Ground fault directionality detection method, device, and computer-readable storage medium

Abstract

A ground fault directionality detection method, a ground fault directionality detection device, and a computer-readable storage medium for use in a small current grounding system are disclosed. The ground fault directionality detection method includes sampling at least a portion of a current signal to obtain a numerical matrix, providing the numerical matrix as input to a ground fault directionality detection neural network, computing, by the ground fault directionality detection neural network, a label vector corresponding to the numerical matrix, the label vector indicating whether the current signal is an upstream fault current signal or a downstream fault current signal, receiving, by the ground fault directionality detection neural network, a ground fault confirmation signal, and outputting, by the ground fault directionality detection neural network, the label vector in a case where the ground fault confirmation signal indicates that the current signal is a ground fault current signal.

Description

Technical Field

[0001] This disclosure relates to methods, devices, and computer-readable storage media for detecting the directionality of ground faults, and more specifically, to methods for detecting the directionality of ground faults in low-current grounding systems. Background Technology

[0002] In a low-current grounding system, when a ground fault occurs on a line, the power equipment on the line (such as ring main units) can operate with the fault for a period of time (usually no more than 2 hours). During this period, it is necessary to diagnose the direction of the ground fault and determine whether the ground fault is a downstream fault (i.e., an intra-zone fault) or an upstream fault (i.e., an extra-zone fault) so that the corresponding maintenance team can be notified to carry out maintenance. Summary of the Invention

[0003] This disclosure relates to a method and apparatus for ground fault directionality detection based on neural networks, requiring only current information and not voltage information.

[0004] According to one aspect of this disclosure, a method for detecting the directionality of a ground fault is disclosed. The method includes: sampling at least a portion of a current signal to obtain a numerical matrix; providing the numerical matrix as input to a ground fault directionality detection neural network; calculating a label vector corresponding to the numerical matrix by the ground fault directionality detection neural network, the label vector indicating whether the current signal is an upstream fault current signal or a downstream fault current signal; receiving a ground fault confirmation signal by the ground fault directionality detection neural network; and outputting the label vector by the ground fault directionality detection neural network when the ground fault confirmation signal indicates that the current signal is a ground fault current signal.

[0005] Optionally, the method further includes: detecting whether the maximum value in the numerical matrix is ​​greater than a preset start-up threshold; and providing the numerical matrix as input to the ground fault directionality detection neural network only when the maximum value in the numerical matrix is ​​greater than or equal to the preset start-up threshold.

[0006] Optionally, the current signal is a three-phase current signal, and a row or column of the numerical matrix corresponds to one of the three-phase current signals.

[0007] Optionally, the ground fault directionality detection neural network is a convolutional neural network.

[0008] Optionally, the ground fault directionality detection neural network is a trained neural network, and the training includes the following steps: Step 1, obtaining a ground fault current signal training sample set for training the ground fault directionality detection neural network; Step 2, providing the training samples in the ground fault current signal training sample set as input to the ground fault directionality detection neural network; Step 3, calculating a label vector corresponding to the training sample by the ground fault directionality detection neural network; Step 4, determining the processing loss of the ground fault directionality detection neural network based on the label vector of the training sample; Step 5: if the processing loss is greater than or equal to a preset processing loss threshold, updating the parameters of the ground fault directionality detection neural network and performing steps 2 to 5 based on the updated ground fault directionality detection neural network, and stopping the training if the processing loss is less than or equal to the preset processing loss threshold.

[0009] Optionally, the ground fault directionality detection neural network includes at least a convolutional layer, a first fully connected layer, a second fully connected layer, an activation layer, and an output layer; and the parameters of the ground fault directionality detection neural network include at least the weight matrix of the convolutional kernels included in the convolutional layer, the first weight matrix included in the first fully connected layer, and the second weight matrix included in the second fully connected layer.

[0010] Optionally, the ground fault directionality detection neural network may further include a pooling layer.

[0011] Optionally, the convolutional layer uses a 5×5 convolutional kernel; and the activation function is one of the sigmoid function, the relu function, and the tanh function.

[0012] Optionally, one or more training samples in the ground fault current signal training sample set are generated by the electromagnetic transient simulation program EMTP.

[0013] According to another aspect of this disclosure, a ground fault directionality detection device is provided. The device includes: a sampling module for sampling at least a portion of a current signal to obtain a numerical matrix; an input module for providing the numerical matrix as input to a ground fault directionality detection neural network; and a classification module configured with the ground fault directionality detection neural network, which calculates a label vector corresponding to the numerical matrix, the label vector indicating whether the current signal is an upstream fault current signal or a downstream fault current signal. The ground fault directionality detection neural network also receives a ground fault confirmation signal and outputs the label vector if the ground fault confirmation signal indicates that the current signal is a ground fault current signal.

[0014] Optionally, the input module is further configured to detect whether the maximum value in the numerical matrix is ​​greater than a preset start-up threshold, and to provide the numerical matrix as input to the ground fault directionality detection neural network only when the maximum value in the numerical matrix is ​​greater than or equal to the preset start-up threshold.

[0015] According to another aspect of this disclosure, a ground fault directionality detection device is provided. The device includes: a processor; and a memory, wherein the memory stores a computer-executable program that, when executed by the processor, causes the device to perform the method as described in any of the preceding claims.

[0016] According to another aspect of this disclosure, a computer-readable storage medium is provided that stores computer instructions thereon, which, when executed by a processor, perform the methods described in any of the preceding claims.

