A power distribution network early fault recording starting feature selection method and a recording starting judgment method

Abstract

The application provides a power distribution network early fault recording starting feature selection method and a recording starting judgment method, and belongs to the technical field of power distribution network early fault identification. The feature selection method comprises the following steps: performing multi-dimensional analysis on the power distribution network early fault recording data, constructing a feature set, and extracting waveform morphology and time-frequency domain features; calculating fault feature benefit cost, considering the calculation complexity and fault saliency; sorting in descending order of benefit cost, and selecting the first M features as the recording starting features. The recording starting judgment method comprises the following steps: collecting a zero sequence current signal, transforming the signal by using an improved variation mode decomposition method, and extracting low-frequency components; selecting transient energy and phase plane trajectory features; extracting the transient energy and phase plane trajectory of the transformed signal based on a sliding time window, and calculating a mutation variable; fusing the Laplace distribution into a fault probability index, and setting a double threshold to trigger the recording. The recording starting method constructed by the application has higher sensitivity and adaptability, and can quickly respond to fault precursors.

Description

TECHNICAL FIELD

[0001] The present application relates to a method for selecting a recording wave starting feature of an early fault of a power distribution network and a method for judging a recording wave starting, and belongs to the technical field of early fault identification of a power distribution network. BACKGROUND

[0002] The power distribution network is an important part of the power system, and its operation reliability directly affects the safety production and daily power consumption of users. Improving the fault detection and disposal capability of the power distribution network not only helps to ensure the continuity of power supply and shorten the power outage time, but also improves the user's power consumption experience. With the increasing requirements for power supply safety and stability, the fault handling mode of the power distribution network is gradually changing from "post-diagnosis" to "pre-warning", which puts higher requirements on fault precursor identification and data acquisition.

[0003] With the continuous advancement of the construction of the smart power distribution network, a large number of high-precision measurement devices are deployed in the power distribution system, providing rich data support for fault monitoring and state evaluation. However, compared with the sudden and energy strong faults in the transmission system, the early faults of the equipment in the power distribution network (such as pre-discharge of insulator pollution flashover, cable terminal creeping, tree line fault, etc.) usually exhibit electrical signals with short duration, weak amplitude, and non-periodic intermittent appearance. Such signals are often overwhelmed by operational noise and difficult to be accurately identified by the starting criterion of traditional recording devices.

[0004] The existing recording wave starting criterion is mostly based on current over-limit, voltage drop, frequency deviation and other significant mutation features, mainly aiming at typical fault events such as short circuit and grounding. Such criterion has the following problems when facing early faults:

[0005] (1) The early faults of the power distribution network may go through different discharge stages in the process of continuous development, showing different fault waveforms. The existing fault recording wave starting criterion that relies on a single feature quantity generally has the problems of insufficient sensitivity and poor applicability. It is still a technical problem to be solved to extract and comprehensively utilize fault features from time domain, frequency domain and time-frequency domain, etc. to improve the identification ability of early faults.

[0006] (2) The increase in the number of features will increase the requirements for the accuracy of the measurement device and the operation capacity of the regional master station. At the same time, the sensitivity of different features to faults is different, and some features may not be significantly responsive to early faults, resulting in a decrease in diagnostic efficiency. Therefore, it is necessary to select features with high sensitivity to the change rule of different types of early faults, low computational complexity and easy real-time processing from numerous features to ensure the rapidity of fault recording wave starting.

[0007] (3) Early fault current amplitude is small, easy to be submerged in system noise, in order to extract effective and reliable fault features, it is necessary to develop high-precision signal processing method; In addition, early fault has a certain development process, and the simple threshold-based recording wave starting method is difficult to identify the fault signal in the early stage of hidden danger, and a reasonable fault quantization method should be proposed to evaluate the fault degree and to design a sensitive fault recording starting criterion.

[0008] Therefore, it is urgent to face the early fault features of the vulnerable equipment of the distribution network, to build a recording wave starting method with higher sensitivity and adaptability, so as to realize the rapid response to the fault precursor. SUMMARY

[0009] The purpose of the present application is to provide a recording wave starting feature selection method and a recording wave starting judgment method for early fault of distribution network, which can capture and record key data in the early stage, provide reliable data basis for subsequent fault mechanism research, feature extraction and intelligent diagnosis model training, thereby significantly improving the ability of distribution network in fault early warning, state evaluation and active operation and maintenance.

