Small current grounding system single-phase grounding fault intelligent line selection positioning method and system

Abstract

The invention discloses an intelligent line selection positioning method and system for a single-phase earth fault of a small-current grounding system. Calculating a true effective value amplification ratio and a fundamental wave effective value amplification ratio positioning transient state from a second cycle, and inputting the true effective value amplification ratio and the fundamental wave effective value amplification ratio positioning transient state into the convolutional neural network for judgment after preprocessing; the steady state identification is used for distinguishing a normal steady state, a grounding steady state and a normal recovery steady state through multi-dimensional detection of abnormal data; the phase difference between the zero-sequence voltage and the zero-sequence current is used for judging the interior / exterior of a grounding steady state boundary; and integrating transient and steady state results to make decisions. According to the method, the problems of high misjudgment rate, narrow application range and the like in the prior art are effectively solved, the accuracy of fault line selection and positioning is improved, the application system range is expanded, the adaptability to system parameter changes is enhanced, and an efficient and reliable scheme is provided for troubleshooting of a small-current grounding system.

Description

Technical Field

[0001] The present invention belongs to the technical field of small - current single - phase grounding fault line selection, and particularly relates to an intelligent line selection and positioning method and system for single - phase grounding faults in a small - current grounding system. Background Art

[0002] A small - current grounding system refers to a three - phase system in which the neutral point is not grounded, grounded through an arc suppression coil, or grounded through a high impedance, also known as a neutral - point indirectly grounded system. When a single - phase grounding fault occurs, since a short - circuit loop cannot be formed, the grounding fault current is often much smaller than the load current, so this system is called a small - current grounding system. A single - phase grounding fault is a common temporary fault, mostly occurring in humid and rainy weather. After a single - phase grounding occurs, the phase - to - ground voltage of the faulty phase decreases, and the phase voltages of the two non - faulty phases increase, but the line voltage remains symmetrical, so it does not affect the continuous power supply of the system to users, and the system can operate with the fault for 1 - 2 hours. However, if the power grid operates for a long time when a single - phase grounding fault occurs, other more serious accidents will be caused, and the fault needs to be removed in time.

[0003] After a single - phase grounding fault occurs, it is necessary to accurately locate the fault line and fault section to facilitate timely fault removal. The most commonly used method for troubleshooting is to manually disconnect circuits one by one until the abnormal characteristics disappear. However, this method will affect the normal power consumption of users and is cumbersome and inefficient when the topological structure of the power grid is relatively complex.

[0004] In the prior art, line selection and positioning include the transient first - half - wave method and group amplitude and phase comparison. The transient first - half - wave method is applicable to power grid systems with ungrounded neutral points and grounded through arc suppression coils. A single - phase grounding fault often occurs at the moment when the phase voltage is close to the maximum value. At this time, the capacitive charge of the faulty phase discharges. When the fault occurs at the moment when the phase voltage passes through zero, the transient inductive current is the largest; when it is close to the maximum value, the transient inductive current is zero, and the transient capacitive current is much larger than the inductive current. This method selects the line by using the fact that the transient zero - sequence current, voltage, and the amplitude and direction of the first half - wave of the faulty line are different from the normal situation. Specifically, at the moment of grounding, the amplitudes of the zero - sequence voltage and current increase significantly, and the directions of the mutation of the zero - sequence voltage and current at the in - zone detection points are opposite, while they are the same outside the zone.

[0005] Group amplitude and phase comparison is applicable to ungrounded neutral - point systems. Its principle is to first compare the zero - sequence currents, select the one with the larger amplitude as a candidate, and then perform phase comparison. The phase of the zero - sequence current at the in - zone detection points lags behind the zero - sequence voltage by 90°, and is opposite to the zero - sequence current at the out - of - zone detection points (the zero - sequence current outside the zone leads the zero - sequence voltage by 90°). However, in a power grid system with a neutral point grounded through an arc suppression coil, due to the compensation effect of the arc suppression coil, the zero - sequence currents at both in - zone and out - of - zone detection points lead the zero - sequence voltage.

[0006] However, the transient first half-wave method is affected by line disturbances, making it difficult to accurately determine the transient position, prone to misjudgment, and difficult to identify the direction of transient mutation. The group amplitude comparison and phase comparison method is only applicable to ungrounded neutral systems. When the neutral point characteristics change, the model cannot be adjusted in time, easily missing boundary faults. Its phase determination has insufficient fault tolerance for system parameter differences, and the complex phase relationship in the steady state can also lead to misjudgment. Therefore, there is an urgent need for a more reliable intelligent line selection and positioning algorithm to reduce the misjudgment rate, expand the applicable range, and enhance the adaptability to system parameter changes. Summary of the Invention

[0007] To solve the deficiencies in the prior art, the present invention provides an intelligent line selection and positioning method and system for single-phase grounding faults in a small current grounding system, solving the problems that the transient first half-wave method is easily affected by disturbances and prone to misjudgment and difficult to identify the mutation direction, as well as the problems of narrow applicable range, low fault tolerance to parameter changes, and easy error in steady-state judgment of the group amplitude comparison and phase comparison method. The present invention uses an innovative algorithm to achieve precise fault line selection and positioning by utilizing global optimal transient positioning, multi-category steady-state recognition, and comprehensive transient and steady-state results, reducing the misjudgment rate, expanding the applicable range, and enhancing the adaptability to system parameter changes.

[0008] The present invention adopts the following technical solutions.

