correlation degree threshold is set according to the fault diagnosis accuracy requirements and historical sample statistical results, and is used to identify weakly correlated components that contribute very little to fault classification.

[0153] Step S504: If the gas correlation degree is greater than the preset correlation degree threshold, then the gas correlation degree is normalized to obtain the diagnostic weight value corresponding to each target partial discharge characteristic gas.

[0154] For target partial discharge characteristic gases whose gas correlation exceeds a preset threshold, their correlation values ​​are summed and normalized to convert the correlation values, which represent the absolute degree of correlation, into weight parameters representing the relative contribution ratio. The calculation formula is as follows: ,in Indicates the first Diagnostic weighting values ​​for target partial discharge characteristic gases. Indicates the first The correlation between the gases This represents the total number of target partial discharge characteristic gases retained; after normalization, the sum of the diagnostic weight values ​​of all retained gases is 1, which can be directly adapted to the computational needs of subsequent weighted fusion of multi-component concentration data.

[0155] Step S505: Generate diagnostic weight data based on the values ​​of each diagnostic weight.

[0156] A one-to-one mapping relationship is established between the retained partial discharge characteristic gases of each target and the corresponding diagnostic weight values, forming a structured diagnostic weight dataset containing gas identifiers and corresponding weight parameters. This dataset can be directly used as the basis for weighted calculations in subsequent insulation life attenuation calculations and fault severity level assessments.

[0157] Specifically, the selected concentration values ​​are input into a pre-trained random forest model. During the feature splitting process at each node of all decision trees in the model, the decrease in Gini index corresponding to each target partial discharge characteristic gas is calculated. Based on this, the average decrease in Gini index of each target partial discharge characteristic gas at all nodes of all decision trees is taken to obtain the gas correlation degree corresponding to each gas. The gas correlation degree is compared with a preset correlation degree threshold. Gas correlation degrees greater than the threshold are normalized to obtain the diagnostic weight value corresponding to each target partial discharge characteristic gas. Finally, all diagnostic weight values ​​are integrated to generate diagnostic weight data.

[0158] This embodiment uses a weighting mechanism based on the decrease in the Gini index. It can rely on the ensemble learning characteristics of random forests to objectively quantify the contribution of each target feature gas to the identification of partial discharge fault types, avoiding the subjective bias of fixed weight criteria. After filtering by correlation threshold, weakly correlated feature components can be further eliminated, simplifying the diagnostic input dimensions. The normalized diagnostic weights can adaptively match the differences in component contributions under different fault scenarios, improving the anti-interference ability and classification accuracy of the fault diagnosis model.

[0159] In one embodiment, the insulation life attenuation of partial discharge is calculated based on diagnostic weight data and screening concentration values ​​to obtain the fault status determination result of the gas-insulated metal-enclosed switchgear. This may include the following steps:

[0160] Step S601: Construct a partial discharge feature vector by weighting the diagnostic weight data and the selected concentration values.

[0161] Furthermore, using the selected target partial discharge characteristic gases as feature dimensions, the corresponding gas concentration values ​​are multiplied element-wise by the weight coefficients in the diagnostic weight data. After weighted mapping, a one-dimensional partial discharge feature vector is constructed, the expression of which is: ,in Represents the characteristic vector of partial discharge. For the first Diagnostic weights for target characteristic gases This represents the screening concentration value for the corresponding gas. This represents the total number of target characteristic gases. Weighted processing can amplify the contribution of components with high fault correlation to the feature vector, weaken the interference of components with low correlation, and make the feature vector more closely match the requirements for fault classification.

[0162] Step S602: Input the partial discharge feature vector into the pre-trained support vector machine classification model to obtain the discharge pattern classification result.

[0163] The support vector machine classification model is trained using historical feature samples of multiple typical partial discharge faults. The input low-dimensional feature vectors are mapped to a high-dimensional feature space through a kernel function to solve for the optimal classification hyperplane among different discharge types. After the partial discharge feature vectors are input into the model, the corresponding discharge mode classification results are output after feature space mapping and classification decision surface determination. These include typical fault types such as metal tip discharge, floating potential body discharge, air gap discharge, and insulator surface discharge, thus realizing the qualitative identification of partial discharge fault types within GIS.