[0010] In order to achieve the above purpose, the present application realizes the following technical scheme:

[0011] A recording wave starting feature selection method for early fault of distribution network, comprising the following steps:

[0012] Performing multi-dimensional fault feature analysis on the recording wave data of early fault of distribution network to construct a multi-dimensional fault feature set, and extracting waveform morphology features and time-frequency domain features;

[0013] Calculating the benefit cost of fault features based on the calculation complexity and fault saliency;

[0014] According to the benefit cost, the fault features are sorted in descending order, and the first M fault features are selected as the recording wave starting features of early fault of distribution network.

[0015] Preferably, the multi-dimensional fault feature set includes waveform morphology features and time-frequency features;

[0016] The waveform features include kurtosis, skewness, waveform factor, peak factor, pulse factor, margin factor and phase plane trajectory;

[0017] The time-frequency domain features include transient energy, total harmonic distortion rate, transient center frequency, power spectrum entropy, wavelet singular entropy and frequency band energy entropy.

[0018] Preferably, the benefit cost The calculation formula is as follows:

[0019]

[0020] Wherein, the number of data sampling points in a single cycle, a fault saliency index, a computational complexity index.

[0021] Preferably, the computational complexity is expressed in big O notation, based on the number of data sampling points in a single cycle, to analyze the number of multiplication and addition operations, the number of floating point operations or the number of iterations required in each feature extraction process, to obtain the computational complexity index; and the fault features with a computational complexity index to number of data sampling points ratio less than 10 are screened out.

[0022] the fault saliency The mutation index is quantified by the following formula:

[0023] ,

[0024] wherein, represents the feature value calculated using half-cycle data after the occurrence of the i th fault; represents the feature value calculated using half-cycle data after the occurrence of the i th fault; is the average of all feature values calculated using a half-cycle sliding time window within the ten cycles before the fault occurrence time; is the standard deviation of all feature values calculated using a half-cycle sliding time window within the ten cycles before the fault occurrence time.

[0025] A recording wave starting judgment method for early faults in a power distribution network, comprising:

[0026] Collecting a zero sequence current signal, transforming the zero sequence current signal based on an improved variational mode decomposition method, and extracting low-frequency components in the zero sequence current signal;

[0027] Selecting early fault identification fault features, including transient energy features and phase plane trajectory features;

[0028] Extracting transient energy and phase plane trajectory of the transformed zero sequence current signal based on a sliding time window, and calculating transient energy mutation and phase plane trajectory mutation;

[0029] Fusing the two kinds of mutation into a fault probability index using Laplace distribution;

[0030] Setting a double threshold value to trigger a recording wave starting condition according to the fault probability index, and starting recording wave when the zero sequence current signal meets the double threshold value to trigger the recording wave starting condition.

[0031] Preferably, the early fault identification fault features are selected by the recording wave starting feature selection method for early faults in a power distribution network.

[0032] Preferably, the improved variational mode decomposition method for transforming the zero sequence current signal comprises the following steps:

[0033] Distinguish the fault current signal in the zero-sequence current signal, wherein the fault current signal is an aliasing of multiple frequency signals;

[0034] The fault current signal is subjected to Hilbert transform and time delay processing to obtain two preprocessed signals. , :

[0035] ,

[0036] in, The fault current after Hilbert transformation. This is the fault current after time delay processing. For fault current Second harmonic angular frequency Sampling time, For fault current Second harmonic phase , This is the maximum frequency of harmonics that the device can collect;

[0037] The sum of the preprocessed signals is decomposed using VMD to extract the low-frequency components of the zero-sequence current signal, including the fundamental frequency and the third harmonic component. These components are then summed as the data source for subsequent fault feature extraction.

[0038] Preferably, the transient energy calculation formula is as follows:

[0039] ,

[0040] in, Transient energy, This refers to the number of data sampling points per cycle. This is the low-frequency component of the zero-sequence current. For sampling points;

[0041] The phase plane trajectory is calculated as follows:

[0042] The transformed zero-sequence current signal is reconstructed into a phasor sequence by performing phase space reconstruction:

[0043] ,

[0044] in, ; express indivual The zero-sequence current phase plane trajectory formed by points in a certain dimension. Indicates the first Zero-sequence current data in each dimension To delay time, For the embedding dimension, sampling points of low frequency component of zero sequence current;

[0045] Adopting Euclidean distance to quantify the degree of the curve deviating from the origin as the characteristic of the phase plane trajectory

[0046]

[0047] wherein, and are respectively two-dimensional zero sequence current data used for constituting the phase plane trajectory, the lengths are respectively , and hysteresis one quarter of a cycle.