[0009] The present invention proposes an intelligent line selection and positioning method for single-phase grounding faults in a small current grounding system, including:

[0010] After a single-phase grounding fault occurs in the small current grounding system, collect the zero-sequence voltage and zero-sequence current at the positions of each detection point; judge the transient process based on the true effective value increase ratio and fundamental wave increase ratio of the zero-sequence voltage and zero-sequence current, and select the optimal transient according to the score and ranking of the transient process for transient positioning;

[0011] Take the feature vectors of each detection point where the transient process appears as input signals, and perform transient discrimination through a trained transient discrimination neural network to judge whether each detection point is inside or outside the fault boundary;

[0012] Exclude abnormal data, calculate the fundamental wave effective value increase ratio of the subsequent cycles starting from the second half cycle after the transient and the normal steady state, and judge whether each detection point is in the grounding steady state;

[0013] For the detection points in the grounding steady state, judge whether the detection point is inside or outside the fault boundary according to the phase difference between the zero-sequence voltage and the zero-sequence current;

[0014] Calculate the ratio of the proportion of inside and outside the boundary in all the cycles in the grounding steady state, and combine the transient results to obtain the final result of whether the detection point is inside or outside the fault boundary.

[0015] Further, the score of the transient process includes the ratio of the true effective value increase of the zero-sequence voltage of the candidate transient and the ratio of the true effective value increase of the zero-sequence current of the candidate transient; for each candidate transient, the smaller value of the ratio of the true effective value increase of its zero-sequence voltage and zero-sequence current is used as the final score of the candidate transient, and the candidate transient with the largest final score is used as the optimal transient, thereby obtaining the transient positioning.

[0016] Further, if the fundamental wave increase ratio exceeds the fundamental wave increase ratio threshold, the cycle is placed in the first candidate set, otherwise it is placed in the second candidate set; after the traversal is completed, the optimal transient is selected from the first candidate set. If the first candidate set is empty, it is selected from the second candidate set; when there are multiple candidate transients in the same candidate set, calculate the ratio of the true effective value increase of the zero-sequence voltage and zero-sequence current of each candidate transient, extract the smaller value between the two, sort them in ascending order, and use the candidate transient corresponding to the largest sort as the optimal transient.

[0017] Further, the network structure adopted by the convolutional neural network is as follows:

[0018] Input layer: The input format is [n, 4, 256], where n represents the number of input samples, 4 represents four channels, and the channel order is: zero-sequence voltage, zero-sequence current, FFT result of zero-sequence voltage, and FFT result of zero-sequence current, and 256 represents the number of nodes in one cycle;

[0019] First convolutional layer: one-dimensional convolution Conv1d, batch normalization BN1d, activation function ReLu, one-dimensional max pooling MaxPool1d, linear transformation Linear in the fully connected layer, activation function softmax;

[0020] The structures of the second convolutional layer and the third convolutional layer are the same as that of the first convolutional layer;

[0021] Output layer: The output is [a, b, c], representing the probability values of [none, within the boundary, outside the boundary] respectively, and the one with the largest probability among the three is selected as the output result.

[0022] Further, the abnormal data includes 5 non-stable states;

[0023] The determination of the non-stable state includes whether the fundamental wave characteristics are obvious, whether there is an upward / downward trend, and whether it has the characteristics of a sine wave; the method for determining whether the fundamental wave characteristics are obvious is to sort the harmonic ratios from large to small, compare the sum of the top 5 with the harmonic ratio threshold, and filter the cycle when the sum of the harmonic ratios is greater than the harmonic ratio threshold; among them, the calculation formula for the harmonic ratio is:

[0024]

[0025] Among them, γ represents the harmonic proportion; X m represents the effective value of the m-th harmonic; base represents the effective value of the fundamental wave; M represents the number of harmonics.

[0026] Furthermore, the method for judging whether there is an upward / downward trend is to introduce the average value of the absolute values of the height differences between all adjacent nodes to standardize the difference. The specific formula is as follows:

[0027]

[0028] Among them, Y represents the rising and falling trend; Y d represents the height value of the d-th point in the cycle; D represents the total number of points.

[0029] Furthermore, whether it has the characteristics of a sine wave: the obviousness of the fundamental wave is measured by calculating the proportion of the effective value of the fundamental wave in the true effective value. If the proportion is less than the proportion threshold, the cycle is filtered. The proportion threshold is set to 2; the calculation formula for the proportion is as follows:

[0030]

[0031] Among them, λ represents the proportion; z0 represents the true effective value; base0 represents the effective value of the fundamental wave.

[0032] Furthermore, calculate the effective value of the fundamental wave of the previous unfiltered cycle before the transient and the first two unfiltered cycles of the overall waveform, and select the one with the largest effective value of the fundamental wave among the three as the normal steady state;

[0033] After determining the normal steady state, starting from the second half cycle of the transient, calculate the proportion of the increase in the effective value of the fundamental wave of the subsequent cycles to the normal steady state. If it exceeds the threshold, the subsequent cycle is considered to be the grounded steady state, otherwise it is the normal steady state restored after grounding;

[0034] When it is recognized that the normal steady state is restored after grounding or all the subsequent cycles have been traversed, stop the steady state recognition;

[0035] Calculate the phase difference between the zero-sequence voltage phase and the zero-sequence current phase of each cycle in the grounded steady state. The range of the phase difference is between [-π, π]; when the phase difference is between [-π, 0.4], it is judged as out of bounds, otherwise it is judged as in bounds.

[0036] Furthermore, distinguish according to the length of the grounded steady state. Within two cycles of the grounded steady state, it is considered that it is very likely to be restored to normal after instantaneous grounding. Regardless of the results of the transient and steady state, the final result to be pushed should be that the final result is judged as out of bounds;

[0037] When the grounding steady state is greater than two cycles, calculate the in-bound ratio and out-of-bound ratio in all grounding steady-state cycles. If the in-bound ratio or out-of-bound ratio is greater than the ratio threshold, the corresponding steady-state result is in-bound or out-of-bound. Combining with the transient result, if the transient is connected to the first grounding steady-state cycle, comprehensively consider the transient and steady-state results; if not connected, the final result is based on the steady-state result.

