GIS equipment insulation fault diagnosis method based on infrared spectrum and machine learning

Abstract

This application discloses a method for diagnosing insulation faults in GIS equipment based on infrared spectroscopy and machine learning. The method includes: acquiring an interferometric time-domain signal containing infrared absorption characteristics of the gas being measured inside the GIS equipment; generating an infrared absorption spectrum based on the interferometric time-domain signal, and extracting the absorption peak characteristics of each target feature component from the infrared absorption spectrum to generate a spectral feature vector; inputting the spectral feature vector into a trained multi-group decoupling model to generate predicted concentration values ​​of each target feature component in the gas being measured; determining the fault type of the GIS equipment based on the predicted concentration values ​​of each target feature component, the concentration variation law, the concentration ratio between different target feature components, and a pre-established mapping model. The mapping model is used to characterize the correspondence between the concentration of each target feature component and its concentration variation law and the insulation fault type. This invention effectively eliminates multi-component spectral overlap interference, has strong anti-interference ability, and is miniaturized, making it suitable for online monitoring in substations.

Description

Technical Field

[0001] This application relates to the field of power equipment condition monitoring technology, and more specifically, to a method and apparatus for diagnosing insulation faults in GIS equipment based on machine learning. Background Technology

[0002] Gas-insulated switchgear (GIS) has advantages such as compact structure, high reliability, and good safety, and is widely used in power systems. Partial discharge is an important indicator of early insulation defects in GIS equipment. Under the influence of partial discharge or local overheating, the SF6 gas, the insulating medium inside the GIS, decomposes, generating various characteristic decomposition components such as SO2, H2S, SOF2, SO2F2, and CF4. The types, contents, and formation rates of these components are closely related to the type and severity of faults within the GIS. Therefore, assessing the operating status of GIS by detecting SF6 characteristic decomposition products has become a key means of discovering potential equipment faults and preventing serious insulation accidents.

[0003] Currently, the main detection methods for SF6 decomposition components, both domestically and internationally, include gas chromatography, electrochemical sensor methods, and infrared spectroscopy.

[0004] While gas chromatography offers high detection accuracy, it typically requires on-site sampling followed by laboratory analysis, resulting in long detection cycles and complex operations, making online continuous monitoring difficult. Electrochemical sensors, despite their small size and fast response, suffer from cross-interference between different components and are prone to "poisoning," leading to a significant decrease in sensitivity with repeated use and a short lifespan. Infrared spectroscopy, with its advantages of being non-contact, fast-response, and highly selective, is currently a research hotspot; however, existing infrared spectroscopy detection devices still face numerous technical challenges in engineering applications. (1) Poor environmental adaptability: Traditional Fourier transform infrared spectrometers (FTIR) and other spectroscopic equipment are bulky and extremely sensitive to strong electromagnetic interference and mechanical vibration at the substation site, making it difficult to operate stably under harsh conditions.

[0005] (2) Multi-component cross-interference: The absorption spectra of various characteristic gases produced by SF6 decomposition often exhibit severe overlap in the infrared band. Traditional non-dispersive infrared (NDIR) technology struggles to achieve accurate decoupling of multiple components without physical separation of the mixed gases, resulting in significant measurement errors and consequently affecting the accuracy of fault diagnosis.

[0006] (3) Low timeliness of online monitoring: Existing high-sensitivity detection technologies often rely on long optical path gas chambers, while traditional gas chambers have complex structures and slow gas exchange rates, making it difficult to meet the needs of real-time in-situ monitoring.

[0007] In summary, developing a miniaturized, highly interference-resistant, and algorithmically effective online diagnostic method for GIS insulation faults is a bottleneck problem that urgently needs to be solved in the field of power engineering. Summary of the Invention

[0008] To address at least one deficiency or improvement need in existing technologies, this invention provides a method and device for diagnosing insulation faults in GIS equipment based on infrared spectroscopy and machine learning. By combining miniaturized fiber optic interferometry with machine learning algorithms, it achieves high-precision online decoupled monitoring of the gas being measured in complex substation environments, thereby enabling early and accurate warning of GIS insulation faults. This overcomes the shortcomings of existing GIS insulation fault detection technologies, such as poor real-time performance of gas chromatography, susceptibility to poisoning and cross-interference of electrochemical sensors, and the large size and difficulty in handling multi-component spectral overlap of traditional infrared spectrometers.

[0009] To achieve the above objectives, according to a first aspect of the present invention, a method for diagnosing insulation faults in GIS equipment based on infrared spectroscopy and machine learning is provided, the method comprising: Acquire interferometric time-domain signals containing infrared absorption characteristics of the gas being measured inside the GIS equipment; Infrared absorption spectra are generated based on the interference time-domain signal, and the characteristics of each absorption peak are extracted from the infrared absorption spectrum to generate spectral feature vectors; the absorption peak characteristics include absorbance, peak height, peak area, and band ratio. The spectral feature vector is input into a pre-trained multi-group decoupling model to generate the concentration prediction values ​​of each target feature component in the gas being tested. The fault type of GIS equipment is determined based on the predicted concentration value, concentration change law, and concentration ratio between different target characteristic components and the pre-established mapping model. The mapping model is used to characterize the correspondence between the concentration of each target characteristic component and its concentration change law and the insulation fault type. The training process of the multi-group decoupling model includes: Acquire multiple first training samples with concentration labels, the first training samples including single-component standard gas samples with different concentrations of a single target feature component and multi-component standard gas samples composed of a mixture of multiple target feature components; The original interferometric time-domain signal of each of the first training samples is acquired and the corresponding original infrared absorption spectrum is generated. The original spectral feature vector is extracted from the original infrared absorption spectrum. The decoupling models are trained based on the original spectral feature vectors.

[0010] Furthermore, the above-mentioned GIS equipment insulation fault diagnosis method uses a decision tree classification model as its mapping model; The fault diagnosis feature vector generated based on the concentration prediction value, concentration change law and concentration ratio between different target feature components is input into the trained decision tree classification model to obtain the insulation fault type and the corresponding classification confidence.

[0011] Furthermore, the training process of the decision tree classification model in the above-mentioned GIS equipment insulation fault diagnosis method includes: Obtain multiple second training samples labeled with different insulation fault types; Obtain the concentration prediction value, concentration change law, and concentration ratio between different target feature components corresponding to each second training sample, and generate the original fault diagnosis feature vector. The decision tree classification model is trained based on the original fault diagnosis feature vector.

