A primary distribution network fault identification and positioning method and related equipment

Abstract

This invention provides a single-level distribution network fault identification and location method and related equipment, comprising: modeling the topology of the target distribution network under fault conditions, and obtaining the fault transient signal of the topology; the fault transient signal includes transient voltage signal and transient current signal; establishing the optimal time-frequency joint distribution expression of the fault transient signal to extract shallow feature quantities of the fault transient signal; constructing a fault location model, inputting the topology, fault transient signal and shallow feature quantities into the fault location model for feature extraction and feature fusion, to obtain the fault identification result and fault location result of the target distribution network; it can extract the spatiotemporal fault features of the distribution network topology and fault transient signal, and complete the fault type identification, fault section location and fault point precise location by integrating a multilayer sensor, realizing single-level fault identification and location, greatly reducing the fault location time and improving the fault location speed.

Description

Technical Field

[0001] This invention relates to the field of power distribution network fault identification and location technology, and in particular to a method and related equipment for primary power distribution network fault identification and location. Background Technology

[0002] High penetration rates of new energy sources and a high proportion of power electronic equipment connected to the grid are the main characteristics of the new generation of power grids. Compared with transmission networks, distribution networks have characteristics such as frequent topology changes, complex disturbances, and diverse loads. The large-scale grid connection of distributed new energy sources further increases the complexity, variability, and uncertainty of the distribution network topology, as well as the harmonics and noise introduced by power electronic equipment. This makes traditional protection systems in distribution networks prone to false tripping and failure to trip. The speed and accuracy of fault location have always been bottleneck problems restricting the fault location technology of distribution systems.

[0003] Existing fault location methods can only locate fault sections when faced with distribution network faults that are structurally variable, operationally variable, and have diverse fault scenarios. The location range is large, and the fault point cannot be accurately located, which leads to increased fault clearing time and seriously endangers the safe and stable operation of the power system.

[0004] Rapid and accurate fault type identification is a prerequisite for restoring power supply to the distribution network. Once a fault occurs in the distribution network, accurate fault type identification is of great significance for fault line selection, fault point location, rapid repair of faulty lines, improving power supply reliability and minimizing power outage losses.

[0005] Existing technologies include: extracting fault signal features using artificial neural networks and combining them with Support Vector Machines (SVM) to identify distribution network fault types; constructing membership functions based on the phase angle relationship of the positive and negative sequence components of three-phase currents, then establishing fuzzy rules and using fuzzy reasoning to classify faults in unbalanced radial distribution networks; proposing a distribution network fault classification method based on an adaptive neural fuzzy reasoning system and analyzing its adaptability under four different operating conditions; proposing a distribution network fault classification method based on binary ant colony algorithm and fuzzy neural network, which overcomes the shortcomings of slow search speed and easy getting trapped in local minima; and proposing a distribution network fault type identification method based on grey relational analysis. However, the aforementioned methods for identifying distribution network fault types are greatly affected by the resistance of fault nodes and the system operating mode. Furthermore, in distribution networks containing distributed renewable energy sources, they fail to effectively integrate distribution network topology and electrical quantity information to improve the feature recognition capability of the classification model and cannot adapt to dynamic changes in the topology.

[0006] To achieve accurate fault location in distribution networks, it is essential to thoroughly train the system with fault data under various fault scenarios, learning the combination patterns of various influencing factors, such as fault type, fault location, transition resistance, fault phase angle, load operation changes, distributed renewable energy switching, and random output changes. The number of sample sets generated by traversing these various fault scenarios is enormous. Extracting deep-level, highly discriminative fault features from such a massive dataset becomes crucial for fault location in distribution networks.

[0007] Deep learning, evolved from machine learning and multi-layer neural networks, is at the forefront of current big data processing and analysis research. It possesses unique advantages in handling high-dimensional, large-scale, nonlinear data with indistinct features and strong correlations, demonstrating a powerful ability to capture representative features from raw data and obtain more robust nonlinear representations. Since its inception in 2006, deep learning has achieved significant application results in speech recognition, computer vision, and natural language processing. However, research on its application in power system fault location started relatively late and is still in its developmental stage.

[0008] This paper compares and analyzes existing fault location methods for power transmission and distribution networks, and explores the unique advantages of deep learning-based machine learning algorithms in fault location. It utilizes convolutional neural networks to detect faulty lines in distribution networks. Addressing the issue of low fault location accuracy in distribution networks, it proposes using two different convolutional neural networks to locate fault sections and fault points respectively. The research results demonstrate that introducing deep learning into fault identification and location in large-scale complex power grids can effectively solve the challenges of high-dimensional data mining and feature extraction in complex power grids, and effectively eliminate the uncertainties caused by manual feature extraction and selection in traditional intelligent algorithms. However, the aforementioned deep learning-based fault location methods only utilize electrical quantity signals and do not consider the impact of distribution network topology on fault location effectiveness, thus failing to adapt to dynamic changes in distribution network topology.