[0017] According to the above aspects of this disclosure, by using a neural network to determine the directionality of a ground fault, only current information is needed and no voltage information is required, which overcomes the disadvantage of the classical ground fault detection method that requires both voltage and current information to perform the detection. Furthermore, due to the significant advantages of neural networks in feature extraction and classification, the accuracy of ground fault directionality detection is improved. Attached Figure Description

[0018] These and / or other aspects, features, and advantages of this disclosure will become clearer and more readily understood from the following description of the disclosure in conjunction with the accompanying drawings, wherein:

[0019] Figure 1 This is a schematic diagram illustrating a usage scenario of the grounding fault directionality detection device according to an embodiment of the present disclosure;

[0020] Figure 2 This is a flowchart of a ground fault directionality detection method according to an embodiment of the present disclosure;

[0021] Figure 3 This is a flowchart of a ground fault directionality detection method according to another embodiment of the present disclosure;

[0022] Figure 4 This is a flowchart of a training method for a ground fault directionality detection neural network according to an embodiment of the present disclosure;

[0023] Figure 5 This is a schematic diagram of the structure of a ground fault directionality detection neural network according to an embodiment of the present disclosure;

[0024] Figure 6 The test results obtained by testing the ground fault directionality detection neural network according to an embodiment of the present disclosure are shown.

[0025] Figure 7 This is a block diagram illustrating a ground fault directionality detection device according to an embodiment of the present disclosure.

[0026] Figure 8 This is a hardware block diagram illustrating a ground fault directionality detection device according to another embodiment of the present disclosure.

[0027] Figure 9 This is a schematic diagram illustrating a computer-readable storage medium according to an embodiment of the present disclosure. Detailed Implementation

[0028] The present disclosure will now be described in detail with reference to exemplary embodiments thereof. However, the present disclosure is not limited to the embodiments described herein, which may be implemented in many different forms. The described embodiments are provided only to make the present disclosure thorough and complete, and to fully convey the concept of the present disclosure to those skilled in the art. Features of the various embodiments described may be combined with or substituted for each other, unless expressly excluded or should be excluded based on the context.

[0029] Currently, methods used to detect the directionality of grounding faults mainly include: The group amplitude and phase comparison method, which first selects three or more lines with the largest zero-sequence current using the "maximum value," then compares the phases of these three zero-sequence currents. The line with a phase different from the other two is the grounded line; if the phases of the zero-sequence currents in all three lines are the same, it is determined to be a busbar grounding. The fifth harmonic determination method, which works by using a neutral point grounded system with an arc suppression coil. Due to the compensation provided by the arc suppression coil, the fundamental component of the zero-sequence current in the system is very small, so line selection analysis based on the fundamental component is ineffective. The method involves several approaches. One approach is to compare the amplitude and phase of the fifth harmonic component of the zero-sequence current across all lines. Another approach is the active power determination method, which uses the principle that for a neutral-point grounded system via an arc-suppression coil, the active component of the zero-sequence current in a faulty line has the opposite polarity to that of a normal line. A third approach is the wavelet transform method, which determines the characteristic frequency band where transient currents are most concentrated based on the principle of maximum energy, and then compares the polarity of the first half-wave of the maximum modulus of each line's characteristic frequency band to determine the faulty line. A fourth approach is the first half-wave determination method, which focuses on the first half-cycle after a fault occurs. The zero-sequence current of a faulty line has the opposite polarity to the zero-continuous current of a normal line; therefore, fault location can be determined by comparing the polarity of the zero-sequence current in the first half-wave. The median resistance method works by connecting a resistor device at the neutral point of the system via a switch, or by connecting a resistor device in parallel with the arc-suppression coil through the arc-suppression coil. When a ground fault occurs, if it is a transient fault, the resistor is not engaged; if it is a permanent ground fault, the switch is closed after a delay to engage the resistor, and the active current generated by this resistor flows only through the faulty line. The pulse current method works by using a pulse current-type small-current ground fault precise location device. A controllable pulse current generator is added to the low-current fault location device. That is, a resistor device that can control and pulse current is added at the neutral point of the system. The multi-frequency admittance method is based on the principle that at the moment of ground fault occurrence, the three-phase discharge current to ground is mainly high-frequency capacitive current. According to a certain weight, the high-frequency component of the current of a specific order is extracted within this time window. Since the arc suppression coil is approximately open-circuited to the high-frequency capacitive current, the high-frequency component is selected for processing. The algorithm can refer to the vector method of ungrounded system and obtain the result based on the phase angle of zero-sequence voltage and zero-sequence current.

[0030] However, some of these methods require additional voltage sensors to obtain voltage information, some require additional resistors or resistor devices to implement, and some have poor reliability. For example, in the first half-wave determination method, it is difficult to select the time window of the waveform used for polarity comparison. If the time window is not selected properly, it may lead to completely opposite judgment results.

[0031] In view of the above, this disclosure provides a method for detecting the directionality of ground faults based on neural networks, which only requires current information of the ground fault and does not require voltage information.

[0032] Figure 1 This is a schematic diagram of a usage scenario for the ground fault directionality detection device 100 according to an embodiment of the present disclosure.

[0033] like Figure 1 The scenario shown serves as an example of a low-current grounding system, where a substation supplies power to loads A, B, and C, such as a business center or office building, via transformers. The transformers distribute voltage to feeders A, B, and C via busbars, supplying power to loads A, B, and C through these feeders. Feeder A is equipped with current transformers CT1 and CT2, and relays R1 and R2; feeder B is equipped with current transformer CT3 and relay R3; and feeder C is equipped with current transformer CT4 and relay R4. Additionally, each feeder also contains circuit breakers Q1–Q4, loads, and other equipment or devices not shown.

[0034] The ground fault directionality detection device 100 according to the embodiments of this disclosure can be integrated as part of a relay in each relay R1 to R4, or it can be integrated in other devices located outside the relay and exchanging data or commands with the relay via a wired or wireless network, or it can be located in a fault path indicator FPI deployed on the line.

