Method and device for determining development stage of high-resistance ground fault of power distribution network

Abstract

This application discloses a method and apparatus for determining the development stage of a high-resistance grounding fault in a distribution network. The method includes: acquiring a target zero-sequence current signal of the distribution network in a target period; determining a target slow characteristic based on the target zero-sequence current signal, wherein the target slow characteristic describes the evolution trend of the high-resistance grounding fault in the distribution network; determining a target confidence level for the target period based on the target slow characteristic, wherein the target confidence level characterizes the probability of a high-resistance grounding fault existing in the distribution network; and determining the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level. This application solves the technical problem of inaccurate determination of the development stage of a high-resistance grounding fault in a distribution network in related technologies.

Description

Technical Field

[0001] This application relates to the field of power systems, and more specifically, to a method and apparatus for determining the development stage of a high-resistance grounding fault in a distribution network. Background Technology

[0002] High-resistance grounding faults in distribution networks are among the most difficult to detect in power systems due to their weak fault current, concealed nonlinear characteristics, and susceptibility to environmental noise and load fluctuations. They not only seriously threaten public safety (e.g., causing wildfires, electric shocks) but can also evolve into serious accidents such as phase-to-phase short circuits if left untreated for extended periods. This represents a critical technological bottleneck that must be overcome to improve the reliability of distribution network power supply. Current methods for determining the development stage of high-resistance grounding faults in distribution networks, such as those relying on steady-state electrical quantities, those based on traditional signal processing (e.g., Fast Fourier Transform, Wavelet Transform), and those based on deep learning, cannot effectively balance the contradiction between "feature extraction under strong noise" and "time series modeling of non-stationary signals." This leads to frequent false alarms and missed detections in complex field environments with low signal-to-noise ratios (e.g., -5dB), failing to meet the actual detection needs of high-resistance grounding faults and their development stages in distribution networks. Therefore, the existing technologies suffer from inaccurate determination of the development stage of high-resistance grounding faults in distribution networks.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This application provides a method and apparatus for determining the development stage of a high-resistance grounding fault in a distribution network, so as to at least solve the technical problem of inaccurate determination of the development stage of a high-resistance grounding fault in a distribution network in related technologies.

[0005] According to one aspect of the embodiments of this application, a method for determining the development stage of a high-resistance grounding fault in a distribution network is provided, comprising: acquiring a target zero-sequence current signal of the distribution network in a target period; determining a target slow feature based on the target zero-sequence current signal, wherein the target slow feature is used to describe the evolution trend of the high-resistance grounding fault in the distribution network; determining a target confidence level in the target period based on the target slow feature, wherein the target confidence level is used to characterize the probability of a high-resistance grounding fault existing in the distribution network; and determining the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level.

[0006] According to another aspect of the embodiments of this application, a device for determining the development stage of a high-resistance grounding fault in a distribution network is provided, comprising: a data acquisition module for acquiring a target zero-sequence current signal of the distribution network in a target period; a first determination module for determining a target slow feature based on the target zero-sequence current signal, wherein the target slow feature is used to describe the evolution trend of the high-resistance grounding fault in the distribution network; a second determination module for determining a target confidence level in the target period based on the target slow feature, wherein the target confidence level is used to characterize the probability of a high-resistance grounding fault existing in the distribution network; and a third determination module for determining the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level.

[0007] According to another aspect of the embodiments of this application, a non-volatile storage medium is provided, which stores multiple instructions, any one of which is adapted to be loaded by a processor to determine the development stage of a high-resistance grounding fault in a distribution network.

[0008] According to another aspect of the embodiments of this application, an electronic device is provided, including: one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the following methods for determining the development stage of a high-resistance grounding fault in a power distribution network.

[0009] According to another aspect of the embodiments of this application, a computer program product is provided that, when executed on a data processing device, is suitable for performing the steps of a method for determining the development stage of a high-resistance grounding fault in a distribution network.

[0010] In this embodiment, a target zero-sequence current signal of the distribution network in a target period is acquired; based on the target zero-sequence current signal, a target slow feature is determined, wherein the target slow feature is used to describe the evolution trend of high-resistance grounding faults in the distribution network; based on the target slow feature, a target confidence level for the target period is determined, wherein the target confidence level is used to characterize the probability of a high-resistance grounding fault existing in the distribution network; and based on the target confidence level, the development stage of the high-resistance grounding fault in the distribution network is determined. This achieves the goal of determining the target slow feature based on the acquired target zero-sequence current signal of the distribution network, determining the target confidence level characterizing the probability of a high-resistance grounding fault existing in the distribution network, and further determining the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level. This achieves the technical effect of improving the accuracy of the target confidence level determination result by determining the target slow feature, thereby improving the accuracy of the high-resistance grounding fault development stage determination result in the distribution network, and thus solving the technical problem of inaccurate determination of the high-resistance grounding fault development stage in the related art. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0012] Figure 1 This is a flowchart of a method for determining the development stage of a high-resistance grounding fault in a distribution network according to an embodiment of this application;

[0013] Figure 2 This is a flowchart of an optional method for determining the development stage of a high-resistance grounding fault in a distribution network, according to an embodiment of this application.

[0014] Figure 3 This is a schematic diagram of a device for determining the development stage of a high-resistance grounding fault in a power distribution network, according to an embodiment of this application.

[0015] Figure 4 This is a structural diagram of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] It should be noted that the information (including but not limited to the first threshold and the second threshold) and data (including but not limited to the target zero-sequence current signal and the initial training zero-sequence current signal sequence) involved in this application are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, and necessary confidentiality measures have been taken. These measures do not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or institutions, providing users with corresponding operation entry points for them to choose to agree to or refuse the automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0019] According to an embodiment of this application, a method embodiment for determining the development stage of a high-resistance grounding fault in a distribution network is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0020] Figure 1 This is a flowchart of a method for determining the development stage of a high-resistance grounding fault in a distribution network according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0021] Step S102: Obtain the target zero-sequence current signal of the distribution network in the target period;

[0022] It is understandable that by acquiring the target zero-sequence current signal within the target period, the complete waveform characteristics and time-varying dynamics of the fault current in the distribution network within a specific time period can be captured comprehensively and continuously, laying a data foundation for the subsequent accurate extraction of target slow features and determination of the development stage of high-resistance grounding faults in the distribution network.