[0038] If both the in-bound ratio and the out-of-bound ratio do not reach the threshold, use the grounding steady-state result of the second half as the steady-state result, and the final result is also based on the steady state.

[0039] The present invention also proposes a single-phase grounding fault intelligent line selection and positioning system for a small current grounding system, including an electrical quantity acquisition module, a transient positioning module, a transient discrimination module, a grounding steady-state discrimination module, and a fault positioning module:

[0040] The electrical quantity acquisition module, after a single-phase grounding fault occurs in the small current grounding system, acquires the zero-sequence voltage and zero-sequence current at the positions of each detection point.

[0041] The transient positioning module judges the transient process based on the ratio of the true effective value increase and the fundamental wave increase ratio of the zero-sequence voltage and zero-sequence current, and selects the optimal transient according to the score and ranking of the transient process for transient positioning.

[0042] The transient discrimination module uses the characteristic vectors of each detection point where a transient process appears as input signals, and performs transient discrimination through a trained transient discrimination neural network to judge whether each detection point is inside or outside the fault boundary.

[0043] The grounding steady-state discrimination module excludes abnormal data, calculates the ratio of the fundamental wave effective value increase of the subsequent cycles starting from the second half cycle after the transient to the normal steady state, and judges whether each detection point is in the grounding steady state.

[0044] The fault positioning module, for the detection points in the grounding steady state, judges whether the detection point is inside or outside the fault boundary according to the phase difference between the zero-sequence voltage and the zero-sequence current; calculates the in-bound and out-of-bound ratio in all the cycles of the grounding steady state, and combines with the transient result to obtain the final result of whether the detection point is inside or outside the fault boundary.

[0045] The beneficial effects of the present invention are as follows, compared with the prior art:

[0046] 1. Introduce a global optimal transient positioning method, so that more representative and more utilizable transients can be found when facing more complex transient situations.

[0047] 2. The transient discrimination uses a convolutional neural network, which is more accurate in discriminating the mutation direction. In theory, when the model has sufficient learning ability and sufficient data volume, the performance can be continuously improved. At the same time, for the transient discrimination, in addition to inside / outside the boundary, "none" is introduced, making the discrimination result more scientific and accurate.

[0048] 3. By comprehensively using transients and steady states, the situation of missed judgment inside the boundary when the system type is unknown is reduced, and the discrimination reliability is higher.

[0049] 4. The part of the phase difference in the group amplitude comparison and phase comparison is optimized, making the discrimination of inside / outside the boundary of the grounding steady state more perfect.

[0050] 5. In the strategy of comprehensively using transients and steady states, the complex steady state situation is also analyzed separately, so that the analysis results of waveforms with more complex situations are closer to the actual situation.

[0051] 6. In the transient and steady state recognition part, an operation to identify abnormal data in units of a single cycle is added. On the one hand, it not only reduces the interference of abnormal data on the recognition result, but also improves the utilization rate of waveforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is the overall step flow chart of the present invention;

[0053] Figure 2 is the method flow chart of the present invention;

[0054] Figure 3 is the schematic diagram of different mutation times out of sync in the present invention;

[0055] Figure 4 is the schematic diagram of the waveform slowly increasing in the present invention;

[0056] Figure 5 is the network structure diagram of the convolutional neural network in the present invention;

[0057] Figure 6 is the schematic diagram of a data missing sample in the present invention;

[0058] Figure 7 is the schematic diagram of waveforms corresponding to different harmonic ratios in the present invention;

[0059] Figure 8 is the schematic diagram of the comparison between rising and falling cycles and normal cycles in the present invention;

[0060] Figure 9 is the schematic diagram of a complex steady state example in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0061] To make the objectives, technical solutions and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only a part of the embodiments of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0062] The present invention combines the transient first half-wave method and the group amplitude ratio and phase comparison method, and adds a filtering mechanism for abnormal data, comprehensively utilizing the transient and steady states of the entire waveform. For the transient part: a method for locating the globally optimal transient is added, and a convolutional neural network is used to discriminate the direction of transient mutation; for the steady state part: the content of steady state recognition, discrimination of in / out of the grounding steady state boundary, and a strategy for comprehensively utilizing the transient and steady states are added.

[0063] The present invention proposes an intelligent line selection and location method for single-phase grounding faults in a small current grounding system, as Figure 1 shown, and gives the overall step flow chart of the present invention. The overall steps of the present invention are as follows:

[0064] 1. Set the initial threshold and parameters, and load the neural network model.

[0065] 2. Read the oscillogram file, and read / generate the correct zero-sequence voltage and zero-sequence current. When there are U0 and I0 channels with good quality in the oscillogram file, directly read them; otherwise, synthesize the zero-sequence from the phase voltage and current for subsequent steps.

[0066] 3. Starting from the second cycle of the zero-sequence, traverse with half of the cycle as the step size and intercept a single cycle.

[0067] 4. Check whether the intercepted cycle belongs to abnormal data.

[0068] 5. If it is not abnormal data, calculate the true effective values of its zero-sequence voltage and zero-sequence current with the previous cycle, and calculate the increase ratio of the true effective values.

[0069] 6. Compare the increase ratio with the threshold (default is 0.08). If it is larger than the threshold, continue to compare the fundamental wave effective value increase ratios of the cycles before and after the transient, and compare this ratio with the threshold (default is also 0.08). If the ratio exceeds the threshold, include this cycle in the first candidate set; otherwise, include it in the second candidate set. At the same time, take the minimum of the ratios of the zero-sequence voltage and zero-sequence current as the score of this transient in the same candidate set.

[0070] 7. Repeat steps 4-6 above until the traversal is complete and two candidate sets and corresponding scores are obtained. If both candidate sets are empty, filter the recording directly; if there is a cycle in the first candidate set, select the one with the highest score; if the first candidate set is empty and the second candidate set has a cycle, select the one with the highest score from the second candidate set in the same way.