[0035] When a ground fault F occurs at a certain point on the line, the ground fault directionality detection device 100 can detect the ground fault current signal I. F Sample and detect the current signal I F The directionality is determined, and a detection result is generated, which can indicate the current signal I. F Is the current signal representing an upstream fault (hereinafter referred to as the upstream fault current signal) or a downstream fault current signal (hereinafter referred to as the downstream fault current signal)? Upstream and downstream faults are distinguished according to the protection range of the protection device (e.g., relays R1-R4 or FPI). For example, for each of relays R1-R4, a fault occurring within its protection range is a downstream fault, while a fault occurring outside its protection range is an upstream fault. For example, when in feeder A... Figure 1 When a fault F occurs at the location shown, it is a downstream fault for relay R1, and an upstream fault for relays R2 to R4. For example, when fault F occurs, the ground fault directionality detection device 100 configured in relays R1 and R2 respectively detects the ground fault current signal I. FDirectional detection is performed. The ground fault directional detection device 100 in relay R1 will produce a detection result indicating a downstream fault, and the ground fault directional detection device 100 in relay R2 will produce a detection result indicating an upstream fault. Considering these detection results, it can be determined that the fault occurs within the protection range of relay R1.

[0036] Specifically, the ground fault directionality detection device 100 is equipped with a ground fault directionality detection neural network as described in the embodiments of this disclosure. Using this neural network, the ground fault current signal I can be detected. F Feature extraction processing is performed, and a detection result is generated based on the extracted features. This detection result indicates whether the ground fault belongs to an upstream or downstream fault. The ground fault directionality detection neural network needs to be pre-trained using the training method for the ground fault directionality detection neural network according to embodiments of this disclosure to be a neural network capable of performing fault detection directionality detection.

[0037] In addition, the ground fault directionality detection device 100 can also output the detection results to a display device (not shown in the figure), which can display the detection results in a user-friendly manner such as text, numbers, and images.

[0038] In the embodiments of this disclosure, since the detection results are generated by the ground fault directionality detection device 100 using a trained ground fault directionality detection neural network, the neural network overcomes the disadvantages such as the need to collect voltage information, the need to add an additional resistor, and the low reliability of the detection results by taking advantage of the significant advantages of the neural network in feature extraction and classification, thus achieving more efficient and accurate ground fault directionality detection.

[0039] Figure 2 This is a flowchart of a ground fault directionality detection method 200 according to an embodiment of the present disclosure.

[0040] Figure 2 The ground fault directionality detection method 200 shown can be derived from... Figure 1 The method is implemented using a ground fault directionality detection device 100, which is equipped with a ground fault directionality detection neural network trained according to the training method of this embodiment. Figure 2 As shown, the ground fault directionality detection method 100 may include steps S201-S205.

[0041] In step S201, at least a portion of the current signal I is sampled to obtain a numerical matrix M. The current signal I in this step can also be a pre-processed current signal, which includes, but is not limited to, converting a large ground fault current signal into a smaller current signal using a current transformer at a certain ratio. Furthermore, in this step, the sampling resolution for sampling the current signal I is k sampling points per power frequency cycle. The value of k can be selected according to actual needs, for example, k = 24, 32, 48, etc., and the sampling range is N cycles, for example, N = 1, 2, etc. The numerical matrix M in this step is an arrangement of the sampled values ​​of the ground fault current signal I, where one row or column corresponds to one phase of the current signal. For example, if the ground fault current signal I is a three-phase current, sampling N cycles will yield a numerical matrix M of size 3 × Nk, where one row of the numerical matrix M corresponds to one phase of the three-phase current signal.

[0042] In step S202, the numerical matrix M is provided as input to the ground fault directionality detection neural network. The ground fault directionality detection neural network is a pre-trained neural network, and its training method will be discussed below. Figure 4 Describe it.

[0043] In step S203, the ground fault directionality detection neural network calculates a label vector V corresponding to the numerical matrix M. This label vector V indicates whether the current signal I is an upstream or downstream fault current signal. In this step, the ground fault directionality detection neural network performs feature extraction processing on the current signal I and generates a detection result based on the extracted features. In other words, the ground fault directionality detection neural network performs a series of operations on the numerical matrix M input in step S203 to obtain the label vector V indicating whether the ground fault is an upstream or downstream fault. The category corresponding to the maximum value in the label vector V is the prediction result of the ground fault directionality detection neural network.

[0044] It is particularly important to note that steps S201 to S203 are based on the assumption that the current signal I is a ground fault current signal. That is, if the current signal I is determined to be a ground fault current signal, the tag vector V calculated in step S203 is valid; if the current signal I is determined not to be a ground fault current signal, the tag vector V calculated in step S203 is invalid. Therefore, steps S204 and S205 further combine the ground fault confirmation signal to determine whether the tag vector V should be output. This is because in low-current grounding systems, the capture opportunity for the ground fault current signal is very short; it can only be captured during a very short fault transient. If steps S201 to S203 are performed only after the current signal is determined to be a ground fault current signal, the capture opportunity for the ground fault current signal may be missed. In other words, the generation process of the ground fault confirmation signal mentioned in steps S204 and S205 below is performed synchronously with steps S201 to S203.

