Early fault voltage and current identification method and system based on multi-time scale transformation

Abstract

The invention relates to a voltage and current identification method and system for an early fault based on multi-scale transformation, and the method specifically comprises the steps: determining the topological information of a line, and collecting the power supply voltage waveform and current waveform of the line; filtering the voltage waveform, calculating an instantaneous voltage energy value of a time window, and calculating a left-right side time window energy difference value of adjacent voltage periods and a left-right side time window energy difference value of the same voltage period according to the instantaneous voltage energy value to serve as a preliminary judgment basis of the arc fault; performing multi-scale transformation on the current waveform, and extracting high-frequency information and low-frequency information characteristics of the current; a neural network model is adopted for modeling, and classification of early fault development periods is achieved; based on the early fault development time period, early fault degree judgment of the cable line is given, and operation and maintenance personnel are assisted in further decision making. According to the invention, related parameters of early faults containing various interference factors in a complex environment can be monitored, and system operation personnel can be helped to take corresponding measures for optimization in time.

Description

Technical Field

[0001] The present invention belongs to the field of fault detection of power transmission line periphery, and more specifically, relates to a method and system for early fault voltage and current identification based on multi-time scale transformation. Background Art

[0003] For cable-overhead line hybrid multi-branch lines, the overvoltage on the lines comes from many aspects, including early faults of the cable system, permanent faults of the cable system, lightning overvoltage of overhead lines, load fluctuations, and operating overvoltage caused by circuit breaker operation. Among them, early faults of the cable system are an important precursor to permanent cable failures. Early faults of the cable system refer to multiple repetitive discharge faults from a single breakdown of the cable or accessories to the action of the system relay protection. During the early fault period, local discharge or even intermittent arc discharge may occur in the discharge channel. In this process, there are situations where the discharge is extinguished or the discharge lasts for a long time. Since the discharge time usually ranges from a fraction of a cycle to four cycles, the relay protection equipment cannot yet take corresponding actions. Existing monitoring methods often have problems of difficulty in capturing signals or misjudging early faults as other transient phenomena.

[0004] As for early fault warning methods, although convolutional neural networks, hybrid neural networks based on attention mechanisms, and early fault prediction methods based on periodic time series have been used to identify early faults in lines, the convolutional neural network-based method requires a large number of sample data sets and training sets to capture the details of early faults. There is a lack of preliminary recognition methods for the typical characteristics of early faults. The output features of the hybrid neural network are fused and classified through a multi-head attention mechanism. There is a lack of separation methods for different time scales of early faults. Building a distribution network early fault prediction network model based on periodic time series is heavily dependent on the accuracy of the data set and the richness of the samples. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the present invention provides a method for identifying early fault voltage and current, which can identify the development process of early faults based on multi-scale transformation.

[0006] The present invention adopts the following technical solution. In a first aspect, the present invention provides a method for early fault voltage and current identification based on multi-time scale transformation, the method comprising the following steps:

[0007] Step 1: According to the line topology information, collect the power supply voltage waveform and current waveform of the line;

[0008] Step 2: Based on the power supply voltage waveform collected in step 1, the power supply voltage waveform is filtered to calculate multiple parameters, and according to the multiple parameters, it is determined whether the power supply voltage waveform meets the early fault condition;

[0009] Step 3: After determining that the power supply voltage waveform meets the early fault condition, a multi-scale wavelet transform is performed on the current waveform to extract the high-frequency information and low-frequency information characteristics of the current;

[0010] Step 4: input the high-frequency information of the current, the low-frequency information of the current and the current sampling samples containing the severity of the early fault obtained in step 3 into the neural network model training to obtain a trained neural network model;

[0011] Step 5: Based on the trained neural network model obtained in step 4, the high-frequency detail coefficients and low-frequency approximation coefficients of the energy distribution characteristics of the current waveform of the real-time sampling target line are identified to determine the severity of the early fault of the target line and obtain the identification result of the early fault.