[0048] Preferably, the fault probability index The calculation formula is:

[0049]

[0050] wherein, is a transient energy mutation quantity, which is obtained by subtracting the transient energy in the last time window from the transient energy in the current time window, is a phase plane trajectory mutation quantity, which is obtained by subtracting the phase plane trajectory in the last time window from the phase plane trajectory in the current time window, is the mean square error of the transient energy mutation quantity in the ten cycles before the fault occurrence moment, is the mean square error of the phase trajectory mutation quantity in the ten cycles before the fault occurrence moment, is the average value of and , is the position parameter of the Laplace distribution function, and are the independent variables of the Laplace distribution function, which are respectively substituted into the time The transient energy mutation quantity and the phase plane trajectory mutation quantity are calculated.

[0051] Preferably, the double-threshold triggering condition is:

[0052] When the number of times that the fault probability is in the double-threshold range within a set time period is greater than a set threshold value, the fault recording is started;

[0053] When the fault probability exceeds the large threshold value in the double threshold value, the fault recording is triggered immediately.

[0054] The present application has the following advantages:

[0055] ​​​​The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0056] The present application proposes a recording starting judgment method for early fault of distribution network, which can effectively extract low-frequency components of zero sequence current, further calculate transient energy and phase plane trajectory features, and use Laplace distribution to combine two different fault features, quantify feature changes before and after fault occurrence through probabilistic index, compared with the problem of insufficient applicability of traditional threshold to early fault, the starting criterion based on fault probability can respond to early fault of distribution network in weak discharge stage. BRIEF DESCRIPTION OF DRAWINGS

[0057] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation of the present application.

[0058] Figure 1 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0059] Figure 2 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0060] Figure 3 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0061] Figure 4 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0062] Figure 5 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0063] Figure 6 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0064] Figure 7 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0065] Figure 8 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0066] Figure 9 The present application is aimed at early fault events of distribution network, based on field recording data, analyzes waveform characteristics and time-frequency energy changes, and selects multi-dimensional fault features from waveform and time-frequency domain, proposes a feature optimization method through calculation complexity and significance index evaluation, and finally selects transient energy and phase plane trajectory as fault features of early fault recording starting criterion, which can better meet the real-time and sensitivity requirements of measurement device starting.

[0067] Figure 10 The schematic diagram for evaluating the fault feature benefit of the application.

[0068] Figure 11 The schematic diagram for calculating the early probability of the distribution network of the application.

[0069] Figure 12 The schematic diagram for the early fault recording waveform of the insulator of the application.

[0070] Figure 13 The schematic diagram for quantifying the probability index based on the Laplace distribution of the application. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0072] Embodiment 1

[0073] This embodiment proposes a recording wave starting feature selection method for early faults of a distribution network, as shown in FIG. 1, including the following steps: Figure 1

[0074] S1: Multi-dimensional fault feature analysis is performed on the early fault recording data of the distribution network to construct a multi-dimensional fault feature set, and waveform shape features and time-frequency domain features are extracted;

[0075] S2: The fault feature benefit cost is calculated based on the calculation complexity and the fault saliency;

[0076] S3: The fault features are sorted in descending order according to the benefit cost, and the first M fault features are selected as the recording wave starting features for early faults of the distribution network.

[0077] As a refinement of the above-described embodiments, the step S1 includes waveform feature analysis and time-frequency feature analysis, and specifically includes:

[0078] S101: Early fault waveform feature analysis based on recording wave data

[0079] As shown in FIG. 2, it is the typical early fault data collected on site, Figures 3-4 Figure 3 Figure 4 It is the crack weak discharge and flashover phenomenon caused by the deterioration and water immersion of the insulator, Figure 5 ​​Tree-line discharge caused by tree contact of overhead line in distribution network. Due to different causes of early fault, the fault waveform also exists differences. For early fault of insulator, it is usually because the insulator performance is reduced, and the high resistance area appears in some part of the insulator, and the local current is generated under the external applied voltage, and the instantaneous flashover is formed when the breakdown voltage is reached. While the tree-line fault generally forms a stable fault connection point at the beginning of the fault, but the fault resistance is as high as tens of kilo-ohms, and the fault current is submerged in the measurement noise. Under the action of continuous arc burning, the arc will gradually carbonize the tree branch contact point, forming a low resistance channel, and the current shows a slow rising trend.