[0071] 8. Interpolate the selected transient cycle and change the length to 256 points. Then calculate FFT on it, splice the interpolated data and FFT results into 4 channels (the channel order is: U0, I0, U0_FFT, I0_FFT), and the data format with a length of 256, and put it into the neural network model for analysis to obtain the transient results.

[0072] 9. First, assume that the normal steady state is the cycle before the transient state. At the same time, check the effective values ​​of the fundamental waves of the first two cycles of the entire recording. Select the one with the largest effective value of the fundamental wave from the three cycles as the normal steady state.

[0073] 10. Start from the position where the transient cycle lags behind by half a cycle, and start traversing with a step size of half a cycle (the first cycle will overlap with the transient by half a cycle).

[0074] 11. Take a single cycle, calculate the sum of the proportions of its first five harmonics, and compare it with the threshold (default 5). If it is less than or equal to the threshold, continue to detect whether there is data missing and rising or falling trends. If there is none, proceed to the next step, otherwise repeat this step.

[0075] 12. Calculate the fundamental effective value, calculate its increase ratio to the normal steady state, and compare the ratio with the threshold (default 0.08). If it is less than the threshold, jump directly to step 15. If it is greater than the threshold, continue to calculate the ratio of the fundamental effective value to the true effective value. Only when the ratio of the zero-sequence voltage and the zero-sequence current is greater than the threshold (default 2), proceed to the next step. Otherwise, continue to detect the next cycle from step 11.

[0076] 13. Determine whether the grounding steady state is inside or outside the boundary and save it.

[0077] 14. Repeat 11-13 until the traversal is complete.

[0078] 15. If the number of cycles of the grounding steady state is less than or equal to the threshold (default 2), it is considered to be an instantaneous grounding or disturbance and is directly pushed out of the boundary. If it is greater than the threshold, proceed to the next step.

[0079] 16. Detect the proportion of inside and outside the boundary in the grounding steady state, select the larger proportion and compare it with the threshold (default 0.7): If it is greater than or equal to the threshold, determine the steady-state result as the corresponding larger proportion and continue with step 17; but if it is smaller than the threshold, continue with step 18.

[0080] 17. Compare the transient state and the first power frequency cycle of the ground fault steady state. If they are connected, comprehensively utilize the transient state and the steady state. If it is within the transient boundary, directly push it inside the boundary; if the transient state is outside the boundary, push it outside the boundary; if there is no transient state, use the steady state as the criterion.

[0081] 18. Divide the ground fault steady state into two equal segments, and output the second half of the ground fault steady state as the final result.

[0082] 19. Output the line selection (the line with in-boundary detection points) and positioning (between the in-boundary and out-of-boundary detection points) results according to the in-boundary or out-of-boundary situation. It should be noted that after obtaining the in-boundary detection points, the results of its upstream detection points need to be searched in the reverse direction. If all the upstream points are out-of-boundary, it indicates that there may be a misjudgment, and this in-boundary data should be changed to filtered.

[0083] Next, each step of the present invention will be described in detail, as Figure 2 shown, and the description content is as follows.

[0084] First, read the zero-sequence voltage U0 and zero-sequence current I0 in the oscillogram file; starting from the second power frequency cycle, calculate the ratio of the true effective value increase of the current cycle to the previous cycle in the zero-sequence voltage U0 and zero-sequence current I0. If it is greater than the true effective value increase ratio threshold, put it into the transient candidate set; calculate the ratio of the fundamental wave increase of the two adjacent power frequency cycles before and after the transient in the transient candidate set. If it is greater than the fundamental wave increase ratio, put the power frequency cycle into the first candidate set, otherwise put it into the second candidate set; based on the first candidate set and the second candidate set, select the optimal transient, and preprocess the power frequency cycles of the optimal transient; input the processed power frequency cycles into a convolutional neural network for three-classification, and the output results include in-boundary, out-of-boundary, and none;

[0085] Specifically, the idea of globally searching for the optimal transient is adopted to locate the transient: if the zero-sequence voltage U0 and zero-sequence current I0 cannot be used, the zero-sequence voltage U0 and zero-sequence current I0 are synthesized by the phase voltage and phase current;

[0086] Starting from the second power frequency cycle, calculate the ratio of the true effective value increase of the current cycle to the previous cycle in the zero-sequence voltage U0 and zero-sequence current I0. The calculation formula is:

[0087]

[0088] where Z represents the ratio of the true effective value increase; z n represents the true effective value of the nth power frequency cycle; N represents the total number of power frequency cycles;

[0089] If the ratio of the true effective value increase is greater than the true effective value increase ratio threshold, put it into the transient candidate set, and calculate the ratio of the fundamental wave increase of the two adjacent power frequency cycles before and after the transient. The calculation formula is:

[0090]

[0091] Among them, B represents the proportion of the fundamental wave amplitude increase; B n represents the effective value of the fundamental wave in the nth cycle;

[0092] If the proportion of the fundamental wave amplitude increase exceeds the fundamental wave amplitude increase threshold, it indicates that the subsequent steady state of the transient also has an upward characteristic, and the possibility of it being non-disturbed is greater. Then, the cycle is put into the first candidate set; otherwise, it is put into the second candidate set. After the traversal is completed, the optimal transient is selected from the first candidate set. If the first candidate set is empty, it is selected from the second candidate set. If both are empty, it means there is no transient. When there are multiple candidate transients in the same candidate set, a scoring ranking is introduced to select the optimal transient. The score needs to be able to represent the comprehensive transient amplitude increase of the zero-sequence voltage and zero-sequence current. The larger the comprehensive amplitude increase, the more obvious the transient characteristics are. The scoring ranking uses the short-board effect: each candidate transient includes the proportion of the true effective value increase corresponding to the zero-sequence voltage U0 and the zero-sequence current I0, and the smaller value between the two is used as the score.