[0012] Furthermore, in the above-mentioned method for diagnosing insulation faults in GIS equipment, the acquisition of interferometric time-domain signals containing infrared absorption characteristics of the gas being measured inside the GIS equipment includes: An anti-resonant hollow fiber probe is connected to the gas chamber of the GIS equipment. After the gas inside the gas chamber diffuses evenly within the core of the anti-resonant hollow fiber probe, the gas inside the core is irradiated with an infrared laser. The time-domain interference signal generated by gas molecules absorbing the infrared laser is collected using a micro-interferometer.

[0013] Furthermore, the aforementioned GIS equipment insulation fault diagnosis method includes an anti-resonant hollow fiber probe comprising a single-mode fiber, an anti-resonant hollow fiber, and an end-face reflector. The anti-resonant hollow fiber contains multiple nested capillary hollow fibers.

[0014] Furthermore, the aforementioned method for diagnosing insulation faults in GIS equipment, which generates infrared absorption spectra based on interferometric time-domain signals, includes: Extract the upper and lower envelopes of the interference time-domain signal, and perform baseline drift correction based on the center lines of the upper and lower envelopes; A dynamic sliding window is used to perform mean filtering on the corrected signal; The filtered signal is reconstructed into an equally spaced light intensity-light sequence using an interpolation method. The light intensity-light sequence is then subjected to a Fourier transform to obtain the infrared absorption spectrum.

[0015] Furthermore, the above-mentioned GIS equipment insulation fault diagnosis method also includes: The absorption peak intensities of each component in the infrared absorption spectrum are self-calibrated using a temperature compensation coefficient. The temperature compensation formula for this self-calibration is as follows: Ccompensated =C raw ×(1+α×(T - T ref )) Among them, C compensated Indicates the calibrated absorption peak intensity; C raw Indicates the original absorption peak intensity; α represents the temperature coefficient; T represents the measured ambient temperature; T ref Indicates the reference temperature.

[0016] According to a second aspect of the present invention, this application also provides a GIS equipment insulation fault diagnosis device based on infrared spectroscopy and machine learning, comprising: The signal acquisition module is configured to acquire interferometric time-domain signals containing infrared absorption characteristics of the gas being measured inside the GIS equipment. The feature extraction module is configured to generate an infrared absorption spectrum based on the interference time-domain signal, extract the features of each absorption peak from the infrared absorption spectrum, and generate a spectral feature vector; the absorption peak features include absorbance, peak height, peak area, and band ratio. The concentration decoupling module is configured to input the spectral feature vector into a preset set of decoupling models to generate the concentration prediction values ​​of each target feature component in the gas being measured. The fault diagnosis module is configured to determine the fault type of the GIS equipment based on the predicted concentration values, concentration variation patterns, and concentration ratios between different target characteristic components, as well as a pre-established mapping model. The mapping model is used to characterize the correspondence between the concentration of each target characteristic component and its concentration variation patterns and the insulation fault type. The preset multi-group decoupling model is any one of the multi-group decoupling models described in the GIS equipment insulation fault diagnosis method.

[0017] Furthermore, the mapping model of the aforementioned GIS equipment insulation fault diagnosis device is a decision tree classification model; The fault diagnosis feature vector generated based on the concentration prediction value, concentration change law and concentration ratio between different target feature components is input into the trained decision tree classification model to obtain the insulation fault type and the corresponding classification confidence.

[0018] Furthermore, the signal acquisition module of the aforementioned GIS equipment insulation fault diagnosis device includes an anti-resonant hollow fiber probe and a miniature interferometer; The anti-resonant hollow fiber probe is connected to the gas chamber of the GIS equipment. After the gas inside the gas chamber diffuses evenly within the core of the anti-resonant hollow fiber probe, the gas inside the core is irradiated with an infrared laser. The micro-interferometer is used to collect the time-domain interference signal generated after gas molecules absorb the infrared laser.

[0019] Furthermore, in the aforementioned GIS equipment insulation fault diagnosis device, the anti-resonant hollow fiber probe includes a ceramic sleeve, an input / output multiplexed single-mode fiber, an anti-resonant hollow fiber, and a total reflection mirror disposed at the far end of the anti-resonant hollow fiber; a graded-index fiber lens is fused to one end of the single-mode fiber near the anti-resonant hollow fiber, and the exit end face of the graded-index fiber lens is coated with a partial reflective film. The incident assembly, consisting of a single-mode fiber and a graded-index fiber lens, is coupled to the anti-resonant hollow fiber using the ceramic sleeve with a gap. The graded-index fiber lens and the anti-resonant hollow fiber have a pre-reserved aperture at the docking position, so that the external gas to be tested can enter the central air core of the anti-resonant hollow fiber through the aperture.

[0020] According to a third aspect of the present invention, this application also provides an electronic device including at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program that, when executed by the processing unit, causes the processing unit to perform the steps of any of the above-described GIS equipment insulation fault diagnosis methods.

[0021] According to a fourth aspect of the present invention, this application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps in the GIS equipment insulation fault diagnosis method according to any one of the first aspects.

[0022] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) Strong decoupling capability of multi-component spectra: By using support vector regression (SVR) to perform nonlinear regression on overlapping absorption spectra in high-dimensional feature space, the measurement error caused by severe overlap of multi-component gas spectra in the mid-infrared band is effectively solved, and the synchronous concentration inversion of multiple characteristic decomposition gases is realized. Multiple component concentrations are output simultaneously in one sampling without physical separation.

[0023] (2) Miniaturization and high environmental adaptability: An integrated sensing probe is constructed by using an anti-resonant hollow fiber probe and a miniature Fizeau interferometer, integrating the gas absorption cell and the spectral detection cavity into the fiber. The probe volume can be controlled within the centimeter range. The fiber itself is unaffected by electromagnetic interference and has strong mechanical vibration resistance, making it fully adaptable to the harsh electromagnetic environment and mechanical disturbances at the substation site.

[0024] (3) High timeliness of online monitoring: The gas sampling method of direct fiber connection is adopted, which eliminates the need for offline sampling and chromatographic separation. The gas diffusion rate in the hollow fiber is fast and the exchange volume is small. The entire process from sampling to output of diagnostic results can be completed within a few minutes. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments 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.

