A power distribution network high resistance ground fault detection method and device

Abstract

The embodiment of the application discloses a kind of distribution network high resistance ground fault detection method and device.The method comprises: based on the zero sequence current of each line in distribution network, determine to be detected line;To the zero sequence current corresponding to the line to be detected and the other zero sequence current corresponding to each other line, cross wavelet coherence analysis is carried out, and cross wavelet power spectral density is obtained;Determine target scale coefficient based on cross wavelet power spectral density;Based on target structure operator, mathematical morphological transformation is carried out to the zero sequence current under target scale coefficient, and the target current signal after transformation is obtained;The estimation of fractal dimension under different time scales is carried out to the target current signal, and determines fractal dimension vector;High resistance ground fault detection is carried out based on mixed Gaussian model and fractal dimension vector, and the fault detection result corresponding to the line to be detected is determined.Through the technical scheme of the embodiment of the application, the accuracy of high resistance ground fault detection in distribution network can be improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to computer technology, and particularly relate to a power distribution network high impedance ground fault detection method and device. BACKGROUND

[0002] The power distribution network has the characteristics of complex structure, many branches and high fault rate, and is easily affected by factors such as wind, tree barriers, animals and humid weather conditions to cause faults. When contacting the tower, tree branches or grassland and other high impedance media, a high impedance ground fault (HIF) is caused. Because the fault transition resistance is above several thousand ohms, the fault current is usually below 50A, so it is difficult to detect HIF based on power frequency quantities such as overcurrent, recloser and fuse. Because it is difficult to detect HIF, the exposed live conductor may cause accidents such as fire and even pose a serious threat to personal safety.

[0003] At present, the high penetration rate of new energy and the high proportion of power electronic equipment access to the power distribution network have produced new fault characteristics. The large-scale grid connection of distributed new energy further increases the variability and uncertainty of the topology structure of the power distribution network, as well as the harmonics and noise introduced by power electronic equipment, which further complicates the harmonic environment of the power distribution network, resulting in significant changes in the fault waveform. Especially in the case of high impedance nonlinear fault of weak conductive complex medium grounding, the nonlinear characteristics such as fault waveform distortion and spectral characteristics are greatly affected. Therefore, there is an urgent need for a sensitive and reliable power distribution network high impedance ground fault detection method. SUMMARY

[0004] Embodiments of the present application provide a power distribution network high impedance ground fault detection method and device to improve the accuracy of high impedance ground fault detection in the power distribution network.

[0005] In a first aspect, embodiments of the present application provide a power distribution network high impedance ground fault detection method, comprising:

[0006] Determining a to-be-detected line from all lines based on the zero sequence current of each line in the power distribution network;

[0007] Performing cross wavelet coherence analysis on the to-be-detected zero sequence current corresponding to the to-be-detected line and the other zero sequence current corresponding to each other line except the to-be-detected line to obtain the cross wavelet power spectrum density between the to-be-detected line and each other line;

[0008] Determining a target scale coefficient corresponding to the to-be-detected zero sequence current based on the cross wavelet power spectrum density;

[0009] Performing mathematical morphological transformation on the to-be-detected zero sequence current at the target scale coefficient based on a target structure operator to obtain a transformed target current signal;

[0010] performing fractal dimension estimation on the target current signal under different time scales to determine a fractal dimension vector;

[0011] performing high-resistance ground fault detection based on the Gaussian mixture model and the fractal dimension vector to determine a fault detection result corresponding to the line to be detected.

[0012] In a second aspect, an embodiment of the present application further provides a power distribution network high-resistance ground fault detection device, comprising:

[0013] a line to be detected determination module configured to determine a line to be detected from all lines based on zero sequence currents of each line in the power distribution network;

[0014] a cross wavelet coherence analysis module configured to perform cross wavelet coherence analysis on a to-be-detected zero sequence current corresponding to the line to be detected and other zero sequence currents corresponding to each other line except the line to be detected to obtain cross wavelet power spectral densities between the line to be detected and each other line;

[0015] a target scale coefficient determination module configured to determine a target scale coefficient corresponding to the to-be-detected zero sequence current based on the cross wavelet power spectral densities;

[0016] a mathematical morphology transformation module configured to perform mathematical morphology transformation on the to-be-detected zero sequence current under the target scale coefficient based on a target structure operator to obtain a transformed target current signal;

[0017] a fractal dimension estimation module configured to perform fractal dimension estimation on the target current signal under different time scales to determine a fractal dimension vector;

[0018] a high-resistance ground fault detection module configured to perform high-resistance ground fault detection based on the Gaussian mixture model and the fractal dimension vector to determine a fault detection result corresponding to the line to be detected.

[0019] An embodiment of the above application has the following advantages or beneficial effects:

[0020] The zero sequence current based on each line in the power distribution network is determined from all lines; cross wavelet coherence analysis is performed on the to-be-detected zero sequence current corresponding to the to-be-detected line and other zero sequence currents corresponding to each other line, cross wavelet power spectral densities between the to-be-detected line and each other line are obtained, the optimal target scale coefficient of the to-be-detected zero sequence current is determined based on the cross wavelet power spectral densities; the mathematical morphology transformation is performed on the to-be-detected zero sequence current under the optimal target scale coefficient based on the target structure operator, the transformed target current signal is obtained, so as to filter out noise and interference signals and enhance the weak response of the high-resistance ground fault at the occurrence moment; the fractal dimension under different time scales of the target current signal is estimated, the fractal dimension vector under different time scales is determined, and the high-resistance ground fault detection is performed based on the mixed Gaussian model and the fractal dimension vector, the fault detection result corresponding to the to-be-detected line is determined, so as to improve the accuracy of the high-resistance ground fault detection in the power distribution network. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0022] Figure 1 is a flow chart of a power distribution network high-resistance ground fault detection method provided by an embodiment of the present application;

[0023] Figure 2 is a flow chart of another power distribution network high-resistance ground fault detection method provided by an embodiment of the present application;

[0024] Figure 3 is a flow chart of still another power distribution network high-resistance ground fault detection method provided by an embodiment of the present application;

[0025] Figure 4 is an example diagram of a double-logarithmic slope fitting related to an embodiment of the present application;

[0026] Figure 5 is a structural schematic diagram of a power distribution network high-resistance ground fault detection device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0027] The present application will be further described in detail below in combination with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application are shown in the drawings, but not all the structures.