[0080] For different types of early faults, the duration or amplitude of the initial stage of the fault often does not meet the action conditions of the relay protection, and the equipment with hidden faults will not be cut off. Although it does not affect normal power supply, it has the risk of fault, and is easy to develop into destructive faults such as single-phase grounding and inter-phase short circuit.

[0081] S102: Time-frequency feature analysis of early fault based on recorded wave data

[0082] When the distribution system is in normal operation, the zero sequence current is basically measurement noise or unbalanced current; and when the distribution network has hidden faults, the current waveform will produce obvious waveform distortion, including current peak and zero-crossing flat shoulder.

[0083] In addition, the panoramic information on the time-frequency domain can provide more sufficient fault information analysis. By using continuous wavelet transform processing Figures 3-4 The fault time-frequency domain waveform shown in Figures 6-7 The analysis results show that when the insulator flashes, there is a clear energy step in the low frequency band and the high frequency band, and the high frequency band energy is mainly caused by unstable arc. In theory, the mutation of high frequency energy compared with low frequency energy is more obvious, but in the early stage of hidden discharge, the current amplitude is too weak, and the high frequency energy is submerged in the noise. The tree-line fault is usually only in the low frequency energy because of the stable and continuous development.

[0084] S103: Early fault may be in different discharge stages during the development process, and the characteristics of the early weak discharge stage are difficult to extract, so the method based on a single fault feature has limitations. Therefore, features should be extracted from both waveform shape and time-frequency distribution. The common fault features in the field of distribution network fault diagnosis include waveform shape features and time-frequency features.

[0085] The waveform features include: kurtosis, skewness, waveform factor, peak factor, pulse factor, margin factor, and phase plane trajectory.

[0086] The time-frequency features include: transient energy, total harmonic distortion rate, transient center frequency, power spectrum entropy, wavelet singular entropy, and frequency band energy entropy.

[0087] As a refinement of the above embodiment, the step S2 comprises:

[0088] S201: Fault feature evaluation index based on computational complexity and saliency

[0089] (1) Computational complexity

[0090] Considering that the recording wave starting criterion requires features to be sensitive and have low computational complexity for fast response to the occurrence of weak early faults in distribution networks, it is necessary to evaluate the computational complexity of each feature first.

[0091] In theoretical computer science, computational complexity refers to the time required to execute an algorithm in a computer. It can be quantified by theoretical estimation: analyzing the calculation process of each feature. The number of multiplication and addition operations, floating point operations, or iterations required is counted. Finally, a comparable index is obtained, and the larger the value, the more complex it is. The computational complexity index is denoted as (corresponding to the th fault feature).

[0092] (2) Fault saliency

[0093] In addition, the more fault features extracted, the higher the precision of the measuring device and the computational capacity of the regional master station required, and the sensitivity of different fault features to faults also varies. This indicates that it is necessary to select optimal fault features to design a fault recording wave starting criterion.

[0094] In order to select features that change significantly before and after the occurrence of early faults, data-driven methods are needed to extract features before and after each fault occurrence using a large amount of simulation data and actual recording wave data. For each feature, the fault saliency index is evaluated using the mutation index:

[0095] ,

[0096] wherein represents the feature value calculated using half-cycle data after the th fault occurrence; is the average of all feature values calculated using a half-cycle sliding time window within the first ten cycles before the fault occurrence time; is the standard deviation of all feature values calculated using a half-cycle sliding time window within the first ten cycles before the fault occurrence time.

[0097] By calculating the mutation index, the extreme degree of feature values exceeding the normal fluctuation range can be quantified, the larger the value, the stronger the ability of the feature to distinguish between faults and normal states, i.e., the higher the fault saliency.

[0098] S202: Multidimensional fault feature ranking and screening method

[0099] On the basis of successfully quantifying the calculation complexity index and fault significance index of each fault feature quantity, to realize the fast and reliable perception of early weak faults of distribution networks, the core of the application is to provide a recording wave starting feature selection method. The method aims to abandon the limitations of single index evaluation, and selects the optimal feature subset from the fault feature set by comprehensively considering the "cost performance" of the features, so as to lay a foundation for constructing an efficient and reliable early fault recording wave starting criterion. Specifically, it includes:

[0100] (1) A large number of fault data sets should be collected on site, including zero sequence current sampling data before and after the fault occurs, so as to extract features during normal operation and features after the fault occurs.