[0026] Figure 1 A schematic flowchart of an embodiment of the GIS equipment insulation fault diagnosis method provided in this application; Figure 2 A schematic diagram of the cross-sectional structure of the anti-resonant hollow-core optical fiber provided in the embodiments of this application; wherein, 21-single-mode optical fiber; 22-graded refractive index fiber lens; 23-anti-resonant hollow-core optical fiber; 24-ceramic sleeve; 25-total reflection mirror; 26-sealing ring; 27-pore; 28-partial reflective film; Figure 3 This application provides a schematic diagram of the interference principle of an anti-resonant hollow fiber probe for embodiments of the present application; wherein, 30-gas to be tested; 31-first reflecting surface; 32-second reflecting surface; Figure 4 A schematic diagram of the optical path structure of the Fizeau interferometer is provided for the embodiments of this application; Figure 5 This is a schematic diagram illustrating the preprocessing effect of the interferometric time-domain signal provided in an embodiment of this application. Figure 6 This is a schematic diagram of the "light intensity-optical path" signal extracted from the interference time-domain signal according to an embodiment of this application; Figure 7 This application provides a flowchart for offline training, online concentration inversion, and GIS insulation fault classification processing of the machine learning model in an embodiment of the present application; wherein, Figure 7 (a) represents the Fizeau measurement channel signal acquired by the first photodetector; Figure 7 (b) represents the FBG reference channel signal acquired by the second photodetector; Figure 8 This is a schematic diagram of the entire process of the GIS equipment insulation fault diagnosis method provided in this embodiment; Figure 9 This is a schematic diagram of the logic structure of the GIS equipment insulation fault diagnosis device provided in the embodiments of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0028] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0029] Furthermore, to avoid obscuring the understanding of the invention by those skilled in the art, well-known or widely used techniques, elements, structures, and processes may not be described or shown in detail. Although the accompanying drawings illustrate exemplary embodiments of the invention, the drawings are not necessarily drawn to scale, and specific features may be enlarged or omitted to better illustrate and explain the invention.

[0030] First, this embodiment provides a method for diagnosing insulation faults in GIS equipment based on infrared spectroscopy and machine learning. Figure 1 This is a schematic flowchart of an embodiment of the GIS equipment insulation fault diagnosis method provided in this application, as shown below. Figure 1 As shown, the fault diagnosis method includes: S1. Collect the interferometric time-domain signal containing infrared absorption characteristics of the gas being measured inside the GIS equipment; The purpose of this step is to acquire the interference time-domain signal generated by the gas being measured inside the GIS equipment after absorbing infrared light, and then obtain the corresponding infrared absorption spectrum through mathematical transformation. However, traditional spectroscopic equipment such as Fourier transform infrared spectrometers (FTIR) are bulky and extremely sensitive to strong electromagnetic interference and mechanical vibration at the substation site, making it difficult to operate stably for a long time under harsh conditions.

[0031] Therefore, in a preferred embodiment, an anti-resonant hollow fiber optic probe combined with a micro-interferometer is used to acquire the infrared spectral interference time-domain signal of the gas being measured inside the GIS equipment. Specifically, the anti-resonant hollow fiber optic probe is connected to the gas chamber of the GIS equipment through the gas sampling interface. After the gas inside the gas chamber has diffused uniformly within the core of the anti-resonant hollow fiber optic probe, the gas within the core is irradiated with an infrared laser; the interference time-domain signal generated by the gas molecules absorbing the infrared laser is acquired using a micro-interferometer.

[0032] In this embodiment, the core of the anti-resonant hollow fiber serves as both a gas absorption cell and an interference cavity. The absorption of infrared light by gas molecules alters the optical properties within the interference cavity, resulting in a high-contrast time-domain interference pattern acquired and output by a micro-interferometer. This pattern is then converted into a digital signal by a photodetector. In a specific example, this micro-interferometer employs a Fizeau interferometer.

[0033] Figure 2 This is a schematic diagram of the anti-resonance hollow fiber probe provided in this embodiment, as shown below. Figure 2 As shown, the anti-resonant hollow fiber probe adopts a single-end reflective structure, including an input / output multiplexed single-mode fiber 21, a graded-index fiber lens 22 fused to the end of the single-mode fiber 21, an anti-resonant hollow fiber 23, a ceramic sleeve 24, and a total reflection mirror 25 disposed at the distal end of the anti-resonant hollow fiber 23. The incident component consisting of the single-mode fiber 21 and the graded-index fiber lens 22 is coupled to the anti-resonant hollow fiber 23 through a gap in the ceramic sleeve 24: the two are coaxially arranged inside the ceramic sleeve 24 and fixed by a sealing ring 26; a gap 27 of width d is reserved at the docking position between the graded-index fiber lens 22 and the anti-resonant hollow fiber 23, allowing the external gas to be measured 30 to enter the central air core of the anti-resonant hollow fiber 23 through the gap 27; the gap 27 serves as both a gas inlet and a coupling gap. The anti-resonant hollow fiber 23 consists of an outer cladding, an anti-resonant glass tube layer, and a central air core. The anti-resonant glass tube layer confines light transmission within the central air core, which also serves as an absorption channel for the gas 30 to be measured. In a specific example, the diameter D of the central air core is 50 μm, the width d of the pre-drilled aperture 27 on the ceramic sleeve 24 is 50 μm, and the optical path length L of the Fizeau interference cavity is 15 cm. This single-ended structure requires only one single-mode fiber to complete both the incident light transmission and the return interference light output, facilitating the sealed installation of the probe at the GIS gas chamber interface.

[0034] Figure 3 This is a schematic diagram of the Fizeau interference principle of the anti-resonant hollow fiber probe provided in this embodiment. Figure 3 As shown, the anti-resonant hollow fiber probe is placed in the gas to be tested 30, and the single-mode fiber 21 receives the incident beam I. inThe graded-index fiber optic lens 22 is used to focus the incident beam I. in Collimation is performed; the exit end face of the graded-index fiber lens 22 is coated with a partial reflective film 28 with a reflectivity of approximately 50%, forming the first reflecting surface 31 of the Fizeau interference cavity; the distal end of the anti-resonant hollow fiber 23 is sealed with a total reflection mirror 25, forming the second reflecting surface 32 of the Fizeau interference cavity. The light I input from the single-mode fiber 21... in After being collimated by the graded-index fiber lens 22, part of the light is reflected by the first reflecting surface 31 to form the first return light I1, and the other part passes through the first reflecting surface 31 and enters the central air core of the anti-resonant hollow fiber 23. After being reflected by the total reflection mirror 25, it returns along the original optical path and forms the second return light I2. I1 and I2 are superimposed in the single-mode fiber 21 to form the output interference light I. out After the gas to be tested 30 enters the central air fiber core through the reserved hole 27 of the ceramic sleeve 24, it produces wavelength-selective absorption of the round-trip propagating light, causing the amplitude and phase of the second returning light I2 to change with the wavelength. Therefore, the output interference light signal carries the absorption spectrum information of the gas to be tested.