[0012] Preferably, the multiple parameters include an instantaneous voltage energy value of a time window, and calculating the instantaneous voltage energy value of the time window includes: filtering the power supply voltage waveform using a bandpass filter, and dividing the filtered power supply voltage waveform into a left time window and a right time window, and calculating the instantaneous voltage energy values ​​of the left time window and the right time window. The calculation formula of the instantaneous energy value is as follows:

[0013]

[0014] In the formula,

[0015] E L (k) is the instantaneous voltage energy value of the left time window in the kth time period,

[0016] E R (k) is the instantaneous voltage energy value of the right time window in the kth time period,

[0017] U L，k (j) is the jth voltage sampling value in the left time window in the kth time period,

[0018] U R，k (j) is the jth voltage sampling value in the right time window in the kth time period,

[0019] n1 represents the number of samples.

[0020] Preferably, the multiple parameters further include energy differences between left and right time windows of adjacent time periods, and calculating the energy differences between left and right time windows of adjacent time periods includes:

[0021] The energy difference between the left time window and the right time window of adjacent time periods is calculated using the following formula:

[0022]

[0023] In the formula,

[0024] ΔE L (k) is the energy difference of the left time window of adjacent time periods,

[0025] ΔE R (k) is the energy difference of the right time window of adjacent time periods.

[0026] Preferably, the multiple parameters further include an energy difference between a left time window and a right time window in the same time period, and calculating the energy difference between the left time window and the right time window in the same time period includes:

[0027] The energy difference between the left time window and the right time window of the same time period is calculated using the following formula:

[0028] ΔE LR (k)=|E L (k)-E R (k)| (3)

[0029] In the formula,

[0030] ΔE LR (k) is the energy difference between the left time window and the right time window in the same time period.

[0031] Preferably, whether the power supply voltage waveform satisfies the early fault conditions is judged based on multiple parameters, including: when the instantaneous voltage energy value of the time window is greater than the set threshold, the energy difference between the left time window and the right time window of adjacent time periods is greater than the set threshold, and the energy difference between the left time window and the right time window of the same time period is greater than the set threshold, it is characterized that the power supply voltage waveform satisfies the early fault conditions at the same time, and the early fault counter is counted by 1, the power supply voltage waveform is continuously collected, and it is repeatedly judged whether the power supply voltage waveform satisfies the early fault conditions at the same time, and when the counter reaches n times, the next step is performed.

[0032] Preferably, the instantaneous voltage energy value of the time window is greater than the set threshold value including the instantaneous energy value of the left time window, the instantaneous voltage energy value E L (k) is greater than the set threshold ε1 and the instantaneous energy value of the right time window, the instantaneous voltage energy value E R (k) is greater than the set threshold ε2, expressed as follows:

[0033]

[0034] In the formula,

[0035] ε1 is the average left time window energy of all sampling sequences,

[0036] ε2 is the average right time window energy of all sampling sequences, expressed as follows:

[0037]

[0038] In the formula,

[0039] N represents the number of sampling sequences,

[0040] E Li represents the signal energy of the left time window of the i-th sampling sequence,

[0041] E Ri represents the signal energy of the right time window of the i-th sampling sequence,

[0042] η1 and η2 are margin coefficients.

[0043] Preferably, the energy difference between the left time window and the right time window of adjacent time periods is greater than a set threshold, including the energy difference between the left time windows of adjacent time periods being greater than a set threshold ε3 and the energy difference between the right time windows of adjacent time periods being greater than a set threshold ε4, which is expressed by the following formula:

[0044]

[0045] In the formula,

[0046] ε3 is the average difference in energy between the left time windows of adjacent sampling sequences,

[0047] ε4 is the average difference in energy between the right time windows of adjacent sampling sequences, expressed as follows:

[0048]

[0049] In the formula,

[0050] N is the number of sampling sequences,

[0051] E Li is the signal energy of the left time window of the i-th sampling sequence,

[0052] E L(i-1) is the signal energy of the left time window of the i-1th sampling sequence,

[0053] E Ri is the signal energy of the right time window of the i-th sampling sequence,

[0054] E R(i-1) is the signal energy of the right time window of the i-1th sampling sequence,

[0055] 113 and 114 are margin coefficients.

[0056] Preferably, the energy difference between the left time window and the right time window of the same time period is greater than the set threshold ε5, which is expressed by the following formula:

[0057] ΔE LR (k)>ε5 (8)

[0058] In the formula,

[0059] ε5 is the average difference between the energy of the left time window and the energy of the right time window of the same sampling sequence, expressed as follows:

[0060]

[0061] In the formula,

[0062] N represents the number of sampling sequences,

[0063] E Li represents the signal energy of the left time window of the i-th sampling sequence,

[0064] E Ri represents the signal energy of the right time window of the i-th sampling sequence,

[0065] η5 is the margin coefficient.