[0101] (2) According to the selected multidimensional fault feature set, the features are extracted based on the on-site fault data set.

[0102] The calculation complexity of different features is analyzed based on the big O notation, and the number of single-cycle data sampling points required for feature extraction is taken as the benchmark. The calculation complexity of all features is sorted from small to large, and the criterion is set to preliminarily screen, and the fault features with shorter calculation time consumption are retained;

[0103] According to the selected fault feature set, the significance index of each fault feature on each fault data is calculated, and the statistical analysis is performed and the median is taken as the significance of each feature;

[0104] Finally, the benefit-cost is proposed, the calculation complexity and the significance of the fault features are ranked and screened, the first two-dimensional fault features are retained, and the benefit-cost formula is:

[0105] ,

[0106] As shown in Table 1, the calculation complexity analysis and sorting results of each fault feature in Table 1 are shown in Table 1. The complexity is based on the big O notation, and the number of single-cycle data sampling points is taken as the benchmark, and is set to =1024, and is sorted from small to large.

[0107] Table 1 Calculation complexity analysis and sorting results of each fault feature

[0108]

[0109] According to the complexity sorting, the above multidimensional fault features can be divided into linear complexity group ​​​​​Log-linear complexity group And high order complexity Generally, the fault starting criterion is configured at the equipment side, and the real-time requirement is high, and the features of the group are preferentially selected according to the criteria .

[0110] The significance indicators of the features extracted from the early fault data on site are counted, and the results are shown in Figure 9 It can be seen that the transient energy and the phase plane fault feature are the most obvious in the sudden change after the fault occurs, and the order of magnitude is far more than other fault features.

[0111] As a refinement of the above embodiment, the step S3 specifically selects the first two fault features as the recording wave starting features of the early fault of the power distribution network. According to the calculation complexity and the evaluation results of the feature significance, the characteristic benefit ratio is calculated according to the benefit cost formula, and the calculation and sorting results are shown in Figure 10 .

[0112] It should be noted that: the present application is aimed at the early fault event of the power distribution network, based on the waveform characteristics and time-frequency energy changes of the field recording wave data, and the multi-dimensional fault features are selected from the waveform and time-frequency domain, a feature optimization method is proposed by calculating the complexity and the significance index evaluation, and finally the transient energy and the phase plane trajectory are selected as the fault features of the recording wave starting criterion of the early fault, which can better meet the real-time and sensitivity requirements of the measurement device starting.

[0113] Embodiment 2

[0114] As shown in Figure 2 , a recording wave starting judgment method for early fault of a power distribution network, comprising:

[0115] S1: collecting a zero sequence current signal, transforming the zero sequence current signal based on an improved variational mode decomposition method, and extracting low-frequency components in the zero sequence current signal;

[0116] S2: selecting early fault identification fault features, including transient energy features and phase plane trajectory features;

[0117] S3: extracting transient energy and phase plane trajectory of the transformed zero sequence current signal based on a sliding time window, and calculating transient energy mutation and phase plane trajectory mutation;

[0118] S4: fusing the two kinds of mutations into a fault probability index by using Laplace distribution;

[0119] S5: setting double thresholds according to the fault probability index to trigger the recording wave starting condition, and if the zero sequence current signal meets the double threshold recording wave starting condition, the recording wave is started.

[0120] As a refinement of the above embodiment, the step S1 specifically comprises, based on an improved variational mode decomposition method (HT-VMD):

[0121] In the field of signal processing, variational mode decomposition (VMD) can decompose the original signal into a plurality of intrinsic mode functions (IMF) with different center frequencies according to the optimal solution of the iterative search variational model. The VMD algorithm can decompose the fault current into noise and effective signals containing different frequencies, which is suitable for the scene of fault feature extraction in this paper.

[0122] However, the traditional VMD algorithm is not sensitive to the signal component with small amplitude, and there is still a problem of similar mode aliasing when the number of decomposition layers is not clear. In view of this problem, the VMD algorithm is improved, and a HT-VMD signal processing method is proposed combined with Hilbert transform (HT).

[0123] The zero sequence current signal transformation step provided by the embodiment is as follows:

[0124] S101: distinguish the fault current signal in the zero sequence current signal, and regard the fault current as aliasing of a plurality of frequency signals at a signal level.