[0045] In step S204, a ground fault confirmation signal is generated by a ground fault directional detection neural network. The ground fault confirmation signal can be a signal calculated by other devices, apparatuses, or components (e.g., other components in a relay) according to a specific algorithm, indicating whether the current signal in steps S201 to S203 is a ground fault current signal. Determining whether a current signal is a ground fault current signal can utilize various principles well-known to those skilled in the art. For example, if the residual voltage signal continuously exceeds a threshold value for a certain duration (without the need for an additional voltage sensor), such as 500 ms, or if the continuously acquired residual current Fourier value is greater than or equal to 3 amperes compared to the value before the fault, then the current signal is considered a ground fault current signal. Since determining whether a current signal is a ground fault current signal is not the subject of this disclosure, it will not be described in detail to avoid obscuring this disclosure.

[0046] In step S205, the ground fault directional detection neural network outputs the tag vector V when the ground fault confirmation signal indicates that the current signal I is a ground fault current signal. In this step, if the ground fault confirmation signal indicates that the current signal I analyzed in steps S201-S203 is a ground fault signal, then the tag vector V calculated in step S203 is considered valid, and tag vector V is output. Conversely, if the ground fault confirmation signal indicates that the current signal I analyzed in steps S201-S203 is not a ground fault signal, then the tag vector V calculated in step S203 is considered invalid, and tag vector V is not output.

[0047] In the embodiments of this disclosure, a ground fault directionality detection neural network is used to determine whether a ground fault is an upstream or downstream fault. Since it only needs to analyze the current signal without collecting voltage information, and the current transformers and relays in the line can themselves collect the current signal, the ground fault detection method according to the embodiments of this disclosure can be conveniently implemented without considering the additional deployment of voltage sensors. This saves costs and makes fault directionality detection easier to perform. Furthermore, because neural networks have significant advantages in feature extraction and feature-based classification, this method also improves the accuracy of ground fault directionality detection.

[0048] Figure 3 This is a flowchart of a ground fault directionality detection method 300 according to another embodiment of the present disclosure.

[0049] In most cases, ground faults occur only occasionally, so it is unnecessary to perform ground fault directionality analysis on the current signal at every moment. However, when a ground fault occurs, the current value of the current signal usually increases. Therefore, to save computational resources, a start threshold can be set for the ground fault directionality detection neural network.

[0050] like Figure 3 As shown, the ground fault directionality detection method 300 may include steps S301-S306. Step S301 in method 300 is related to... Figure 2 The steps are the same as S201, except that a preset start threshold judgment operation is added at step 302.

[0051] In step S302, it is determined whether the maximum value in the numerical matrix M obtained in step S301 is greater than a preset start-up threshold. This maximum value corresponds to the maximum current value in the sampled current signal I. The preset start-up threshold T... enable This is the threshold value for activating the ground fault directionality detection neural network for feature extraction and classification; this T... enable It can be set according to actual needs; for example, it can be set to the abrupt change in the sampled value of the current signal I, X. delta X delta Including but not limited to satisfying X delta =X k –X k-2N Where K is the number of sampling points per cycle, for example, k = 48. This equation means using the signal quantity X collected when the fault just occurred. k Subtract the signal quantity X under normal load conditions before the fault k-2N This yields the quantity that best reflects the fault characteristics; for example, the signal quantity from two sampling periods ago can be used as a reference quantity for the normal compliance state. Furthermore, the preset start-up threshold T... enableIt can be modified manually, for example, by adjusting it according to the power environment within the relay's protection range.

[0052] If the judgment result of step S302 is that the maximum value in the numerical matrix M is greater than or equal to the preset start threshold T enable If the condition is met, then method 300 proceeds to step S303. Similar to step S202 of method 200, in step S303, the numerical matrix M is provided as input to the ground fault directivity detection neural network. Also similar to step S203 of method 200, in step S304, the ground fault directivity detection neural network calculates a label vector V corresponding to the numerical matrix M. This label vector V indicates whether the current signal I is an upstream fault current or a downstream fault current. If the judgment result of step S302 is that the maximum value in the numerical matrix M is less than a preset start-up threshold T... enable Then it waits for the next numerical matrix M obtained in step 301. It should be understood that... Figure 3 Step S301 can be executed continuously, that is, the current signal I is continuously acquired (for example, each acquisition range is one cycle of the current signal I). Therefore, steps S302 to S306 are repeated for each numerical matrix acquired in step S301.

[0053] Steps S305 and S306 are respectively with Figure 2 Steps S204 and S205 are the same, and will not be repeated here.

[0054] In the embodiments of this disclosure, by setting the ground fault directional detection neural network to be activated only when a preset activation threshold is met, computational resources are saved.

[0055] Figure 4 This is a flowchart of a training method 400 for a ground fault directionality detection neural network according to an embodiment of the present disclosure.

[0056] like Figure 4 As shown, the training method 400 for the ground fault directional detection neural network includes steps S401 to S407.

[0057] In step S401, a training sample set of ground fault current signals is obtained for training the ground fault directionality detection neural network. In embodiments of this disclosure, each sample in the training sample set can be a ground fault current signal collected from an actual ground fault occurrence scenario or a ground fault current signal generated by an electromagnetic transient simulation program (EMTP) or other software programs according to a specific algorithm. These ground fault current signals can include current signals representing upstream faults and current signals representing downstream faults. For example, when the collected ground fault current signals from actual ground fault occurrence scenarios are too few to meet the sample size required for training the ground fault directionality detection neural network, training samples generated by EMTP can be used as a supplement. For each sample in the training sample set, regardless of its source (either of the two aforementioned cases), the true label vector of the sample is known, and the label vector calculated by the ground fault directionality detection neural network should be consistent with the true label vector of the sample. Therefore, the goal of training the ground fault directionality detection neural network is to minimize the difference between the label vector calculated by the ground fault directionality detection neural network and its true label vector for each sample in the training sample set.

[0058] In step S402, training samples from the current signal training sample set are provided as input to the ground fault directionality detection neural network. The input training samples in this step can be all samples, a portion of the samples, or just one sample from the training sample set, depending on factors such as the size of the sample set, computational resource requirements, and the required network prediction accuracy.