[0066] Preferably, performing a multi-scale wavelet transform on the current waveform includes:

[0067] The current waveform is continuously decomposed by discrete wavelet transform to obtain K-layer wavelet transform. The m-th low-frequency approximate coefficient a of the low-frequency component cAi of the i-th layer wavelet transform is i [m] and the mth high-frequency detail coefficient d of the high-frequency component cDi of the i-th layer wavelet transform i [m], expressed as follows:

[0068]

[0069] in,

[0070] K represents the number of wavelet transform layers,

[0071] h1[k0] represents the low-pass filter coefficient when the filter length is k0,

[0072] h0[k0] represents the high-pass filter coefficient when the filter length is k0.

[0073] Preferably, the high-frequency information and low-frequency information of the current are divided into a test set and a data set, and current sampling samples containing the severity of early faults are given. Adaptive learning is performed in the BP neural network, and the condition for ending the training is set to that the maximum error of the training set is less than 0.05. After the training is completed, a neural network model for judging the severity of early faults is obtained.

[0074] The second aspect of the present invention provides a voltage and current identification system for early faults based on multi-scale transformation, which is used to run the voltage and current identification method for early faults based on multi-scale transformation described in the first aspect, comprising:

[0075] Acquisition module: used to collect the power supply voltage waveform and current waveform of the line according to the line topology information;

[0076] Judgment module: used to filter the power supply voltage waveform to calculate multiple parameters and determine whether the power supply voltage waveform meets the early fault conditions;

[0077] Extraction module: used to determine whether the power supply voltage waveform meets the early fault conditions, and then perform multi-scale wavelet transform on the current waveform to extract the high-frequency and low-frequency information characteristics of the current;

[0078] Training module: used to input the high-frequency information of current, the low-frequency information of current and the current sampling samples containing the severity of early faults into the neural network model training to obtain the trained neural network model;

[0079] Identification module: It is used to identify the high-frequency detail coefficients and low-frequency approximation coefficients of the energy distribution characteristics of the current waveform of the real-time sampling target line according to the trained neural network model, determine the severity of the early fault of the target line, and obtain the identification result of the early fault.

[0080] A computer device comprising: one or more processors;

[0081] The processor is used to store one or more programs;

[0082] When the one or more programs are executed by the one or more processors, the voltage and current identification method for early faults based on multi-scale transformation is implemented.

[0083] A computer-readable storage medium, characterized in that a computer program is stored thereon, and when the computer program is executed, the voltage and current identification method of early faults based on multi-scale transformation is implemented. The beneficial effect of the present invention is that, compared with the prior art, after the method comprehensively analyzes the voltage and current waveform sequence data of the cable, the method can be based on the instantaneous energy difference between the left and right time windows of adjacent voltage cycles and the energy difference between the left and right time windows of the same voltage cycle as the preliminary judgment basis of the early arc fault to make a preliminary judgment of the early fault, and the instantaneous voltage energy value feature and multi-scale transformation are used to separate the low-frequency information feature and high-frequency information feature after the early fault occurs. When the sample data set is lacking in richness, it can still be classified and identified based on the typical features of the early fault waveform, which is convenient for subsequent classification and identification based on the typical features of different frequency bands; in the case of a large interference degree and a small sample, the preliminary judgment of the early fault can still be achieved with a high accuracy. In view of the different time scale characteristics of early arc faults in different development stages, the present invention extracts multiple time scale characteristics of arc based on the arc development mechanism, increases the detection basis of early arc fault signals, reduces the possibility of false alarms caused by interference, and can realize the monitoring of relevant parameters of early faults in complex environments with multiple interference factors, helping system operators to have a more comprehensive understanding of the operating quality of the cable and take corresponding measures in time to optimize it. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 is a flow chart of the present invention;

[0085] Figure 2 This is a preliminary method for judging early-stage faults of the present invention;

[0086] Figure 3 It is a schematic diagram of wavelet transformation of the present invention;

[0087] Figure 4 The present invention is based on the early fault identification of the neural network model. DETAILED DESCRIPTION

[0088] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0089] In view of the problems existing in the prior art, the purpose of the present invention is to propose a method for identifying early fault voltage and current. After comprehensive analysis of the voltage and current of the line, the method can study the development process of early faults based on multi-scale transformation and energy calculation to obtain the identification results of early faults.