[0093] Furthermore, after selecting the interval where the optimal transient is located, the transient needs to be discriminated. The discrimination of the transient in the present invention not only includes inside the boundary and outside the boundary, but also "none". Although the transient characteristics of some waveforms are very obvious, it is difficult for humans to distinguish whether the transient mutation is reversed. For example, the mutation times of the zero-sequence voltage and zero-sequence current of the transient are not synchronized (see Figure 3 ), the waveform increases slowly (see Figure 4 ), etc. Therefore, the transient of such waveforms should output "none" (meaning that the transient cannot be distinguished inside or outside the boundary) to prevent interfering with the overall judgment result. In the part of transient result discrimination, the present invention innovatively introduces a convolutional neural network model in deep learning to complete the task of three-class classification (the three classes are: inside the boundary, outside the boundary, and none). After preprocessing the selected transient cycle (the preprocessing is mainly to meet the input format requirements of the convolutional neural network), it is input into the convolutional neural network model, and the output is the three-class classification result. The one with the highest probability is selected as the transient discrimination result.

[0094] The specific steps for transient discrimination are as follows: preprocess the cycle of the optimal transient, and the preprocessing includes resampling, normalization, and discrete Fourier transform (FFT); the resampling sets the number of nodes in a single cycle to 256; the normalization controls the thresholds of voltage and current within the range of [0, 1]; the FFT is used to extract features of different frequencies; the resampling is used to unify the frequency of the cycle, and linear interpolation is used to supplement N points between every two adjacent points for upsampling, or one point is taken every N points for downsampling, where N represents the value obtained by rounding the quotient of the expected frequency and the actual frequency; the default frequency is set to 12,800 Hz. After unifying the frequency, there may still be a problem of inconsistent number of sampling points. When there is a situation where the specific actual frequency and the default frequency are not in an integer multiple relationship, the value of the last point of the cycle is used for deletion and filling. This operation method is because resampling is only used for the transient part, and the mutation of the transient generally occurs in the front or middle of the transient cycle. In order not to affect transient discrimination by adding or deleting from the end, and at the same time to avoid generating new mutations and false judgments when adding nodes, the straight line is filled directly according to the value of the last point.

[0095] The network structure diagram of the convolutional neural network adopted is as Figure 5 shown. The network structure adopted by the convolutional neural network is as follows:

[0096] Input layer: The input format is [n, 4, 256], where n represents the number of input samples, 4 represents four channels, and the channel order is: zero-sequence voltage, zero-sequence current, zero-sequence voltage FFT result, and zero-sequence current FFT result. 256 represents the number of nodes in a cycle (if the original number of cycle nodes is inconsistent, resampling is used for preprocessing):

[0097] First convolutional layer: one-dimensional convolution Conv1d, batch normalization BN1d, activation function ReLu, one-dimensional max pooling MaxPool1d, linear transformation Linear in the fully connected layer, activation function softmax;

[0098] The second convolutional layer and the third convolutional layer have the same structure as the first convolutional layer;

[0099] Output layer: The output is [a, b, c], representing the probability values of [none, within the boundary, outside the boundary] respectively, and the one with the largest probability among the three is selected as the output result.

[0100] Secondly, a steady-state analysis is performed. The steady state is divided into three categories: normal steady state, grounded steady state, and normal steady state restored after grounding. Abnormal data is excluded by detecting the interpolation of adjacent points of zero-sequence current, harmonic ratio, cycle rise and fall trend, and sine wave characteristics; the fundamental effective value of the first two unfiltered cycles before the transient and the complete waveform is compared, and the largest of the three is taken as the normal steady state; starting from the latter half cycle after the transient, the increase ratio of the fundamental effective value of the subsequent cycles to the normal steady state is calculated to determine whether it belongs to the grounded steady state or the normal steady state restored after grounding; the phase difference between the zero-sequence voltage phase and the zero-sequence current phase of each cycle in the grounded steady state is calculated, and when the phase difference is in [-π, 0.4], it is judged as outside the boundary, otherwise it is judged as inside the boundary.

[0101] Among them, the normal steady state refers to the stable state in the normal state before grounding; the grounded steady state refers to the stable state after grounding and before the fault is eliminated; the normal steady state restored after grounding refers to the stable state after the fault is eliminated after grounding. The normal steady state and the normal steady state restored after grounding should theoretically be the same and do not necessarily have the periodic characteristics of a sine wave, and may also be a straight line (because theoretically I0 should be 0 when there is no fault); the fundamental effective value of the grounded steady state will be greater than the two, and it must have the periodic property of a sine wave to facilitate the analysis of the inside and outside of the boundary through the phase difference.

[0102] Among the three types of steady states, only the grounded steady state is used for boundary inside / outside discrimination, but the other two types of steady states have an auxiliary role in identifying the grounded steady state, so all three types of steady states need to be identified in the algorithm. During the process of identifying the three types of steady states, the interference of abnormal data needs to be excluded, and the present invention excludes abnormal data through abnormal filtering. Abnormal data includes data loss and non-steady states;

[0103] The specific steps for determining data loss are as follows. Data loss is mainly manifested as a sudden large increase / decrease straight line replacing the original normal stable state in the middle of a stable waveform, as Figure 6 shown. The identification of data loss is relatively simple. By using the characteristic that the zero-sequence current value is small, it can be detected whether the interpolation of two adjacent points of the zero-sequence current in a cycle exceeds a specified large threshold, and it can be considered as data loss. This type of abnormality needs to be detected and abnormal cycles excluded both in the transient and steady states.