[0035] In a preferred embodiment, an optical fiber capillary 29 is sleeved on the outside of the single-mode optical fiber 21 and the anti-resonant hollow optical fiber 23 to better achieve end-face coupling between the segmented optical fibers.

[0036] Interference light I output from single-mode fiber 21 out The data was collected by the Fizeau interferometer and converted into digital signals by a photodetector. Figure 4 This is a schematic diagram of the optical path structure of the interferometric measurement system provided in this embodiment, as shown below. Figure 4 As shown, the measurement system includes an anti-resonant hollow fiber probe, a Fizeau interferometer, a broadband light source, a first circulator, a coupler, a second circulator, a fiber Bragg grating (FBG), a first photodetector, and a second photodetector. All these components are connected via single-mode fiber. The broadband light source provides the input light I... in The light is input from the first port of the first circulator and transmitted to the anti-resonant hollow fiber probe through its second port; the output interference light I generated by the anti-resonant hollow fiber probe... out After being acquired by the Fizeau interferometer, the light returns from the second port of the first circulator and is output from the third port to the coupler. The coupler will output the interference light I. out It is divided into a measurement branch and a reference branch: among which, the interference light I in the measurement branch out The signal is directly fed into the first photodetector to obtain the measurement signal I1(t) carrying gas absorption information; the interference light I in the reference branch... outThe light is transmitted to the FBG via the second circulator. The corresponding port of the FBG and the second circulator is connected via a single-mode fiber. The narrowband reflected light is then directed by the second circulator into the second photodetector to obtain the reference signal I2(t). In this embodiment, the two photodetectors are not used to perform simple repeated measurements on the same signal. The reference signal I2(t) output by the reference channel is used to calibrate the instantaneous wavelength or optical path coordinates during the laser scanning process, correct scanning nonlinearity and light source drift, thereby improving the accuracy of subsequent spectral reconstruction of the measurement signal I1(t).

[0037] In a specific example, a broadband C+L band light source with a wavelength scanning range of 1520 nm to 1620 nm is used. This laser is injected into a single-mode fiber link via a first circulator; the output interference light I returned by the anti-resonant hollow fiber probe... out After being split by a coupler, the measurement signal and the FBG reference signal are acquired by two InGaAs photodetectors respectively, with a sampling frequency of 10MHz and a single acquisition duration of 100ms.

[0038] S2. Generate an infrared absorption spectrum based on the interference time-domain signal, extract the absorption peak features of each target characteristic component from the infrared absorption spectrum, and generate a spectral feature vector; In this step, after obtaining the interference time-domain signal corresponding to the gas being measured, the corresponding infrared absorption spectrum is obtained through Fourier transform.

[0039] In a preferred embodiment, generating an infrared absorption spectrum based on an interference time-domain signal includes: Extract the upper and lower envelopes of the interference time-domain signal, and perform baseline drift correction based on the center lines of the upper and lower envelopes; A dynamic sliding window is used to perform mean filtering on the corrected signal; The filtered signal is reconstructed into an equally spaced light intensity-light sequence using interpolation. The light intensity-light sequence is then subjected to Fourier transform to obtain the infrared absorption spectrum.

[0040] In this embodiment, the acquired interference time-domain signal is first preprocessed to extract the upper and lower envelopes of the signal. The center line of the envelope is taken as the interference light intensity after eliminating environmental disturbances such as light source power fluctuations and fiber bending losses, thus achieving baseline drift correction. The center line is obtained by the arithmetic mean of the upper and lower envelopes, representing the effective interference light intensity signal after eliminating baseline drift caused by light source fluctuations and fiber bending. Then, the size of the filtering window is adaptively adjusted according to the local fluctuation rate of the signal, and dynamic sliding window mean filtering is performed (for example, taking about 50 points in the area of ​​severe fluctuation and about 10 points in the area of ​​stability), which preserves spectral detail information while suppressing random noise.

[0041] Figure 5This is a schematic diagram illustrating the extraction of upper and lower envelopes of an interferometric time-domain signal according to an embodiment of the present invention. Figure 6 This is a schematic diagram illustrating the preprocessing effect of the interferometric time-domain signal provided in this embodiment. Figure 6 This demonstrates the effect of eliminating baseline drift after extracting the upper and lower envelopes and removing the center line from the original interference signal, and compares the results. Figure 5 and Figure 6 It is evident that the baseline of the processed signal is effectively flattened, eliminating slowly changing drift components while preserving high-frequency interference fringe details, laying the foundation for subsequent accurate spectral inversion.

[0042] After preprocessing, the filtered signal undergoes spectral inversion and self-calibration. The preprocessed signal is reconstructed into an equally spaced "intensity-optical path" sequence using either interpolation (when sampling points are uniformly distributed) or the Lomb-Scargle method (when sampling points are non-uniformly distributed). A Fast Fourier Transform (FFT) is then performed on the reconstructed signal to obtain the infrared absorption spectrum in the wavenumber domain. In a specific example, cubic spline interpolation is used to reconstruct the preprocessed signal into an equally spaced (approximately 0.01 nm interval, corresponding to an optical path difference of 5 nm) "intensity-optical path" sequence.

[0043] Figure 7 This is a schematic diagram of the dual-channel signal and its optical path coordinate calibration results provided in this embodiment, wherein... Figure 7 (a) represents the Fizeau measurement channel signal acquired by the first photodetector: the horizontal axis of the upper figure is the sampling time t, showing the measurement interference signal I1(t) obtained with laser scanning; the horizontal axis of the lower figure is the equally spaced optical path coordinate δ after calibration by the FBG reference signal, showing the measurement signal I1(δ) after resampling, where the local magnified box is used to compare the changes in the density of the interference fringes before and after calibration. Figure 7 (b) shows the FBG reference channel signal acquired by the second photodetector: the upper figure shows the reference signal I2(t), and the lower figure shows the reference signal I2(δ) after being transformed to equally spaced optical path coordinates. The reference channel is used to establish the mapping relationship between sampling time t and optical path coordinate δ. After the measurement channel is resampled through this mapping, it forms an equally spaced optical path sequence, which serves as the input for fast Fourier transform and absorption spectrum reconstruction.