[0023] Step S104: Based on the target zero-sequence current signal, determine the target slow characteristics, wherein the target slow characteristics are used to describe the evolution trend of high-resistance grounding faults in the distribution network.

[0024] It is understandable that slow extraction of the target zero-sequence current signal yields target slow features used to describe the evolution trend of high-resistance grounding faults in the distribution network. By extracting these target slow features, high-frequency random noise interference can be effectively suppressed and redundant correlations between features can be eliminated. This allows for the accurate removal and retention of low-frequency slow-changing signals that characterize the essential evolution trend of high-resistance grounding faults, thereby improving the identifiability and purity of high-resistance grounding fault features in a noisy environment.

[0025] In one optional embodiment, determining the target slow feature based on the target zero-sequence current signal includes: determining a target projection weight matrix, wherein the target projection weight matrix is ​​used to linearly map the high-dimensional zero-sequence current signal into a low-dimensional slow feature; and determining the target slow feature based on the target projection weight matrix and the target zero-sequence current signal.

[0026] It is understandable that a linear projection mapping of the target zero-sequence current signal using a target projection weight matrix yields the target's slow characteristics. By using a target projection weight matrix to linearly project the target zero-sequence current signal, the high-dimensional, noisy original current signal can be efficiently reduced in dimensionality and mapped to a low-dimensional, slowly varying feature space. This not only eliminates fast random noise interference but also accurately preserves and highlights the slow evolution trend that characterizes the essence of high-resistance grounding faults, laying the foundation for determining the development stage of high-resistance grounding faults in the subsequent distribution network.

[0027] In one optional embodiment, determining the target projection weight matrix includes: acquiring multiple initial training zero-sequence current signals of the distribution network arranged in order from front to back according to the acquisition period; performing noise processing on the multiple initial training zero-sequence current signals to obtain multiple target training zero-sequence current signals; and determining the target projection weight matrix based on the multiple target training zero-sequence current signals.

[0028] It is understandable that high-precision current transformers or a constructed distribution network simulation model are used to acquire multiple initial training zero-sequence current signals of the distribution network under various operating conditions (including normal, disturbance, and high-resistance grounding faults), arranged in chronological order according to the acquisition cycle. However, due to the complexity of actual distribution network conditions and the scarcity of high-resistance grounding fault samples, directly acquired initial training zero-sequence current signals often suffer from weak noise interference, a single sample distribution, and unclear fault characteristics. To simulate real high-noise environments (such as low signal-to-noise ratio conditions), noise is added to multiple initial training zero-sequence current signals to obtain multiple target training zero-sequence current signals. Based on these multiple target training zero-sequence current signals, a target projection weight matrix is ​​determined. By constructing target training zero-sequence current signals that include multiple operating conditions and strong noise interference, a realistic and harsh operating environment can be simulated to enhance data diversity and noise resistance representativeness. This makes the target projection weight matrix determined based on the target training zero-sequence current signals more robust, ensuring the accuracy of the target slow feature extraction results.

[0029] Optionally, an IEEE (Institute of Electrical and Electronics Engineers) 33-node distribution network simulation model can be constructed, or high-precision current transformers can be deployed at actual distribution network monitoring points to collect zero-sequence current signals of distribution network lines (e.g., initial training zero-sequence current signals). To address the scarcity of high-resistivity grounding fault samples, a simulation dataset containing various fault media (such as dry road surfaces, grass, and asphalt) and different noise intensities can be established.

[0030] Optionally, the improved Mayr arc model can be used to simulate the arc dynamics at a high-resistance grounding fault, generating an original zero-sequence current signal sequence (i.e., a sequence composed of initial training zero-sequence current signals) containing "zero rest" and waveform distortion characteristics.

[0031] Optionally, in order to simulate harsh real-world operating conditions, the original zero-sequence current signal... High-intensity Gaussian white noise superimposed on the middle A low signal-to-noise ratio input signal (i.e., the target training zero-sequence current signal) is constructed. The input signal is constructed as follows:

[0032]

[0033] in, A zero-sequence current signal is trained for a target period of t. The initial training zero-sequence current signal has a period of t. This is the noise intensity adjustment factor (corresponding to SNR (Signal-to-Noise Ratio) = -5dB to 30dB). It is a random noise sequence that follows a standard normal distribution.

[0034] In one optional embodiment, determining the target projection weight matrix based on multiple target training zero-sequence current signals includes: for any target training zero-sequence current signal among the multiple target training zero-sequence current signals, determining the current slow feature corresponding to the current iteration number based on any target training zero-sequence current signal and the previous projection weight matrix, wherein the previous projection weight matrix was determined in the previous iteration number before the current iteration number; and determining the current instantaneous gradient corresponding to the current iteration number based on the current slow feature, any target training zero-sequence current signal, and the previous slow feature, wherein the previous slow feature was determined in the previous iteration number based on the previous target training zero-sequence current signal located before any target training zero-sequence current signal among the multiple target training zero-sequence current signals. Given a defined signal, the current instantaneous gradient is used to characterize the direction and magnitude of the adjustment of the projection weight matrix to minimize the rate of change of the slow feature. Based on the current instantaneous gradient, the previous projection weight matrix is ​​iteratively updated to obtain the updated projection weight matrix. The updated projection weight matrix is ​​orthogonalized to obtain the current projection weight matrix corresponding to the current iteration number. Based on the next target training zero-sequence current signal after any target training zero-sequence current signal among multiple target training zero-sequence current signals, the current projection weight matrix is ​​iteratively updated by updating the previous projection weight matrix until the iteration stopping condition is met, and the projection weight matrix obtained by the last iteration update is determined as the target projection weight matrix.

[0035] It can be understood that, for any target training zero-sequence current signal among multiple target training zero-sequence current signals, the current slow feature corresponding to the current iteration number is determined based on the target training zero-sequence current signal and the previous projection weight matrix (i.e., the projection weight matrix determined in the previous iteration number before the current iteration number). Based on the current slow feature, the target training zero-sequence current signal, and the previous slow feature (i.e., the slow feature determined in the previous iteration number), the current instantaneous gradient corresponding to the current iteration number is determined. Further, based on the current instantaneous gradient, the previous projection weight matrix is ​​iteratively updated to obtain the updated projection weight matrix, and the updated projection weight matrix is ​​further orthogonalized to obtain the current projection weight matrix corresponding to the current iteration number. This iterative update of the current projection weight matrix corresponding to the current iteration number continues in the above manner until the iteration stopping condition is met, and the projection weight matrix obtained from the last iteration update is determined as the target projection weight matrix. Through the above process, the online adaptive iterative update of the projection weight matrix can be achieved. While ensuring that the extracted slow feature components are orthogonal and uncorrelated, it dynamically adapts to the time-varying characteristics of the zero-sequence current signal of the distribution network, and finally converges to obtain the target projection weight matrix that is highly robust to strong noise and can accurately capture the slow trend of faults.

