A method and system for detecting inter-turn insulation short-circuit faults in dry-type air-core reactors

Abstract

This invention relates to the field of fault detection technology, specifically to a method and system for detecting inter-turn insulation short-circuit faults in dry-type air-core reactors. The method involves arranging a detection coil array radially at the bottom of the reactor and performing segmented autocorrelation analysis on the signal to obtain a noise probability estimate. By combining the analysis of energy differences between different measurement points in the same frequency band and the energy attenuation characteristics of the radially distributed frequency band, characteristic frequency bands are selected. Adaptive wavelet denoising is then performed using these characteristic frequency bands, selecting the optimal wavelet basis for denoising based on the energy proportion of the characteristic frequency band. Fault determination is then performed based on the spatial distribution of the measurement points. This invention utilizes adaptive wavelets based on decomposition goodness, leveraging the local signal autocorrelation regularity and the attenuation characteristics of the radial magnetic field distribution to enhance the retention and identification of potential fault signals, thereby improving fault identification capabilities and increasing the accuracy and reliability of inter-turn insulation short-circuit fault detection in dry-type air-core reactors.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, specifically to a method and system for detecting inter-turn insulation short-circuit faults in dry-type air-core reactors. Background Technology

[0002] Dry-type air-core reactors are widely used as key equipment in power systems for reactive power compensation, current limiting, and filtering due to their advantages such as good linearity, simple structure, and convenient maintenance. Inter-turn insulation short circuits are one of the most common fault types. While minor early inter-turn short circuits may not directly cause equipment tripping, they can lead to localized overheating of the reactor, accelerating insulation degradation, and in severe cases, potentially causing fires and threatening the safe and stable operation of the power system.

[0003] Currently, inter-turn short-circuit fault diagnosis is often achieved by detecting local magnetic field changes caused by short-circuit loops in the spatial magnetic field. However, the complex dynamic electromagnetic environment generated by high-voltage, high-current equipment operating in substations for extended periods exhibits high-frequency components that are quite similar to inter-turn short-circuit fault signals under strong electromagnetic interference, which can easily lead to the fault signals being submerged. This is especially true for early, minor inter-turn insulation short-circuit faults with few short-circuit turns and weak fault signals, where the detection sensitivity and anti-interference capability are insufficient. The weak magnetic field signals collected by the detection coil are severely interfered with, resulting in a reduced signal-to-noise ratio and making it easy to miss or misdiagnose faults, thus affecting the accuracy of fault identification. Summary of the Invention

[0004] To address the technical problems in existing technologies, such as insufficient detection sensitivity and anti-interference capability, severe interference from weak magnetic field signals collected by the detection coil leading to reduced signal-to-noise ratio, and easy misjudgment or missed detection, thus affecting the accuracy of fault identification, the present invention aims to provide a method and system for detecting inter-turn insulation short-circuit faults in dry-type air-core reactors. The specific technical solution adopted is as follows:

[0005] This invention provides a method for detecting inter-turn insulation short-circuit faults in dry-type air-core reactors, the method comprising:

[0006] At the measurement points of each detection coil arranged radially below the bottom of the dry-type air-core reactor, the fundamental filtered induced voltage debase signal is simultaneously acquired; the measurement point directly below the center of the bottom of the dry-type air-core reactor is the measurement origin.

[0007] All induced voltage removal signals are uniformly divided into different local signal segments according to time sequence; for each local signal segment of each measurement point, the noise probability estimate of the measurement point in the local signal segment is determined by autocorrelation intensity analysis;

[0008] The spectrum of each local signal segment is uniformly divided into different frequency bands in the spectrum diagram. At each frequency band, the frequency band energy difference and noise probability estimate of each local signal segment between every two measurement points are analyzed to determine the noise dominance of the frequency band. The frequency band energy attenuation characteristics of each local signal segment at each frequency band are analyzed in a radial distribution order starting from the measurement origin. Combined with the noise dominance, the fault significance value of each frequency band is obtained. Characteristic frequency bands are selected by screening the fault significance values.

[0009] Adaptive wavelet denoising is performed on the induced voltage signal at each measurement point, including: at each decomposition layer, the signal is decomposed using different wavelet basis functions, and the optimal wavelet basis is determined by the energy proportion of the corresponding characteristic frequency band in the spectrum of the decomposed wavelet coefficients; wavelet reconstruction is performed using the optimal wavelet basis at each layer to obtain the wavelet reconstructed signal at each measurement point.

[0010] Fault detection is performed based on the positional distribution of each measurement point relative to the measurement origin and the signal energy of the wavelet reconstructed signal.

[0011] Furthermore, the method for obtaining the noise probability estimate includes:

[0012] For any local signal segment, the autocorrelation peak value and corresponding time delay of the local signal segment are obtained through the autocorrelation function; based on the deviation between the autocorrelation peak value of the local signal segment and the maximum autocorrelation peak value among all local signal segments, the correlation strength index of the local signal segment is obtained, and the deviation of the autocorrelation peak value is negatively correlated with the correlation strength index.

[0013] Based on the deviation between the time delay corresponding to the autocorrelation peak value of the local signal segment and the mean time delay corresponding to the autocorrelation peak value of all local signal segments, the correlation delay index of the local signal segment is obtained. The deviation of the time delay is negatively correlated with the correlation delay index.