[0028] Figure 1 A flow chart of a method for detecting high resistance grounding fault of a power distribution network is provided for an embodiment of the present application. The embodiment can be applied to detect whether a high resistance grounding fault occurs in a line of the power distribution network. The method can be executed by a device for detecting high resistance grounding fault of the power distribution network. The device can be realized by software and / or hardware, and integrated in an electronic device. As shown in the figure, the method specifically includes the following steps: Figure 1

[0029] S110, determining a to-be-detected line from all lines based on zero sequence currents of each line in the power distribution network.

[0030] The zero sequence current refers to a leakage current flowing in a loop when a shock or leakage fault occurs in a circuit. At this time, the three-phase current vector passing through the mutual inductor is not equal to zero, and the current generated thereby is the zero sequence current. The zero sequence current can refer to a fault signal generated when a line fault occurs. The to-be-detected line refers to a line that is likely to generate a high resistance grounding fault in all lines. Other lines except the to-be-detected line will not generate a high resistance grounding fault, so that only the to-be-detected line needs to be detected for a high resistance grounding fault.

[0031] Specifically, the zero sequence currents of each line in a certain time range are collected based on a preset sampling rate (such as 1.25 MHz) and a preset sampling interval (such as 0.8 μs). For example, when a fault detection threshold is reached, a fault detection operation is started, and the zero sequence currents of all lines in a time range from 1 ms before the start of the fault detection operation to 15 ms after the start are obtained. It should be noted that the obtained zero sequence current is a time domain signal, which includes a plurality of current values sampled. By comparing the zero sequence currents of each line, the line that is most likely to generate a high resistance grounding fault in all lines is determined as the to-be-detected line.

[0032] Exemplarily, the step S110 can include: determining an energy of each zero sequence current based on the zero sequence current of each line in the power distribution network; and determining a line with the largest energy of the zero sequence current as the to-be-detected line.

[0033] Specifically, for each line, the absolute values of all current values in the zero sequence current of the line are added, and the addition result is obtained as the energy of the zero sequence current. For example, the energy E of the zero sequence current of the mth line can be determined by the following formula: m

[0034]

[0035] where i m ​​is the zero sequence current of the mth line, j is the sampling point sequence number, N is the total number of sampling points, and M is the total number of lines. By comparing the energy of the zero sequence current of each line, the line x with the maximum energy is taken as the to-be-detected line, that is, E x m , m e [1, M].

[0036] S120, cross wavelet coherence analysis is performed on the to-be-detected zero sequence current corresponding to the to-be-detected line and the other zero sequence current corresponding to each other line except the to-be-detected line, to obtain the cross wavelet power spectrum density between the to-be-detected line and each other line.

[0037] wherein the to-be-detected zero sequence current refers to the zero sequence current of the to-be-detected line. The other zero sequence current refers to the zero sequence current of the other line. The cross wavelet coherence analysis refers to cross wavelet transform of two energy-limited non-stationary signals in the time-frequency domain to determine the time-frequency correlation characteristics between the two signals. If the cross wavelet coherence of a certain frequency band of the two signals is large, it indicates that the content of the frequency band component in the two signals is large. Since noise has randomness and mutual independence, cross wavelet transform can effectively shield the influence of noise, and thus can accurately select the best scale coefficient in the process of detecting high-resistance grounding faults in the power distribution network.

[0038] Specifically, for each other line m, the to-be-detected zero sequence current i x (t) of the to-be-detected line x is wavelet transformed to obtain the wavelet transform result W x (a, b). For example, wherein ψ is a mother wavelet function, a (a > 0) is a scale factor, b is a time shift, and * represents a complex conjugate. The mother wavelet function can be, but is not limited to, a complex wavelet function, i.e., a Morlet wavelet. The Morlet wavelet has good time-frequency localization characteristics, and its mathematical expression is: Similarly, the other zero sequence current i m (t) of the other line m is wavelet transformed to obtain the wavelet transform result W m (a, b). Based on the wavelet transform result W x (a, b) of the to-be-detected line x and the wavelet transform result W m (a, b) of the other line m, cross transform is performed to obtain the cross wavelet transform result W xm (a, b), that is, W xm (a, b) = W x (a, b) W m * (a, b). Based on the cross wavelet transform result W xm (a, b), the cross wavelet power spectrum density C xm ​(a, b), i.e. C xm (a, b) = W xm * (a, b)W xm (a, b) = |W xm (a, b) 2 The cross wavelet power spectrum density C xm (a, b) is larger, indicating that the to-be-detected zero sequence current i x (t) of the to-be-detected line x and the other zero sequence current i m (t) of the other line m is larger.

[0039] It should be noted that the cross wavelet power spectrum density C xm (a, b) = |W xm (a, b) 2 , i.e. the amplitude square of the cross wavelet transform of the signal, which is an energy distribution with the scale factor a and the time shift b. Since the scale factor a indirectly corresponds to the frequency (the smaller a is, the higher the frequency is), the cross wavelet power spectrum density is essentially also a time-frequency distribution composed of the scale factor a and the time shift b.

[0040] S130, determining a target scale coefficient corresponding to the to-be-detected zero sequence current based on the cross wavelet power spectrum density.

[0041] The target scale coefficient refers to the best scale coefficient for processing the to-be-detected zero sequence current, i.e. the best scale coefficient of the fault feature. Specifically, based on the cross wavelet power spectrum density between the to-be-detected line and each other line, a time-frequency information region with a higher correlation degree between the to-be-detected line and each other line can be determined, and the best target scale coefficient can be determined based on the time-frequency information region, so as to realize the selection of the adaptive best scale coefficient.

[0042] It should be noted that since the power distribution network fault signal contains strong background noise and harmonic interference components, the cross wavelet coherence analysis can be used to select the best scale coefficient of the fault feature, so as to reduce noise interference and enhance the fault feature in the signal.

[0043] S140, performing mathematical morphology transformation on the to-be-detected zero sequence current under the target scale coefficient based on a target structure operator, to obtain a transformed target current signal.