[0104] Detect whether the interpolation of two adjacent points of the zero-sequence current in a cycle exceeds the specified threshold. If so, it is considered as data loss; among them, the specified threshold is set to 1000.

[0105] The specific steps for determining the non - steady state are the distinctness of the fundamental wave characteristics, the presence or absence of an upward / downward trend, and the presence of sine - wave characteristics. When identifying the grounded steady state, all three items need to be detected to exclude anomalies because the grounded steady state uses the sine - wave phase information of the fundamental wave to judge inside and outside the boundary, and the grounded steady state with an upward / downward trend will also lead to incorrect phase identification; when identifying the normal steady state and the normal steady state restored after grounding, since the identification of these two types of steady states only depends on the magnitude of the fundamental - wave effective value and has nothing to do with the sine - wave characteristics of the fundamental wave, and the presence of an upward / downward trend may cause these two types of steady states to be misidentified as the grounded steady state, so only the second item needs to be detected.

[0106] The specific steps for judging whether the fundamental - wave characteristics are distinct (see Figure 7 ) are as follows: sort the harmonic ratios from large to small, compare the sum of the harmonic ratios of the top 5 with the harmonic - ratio threshold. The harmonic - ratio threshold is set to 5. When the sum of the harmonic ratios is greater than the harmonic - ratio threshold, the cycle is filtered; where the formula for the harmonic ratio is:

[0107]

[0108] where γ represents the harmonic ratio; X m represents the effective value of the m - th harmonic; base represents the fundamental - wave effective value; M represents the order of the harmonic.

[0109] The specific steps for judging whether there is an upward / downward trend are related to the detection of the upward / downward trend, the first and last nodes of the cycle, and the average value of the height differences between all adjacent nodes. For a sine cycle, the heights of the starting point and the ending point should be the same. If the offset between them is too large, it will cause the overall waveform to have the characteristics of rising and falling, as shown in Figure 8 . However, it is still inappropriate to rely solely on the difference between the last point and the first point as the criterion for identifying whether the waveform has the characteristics of rising and falling because the thresholds of different waveforms are different. For example, the waveform threshold of zero - sequence voltage is much larger than that of zero - sequence current. Therefore, the average value of the absolute values of the height differences between all adjacent nodes is introduced to standardize the difference. The specific formula is as follows:

[0110]

[0111] where Y represents the rising - falling trend; Y d represents the height value of the d - th point in the cycle; D represents the total number of points.

[0112] The specific steps to determine whether it has a sine wave feature are as follows. Whether it has a sine wave feature mainly detects whether the fundamental sine wave feature is obvious. The obvious degree of the fundamental wave feature has been identified before, but it is only compared with the harmonics. When the waveform of a partial cycle is almost a straight line, without obvious phase features, both the fundamental wave and the harmonics are very small and the harmonic ratio is not large, for the said cycle, the obvious degree of the fundamental wave is measured by calculating the ratio of the fundamental wave effective value to the true effective value. If the ratio is less than the ratio threshold, the said cycle is filtered. The ratio threshold is set to 2. The calculation formula of the ratio is as follows:

[0113]

[0114] Where, λ represents the said ratio; z0 represents the true effective value; base0 represents the fundamental wave effective value.

[0115] Furthermore, the specific steps of steady-state identification are as follows. While filtering out abnormal data, it is also necessary to distinguish three types of steady states. First is the determination of normal steady state. The normal steady state must be before the transient. One way is to regard the nearest unfiltered cycle before the transient as the normal steady state. However, considering that the transient is found through global search for the optimal transient, in the case of multiple transients, the last unfiltered cycle before the finally determined transient is very likely not in the normal state but has already changed partially. Therefore, another method is to use the first few unfiltered cycles of the waveform as the normal steady state. Combining the two methods, the present invention selects the one with the largest fundamental wave effective value among the fundamental wave effective values of the last unfiltered cycle before the transient and the first two unfiltered cycles of the overall waveform by calculating and comparing them as the normal steady state.

[0116] After determining the said normal steady state, starting from the second half cycle after the transient (because the transient time is short, and the transient cycle found often contains the waveform of half of the subsequent steady state), gradually test whether the subsequent cycles are abnormal. If there is no abnormality, calculate the ratio of the increase in the fundamental wave effective value of the said subsequent cycle to the normal steady state. If it exceeds the threshold of 0.08, the said subsequent cycle is considered as the grounded steady state, otherwise it is the normal steady state after grounding recovery. Once the normal steady state after grounding recovery is identified or all the said subsequent cycles are traversed, the steady-state identification stops.

[0117] Specifically, the specific steps for discrimination inside / outside the grounded steady state are as follows. Calculate the phase difference between the zero-sequence voltage phase and the zero-sequence current phase of each cycle in the grounded steady state. The range of the phase difference is between [-π, π]. Positive represents that the current lags, and negative represents that the current leads. When the phase difference is between [-π, 0.4], it is judged as outside the boundary, otherwise it is judged as inside the boundary.

[0118] Finally, use the transient-steady state comprehensive strategy, such as Figure 1As shown in "Comprehensive Strategy: Steps 15-18", calculate the proportion of inside and outside the boundary in all cycles of the grounding steady state, and compare it with the proportion ratio threshold; if it is greater than the proportion ratio threshold, combine the transient results to determine whether there is a connection with the first cycle of the grounding steady state to obtain the final result of inside or outside the boundary; if it is not greater than the proportion ratio threshold, the final result is based on the steady state.

[0119] Specifically, distinguish according to the length of the grounding steady state. Within two cycles of the grounding steady state, it is considered that it is very likely to return to normal after instantaneous grounding. Regardless of the transient and steady state results, the final result pushed should be outside the boundary.