[0009] Recently, Gori et al. proposed graph neural networks, which can directly aggregate and extract features from nodes, edges, and the graph as a whole, preserving the topological structure and node attribute information of the graph to the greatest extent, and can be applied to the field of graph data. It has achieved good application results in many structured scenarios (such as social networks, physical systems, compound molecules, knowledge graphs, etc.). In graph neural networks, the information transmission between nodes is constrained by topological connections, which is similar to the energy flow in the physical model of a power system. Therefore, the topological connections of a power system can be directly modeled into a graph neural network. However, research on graph neural networks in fault location is scarce and still in its early stages of development; existing technologies have proposed methods for fault location in distribution networks based on graph attention networks. The aforementioned fault location methods using graph neural networks can only extract spatial features of graph data to achieve segment location, but cannot extract the temporal characteristics and spatiotemporal correlation characteristics of graph data. Moreover, they cannot escape the uninterpretability inherent in deep networks.

[0010] In summary, with the grid connection of large-scale distributed renewable energy sources and the introduction of a large number of power electronic devices, the accurate location of distribution networks faces new problems and challenges, mainly including: 1) The grid connection of distributed renewable energy sources causes the distribution network topology to become complex and variable, and the transient state of faults is random and complex, making it difficult to accurately extract fault features; 2) The fault features of random noise interference introduced by power electronic devices cannot be reliably extracted; 3) Traditional fault location methods based on transient measurement information only use transient signals for feature extraction and cannot adapt to the dynamic changes in the distribution network topology. Summary of the Invention

[0011] This invention provides a method and related equipment for primary distribution network fault identification and location, with the aim of improving the accuracy of fault identification and location in distribution networks.

[0012] To achieve the above objectives, the present invention provides a method for primary distribution network fault identification and location, comprising:

[0013] Step 1: Model the topology of the target distribution network under fault conditions and obtain the fault transient signals of the topology; the fault transient signals include transient voltage signals and transient current signals.

[0014] Step 2: Establish the optimal time-frequency joint distribution expression of the fault transient signal, and extract the shallow feature quantity of the fault transient signal through the optimal time-frequency joint distribution expression;

[0015] Step 3: Construct a fault location model. Input the topology, fault transient signals, and shallow feature quantities into the fault location model for feature extraction and feature fusion to obtain the fault identification and fault location results of the target distribution network.

[0016] The fault location model includes an input layer, a shared layer, a fully connected layer, and a multilayer perceptron. The output of the input layer is connected to the input of the shared layer, the output of the shared layer is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the multilayer perceptron.

[0017] Furthermore, the shared layer includes: a graph attention network for extracting spatial features and a long short-term memory network for extracting temporal features;

[0018] The graph attention network includes a first graph attention layer, a second graph attention layer, a third graph attention layer, and a first fully connected layer connected in sequence;

[0019] The input terminal of the attention layer in the first diagram is connected to the first output terminal of the input layer as the input terminal of the shared layer.

[0020] The Long Short-Term Memory (LSTM) network comprises multiple LSM modules connected to a second fully connected layer. The multiple LSM modules are connected sequentially. The inputs of the multiple LSM modules are all connected to the outputs of the first fully connected layer. The outputs of the multiple LSM modules are all connected to the inputs of the second fully connected layer. The input of the second fully connected layer is connected to the second output of the input layer. The output of the second fully connected layer is connected to the input of the fully connected layer as the output of the shared layer.

[0021] Furthermore, the topology, fault transient signals, and shallow features are all input into the input layer. The topology and fault transient signals are input into the graph attention network, and features are extracted using the first, second, and third graph attention layers. The features are then aggregated through the first fully connected layer to obtain the spatial features of the fault transient signals.

[0022] Spatial features are input into the long short-term memory module to mine the variation pattern of fault transient signals in the time dimension and obtain the temporal features of fault transient signals;

[0023] The temporal features and shallow features are input into the second fully connected layer for aggregation to obtain the aggregation result.

[0024] The aggregation results are input to the multilayer sensor through the fully connected layer for fault identification and location, resulting in fault identification and location results for the target distribution network.

[0025] Furthermore, the fully connected layer includes a third fully connected layer, a fourth fully connected layer, and a fifth fully connected layer;

[0026] The multilayer sensor includes a first multilayer sensor for identifying fault types, a second multilayer sensor for locating fault sections, and a third multilayer sensor for locating fault points.