[0044] The mathematical morphology is a nonlinear signal processing and analysis method, which starts from the object features and geometric structure of the research object, and investigates the mutual correlation between the parts of the signal. Therefore, the mathematical morphology can effectively reflect the local shape characteristics of the zero sequence current signal, and in the calculation process, the noise and interference signals in the signal can be effectively filtered by selecting the structure operator, so that the fault characteristics are more obvious. Moreover, the mathematical morphology is completed by simple addition, subtraction and comparison operations, and compared with the wavelet transform and other methods, the mathematical morphology has more advantages in calculation speed and can better meet the real-time requirements of the power system. The structure operator plays a role of window filtering in the mathematical morphology transformation, and its scale and shape have important influence on the transformation result. For example, the shape of the structure operator can be circular, semicircular, straight line, curve, triangle, polygon and their mutual combination. According to the structural characteristics of the signal, the corresponding structure operator is selected to effectively decompose the signal by using the mathematical morphology. The target structure operator refers to the structure operator used in the high resistance grounding fault detection process, so that under the condition of a certain decomposition times, the steepness and gentleness of different fault pulses can be effectively distinguished, and the fault characteristics are amplified.

[0045] Specifically, based on the optimal target scale coefficient and the target structure operator, a morphological filter can be constructed. By the morphological filter, the mathematical morphology transformation of the zero sequence current to be detected is performed under the target scale coefficient, and the target current signal transformed under the optimal target scale coefficient is obtained, so that the ideal decomposition result is obtained, the noise and interference signals in the waveform are filtered, and the fault characteristics are more obvious.

[0046] S150, estimating the fractal dimension under different time scales of the target current signal, and determining a fractal dimension vector.

[0047] The fractal dimension is used for estimation because the target current signal after the morphological transformation is relatively complex. The fractal dimension is originally used to characterize the geometric pattern generated by the abstract recursive process called fractal process, and has the advantage of simple calculation. The fractal dimension is essentially an approximate sequence related to the decreasing scale, which is a geometric factor of the simple figure forming the approximation. The fractal dimension vector is composed of the fractal dimensions estimated under different time scales.

[0048] Specifically, the target current signal is transformed under different time scales to obtain a reference current signal under each time scale. The fractal dimension of the target current signal and the reference current signal under each time scale can be estimated by using a box counting method, the fractal dimension corresponding to the target current signal and the fractal dimension corresponding to each reference current signal are determined, and all the fractal dimensions are combined to obtain a fractal dimension vector. It should be noted that since the fault signal is not strictly self-similar, different fault signals may have the same fractal dimension calculated at a certain time scale. Therefore, a single time scale fractal dimension is not sufficient to represent the complexity and randomness of the signal, and thus multiple time scale fractal dimensions need to be determined. Since the fault characteristics in the zero sequence current to be detected are still not obvious, the fault feature enhancement method of mathematical morphology and fractal dimension is used to amplify the weak response of the fault when the fault occurs, to provide more effective feature information for subsequent detection, and thus to improve the accuracy of fault detection.

[0049] S160, performing high resistance ground fault detection based on the Gaussian mixture model and the fractal dimension vector, and determining a fault detection result corresponding to the line to be detected.

[0050] The Gaussian mixture model (GMM) is a clustering method based on a probability model. The Gaussian mixture model can assume that the input sample is subject to K unknown Gaussian distributions, and samples subject to the same distribution are clustered into a class. The Gaussian mixture model can be regarded as a linear combination of K Gaussian distributions, that is, where x represents a state variable; K represents the number of sub-distributions in the Gaussian mixture model, that is, the number of clusters; π k represents a mixing coefficient (that is, a weight) and satisfies 0≤π k ≤1 and N(x|μ k ,Σ k ) represents a Gaussian distribution, μ k and μ k represent the mean and covariance of the kth sub-model. The Gaussian mixture model can perform unsupervised learning clustering on sample data, thereby dividing the sample data into two Gaussian sub-models of normal and fault. The sample data can include fractal dimension vectors corresponding to multiple sample zero sequence currents. Since the Gaussian mixture model is used to detect whether there is a high resistance ground fault in the line, the sample data only needs to be divided into two categories, that is, one category of normal data and another category of fault data. For example, the trained Gaussian mixture model includes a normal Gaussian sub-model and a fault Gaussian sub-model.

[0051] Specifically, the fractal dimension vector corresponding to the zero sequence current to be detected is input into the mixed Gaussian model after training, and the mixed Gaussian model can divide the input fractal dimension vector into a normal Gaussian sub-model or a fault Gaussian sub-model based on the cluster centers corresponding to the normal Gaussian sub-model and the cluster centers corresponding to the fault Gaussian sub-model, for example, determine the Euclidean distance between the input fractal dimension vector and the cluster center of each sub-model, and divide the input fractal dimension vector into the sub-model with the smallest distance, so that the mixed Gaussian model can accurately detect whether the line to be detected has a high resistance ground fault and obtain a fault detection result.

[0052] For example, step S160 can include inputting the fractal dimension vector into the mixed Gaussian model for high resistance ground fault detection, and determining the fault detection result corresponding to the line to be detected based on the output of the mixed Gaussian model.

[0053] The mixed Gaussian model is a normal Gaussian sub-model and a fault Gaussian sub-model obtained by clustering sample data based on the maximum expectation algorithm in advance. Because the probability density distributions of fault data and normal data are significantly different, two Gaussian sub-models with large differences in mean and variance can accurately identify fault data and normal data.

[0054] Specifically, by inputting the fractal dimension vector corresponding to the zero sequence current to be detected into the clustered mixed Gaussian model, the mixed Gaussian model re-clusters the input new fractal dimension vector and outputs the probability value of the vector data belonging to fault data. If the probability value output by the mixed Gaussian model is greater than or equal to a preset threshold, it is determined that the fault detection result is that the line to be detected has a high resistance ground fault, otherwise it is determined that the fault detection result is that the line to be detected has no high resistance ground fault.