[0164] Step S603: Combine the classification results of multiple consecutive discharge modes with the corresponding actual concentration values ​​to extract the degradation trend index.

[0165] Discharge mode classification results and time-series data of actual concentrations of each characteristic gas were collected over N consecutive detection periods. First, the number of consecutive periods in which the same discharge mode occurred was counted to obtain the mode persistence coefficient. Then, the time-series rate of change of concentration of each characteristic gas was fitted using the least squares method. The average rate of change of concentration was weighted and fused with the mode persistence coefficient to calculate the degradation trend index, the expression of which is: .in As a deterioration trend indicator, For the first The change in concentration of a gas per unit time. The number of cycles in the same discharge mode. To calculate the total number of cycles, , This is a preset weighting coefficient. This indicator can quantitatively characterize the rate of insulation degradation and the stability of fault modes, eliminating the risk of misjudgment caused by fluctuations in a single test.

[0166] Step S604: Determine whether the degradation trend index is greater than the preset degradation trend threshold.

[0167] Step S605: If the degradation trend index is greater than the preset degradation trend threshold, the insulation life attenuation of the gas-insulated metal-enclosed switchgear is calculated based on the degradation trend index.

[0168] If the degradation trend index exceeds the preset degradation trend threshold, the insulation life attenuation of the gas-insulated metal-enclosed switchgear is calculated based on the degradation trend index. The degradation trend index is compared with the preset degradation trend threshold. When the index exceeds the threshold, it is determined that the insulation degradation rate exceeds the normal aging level, posing a risk of accelerated degradation. Based on the baseline remaining insulation life under normal operating conditions, and considering the magnitude and duration of the degradation trend index exceeding the threshold, the additional life loss due to abnormal degradation, i.e., the insulation life attenuation, is calculated, expressed as follows: ,in This represents the insulation life attenuation. The reference remaining insulation life of the equipment. To preset the degradation trend threshold, This represents the duration of abnormal degradation. This quantitative result can intuitively reflect the degree of damage that partial discharge inflicts on the insulation life of equipment.

[0169] Step S606: Based on the insulation life attenuation, generate the fault status determination result of the gas-insulated metal-enclosed switchgear.

[0170] Optionally, the fault status determination results include discharge mode classification results, insulation life attenuation, fault severity level, and warning level.

[0171] Based on the insulation life attenuation, a fault status assessment result is generated for gas-insulated metal-enclosed switchgear. Integrating two core data points—discharge mode classification results and insulation life attenuation—and matching them with preset fault level classification rules, the severity level and warning level of the current fault are determined. The fault severity level is divided into four levels: minor, moderate, severe, and critical; the warning level is divided into four levels: blue, yellow, orange, and red. The final output is a structured fault status assessment result containing the discharge mode classification result, insulation life attenuation, fault severity level, and warning level, providing equipment maintenance personnel with a multi-dimensional basis for insulation status assessment.

[0172] Specifically, based on diagnostic weight data and selected concentration values, the selected concentration values ​​are weighted using the diagnostic weights corresponding to the characteristic gases of each target partial discharge as weighting coefficients to construct a partial discharge feature vector. This feature vector is then input into a pre-trained support vector machine classification model, and after high-dimensional feature space mapping and classification decision surface determination, the discharge mode classification result is obtained. Combining the discharge mode classification results and corresponding actual concentration values ​​from multiple consecutive time-series detections, a degradation trend index is extracted through time-series trend fitting. This index is used to quantitatively characterize the development rate and evolution degree of insulation degradation within the equipment. The degradation trend index is compared with a preset degradation trend threshold. If the index is greater than the threshold, the insulation life attenuation of the gas-insulated metal-enclosed switchgear is calculated based on the degradation trend index and the equipment insulation life benchmark model. Finally, the discharge mode classification results and insulation life attenuation are integrated, matched with the corresponding fault severity level and warning level, to generate a fault status determination result containing multi-dimensional diagnostic information.