[0090] To achieve the above object, the present invention provides a method for identifying early fault voltage and current based on multi-scale transformation, such as Figure 1 As shown, the method comprises the following steps:

[0091] Step 1: According to the line topology information, collect the line power supply voltage waveform and current waveform;

[0092] Step 2: Filter the power supply voltage waveform to calculate multiple parameters, including the instantaneous voltage energy value of the time window, the energy difference between the left and right time windows of adjacent time periods, and the energy difference between the left time window and the right time window of the same time period. According to the multiple parameters, an early fault condition is constructed to determine whether the power supply voltage waveform meets the early fault condition.

[0093] Furthermore, the flowchart of the preliminary judgment method of early fault is as follows: Figure 2 As shown, the early fault conditions include that the instantaneous voltage energy value of the time window is greater than the set threshold, the energy difference between the left time window and the right time window of adjacent time periods is greater than the set threshold, and the energy difference between the left time window and the right time window of the same time period is greater than the set threshold. When the power supply voltage waveform satisfies the early fault conditions at the same time, the early fault counter is incremented by 1, the power supply voltage waveform is continuously collected, and it is repeatedly determined whether the power supply voltage waveform satisfies the early fault conditions at the same time. When the counter reaches n times, the next step is performed.

[0094] In step 2, the voltage waveform signal is filtered by a bandpass filter, and the filtered voltage waveform signal is set to have a left time window from before the voltage crosses zero to the maximum voltage amplitude, and a right time window from after the voltage crosses zero to the maximum voltage amplitude, and the instantaneous voltage energy value is calculated. The calculation formula is as follows:

[0095]

[0096] In the formula, E L (k) is the instantaneous voltage energy value of the left time window in the kth time period, E R (k) is the instantaneous voltage energy value of the right time window in the kth time period, U L，k (j) is the jth sampling value of the voltage in the left time window of the kth time period, U R，k (j) is the j-th sampling value of the voltage in the right time window of the k-th time period, and n1 represents the sampling number.

[0097] Furthermore, the energy difference between the left and right time windows of adjacent time periods is calculated:

[0098]

[0099] ΔE L (k) is the energy difference between the left time windows of adjacent time periods, ΔE R (k) is the energy difference of the right time window of adjacent time periods.

[0100] The energy difference between the left and right time windows in the same time period is:

[0101] ΔE LR (k)=|E L (k)-E R (k)| (3)

[0102] Furthermore, since the arc voltage waveform will have two obvious rising processes at the moment of arc generation and extinction, these two pulses are transmitted through the cable line and affect the energy of the power supply voltage waveform. Normally, when there is no early fault arc, the energy difference between the left and right time windows of adjacent time periods and the energy difference between the left and right time windows of the same time period should be close to 0. When an arc fault occurs, the instantaneous voltage energy values ​​of the left and right time windows of the same time period will increase. The instantaneous voltage energy value E of the left time window of the kth time period is L (k) and the instantaneous energy value E of the right time window of the kth time period R (k) Exceeding the critical values ​​ε1 and ε2, the instantaneous voltage energy value of the time window is greater than the set threshold, including the instantaneous energy value of the left time window, the instantaneous voltage energy value E L (k) is greater than the set threshold ε1 and the instantaneous energy value of the right time window, the instantaneous voltage energy value E R (k) is greater than the set threshold ε2, expressed as follows:

[0103]

[0104] The left critical value is represented by the average left time window energy ε1 of all sampling sequences, and the right critical value is represented by the average right time window energy ε2 of all sampling sequences.

[0105] The signal under normal conditions is sampled multiple times, and the average left time window energy ε1 and the average right time window energy ε2 of all sampling sequences are calculated. The specific calculation formula is shown as follows:

[0106]

[0107] Where N represents the number of sampling sequences, E Lirepresents the signal energy of the left time window of the i-th sampling sequence, E Ri It represents the signal energy in the right time window of the i-th sampling sequence, and η1 and η2 are margin coefficients.

[0108] Furthermore, when an early arc fault occurs, the instantaneous energy fluctuates significantly in different power supply cycles. The energy difference between the left and right time windows of adjacent time cycles is used for judgment.