[0036] Optionally, in the initial stage of iterative optimization, a projection weight matrix needs to be initialized as the starting point for iterative updates, providing a stable search benchmark for subsequent iterative updates based on recursive exponential gradient descent.

[0037] Optionally, the above iteration stopping condition can be that the change in the projection weight matrix between two adjacent iterations is less than a preset threshold, or that the number of iterations reaches a preset maximum limit.

[0038] Optionally, a linear mapping function is defined to embed the high-dimensional or time-delayed input signal. Mapped to low-dimensional slow features The mapping formula is:

[0039]

[0040] in, This is the slow feature vector extracted at period t (i.e., the slow feature represented in vector form). The input signal vector is a period t (i.e., the target training zero-sequence current signal represented in vector form). Let be the projection weight matrix to be solved.

[0041] Optionally, an exponential forgetting factor is introduced to address the non-stationary characteristics of distribution network fault signals. We construct an objective function that minimizes the first-order difference energy of slow feature vectors to find the feature directions that change the slowest. Objective function Defined as:

[0042]

[0043] in, for The exponential forgetting factor of the cycle (ranging from 0.9 to 0.99) is used to strengthen the weight of the slow feature vector of the current cycle and forget the slow feature vector of the previous cycle. for The time derivative of a periodic slow eigenvector can be discretized and approximated as a first-order difference. for Slow eigenvectors with periodicity Periodic index variable. The objective function. A mathematical extremum criterion is defined for extracting the physical objective of "slow features," namely, minimizing the first-order difference energy of the slow feature vector.

[0044] Optionally, the constructed objective function establishes a global optimization criterion for extracting slow features. Since the fault signals (e.g., zero-sequence current signals) corresponding to high-resistivity grounding faults in distribution networks exhibit strongly non-stationary flowing data characteristics, traditional eigenvalue decomposition methods cannot meet the requirements of online real-time computation. Therefore, solving the above objective function is transformed into an online stochastic optimization process. Specifically, the instantaneous gradient is derived by taking the partial derivative of the objective function with respect to the projection weight matrix, and then the projection weight matrix is ​​iteratively updated. By calculating this instantaneous gradient and substituting it into the stochastic gradient descent framework, the online iterative update of the projection weight matrix is ​​achieved, thereby avoiding the offline batch processing delay of traditional slow feature analysis.

[0045] Alternatively, the weight matrix can be updated online using stochastic gradient descent. Orthogonalization constraints are applied to prevent feature collapse. The weight update formula is as follows:

[0046]

[0047] in, For learning rate, This is the instantaneous gradient of the loss function with respect to the weights. This is the previous projection weight matrix for the corresponding iteration number. This updates the projection weight matrix for the corresponding iteration number. After updating, the Gram-Schmidt algorithm is used to... Orthogonalization is performed to obtain the current projection weight matrix for the corresponding iteration number, ensuring that the extracted slow feature components are uncorrelated, thereby characterizing the fault trend from different dimensions.

[0048] Step S106: Based on the target slow characteristics, determine the target confidence level of the target period, wherein the target confidence level is used to characterize the probability of a high-resistance grounding fault in the distribution network;

[0049] It is understandable that, based on the extracted target slow features, a target fault diagnosis model is used to determine the target confidence level, which is used to characterize the probability of a high-resistance grounding fault in the distribution network during the target period.

[0050] In one optional embodiment, determining the target confidence level of the target period based on the target slow features includes: obtaining the target confidence level by using a target fault diagnosis model based on the target slow features, wherein the target fault diagnosis model has pre-learned the correlation between the target slow features and the target confidence level.

[0051] It is understandable that, based on the slow characteristics of the target, a target fault diagnosis model (such as a fault diagnosis model built on Bi-LSTM (Bidirectional Long Short-Term Memory Network)) is used to obtain the target confidence.

[0052] Optionally, a target confidence level can be determined using a target fault diagnosis model based on Bi-LSTM (Bidirectional Long Short-Term Memory Network), which aims to capture the long-distance temporal dependence before and after the occurrence of a high-resistance grounding fault.

[0053] Alternatively, the target fault diagnosis model based on Bi-LSTM can be modeled in the following way:

[0054] For each time step in the input sequence of the model, the LSTM unit controls the information flow through the input gate, forget gate, and output gate. The forget gate outputs at period t. The calculation formula is:

[0055]

[0056] in, It is the Sigmoid activation function. and These are the weight matrix and bias term of the forget gate, respectively. This is a hidden state with a period of t-1. It is a slow characteristic of period t.

[0057] The input gate output at period t With candidate cell state The calculation formula is:

[0058]

[0059]

[0060] in, The candidate cell states for period t; and These are the weight matrices for the input gate and the candidate state, respectively; and These are the corresponding bias terms; This is the hyperbolic tangent activation function, used to scale values ​​to the range [-1, 1].

[0061] Cell state in cycle t The update formula is:

[0062]

[0063] in, This represents the cell state after period t. For the cell state at period t-1, this step enables the long-term retention of historical information and the integration of new information.

[0064] Output gate output with period t and the final hidden state The calculation formula is:

[0065]

[0066]

[0067] in, and These are the weight matrix and bias term of the output gate, respectively; The hidden state output for period t contains filtered temporal feature information and will be passed to the next period and the output layer.

[0068] Construct forward LSTM layers (for processing sequences) respectively Where T is the number of cycles in the input sequence) and the backward LSTM layer (processing the sequence) The two are then concatenated in the hidden state of the last cycle to form a global feature vector. :

[0069]

[0070] in, Features extracted from the forward LSTM layer These are the inverse inference features extracted from the backward LSTM layer.

[0071] A sample library containing various operating conditions, including normal operation, line switching, and high-resistance grounding faults (tree branches, ground, insulator flashover), is generated and divided into training and test sets in an 8:2 ratio.