[0055] The technical scheme of the embodiment determines the line to be detected from all lines based on the zero sequence current of each line in the power distribution network, performs cross wavelet coherence analysis on the zero sequence current to be detected corresponding to the line to be detected and the other zero sequence currents corresponding to each other line, obtains the cross wavelet power spectral density between the line to be detected and each other line, determines the best target scale coefficient of the zero sequence current to be detected based on the cross wavelet power spectral density, performs mathematical morphological transformation on the zero sequence current to be detected at the best target scale coefficient based on a target structure operator, obtains a transformed target current signal, thereby filtering out noise and interference signals and enhancing the weak response of the high resistance ground fault at the occurrence moment, estimates the fractal dimension at different time scales of the target current signal, determines the fractal dimension vector at different time scales, and performs high resistance ground fault detection based on the mixed Gaussian model and the fractal dimension vector to determine the fault detection result corresponding to the line to be detected, thereby improving the accuracy of high resistance ground fault detection in the power distribution network.

[0056] Figure 2 The flow chart of another power distribution network high resistance ground fault detection method provided by an embodiment of the present application is based on the above embodiments, and the determination process of the target scale coefficient and the mathematical morphology transformation process are described in detail. The explanations of the same or corresponding terms in the above embodiments are not repeated here.

[0057] Referring to Figure 2 , the another power distribution network high resistance ground fault detection method provided by the embodiment specifically includes the following steps:

[0058] S210, determining a to-be-detected line from all lines based on the zero sequence current of each line in the power distribution network.

[0059] S220, performing cross wavelet coherence analysis on the to-be-detected zero sequence current corresponding to the to-be-detected line and the other zero sequence current corresponding to each other line except the to-be-detected line, to obtain the cross wavelet power spectrum density between the to-be-detected line and each other line.

[0060] S230, determining the time-frequency information region when the correlation degree between the to-be-detected line and each other line is greater than a preset correlation degree based on the cross wavelet power spectrum density.

[0061] The time-frequency information region refers to a region composed of all time-frequency information meeting the condition, that is, a high correlation degree time-frequency region between the to-be-detected line and the other line. Each time-frequency information includes a scale factor a and a time shift b.

[0062] Specifically, in order to unify the selection criteria in all cases, the cross wavelet power spectrum densities are normalized. The normalized cross wavelet power spectrum density is actually a "significance level" in a statistical sense. The normalized cross wavelet power spectrum density can be regarded as the correlation degree between the to-be-detected line and the other line. For each other line, the cross wavelet power spectrum density C xm (a,b) between the to-be-detected line x and each other line m can be determined by the following formula: xm

[0063]

[0064] wherein σ x is the standard deviation of the zero sequence current of the to-be-detected line x; σ m is the standard deviation of the zero sequence current of the other line m. p refers to the preset correlation degree, for example, in order to obtain the best scale coefficient, p = 0.05. The above formula can be regarded as the time-frequency information region B​xm The inner time-frequency information is tested by a p-significance level.

[0065] S240, determining an overlapping region between all time-frequency information regions, and determining a target scale factor based on a number of overlapping regions.

[0066] The overlapping region can refer to an intersection region between all time-frequency information regions. The number of overlapping regions can be one or more. The target scale factor can refer to an optimal scale factor a. Specifically, based on the number of overlapping regions between all time-frequency information regions, the optimal target scale factor can be accurately selected adaptively.

[0067] For example, the step S240 of "determining a target scale factor based on the number of overlapping regions" can include: if there is only one overlapping region, determining a target time-frequency information when the cross wavelet power spectrum density is maximum in the overlapping region, and determining the scale factor in the target time-frequency information as the target scale factor; if there are at least two non-connected overlapping regions, determining a candidate time-frequency information when the cross wavelet power spectrum density is maximum in each overlapping region, and determining the minimum scale factor in all candidate time-frequency information as the target scale factor.

[0068] Specifically, in the case of only one overlapping region, the overlapping region can be defined as the selected optimal time-frequency window, and the cross wavelet power spectrum density corresponding to each time-frequency information in the overlapping region is determined, and the time-frequency information when the cross wavelet power spectrum density is maximum is determined as the target time-frequency information, and the scale factor a in the target time-frequency information is determined as the optimal target scale factor. In the case of at least two non-connected overlapping regions, the cross wavelet power spectrum density corresponding to each time-frequency information in each overlapping region is determined, and the time-frequency information when the cross wavelet power spectrum density is maximum in each overlapping region is determined as the candidate time-frequency information. The scale factors a in all candidate time-frequency information are compared, and the minimum scale factor a is determined as the optimal target scale factor.

[0069] S250, determining a scale coefficient corresponding to the to-be-detected zero sequence current based on the target scale factor, and performing integer processing on the scale coefficient to obtain an integer-processed target scale coefficient.

[0070] Specifically, the cross wavelet transform is a continuous wavelet transform based on a complex wavelet function, and the numerical relationship between the scale factor a and the degree coefficient j is a = 2 j Accordingly, the optimal scale coefficient j corresponding to the target scale factor a can be determined. Since the determined optimal scale coefficient j is in decimal form, the optimal scale coefficient needs to be integer-processed, and the integer obtained is used as the final optimal target scale coefficient, so as to perform mathematical morphology transformation and the like based on the integer form of the target scale coefficient subsequently.

[0071] S260, under the target scale coefficient, the to-be-detected zero sequence current is subjected to a transformation process of expansion operation and corrosion operation under the action of the target structure operator for a target decomposition number, to obtain a transformed target current signal, wherein the target scale coefficient is equal to the target decomposition number.

[0072] wherein the scale coefficient is the morphological decomposition number. The target structure operator can be a piecewise function composed of a straight line and a cosine. When an arc fault occurs, there is unstable combustion of the arc, which causes serious distortion of the current waveform, and during the transient process after the fault occurs, the zero sequence current waveform still has certain fluctuation distortion. In order to avoid misjudging the transient process after the fault as the fault occurrence time, the selected target structure operator should focus on extracting high-frequency information as much as possible, and screening out nonlinear high-frequency signals with non-stationary changes. The straight line structure in the target structure operator can retain the high-frequency part of the non-stationary signal, and the cosine structure can filter out signal abrupt edges, playing a role in smoothing the signal, so that the target structure operator combining the cosine and the straight line can more effectively distinguish the steepness and gentleness of different fault pulses under the condition of a certain decomposition number, and amplify the fault characteristics. For example, the target structure operator is defined as follows:

[0073]

[0074] wherein G n is the nth layer structure operator function, A p is the highest amplitude of the target structure operator, and x is the position variable of the target structure operator. m n is the division position of the piecewise function, i.e. L n is the length of the target structure operator.