[0059] In step S403, the ground fault directionality detection neural network calculates the label vector corresponding to the training sample. As mentioned above, the ground fault directionality detection neural network extracts features from the training sample input in step S402 and classifies them according to the extracted features to obtain a label vector V indicating whether the current signal corresponding to the training sample is an upstream fault current signal or a downstream fault current signal.

[0060] In step S404, the processing loss of the ground fault directionality detection neural network is determined based on the label vector of the training sample. For the training sample input in step S402, the label vector output by the ground fault directionality detection neural network can be compared with its known true label vector. For example, it is known that for a training sample that is actually an upstream fault, its true label vector is [1,0], indicating that the probability of the training sample being an upstream fault current signal is 1 and the probability of it being a downstream fault current signal is 0; for a training sample that is actually a downstream fault, its true label vector is [0,1], indicating that the probability of the training sample being an upstream fault current signal is 0 and the probability of it being a downstream fault current signal is 1. Before the ground fault directionality detection neural network is fully trained, the output label vector of the training samples may differ significantly from the true label vector of the training samples. For a training sample known to be an upstream fault current signal, the output label vector might be [0.2, 0.8], indicating that the neural network considers the sample to be an upstream fault current signal with a probability of 0.2 and a downstream fault current signal with a probability of 0.8. This result shows that the neural network's detection of this sample is completely wrong. For example, if the calculated label vector is [0.7, 0.3], it indicates that the ground fault detection neural network considers the training sample to be a downstream fault current signal with a probability of 0.7 and a probability of 0.3. Although this result shows that the training sample is more likely to be an upstream fault current signal than a downstream fault current signal, the difference between it and the true label vector is still relatively large, meaning the accuracy is low, and the training objective of minimizing the difference between the label vector calculated by the neural network and the true label vector has not yet been achieved. The difference between the label vector output by the ground fault detection neural network and the true label vector can be measured by the processing loss L. The determination of the processing loss L is related to the selection of the loss function used in the neural network for ground fault directionality detection. For example, the loss function can be one of the following: mean squared error function, cross-entropy loss function, L1 loss function, L2 loss function, etc. The processing loss L corresponds to whether the training samples input in step 402 are all samples, a portion of the training sample set, or a single sample. It can be determined based on Batch Gradient Descent (BGD), Mini-Batch Gradient Descent (SGD), or Stochastic Gradient Descent (MBGD).

[0061] In step S405, it is determined whether the processing loss L is greater than or equal to the preset processing loss threshold T. L Preset processing loss threshold T LThe value can be selected based on the prediction accuracy of the ground fault directionality detection neural network, taking into account the protection range involved by the ground fault directionality detection equipment. For example, 0.01 or 0.02 can be selected. If it is determined that the processing loss L is greater than or equal to the preset processing loss threshold T... L If the training of the ground fault directional detection neural network is not yet complete, the process proceeds to step S406. Conversely, if the processing loss L is determined to be less than the preset processing loss threshold T, the process continues. L If the training of the ground fault directional detection neural network is completed, then proceed to step S407.

[0062] In step S406, the parameters of the ground fault directional detection neural network are updated. This step can be performed using the backpropagation algorithm to update one or more parameters in the ground fault directional detection neural network. These parameters mainly include the weight parameters of each layer of the neural network, such as the weight matrix of one or more convolutional kernels in the convolutional layer, the weight matrix of the first and / or second fully connected layer, etc. After the update is completed, steps S402 to S405 are repeated until it is determined in step S405 that the processing loss L is less than the preset processing loss threshold T. L In step S407, the training of the ground fault directionality detection neural network is stopped.

[0063] Figure 5 This is a schematic diagram of the structure of a ground fault directionality detection neural network according to an embodiment of the present disclosure.

[0064] like Figure 5 As shown, the ground fault directionality detection neural network can be a convolutional neural network including convolutional layers, a first fully connected layer, a second fully connected layer, an activation layer, and an output layer. The convolutional layer can include n convolutional kernels: kernel 1, kernel 2, ..., kernel n. The numerical matrix input to the ground fault directionality detection neural network is M.

[0065] After the numerical matrix M is input into the ground fault directionality detection neural network, it is first processed by a convolutional layer. Each of the convolutional kernels 1 to n in the convolutional layer is convolved with the numerical matrix M to obtain the convolution result of the first convolutional layer, i.e., the numerical matrix C. A row in the numerical matrix C corresponds to the convolution result of one of the convolutional kernels 1 to n with the numerical matrix M.

[0066] Next, the numerical matrix C is expanded into a row or a column. The expanded result is the first fully connected layer, denoted as the numerical matrix U1.

[0067] Next, the numerical matrix U1 is multiplied by the weight matrix W1 of the first fully connected layer to obtain the second fully connected layer, denoted as the numerical matrix U2.

[0068] Next, the numerical matrix U2 is multiplied by the weight matrix W2 of the second fully connected layer to obtain the weighted composite vector S.

[0069] Finally, the weighted sum vector S is transformed into a label vector V using the activation function G(z) of the activation layer. The label vector V indicates whether the detected current signal is an upstream or downstream fault current signal. The activation function G(z) of the activation layer can include, but is not limited to, the sigmoid function, the ReLU function, and the tanh function.