[0072] Construct a fully connected classification layer to integrate the global feature vector. This is mapped to fault category probabilities (i.e., confidence levels). The Negative Log-Likelihood Loss (NLLLoss) function is used as the optimization objective, and the Adam optimizer is used to iteratively update the Bi-LSTM network parameters. Loss Function The formula is:

[0073]

[0074] in, Batch size This represents the number of operating condition categories (e.g., normal / faulty). This is the true label (0 or 1) for the c-th working condition in the i-th batch. Let be the probability of the fault category predicted by the model under the c-th working condition in the i-th batch.

[0075] Step S108: Based on the target confidence level, determine the development stage of high-resistance grounding faults in the distribution network.

[0076] It is understandable that by using target confidence levels, the development stage of high-resistance grounding faults in distribution networks can be accurately determined, thereby realizing a proactive operation and maintenance mode that shifts from "post-tripping" to "pre-warning and in-process monitoring," improving the early detection capability of high-resistance grounding faults and effectively preventing secondary disasters.

[0077] In one optional embodiment, determining the high-resistance grounding fault development stage of the distribution network based on a target confidence level includes: if the target confidence level is less than a first threshold, determining the high-resistance grounding fault development stage as the distribution network does not have a high-resistance grounding fault; if the target confidence level is greater than or equal to the first threshold and less than a second threshold, and the trend of the target confidence level is upward, determining the high-resistance grounding fault development stage as the distribution network is in the high-resistance grounding fault evolution period, wherein the trend is obtained based on the target confidence level and the historical confidence level determined in historical periods prior to the target period; if the target confidence level is greater than or equal to the second threshold, and the historical confidence level in a preset number of historical periods prior to the target period is greater than or equal to the second threshold, determining the high-resistance grounding fault development stage as the distribution network is in the high-resistance grounding fault stability period.

[0078] It is understandable that if the target confidence level is less than the first threshold, it indicates that the distribution network is in normal operation and no significant fault characteristics have been detected. In this case, the high-resistance grounding fault development stage of the distribution network is determined to be that there is no high-resistance grounding fault in the distribution network. If the target confidence level is greater than or equal to the first threshold and less than the second threshold, and the trend of the target confidence level (obtained by trend analysis of the target confidence level for the target period and the historical confidence level determined in the historical period before the target period) is upward, it indicates that the distribution network has shown weak fault signs but has not yet formed a stable breakdown, and the arc is in an unstable burning state. In this case, the high-resistance grounding fault development stage of the distribution network is determined to be that the distribution network is in the high-resistance grounding fault evolution period. If the target confidence level is greater than or equal to the second threshold, and the historical confidence level for the preset number of historical periods before the target period is greater than or equal to the second threshold, it indicates that the distribution network has experienced a continuous and stable high-resistance grounding fault, and the arc burning has entered a mature steady state. In this case, the high-resistance grounding fault development stage of the distribution network is determined to be that the distribution network is in the high-resistance grounding fault stability period. By using a multi-dimensional threshold determination mechanism based on target confidence and its temporal variation trend, it is possible to accurately identify and differentiate the three full life cycle states of the distribution network: normal operation, early hidden danger evolution, and steady-state fault. This solves the problem that traditional methods are difficult to capture weak precursors of high-resistivity grounding faults and have a high false alarm rate, thereby improving the sensitivity and accuracy of high-resistivity grounding fault identification in the distribution network.

[0079] Alternatively, it can be based on the target confidence level. Based on the time-varying characteristics, the development stages of high-resistance grounding faults in distribution networks are divided into three stages:

[0080] During normal operation (i.e., when there is no high-resistance grounding fault in the distribution network): when When the threshold (i.e., the first threshold, such as 0.1) is reached, it is determined that there is no high-resistance grounding fault in the distribution network, and the distribution network is in normal operation.

[0081] Early stage of potential problem evolution (Pre-fault, i.e., the distribution network is in the high-resistance grounding fault evolution period): when (in When the second threshold (e.g., 0.9) is set and the trend of change is continuously rising, the distribution network is determined to be in the high-resistance grounding fault evolution period. At this time, the electric arc is in an unstable burning state, and a level-two warning is issued.

[0082] Steady-state fault phase (Fault, i.e., the distribution network is in the steady-state period of a high-resistance ground fault): when If the duration exceeds the set threshold (e.g., 3 (i.e., the preset number) cycles), the distribution network is determined to be in the stable period of high-resistance grounding fault. At this time, the distribution network is diagnosed with a high-resistance grounding fault and a first-level alarm is issued.

[0083] Through the above steps S102 to S108, the goal is to determine the target confidence level, which characterizes the probability of a high-resistance grounding fault in the distribution network, by determining the target slow characteristics based on the acquired target zero-sequence current signal of the distribution network, and further determine the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level. This achieves the technical effect of improving the accuracy of the target confidence level determination result by determining the target slow characteristics, thereby improving the accuracy of the determination result of the development stage of the high-resistance grounding fault in the distribution network, and thus solving the technical problem of inaccurate determination of the development stage of the high-resistance grounding fault in the distribution network in related technologies.

[0084] A method for determining the development stage of a high-resistance ground fault in a distribution network based on steady-state electrical quantities has been provided in the related technologies. However, the method suffers from insufficient sensitivity and non-significant extracted fault characteristics. The weakness of the high-resistance ground fault current (usually less than the load current) makes its electrical characteristics easily obscured or confused by transient disturbances such as normal three-phase imbalance, large load switching, and line switch operation in the distribution network. This leads to the following problems when determining the development stage of a high-resistance ground fault: (1) If the protection setting is reduced to increase sensitivity, it is easily affected by normal unbalanced current and may cause false tripping; (2) If the protection setting is increased to avoid false tripping, it may cause a large number of faults to be missed or not tripped. Therefore, this method has limitations in its ability to distinguish between weak faults (such as high-resistance ground faults) and normal operating disturbances.

[0085] The related technology provides a method for determining the development stage of high-resistance grounding faults in distribution networks based on traditional signal processing (FFT) and traditional slow feature analysis (SFA). However, it has the following shortcomings: (1) The actual distribution network is filled with strong random white noise and background harmonics. When the signal-to-noise ratio is as low as -5dB, the weak high-frequency characteristics of the electric arc are often completely aliased or submerged by the strong noise background, which makes it impossible for the frequency domain analysis method to effectively extract fault features; (2) Although traditional slow feature analysis (SFA) has the theoretical ability to extract slow changing trends (i.e., slow features), it is based on the assumption of linear transformation and usually adopts offline batch processing mode. This makes it unable to adapt to the dynamic changes of distribution network load in real time and difficult to effectively capture the dynamic evolution law of electric arcs with strong nonlinear characteristics.