[0173] This embodiment constructs a complete analysis chain from qualitative identification of discharge modes to quantitative assessment of insulation degradation. The construction of weighted feature vectors highlights the diagnostic value of highly correlated characteristic gases, and the support vector machine classification model ensures the accuracy of discharge mode identification. The degradation trend indicators extracted based on continuous time-series data can reflect the dynamic development trend of insulation faults and reduce the interference of single detection fluctuations on diagnostic results. The quantitative output of insulation life attenuation and the grading judgment mechanism can output diagnostic conclusions from multiple dimensions such as fault type, severity, and development trend, providing comprehensive data support for the insulation condition assessment and condition-based maintenance strategy formulation of GIS equipment.

[0174] In one embodiment, such as Figure 2 As shown, this application also provides a multi-component spectroscopic detection system for SF6 decomposition products in GIS partial discharge faults. The system may include:

[0175] The spectral preprocessing module 701 is used to acquire the initial spectral data of the SF6 decomposition mixture in the three-phase gas chamber of the gas-insulated metal-enclosed switchgear, and to perform baseline correction processing on the initial spectral data through a support vector machine to obtain corrected spectral data.

[0176] The spectral unmixing enhancement module 702 is used to perform blind source separation based on the corrected spectral data using an independent component analysis algorithm to obtain independent spectral data corresponding to each partial discharge characteristic gas; it is also used to calculate the baseline drift based on the absorption peak characteristics of the independent spectral data, reconstruct the spectrum through partial least squares concentration inversion and calculate the spectral residual, and perform signal enhancement on the reconstructed spectrum with excessive spectral residual to obtain enhanced spectral data.

[0177] The concentration quantitative screening module 703 is used to extract the absorption peak characteristic values ​​of each partial discharge characteristic gas based on the enhanced spectral data, determine whether the absorption peak characteristic values ​​meet the preset characteristic matching conditions, and if so, determine the actual concentration value of each partial discharge characteristic gas; it is also used to obtain the actual concentration value, calculate the diagnostic contribution value corresponding to each partial discharge characteristic gas, and if the diagnostic contribution value is less than the preset contribution threshold, remove the corresponding partial discharge characteristic gas to obtain the screening concentration value of the target partial discharge characteristic gas.

[0178] The fault diagnosis and assessment module 704 is used to input the screening concentration values ​​into the random forest model, calculate the diagnostic weight values ​​of the characteristic gases of partial discharge for each target, and obtain diagnostic weight data; it is also used to calculate the insulation life attenuation of partial discharge based on the diagnostic weight data and the screening concentration values, and obtain the fault status judgment result of the gas-insulated metal-enclosed switchgear.

[0179] The aforementioned multi-component spectral detection system for SF6 decomposition products in GIS partial discharge faults consists of a spectral preprocessing module, a spectral demixing and enhancement module, a concentration quantitative screening module, and a fault diagnosis and assessment module, which sequentially handle data flow. The spectral preprocessing module acquires the initial spectral data of the SF6 decomposition mixture in the three-phase gas chamber of a gas-insulated metal-enclosed switchgear. It performs baseline correction processing on the initial spectral data using a support vector machine, outputting corrected spectral data and completing baseline interference suppression and data quality preprocessing of the original spectrum. The spectral demixing and enhancement module receives the corrected spectral data, performs blind source separation using an independent component analysis algorithm, decouples the overlapping absorption spectra of the multi-component gases, and obtains independent spectral data corresponding to each partial discharge characteristic gas. It further calculates the baseline drift based on the absorption peak characteristics of the independent spectral data, generates a reconstructed spectrum through partial least squares concentration inversion, and calculates the spectral residual. For reconstructed spectra with spectral residuals exceeding a preset threshold, it performs signal enhancement processing, outputting enhanced spectral data to improve the signal-to-noise ratio of trace fault characteristic signals. The concentration quantification screening module receives enhanced spectral data, extracts the absorption peak characteristics of each partial discharge characteristic gas, and determines the actual concentration value of each characteristic gas after verification under preset characteristic matching conditions. It calculates the diagnostic contribution value corresponding to each characteristic gas, removes characteristic gases with weak correlation to the fault by using a preset contribution threshold, and outputs the target partial discharge characteristic gas and its screened concentration value, completing the quantitative concentration calculation and secondary screening of fault features. The fault diagnosis and assessment module receives the screened concentration values, inputs them into a random forest model to calculate the diagnostic weight value of each target characteristic gas, and generates diagnostic weight data. Combining the diagnostic weight data and the screened concentration values, it calculates the insulation life attenuation corresponding to the partial discharge, and finally outputs the fault status judgment result of the gas-insulated metal-enclosed switchgear.