[0070] It is important to note that Figure 5 The specific structure of the ground fault directionality detection neural network shown is merely an example for illustrative purposes. In specific environments, this structure can be modified without departing from the scope of this disclosure. For example, to reduce the amount of data to be computed, a pooling layer can be included after the convolutional layer. The pooling layer can use average pooling or max pooling to reduce the dimensionality of the numerical matrix C resulting from the convolutional layer. Furthermore, the number of convolutional layers, pooling layers, fully connected layers, activation layers, etc., can be modified according to the training status of the ground fault directionality detection neural network. For instance, when overfitting occurs during training, the structure of the ground fault directionality detection neural network can be simplified, such as by reducing the number of convolutional layers. Additionally, the number and size of the convolutional kernels included in each convolutional layer can be modified, and the sizes of the convolutional kernels included in each convolutional layer can be different. This disclosure does not exhaustively list all possible modifications.

[0071] To make it easier to understand, a detailed example is given below. Figure 5 The ground fault detection neural network shown is based on Figure 4 The training process of the training method shown is as well as the process of using the trained ground fault detection neural network to calculate the label vector V corresponding to the current signal I.

[0072] For ease of explanation, the following settings are made for this example: the sampling resolution is k = 48, the sampling range is N = 1 cycles, the convolutional layer of the ground fault directionality detection neural network contains 2 convolutional kernels, i.e., n = 2, the size of convolutional kernel 1 and kernel 2 is 5×5, and the cross-entropy loss function is used. Processing loss threshold T L Set to 0.01, the activation function used for the activation layer is σ(z) = 1 / (1+e^(z-1)). -z ).

[0073] This example illustrates Figure 5 The ground fault directionality detection neural network shown is based on Figure 4 The training process of the training method shown.

[0074] Corresponding to step S401, 4000 numerical matrices representing the sampled ground fault current signals are generated using EMTP as a training sample set. This training sample set includes 2000 numerical matrices corresponding to the upstream fault current signals and 2000 numerical matrices corresponding to the downstream fault current signals. These 4000 numerical matrices are randomly sorted and sequentially numbered M1 to M... 4000 For an upstream fault current signal sample, its true label vector is [1,0], indicating that the probability of the sample being an upstream fault current signal is 1 and the probability of it being a downstream fault current signal is 0; for a downstream fault current signal sample, its true label vector should be [0,1], indicating that the probability of the sample being an upstream fault current signal is 0 and the probability of it being a downstream fault current signal is 1.

[0075] Corresponding to step S402, the training sample M1 is provided to the ground fault directionality detection neural network. It is known that M1 is an upstream fault current signal sample, and its true label vector is [1,0].

[0076] Corresponding to step S403, the ground fault directionality detection neural network outputs a label vector y1 corresponding to M1, for example, [0.4, 0.6]. That is to say, at this time, the ground fault directionality detection neural network considers that the probability of sample M1 being an upstream fault current signal is 0.4 and the probability of it being a downstream fault current signal is 0.6. This indicates that the classification of sample M1 by the ground fault directionality detection neural network at this time is completely wrong.

[0077] Corresponding to step S404, based on the loss function The ground fault directional detection neural network can determine its processing loss L1 = -(1×log0.4 + 0×log0.6) = 0.91.

[0078] Corresponding to step S405, the ground fault directional detection neural network determines that its processing loss L1 = 0.91 ≥ T. L =0.01.

[0079] Corresponding to step S406, the ground fault directional detection neural network updates its parameters according to the backpropagation algorithm. These parameters include a 5×5 weight matrix for convolution kernels 1 and 2 of the convolutional layer, a 20×44 weight matrix W1 for the first fully connected layer, and a 2×20 weight matrix W2 for the second fully connected layer.

[0080] After the update is complete, steps S402 to S406 are repeated, that is, the training sample M2 is provided to the ground fault directionality detection neural network, and then feature extraction, classification, and processing loss calculation of the training sample M2 are performed. This continues until the processing loss L is determined in step S405. <T LIf the training ends, proceed to step S407, which means the training is complete.

[0081] After training, we will obtain a 2×5×5 weight matrix corresponding to the two convolutional kernels of the convolutional layer, a 20×44 weight matrix corresponding to the weight matrix W1 of the first fully connected layer, and a 2×20 weight matrix corresponding to the weight matrix W2 of the second fully connected layer. These parameters, together with the previously set structure, loss function, activation function, etc. of the ground fault directionality detection neural network, constitute the trained ground fault directionality detection neural network.

[0082] It should be noted that the above simple examples are for illustrative purposes only and do not limit this disclosure. Various modifications and changes can be made to them according to actual needs. For example, multiple samples can be output each time in step S402, and the processing loss calculated in step S404 is the average of the processing losses of these samples, etc. This disclosure will not exhaustively list all such examples.

[0083] The following example continues to illustrate the process of using a trained ground fault directionality detection neural network to detect and calculate the label vector V corresponding to the current signal I.

[0084] refer to Figure 1 and Figure 4 Taking the detection process performed by the ground fault directional detection device 100 configured in relay R1 as an example, the ground fault detection device 100 detects the three-phase current signal I of the transformer converted by the current transformer. According to the sampling resolution k = 48, the current signal is sampled for N = 1 cycles to obtain a numerical matrix M of size 3 × 48. Each row in M ​​represents the sampled value of one phase, such as... Figure 4 Middle I ph1 I ph2 I ph3 As shown.

[0085] After the numerical matrix M is input into the ground fault directionality detection neural network, the two convolutional kernels 1 and 2 contained in the convolutional layer are convolved with it respectively. The convolution result of each convolutional kernel with M is a numerical vector containing 22 values. Therefore, the convolution result of the two convolutional kernels with M is a numerical matrix C of size 2×22.

[0086] Next, the numerical matrix C is expanded into a row or a column of the first fully connected layer. In this example, it is expanded into a column, resulting in a numerical matrix U1 of size 44×1.