[0109] The energy difference between the left time window and the right time window of adjacent time periods is greater than the set threshold, including the energy difference between the left time window of adjacent time periods is greater than the set threshold ε3 and the energy difference between the right time windows of adjacent time periods is greater than the set threshold ε4, which is expressed by the following formula:

[0110]

[0111] The signal under normal conditions is sampled multiple times, and the average difference ε3 of the energy in the left time window and the average difference ε4 of the energy in the right time window of adjacent sampling sequences are calculated. The specific calculation formula is shown as follows:

[0112]

[0113] Where N represents the number of sampling sequences, E Li represents the signal energy of the left time window of the i-th sampling sequence, E L(i-1) represents the signal energy of the left time window of the i-1th sampling sequence, E Ri represents the signal energy of the right time window of the i-th sampling sequence, E R(i-1) It represents the signal energy in the right time window of the i-1th sampling sequence, and η3 and η4 are margin coefficients.

[0114] Furthermore, when an early arc fault occurs, the instantaneous energy of the signal in the left time window is approximately equal to the instantaneous energy of the signal in the right time window. However, when the load fluctuates or the load is disturbing, ΔE LR (k) is large, so the energy difference between the left and right time windows of the same power supply time period can be used to well judge the occurrence of early faults. The energy difference between the left time window and the right time window of the same time period is greater than the set threshold ε5, which is expressed as follows:

[0115] ΔE LR (k)>ε5 (8)

[0116] The signal under normal conditions is sampled multiple times, and the average difference ε5 between the energy of the left time window and the energy of the right time window of the same sampling sequence is calculated. The specific calculation formula is shown as follows:

[0117]

[0118] Where N represents the number of sampling sequences, E Li represents the signal energy of the left time window of the i-th sampling sequence, E Ri represents the signal energy in the right time window of the i-th sampling sequence, and η5 is the margin coefficient.

[0119] Based on the above three judgment conditions, the power supply voltage is judged sequentially. When the three conditions are met at the same time, the early fault counter is incremented by one. When the counter reaches n times (n can be 3), it is preliminarily considered that an early fault has occurred, and the next step of judgment and analysis is carried out.

[0120] Step 3: After determining that the power supply voltage waveform in step 2 meets the early fault condition, multi-scale wavelet transform is performed on the current waveform to extract the high-frequency information and low-frequency information characteristics of the current;

[0121] Furthermore, in step 3, the current waveform is subjected to multi-scale wavelet transform, and then the high-frequency information and low-frequency information features of the current are extracted. Wavelet transform fully highlights the characteristics of the signal in certain aspects through a series of transformations, and can analyze the localization of time frequency, and can also gradually refine the signal at multiple scales through telescoping and translation operations, and finally achieve time subdivision at high frequencies and time subdivision at low frequencies.

[0122] Using discrete wavelet transform, that is, using orthogonal wavelet basis to decompose the signal into multiple components of different scales, similar to using high-pass filter and low-pass filter to continuously decompose the current signal, we can get the high-frequency component cDi and low-frequency component cAi under K-layer wavelet transform. The decomposition process is as follows: Figure 3 As shown. h0 and h1 represent the high-pass and low-pass filter coefficients respectively. If the signal length is m and the filter length is k0, then the m-th low-frequency approximate coefficient a of the low-frequency component cAi of the i-th layer wavelet transform is i [m] and the mth high-frequency detail coefficient d of the high-frequency component cDi of the i-th layer wavelet transform i [m] is

[0123]

[0124] in,

[0125] K represents the number of wavelet transform layers,

[0126] h1[k0] represents the low-pass filter coefficient when the filter length is k0,

[0127] h0[k0] represents the high-pass filter coefficient when the filter length is k0.

[0128] Through multi-scale transformation, the high-frequency and low-frequency information of the signal can be well preserved. By analyzing the energy values ​​of each frequency band, it is helpful to use the neural network model to realize early fault identification.

[0129] Step 4: Input the high-frequency information of the current, the low-frequency information of the current and the current sampling samples containing the severity of the early fault obtained in step 3 into the neural network model training to obtain the trained neural network model; divide the high-frequency information and low-frequency information of the current into a test set and a data set, give the current sampling samples containing the severity of the early fault, perform adaptive learning in the BP neural network to update the weights and thresholds, set the condition for the end of training to be that the maximum error of the training set is less than 0.05, and obtain the neural network model for judging the severity of early faults after the training is completed.