[0014] By combining the relevant intensity and delay indices of the local signal segment, the noise probability estimate of the local signal segment is obtained.

[0015] Furthermore, the method for obtaining the noise dominance includes:

[0016] For any frequency band, every two different measurement points are considered as a measurement group, and each measurement group is considered as an analysis group. For any local signal segment of the analysis group, based on the difference in frequency band energy between the two measurement points in the analysis group and the sum of the noise probability estimates of the two measurement points in the local signal segment, the spectral energy approximation of the analysis group in the local signal segment is obtained. The difference in frequency band energy is negatively correlated with the spectral energy approximation, and the sum of the noise probability estimates is positively correlated with the spectral energy approximation.

[0017] By fusing the spectral energy approximation of all local signal segments in the analysis group, the spectral characteristic approximation of the analysis group is obtained; by fusing the spectral characteristic approximation of all measurement groups in the frequency band, the noise dominance of the frequency band is obtained.

[0018] Furthermore, the method for obtaining the fault significance value includes:

[0019] Starting from the measurement origin, the remaining measurement points are arranged in ascending order of their distance from the measurement origin to obtain the measurement distribution sequence. For any frequency band, the frequency band energy of each local signal segment at the measurement point is arranged in the order of the measurement distribution sequence to obtain the energy arrangement sequence of each local signal segment.

[0020] Obtain the slope of the frequency band energy in the energy arrangement sequence of each local signal segment; count the proportion of local signal segments with negative slopes, and use this as the magnetic attenuation of that frequency band;

[0021] By integrating the magnetic attenuation and noise dominance of the frequency band, the fault significance value of the frequency band is obtained. The noise dominance is negatively correlated with the fault significance value, while the magnetic attenuation is positively correlated with the fault significance value.

[0022] Furthermore, the step of filtering characteristic frequency bands through fault significance values ​​includes:

[0023] The segmentation threshold for the significant fault values ​​of all frequency bands is obtained using the Otsu's method; frequency bands with significant fault values ​​higher than the segmentation threshold are used as feature frequency bands.

[0024] Furthermore, the method for obtaining the optimal wavelet basis includes:

[0025] In each decomposition layer, for each wavelet basis function, the wavelet coefficients after decomposition are used as the ratio of the sum of the frequency band energies of all characteristic frequency bands in the frequency domain transformed spectrum to the sum of the frequency band energies of all frequency bands. The wavelet basis function with the highest decomposition quality is taken as the optimal wavelet basis function for each decomposition layer.

[0026] Furthermore, the fault detection based on the positional distribution of each measurement point relative to the measurement origin and the signal energy of the wavelet reconstructed signal includes:

[0027] Based on the signal value of the wavelet reconstructed signal at each measurement point, the wavelet coefficient energy at each measurement point is obtained; the measurement origin and the preset number of measurement points closest to the measurement origin are taken as neighboring measurement points; when the wavelet coefficient energy of the neighboring measurement points is greater than the preset energy threshold, a fault alarm is triggered.

[0028] Otherwise, the number of measurement points with wavelet coefficient energy greater than the preset energy threshold is counted as the suspected number. When the proportion of suspected number exceeds the preset proportion of fault number, a fault alarm is triggered.

[0029] Furthermore, the method for obtaining the induced voltage debase signal includes:

[0030] The induced voltage signal at each measurement point is collected, and the induced voltage signal is notched to remove the power frequency fundamental component, thus obtaining the induced voltage de-fundamental signal.

[0031] Furthermore, the method for determining the frequency band energy includes:

[0032] In the spectrum diagram of each local signal segment, the root mean square of the amplitude corresponding to all frequencies within each frequency band is taken as the frequency band energy of the local signal segment in each frequency band.

[0033] The present invention also provides a dry-type air-core reactor inter-turn insulation short-circuit fault detection system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described dry-type air-core reactor inter-turn insulation short-circuit fault detection method.

[0034] The present invention has the following beneficial effects:

[0035] This invention utilizes a radially arranged array of detection coils at the bottom of the reactor and performs segmented signal analysis. It leverages the strong autocorrelation of background noise and the suddenness of fault signals to construct a noise probability estimate. Combining the physical characteristics of radial attenuation of the fault magnetic field signal and the spatial consistency of background noise, it analyzes the energy differences between different measurement points in the same frequency band and the radially distributed frequency band energy attenuation characteristics to screen out characteristic frequency bands more likely to characterize the fault, enhancing subsequent fault signal identification. Furthermore, adaptive wavelet denoising is performed using these characteristic frequency bands, selecting the optimal wavelet basis based on the energy proportion of the characteristic frequency band for denoising, more effectively preserving the time-frequency characteristics of inter-turn short-circuit faults. Finally, a comprehensive fault determination is made based on the spatial distribution of measurement points and the signal energy of the wavelet-reconstructed signal. This invention utilizes adaptive wavelets based on decomposition goodness, leveraging the autocorrelation regularity of local signals and the attenuation characteristics of the radial magnetic field distribution to enhance the preservation and identification of possible fault signals. While suppressing complex background noise interference in substations, it improves fault identification capabilities and enhances the accuracy and reliability of detecting inter-turn insulation short-circuit faults in dry-type air-core reactors. Attached Figure Description