[0027] The inputs of the third, fourth, and fifth fully connected layers are all connected to the output of the second fully connected layer. The output of the third fully connected layer is connected to the input of the first multilayer sensor. The output of the fourth fully connected layer is connected to the input of the second multilayer sensor. The output of the fifth fully connected layer is connected to the input of the third multilayer sensor. The outputs of the first, second, and third multilayer sensors all serve as the outputs of the fault location model.

[0028] Furthermore, modeling the topology of the target distribution network under fault conditions includes:

[0029] The electrical nodes and lines of the target distribution network are equivalent to the vertices and edges of a graph;

[0030] The topology of the target distribution network is obtained by modeling the topology under fault conditions using an adjacency matrix.

[0031] Furthermore, the process before step 2 also includes:

[0032] Time-domain denoising of fault transient signals;

[0033] The time-domain denoised transient fault signal is subjected to deviation standardization processing;

[0034] Establish the optimal time-frequency joint distribution expression for the fault transient signal after deviation standardization.

[0035] Furthermore, the fault transient signal after deviation standardization is:

[0036]

[0037] Among them, z k This is the fault transient signal after filtering and noise reduction.

[0038] The present invention also provides a primary distribution network fault identification and location device, comprising:

[0039] The acquisition module is used to model the topology of the target distribution network under fault conditions and acquire the fault transient signals of the topology; the fault transient signals include transient voltage signals and transient current signals;

[0040] The shallow feature extraction module is used to establish the optimal time-frequency joint distribution expression of the fault transient signal and extract the shallow feature quantity of the fault transient signal through the optimal time-frequency joint distribution expression;

[0041] The feature extraction and fusion module is used to construct the fault location model. It inputs the topology, fault transient signals and shallow feature quantities into the fault location model for feature extraction and feature fusion to obtain the fault identification results and fault location results of the target distribution network.

[0042] The fault location model includes an input layer, a shared layer, a fully connected layer, and a multilayer perceptron. The output of the input layer is connected to the input of the shared layer, the output of the shared layer is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the multilayer perceptron.

[0043] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for identifying and locating faults in a primary distribution network.

[0044] The present invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for identifying and locating faults in a primary power distribution network.

[0045] The above-described solution of the present invention has the following beneficial effects:

[0046] This invention models the topology of a target distribution network under fault conditions to obtain transient fault signals of the topology. These transient signals include transient voltage and current signals. An optimal time-frequency joint distribution expression for the transient fault signals is established, and shallow features of the transient fault signals are extracted using this expression. A fault location model is constructed, and the topology, transient fault signals, and shallow features are input into the model for feature extraction and fusion to obtain fault identification and location results for the target distribution network. The fault location model extracts spatiotemporal fault features by fusing the distribution network topology and corresponding measurement point transient fault signals. By integrating a multilayer sensor, fault type identification, fault section location, and precise fault point location are achieved, enabling fault identification and location in a primary distribution network. This significantly reduces fault location time, increases fault location speed, and enhances the adaptability and robustness of the location algorithm to changes in the distribution network topology.

[0047] Other beneficial effects of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating an embodiment of the present invention;

[0049] Figure 2 This is a model structure diagram of an embodiment of the present invention. Detailed Implementation

[0050] To make the technical problems, solutions, and advantages of this invention clearer, a detailed description will be provided below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0051] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0052] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a locking connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0053] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0054] This invention addresses existing problems by providing a method and related equipment for primary distribution network fault identification and location.

[0055] like Figure 1 As shown, an embodiment of the present invention provides a method for fault identification and location in a primary distribution network, comprising:

[0056] Step 1: Model the topology of the target distribution network under fault conditions and obtain the fault transient signals of the topology; the fault transient signals include transient voltage signals and transient current signals.

[0057] Step 2: Establish the optimal time-frequency joint distribution expression of the fault transient signal, and extract the shallow feature quantity of the fault transient signal through the optimal time-frequency joint distribution expression;

[0058] Step 3: Construct a fault location model. Input the topology, fault transient signals, and shallow feature quantities into the fault location model for feature extraction and feature fusion to obtain the fault identification and fault location results of the target distribution network.

[0059] Specifically, modeling the topology of the target distribution network under fault conditions includes:

[0060] The electrical nodes and lines of the target distribution network are equivalent to the vertices and edges of a graph;

[0061] The topology of the target distribution network is obtained by modeling the topology under fault conditions using an adjacency matrix.