[0130] Furthermore, the step 4 adopts a neural network model to achieve the classification of early fault development periods. For the specific process, refer to Figure 4 . After an early fault occurs, the high-frequency component of the line current increases. The wavelet decomposition used in step 3 is used to calculate the energy index of each layer, and the high-frequency detail coefficients and low-frequency approximate coefficients of the energy distribution characteristics when the early fault occurs are extracted. Adaptive learning is performed in the BP neural network to update the weights and thresholds. Based on the self-learning, self-adaptation, nonlinear mapping and good generalization capabilities of the BP neural network, the development section of the early fault can be well identified. Based on the high-frequency detail coefficients and low-frequency approximate coefficients of the energy distribution characteristics when the early fault occurs obtained in step 3, the test set and the data set are divided, and the current sampling samples containing the severity of the early fault are given. The condition for the end of training is set to set the maximum error of the training set to be less than 0.05. The neural network model is trained to obtain a trained neural network model for judging the severity of the early fault.

[0131] Step 5: Use the trained neural network model to identify the high-frequency detail coefficients and low-frequency approximation coefficients of the energy distribution characteristics of the current waveform of the real-time sampling target line, determine the severity of the early fault of the target line, obtain the identification results of the early fault, and assist the operation and maintenance personnel in making further decisions.

[0132] Furthermore, step 5 is based on the development time period of the early fault. When the early cable fault develops seriously, it may affect the normal operation of the line and the normal operation of the user's equipment, and may cause user complaints. The operation and maintenance personnel can promptly replace the cable line and deactivate the cable line for inspection to further inspect the cable line to ensure the stability and reliability of power supply.

[0133] Example 2

[0134] Embodiment 2 of the present invention provides an early fault voltage and current identification system based on multi-time scale transformation, which is used to run the early fault voltage and current identification method based on multi-time scale transformation described in Embodiment 1, including:

[0135] Acquisition module: used to collect the power supply voltage waveform and current waveform of the line according to the line topology information;

[0136] Judgment module: used to filter the power supply voltage waveform to calculate multiple parameters and determine whether the power supply voltage waveform meets the early fault conditions;

[0137] Extraction module: used to determine whether the power supply voltage waveform meets the early fault conditions, and then perform multi-scale wavelet transform on the current waveform to extract the high-frequency and low-frequency information characteristics of the current;

[0138] Training module: used to input the high-frequency information of current, the low-frequency information of current and the current sampling samples containing the severity of early faults into the neural network model training to obtain the trained neural network model;

[0139] Identification module: It is used to identify the high-frequency detail coefficients and low-frequency approximation coefficients of the energy distribution characteristics of the current waveform of the real-time sampling target line according to the trained neural network model, determine the severity of the early fault of the target line, and obtain the identification result of the early fault.

[0140] Example 3

[0141] Embodiment 3 of the present invention provides a computer device, including: one or more processors;

[0142] The processor is used to store one or more programs;

[0143] When the one or more programs are executed by the one or more processors, the voltage and current identification method for early faults based on multi-scale transformation is implemented.

[0144] Example 4

[0145] Embodiment 4 of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the voltage and current identification method for early faults based on multi-scale transformation is implemented.

[0146] The beneficial effect of the present invention is that, compared with the prior art, after the method comprehensively analyzes the voltage and current waveform sequence data of the cable, the method can be based on the instantaneous energy difference between the left and right time windows of adjacent voltage cycles and the energy difference between the left and right time windows of the same voltage cycle as the preliminary judgment basis for early arc faults. The low-frequency information characteristics and high-frequency information characteristics after the early fault occur are separated by using the instantaneous voltage energy value characteristics and multi-scale transformation. When the sample data set is lacking in richness, it can still be classified and identified based on the typical characteristics of the early fault waveform, which is convenient for subsequent classification and identification based on the typical characteristics of different frequency bands; in the case of a large interference degree and a small sample, the preliminary judgment of the early fault can still be achieved with a high accuracy. In view of the different time scale characteristics of early arc faults at different development stages, the present invention extracts multiple time scale characteristics of the arc based on the arc development mechanism, increases the detection basis of early arc fault signals, reduces the possibility of false alarms caused by interference, and can realize the monitoring of relevant parameters of early faults with multiple interference factors in complex environments, helping system operators to more comprehensively understand the operation quality of the cable and take corresponding measures to optimize it in time.