[0075] wherein the length of the target structure operator is determined based on the target scale coefficient. Since the length of the structure operator has a great influence on the result of the morphological decomposition of the signal, the morphological filter constructed by using a small-scale structure operator can detect the edge details of a complex signal and retain its detailed information, but the decomposition effect is poor; while the morphological filter constructed by using a large-scale structure operator has good decomposition efficiency, but the decomposed signal is relatively rough and may blur the details, resulting in waveform distortion. In view of this, according to the corresponding relationship between the length of the structure operator and the wavelet scale coefficient constructed by pre-simulation, the length of the target structure operator corresponding to the target scale coefficient is determined, so as to obtain a target structure operator with appropriate scale. For example, Table 1 shows the corresponding relationship between the length of the structure operator and the wavelet scale coefficient, based on which it can be determined that the length of the target structure operator is 15 to achieve a relatively ideal decomposition effect.

[0076] Table 1 Corresponding relationship between structure operator length and wavelet scale coefficient, frequency domain

[0077] Frequency band / kHz Wavelet scale coefficient Structure operator length >625 d1~d2 1～8 312～625 3 8～16 156～312 4 16～32 78～156 5 32～64 39～78 6 64～128

[0078] Specifically, the target scale coefficient epsilon is determined as a target decomposition order, and the to-be-detected zero sequence current f is subjected to a transformation process of epsilon times of dilation operation and erosion operation under the action of the target structure operator g, to obtain a target current signal after transformation under the target scale coefficient epsilon, that is:

[0079]

[0080]

[0081] Wherein, the to-be-detected zero sequence current signal f is subjected to one time of dilation operation under the action of the target structure operator g, and is expressed as: The to-be-detected zero sequence current signal f is subjected to one time of erosion operation under the action of the target structure operator g, and is expressed as: (fΘg)(x) = min{f(x+y)-g(y)|x+y∈D f ,y∈D g}, D f and D g are variable definition domains of the to-be-detected zero sequence current signal f and the target structure operator g respectively. The dilation operation is to strengthen the positive pulse of the signal and filter out the negative pulse, that is, to widen the wave head; the erosion operation is to strengthen the negative pulse of the signal and reduce the positive pulse, which is equivalent to widening the valley.

[0082] S270, estimating the fractal dimension under different time scales of the target current signal to determine a fractal dimension vector.

[0083] S280, performing high-resistance ground fault detection based on the mixed Gaussian model and the fractal dimension vector to determine a fault detection result corresponding to the to-be-detected line.

[0084] The technical scheme of the embodiment determines the time-frequency information region when the correlation degree between the to-be-detected line and each other line is greater than a preset correlation degree based on the cross wavelet power spectral density, determines the optimal target scale factor based on the number of overlapping regions between the time-frequency information regions, determines the scale coefficient corresponding to the to-be-detected zero sequence current based on the target scale factor, and performs integer processing on the scale coefficient to obtain an integer-processed target scale coefficient, so as to realize adaptive selection of the optimal scale coefficient. Through the transformation process of the to-be-detected zero sequence current under the action of the target structure operator under the target scale coefficient, the dilation operation and the erosion operation of the target decomposition order are obtained, and the target current signal after transformation is obtained, so as to realize morphological transformation under the optimal scale coefficient, further improve the decomposition effect, and further enhance the fault feature.

[0085] Figure 3This is a flowchart of another method for detecting high-resistance grounding faults in a distribution network according to an embodiment of the present invention. Based on the above embodiments, this embodiment optimizes the step of "estimating the fractal dimension of the target current signal at different time scales and determining the fractal dimension vector". Explanations of terms that are the same as or corresponding to those in the above embodiments are not repeated here.

[0086] See Figure 3 The other method for detecting high-resistance grounding faults in distribution networks provided in this embodiment specifically includes the following steps:

[0087] S310. Based on the zero-sequence current of each line in the distribution network, determine the line to be tested from all lines.

[0088] S320. Perform cross-wavelet coherence analysis on the zero-sequence current corresponding to the line under test and the other zero-sequence currents corresponding to each other line except the line under test to obtain the cross-wavelet power spectral density between the line under test and each other line.

[0089] S330. Based on the cross-wavelet power spectral density, determine the target scale coefficient corresponding to the zero-sequence current to be detected.

[0090] S340. Based on the target structure operator, a mathematical morphological transformation is performed on the zero-sequence current to be detected under the target scale coefficient to obtain the transformed target current signal.

[0091] S350: Perform time-domain processing on the target current signal at different time scales to obtain the reference current signal at each time scale.

[0092] Specifically, to obtain complementary and rich fault information at different time scales, multi-time-scale operations are performed on the morphologically transformed target current signal to calculate the fractal dimension at different time scales. This can be achieved by downsampling and smoothing the target current signal to obtain reference current signals at different time scales. For example, given a target current signal x = {x1, x2, ..., x...} N}, where x i Here, is the value of timestamp i, and N is the time step length of the signal. A reference current signal {y(τ)} at time scale τ is constructed by averaging the data points in the target current signal x using a non-overlapping window of length τ. Each element in the reference current signal {y(τ)} can be calculated using the following formula: in, By changing the time scale τ, the reference current signal {y(τ)} at different time scales τ can be obtained.

[0093] S360, estimate the fractal dimension of the target current signal to obtain the fractal dimension corresponding to the target current signal, and estimate the fractal dimension of each reference current signal at each time scale to obtain the fractal dimension corresponding to each reference current signal.

[0094] Specifically, the reference current signal at the time scale τ is 1, that is, the target current signal. The fractal dimension estimation process of the target current signal and each reference current is the same. For example, the fractal dimension of the target current signal and each reference current can be estimated by using the box counting method in the fractal dimension estimation method. The box counting method is measured by calculating the minimum number of boxes covering the image surface. The fractal dimension is estimated by calculating the minimum number of boxes covering the waveform pattern, and the signal time domain feature is extracted by using the differential box dimension.