[0062] In this embodiment of the invention, various typical distribution network models are built using EMTP and MATLAB simulation platforms and a distribution network laboratory to simulate the network topology of actual distribution networks under various fault scenarios. A signal acquisition module is used to detect fault transient signals under different topologies, operating modes, fault types, fault locations, transition resistances, initial phase angles, distributed renewable energy switching and output random changes, load fluctuations, and noise interference. A 50MHz digital oscilloscope is used to record multiple sets of measured fault signal waveforms to identify the differences in fault transient signal variations under different fault conditions. Fault characteristics are analyzed in the time domain, frequency domain, and time-frequency domain. The characteristics of distribution network faults involving distributed renewable energy are compared with those of traditional distribution network faults. The characteristic parameters of various faults are analyzed to reveal the mapping mechanism between the integrated network topology and the corresponding measurement point fault transient signals, fault types, and fault locations. By comparing and analyzing the differences between simulated fault transient signals and field-measured fault transient signals, we can further study the mechanism of these differences, improve and refine the theoretical research results, improve and revise the distribution network fault simulation model, and construct a distribution network fault rule knowledge base to provide a complete theoretical basis for the rapid identification and accurate location of distribution network faults.

[0063] The topology of the target distribution network is modeled based on graph theory. The electrical nodes and lines of the target distribution network are equivalent to the vertices and edges of the graph. The topology of the distribution network is represented by an adjacency matrix. In the structure, the node characteristics are determined by their own characteristics and the characteristics of the nodes connected to them. When the topology nodes or branches of the distribution network change, the new network adjacency matrix and node characteristics will also change accordingly.

[0064] Specifically, the steps preceding step 2 also include:

[0065] Time-domain denoising of fault transient signals;

[0066] The time-domain denoised transient fault signal is subjected to deviation standardization processing;

[0067] Establish the optimal time-frequency joint distribution expression for the fault transient signal after deviation standardization.

[0068] Because fault transient signals contain rich fault information and can essentially reflect fault characteristics, they have advantages over switch quantity information, which has problems such as missed reporting and false tripping, in terms of fault tolerance, accuracy and completeness. However, the noise from power electronic equipment seriously interferes with the extraction of transient signal features, so noise reduction processing is required.

[0069] The transient voltage or current signals acquired by the measurement system are often interfered with by various random noises, which can be described as:

[0070] y k =z k +v k (1)

[0071] Among them, y k Represents a transient signal (voltage or current signal), {y k ∈R, k=1,2,…,K}, where K is the sampling length. Z represents Gaussian random white noise. k The precise mathematical expression for a pure fault transient signal is often very complex. In practical applications, it can be represented as the sum of the fundamental component, multiple harmonic components, and the attenuated DC component, i.e.:

[0072]

[0073] Where N represents the highest order of the harmonic, and A n φ n ω represents the amplitude and phase of the fundamental wave (n=1) or the nth harmonic, respectively. n = 2πnf0, where f0 represents the fundamental frequency, typically taken as 50Hz, T s Indicates the sampling period, A d exp(-αkT s ) represents the DC component.

[0074] The DC component A in the above formula d exp(-αkT s Using a Taylor series approximation, we can obtain z k :

[0075]

[0076] To use the Kalman filter algorithm to filter and denoise transient fault signals, state-space modeling is required, and a 2N+2 dimensional state variable X is selected. k =[x 1,k ,x 2,k ,…,x2N-1,k ,x 2N,k ,x 2N+1,k ,x 2N+2, k] T The details are as follows:

[0077] X k =[A1cosφ1,A1sinφ1,…,A n cosφ n A n sinφ n ,…,A N cosφ N A N sinφ N A d A d α] T (4)

[0078] The following state-space model can then be established:

[0079]

[0080] Where, η k v k Let H represent the system process noise and measurement noise, respectively, which follow Gaussian random processes with mean 0 and variances Q and R, respectively. k The system observation matrix is ​​represented by the following:

[0081] H k =[sin(ω1kT)] s ),cos(ω1kT s ),…,sin(ω n kT s ),cos(ω n kT s ),…,sin(ω N kT s ),cos(ω N kT s ),1,-kT s (6)

[0082] When a fault occurs, the corresponding transient signal undergoes a sudden change. Traditional Kalman filtering algorithms cannot respond to this sudden change in a timely manner, thus losing their ability to track it promptly. Therefore, in this embodiment of the invention, a strong tracking Kalman filtering algorithm with a time-varying fading factor is employed to achieve fast and accurate tracking of the sudden change components of the fault transient signal and to extract conventional feature quantities such as amplitude and phase.

[0083]

[0084] Among them, yk|k A is the filtered and denoised fault transient signal. n φ n These are the extracted amplitude and phase, respectively. Representing state variables x respectively 2n-1,k x 2n,k The estimated value.

[0085] This algorithm introduces a time-varying fading factor to adjust the state prediction error covariance matrix and gain matrix in real time, keeping the residual sequence approximately orthogonal. This maximizes the extraction of effective information from the residuals and improves the Kalman filter algorithm's tracking ability and robustness in the face of sudden signal changes.

[0086] As can be seen from the state-space model, the estimated residual sequence e k The difference between the actual measured value and the estimated value of the output signal is:

[0087]

[0088] in, State variable X k The estimated value.