[0036] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A flowchart of a method for detecting inter-turn insulation short-circuit faults in a dry-type air-core reactor, provided in one embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of a measurement origin arrangement provided in one embodiment of the present invention;

[0039] Please label the following reference numerals on the diagram, referring to the attached figures: 1. Dry-type hollow resistor; 2. Insulator; 3. Measurement origin; 4. Radial direction originating from the measurement origin. Detailed Implementation

[0040] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method and system for detecting inter-turn insulation short-circuit faults in a dry-type air-core reactor according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0042] The following description, in conjunction with the accompanying drawings, details the specific scheme of the method and system for detecting inter-turn insulation short-circuit faults in a dry-type air-core reactor provided by the present invention.

[0043] Please see Figure 1 The diagram illustrates a flowchart of a method for detecting inter-turn insulation short-circuit faults in a dry-type air-core reactor according to an embodiment of the present invention. The method includes the following steps:

[0044] S1: At the measurement points of each detection coil arranged radially below the bottom of the dry-type air-core reactor, the fundamental filtered induced voltage debase signal is simultaneously acquired; the measurement point directly below the center of the bottom of the dry-type air-core reactor is the measurement origin.

[0045] When a dry-type air-core reactor is in operation, the current will pass through each layer of coils of the reactor and generate a corresponding magnetic field. When an inter-turn insulation short circuit fault occurs, the spatial magnetic field distribution around the reactor will inevitably change. Moreover, compared with electrical quantity parameters, magnetic field distribution measurement can be carried out without contact detection, and there is no need to make secondary adjustments to the reactor layout.

[0046] In practical use, dry-type air-core reactors require insulators for support, and a certain amount of space is reserved at the bottom of the reactor. In this embodiment of the invention, the measurement origin is the measurement point directly below the center of the bottom of the dry-type air-core reactor. Please refer to [link / reference]. Figure 2 The figure shows a schematic diagram of a measurement origin arrangement provided by an embodiment of the present invention. In the figure, 1 is a dry-type hollow resistor, 2 is an insulator, 3 represents the measurement origin, and 4 represents the radial direction starting from the measurement origin. A measurement point is arranged every 100mm in the radial direction to form a magnetic array of the dry-type hollow reactor. A detection coil is placed at each measurement point.

[0047] Each detection coil is connected to a host computer via a voltage signal acquisition module. The magnetic array's detection coils synchronously acquire induced voltage signals for a preset duration. In this embodiment, the preset duration is 2 seconds, the sampling rate is 5 kHz, and the host computer is a microcontroller or computer. The host computer includes a wireless signal communication receiving module, which acquires the collected induced voltage signals via wireless communication. Specific acquisition settings can be adjusted by the implementer according to the specific implementation scenario and are not limited here.

[0048] Dry-type air-core reactors consist only of windings, which are actually hollow inductor coils. During normal operation, the current in all winding layers generates an alternating magnetic field around the reactor. Inter-turn insulation short circuits are one of the main causes of failure in dry-type air-core reactors. When the inter-turn insulation of the reactor coil is damaged, adjacent turns at the location of the insulation failure come into contact with each other under the action of electromagnetic force, thus forming a short-circuit loop. The short-circuit loop generates an induced current, i.e., a short-circuit current, under the action of the surrounding alternating magnetic field.

[0049] Considering that the induced voltage of the detection coil of a dry-type air-core reactor is a 50Hz sinusoidal signal during normal operation, when an inter-turn short circuit occurs, adjacent turns of the reactor continuously collide and separate. The magnetic field generated by the short-circuit current induces a corresponding voltage signal in the detection coil, which is superimposed on the 50Hz sinusoidal voltage. This manifests as a local sharp portion in the induced voltage waveform of the detection coil, causing local distortion of the induced voltage waveform. Simultaneously, the numerous high-voltage, high-current devices operating long-term in the substation where the dry-type air-core reactor is located will generate interference noise in the induced voltage signal of the detection coil. Both the interference noise and the inter-turn short-circuit fault signal appear as high-frequency signals relative to the 50Hz fundamental voltage.

[0050] Therefore, in order to facilitate the subsequent targeted improvement of fault signal detection, the acquired signal is first subjected to fundamental frequency filtering. In this embodiment of the invention, the induced voltage signal is subjected to notch filtering, and the induced voltage signal is used as the input of the notch filter. The notch frequency of the notch filter is set to 50Hz to remove the power frequency fundamental frequency component and obtain the induced voltage de-fundamental signal.

[0051] S2: Divide all induced voltage removal signals into different local signal segments according to the time sequence; for each local signal segment of each measurement point, determine the noise probability estimate of the measurement point in the local signal segment through autocorrelation intensity analysis.

[0052] The interference source equipment in the substation where the dry-type air-core controller is located is usually in a steady-state operation mode during local time periods. That is, the electrical parameters such as voltage and current are stable, and the electromagnetic coupling path between the interference source equipment and the detection coil is also time-invariant. The interference noise component in the induced voltage debase signal is mainly steady-state interference with fixed frequency characteristics in the substation, which usually has strong autocorrelation during local time periods.