[0095] Exemplarily, the "estimating the fractal dimension of the target current signal to obtain the fractal dimension corresponding to the target current signal" in step S360 can include: converting the target current signal into a two-dimensional grid, the horizontal coordinate of the two-dimensional grid being time and the vertical coordinate being gray value; superimposing a square small box with a preset length on each grid in the two-dimensional grid to determine the total number of boxes covering the entire two-dimensional grid, wherein the number of preset lengths is multiple; fitting based on the multiple preset lengths and the total number of boxes corresponding to each preset length to determine the fractal dimension corresponding to the target current signal.

[0096] Specifically, the target current signal with a length of N can be divided into n components, and the gray value of the signal at time x is f(x), so that the target current signal can be regarded as a two-dimensional grid (i.e. a two-dimensional object), the horizontal coordinate being time and the vertical coordinate being gray value. The target current signal is composed of n grids. A square small box with a preset length r is superimposed on each grid. If in the ith grid, the mth box contains the minimum gray value in the grid, and the lth box contains the maximum gray value in the grid, then the number of boxes n(i) covering the ith grid is: n(i) = l-m+1. The number of boxes covering each grid is added to obtain the total number of boxes N covering the entire two-dimensional grid. r , that is The fractal dimension corresponding to the preset length r The fractal dimension can be obtained by data fitting. For example, by least squares fitting of the logarithm of the total number of boxes and the logarithm of the preset length (i.e. {log(1 / r), log(N r )}, the slope of the fitted straight line is the fractal dimension D M . For example, see Figure 4by changing the preset length, the total number of boxes corresponding to each preset length can be calculated, and the least square fitting is performed on the logarithm of the total number of boxes and the logarithm of the preset length (i.e., {log(1 / r), log(N r The slope of the straight line in the double logarithmic graph is determined as the fractal dimension corresponding to the target current signal.

[0097] Similarly, the "estimating the fractal dimension of the reference current signal at each time scale to obtain the fractal dimension corresponding to each reference current signal" in step S360 can include: for the reference current signal at each time scale, converting the reference current signal into a two-dimensional grid, the horizontal coordinate of the two-dimensional grid being time and the vertical coordinate being gray value; superimposing a square small box with a preset length on each grid in the two-dimensional grid to determine the total number of boxes covering the entire two-dimensional grid; and performing fitting based on the plurality of preset lengths and the total number of boxes corresponding to each preset length to determine the fractal dimension corresponding to the reference current signal.

[0098] S370, combining the fractal dimension corresponding to the target current signal and the fractal dimensions corresponding to all reference current signals to obtain a fractal dimension vector.

[0099] Specifically, the combination result of the fractal dimension corresponding to the target current signal and the fractal dimensions corresponding to all reference current signals is taken as the fractal dimension vector. For example, the fractal dimension vector is: wherein sFD0 is the fractal dimension corresponding to the target current signal, i.e., the fractal dimension of the original signal without downsampling and smoothing, sFD τ is the fractal dimension corresponding to the reference current signal with the time scale τ.

[0100] S380, performing high-resistance ground fault detection based on the Gaussian mixture model and the fractal dimension vector to determine the fault detection result corresponding to the to-be-detected line.

[0101] The technical scheme of the embodiment can obtain the reference current signal at each time scale by performing time domain processing on the target current signal at different time scales, estimate the fractal dimension of the target current signal to obtain the fractal dimension corresponding to the target current signal, estimate the fractal dimension of the reference current signal at each time scale to obtain the fractal dimension corresponding to each reference current signal, and combine the fractal dimension corresponding to the target current signal and the fractal dimensions corresponding to all reference current signals, so that a more accurate fractal dimension vector can be obtained, and the accuracy of fault detection is further improved.

[0102] On the basis of the above technical schemes, the estimation of the fractal dimension of the target current signal at different time scales to determine the fractal dimension vector can include:

[0103] Sliding the sliding window on the target current signal to obtain a target current sub-signal in a current sliding window; and estimating the fractal dimension of the target current sub-signal under different time scales to determine a fractal dimension vector corresponding to the current sliding window.

[0104] The target current sub-signal refers to a signal in the target current signal within the current sliding window. Specifically, the target current signal is detected by using the sliding window, so that the target current signal can be divided into shorter time windows, and the fractal dimension of the target current sub-signal in the time window is estimated under different time scales to obtain the fractal dimension vector corresponding to the current sliding window (for details of the estimation process, refer to steps S360 and S370 described above, which will not be repeated here). Based on the mixed Gaussian model and the fractal dimension vector corresponding to the current sliding window, the high-resistance grounding fault detection is performed to determine the fault detection result of the line to be detected in the current time window, so that it can be adaptively determined whether there is a high-resistance grounding fault in each time window, meeting the real-time detection requirement, and further the high-resistance grounding fault can be found in time.

[0105] The following is an embodiment of a power distribution network high-resistance grounding fault detection device provided by the embodiment of the application. The device and the power distribution network high-resistance grounding fault detection method of each embodiment described above belong to the same inventive concept. Details not described in the embodiment of the power distribution network high-resistance grounding fault detection device can be referred to the embodiment of the power distribution network high-resistance grounding fault detection method described above.

[0106] Figure 5 A structure diagram of a power distribution network high-resistance grounding fault detection device provided by the embodiment of the application. The embodiment can be applied to detect whether a high-resistance grounding fault occurs in a line in a power distribution network. As shown in the figure, the device specifically includes: a line to be detected determination module 510, a cross wavelet coherence analysis module 520, a target scale coefficient determination module 530, a mathematical morphological transformation module 540, a fractal dimension estimation module 550, and a high-resistance grounding fault detection module 560. Figure 5

[0107] ​The zero-sequence current of each line in the power distribution network is determined as the to-be-detected line, cross wavelet coherence analysis is performed on the to-be-detected zero-sequence current corresponding to the to-be-detected line and other zero-sequence currents corresponding to each other line except the to-be-detected line, cross wavelet power spectrum densities between the to-be-detected line and each other line are obtained, a target scale coefficient corresponding to the to-be-detected zero-sequence current is determined based on the cross wavelet power spectrum densities, mathematical morphological transformation is performed on the to-be-detected zero-sequence current at the target scale coefficient based on a target structure operator, a transformed target current signal is obtained, fractal dimension estimation is performed on the target current signal at different time scales, a fractal dimension vector is determined, high-resistance grounding fault detection is performed based on a mixture Gaussian model and the fractal dimension vector, and a fault detection result corresponding to the to-be-detected line is determined.