[0089] When the distribution network is operating normally, the filter residuals are a set of zero-mean, uncorrelated random variables that approximately follow a normal distribution, and the absolute value of the residuals |e k |Very small; when a fault occurs in the distribution network, the expected value of the filter residual is not zero, its distribution no longer follows a normal distribution, and the absolute value of the residual is |e k The magnitude is relatively large. Therefore, based on the normal distribution single-sample hypothesis testing theory, using the filter residual as a condition for the test statistic, it is possible to quickly and effectively detect whether a fault has occurred. It can effectively deal with the detection of weak faults (such as high-resistance grounding faults) and has high sensitivity.

[0090] To reduce the impact of differences in data magnitude, deviation normalization is applied to the filtered fault transient signal, i.e.:

[0091]

[0092] After standardization, the data from different dimensions are numerically comparable, which can greatly improve the model convergence speed and accuracy, and improve the accuracy of fault identification and location.

[0093] This invention employs time-domain analysis to extract time-domain fault features such as phase, time, amplitude, and decay factor from transient fault signals. Based on the time-frequency plot, energy density spectrum, and marginal spectrum of the transient fault signal, it proposes an optimal time-frequency joint distribution expression for the transient fault signal, representing the one-dimensional time signal as a time-frequency joint distribution based on an adaptive kernel function.

[0094]

[0095] Where g(θ,τ) is the kernel function, A y (θ,τ) is a fuzzy function, defined as:

[0096]

[0097] Where y(u) is the original signal z k Given different g(θ,τ) values ​​for the analytic signal, different forms of time-frequency distributions can be obtained.

[0098] To obtain the optimal time-frequency joint expression of the signal, we only need to solve for the optimal kernel function of the signal, that is, to solve the following optimization problem:

[0099]

[0100] Based on the above optimal time-frequency joint distribution expression, shallow fault characteristics such as frequency, amplitude, polarity, and energy of the fault transient signal can be extracted from the time-frequency domain.

[0101] Specifically, such as Figure 2 As shown, in order to improve the feature recognition capability of the location model, this embodiment of the invention introduces the distribution network topology when performing deep feature extraction on the fault transient signal, and constructs a fault location model that integrates the distribution network topology and the fault transient signal of the corresponding measurement point, including: an input layer, a shared layer, a fully connected layer and a multilayer sensor. The output end of the input layer is connected to the input end of the shared layer, the output end of the shared layer is connected to the input end of the fully connected layer, and the output end of the fully connected layer is connected to the input end of the multilayer sensor, so as to fully explore the spatiotemporal fault features.

[0102] The shared layer includes a graph attention network for extracting spatial features and a long short-term memory network for extracting temporal features; the graph attention network includes a first graph attention layer, a second graph attention layer, a third graph attention layer and a first fully connected layer connected in sequence;

[0103] The input terminal of the attention layer in the first diagram is connected to the first output terminal of the input layer as the input terminal of the shared layer.

[0104] The Long Short-Term Memory (LSTM) network comprises multiple LSM modules connected to a second fully connected layer. The multiple LSM modules are connected sequentially. The inputs of the multiple LSM modules are all connected to the outputs of the first fully connected layer. The outputs of the multiple LSM modules are all connected to the inputs of the second fully connected layer. The input of the second fully connected layer is connected to the second output of the input layer. The output of the second fully connected layer is connected to the input of the fully connected layer as the output of the shared layer.

[0105] Specifically, the topology, fault transient signal, and shallow feature quantities are all input into the input layer. The topology and fault transient signal are input into the graph attention network, and feature extraction is performed using the first graph attention layer, the second graph attention layer, and the third graph attention layer. The spatial features of the fault transient signal are then aggregated through the first fully connected layer.

[0106] Spatial features are input into the long short-term memory module to mine the variation pattern of fault transient signals in the time dimension and obtain the temporal features of fault transient signals;

[0107] The temporal features and shallow features are input into the second fully connected layer for aggregation to obtain the aggregation result.

[0108] The aggregation results are input to the multilayer sensor through the fully connected layer for fault identification and location, resulting in fault identification and location results for the target distribution network.

[0109] In this embodiment of the invention, the fault location model integrates a two-level network structure: a graph attention network and a long short-term memory network. The former can effectively fuse the distribution network topology (adjacency matrix) and the corresponding fault transient signals (node ​​features) of the measurement points. By using the graph attention layer to obtain new node features through aggregation operations between nodes, it effectively captures the mutual influence generated by the information flow between nodes and realizes the extraction of data spatial correlation features. The latter can process data according to time, and obtain the change pattern of data in the time dimension through layer-by-layer training, fully mining the relevant features in time.