[0147] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described action sequence, because according to the present application, some steps of certain factors can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the factor embodiments described in the specification are all preferred embodiments, and the actions and units involved are not necessarily required by the present application.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A voltage and current identification method for early faults based on multi-scale transformation, characterized in that: include: Step 1: According to the line topology information, collect the power supply voltage waveform and current waveform of the line; Step 2: Based on the power supply voltage waveform collected in step 1, the power supply voltage waveform is filtered to calculate multiple parameters, and according to the multiple parameters, it is determined whether the power supply voltage waveform meets the early fault condition; Step 3: After determining that the power supply voltage waveform meets the early fault condition, a multi-scale wavelet transform is performed on the current waveform to extract the high-frequency information and low-frequency information characteristics of the current; Step 4: input the high-frequency information of the current, the low-frequency information of the current and the current sampling samples containing the severity of the early fault obtained in step 3 into the neural network model training to obtain a trained neural network model; Step 5: Based on the trained neural network model obtained in step 4, the high-frequency detail coefficients and low-frequency approximation coefficients of the energy distribution characteristics of the current waveform of the real-time sampling target line are identified to determine the severity of the early fault of the target line and obtain the identification result of the early fault.

2. The voltage and current identification method for early faults based on multi-scale transformation according to claim 1 is characterized in that: The multiple parameters include an instantaneous voltage energy value of a time window. The instantaneous voltage energy value of the time window is calculated, including: filtering the power supply voltage waveform with a bandpass filter, and dividing the filtered power supply voltage waveform into a left time window and a right time window, and calculating the instantaneous voltage energy values ​​of the left time window and the right time window. The calculation formula of the instantaneous energy value is as follows: In the formula, E L (k) is the instantaneous voltage energy value of the left time window in the kth time period, E R (k) is the instantaneous voltage energy value of the right time window in the kth time period, U L,k (j) is the jth voltage sampling value in the left time window in the kth time period, U R,k (j) is the jth voltage sampling value in the right time window in the kth time period, n1 represents the number of samples.

3. The voltage and current identification method for early faults based on multi-scale transformation according to claim 2 is characterized in that: The multiple parameters also include energy differences between left and right time windows of adjacent time periods. Calculating the energy differences between left and right time windows of adjacent time periods includes: The energy difference between the left time window and the right time window of adjacent time periods is calculated using the following formula: In the formula, ΔE L (k) is the energy difference of the left time window of adjacent time periods, ΔE R (k) is the energy difference of the right time window of adjacent time periods.

4. The voltage and current identification method for early faults based on multi-scale transformation according to claim 3 is characterized in that: The multiple parameters also include an energy difference between a left time window and a right time window of the same time period. Calculating the energy difference between the left time window and the right time window of the same time period includes: The energy difference between the left time window and the right time window of the same time period is calculated using the following formula: ΔE LR (k)=|E L (k)-E R (k)| (3) In the formula, ΔE LR (k) is the energy difference between the left time window and the right time window in the same time period.

5. The voltage and current identification method for early faults based on multi-scale transformation according to claim 4 is characterized in that: Based on multiple parameters, determine whether the power supply voltage waveform meets the early fault conditions, including: when the instantaneous voltage energy value of the time window is greater than the set threshold, the energy difference between the left time window and the right time window of adjacent time periods is greater than the set threshold, and the energy difference between the left time window and the right time window of the same time period is greater than the set threshold, it is characterized that the power supply voltage waveform meets the early fault conditions at the same time, and the early fault counter is counted by 1, the power supply voltage waveform is continuously collected, and it is repeatedly determined whether the power supply voltage waveform meets the early fault conditions at the same time. When the counter reaches n times, proceed to the next step.

6. The voltage and current identification method of early faults based on multi-scale transformation according to claim 5 is characterized in that: The instantaneous voltage energy value of the time window is greater than the set threshold, including the instantaneous energy value of the left time window, the instantaneous voltage energy value E L (k) is greater than the set threshold ε1 and the instantaneous energy value of the right time window, the instantaneous voltage energy value E R (k) is greater than the set threshold ε2, expressed as follows: In the formula, ε1 is the average left time window energy of all sampling sequences, ε2 is the average right time window energy of all sampling sequences, expressed as follows: In the formula, N represents the number of sampling sequences, E Li represents the signal energy of the left time window of the i-th sampling sequence, E Ri represents the signal energy of the right time window of the i-th sampling sequence, η1 and η2 are margin coefficients.