[0086] Related technologies offer a method for determining the development stage of high-resistivity grounding faults in distribution networks based on deep learning (such as SVM (Support Vector Machine)) or local deep learning (such as 1D-CNN (One-Dimensional Convolutional Neural Network)). While this method introduces data-driven thinking and avoids the difficulty of setting manual thresholds to some extent, its core flaw lies in the feature fragmentation caused by the single dimension of temporal modeling. This method can effectively identify waveform segments at a certain moment, but it cannot effectively trace the temporal logic of fault evolution. For example, SVM models have poor generalization ability, with a noise resistance score of only 0.60 in noisy environments, making them prone to misjudgment. While 1D-CNN can extract local features, it ignores the fact that high-resistance grounding faults are a long-term process that includes "early hidden dangers - zero-rest extension - steady-state breakdown". Due to the lack of information in the local spatial dimension and the lack of capturing the bidirectional temporal dependence before and after the fault, it is impossible to effectively establish the logical connection between the fault precursors and the subsequent state, resulting in a high false negative rate and failing to provide a sufficiently reliable early warning basis for high-resistance grounding faults for full life cycle management.

[0087] In summary, none of the relevant technologies have fundamentally resolved the contradiction between "difficulty in feature extraction under strong noise" and "lack of time-series evolution logic" in the process of determining the development stage of high-resistance grounding faults. There is an urgent need for a new technical solution that can break through the limitations of linear analysis and achieve dynamic slow feature extraction and cross-validation of bidirectional time-series modeling.

[0088] In view of this, based on the above embodiments and optional embodiments, this application proposes an optional implementation method for determining the development stage of high-resistance grounding faults in distribution networks. This optional implementation method can be understood as a high-resistance grounding fault identification method for distribution networks based on recursive exponential slow feature analysis and Bi-LSTM. It aims to solve the problems of poor robustness in diagnosing high-resistance grounding faults and their development stages, as well as feature overload or missing temporal logic, caused by the inability of existing data-driven technologies to effectively extract weak fault features from strong noise backgrounds and the difficulty in capturing the long-term temporal dependencies of fault evolution. Its core significance lies in creatively proposing a diagnostic paradigm of "dynamic slow feature extraction + bidirectional temporal modeling": using recursive exponential slow feature analysis (RESFA) as a highly efficient "soft filter" to extract the slowly changing fault essence trend (slow features) from the target zero-sequence current signal containing rapidly changing noise; and using a Bi-LSTM network as a deep parser for temporal logic, using the slow feature sequence enhanced by RESFA to accurately capture and confirm the bidirectional temporal dependencies of high-resistance grounding fault occurrence.

[0089] Figure 2 This is a flowchart of an optional method for determining the development stage of a high-resistance grounding fault in a distribution network, provided according to an embodiment of this application. Figure 2 As shown, the steps of the distribution network high-resistivity grounding fault identification method based on recursive exponential slow feature analysis and Bi-LSTM include:

[0090] Step S1: Acquisition of zero-sequence current signal of distribution network and construction of high-fidelity simulation environment.

[0091] Construct an IEEE (Institute of Electrical and Electronics Engineers) 33-node distribution network simulation model, or deploy high-precision current transformers at actual distribution network monitoring points to collect zero-sequence current signals (e.g., initial training zero-sequence current signals) of distribution network lines. To address the scarcity of high-resistivity grounding fault samples, establish a simulation dataset containing various fault media (such as dry road surfaces, grass, and asphalt) and different noise intensities.

[0092] Step S11: The fault current signal (i.e., the initial training zero-sequence current signal) is generated.

[0093] The improved Mayr arc model is used to simulate the dynamic characteristics of the arc at a high-resistivity grounding fault, and the original zero-sequence current signal sequence containing "zero rest" and waveform distortion features (i.e. the sequence composed of the initial training zero-sequence current signal) is generated.

[0094] Step S12, strong noise environment simulation and data augmentation.

[0095] To simulate harsh real-world operating conditions, the original zero-sequence current signal... High-intensity Gaussian white noise superimposed on the middle A low signal-to-noise ratio input signal (i.e., the target training zero-sequence current signal) is constructed. The input signal is constructed as follows:

[0096]

[0097] in, A zero-sequence current signal is trained for a target period of t. The initial training zero-sequence current signal has a period of t. This is the noise intensity adjustment factor (corresponding to SNR (Signal-to-Noise Ratio) = -5dB to 30dB). It is a random noise sequence that follows a standard normal distribution.

[0098] Step S2: Dynamic slow feature extraction based on Recursive Exponential Slow Feature Analysis (RESFA).

[0099] Online processing is performed on the noisy target training zero-sequence current signal stream obtained in step S1. This step aims to use the principle of slow feature analysis to extract the slowly changing fault essence trend features (i.e., slow features, such as fundamental envelope and zero-sequence evolution trend) from the rapidly changing random noise background.

[0100] Step S21, slow feature map construction.

[0101] Define a linear mapping function to embed the high-dimensional or time-delayed input signal. Mapped to low-dimensional slow features The mapping formula is:

[0102]

[0103] in, This is the slow feature vector extracted at period t (i.e., the slow feature represented in vector form). The input signal vector is a period t (i.e., the target training zero-sequence current signal represented in vector form). Let be the projection weight matrix to be solved.

[0104] Step S22: Construct the exponentially weighted objective function.

[0105] To address the non-stationary characteristics of fault signals in distribution networks, an exponential forgetting factor is introduced. We construct an objective function that minimizes the first-order difference energy of slow feature vectors to find the feature directions that change the slowest. Objective function Defined as:

[0106]

[0107] in, for The exponential forgetting factor of the cycle (ranging from 0.9 to 0.99) is used to strengthen the weight of the slow feature vector of the current cycle and forget the slow feature vector of the previous cycle. for The time derivative of a periodic slow eigenvector can be discretized and approximated as a first-order difference. for Slow eigenvectors with periodicity Periodic index variable. The objective function. A mathematical extremum criterion is defined for extracting the physical objective of "slow features," namely, minimizing the first-order difference energy of the slow feature vector.