[0125] ,

[0126] Among them, is the amplitude of the signal with different frequencies, is the subharmonic angular frequency of the fault current, is the subharmonic phase of the fault current, is the sampling time.

[0127] S102: Hilbert transform is performed on the fault current information, and the physical meaning is that the phase of all frequency components of the signal is delayed by 90°, and the above formula becomes:

[0128] ,

[0129] Among them, is the fault current after Hilbert transform.

[0130] S103: the fault current signal is delayed at the same time, as a time delay function:

[0131] ,

[0132] Among them,​ represents the power frequency cycle.

[0133] S104: For early faults, the waveform behaves like a high-resistance nonlinear fault signal, and according to traditional research, the even harmonic components are few. The pre-processing formula for the original signal is as follows:

[0134]

[0135] wherein, is the maximum frequency of the harmonics that can be collected by the device, the value is a series of odd numbers; are two signals obtained after pre-processing.

[0136] After pre-processing the original signal by Hilbert transform and time delay function, the harmonic components with similar frequencies can be split, wherein contains first, fifth, seventh and other harmonics, and contains third, ninth and other harmonics.

[0137] Further, VMD decomposition can be performed on According to the signal composition characteristics, the number of decomposition layers can be set to 4. At this time, the two first layer IMF1 components after decomposition mainly contain the fundamental frequency and third harmonic components (i.e. low frequency components) of the original zero sequence current. Adding the two signals as the data source for subsequent fault feature extraction.

[0138] As a refinement of the above embodiment, the step S2 is combined with the embodiment 1, and finally the transient energy and the phase plane trajectory are selected as the fault features of the response time-frequency energy and waveform morphology change.

[0139] According to Figures 6-7 , the low-frequency energy of early faults of any discharge type will have a more obvious fluctuation. In addition, Figure 3 Figure 5 The early stage of the hidden danger of early faults shows that the fault waveform changes very weakly and is easily submerged in the measurement noise. Therefore, a high-precision signal processing method needs to be designed to extract more critical and effective low-frequency components in the zero sequence current. Then, based on the zero sequence current low-frequency component, the transient energy and the phase plane trajectory features are extracted, and the two fault features are used to develop a recording wave starting criterion suitable for early faults.

[0140] As a refinement of the above embodiment, the transient energy calculation method in step S3 is as follows:

[0141] The first half wave after the fault usually contains rich transient information, and the transient energy ​​​​​It can describe the energy difference in the fault waveform before and after the fault, and the calculation formula is:

[0142] ,

[0143] In the formula, This refers to the number of data sampling points per cycle. This is the low-frequency component of the zero-sequence current. These are the sampling points.

[0144] The phase plane trajectory is calculated as follows:

[0145] Phase space reconstruction is an analytical method used to study time series. According to the time delay theorem, for a zero-sequence current time series of length N... When the delay time is Embedding dimension is At that time, it can be reconstructed into Phase space. After reconstruction. The length is of A sequence of phasors, expressed as:

[0146] ,

[0147] In the formula, ; express indivual The zero-sequence current phase plane trajectory formed by points in a certain dimension. Indicates the first Zero-sequence current data in each dimension.

[0148] When considering early fault flashover, the discharge time is typically maintained within a quarter cycle, with a delay time... Using T / 4, embedding dimension Setting it to 2 allows for the formation of a two-dimensional phase plane trajectory.

[0149] The Euclidean distance is used to quantify the degree of deviation of the phase plane curve from the origin as a feature of the phase plane trajectory. It directly reflects the straight-line distance of all data points from the origin, and the larger the value, the greater the overall deviation.

[0150] ,

[0151] In the formula, and These are the two zero-sequence currents used to construct the phase plane trajectory, with lengths of [missing information]. ,and Lag A quarter cycle.

[0152] As a refinement of the above embodiment, the step S4 fault probability index calculation is as follows:

[0153] For the collected zero sequence current, first, the original signal is transformed by HT-VMD, and then the transient energy and phase plane trajectory of the transformed signal are extracted based on the sliding time window, and the transient energy and phase plane trajectory are calculated by and The energy index and the shape index are designed.

[0154] Because the fault feature is weak, it is difficult to directly design a threshold as the starting criterion triggering condition for fault hidden danger detection. According to the requirement of reliable and continuous normal operation of the distribution network, the occurrence of the early fault event can be regarded as a probabilistic event, the above two fault indexes are probabilistically quantified, and the moment when the probability increases is regarded as the moment when the early fault occurs according to the continuous detection of the sliding time window.