[0120] When the grounding steady state is greater than two cycles, the result of the grounding steady state has reference value. Calculate the proportion ratio of inside the boundary and the proportion ratio of outside the boundary in all cycles of the grounding steady state. If the proportion ratio of inside the boundary or the proportion ratio of outside the boundary is greater than the proportion ratio threshold of 0.7, the steady state result shall prevail. Then consider the transient. If the transient is connected with the first cycle of the grounding steady state, comprehensively consider the transient and steady state results (the comprehensive consideration method is shown in Table 1); if they are not connected together, the final result is based on the steady state result.

[0121] Table 1 Comprehensive consideration plan for transient and steady state

[0122] Situation Transient Steady state Final result 1 Inside the boundary Inside the boundary Inside the boundary 2 Outside the boundary Outside the boundary Outside the boundary 3 Inside the boundary Outside the boundary Inside the boundary 4 Outside the boundary Inside the boundary Inside the boundary 5 None Inside the boundary Inside the boundary 6 None Outside the boundary Outside the boundary

[0123] Situations 1, 2, 5, and 6 in Table 1 are relatively simple. When the transient and steady state results are the same in situations 1 and 2, they are directly output. When situations 5 and 6 occur, the steady state shall prevail. However, when the transient and steady state results are opposite, for example: situation 3 is the case where the system model is a neutral point grounded through an arc suppression coil system, which is classified as inside the boundary; situation 4 is caused by the reverse installation of the voltage direction and current direction acquisition devices of the equipment, so this type of waveform should be inside the boundary.

[0124] If both the proportion ratio of inside the boundary and the proportion ratio of outside the boundary do not reach the threshold, it is very likely that it is outside the boundary at the beginning and then becomes inside the boundary, etc. See Figure 9 (U0 above and I0 below. At the beginning, both the transient and the grounding steady state are outside the boundary, and then the grounding steady state changes to inside the boundary). At this time, the transient loses its meaning, and the steady state should also be based on the subsequent one. Therefore, the grounding steady state result of the second half is used as the steady state result, and the final result is also based on the steady state.

[0125] The present invention also proposes a single-phase grounding fault intelligent line selection and positioning system for a small current grounding system, including an electrical quantity acquisition module, a transient positioning module, a transient discrimination module, a grounding steady state discrimination module, and a fault positioning module:

[0126] The electrical quantity acquisition module collects the zero-sequence voltage and zero-sequence current at the positions of each detection point after a single-phase grounding fault occurs in the small-current grounding system;

[0127] The transient positioning module judges the transient process based on the ratio of the true effective value increase and the fundamental wave increase ratio of the zero-sequence voltage and zero-sequence current, and selects the optimal transient according to the score and ranking of the transient process for transient positioning;

[0128] The transient discrimination module uses the characteristic vectors of each detection point where a transient process occurs as input signals, and performs transient discrimination through a trained transient discrimination neural network to judge whether each detection point is inside or outside the fault boundary;

[0129] The grounding steady-state discrimination module excludes abnormal data, calculates the ratio of the fundamental wave effective value increase of the subsequent cycles starting from the second half cycle after the transient to the normal steady state, and judges whether each detection point is in the grounding steady state;

[0130] The fault location module, for the detection points in the grounding steady state, judges whether the detection point is inside or outside the fault boundary according to the phase difference between the zero-sequence voltage and the zero-sequence current; calculates the ratio of the inside and outside the boundary in all the cycles of the grounding steady state, and combines the transient results to obtain the final result of whether the detection point is inside or outside the fault boundary. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: still can modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement without departing from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. An intelligent line selection and location method for single-phase grounding fault in a small current grounding system, characterized in that: include: After a single-phase grounding fault occurs in a small current grounding system, the zero-sequence voltage and zero-sequence current at each detection point are collected; The transient process is judged based on the true effective value increase ratio and fundamental wave increase ratio of zero-sequence voltage and zero-sequence current, and the optimal transient is selected according to the score and ranking of the transient process for transient location; Taking the characteristic vector of each detection point where the transient process occurs as the input signal, the transient discrimination is performed through the trained transient discrimination neural network to determine whether each detection point is inside or outside the fault boundary; Exclude abnormal data, calculate the increase ratio of the subsequent cycles starting from the second half of the transient state to the fundamental wave effective value of the normal steady state, and determine whether each detection point is in the grounding steady state; For the detection point in the grounding steady state, the phase difference between the zero-sequence voltage and the zero-sequence current is used to determine whether the detection point is inside or outside the fault boundary. The ratio of the proportion of the internal and external faults in all the grounding steady-state cycles is calculated, and combined with the transient results, the final result of whether the detection point is located inside or outside the fault boundary is obtained.

2. The intelligent line selection and location method for single-phase grounding fault in a small current grounding system according to claim 1 is characterized in that: The score of the transient process includes the true effective value increase ratio of the zero-sequence voltage of the candidate transient and the true effective value increase ratio of the zero-sequence current of the candidate transient; for each candidate transient, the smaller value of the true effective value increase ratio of the zero-sequence voltage and the zero-sequence current is used as the final score of the candidate transient, and the candidate transient with the largest final score is used as the optimal transient, thereby obtaining transient positioning.

3. The intelligent line selection and location method for single-phase grounding fault in a small current grounding system according to claim 2 is characterized in that: If the fundamental wave increase ratio exceeds the fundamental wave increase ratio threshold, the cycle is placed in the first candidate set, otherwise it is placed in the second candidate set; after the traversal is completed, the optimal transient is selected from the first candidate set, if the first candidate set is empty, it is selected from the second candidate set; when there are multiple candidate transients in the same candidate set, the true effective value increase ratio of the zero-sequence voltage and zero-sequence current of each candidate transient is calculated, the smaller value between the two is extracted and sorted in ascending order, and the candidate transient corresponding to the largest sort is taken as the optimal transient.