[0044] After obtaining the infrared absorption spectrum, background correction is performed using a blank reference spectrum without the target component, matched to the current instrument, temperature, pressure, and effective optical path, to correct the instrument response and baseline. Then, absorption peak features of each target component are extracted from the corrected infrared absorption spectrum to generate a spectral feature vector. This absorption peak feature is used for subsequent component concentration decoupling. In one specific example, the absorption peak feature includes one or more of normalized absorbance, peak height, peak area, and band ratio. These features are concatenated in a specific order to construct the spectral feature vector. In an optional embodiment, the spectral feature vector also includes temperature and pressure features.

[0045] This embodiment also incorporates a temperature compensation coefficient to self-calibrate the absorption peak intensities of each component in the infrared absorption spectrum. The temperature compensation formula for this self-calibration is as follows: C compensated =C raw ×(1+α×(T - T ref )) Among them, C compensated Indicates the calibrated absorption peak intensity; C raw Indicates the original absorption peak intensity; α represents the temperature coefficient; T represents the measured ambient temperature; T ref Indicates the reference temperature.

[0046] In a specific example, the measured ambient temperature is 35°C, the reference temperature Tref is 25°C, and the temperature compensation coefficient α = 0.003 / °C is used to perform temperature compensation self-calibration on the absorption peak intensity of each component.

[0047] In this embodiment, the target characteristic components include three gases: sulfur hexafluoride (SF6), hydrogen sulfide (H2S), and oxygen (O2). SF6 is the primary insulating and arc-extinguishing medium in GIS; abnormal insulation, partial discharge, or changes in sealing conditions can cause changes in the concentration of SF6, abnormal H2S components, and environmental O2 components. H2S can be considered an abnormal gas related to insulation faults, while O2 reflects the presence of residual air or intrusion of external air in the gas chamber. Therefore, this embodiment selects SF6, H2S, and O2 as the three gases to be measured, and assesses the GIS operating status by analyzing their concentrations, component ratios, and trends.

[0048] It is understood that the foregoing is merely an illustrative example of this embodiment and does not constitute a limitation on the type of target feature components.

[0049] S3. Input the spectral feature vector into the trained multi-group decoupling model to generate the concentration prediction value of each target feature component in the gas being tested; The training process for the multi-group decoupling model includes: Acquire multiple first training samples with concentration labels. The first training samples include the original infrared absorption spectra of single-component standard gas samples with different concentrations of a single target feature component and multi-component standard gas samples composed of a mixture of multiple target feature components. The original spectral feature vector is extracted from the original infrared absorption spectrum, and multiple decoupling models are trained based on the original spectral feature vector.

[0050] In this embodiment, the multi-component decoupling model adopts the support vector regression (SVR) model. The training sample dataset is constructed using the infrared absorption spectra of the standard test gases SF6, H2S and O2 at known concentrations, as well as the pairwise and three-component standard test gases. The training set, validation set and independent test set are divided according to the gas mixing or measurement batch to train the support vector regression (SVR) multi-component concentration decoupling model.

[0051] Specifically, according to the planned gas mixing matrix, single-component standard gases of SF6, H2S, and O2 at different concentrations were prepared. These three single-component standard gases at different concentrations were mixed to prepare three-component standard analytes, and the three single-component standard gases were also combined in pairs to prepare three two-component standard analytes. The prepared gas samples were repeatedly collected under different temperatures, pressures, and measurement dates. For each gas sample, the true concentration of each component was measured, and a concentration label was generated. Infrared laser irradiation was used to measure the original interference signal. For each sample, information including the original interference signal, reference channel, true concentration, temperature, pressure, instrument serial number, calibration version, and measurement batch was simultaneously saved.

[0052] A consistent integrity check is performed on all gas samples, and envelope extraction is performed on their original interference signals to correct the baseline. Adaptive filtering is used to suppress random noise, and then the original infrared absorption spectrum is generated through optical path mapping and spectral reconstruction. It can be understood that the above processing of the original interference signal is exactly the same as the processing of the interference time domain signal of the gas to be measured in step S1, and will not be repeated here.

[0053] After obtaining the original infrared absorption spectrum, normalized absorbance, peak height, peak area and band ratio are extracted from the spectrum through wavenumber calibration and absorbance conversion. Temperature and pressure are used as compensation features, and the above features are spliced ​​together to form the original spectral feature vector.

[0054] The gas samples were grouped into training, validation, and independent test sets according to gas batch or measurement date to avoid duplicate measurements of the same batch of standard gas being included in both sets. The multi-component concentration decoupling model was trained using the training set, and a radial basis function (RBF) kernel SVR regressor was established for each target characteristic component. Cross-validation was used to select the penalty coefficient C, kernel width γ, and insensitive loss parameter ε, with the goal of minimizing error. The validation set was used to determine the characteristic intervals and model parameters, while the independent test set was used only for final evaluation.

[0055] After the model training is completed, the preprocessing parameters of the interference signal, the feature splicing order of the spectral feature vectors, and the model version are frozen and deployed online.

[0056] After performing the same preprocessing and feature stitching as in the training phase on the interferometric time-domain signal of the gas being measured, the generated spectral feature vector is input into a multi-group decoupled model that has been trained. The model outputs the concentration prediction value of each target feature component and the corresponding uncertainty. The uncertainty characterizes the measurement uncertainty of the concentration prediction result, and this uncertainty is not a probability directly given by the SVR regressor model. Specifically, this embodiment uses multiple Bootstrap resampling trained SVR sub-models for each target feature component, takes the mean of the prediction values ​​of each sub-model as the concentration prediction value, and calculates the standard deviation smodel of the sub-model prediction value to characterize the model uncertainty; multiple repeated measurements are performed on the same sample, and the standard deviation srep of repeated measurements is calculated to characterize the measurement repeatability; the root mean square (RMSE) of the prediction residuals of the independent validation set is used to characterize the calibration error. These three factors are combined to form the standard uncertainty u = √(RMSE² + smodel² + srep²), and the expanded uncertainty at approximately 95% coverage probability is given as U = 1.96u. The concentration result is expressed as "predicted concentration ± U". Furthermore, the system generates data quality indicators based on spectral residuals, the distance between the distribution of input data and training data, repeated measurement dispersion, and expanded uncertainty U; measured gases that exceed the calibration range or whose uncertainty exceeds a preset threshold are not included in subsequent fault classification.