7. The voltage and current identification method for early faults based on multi-scale transformation according to claim 5 is characterized in that: The energy difference between the left time window and the right time window of adjacent time periods is greater than the set threshold, including the energy difference between the left time window of adjacent time periods is greater than the set threshold ε3 and the energy difference between the right time windows of adjacent time periods is greater than the set threshold ε4, which is expressed by the following formula: In the formula, ε3 is the average difference in energy between the left time windows of adjacent sampling sequences, ε4 is the average difference in energy between the right time windows of adjacent sampling sequences, expressed as follows: In the formula, N is the number of sampling sequences, E Li is the signal energy of the left time window of the i-th sampling sequence, E L(i-1) is the signal energy of the left time window of the i-1th sampling sequence, E Ri is the signal energy of the right time window of the i-th sampling sequence, E R(i-1) is the signal energy of the right time window of the i-1th sampling sequence, η3 and η4 are margin coefficients.

8. The voltage and current identification method for early faults based on multi-scale transformation according to claim 5 is characterized in that: The energy difference between the left time window and the right time window of the same time period is greater than the set threshold ε5, which is expressed by the following formula: DE LR (k)>ε5 (8) In the formula, ε5 is the average difference between the energy of the left time window and the energy of the right time window of the same sampling sequence, expressed as follows: In the formula, N represents the number of sampling sequences, E Li represents the signal energy of the left time window of the i-th sampling sequence, E Ri represents the signal energy of the right time window of the i-th sampling sequence, η5 is the margin coefficient.

9. The voltage and current identification method of early faults based on multi-scale transformation according to claim 1 is characterized in that: Perform multi-scale wavelet transform on the current waveform, including: The current waveform is continuously decomposed by discrete wavelet transform to obtain K-layer wavelet transform. The m-th low-frequency approximate coefficient a of the low-frequency component cAi of the i-th layer wavelet transform is i [m] and the mth high-frequency detail coefficient d of the high-frequency component cDi of the i-th layer wavelet transform i [m], expressed as follows: in, K represents the number of wavelet transform layers, h1[k0] represents the low-pass filter coefficient when the filter length is k0, h0[k0] represents the high-pass filter coefficient when the filter length is k0.

10. The voltage and current identification method of early faults based on multi-scale transformation according to claim 1, characterized in that: The high-frequency information and low-frequency information of the current are divided into a test set and a data set, and current sampling samples containing the severity of early faults are given. Adaptive learning is performed in the BP neural network. The condition for the end of training is set that the maximum error of the training set is less than 0.

05. After the training, a neural network model for judging the severity of early faults is obtained.

11. A voltage and current identification system for early faults based on multi-scale transformation, used for running a voltage and current identification method for early faults based on multi-scale transformation as claimed in any one of claims 1 to 10, characterized in that: Acquisition module: used to collect the power supply voltage waveform and current waveform of the line according to the line topology information; Judgment module: used to filter the power supply voltage waveform to calculate multiple parameters and determine whether the power supply voltage waveform meets the early fault conditions; Extraction module: used to determine whether the power supply voltage waveform meets the early fault condition, and then perform multi-scale wavelet transform on the current waveform to extract the high-frequency and low-frequency information characteristics of the current; Training module: used to input the high-frequency information of current, the low-frequency information of current and the current sampling samples containing the severity of early faults into the neural network model training to obtain the trained neural network model; Identification module: It is used to identify the high-frequency detail coefficients and low-frequency approximation coefficients of the energy distribution characteristics of the current waveform of the real-time sampling target line according to the trained neural network model, determine the severity of the early fault of the target line, and obtain the identification result of the early fault.

12. A computer device, characterized in that: include: one or more processors; The processor is used to store one or more programs; When the one or more programs are executed by the one or more processors, the voltage and current identification method for early faults based on multi-scale transformation as described in any one of claims 1 to 10 is implemented.

13. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, the voltage and current identification method for early faults based on multi-scale transformation as described in any one of claims 1 to 10 is implemented.

Images