[0053] Therefore, by dividing local time periods and analyzing the autocorrelation intensity, the degree to which the time period signal is affected by periodic steady-state electromagnetic noise can be reflected. In this embodiment of the invention, the induced voltage debasement signal of all measurement points is divided into a preset number of time windows as local signal segments according to the time series. The preset number of time windows is set to 10. There is no overlap between the local signal segments, and the division of all measurement points is consistent and uniform. The number of divisions can be adjusted by the implementer and is not limited here.

[0054] Furthermore, in this embodiment of the invention, for any local signal segment, the autocorrelation peak value and corresponding time delay of the local signal segment are obtained through the autocorrelation function. The autocorrelation function can effectively reveal the similarity of the signal itself under time shift, and it is suitable for identifying periodic or quasi-periodic components. In a specific embodiment of the invention, by setting the time delay of the autocorrelation function to 5ms to 100ms, with a total of 20 time delay increments of 5ms each, the autocorrelation value of the local signal segment and itself under each time delay is calculated, and the maximum autocorrelation value is extracted as the autocorrelation peak value of the local signal segment. At the same time, the time delay corresponding to the peak value is recorded as the corresponding time delay. It should be noted that autocorrelation function analysis is a well-known technique in the art. Its core is to determine the periodicity of the signal by the correlation between the signal and its own delayed signal. This invention only limits the specific range and increment settings of the time delay; the other conventional calculation procedures are not described in detail here.

[0055] Based on the deviation between the autocorrelation peak value of a local signal segment and the maximum autocorrelation peak value among all local signal segments, the correlation strength index of that local signal segment is obtained. The smaller the deviation, the larger the correlation strength index. The deviation of the autocorrelation peak value is negatively correlated with the correlation strength index. In a specific embodiment of the present invention, among all local signal segments at all measurement points, the largest autocorrelation peak value is taken as the peak extremum. The difference between the autocorrelation peak value and the peak extremum value of the local signal segment is negatively correlated and normalized to obtain the correlation strength index. This indicates that the stronger the autocorrelation of the local signal segment, the better it matches the periodic characteristics of the steady-state interference noise of the substation, and the better the correlation strength index. The difference is calculated as the absolute value of the difference between data points.

[0056] It should be noted that normalization and negative correlation mapping are techniques well known to those skilled in the art. Negative correlation mapping can take the form of an inverse proportional function or a negative exponential power function with the natural constant as the base. The normalization function can be linear normalization or standard normalization, etc. Specific methods are not limited or elaborated here.

[0057] Furthermore, based on the deviation between the time delay corresponding to the autocorrelation peak of the local signal segment and the mean time delay corresponding to the autocorrelation peak of all local signal segments, the correlation delay index of the local signal segment is obtained. The smaller the time delay deviation, the more consistent the location of the autocorrelation peak of the local signal segment is with most signal segments. The larger the correlation delay index, the more negatively correlated the time delay deviation is with the correlation delay index. In a specific embodiment of the present invention, the mean time delay corresponding to the autocorrelation peak of all local signal segments at all measurement points is used as the time delay mean. The difference between the time delay corresponding to the autocorrelation peak of the local signal segment and the time delay mean is negatively correlated and normalized to obtain the correlation delay index. The larger the correlation delay index, the closer the peak time delay of the local signal segment is to the global mean, the more significant the periodicity of the signal, and the better it matches the time-domain characteristics of steady-state interference noise.

[0058] Finally, by combining the correlation intensity index and correlation delay index of the local signal segment, a noise probability estimate for the local signal segment is obtained. The larger the correlation intensity index and correlation delay index, the more the local signal segment conforms to the characteristics of periodic steady-state noise in both intensity and temporal distribution. In one specific embodiment of the invention, the product of the correlation intensity index and the correlation delay index is used as the noise probability estimate for the local signal segment, quantifying the reliability of the local signal segment as a background noise sample. A larger noise probability estimate indicates that the local signal segment contains more steady-state interference noise components.

[0059] S3: Divide the spectrum of each local signal segment into different frequency bands in the spectrum diagram; at each frequency band, analyze the frequency band energy difference and noise probability estimate of each local signal segment between every two measurement points to determine the noise dominance of the frequency band; analyze the frequency band energy attenuation characteristics of each local signal segment in each frequency band along the radial distribution order starting from the measurement origin, and combine the noise dominance to obtain the fault significance value of each frequency band; select characteristic frequency bands through the fault significance value.

[0060] In the complex electromagnetic environment of substations, the inter-turn short-circuit fault signal and background noise of dry-type air-core reactors often overlap in the frequency domain, making effective separation difficult with simple time-domain or frequency-domain analysis. However, the two exhibit significant physical differences in spatial distribution. The background noise source typically manifests as far-field interference relative to the detection array, with its energy exhibiting strong spatial consistency across different magnetic measurement points. In contrast, the inter-turn short-circuit fault point is a near-field point source relative to the detection array, and the magnetic field signal it generates exhibits a significant gradient attenuation characteristic with increasing distance.