[0108] The objective function constructed in step S22 establishes the global optimization criterion for extracting slow features. Since the fault signals (e.g., zero-sequence current signals) corresponding to high-resistivity grounding faults in distribution networks have strongly non-stationary flowing data characteristics, traditional eigenvalue decomposition methods cannot meet the requirements of online real-time computation. Therefore, the solution of the above objective function is transformed into an online stochastic optimization process. Specifically, the instantaneous gradient is derived by taking the partial derivative of the objective function in step S22 with respect to the projection weight matrix, and then the projection weight matrix is ​​iteratively updated. By calculating this instantaneous gradient and substituting it into the stochastic gradient descent framework, the online iterative update of the projection weight matrix is ​​realized, thereby avoiding the offline batch processing delay of traditional slow feature analysis.

[0109] Step S23: Recursive gradient update and orthogonalization.

[0110] The weight matrix is ​​updated online using stochastic gradient descent. Orthogonalization constraints are applied to prevent feature collapse. The weight update formula is as follows:

[0111]

[0112] in, For learning rate, This is the instantaneous gradient of the loss function with respect to the weights. This is the previous projection weight matrix for the corresponding iteration number. This updates the projection weight matrix for the corresponding iteration number. After updating, the Gram-Schmidt algorithm is used to... Orthogonalization is performed to obtain the current projection weight matrix for the corresponding iteration number, ensuring that the extracted slow feature components are uncorrelated, thereby characterizing the fault trend from different dimensions.

[0113] Traditional slow feature analysis typically solves for the objective function directly by performing generalized eigenvalue decomposition on the covariance matrix. This is a static, offline batch processing method. To adapt to the real-time streaming data characteristics of fault signals corresponding to high-resistivity grounding faults in distribution networks, an instantaneous gradient is introduced. By substituting this instantaneous gradient into a stochastic gradient descent framework, online solution of the objective function and dynamic iterative update of the projection weight matrix are achieved.

[0114] Step S3: Bidirectional temporal feature modeling based on Bi-LSTM.

[0115] The slow features extracted in step S2 are input into a bidirectional long short-term memory (Bi-LSTM) network to capture long-distance temporal dependencies before and after a high-resistance grounding fault occurs.

[0116] Step S31, LSTM cell gating calculation.

[0117] For each time step in the input sequence of the model, the LSTM unit controls the information flow through the input gate, forget gate, and output gate. The forget gate outputs at period t. The calculation formula is:

[0118]

[0119] in, It is the Sigmoid activation function. and These are the weight matrix and bias term of the forget gate, respectively. This is a hidden state with a period of t-1. It is a slow characteristic of period t.

[0120] The input gate output at period t With candidate cell state The calculation formula is:

[0121]

[0122]

[0123] in, The candidate cell states for period t; and These are the weight matrices for the input gate and the candidate state, respectively; and These are the corresponding bias terms; This is the hyperbolic tangent activation function, used to scale values ​​to the range [-1, 1].

[0124] Cell state in cycle t The update formula is:

[0125]

[0126] in, This represents the cell state after period t. For the cell state at period t-1, this step enables the long-term retention of historical information and the integration of new information.

[0127] Output gate output with period t and the final hidden state The calculation formula is:

[0128]

[0129]

[0130] in, and These are the weight matrix and bias term of the output gate, respectively; The hidden state output for period t contains filtered temporal feature information and will be passed to the next period and the output layer.

[0131] Step S32, bidirectional feature fusion.

[0132] Construct forward LSTM layers (for processing sequences) respectively Where T is the number of cycles in the input sequence) and the backward LSTM layer (processing the sequence) The two are then concatenated in the hidden state of the last cycle to form a global feature vector. :

[0133]

[0134] in, Features extracted from the forward LSTM layer These are the inverse inference features extracted from the backward LSTM layer.

[0135] Step S4: Offline training of the fault diagnosis model.

[0136] Step S41: Sample library construction and partitioning.

[0137] A sample library containing various operating conditions, including normal operation, line switching, and high-resistance grounding faults (tree branches, ground, insulator flashover), is generated and divided into training and test sets in an 8:2 ratio.

[0138] Step S42, Model Training and Parameter Optimization.

[0139] Construct a fully connected classification layer to integrate the global feature vector. This is mapped to fault category probabilities (i.e., confidence levels). The Negative Log-Likelihood Loss (NLLLoss) function is used as the optimization objective, and the Adam optimizer is used to iteratively update the Bi-LSTM network parameters. Loss Function The formula is:

[0140]

[0141] in, Batch size This represents the number of operating condition categories (e.g., normal / faulty). This is the true label (0 or 1) for the c-th working condition in the i-th batch. Let be the probability of the fault category predicted by the model under the c-th working condition in the i-th batch.

[0142] Step S5: Identification and collaborative decision-making of high-resistance grounding fault development stages throughout the entire life cycle.

[0143] Step S51, real-time probabilistic reasoning.

[0144] The real-time acquired and RESFA-processed data (i.e., the target zero-sequence current signal) is input into the trained Bi-LSTM model (i.e., the target fault diagnosis model), and the target confidence score for the target period is output. .

[0145] Step S52: Divide the development stages of high-resistance grounding faults throughout their entire life cycle.

[0146] Based on target confidence level Based on the time-varying characteristics, the development stages of high-resistance grounding faults in distribution networks are divided into three stages:

[0147] During normal operation (i.e., when there is no high-resistance grounding fault in the distribution network): when When the threshold (i.e., the first threshold, such as 0.1) is reached, it is determined that there is no high-resistance grounding fault in the distribution network, and the distribution network is in normal operation.

[0148] Early stage of potential problem evolution (Pre-fault, i.e., the distribution network is in the high-resistance grounding fault evolution period): when (in When the second threshold (e.g., 0.9) is set and the trend of change is continuously rising, the distribution network is determined to be in the high-resistance grounding fault evolution period. At this time, the electric arc is in an unstable burning state, and a level-two warning is issued.

[0149] Steady-state fault phase (Fault, i.e., the distribution network is in the steady-state period of a high-resistance ground fault): when If the duration exceeds the set threshold (e.g., 3 (i.e., the preset number) cycles), the distribution network is determined to be in the stable period of high-resistance grounding fault. At this time, the distribution network is diagnosed with a high-resistance grounding fault and a first-level alarm is issued.