[0108] The technical scheme of the embodiment determines the to-be-detected line based on the zero-sequence current of each line in the power distribution network, performs cross wavelet coherence analysis on the to-be-detected zero-sequence current corresponding to the to-be-detected line and other zero-sequence currents corresponding to each other line, obtains cross wavelet power spectrum densities between the to-be-detected line and each other line, determines a best target scale coefficient of the to-be-detected zero-sequence current based on the cross wavelet power spectrum densities, performs mathematical morphological transformation on the to-be-detected zero-sequence current at the best target scale coefficient based on a target structure operator, and obtains a transformed target current signal, thereby filtering out noise and interference signals and enhancing a weak response of a high-resistance grounding fault at a moment of occurrence, estimates fractal dimensions of the target current signal at different time scales, determines a fractal dimension vector at different time scales, performs high-resistance grounding fault detection based on a mixture Gaussian model and the fractal dimension vector, and determines a fault detection result corresponding to the to-be-detected line, thereby improving the accuracy of high-resistance grounding fault detection in the power distribution network.

[0109] Optionally, the to-be-detected line determination module 510 is specifically configured to:

[0110] Based on the zero-sequence current of each line in the power distribution network, the energy of each zero-sequence current is determined, and a line with the largest energy of the zero-sequence current is determined as the to-be-detected line.

[0111] Optionally, the target scale coefficient determination module 530 comprises:

[0112] The time-frequency information region determination unit is configured to determine a time-frequency information region in which a correlation degree between the to-be-detected line and each other line is greater than a preset correlation degree based on the cross wavelet power spectrum density.

[0113] The target scale factor determination unit is configured to determine an overlapping region between all the time-frequency information regions, and determine a target scale factor based on a number of the overlapping region.

[0114] The target scale coefficient determination unit is configured to determine a scale coefficient corresponding to the to-be-detected zero sequence current based on the target scale factor, and perform an integer processing on the scale coefficient to obtain an integer-processed target scale coefficient.

[0115] Optionally, the target scale factor determination unit is specifically configured to:

[0116] If there is only one overlapping region, a target time-frequency information in which the cross wavelet power spectrum density is maximum in the overlapping region is determined, and a scale factor in the target time-frequency information is determined as the target scale factor; if there are at least two unconnected overlapping regions, to-be-selected time-frequency information in which the cross wavelet power spectrum density is maximum in each overlapping region is determined, and a minimum scale factor in all the to-be-selected time-frequency information is determined as the target scale factor.

[0117] Optionally, the target structure operator is a piecewise function composed of a cosine and a straight line, and a length of the target structure operator is determined based on the target scale coefficient.

[0118] The mathematical morphological transformation module 540 is specifically configured to perform a transformation processing of an expansion operation and an erosion operation of a target decomposition number on the to-be-detected zero sequence current under the target structure operator to obtain a transformed target current signal, where the target scale coefficient is equal to the target decomposition number.

[0119] Optionally, the fractal dimension estimation module 550 comprises:

[0120] The time-frequency processing unit is configured to perform a time domain processing of different time scales on the target current signal to obtain a reference current signal under each time scale.

[0121] The fractal dimension estimation unit is configured to estimate a fractal dimension of the target current signal to obtain a fractal dimension corresponding to the target current signal, and estimate a fractal dimension of each reference current signal under each time scale to obtain a fractal dimension corresponding to each reference current signal.

[0122] The fractal dimension combination unit is configured to combine the fractal dimension corresponding to the target current signal and the fractal dimensions corresponding to all the reference current signals to obtain a fractal dimension vector.

[0123] Optionally, the fractal dimension estimation unit is specifically used for:

[0124] The target current signal is converted into a two-dimensional grid, the horizontal coordinate of the two-dimensional grid is time, and the vertical coordinate is a gray value; a square small box with a preset length is superimposed on each grid in the two-dimensional grid, and the total number of boxes covering the entire two-dimensional grid is determined, wherein the number of the preset length is a plurality; based on the plurality of preset lengths and the total number of boxes corresponding to each preset length, fitting is performed to determine the fractal dimension corresponding to the target current signal.

[0125] Optionally, the high-resistance ground fault detection module 560 is specifically used for:

[0126] The fractal dimension vector is input into a mixed Gaussian model for high-resistance ground fault detection, the mixed Gaussian model being a normal Gaussian sub-model and a fault Gaussian sub-model obtained by clustering sample data based on a maximum expectation algorithm in advance; based on the output of the mixed Gaussian model, a fault detection result corresponding to the to-be-detected line is determined.

[0127] Optionally, the fractal dimension estimation module 550 is specifically used for:

[0128] A sliding window is slid on the target current signal to obtain a target current sub-signal in a current sliding window; fractal dimension estimation of the target current sub-signal under different time scales is performed to determine a fractal dimension vector corresponding to the current sliding window.

[0129] The power distribution network high-resistance ground fault detection device provided in the embodiments of the present application can perform the power distribution network high-resistance ground fault detection method provided in any of the embodiments of the present application, and has corresponding functional modules and beneficial effects for performing the power distribution network high-resistance ground fault detection method.

[0130] It should be noted that in the embodiments of the power distribution network high-resistance ground fault detection device described above, each unit and module included is only divided according to functional logic, but is not limited to the above division, as long as the corresponding functions can be implemented; in addition, the specific names of each functional unit are only for convenient mutual differentiation, and are not used to limit the protection scope of the present application.

[0131] Those skilled in the art should understand that the modules or steps of the present application described above can be realized by general computing devices, which can be centralized on a single computing device or distributed on a network composed of multiple computing devices, and optionally, they can be realized by program codes executable by the computing devices, so that they can be stored in storage devices and executed by the computing devices, or they can be respectively manufactured into individual integrated circuit modules, or multiple modules or steps among them can be manufactured into a single integrated circuit module to realize. Thus, the present application is not limited to any specific combination of hardware and software.