[0057] S4. Determine the fault type of the GIS equipment based on the predicted concentration value, concentration change law, concentration ratio between different target characteristic components, and the pre-established mapping model. The mapping model is used to characterize the correspondence between the concentration of each target characteristic component and its concentration change law and the insulation fault type.

[0058] In this step, based on a large amount of known data (simulation test data or field data confirmed by detection methods) with clear fault types, a mapping model is pre-established between the concentrations of the three tested gases SF6, H2S and O2, their changes over time, the H2S / SF6 and O2 / SF6 concentration ratios, and the insulation fault types. This model is then saved in the form of mapping function formulas or mapping lookup tables.

[0059] Step S3 obtains the concentration values ​​of SF6, H2S and O2 in the output gas being tested, and calculates the concentration change rate of SF6, H2S and O2, as well as the concentration ratio of H2S / SF6 and O2 / SF6 by continuous measurement. The target mapping result, i.e. the insulation fault type, is obtained by looking up the mapping lookup table or calling the mapping function.

[0060] In a more preferred embodiment, the above mapping model adopts a decision tree classification model, and the mapping model is deeply characterized by nonlinear fitting. The fault diagnosis feature vector generated based on the concentration prediction value, concentration change law and concentration ratio between different target feature components is input into the trained decision tree classification model to obtain the insulation fault type and the corresponding classification confidence.

[0061] In this embodiment, for the continuous measurement results that pass the quality inspection in step S3, the concentrations, concentration change patterns, H2S / SF6 and O2 / SF6 concentration ratios, and data quality indicators of the three measured gases SF6, H2S and O2 are combined to form a fault diagnosis feature vector in a fixed order. This fault diagnosis feature vector is then input into a pre-trained decision tree classification model, and the model outputs the fault classification result and the corresponding confidence level.

[0062] Furthermore, the training process of this decision tree classification model includes: Obtain multiple second training samples labeled with different insulation fault types; Obtain the predicted concentration value, concentration variation law, and concentration ratio between different target feature components corresponding to each second training sample, and generate the original fault diagnosis feature vector; The decision tree classification model is trained based on the original fault diagnosis feature vector.

[0063] Specifically, a training dataset with fault labels is first established, which includes at least corona discharge, spark discharge, and arc discharge. The second training samples in the training dataset are sourced from simulated test data with clear fault types or field data confirmed by maintenance, disassembly, and other detection methods. Samples whose fault types cannot be confirmed are not used as supervised training labels.

[0064] For each second training sample, an original fault diagnosis feature vector is generated. Specifically, the concentrations of the three measured gases SF6, H2S, and O2, the concentration difference or rate of change between adjacent measurement times, the H2S / SF6 and O2 / SF6 concentration ratios, temperature and pressure, and the data quality indicators of concentration inversion are concatenated as diagnostic features to generate the original fault diagnosis feature vector.

[0065] The training dataset was divided into training, validation, and independent test sets. A decision tree classification model was fitted using the training set. The validation set was used to select the maximum tree depth, minimum number of leaf node samples, and classification threshold, with clearly unreasonable branches limited by the GIS fault gas response mechanism. The independent test set was used to calculate the accuracy, recall, precision, F1 score, and confusion matrix for each fault category. The test set was not used to modify classification features or thresholds. Once the independent tests met the pre-set error and classification thresholds, the feature concatenation order of the diagnostic feature vectors and the model parameters were frozen, and the model was deployed online.

[0066] During online classification, the decision tree classification model predicts the fault category based on the input fault diagnosis feature vector and outputs the classification confidence score formed by the leaf node sample distribution. When the concentration inversion quality is unqualified, the input data exceeds the coverage of the training data, or the classification confidence score is lower than the preset threshold, the model does not output a definite fault conclusion, but instead outputs a prompt for retesting or manual review.

[0067] Figure 8 This is a schematic diagram of the entire process of the GIS equipment insulation fault diagnosis method provided in this embodiment, as shown below. Figure 8 As shown, this method comprises two main stages: offline modeling and validation, and online measurement and classification. The offline modeling and validation stage mainly includes the following steps:

[0068] (1) Establish a multi-component calibration dataset; configure standard mixed gases of SF6, H2S and O2 with single components, pairwise combinations and three components covering zero point, low concentration, medium concentration and high concentration, and repeatedly collect them under different temperatures, pressures and measurement dates; (2) Acquire the time-domain interference signal of each sample; for each sample, save the original interference signal, reference channel, true concentration of the three gases being measured, temperature, pressure, instrument number, calibration version and measurement batch.

[0069] (3) All samples were subjected to consistent integrity checks, envelope extraction, baseline correction, adaptive filtering, optical path mapping, spectral reconstruction, wavenumber calibration, and absorbance conversion. Normalized absorbance, peak height, peak area, and band ratio were extracted from the identifiable bands of the target gas, and temperature and pressure were used as compensation features. All standardized parameters were calculated only from the training set and used as is in the validation set, test set, and online data.

[0070] (4) Group the gas mixing batches or measurement dates into training sets, validation sets, and independent test sets to avoid repeated measurements of the same batch of standard gases from entering both the training and test sets simultaneously. Establish a radial basis kernel SVR regressor for each target component, and select the penalty coefficient C, kernel width γ, and insensitive loss parameter ε through cross-validation; use the validation set to determine the feature interval and model parameters, and use the independent test set only for the final evaluation.

[0071] (5) Establish a dataset with fault labels. The labels should include at least corona discharge, spark discharge and arc discharge. The samples should be sourced from simulated test data with clear fault conditions or field data confirmed by maintenance, disassembly and other detection methods. Samples whose fault type cannot be confirmed should not be used as supervised training labels.

[0072] (6) Generate diagnostic features for each sample, including the concentrations of the three gases SF6, H2S, and O2, the H2S / SF6 and O2 / SF6 concentration ratios, the concentration difference or rate of change between adjacent measurement times, temperature and pressure, and the concentration inversion quality level. The rules for handling missing values, outliers, and class imbalances in the training data are fixed before model training.

[0073] (7) Fit a decision tree using the training set, select the maximum tree depth, minimum number of leaf node samples, and classification threshold using the validation set, and limit obviously unreasonable branches based on the GIS fault gas response mechanism. The accuracy, recall, precision, F1 score, and confusion matrix of each fault category are calculated separately on the independent test set. The test set is not used to modify classification features or thresholds.

[0074] (8) After passing the independent test, freeze the preprocessing parameters, feature order and model version of each step and deploy it online.