[0150] Step S6: Generation and reporting of structured fault diagnosis conclusions.

[0151] Based on the final decision result of step S5, a structured diagnostic report is generated.

[0152] Step S61: Generate hierarchical alarms.

[0153] If the fault is determined to be in the "steady-state fault stage", a first-level trip command and fault report will be generated. The report will include the time of occurrence of the high-resistance grounding fault, the arc intensity assessment and the target confidence level. If the fault is determined to be in the "early hidden danger evolution stage", a second-level early warning signal will be generated to remind maintenance personnel to pay attention to abnormal fluctuations in the distribution network lines.

[0154] Step S62: Report the conclusion.

[0155] The diagnostic results are uploaded to the distribution master station system in encrypted message form through the communication module of the distribution automation terminal (FTU / DTU), providing a basis for decision-making for fault isolation and emergency repair.

[0156] It should be noted that, compared with related technologies, this embodiment has at least the following advantages:

[0157] (1) Strong noise-resistant feature extraction based on RESFA: The slow feature analysis theory is innovatively introduced into the transient analysis of power systems, and a recursive exponential update algorithm is proposed for streaming data. Simulation results show that even under extreme conditions with a signal-to-noise ratio as low as -5dB, RESFA can still effectively suppress high-frequency random noise and restore the "zero-sequence" evolution trend of the zero-sequence current signal. Compared with traditional SVM and 1D-CNN methods, the noise resistance index of this method is improved by more than 35% (reaching 0.95).

[0158] (2) Bi-LSTM-based bidirectional time series modeling: By utilizing the bi-directional memory mechanism of Bi-LSTM, the target fault diagnosis model no longer views a single waveform in isolation, but regards the high-resistance grounding fault as a time series with causal logic. This allows for full utilization of the historical information (forward) and evolution trend (backward) of the high-resistance grounding fault, with an F1 score as high as 0.97, which is significantly better than the traditional method's 0.82~0.90, greatly reducing the missed detection rate of high-resistance grounding faults.

[0159] (3) Millisecond-level early warning capability: Thanks to the sensitivity of RESFA to weak features and the timing capture capability of Bi-LSTM, the method in this embodiment can achieve high-confidence identification and early warning of high-resistance grounding faults at the early stage of fault feature appearance (about 0.04s, i.e., within 2 power frequency cycles). This speed is an order of magnitude faster than traditional overcurrent protection, which buys valuable time for fault isolation, arc suppression and prevention of secondary disasters such as wildfires.

[0160] (4) Low computational cost and easy deployment engineering value: The core of the RESFA algorithm only involves matrix multiplication and addition and subtraction, with extremely low computational overhead; the Bi-LSTM model processes the extracted low-dimensional features (only 4 dimensions), with a small network size and fast convergence speed, making the algorithm very suitable for embedding into FTUs, smart circuit breakers or edge computing gateways with limited computing power, and has broad engineering application prospects and promotion value.

[0161] The above-mentioned optional implementation methods achieve at least the following effects: By constructing a collaborative decision-making mechanism of "slow feature extraction" and "temporal memory reasoning", the problems of fault feature ambiguity and memory forgetting caused by traditional linear methods or unidirectional models are overcome. This not only improves the accuracy (F1>0.97) and anti-interference ability of fault identification in low signal-to-noise ratio environments, but also enables fine division and determination of the entire life cycle (normal-early evolution-steady-state fault) of high-resistance grounding faults. This is of great significance for realizing millisecond-level early warning of high-resistance grounding faults in distribution networks, supporting precise and proactive operation and maintenance of distribution networks, and preventing major electrical fire accidents.

[0162] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0163] This embodiment also provides a device for determining the development stage of a high-resistance grounding fault in a distribution network. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0164] According to an embodiment of this application, an apparatus embodiment for implementing a method for determining the development stage of a high-resistance grounding fault in a distribution network is also provided. Figure 3 This is a schematic diagram of a device for determining the development stage of a high-resistance grounding fault in a distribution network, according to an embodiment of this application. Figure 3 As shown, the device for determining the development stage of a high-resistance grounding fault in the power distribution network includes a data acquisition module 302, a first determination module 304, a second determination module 306, and a third determination module 308. The device will be described below.

[0165] Data acquisition module 302 is used to acquire the target zero-sequence current signal of the distribution network in the target period;

[0166] The first determining module 304, connected to the data acquisition module 302, is used to determine the target slow characteristics based on the target zero-sequence current signal, wherein the target slow characteristics are used to describe the evolution trend of high-resistance grounding faults in the distribution network.

[0167] The second determining module 306, connected to the first determining module 304, is used to determine the target confidence level of the target period based on the target slow characteristics, wherein the target confidence level is used to characterize the probability that there is a high-resistance grounding fault in the distribution network;

[0168] The third determining module 308, connected to the second determining module 306, is used to determine the development stage of high-resistance grounding faults in the distribution network based on the target confidence level.

[0169] This application provides a device for determining the development stage of a high-resistance grounding fault in a distribution network. By setting up a data acquisition module 302, a first determination module 304, a second determination module 306, and a third determination module 308, the device aims to determine the target confidence level, which characterizes the probability of a high-resistance grounding fault in the distribution network, based on the target slow characteristics determined by the acquired target zero-sequence current signal of the distribution network. Furthermore, it determines the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level. This achieves the technical effect of improving the accuracy of the target confidence level determination result by determining the target slow characteristics, thereby improving the accuracy of the high-resistance grounding fault development stage determination result in the distribution network. This solves the technical problem of inaccurate determination results of the high-resistance grounding fault development stage in the related art.

[0170] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0171] It should be noted that the data acquisition module 302, the first determining module 304, the second determining module 306, and the third determining module 308 mentioned above correspond to steps S102 to S108 in the embodiments. The instances and application scenarios implemented by the above modules and their corresponding steps are the same, but they are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run in a computer terminal.

[0172] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0173] The device for determining the development stage of high-resistance grounding faults in the aforementioned power distribution network may also include a processor and a memory. The data acquisition module 302, the first determination module 304, the second determination module 306, the third determination module 308, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0174] The processor contains a core that retrieves the corresponding program unit from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0175] This application provides a non-volatile storage medium storing a program that, when executed by a processor, implements a method for determining the development stage of a high-resistance grounding fault in a power distribution network.