[0087] Next, the numerical matrix U1 is multiplied by the 20×44 weight matrix W1 of the first fully connected layer to obtain a 20×1 numerical matrix U2, which serves as the second fully connected layer.

[0088] Next, the numerical matrix U2 is multiplied by the weight matrix W2 of the second fully connected layer, which is 2×20, to obtain the weighted composite vector S.

[0089] Finally, the weighted sum vector S is transformed into a label vector V using the activation function G(z) of the activation layer. The activation function G(z) of the activation layer can be, but is not limited to, one of the sigmoid function, ReLU function, or tanh function.

[0090] Figure 6 The test results obtained by testing the ground fault directionality detection neural network according to an embodiment of the present disclosure are shown. Figure 6 The left side of the image shows an example of a ground fault current signal sample belonging to a downstream fault in the test sample set. The horizontal axis represents the index of the sampling point, and the vertical axis represents the current value in amperes (A). (Combined with...) Figure 4 and Figure 5 Similarly, in the described example, the ground fault current signal sample is sampled for one period at a sampling resolution k = 48, resulting in a 3×48 numerical matrix. This numerical matrix is ​​then input into the ground fault directivity detection neural network. The label vector calculated by the ground fault directivity detection neural network is shown in the figure as [1.5835431468075839e]. -06 The maximum value of 0.999998671408647 in the label vector corresponds to the downstream fault category, so the ground fault directional detection neural network predicts this sample as a downstream fault current signal. Figure 6 The right side of the image shows an example of a ground fault current signal sample belonging to an upstream fault in the test sample set. The horizontal axis represents the index of the sampling point, and the vertical axis represents the current value in amperes (A). This sample is also sampled for one period at a sampling resolution k = 48, resulting in a 3×48 numerical matrix. This numerical matrix is ​​input into a ground fault directivity detection neural network. The label vector calculated by the ground fault directivity detection neural network is shown in the figure: [0.9999976969348218, 3.2066364547510186e]. -06 The maximum value of 0.9999976969348218 in the label vector V corresponds to the upstream fault category, so the ground fault directional detection neural network predicts this sample as an upstream fault current signal.

[0091] It is important to note that Figure 6 For illustrative purposes, only two test samples are shown, but the number of samples in the test sample set can be determined according to actual needs. For example, it can be a certain proportion of the number of samples in the training sample set, such as 20%, 30%, etc.

[0092] This disclosure demonstrates, through testing with the aforementioned exemplary convolutional neural network, that the ground fault directionality detection method according to embodiments of this disclosure can achieve ground fault directionality detection solely by extracting features from the ground fault current signal. Obviously, those skilled in the art will understand that the ground fault directionality detection method of this disclosure is not limited to the aforementioned exemplary convolutional neural network, and other suitable neural networks can also be employed.

[0093] Figure 7 This is a block diagram illustrating a ground fault directionality detection device 700 according to an embodiment of the present disclosure.

[0094] like Figure 7 As shown, the ground fault directionality detection device 700 includes a sampling module 701, an input module 702, and a classification module 703. These three modules can be configured to be in the same physical location, for example, all integrated within... Figure 1 The relays R1-R4 shown can also be integrated into other computing devices on the line, such as fault path indicators (FPI). Alternatively, one or two of these three modules can be configured to be in different physical locations from the other two or one and connected via wired or wireless communication networks to transmit data or commands between them. For example, sampling module 701 can be integrated into relays R1-R4 or fault path indicators (FPI), while input module 702 and classification module 703 can be integrated into computing devices such as computers or servers in a control room. Since power equipment can operate with faults for a period of time in low-current grounding systems, the time spent on data and command transmission between sampling module 701 and input module 702 and / or classification module 703 can be disregarded.

[0095] The sampling module 701 is configured to sample at least a portion of the current signal I to obtain a numerical matrix M. Its sampling resolution is k sample points per power frequency cycle, for example, k = 24, 32, 48, etc. Furthermore, the sampling range for each iteration can be N cycles of the ground fault current signal I, for example, N = 1, 2, etc.

[0096] The input module 702 provides the numerical matrix M obtained by the sampling module 701 as input to the ground fault directivity detection neural network. Furthermore, the input module 702 can also detect whether the maximum value in the numerical matrix M is greater than a preset start-up threshold, and only provides the numerical matrix M as input to the ground fault directivity detection neural network when the maximum value is greater than or equal to the preset start-up threshold.

[0097] The classification module 703 is configured with a ground fault directionality detection neural network as described in the embodiments of the present disclosure, and the ground fault directionality detection neural network outputs a label vector V corresponding to the numerical matrix M. The label vector V indicates whether the current signal I is an upstream fault current signal or a downstream fault current signal. In addition, the ground fault directionality detection neural network also receives a ground fault confirmation signal and outputs the label vector V when the ground fault confirmation signal indicates that the current signal is a ground fault current signal.

[0098] Obviously, the above description of the ground fault directionality detection method also applies to the ground fault directionality detection equipment 700, and will not be repeated here.

[0099] Figure 8 This is a hardware block diagram illustrating a ground fault directionality detection device 800 according to another embodiment of the present disclosure.

[0100] An electronic device 800 according to an embodiment of this disclosure includes at least a memory for storing computer program instructions and a processor for executing the computer program instructions. When the computer program instructions are loaded and run by the processor, the processor executes the training method and ground fault detection method of the ground fault directionality detection neural network described above.

[0101] Figure 7 The ground fault directionality detection device 800 shown may specifically include a central processing unit (CPU) 801, a graphics processing unit (GPU) 802, and a main memory 803. These units are interconnected via a bus 804. The CPU 801 and / or GPU 802 can be used as the aforementioned processor, and the main memory 803 can be used as the aforementioned memory for storing computer program instructions. Furthermore, the electronic device 800 may also include a communication unit 805, a storage unit 806, an output unit 808, an input unit 808, and an external device 809, which are also connected to the bus 804.