[0075] The online measurement and classification phase mainly includes the following steps: (1) Using an anti-resonant hollow fiber probe combined with a miniature Fizeau interferometer, the infrared spectral interference time-domain signal containing infrared absorption characteristics of the mixed gas inside the GIS equipment was collected.

[0076] (2) Perform the same preprocessing and feature extraction steps as the training stage on the collected interference time-domain signal to obtain the infrared absorption spectrum. Construct the spectral feature vector according to the fixed feature interval and processing order in the training stage. Input it into the pre-trained SVR multi-group decoupling model to output the concentrations of the three measured gases SF6, H2S and O2. Calculate the uncertainty and generate data quality indicators.

[0077] (3) For continuous measurement results that pass the quality inspection, combine the concentrations of the three measured gases SF6, H2S and O2, the H2S / SF6 and O2 / SF6 concentration ratios, the concentration change rate per unit time and data quality indicators to form a fault diagnosis feature vector in a fixed order, and input it into a pre-trained decision tree classification model.

[0078] (4) Output the classification confidence score formed by the fault category and leaf node sample distribution during online classification.

[0079] It should be noted that although the operations of the methods of the embodiments of this specification are described in a specific order in the above embodiments, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowcharts may be executed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0080] Figure 9 This is a schematic diagram of the logical structure of a GIS equipment insulation fault diagnosis device based on infrared spectroscopy and machine learning provided in this embodiment. This device can be implemented in software and / or hardware and can be integrated into electronic devices; for example... Figure 9 As shown, the device includes a signal acquisition module, a feature extraction module, a concentration decoupling module, and a fault diagnosis module; among which, The signal acquisition module is used to acquire interferometric time-domain signals containing infrared absorption characteristics of the gas being measured inside the GIS equipment; In a preferred embodiment, the signal acquisition module includes an anti-resonant hollow fiber probe, a miniature Fizeau interferometer, a light source, and a photodetector. Specifically, the anti-resonant hollow fiber probe is connected to the gas chamber of the GIS equipment. After the gas inside the gas chamber diffuses evenly within the core of the anti-resonant hollow fiber probe, the light source irradiates the gas within the core with an infrared laser. The interference time-domain signal generated by the gas molecules absorbing the infrared laser is acquired by the miniature interferometer and converted into a digital signal by the photodetector.

[0081] For further details, please refer to [link / reference]. Figure 4 The signal acquisition module also includes a broadband light source, a first circulator, a coupler, a second circulator, a fiber Bragg grating (FBG), a first photodetector, and a second photodetector. All these components are connected via single-mode optical fiber. The broadband light source provides the input light I... in The light is input from the first port of the first circulator and transmitted to the anti-resonant hollow fiber probe through its second port; the output interference light I generated by the anti-resonant hollow fiber probe... outAfter being acquired by the Fizeau interferometer, the light returns from the second port of the first circulator and is output from the third port to the coupler. The coupler will output the interference light I. out It is divided into a measurement branch and a reference branch: among which, the interference light I in the measurement branch out The signal is directly fed into the first photodetector to obtain the measurement signal I1(t) carrying gas absorption information, i.e., the interference time-domain signal; the interference light I in the reference branch... out The light is transmitted to the FBG via the second circulator. The corresponding port of the FBG and the second circulator are connected through a single-mode optical fiber. The narrowband reflected light is then directed by the second circulator into the second photodetector to obtain the reference signal I2(t).

[0082] The feature extraction module generates an infrared absorption spectrum based on the interference time-domain signal output by the signal acquisition module, extracts the absorption peak features of each target feature component from the infrared absorption spectrum, and generates a spectral feature vector. The concentration decoupling module inputs the above-mentioned spectral feature vector into a preset multi-group decoupling model to generate the concentration prediction value of each target feature component in the gas being measured; wherein, the above-mentioned preset multi-group decoupling model is the multi-group decoupling model described in the GIS equipment insulation fault diagnosis method.

[0083] The fault diagnosis module determines the fault type of GIS equipment based on the predicted concentration value of each target characteristic component, the concentration change law, the concentration ratio between different target characteristic components, and the pre-established mapping model. The mapping model is used to characterize the correspondence between the concentration of each target characteristic component and its concentration change law and the insulation fault type. The mapping model is established based on a large amount of known data (simulation test data or field data confirmed by detection methods) with clear fault types. It maps the concentrations of the three tested gases, SF6, H2S and O2, the concentration change over time, and the H2S / SF6 and O2 / SF6 concentration ratios to the insulation fault types. The mapping model is saved in the form of mapping function formulas or mapping lookup tables for the fault diagnosis module to call or look up.

[0084] The signal acquisition module continuously acquires the interference time-domain signal of the gas under test at set time intervals. After the concentration decoupling module performs concentration inversion to generate the concentration values ​​of SF6, H2S and O2 in the gas under test, the fault diagnosis module calculates the concentration change rate of SF6, H2S and O2, as well as the concentration ratio of H2S / SF6 and O2 / SF6 based on the above continuous measurement. The target mapping result, i.e. the insulation fault type, is obtained by looking up the mapping lookup table or calling the mapping function.

[0085] In a preferred embodiment, the mapping model described above is a decision tree classification model. Fault diagnosis feature vectors generated based on the predicted concentration values, concentration variation patterns, and concentration ratios between different target feature components are input into the trained decision tree classification model to obtain the insulation fault type and its corresponding classification confidence level. Specifically, the decision tree classification model described above is the decision tree classification model used in the GIS equipment insulation fault diagnosis method.

[0086] Specific limitations regarding the GIS equipment insulation fault diagnosis device can be found in the limitations of the GIS equipment insulation fault diagnosis method described above, and will not be repeated here. Each module in the aforementioned GIS equipment insulation fault diagnosis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0087] This embodiment also provides an electronic device, which includes at least one processor and at least one memory. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of the method for diagnosing insulation faults in GIS equipment in Embodiment 1. The specific steps are described in Embodiment 1 and will not be repeated here. In this embodiment, the types of processor and memory are not specifically limited. For example, the processor can be a microprocessor, a digital information processor, an on-chip programmable logic system, etc.; the memory can be volatile memory, non-volatile memory, or a combination thereof.

[0088] The electronic device can also communicate with one or more external devices (such as a keyboard, pointing terminal, display, etc.), one or more terminals that enable users to interact with the electronic device, and / or any terminal that enables the electronic device to communicate with one or more other computing terminals (such as a network card, modem, etc.). This communication can be performed via an input / output (I / O) interface. Furthermore, the electronic device can also communicate with one or more networks (such as a Local Area Network (LAN), a Wide Area Network (WAN), and / or a public network, such as the Internet) via a network adapter.