[0061] To distinguish between the high-frequency signal of inter-turn short-circuit fault superimposed on the induced voltage removal signal and the electromagnetic interference noise of the substation, a unified benchmark needs to be established for frequency domain analysis. Therefore, in this embodiment of the invention, frequency bands are uniformly divided. In this embodiment, the induced voltage removal signal in a local signal segment is subjected to a Fast Fourier Transform to obtain a spectrum. The horizontal axis of the spectrum represents frequency, and the vertical axis represents the corresponding amplitude. Multiple frequency bands are obtained using an equal-width division method. Each frequency band is a continuous frequency interval with a preset frequency spacing in the spectrum. In one specific embodiment of the invention, each 5Hz interval can be considered a frequency band interval, and the specific value can be adjusted by the implementer.

[0062] The energy distribution corresponding to the frequency bands in different local signal segments characterizes the intensity features of the signal within that frequency range, reflecting the accumulation of fault signals or noise in the frequency domain during that period. Therefore, in the spectrum diagram of each local signal segment, the root mean square of the amplitude corresponding to all frequencies within each frequency band is taken as the frequency band energy of the local signal segment in each frequency band, thereby quantifying the signal component intensity of each frequency band.

[0063] Considering that the energy distribution of substation interference noise is uniform in space, the frequency band energy difference between different measurement points in the same frequency band is small, and the local signal segment that is more severely affected by steady-state noise has a higher noise probability estimate, by analyzing the frequency band energy difference and noise probability estimate of each local signal segment between every two measurement points, the degree to which the frequency band is dominated by interference noise can be quantified.

[0064] Preferably, in this embodiment of the invention, the method for obtaining noise dominance includes:

[0065] For any frequency band, each pair of different measurement points is considered as a measurement group, and each measurement group is considered as an analysis group. For any local signal segment of the analysis group, based on the difference in frequency band energy between the two measurement points in the analysis group in that local signal segment, and combined with the sum of the noise probability estimates of the two measurement points in that local signal segment, the approximation of the spectral energy of the analysis group in that local signal segment is obtained.

[0066] In one specific embodiment of the present invention, the difference in frequency band energy between two measurement points in the local signal segment within the analysis group is negatively correlated and mapped to obtain an energy approximation index. A larger energy approximation index indicates that the energy levels of the two measurement points in that frequency band are closer, meaning the frequency component is more uniformly distributed between the two points in space, conforming to the spatial consistency characteristics of far-field background noise. Furthermore, the sum of the noise probability estimates of the two measurement points in the local signal segment is used as the approximation analysis confidence level. A higher approximation analysis confidence level indicates a higher degree to which the signals at both measurement points are background noise during that time period. Finally, the product of the energy approximation index and the approximation analysis confidence level is used as the spectral energy approximation of the local signal. A higher spectral energy approximation level indicates that the spectral characteristics of the analysis group in that frequency band are more dominated by interference noise during that local time period.

[0067] Furthermore, the spectral energy approximation of all local signal segments in the analysis group is integrated to obtain the spectral feature approximation of the analysis group. In this embodiment of the invention, the mean of the spectral energy approximation of all local signal segments in the analysis group is used as the spectral feature approximation of the analysis group, which reflects the overall spectral approximation level of the analysis group in this frequency band and avoids interference from local fluctuations on the results. The larger the mean, the more similar the overall spectral features of the two measurement points corresponding to the analysis group are in this frequency band, and the more likely they are to be dominated by the same steady-state interference noise.

[0068] Finally, the spectral feature approximation of all measurement groups in the frequency band is fused to obtain the noise dominance of the frequency band. In this embodiment of the invention, the sum of the spectral feature approximation of all measurement groups in the frequency band is taken as the noise dominance of the frequency band. The greater the noise dominance, the higher the spectral similarity of the frequency band in more measurement groups. That is, the signal characteristics of the frequency band are more consistent with the spatial distribution law of steady-state interference noise in the substation, that is, the energy distribution is uniform and the frequency domain characteristics of different measurement points are consistent, and the degree of dominance of interference noise is stronger.

[0069] Meanwhile, the magnetic field energy of the inter-turn short-circuit fault signal decays radially with a gradient, which means that the frequency band energy gradually decreases as the measurement point moves away from the measurement origin. Therefore, according to the radial distribution order of the measurement point starting from the origin, the frequency band energy attenuation characteristics of each local signal segment in each frequency band are analyzed. Then, this attenuation characteristic is combined with the noise dominance. The lower the noise dominance and the more the attenuation characteristic matches the fault pattern, the higher the probability that the frequency band contains fault information. Thus, the fault significance value characterizing the correlation between the frequency band and the fault is obtained.

[0070] In this embodiment of the invention, the measurement origin is used as the starting point, and the remaining measurement points are arranged in ascending order of their distance from the measurement origin to obtain a measurement distribution sequence. This sequence provides a matching relationship for subsequent energy distribution analysis and the spatial attenuation law of the fault magnetic field. For example, the total number of measurement points is 6, and the measurement distribution sequence corresponds to the magnetic measurement points numbered 1 to 6, with horizontal distances from the measurement origin of 0mm, 100mm, 200mm, 300mm, 400mm, and 500mm, respectively.

[0071] For any given frequency band, the energy of each local signal segment at the measurement point is arranged in the order of the measurement distribution sequence to obtain the energy arrangement sequence of each local signal segment. This sequence directly reflects the radial distribution trend of the frequency band energy.