[0132] It is noted that the above merely describes the preferred embodiments of the present application and the principles of the applied technology. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and various obvious changes, re-adjustments and substitutions can be made by those skilled in the art without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and more other equivalent embodiments can be included without departing from the concept of the present application, and the scope of the present application is determined by the appended claims.

Claims

1. A method for detecting high impedance ground fault in a power distribution network, characterized in that, The method comprises the following steps: determining a to-be-detected line from all lines based on zero sequence currents of each line in a power distribution network; performing cross wavelet coherence analysis on a to-be-detected zero sequence current corresponding to the to-be-detected line and other zero sequence currents corresponding to each other line except the to-be-detected line, to obtain cross wavelet power spectrum densities between the to-be-detected line and each other line; determining a target scale coefficient corresponding to the to-be-detected zero sequence current based on the cross wavelet power spectrum densities; performing mathematical morphological transformation on the to-be-detected zero sequence current at the target scale coefficient based on a target structure operator, to obtain a transformed target current signal; performing estimation of fractal dimensions of the target current signal at different time scales, to determine a fractal dimension vector; performing high-resistance ground fault detection based on a Gaussian mixture model and the fractal dimension vector, to determine a fault detection result corresponding to the to-be-detected line.

2. The method of claim 1, wherein, The method of determining a to-be-detected line from all lines based on zero sequence currents of each line in a power distribution network comprises the following steps: determining an energy of each zero sequence current based on zero sequence currents of each line in a power distribution network; determining a line with the largest energy of zero sequence current as the to-be-detected line.

3. The method of claim 1, wherein, The method of determining a target scale coefficient corresponding to the to-be-detected zero sequence current based on the cross wavelet power spectrum densities comprises the following steps: determining a time-frequency information region when a correlation degree between the to-be-detected line and each other line is greater than a preset correlation degree based on the cross wavelet power spectrum densities; determining an overlapping region between all time-frequency information regions, and determining a target scale factor based on a number of the overlapping regions; determining a scale coefficient corresponding to the to-be-detected zero sequence current based on the target scale factor, and performing rounding processing on the scale coefficient, to obtain a rounded target scale coefficient.

4. The method of claim 3, wherein, The method of determining a target scale factor based on the number of overlapping regions comprises the following steps: if there is only one overlapping region, determining a target time-frequency information when a cross wavelet power spectrum density in the overlapping region is the largest, and determining a scale factor in the target time-frequency information as the target scale factor; if there are at least two unconnected overlapping regions, determining to-be-selected time-frequency information when a cross wavelet power spectrum density in each overlapping region is the largest, and determining a smallest scale factor in all to-be-selected time-frequency information as the target scale factor.

5. The method of claim 1, wherein, The target structure operator is a piecewise function composed of a cosine and a straight line, and a length of the target structure operator is determined based on the target scale coefficient. The method of performing mathematical morphological transformation on the to-be-detected zero sequence current at the target scale coefficient based on the target structure operator, to obtain a transformed target current signal, comprises the following steps: performing transformation processing of dilation operation and corrosion operation of a target decomposition number on the to-be-detected zero sequence current under the target structure operator at the target scale coefficient, to obtain the transformed target current signal, wherein the target scale coefficient is equal to the target decomposition number.

6. The method of claim 1, wherein, The method of performing estimation of fractal dimensions of the target current signal at different time scales, to determine a fractal dimension vector, comprises the following steps: The target current signal is subjected to time-domain processing of different time scales to obtain a reference current signal under each time scale; The target current signal is subjected to fractal dimension estimation to obtain a fractal dimension corresponding to the target current signal, and each reference current signal is subjected to fractal dimension estimation to obtain a fractal dimension corresponding to each reference current signal; The fractal dimensions corresponding to the target current signal and all reference current signals are combined to obtain a fractal dimension vector.

7. The method of claim 6, wherein, The target current signal is subjected to fractal dimension estimation to obtain a fractal dimension corresponding to the target current signal, including: The target current signal is converted into a two-dimensional grid, with time as the horizontal coordinate and gray value as the vertical coordinate; A square box with a preset length is superimposed on each grid in the two-dimensional grid to determine the total number of boxes covering the entire two-dimensional grid, wherein the number of preset lengths is multiple; Based on the multiple preset lengths and the total number of boxes corresponding to each preset length, the fractal dimension corresponding to the target current signal is determined.

8. The method of claim 1, wherein, The mixed Gaussian model and the fractal dimension vector are used for high-resistance ground fault detection to determine a fault detection result corresponding to the line to be detected, including: The fractal dimension vector is input into the mixed Gaussian model for high-resistance ground fault detection, and the mixed Gaussian model is a normal Gaussian sub-model and a fault Gaussian sub-model obtained by clustering sample data based on the maximum expectation algorithm in advance; Based on the output of the mixed Gaussian model, a fault detection result corresponding to the line to be detected is determined.

9. The method of claim 1, wherein, The target current signal is subjected to fractal dimension estimation of different time scales to determine a fractal dimension vector, including: A sliding window is slid on the target current signal to obtain a target current sub-signal in the current sliding window; The target current sub-signal is subjected to fractal dimension estimation of different time scales to determine a fractal dimension vector corresponding to the current sliding window.

10. A device for detecting high impedance ground faults in a power distribution network, characterized by, Including: A line to be detected line determination module is configured to determine a line to be detected from all lines based on the zero sequence current of each line in a power distribution network; A cross wavelet coherence analysis module is configured to perform cross wavelet coherence analysis on a to-be-detected zero sequence current corresponding to the line to be detected and other zero sequence currents corresponding to each other line except the line to be detected to obtain a cross wavelet power spectral density between the line to be detected and each other line; A target scale coefficient determination module is configured to determine a target scale coefficient corresponding to the to-be-detected zero sequence current based on the cross wavelet power spectral density; A mathematical morphology transformation module is configured to perform mathematical morphology transformation on the to-be-detected zero sequence current at the target scale coefficient based on a target structure operator to obtain a transformed target current signal; A fractal dimension estimation module is configured to perform fractal dimension estimation of different time scales on the target current signal to determine a fractal dimension vector. The high-resistance ground fault detection module is configured to perform high-resistance ground fault detection based on a Gaussian mixture model and the fractal dimension vector, and determine a fault detection result corresponding to the line to be detected.

Images