[0180] This embodiment is adaptable to the SF6 decomposition product detection requirements under complex on-site GIS conditions. The two-level baseline correction and signal enhancement mechanism effectively suppresses environmental interference, improves spectral data quality and the ability to identify weak characteristic signals; the blind source separation method of independent component analysis eliminates spectral cross-interference of multi-component gases at its source, ensuring the accuracy of simultaneous quantitative detection of multiple components; the mechanism combining contribution screening and random forest adaptive weight allocation simplifies diagnostic feature dimensions, optimizes component diagnostic weights, and improves the anti-interference capability and accuracy of fault diagnosis; quantitative insulation life attenuation assessment and multi-dimensional fault status output provide reliable data support for insulation status assessment and condition-based maintenance strategy formulation for GIS equipment.

[0181] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0182] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the multi-component spectroscopic detection method for SF6 decomposition products of GIS partial discharge faults as described above.

[0183] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0184] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0185] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A multi-component spectroscopic detection method for SF6 decomposition products in GIS partial discharge faults, characterized in that, The method includes: Initial spectral data of SF6 decomposition mixture in the three-phase gas chamber of a gas-insulated metal-enclosed switchgear are obtained, and baseline correction data is performed on the initial spectral data using a support vector machine to obtain corrected spectral data. Based on the corrected spectral data, blind source separation is performed using an independent component analysis algorithm to obtain independent spectral data corresponding to each partial discharge characteristic gas. The baseline drift is calculated based on the absorption peak characteristics of the independent spectral data. The spectrum is reconstructed by partial least squares concentration inversion and the spectral residual is calculated. Signal enhancement is performed on the reconstructed spectrum with the spectral residual exceeding the standard to obtain enhanced spectral data. Based on the enhanced spectral data, the absorption peak characteristic values ​​of each of the partial discharge characteristic gases are extracted, and it is determined whether the absorption peak characteristic values ​​meet the preset characteristic matching conditions. If they do, the actual concentration values ​​of each of the partial discharge characteristic gases are determined. The actual concentration value is obtained, and the diagnostic contribution value corresponding to each partial discharge characteristic gas is calculated. If the diagnostic contribution value is less than the preset contribution threshold, the corresponding partial discharge characteristic gas is removed to obtain the screening concentration value of the target partial discharge characteristic gas. The selected concentration values ​​are input into a random forest model to calculate the diagnostic weight values ​​of each target partial discharge characteristic gas, thereby obtaining diagnostic weight data. Based on the diagnostic weight data and the screening concentration value, the insulation life attenuation of partial discharge is calculated to obtain the fault status judgment result of the gas-insulated metal-enclosed switchgear.