[0072] The slope of the frequency band energy in the energy arrangement sequence of each local signal segment is obtained. In one specific embodiment of the present invention, the slope is obtained after performing linear fitting on the frequency band energy in the energy arrangement sequence. The positive or negative value of the slope directly reflects the radial trend of energy change. A negative slope indicates that the energy attenuates with increasing distance, which matches the characteristics of the fault magnetic field. It should be noted that the method of obtaining the slope by linear fitting is a well-known technique to those skilled in the art, such as the least squares linear fitting algorithm, and will not be elaborated here.

[0073] Therefore, the proportion of local signal segments with negative slopes is used as the magnetic attenuation of the frequency band. In this embodiment of the invention, the ratio of the number of local signal segments with negative slopes to the total number of local signal segments is used as the magnetic attenuation. The larger the magnetic attenuation, the more local signal segments in the frequency band have energy distributions that conform to the radial gradient attenuation law of the magnetic field of the inter-turn short circuit fault, and the higher the probability that the frequency band carries fault information.

[0074] Finally, the magnetic attenuation degree and noise dominance degree of the frequency band are integrated to obtain the fault significance value of the frequency band. The noise dominance degree is negatively correlated with the fault significance value. In a specific embodiment of the present invention, the product of the noise dominance degree of the frequency band after negative correlation mapping and the magnetic attenuation degree is used as the fault significance value of the frequency band, which characterizes the fault feature fit and interference suppression degree of the frequency band.

[0075] Finally, feature frequency bands with concentrated fault information and few interference components are selected based on fault significance values. In this embodiment of the invention, the segmentation threshold for fault significance values ​​of all frequency bands is obtained using the maximum inter-class variance method. This method iterates through all possible thresholds, calculates the inter-class variance of fault significance values ​​in frequency bands on both sides of the threshold, and selects the threshold with the largest inter-class variance as the optimal segmentation threshold. Frequency bands with fault significance values ​​higher than the segmentation threshold are selected as feature frequency bands. The selected frequency bands can reflect the high-frequency signal components of inter-turn short-circuit faults to the greatest extent, providing a key distribution part for subsequent wavelet denoising and extraction of fault signals. It should be noted that the maximum inter-class variance method is a well-known technique in the art and will not be elaborated here.

[0076] S4: Adaptive wavelet denoising is performed on the induced voltage signal at each measurement point, including: at each decomposition layer, the signal is decomposed by different wavelet basis functions, and the optimal wavelet basis is determined by the energy proportion of the characteristic frequency band corresponding to the decomposed wavelet coefficients in the spectrum; wavelet reconstruction is performed using the optimal wavelet basis of each layer to obtain the wavelet reconstructed signal of each measurement point.

[0077] Considering that the choice of wavelet basis function directly determines the wavelet denoising effect, an adaptive wavelet denoising strategy is adopted. Combining the selected feature frequency bands with concentrated fault information and few interference components, the energy ratio of the coefficients after wavelet decomposition in the feature frequency band is used as the core evaluation standard for wavelet basis fit. That is, the higher the decomposition quality, the more accurately the corresponding wavelet basis can focus the fault signal on the feature frequency band, while suppressing the residual noise in non-feature frequency bands.

[0078] Therefore, in this embodiment of the invention, the signal is decomposed by different wavelet basis functions at each decomposition layer. Wavelet coefficients are obtained after each wavelet basis function is decomposed. The wavelet basis functions can be selected from the db series and sym series wavelet basis functions, such as db2-db20 and sym2-sym20. The implementer can adjust the selection range according to the implementation scenario.

[0079] The goodness-of-decomposition of the wavelet coefficients is calculated by taking the sum of the frequency band energies of all characteristic frequency bands in the frequency domain transformed spectrum as the proportion of the sum of the frequency band energies of all frequency bands. A higher goodness-of-decomposition indicates that the corresponding wavelet basis can more effectively concentrate signal energy in the fault characteristic frequency band during this decomposition layer. The wavelet basis function with the highest goodness-of-decomposition is selected as the optimal fundamental wavelet for each decomposition layer, ensuring that this layer maximizes the extraction of fault features and mitigates interference effects.

[0080] After selecting the wavelet basis with the highest decomposition goodness as the optimal wavelet basis for each layer, the signal is reconstructed layer by layer using the optimal wavelet basis of each layer. Finally, the wavelet reconstructed signal of each measurement point is obtained, providing a clean signal with high signal-to-noise ratio and sufficient retention of fault features for subsequent fault detection based on signal energy. It should be noted that the core process of wavelet decomposition and reconstruction is a technique well known to those skilled in the art. Its key improvement lies in adapting the frequency domain characteristics of signals at different layers by selecting independent basis at each layer. Other conventional processing procedures will not be elaborated here.

[0081] S5: Fault detection is performed based on the positional distribution of each measurement point and the measurement origin, as well as the signal energy of the wavelet reconstructed signal.