4. The intelligent line selection and location method for single-phase grounding fault in a small current grounding system according to claim 1 is characterized in that: The network structure adopted by the convolutional neural network is: Input layer: The input format is [n,4,256], where n represents the number of input samples, 4 represents the order of four channels: zero-sequence voltage, zero-sequence current, zero-sequence voltage FFT result and zero-sequence current FFT result, and 256 represents the number of nodes in one cycle; The first convolutional layer: one-dimensional convolution Conv1d, batch normalization BN1d, activation function ReLu, one-dimensional maximum pooling MaxPool1d, linear transformation Linear in the fully connected layer, activation function softmax; The second convolutional layer and the third convolutional layer have the same structure as the first convolutional layer; Output layer: Output [a, b, c], representing the probability values ​​of [none, within bounds, outside bounds] respectively, and select the one with the highest probability as the output result.

5. The intelligent line selection and location method for single-phase grounding fault in a small current grounding system according to claim 1 is characterized in that: Abnormal data include non-steady states; The determination of the unstable state includes whether the fundamental wave characteristics are obvious, whether there is an upward / downward trend, and whether it has a sine wave characteristic; the determination method of whether the fundamental wave characteristics are obvious is to sort the harmonic proportions from large to small, compare the sum of the top 5 with the harmonic proportion threshold, and filter the cycle when the sum of the harmonic proportions is greater than the harmonic proportion threshold; wherein, the calculation formula of the harmonic proportion is: Wherein, γ represents the harmonic proportion; X m It represents the effective value of the mth harmonic; base represents the effective value of the fundamental wave; M represents the order of the harmonic.

6. The intelligent line selection and location method for single-phase grounding fault in a small current grounding system according to claim 5 is characterized in that: The method to determine whether there is an upward / downward trend is to introduce the absolute mean of the height differences of all adjacent nodes to standardize the differences. The specific formula is as follows: Among them, Y represents the upward and downward trend; Y d represents the height value of the dth point in the cycle; D represents the total number of points.

7. The intelligent line selection and location method and system for single-phase grounding fault in a small current grounding system according to claim 5 is characterized in that: Whether it has sine wave characteristics: The prominence of the fundamental wave is measured by calculating the proportion of the fundamental wave effective value in the true effective value. If the proportion is less than the proportion threshold, the cycle is filtered. The proportion threshold is set to 2. The calculation formula of the proportion is as follows: Wherein, λ represents the proportion; z0 represents the true effective value; and base0 represents the fundamental effective value.

8. The intelligent line selection and location method for single-phase grounding fault in a small current grounding system according to claim 1 is characterized in that: Calculate and compare the fundamental effective value of the unfiltered cycle before the transient state and the first two unfiltered cycles of the overall waveform, and select the one with the largest fundamental effective value among the three as the normal steady state; After determining the normal steady state, starting from the second half of the transient state, the increase ratio of the fundamental effective value of the subsequent cycle to the normal steady state is calculated. If it exceeds the threshold, it is considered that the subsequent cycle is a grounding steady state, otherwise it is considered to be a normal steady state after grounding; When the grounding is detected and the normal steady state is restored or all the subsequent cycles are traversed, the steady state identification is stopped; The phase difference between the zero-sequence voltage phase and the zero-sequence current phase of each cycle in the grounded steady state is calculated, and the range of the phase difference is between [-π, π]; when the phase difference is between [-π, 0.4], it is judged to be out of bounds, otherwise it is judged to be in bounds.

9. The intelligent line selection and location method for single-phase grounding fault in a small current grounding system according to claim 1 is characterized in that: According to the length of the grounding steady state, it is distinguished. Within two cycles of the grounding steady state, it is considered that it is very likely to be a transient grounding and then return to normal. Regardless of the results of the transient state and the steady state, the final result pushed should be the final result judged as out of bounds; When the grounding steady state is greater than two cycles, the in-boundary proportion ratio and the out-boundary proportion ratio of all grounding steady state cycles are calculated. If the in-boundary proportion ratio or the out-boundary proportion ratio is greater than the proportion ratio threshold, the corresponding steady state result is in-boundary or out-boundary. Combined with the transient result, if the transient and the first grounding steady state cycle are connected together, the transient steady state result is comprehensively considered. If they are not connected together, the final result shall be based on the steady-state result; If both the in-bounds ratio and the out-bounds ratio do not reach the threshold, the grounding steady-state result in the second half is taken as the steady-state result, and the final result is also based on the steady-state result.

10. An intelligent line selection and positioning system for single-phase grounding fault in a small current grounding system based on the method according to any one of claims 1 to 9, comprising an electrical quantity acquisition module, a transient positioning module, a transient discrimination module, a grounding steady-state discrimination module and a fault positioning module, characterized in that: The electrical quantity acquisition module collects the zero-sequence voltage and zero-sequence current at each detection point after a single-phase grounding fault occurs in a small current grounding system; The transient positioning module determines the transient process based on the true effective value increase ratio and fundamental wave increase ratio of the zero-sequence voltage and zero-sequence current, and selects the optimal transient according to the score and ranking of the transient process for transient positioning; The transient discrimination module uses the feature vector of each detection point where the transient process occurs as the input signal, and performs transient discrimination through the trained transient discrimination neural network to determine whether each detection point is inside or outside the fault boundary; The grounding steady-state identification module excludes abnormal data, calculates the increase ratio of the subsequent cycles starting from the second half of the transient state to the fundamental effective value of the normal steady state, and determines whether each detection point is in the grounding steady state; The fault location module determines whether the detection point is inside or outside the fault boundary according to the phase difference between the zero-sequence voltage and the zero-sequence current for the detection point in the grounding steady state; calculates the proportion of the inside and outside of the boundary in all the cycles of the grounding steady state, and combines the transient results to obtain the final result of whether the detection point is inside or outside the fault boundary.

Images