2. The method according to claim 1, characterized in that, The step of performing blind source separation using an independent component analysis algorithm based on the corrected spectral data to obtain independent spectral data corresponding to each partial discharge characteristic gas includes: Based on the corrected spectral data, an initial spectral matrix is ​​constructed; The initial spectral matrix is ​​iteratively calculated using the independent component analysis algorithm to obtain the target separation matrix; Blind source separation is performed on the initial spectral matrix based on the target separation matrix to obtain candidate independent components; Matching and judgment are performed by calculating the Pearson correlation coefficient between each candidate independent component and the target absorption feature corresponding to each partial discharge characteristic gas in the absorption feature library: in, Indicates the first The candidate independent component and the first The correlation coefficient of the absorption characteristics of each target Indicates the first The candidate independent components in the th... The value at each wavelength point Indicates the first The average of the candidate independent components, Indicates the first The target absorption feature in the first The value at each wavelength point Indicates the first The average value of the absorption characteristics of each target; The absorption feature library is a pre-constructed database that stores the target absorption spectral features corresponding to each of the partial discharge feature gases; If the candidate independent component matches any of the target absorption features, then the candidate independent component is determined to be the independent spectral data corresponding to the partial discharge characteristic gas.

3. The method according to claim 2, characterized in that, The step of using independent component analysis (ICA) to iteratively calculate the initial spectral matrix to obtain the target separation matrix includes: The initial spectral matrix is ​​centered to obtain the centered spectral matrix: in, Represents a vector consisting entirely of 1s. Represents the initial spectral matrix. This indicates the total number of spectral wavelength sampling points. Indicates the number of consecutive samples. Indicates the first The corrected spectral vector obtained from the second sampling; The centered spectral matrix is ​​whitened to obtain a whitened spectral matrix: in, express eigenvector matrix, express The eigenvalue diagonal matrix; The FastICA iterative algorithm is used to solve for the separation vector in the whitening spectral matrix to obtain the target separation matrix: in, Indicates the first The separation vector of the next iteration Represents a nonlinear function. , Represents the mathematical expectation operation, iterating to... When the time stops, the resulting separation vectors constitute the target separation matrix. , This indicates the number of gas types characteristic of partial discharge.

4. The method according to claim 1, characterized in that, The enhanced spectral data is generated through the following process: Extract the spectral absorption peak features corresponding to the independent spectral data; The spectral absorption peak characteristics include absorption peak position, absorption peak height, absorption peak area, absorption peak half width at half maximum (FWHM), and absorption peak shape coefficient. The baseline shift is calculated based on the baseline regions on both sides of the spectral absorption peak characteristics. The formula for calculating the baseline drift is: in, This indicates the baseline drift amount. The wavelength represents the starting wavelength of the baseline region corresponding to the spectral absorption peak characteristics. This indicates the termination wavelength of the baseline region. This indicates that the independent spectral data is in The absorbance value at that point This indicates that the independent spectral data is in The absorbance value at that location; If the baseline drift is greater than a preset drift threshold, then the partial least squares algorithm is used to perform concentration inversion calculation on the independent spectral data to obtain the concentration inversion matrix; A reconstructed spectrum is generated based on the concentration inversion matrix; The spectral residual between the reconstructed spectrum and the independent spectral data was calculated. in, This represents the spectral residual. Represents the first independent spectral data matrix. The first sample absorbance values ​​at each wavelength point In the reconstructed spectral matrix, the first... The first sample Absorbance values ​​at each wavelength point; If the spectral residual is greater than a preset residual threshold, then signal enhancement processing is performed on the reconstructed spectrum to obtain enhanced spectral data.

5. The method according to claim 1, characterized in that, The expression for the diagnostic contribution value corresponding to each of the partial discharge characteristic gases is as follows: in, Indicates the first The diagnostic contribution values ​​corresponding to the characteristic gases of partial discharge. Indicates the first The actual concentration values ​​of the characteristic gases of partial discharge. Indicates the first Preset critical fault concentration values ​​for a type of partial discharge characteristic gas. This represents the total number of types of gases characteristic of partial discharge.