[0176] This application provides an electronic device. Figure 4 This is a structural diagram of an electronic device provided according to an embodiment of this application. For example... Figure 4 As shown, the electronic device may include: one or more ( Figure 4 (Only one is shown) a processor 402, a memory 404, a memory controller, and a peripheral interface, wherein the peripheral interface is connected to an RF module, an audio module, and a display. The electronic device includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: acquiring a target zero-sequence current signal of the distribution network in a target period; determining a target slow characteristic based on the target zero-sequence current signal, wherein the target slow characteristic is used to describe the evolution trend of high-resistance grounding faults in the distribution network; determining a target confidence level for the target period based on the target slow characteristic, wherein the target confidence level is used to characterize the probability of a high-resistance grounding fault existing in the distribution network; and determining the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level. The device in this document could be a server, PC, etc.

[0177] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having the following method steps: acquiring a target zero-sequence current signal of the distribution network in a target period; determining a target slow feature based on the target zero-sequence current signal, wherein the target slow feature is used to describe the evolution trend of high-resistance grounding faults in the distribution network; determining a target confidence level for the target period based on the target slow feature, wherein the target confidence level is used to characterize the probability of a high-resistance grounding fault existing in the distribution network; and determining the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level.

[0178] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0179] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0180] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0181] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0182] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0183] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0184] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0185] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0186] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for determining the development stage of a high-resistance grounding fault in a distribution network, characterized in that, include: Acquire the target zero-sequence current signal of the distribution network in the target period; Based on the target zero-sequence current signal, the target slow characteristics are determined, wherein the target slow characteristics are used to describe the evolution trend of high-resistance grounding faults in the distribution network; Based on the target slow characteristic, the target confidence level of the target period is determined, wherein the target confidence level is used to characterize the probability that the distribution network has a high-resistance grounding fault; Based on the target confidence level, the development stage of the high-resistance grounding fault in the distribution network is determined.

2. The method according to claim 1, characterized in that, The determination of the target slow characteristics based on the target zero-sequence current signal includes: Determine the target projection weight matrix, wherein the target projection weight matrix is ​​used to linearly map the high-dimensional zero-sequence current signal into a low-dimensional slow feature; Based on the target projection weight matrix and the target zero-sequence current signal, the slow characteristics of the target are determined.

3. The method according to claim 2, characterized in that, The determination of the target projection weight matrix includes: Acquire multiple initial training zero-sequence current signals of the power distribution network arranged in order from front to back according to the acquisition period; The multiple initial training zero-sequence current signals are respectively subjected to noise addition processing to obtain multiple target training zero-sequence current signals; The target projection weight matrix is ​​determined based on the zero-sequence current signals trained on the multiple targets.

4. The method according to claim 3, characterized in that, The step of determining the target projection weight matrix based on the zero-sequence current signals trained on the multiple targets includes: For any target training zero-sequence current signal among the plurality of target training zero-sequence current signals, based on any target training zero-sequence current signal and the previous projection weight matrix, the current slow feature corresponding to the current iteration number is determined, wherein the previous projection weight matrix is ​​determined in the previous iteration number before the current iteration number; Based on the current slow feature, the zero-sequence current signal of any target training, and the previous slow feature, the current instantaneous gradient corresponding to the current iteration number is determined. The previous slow feature is determined in the previous iteration number based on the previous target training zero-sequence current signal located before the zero-sequence current signal of any target training among the multiple target training zero-sequence current signals. The current instantaneous gradient is used to characterize the direction and magnitude of the adjustment of the projection weight matrix in order to minimize the rate of change of the slow feature. Based on the current instantaneous gradient, the previous projection weight matrix is ​​iteratively updated to obtain the updated projection weight matrix; The updated projection weight matrix is ​​orthogonalized to obtain the current projection weight matrix corresponding to the current iteration number; Based on the next target training zero-sequence current signal located after any one of the target training zero-sequence current signals among the multiple target training zero-sequence current signals, the current projection weight matrix is ​​iteratively updated by updating the previous projection weight matrix until the iteration stopping condition is met, and the projection weight matrix obtained by the last iteration update is determined as the target projection weight matrix.

5. The method according to claim 1, characterized in that, Determining the target confidence level of the target period based on the target slow feature includes: Based on the target slow features, a target fault diagnosis model is used to obtain the target confidence level, wherein the target fault diagnosis model has pre-learned the correlation between the target slow features and the target confidence level.

6. The method according to any one of claims 1 to 5, characterized in that, The determination of the high-resistance grounding fault development stage of the distribution network based on the target confidence level includes: If the target confidence level is less than the first threshold, the high-resistance grounding fault development stage is determined to be that there is no high-resistance grounding fault in the distribution network. If the target confidence level is greater than or equal to the first threshold and less than the second threshold, and the trend of the target confidence level is an upward trend, then the high-resistance ground fault development stage is determined to be the high-resistance ground fault evolution period of the distribution network. The trend is obtained based on the target confidence level and the historical confidence level determined by the historical period before the target period. If the target confidence level is greater than or equal to the second threshold, and the historical confidence level of a preset number of historical periods before the target period is greater than or equal to the second threshold, then the high-resistance ground fault development stage is determined to be the high-resistance ground fault stable period of the distribution network.

7. A device for determining the development stage of a high-resistance grounding fault in a power distribution network, characterized in that, include: The data acquisition module is used to acquire the target zero-sequence current signal of the distribution network in the target period; The first determining module is used to determine the target slow characteristics based on the target zero-sequence current signal, wherein the target slow characteristics are used to describe the evolution trend of the high-resistance grounding fault in the distribution network; The second determining module is used to determine the target confidence level of the target period based on the target slow characteristics, wherein the target confidence level is used to characterize the probability that the distribution network has a high-resistance grounding fault; The third determining module is used to determine the development stage of the high-resistance grounding fault in the distribution network based on the target confidence level.

8. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores multiple instructions, which are adapted to be loaded by a processor and executed by the method for determining the development stage of a high-resistance grounding fault in a distribution network as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, include: One or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for determining the development stage of a high-resistance grounding fault in a distribution network as described in any one of claims 1 to 6.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the method for determining the development stage of high-resistance grounding faults in a distribution network as described in any one of claims 1 to 6.

Images