[0110] When training the fault location model, its input consists of two parts: the first part is the topological structure information of the graph, i.e., the adjacency matrix; the second part is the node features corresponding to the graph structure, i.e., the three-phase transient voltage and current. The input of the graph neural network integrates the distribution network topology and the corresponding fault transient signals at the measurement points, achieving effective fusion of the topological graph structure and feature data. It fully considers the spatial correlation of transient signals from multiple measurement points, and improves the feature recognition capability and accuracy of the location model based on the fusion of the distribution network topology.

[0111] Meanwhile, to better adapt to dynamic changes in distribution network topology, an attention mechanism is introduced into the graph neural network. When the distribution network topology changes, the adjacency matrix and node characteristics of the new network change, and the interaction between nodes also changes accordingly. The attention mechanism can adaptively assign different attention coefficients (weights) to neighboring nodes based on their neighborhood characteristics (the correlation of fault characteristics between adjacent nodes), thereby capturing the differences in the degree of influence between nodes, mining the spatial correlation of the topology, and thus optimizing node characteristics to reflect the spatial correlation of transient signals. Therefore, the graph neural network with the attention mechanism pays more attention to the electrical characteristics of neighboring nodes and can adaptively adjust the attention coefficients between nodes when the distribution network topology changes, enhancing the model's generalization ability and robustness on different distribution network structures.

[0112] Furthermore, the extracted spatial feature information is transformed into time-corresponding vectors, and long short-term memory networks are used to mine its time-related potential features. This enables in-depth mining of the essential features of the distribution network topology and corresponding measurement point fault transient signals in both the spatial and temporal domains, thereby improving the feature recognition capability of the fault location model.

[0113] The key to how fault location models learn deep-seated, inherent characteristics from massive spatiotemporal data lies in the design of the model's structural depth (including the number of network layers and hidden layer nodes). This invention provides a multi-dimensional model evaluation system that considers minimizing error (the difference between the model's actual output value and the ideal output value) and the degree of error whitening. An adaptive hidden layer optimization mechanism is proposed to determine the number of model layers. After determining the number of layers, the number of hidden layer nodes is further determined by studying the relationship between the number of hidden layer nodes and the estimation accuracy.

[0114] Because existing deep learning networks generally suffer from overfitting, meaning the model has good accuracy but poor generalization ability, this invention proposes a regularization parameter optimization method by adding an L2 regularization term to the original loss function.

[0115]

[0116] In the formula, L represents the number of training data pairs, and Y represents the expected output data vector. The output vector of the model represents the training data. λ represents the regularization coefficient, which controls the relative importance of the two terms in the above formula. W represents all the weight parameters of the model. The first term is the mean squared error term, and the second term is the regularization of the model's weight parameters W, i.e., the network weight decay term. Its purpose is to reduce the magnitude of weight changes and prevent the learned network from overfitting.

[0117] Through the above-mentioned fault location modeling and optimization learning, the spatiotemporal characteristics of graph data (distribution network topology and transient electrical signals) can be automatically learned. By introducing the distribution network topology, the feature recognition capability of the model is improved. At the same time, in order to cope with the dynamic changes of the topology in the fault location algorithm, an attention mechanism is introduced. Nodes with different degrees of influence are assigned different weight parameters, thereby capturing the different degrees of interaction between electrical nodes and enhancing the model's generalization ability and robustness on different distribution network structures.

[0118] Finally, to further highlight fault information, a feature enhancement technique that integrates conventional shallow features and deep features is proposed, which to some extent eliminates redundant and spurious features in the original high-dimensional features and improves the clustering ability of the features.

[0119] Specifically, the fully connected layer includes the third fully connected layer, the fourth fully connected layer, and the fifth fully connected layer;

[0120] The multilayer sensor includes a first multilayer sensor for identifying fault types, a second multilayer sensor for locating fault sections, and a third multilayer sensor for locating fault points.

[0121] The inputs of the third, fourth, and fifth fully connected layers are all connected to the output of the second fully connected layer. The output of the third fully connected layer is connected to the input of the first multilayer sensor. The output of the fourth fully connected layer is connected to the input of the second multilayer sensor. The output of the fifth fully connected layer is connected to the input of the third multilayer sensor. The outputs of the first, second, and third multilayer sensors all serve as the outputs of the fault location model.

[0122] In this embodiment of the invention, three multilayer perceptrons are used to perform classification and regression of multiple different sub-tasks, thereby improving the speed and accuracy of fault identification and location in power distribution networks. The first two multilayer perceptrons are used for category prediction, with outputs as discrete variables, and respectively complete fault type identification and fault branch (section) location; the last multilayer perceptron is used for regression prediction, with outputs as continuous variables, and completes the precise location of the fault point.