6. The method according to claim 1, characterized in that, The step involves inputting the selected concentration values ​​into a random forest model to calculate the diagnostic weight values ​​for each of the target partial discharge characteristic gases, thereby obtaining diagnostic weight data, including: The selected concentration values ​​are input into a pre-trained random forest model. When performing feature splitting at each node of all decision trees in the random forest model, the decrease in Gini index corresponding to each of the target partial discharge characteristic gases is calculated: in, Indicates the first The characteristic gas of the target partial discharge in the first The decrease in Gini index corresponding to feature splitting of each decision tree node. Indicates the first before the split The Gini index of each node, , This indicates the total number of partial discharge fault types. Indicates the first Among the nodes belonging to the node The proportion of samples with this type of fault to the total number of samples in the node. Indicates the first before the split The total number of samples at each node. This represents the number of samples in the left child node after the split. The Gini index of the left child node after the split. This represents the number of samples in the right child node after the split. This represents the Gini index of the right child node after the split; The gas correlation degree corresponding to each target partial discharge characteristic gas is obtained by averaging the decrease of the Gini index of all nodes in all decision trees corresponding to each target partial discharge characteristic gas. Determine whether the correlation degree of each gas is greater than a preset correlation degree threshold; If the gas correlation degree is greater than the preset correlation degree threshold, then the gas correlation degree is normalized to obtain the diagnostic weight value corresponding to each of the target partial discharge characteristic gases. Based on the diagnostic weight values ​​described above, diagnostic weight data is generated.

7. The method according to claim 1, characterized in that, The step of calculating the insulation life attenuation of partial discharge based on the diagnostic weight data and the screening concentration value, and obtaining the fault status determination result of the gas-insulated metal-enclosed switchgear, includes: Based on the diagnostic weight data and the screening concentration values, a weighted partial discharge feature vector is constructed. The partial discharge feature vector is input into a pre-trained support vector machine classification model to obtain the discharge pattern classification result. By combining the classification results of multiple consecutive discharge modes with the corresponding actual concentration values, a degradation trend index is extracted. Determine whether the degradation trend index is greater than a preset degradation trend threshold; If the degradation trend index is greater than the preset degradation trend threshold, the insulation life attenuation of the gas-insulated metal-enclosed switchgear is calculated based on the degradation trend index. Based on the insulation life attenuation, a fault status determination result for the gas-insulated metal-enclosed switchgear is generated. The fault status determination results include discharge mode classification results, insulation life attenuation, fault severity level, and warning level.

8. A multi-component spectroscopic detection system for SF6 decomposition products in GIS partial discharge faults, characterized in that the system comprises: The spectral preprocessing module is used to acquire the initial spectral data of the SF6 decomposition mixture in the three-phase gas chamber of the gas-insulated metal-enclosed switchgear, and to perform baseline correction processing on the initial spectral data using a support vector machine to obtain corrected spectral data. The spectral unmixing and enhancement module is used to perform blind source separation using an independent component analysis algorithm based on the corrected spectral data to obtain independent spectral data corresponding to each partial discharge characteristic gas; it is also used to calculate the baseline drift based on the absorption peak characteristics of the independent spectral data, reconstruct the spectrum through partial least squares concentration inversion and calculate the spectral residual, and perform signal enhancement on the reconstructed spectrum with the spectral residual exceeding the standard to obtain enhanced spectral data. The concentration quantitative screening module is used to extract the absorption peak feature values ​​of each of the partial discharge feature gases based on the enhanced spectral data, determine whether the absorption peak feature values ​​meet the preset feature matching conditions, and if so, determine the actual concentration value of each of the partial discharge feature gases; it is also used to obtain the actual concentration value, calculate the diagnostic contribution value corresponding to each of the partial discharge feature gases, and if the diagnostic contribution value is less than the preset contribution threshold, remove the corresponding partial discharge feature gas to obtain the screening concentration value of the target partial discharge feature gas. The fault diagnosis and assessment module is used to input the selected concentration values ​​into a random forest model, calculate the diagnostic weight values ​​of each of the target partial discharge characteristic gases, and obtain diagnostic weight data; it is also used to calculate the insulation life decay of partial discharge based on the diagnostic weight data and the selected concentration values, and obtain the fault status judgment result of the gas-insulated metal-enclosed switchgear.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images