[0155] The Laplace distribution can evaluate random possibility events and provide a natural extension for general random processes. In addition, the Laplace distribution can provide a linear combination of two different indexes without losing the effectiveness of the features. The above two fault indexes are effectively merged by using the Laplace distribution, and the real-time probability of the occurrence of the fault of the distribution network can be estimated:

[0156] ,

[0157] wherein, is the transient energy mutation variable, which is obtained by subtracting the transient energy in the last time window from the transient energy in the current time window, is the phase plane trajectory mutation variable, which is obtained by subtracting the phase plane trajectory in the last time window from the phase plane trajectory in the current time window, is the mean square error of the transient energy mutation variable in the ten cycles before the fault occurrence moment, is the mean square error of the phase trajectory mutation variable in the ten cycles before the fault occurrence moment, is the average value of and , is the location parameter of the Laplace distribution function, and are the independent variables of the Laplace distribution function, which are respectively substituted into the moment t The transient energy mutation variable and the phase plane trajectory mutation variable are calculated.

[0158] In this embodiment, it is considered that the two indexes are equally important, the mean value in the distribution is set to 0, the mean square error is calculated by using the measurement noise during normal operation, and the Laplace probability density distribution diagram is as shown in Figure 11 .

[0159] As a refinement of the above embodiment, the double threshold triggering recording wave starting condition is set according to the fault probability index in the step S5, and the specific mode is as follows:

[0160] Based on the zero sequence current low frequency component as a signal input, the energy index and the shape index are extracted in a half cycle time window, and the feature is quantified through the cumulative distribution function, which is used to evaluate the system operation state. The mutation of the two indexes is weak in normal operation, and the calculated fault probability is low; when the fault occurs, the zero sequence current gradually increases. As shown in Figures 12-13 Figure 12 is the recording wave form of the early fault development process of the insulator, Figure 13 is the probability index calculation result, and it can be seen that the probability is obviously highlighted in the weak discharge stage, and the probability index is close to 1 after the insulator breakdown. Through the mutation of the probability index, the occurrence of the early fault of the distribution network can be sensitively identified, and the probability size can reflect the fault degree.

[0161] Therefore, two triggering thresholds are designed, and When the calculated real-time fault probability meets more than and less than for more than 5 times in a set time window (1 second in this embodiment), it is considered that there is a device fault hidden danger problem in the distribution network, and the fault recording function of the measuring device is started when the next criterion is met; when the fault probability is greater than at a certain moment, it is considered that the distribution network equipment has an early fault due to insulation deterioration, accompanied by obvious discharge phenomenon, and the fault recording function of the measuring device is started at this time.

[0162] It should be noted that: the present application can effectively extract the zero sequence current low frequency component, further calculate the transient energy and the phase plane trajectory feature, and utilize the Laplace distribution combined with two different fault features to quantify the feature change before and after the fault occurs. Compared with the problem that the traditional threshold is not suitable for early faults, the starting criterion based on the fault probability can respond to the early fault of the distribution network in the weak discharge stage.

[0163] Finally, it should be noted that: the above only describes the preferred embodiments of the present application, and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, and those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.​

Claims

1. A method for selecting waveform initiation features of early faults in a distribution network, characterized in that, Includes the following steps: Multidimensional fault feature analysis is performed on early fault recording data of distribution network to construct a multidimensional fault feature set and extract waveform morphology features and time-frequency domain features. Calculate the cost-benefit analysis of fault characteristics based on computational complexity and fault saliency; The fault features are sorted in descending order according to their cost-benefit ratio, and the top M fault features are selected as the waveform recording start-up features for early faults in the distribution network. The benefits and costs The calculation formula is as follows: in, This refers to the number of data sampling points per cycle. As a fault significance indicator, This is a metric for computational complexity. The computational complexity is represented using Big O notation. Based on the number of data sampling points in a single cycle, the number of multiply-accumulate operations, floating-point operations, or iterations required in each feature extraction process are analyzed to obtain a computational complexity index. Fault features with a computational complexity index ratio of less than 10 to the number of data sampling points in a single cycle are eliminated. The significance of the fault Quantification is achieved through a mutation index, with the specific formula as follows: ， in, Representing the The characteristic value calculated using half-cycle data after a fault occurs; It is the average value of all characteristic values ​​calculated using a half-cycle sliding time window within the ten cycles prior to the time of the fault occurrence; It is the standard deviation of all eigenvalues ​​calculated using a half-cycle sliding time window within the ten cycles preceding the fault occurrence.