[0123] In this invention, a multilayer perceptron, which can be a classifier, is used. The first classifier identifies fault types sensitively by mining the nonlinear mapping relationship between the integrated network topology, corresponding measurement point fault transient signals, and fault types. Its output includes 10 different types of faults (A-phase ground fault, B-phase ground fault, C-phase ground fault, AB-phase ground fault, AC-phase ground fault, BC-phase ground fault, AB-phase fault, AC-phase fault, BC-phase fault, and ABC-phase fault). The second classifier locates the fault branch or section by mining the nonlinear mapping relationship between the integrated network topology, corresponding measurement point fault transient signals, and fault branches. Its output is the branch or section where the fault is located, with the output dimension being the number of branches or sections in the distribution network system. The third regressor accurately locates the fault point by mining the nonlinear mapping relationship between the integrated network topology, corresponding measurement point fault transient signals, and fault location. Its output is the precise fault location.

[0124] Based on the fact that the attention mechanism can effectively capture the correlation between input data and output results, the attention coefficients learned by the hidden layer of the network are visualized to form a heat map. The reliability of the model and the guiding role of the attention mechanism in model training are analyzed through the heat map, and the understandable fault information of the model is mined out. The potential correlation between this understandable fault information and the fault location is explored, and a reasonable explanation is given for the fault location result, so as to realize the retrospective analysis of the fault location result.

[0125] Compared to traditional two-stage distribution network fault location algorithms, this invention model the topology of the target distribution network under fault conditions to obtain the transient fault signals of the topology. These transient signals include transient voltage and current signals. An optimal time-frequency joint distribution expression for the transient fault signals is established, and shallow features of the transient fault signals are extracted using this expression. A fault location model is constructed, and the topology, transient fault signals, and shallow features are input into the model for feature extraction and fusion to obtain the fault identification and location results for the target distribution network. The fault location model extracts spatiotemporal fault features that fuse the distribution network topology and corresponding measurement point transient fault signals. By integrating a multilayer sensor, fault type identification, fault section location, and precise fault point location are achieved, realizing a single-stage fault identification and location. This significantly reduces fault location time, increases fault location speed, and enhances the adaptability and robustness of the location algorithm to changes in the distribution network topology.

[0126] This invention also provides a primary distribution network fault identification and location device, comprising:

[0127] The acquisition module is used to model the topology of the target distribution network under fault conditions and acquire the fault transient signals of the topology; the fault transient signals include transient voltage signals and transient current signals;

[0128] The shallow feature extraction module is used to establish the optimal time-frequency joint distribution expression of the fault transient signal and extract the shallow feature quantity of the fault transient signal through the optimal time-frequency joint distribution expression;

[0129] The feature extraction and fusion module is used to construct the fault location model. It inputs the topology, fault transient signals and shallow feature quantities into the fault location model for feature extraction and feature fusion to obtain the fault identification results and fault location results of the target distribution network.

[0130] The fault location model includes an input layer, a shared layer, a fully connected layer, and a multilayer perceptron. The output of the input layer is connected to the input of the shared layer, the output of the shared layer is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the multilayer perceptron.

[0131] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments in the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of the embodiments of the present invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0133] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for identifying and locating faults in a primary distribution network.

[0134] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the embodiments of the present invention can implement all or part of the processes in the methods described above by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a building device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0135] This invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for identifying and locating faults in a primary power distribution network.

[0136] It should be noted that the terminal device can be a mobile phone, tablet computer, laptop computer, Ultra-mobile Personal Computer (UMPC), netbook, Personal Digital Assistant (PDA), etc. For example, the terminal device can be a station (ST) in a WLAN, a cellular phone, cordless phone, Session Initiation Protocol (SIP) phone, Wireless Local Loop (WLL) station, PDA, handheld device with wireless communication capabilities, computing device or other processing device connected to a wireless modem, computer, laptop computer, handheld communication device, handheld computing device, satellite wireless device, etc. This embodiment of the invention does not impose any restrictions on the specific type of terminal device.

[0137] The processor referred to can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0138] In some embodiments, the memory may be an internal storage unit of the terminal device, such as a hard drive or RAM. In other embodiments, the memory may be an external storage device of the terminal device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital Card (SD), or Flash Card. Furthermore, the memory may include both internal and external storage units of the terminal device. The memory is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory can also be used to temporarily store data that has been output or will be output.

[0139] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0140] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for fault identification and location in a primary distribution network, characterized in that, include: Step 1: Model the topology of the target distribution network under fault conditions and obtain the fault transient signal of the topology. The fault transient signal includes transient voltage signal and transient current signal; Step 2: Establish the optimal time-frequency joint distribution expression of the fault transient signal, and extract the shallow feature quantity of the fault transient signal through the optimal time-frequency joint distribution expression; Step 3: Construct a fault location model. Input the topology, the fault transient signal, and the shallow feature quantity into the fault location model for feature extraction and feature fusion to obtain the fault identification result and fault location result of the target distribution network. The fault location model includes an input layer, a shared layer, a fully connected layer, and a multilayer sensor. The output of the input layer is connected to the input of the shared layer, the output of the shared layer is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the multilayer sensor.