[0082] Considering that the wavelet reconstructed signal after adaptive wavelet denoising has effectively preserved the sharp pulse characteristics of the inter-turn short circuit fault and suppressed most of the background noise, its signal energy is directly related to the severity of the fault. That is, the more severe the fault, the stronger the magnetic field generated by the short circuit current and the greater the induced signal energy. At the same time, since the magnetic field of the inter-turn short circuit fault exhibits a gradient decay characteristic along the radial direction, the closer to the measurement origin, the more significant the change in magnetic field and the higher the signal energy. Near-site points are more sensitive to minor faults. Therefore, it is necessary to combine the positional distribution pattern of each measurement point with the measurement origin and the energy of the wavelet reconstructed signal for fault detection.

[0083] In this embodiment of the invention, the wavelet coefficient energy at each measurement point is obtained based on the signal value of the wavelet reconstructed signal at each measurement point. Specifically, the sum of the squares of the amplitudes of each point in the wavelet reconstructed signal sequence at that measurement point is calculated as the wavelet coefficient energy, which is an energy index characterizing the intensity of magnetic field fluctuations at that location. This quantifies the strength of fault features in the wavelet reconstructed signal, and the greater the energy, the more significant the fault signal.

[0084] Furthermore, the measurement origin and a preset number of measurement points closest to the measurement origin are designated as neighboring measurement points. When the wavelet coefficient energy of all neighboring measurement points is greater than a preset energy threshold, it indicates that the dry-type air-core reactor has a minor inter-turn insulation short-circuit fault. Because the neighboring measurement points are closest to the reactor axis, they are most sensitive to minor faults with few short-circuit turns and weak magnetic field changes, thus triggering a fault alarm. In this embodiment of the invention, the preset number is set to 2, and the preset energy threshold is 1.25 times the historical maximum wavelet coefficient energy of the measurement point under fault-free operation. The specific value is obtained through factory calibration of the equipment or statistical analysis of long-term steady-state operation data. The implementer can adjust the specific value as needed, and no restrictions are imposed here.

[0085] Otherwise, it indicates that the characteristics of a minor fault are not obvious, and further judgment is needed to determine whether a serious inter-turn insulation short-circuit fault exists. Therefore, the magnetic field variation range of a serious fault is wide, and significant energy changes will occur at multiple measurement points. The number of measurement points with wavelet coefficient energy greater than a preset energy threshold is counted as the suspected fault count. A higher suspected fault count indicates a wider fault impact range and a higher severity. When the proportion of suspected faults exceeds the preset fault count proportion, it indicates that the dry-type air-core reactor has a serious inter-turn insulation short-circuit fault, and a fault alarm is triggered. In this embodiment of the invention, the preset fault count proportion is 50%, and the specific value can be adjusted by the implementer.

[0086] In summary, this invention utilizes a radially arranged array of detection coils at the bottom of the reactor and performs segmented signal analysis. It leverages the strong autocorrelation of background noise and the suddenness of fault signals to construct a noise probability estimate. Combining the physical characteristics of radial attenuation of the fault magnetic field signal and the spatial consistency of background noise, it analyzes the energy differences between different measurement points in the same frequency band and the radially distributed frequency band energy attenuation characteristics to screen out characteristic frequency bands more likely to characterize the fault, enhancing subsequent fault signal identification. Furthermore, adaptive wavelet denoising is performed using these characteristic frequency bands, selecting the optimal wavelet basis for denoising based on the energy proportion of the characteristic frequency band, more effectively preserving the time-frequency characteristics of inter-turn short-circuit faults. Finally, a comprehensive fault determination is made based on the spatial distribution of measurement points and the signal energy of the wavelet-reconstructed signal. This invention utilizes adaptive wavelets based on decomposition goodness, leveraging the local signal autocorrelation regularity and the attenuation characteristics of the radial magnetic field distribution to enhance the preservation and identification of possible fault signals. While suppressing complex background noise interference in substations, it improves fault identification capabilities and enhances the accuracy and reliability of detecting inter-turn insulation short-circuit faults in dry-type air-core reactors.

[0087] The present invention also provides a dry-type air-core reactor inter-turn insulation short-circuit fault detection system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described dry-type air-core reactor inter-turn insulation short-circuit fault detection method.

[0088] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0089] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A dry-type air-core reactor inter-turn insulation short-circuit fault detection method, characterized by, The method comprises: Synchronously collecting fundamental wave filtered induced voltage debase signals at the measuring points of each detection coil arranged in the radial direction below the bottom of the dry-type air-core reactor; a measuring origin is a measuring point directly below the center of the bottom of the dry-type air-core reactor; Dividing all induced voltage debase signals into different local signal segments in time sequence; determining noise probability estimates of the measuring points in the local signal segments through autocorrelation intensity analysis of each local signal segment of each measuring point; Uniformly dividing the frequency spectrum of each local signal segment into different frequency bands in a frequency spectrum diagram; analyzing the energy difference of each local signal segment and the noise probability estimate between each two measuring points in each frequency band to determine the noise dominance degree in the frequency band; analyzing the frequency band energy attenuation characteristics of each local signal segment in each frequency band in the radial distribution sequence starting from the measuring origin at the measuring point to obtain the fault saliency value of each frequency band in combination with the noise dominance degree; and screening out characteristic frequency bands through the fault saliency value. Performing adaptive wavelet denoising on the induced voltage debase signals of each measuring point, including: decomposing the signals through different wavelet basis functions at each decomposition layer to determine the optimal wavelet basis according to the energy proportion of the corresponding characteristic frequency band of the wavelet coefficients in the frequency spectrum after decomposition; and performing wavelet reconstruction using the optimal wavelet basis of each layer to obtain the wavelet reconstruction signals of each measuring point. Performing fault detection based on the position distribution of each measuring point and the measuring origin and the signal energy of the wavelet reconstruction signals.