2. The method for selecting waveform recording start-up features for early faults in distribution networks according to claim 1, characterized in that, The multidimensional fault feature set includes waveform morphology features and time-frequency features; The waveform features include: kurtosis, skewness, waveform factor, peak factor, impulse factor, margin factor, and phase plane trajectory; The time-frequency domain features include: transient energy, total harmonic distortion rate, transient centroid frequency, power spectral entropy, wavelet singular entropy, and frequency band energy entropy.

3. A method for initiating waveform recording judgment of early faults in a distribution network, characterized in that, include: The zero-sequence current signal is acquired, and the zero-sequence current signal is transformed based on the improved variational mode decomposition method to extract the low-frequency component in the zero-sequence current signal. The method for selecting early fault identification features of distribution network early faults using the waveform recording start-up feature selection method described in claim 2 includes transient energy features and phase plane trajectory features. The transient energy and phase plane trajectory of the transformed zero-sequence current signal are extracted based on the sliding time window, and the transient energy change and phase plane trajectory change are calculated. The two mutation parameters are fused into a failure probability index using the Laplace distribution; Based on the fault probability index, a dual-threshold trigger waveform recording start condition is set. If the zero-sequence current signal meets the dual-threshold trigger waveform recording start condition, waveform recording is started. The failure probability index The calculation formula is: ， in, The transient energy change is calculated by subtracting the transient energy in the current time window from the transient energy in the previous time window. The abrupt change in the phase plane trajectory is obtained by subtracting the phase plane trajectory within the current time window from the phase plane trajectory within the previous time window. This represents the mean square error of transient energy fluctuations within the ten periods prior to the fault occurrence. This represents the mean square error of the phase trajectory abrupt changes within the ten periods prior to the fault occurrence. for and The average value, The location parameters of the Laplace distribution function, and Let be the independent variable of the Laplace distribution function, and substitute them into the time intervals respectively. Calculated transient energy mutation and phase plane trajectory mutation amount .

4. The method for initiating waveform recording judgment of early faults in distribution networks according to claim 3, characterized in that, The improved variational mode decomposition method for transforming zero-sequence current signals includes the following steps: Distinguish the fault current signal in the zero-sequence current signal, wherein the fault current signal is an aliasing of multiple frequency signals; The fault current signal is subjected to Hilbert transform and time delay processing to obtain two preprocessed signals. , : ， in, The fault current after Hilbert transformation. This is the fault current after time delay processing. For fault current Second harmonic angular frequency Sampling time, For fault current Second harmonic phase , This is the maximum frequency of harmonics that the device can collect; The sum of the preprocessed signals is decomposed using VMD to extract the low-frequency components of the zero-sequence current signal, including the fundamental frequency and the third harmonic component. These components are then summed as the data source for subsequent fault feature extraction.

5. The method for initiating waveform recording judgment of early faults in distribution networks according to claim 4, characterized in that, The formula for calculating transient energy is as follows: ， in, Transient energy, This refers to the number of data sampling points per cycle. This is the low-frequency component of the zero-sequence current. For sampling points; The phase plane trajectory is calculated as follows: The transformed zero-sequence current signal is reconstructed into a phasor sequence by performing phase space reconstruction: ， in, ; express indivual The zero-sequence current phase plane trajectory formed by points in a certain dimension. Indicates the first Zero-sequence current data in each dimension To delay time, For the embedding dimension, These are sampling points for the low-frequency component of the zero-sequence current. The Euclidean distance is used to quantify the degree of deviation of the phase plane curve from the origin as a feature of the phase plane trajectory. : ， In the formula, and These are the two dimensions of zero-sequence current data used to construct the phase plane trajectory, with lengths of [missing information]. ,and Lag A quarter cycle.

6. The method for initiating waveform recording judgment of early faults in distribution networks according to claim 5, characterized in that, The dual-threshold triggered waveform recording start condition is: When the number of times the fault probability falls within the range of two thresholds within a set time period exceeds the set threshold, fault recording is initiated. When the probability of failure exceeds the larger threshold of the two thresholds, fault recording is triggered immediately.

Images