2. The method for primary distribution network fault identification and location according to claim 1, characterized in that, The shared layer includes: a graph attention network for extracting spatial features and a long short-term memory network for extracting temporal features; The graph attention network includes a first graph attention layer, a second graph attention layer, a third graph attention layer, and a first fully connected layer connected in sequence. The input terminal of the first attention layer is connected to the first output terminal of the input layer as the input terminal of the shared layer. The Long Short-Term Memory (LSTM) network includes multiple LSM modules connected to a second fully connected layer. The multiple LSM modules are connected sequentially, and the input terminals of each LSM module are connected to the output terminal of the first fully connected layer. The output terminals of each LSM module are connected to the input terminal of the second fully connected layer. The input terminal of the second fully connected layer is connected to the second output terminal of the input layer. The output terminal of the second fully connected layer serves as the output terminal of the shared layer and is connected to the input terminal of the fully connected layer.

3. The method for primary distribution network fault identification and location according to claim 2, characterized in that, Step 3 includes: The topology, the fault transient signal, and the shallow feature quantity are all input into the input layer. The topology and the fault transient signal are input into the graph attention network. Feature extraction is performed using the first graph attention layer, the second graph attention layer, and the third graph attention layer. The spatial features of the fault transient signal are obtained by aggregation through the first fully connected layer. The spatial features are input into the long short-term memory module to mine the variation pattern of the fault transient signal in the time dimension and obtain the time features of the fault transient signal. The temporal features and the shallow feature values ​​are input into the second fully connected layer for aggregation to obtain the aggregation result. The aggregation result is input to the multilayer sensor through the fully connected layer for fault identification and location, thereby obtaining the fault identification result and fault location result of the target distribution network.

4. The method for primary distribution network fault identification and location according to claim 3, characterized in that, The fully connected layer includes a third fully connected layer, a fourth fully connected layer, and a fifth fully connected layer; The multilayer sensor includes a first multilayer sensor for identifying fault types, a second multilayer sensor for locating fault sections, and a third multilayer sensor for locating fault points. The input terminals of the third fully connected layer, the fourth fully connected layer, and the fifth fully connected layer are all connected to the output terminal of the second fully connected layer. The output terminal of the third fully connected layer is connected to the input terminal of the first multilayer sensor. The output terminal of the fourth fully connected layer is connected to the input terminal of the second multilayer sensor. The output terminal of the fifth fully connected layer is connected to the input terminal of the third multilayer sensor. The output terminals of the first multilayer sensor, the second multilayer sensor, and the third multilayer sensor all serve as the output terminals of the fault location model.

5. The method for primary distribution network fault identification and location according to claim 1, characterized in that, Modeling the topology of the target distribution network under fault conditions includes: The electrical nodes and lines of the target distribution network are equivalent to the vertices and edges of a graph; The topology of the target distribution network is obtained by modeling the topology under fault conditions using an adjacency matrix.

6. The method for fault identification and location in a primary distribution network according to claim 1, characterized in that, The steps preceding step 2 also include: The fault transient signal is denoised in the time domain; The time-domain denoised transient fault signal is subjected to deviation standardization processing; Establish the optimal time-frequency joint distribution expression for the fault transient signal after deviation standardization.

7. The method for primary distribution network fault identification and location according to claim 6, characterized in that, The fault transient signal after deviation standardization is: Among them, z k This is the fault transient signal after filtering and noise reduction.

8. A primary-level distribution network fault identification and location device, characterized in that, include: The acquisition module is used to model the topology of the target distribution network under fault conditions and acquire the fault transient signal of the topology. The fault transient signal includes transient voltage signal and transient current signal; The shallow feature extraction module is used to establish the optimal time-frequency joint distribution expression of the fault transient signal, and extract the shallow feature quantity of the fault transient signal through the optimal time-frequency joint distribution expression; The feature extraction and fusion module is used to construct a fault location model. The topology, the fault transient signal, and the shallow feature quantity are input into the fault location model for feature extraction and feature fusion to obtain the fault identification result and fault location result of the target distribution network. The fault location model includes an input layer, a shared layer, a fully connected layer, and a multilayer sensor. The output of the input layer is connected to the input of the shared layer, the output of the shared layer is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the multilayer sensor.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the primary distribution network fault identification and location method as described in any one of claims 1 to 7.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the primary distribution network fault identification and location method as described in any one of claims 1 to 7.

Images