2. The method according to claim 1, wherein The method for obtaining the noise probability estimate comprises: For any one local signal segment, obtaining the autocorrelation peak value and the corresponding time delay of the local signal segment through the autocorrelation function; obtaining the correlation intensity index of the local signal segment according to the deviation between the autocorrelation peak value of the local signal segment and the maximum autocorrelation peak value in all local signal segments, the deviation of the autocorrelation peak value being negatively correlated with the correlation intensity index; obtaining the correlation time delay index of the local signal segment according to the deviation between the autocorrelation peak value corresponding time delay of the local signal segment and the average of the autocorrelation peak value corresponding time delays of all local signal segments, the deviation of the time delay being negatively correlated with the correlation time delay index; obtaining the noise probability estimate of the local signal segment in combination with the correlation intensity index and the correlation time delay index of the local signal segment.

3. The method according to claim 1, wherein The method for obtaining the noise dominance degree comprises: For any one frequency band, taking each two different measuring points as a measuring group and sequentially taking each measuring group as an analysis group, for any one local signal segment of the analysis group, obtaining the spectral energy approximation of the analysis group in the local signal based on the energy difference of the frequency band between the two measuring points in the analysis group in the local signal segment in the frequency band and in combination with the sum of the noise probability estimates of the two measuring points in the local signal segment, the energy difference of the frequency band being negatively correlated with the spectral energy approximation, the sum of the noise probability estimates being positively correlated with the spectral energy approximation; fusing the spectral energy approximations of all local signal segments in the analysis group to obtain the spectral feature approximation of the analysis group; and fusing the spectral feature approximations of all measuring groups in the frequency band to obtain the noise dominance degree of the frequency band.

4. The method according to claim 1, wherein The method for obtaining the fault saliency value comprises: Taking the measurement origin as a starting point, the rest of the measurement points are arranged in order of distance from the measurement origin from small to large to obtain a measurement distribution sequence; for any frequency band, the frequency band energy of each local signal segment at the measurement point in the frequency band is arranged in order of the measurement distribution sequence to obtain an energy arrangement sequence of each local signal segment; Obtain the slope of the frequency band energy in the energy arrangement sequence of each local signal segment; and count the proportion of the number of local signal segments with negative slopes as the magnetic variation attenuation degree of the frequency band. Fuse the magnetic variation attenuation degree and the noise dominance degree of the frequency band to obtain the fault saliency value of the frequency band, wherein the noise dominance degree and the fault saliency value are negatively correlated, and the magnetic variation attenuation degree and the fault saliency value are positively correlated.

5. The method according to claim 1, wherein The feature frequency band is screened out through the fault saliency value, and the method comprises: Obtain the segmentation threshold of the fault saliency value of all frequency bands by using the maximum inter-class variance method; and take the frequency band with a fault saliency value higher than the segmentation threshold as the feature frequency band.

6. The method according to claim 1, wherein The method for obtaining the optimal wavelet basis comprises: In each decomposition layer, for each wavelet coefficient after decomposition of each wavelet basis function, the proportion of the sum of the frequency band energies of all feature frequency bands in the frequency spectrum diagram after frequency domain transformation of the wavelet coefficient to the sum of the frequency band energies of all frequency bands is taken as the decomposition degree of the wavelet coefficient; and the wavelet basis function with the maximum decomposition degree is taken as the optimal wavelet basis of each layer decomposition.

7. The method according to claim 1, wherein The fault detection is performed based on the position distribution of each measurement point and the measurement origin and the signal energy of the wavelet reconstruction signal, and the method comprises: Obtain the wavelet coefficient energy at each measurement point based on the signal value of the wavelet reconstruction signal at each measurement point; take the measurement origin and the nearest preset number of measurement points from the measurement origin as adjacent measurement points; and when the wavelet coefficient energies of the adjacent measurement points are all greater than a preset energy threshold, perform a fault alarm; Otherwise, count the number of measurement points with a wavelet coefficient energy greater than the preset energy threshold in all measurement points as a suspected number, and when the proportion of the suspected number exceeds a preset fault number proportion, perform a fault alarm.

8. The method according to claim 3, wherein The method for obtaining the induced voltage debase signal comprises: Collect the induced voltage signal at each measurement point, perform notch filtering on the induced voltage signal to remove the fundamental wave component, and obtain the induced voltage debase signal.

9. The method according to claim 1, wherein The method for determining the frequency band energy comprises: In the frequency spectrum diagram of each local signal segment, take the root mean square of the amplitudes of all frequencies in each frequency band as the frequency band energy of the local signal segment in each frequency band.

10. A dry-type air-core reactor turn-to-turn insulation short-circuit fault detection system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the dry-type air-core reactor turn-to-turn insulation short-circuit fault detection method according to any one of claims 1-9.

Images