[0102] Figure 9 This is a schematic diagram illustrating a computer-readable storage medium according to embodiments of the present disclosure. Figure 9 As shown, a computer-readable storage medium 900 according to an embodiment of the present disclosure stores computer program instructions 901 thereon. When the computer program instructions 901 are executed by a processor, the training method and the ground fault directionality detection method of the ground fault directionality detection neural network according to an embodiment of the present disclosure, as described above with reference to the accompanying drawings, are executed.

[0103] The computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, optical disk, magnetic disk, etc.

[0104] The above description, with reference to the accompanying drawings, outlines a neural network-based method and apparatus for detecting the directionality of ground faults according to embodiments of the present disclosure. Leveraging the powerful feature extraction and classification capabilities of neural networks, the neural network-based method, apparatus, and computer-readable storage medium for detecting the directionality of ground faults can accurately determine whether a ground fault is upstream or downstream, requiring only current information and not voltage information. This detection result can help pinpoint the source of the ground fault, allowing the appropriate maintenance team to be notified for repairs.

[0105] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0106] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0107] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0108] Additionally, as used herein, the “or” used in a list of items beginning with “at least one” indicates a separate list, such that a list of, for example, “at least one of A, B, or C” means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word “exemplary” does not imply that the described example is preferred or better than other examples.

[0109] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0110] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0111] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0112] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for ground fault directionality detection, comprising: sampling at least a portion of a three-phase current signal to obtain a numerical matrix, one row or one column of the numerical matrix corresponding to one phase of the three-phase current signal; detecting whether a maximum value in the numerical matrix is greater than a preset activation threshold; in a case where the maximum value in the numerical matrix is greater than or equal to the preset activation threshold, providing the numerical matrix as an input to a ground fault directionality detection neural network, and in a case where the maximum value in the numerical matrix is less than the preset activation threshold, detecting whether a maximum value in a next numerical matrix is greater than the preset activation threshold; in response to the input provided to the ground fault directionality detection neural network, calculating, by the ground fault directionality detection neural network, a label vector corresponding to the input, wherein the label vector indicates whether the three-phase current signal is an upstream fault current signal or a downstream fault current signal; receiving, by the ground fault directionality detection neural network, a ground fault confirmation signal; and in a case where the ground fault confirmation signal indicates that the three-phase current signal is a ground fault current signal, outputting, by the ground fault directionality detection neural network, the label vector.

2. The method for ground fault directionality detection of claim 1, wherein the ground fault directionality detection neural network employs a convolutional neural network.

3. The method for ground fault directionality detection of claim 1, wherein the ground fault directionality detection neural network is a trained neural network, and the training comprises the following steps: Step 1, obtaining a ground fault current signal training sample set for training the ground fault directionality detection neural network; Step 2, providing a training sample in the ground fault current signal training sample set as an input to the ground fault directionality detection neural network; Step 3, calculating, by the ground fault directionality detection neural network, a label vector corresponding to the training sample; Step 4, determining a processing loss of the ground fault directionality detection neural network based on the label vector of the training sample; Step 5: if the processing loss is greater than or equal to a preset processing loss threshold, updating parameters of the ground fault directionality detection neural network and performing steps 2 to 5 based on the updated ground fault directionality detection neural network, and if the processing loss is less than or equal to the preset processing loss threshold, stopping the training.

4. The method for ground fault directionality detection of claim 3, wherein the ground fault directionality detection neural network comprises at least a convolutional layer, a first fully connected layer, a second fully connected layer, an activation layer, and an output layer; and the parameters of the ground fault directionality detection neural network comprise at least a weight matrix of a convolution kernel included in the convolutional layer, a first weight matrix included in the first fully connected layer, and a second weight matrix included in the second fully connected layer.

5. The method for ground fault directionality detection of claim 4, wherein the ground fault directionality detection neural network further comprises a pooling layer.

6. The method for ground fault directionality detection of claim 5, wherein ​ The convolution layer adopts a convolution kernel with a size of 5*5; and The activation function of the activation layer is one of a sigmoid function, a relu function, and a tanh function. 7.The ground fault direction detection method of claim 3, wherein One or more training samples in the set of training samples of the ground fault current signal are generated by an electromagnetic transient program (EMTP). 8.A ground fault direction detection device, comprising: a sampling module configured to sample at least a portion of a three-phase current signal to obtain a numerical matrix, one row or one column of the numerical matrix corresponding to one phase of the three-phase current signal; an input module configured to detect whether a maximum value in the numerical matrix is greater than a preset start threshold, in a case where the maximum value in the numerical matrix is greater than or equal to the preset start threshold, provide the numerical matrix as an input to a ground fault direction detection neural network, and in a case where the maximum value in the numerical matrix is less than the preset start threshold, detect whether a maximum value in a next numerical matrix is greater than the preset start threshold; and a classification module configured with the ground fault direction detection neural network, wherein, in response to an input provided to the ground fault direction detection neural network, a label vector corresponding to the input is calculated by the ground fault direction detection neural network, the label vector indicating whether the three-phase current signal is an upstream fault current signal or a downstream fault current signal, the ground fault direction detection neural network further receiving a ground fault confirmation signal and outputting the label vector in a case where the ground fault confirmation signal indicates that the three-phase current signal is a ground fault current signal. 9.A ground fault direction detection device, comprising: a processor; and a memory, wherein the memory has stored therein a computer executable program which, when executed by the processor, causes the device to perform the method of any one of claims 1-7. 10.A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the method of any one of claims 1-7.

Images