[0089] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0090] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0091] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0092] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0093] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0095] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0096] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0097] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0098] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for diagnosing insulation faults in GIS equipment based on infrared spectroscopy and machine learning, characterized in that, include: Acquire interferometric time-domain signals containing infrared absorption characteristics of the gas being measured inside the GIS equipment; Infrared absorption spectra are generated based on the interference time-domain signal, and absorption peak features of each target characteristic component are extracted from the infrared absorption spectra to generate spectral feature vectors; The spectral feature vector is input into a pre-trained multi-group decoupling model to generate the concentration prediction values ​​of each target feature component in the gas being tested. The fault type of GIS equipment is determined based on the predicted concentration value, concentration change law, and concentration ratio between different target characteristic components and the pre-established mapping model. The mapping model is used to characterize the correspondence between the concentration of each target characteristic component and its concentration change law and the insulation fault type. The training process of the multiple decoupling models includes: Acquire multiple first training samples with concentration labels. The first training samples include the original infrared absorption spectra of single-component standard gas samples with different concentrations of a single target feature component and multi-component standard gas samples composed of a mixture of multiple target feature components. The original spectral feature vector is extracted from the original infrared absorption spectrum, and multiple decoupling models are trained based on the original spectral feature vector.

2. The GIS equipment insulation fault diagnosis method as described in claim 1, characterized in that, The mapping model is a decision tree classification model; The fault diagnosis feature vector generated based on the concentration prediction value, concentration change law and concentration ratio between different target feature components is input into the trained decision tree classification model to obtain the insulation fault type and the corresponding classification confidence.

3. The GIS equipment insulation fault diagnosis method as described in claim 2, characterized in that, The training process of the decision tree classification model includes: Obtain multiple second training samples labeled with different insulation fault types; Obtain the predicted concentration value, concentration variation law, and concentration ratio between different target feature components corresponding to each second training sample, and generate the original fault diagnosis feature vector; The decision tree classification model is trained based on the original fault diagnosis feature vector.

4. The GIS equipment insulation fault diagnosis method as described in claim 1 or 3, characterized in that, The interferometric time-domain signal containing infrared absorption characteristics of the gas being measured inside the GIS acquisition equipment includes: An anti-resonant hollow fiber probe is connected to the gas chamber of the GIS equipment. After the gas inside the gas chamber diffuses evenly within the core of the anti-resonant hollow fiber probe, the gas inside the core is irradiated with an infrared laser. The time-domain interference signal generated by gas molecules absorbing the infrared laser is collected using a micro-interferometer.

5. The GIS equipment insulation fault diagnosis method as described in claim 1 or 3, characterized in that, The generation of infrared absorption spectra based on interference time-domain signals includes: Extract the upper and lower envelopes of the interference time-domain signal, and perform baseline drift correction based on the center lines of the upper and lower envelopes; A dynamic sliding window is used to perform mean filtering on the corrected signal; The filtered signal is reconstructed into an equally spaced light intensity-light sequence using an interpolation method. The light intensity-light sequence is then subjected to a Fourier transform to obtain the infrared absorption spectrum.

6. The GIS equipment insulation fault diagnosis method as described in claim 5, characterized in that, Also includes: The absorption peak intensities of each component in the infrared absorption spectrum are self-calibrated using a temperature compensation coefficient. The temperature compensation formula for this self-calibration is as follows: C compensated =C raw ×(1+α×(T - T ref )) Among them, C compensated Indicates the calibrated absorption peak intensity; C raw Indicates the original absorption peak intensity; α represents the temperature coefficient; T represents the measured ambient temperature; T ref Indicates the reference temperature.

7. A GIS equipment insulation fault diagnosis device based on infrared spectroscopy and machine learning, characterized in that, include: The signal acquisition module is configured to acquire interferometric time-domain signals containing infrared absorption characteristics of the gas being measured inside the GIS equipment. The feature extraction module is configured to generate an infrared absorption spectrum based on the interference time-domain signal, extract the absorption peak features of each target feature component from the infrared absorption spectrum, and generate a spectral feature vector. The concentration decoupling module is configured to input the spectral feature vector into a preset set of decoupling models to generate the concentration prediction values ​​of each target feature component in the gas being measured. The fault diagnosis module is configured to determine the fault type of the GIS equipment based on the predicted concentration values, concentration variation patterns, and concentration ratios between different target characteristic components, as well as a pre-established mapping model. The mapping model is used to characterize the correspondence between the concentration of each target characteristic component and its concentration variation patterns and the insulation fault type. The preset multi-group decomposition and coupling model is the multi-group decomposition and coupling model described in any one of claims 1-6.

8. The GIS equipment insulation fault diagnosis device as described in claim 7, characterized in that, The mapping model is a decision tree classification model; The fault diagnosis feature vector generated based on the concentration prediction value, concentration change law and concentration ratio between different target feature components is input into the trained decision tree classification model to obtain the insulation fault type and the corresponding classification confidence.

9. The GIS equipment insulation fault diagnosis device as described in claim 7, characterized in that, The signal acquisition module includes an anti-resonant hollow fiber probe and a miniature interferometer; The anti-resonant hollow fiber probe is connected to the gas chamber of the GIS equipment. After the gas inside the gas chamber diffuses evenly within the core of the anti-resonant hollow fiber probe, the gas inside the core is irradiated with an infrared laser. The micro-interferometer is used to collect the time-domain interference signal generated after gas molecules absorb the infrared laser.

10. The GIS equipment insulation fault diagnosis device as described in claim 8, characterized in that, The anti-resonant hollow fiber probe includes a ceramic sleeve, an input / output multiplexed single-mode fiber, an anti-resonant hollow fiber, and a total reflection mirror disposed at the far end of the anti-resonant hollow fiber; a graded refractive index fiber lens is fused to one end of the single-mode fiber near the anti-resonant hollow fiber, and the exit end face of the graded refractive index fiber lens is coated with a partial reflective film. The incident assembly, consisting of a single-mode fiber and a graded-index fiber lens, is coupled to the anti-resonant hollow fiber using the ceramic sleeve with a gap. The graded-index fiber lens and the anti-resonant hollow fiber have a pre-reserved aperture at the docking position, so that the external gas to be tested can enter the central air core of the anti-resonant hollow fiber through the aperture.

Images