Power distribution network fault positioning method based on FSGATv2 and topological similarity

Abstract

The invention provides a power distribution network fault positioning method based on FSGATv2 and topological similarity, and belongs to the technical field of power distribution network fault positioning. The small sample problem of topological structure change in the process from simulation to practical application of the model is solved; firstly, a feature selection module is embedded into a GATv2 network, and the sensitivity degree of the model to fault features is improved. Secondly, classifying the topology of the power distribution network based on topological similarity so as to improve the topological generalization capability of the fault positioning model; meanwhile, a domain self-adaptive transfer learning method for optimizing a dynamic loss function is adopted, so that the distribution difference between data of a source domain and a target domain is reduced, and the problem of few-sample fault positioning is effectively solved; the method is applied to fault location under different operation conditions of the power distribution network.

Description

Technical Field

[0001] The invention provides a distribution network fault location method based on FSGATv2 and topology similarity, belonging to the technical field of distribution network fault location. Background Art

[0002] The distribution network has many branch lines and complex line structures. When a small current grounding system is used, there are problems such as weak fault current and unclear characteristics. The information provided by traditional measurement devices is incomplete, the measurement error is large, and data loss and distortion are easy to occur. In addition, due to load switching, line loss optimization and fault reconstruction, the network topology changes frequently. The high penetration rate of distributed generation (DG) such as wind power and photovoltaic power leads to large differences in the characteristics of active distribution networks and traditional distribution networks in terms of power flow distribution, fault current level and polarity. How to achieve rapid fault detection and location in the operation scenario of new distribution networks with high DG penetration is the key to improving the reliability of distribution network operation. Traditional fault location methods such as impedance method, injection method and traveling wave method have the advantages of being simple and easy to implement in scenarios where the distribution network structure does not change. However, when the distribution network structure changes, the method of manually extracting weak characteristic signals relies on empirical values ​​and needs to regenerate the network matrix or sample library, which not only greatly increases the amount of calculation, but also has a high dependence on alarm information, affecting the accuracy of fault location.

[0003] In recent years, data-driven deep learning methods have been attracting the attention of scholars at home and abroad because they can quickly process complex information and have good fault tolerance. Later, with the application of micro phasor measurement unit (μPMU) in distribution networks, a reliable data foundation has been provided for distribution network fault location technology based on deep learning. Some scholars use a fixed number of interpolations to replace the measurement values ​​of a specific branch and apply the gradient boosting tree model to the fault location problem of low-voltage smart distribution networks. Some scholars combine variational mode decomposition (VMD) with convolutional neural networks (CNN) to achieve accurate fault location in networks with different line structures. Some scholars use distributed phasor measurement units (D-PMU) and Petri nets for distribution network fault diagnosis. Based on the phasor distribution characteristics of line voltage and current, some scholars use μPMU measurement data to locate and verify single-phase grounding faults in active distribution networks. However, while the above-mentioned method using measurement device data and neural networks accurately locates faults, it cannot take into account the changes in the topology of the distribution network. The operation mode of active distribution network is complex and changeable, and the access location and capacity of distributed power sources are easy to change in the staged development of power system. How to effectively realize fault location under few samples, complex topological changes and different operating conditions is a difficult problem that needs to be solved urgently. Compared with convolutional neural network, graph neural network (GNN) can extend the neural network method to the spatial graph domain and realize the combination of deep learning and graph data. At present, the most widely used GNNs include graph convolutional neural network (GCN) and graph attention network (GAT). GCN has been studied in the field of power system fault diagnosis. Some scholars consider the spatial fault characteristics between nodes and propose a distribution network fault diagnosis model based on deep GCN, which has anti-interference and generalization performance. Some scholars combine the measurement values ​​of different nodes with the network topology structure and use GCN to construct a distribution network fault location framework with high positioning accuracy. However, none of the above methods consider the universality and practical performance of the model under the change of network topology structure.

[0004] Compared with GCN, GAT uses the attention mechanism in computer vision to make it pay more attention to neighbor nodes and meet the requirements of inductive learning tasks. However, GAT is a restricted static attention mechanism, and each node processes its neighbor nodes in its own representation as a query. In the distribution network fault location task problem with dynamic node selection and frequent topological structure changes, the robustness to noise and topological generalization ability are poor. Therefore, GATv2 with a dynamic attention mechanism was proposed. GATv2 has stronger global node weight update ability and topological generalization performance, but the application of GATv2 in the field of power systems is still in its infancy. In addition, the distribution network fault location model based on graph neural network is highly dependent on massive fault data samples, but the failure of the distribution network is a small sample event, and it is difficult to obtain historical operating data under various fault conditions. Due to the difference in data distribution, the fault diagnosis model trained with labeled data in one scenario cannot classify the unlabeled data obtained from other scenarios. With the development of the new generation of artificial intelligence technology, transfer learning can promote the successful application trained in one scenario to other scenarios for fault diagnosis. Based on a large number of simulation sample training, the machine learning model is trained, and then combined with the actual fault samples of the distribution network, the trained machine learning model is migrated to the actual fault analysis and location tasks. Considering the actual factors of weak fault current and complex working conditions in the distribution network, some scholars proposed a distribution network fault diagnosis model based on sample transfer learning, and verified and tested the effectiveness of the model. Some scholars proposed a distribution network fault location framework based on transfer learning, and verified the fault location effect in the target domain using the collected data of the phasor measurement unit (PMU). However, the above methods all assume that the distribution of source domain data and target domain data is the same for the application of transfer learning, and do not take into account the differences between simulation and actual data. Summary of the invention

[0005] In order to achieve high-precision and strong robustness positioning of distribution network faults and solve the small sample problem of topological structure changes in the process of model from simulation to practical application, the present invention proposes a distribution network fault positioning method based on FSGATv2 and topological similarity.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a distribution network fault location method based on FSGATv2 and topology similarity, comprising the following steps: S1: Use graph structure to represent the actual distribution network topology, collect line zero-sequence current data as the target domain data set through μPMU, and build FSGATv2 model; FSGATv2 model is based on GATv2 network, and FS module for feature screening is embedded in GATv2 network; S2: Build a simulation model with the same operating conditions as the actual distribution network, simulate various faults and different numbers of μPMUs, collect and extract the statistical features of zero-sequence current, input them into the FS module for feature selection, and generate a labeled source domain data set; S3: Classify distribution network topologies based on topological similarity; S4: For the new topology distribution network, firstly perform category judgment according to step S3; then select a distribution network with a clear topology structure, complete historical data, and complete fault scenario in the corresponding category as the central distribution network, and use the FSGATv2 model to extract the transferable features of the central distribution network, guide the classifier to output the correct prediction target, and obtain the cross entropy loss function of the classifier on the source domain data set; S5: A domain adaptive transfer learning method with optimized dynamic loss function is used to measure the data distribution difference between the source domain and the target domain. The knowledge learned by the FSGATv2 model from the source domain data is adaptively transferred to the target domain data to complete fault location in the target domain.

[0007] The graph structure of the actual distribution network topology in step S1 is represented as: G =( A , X ),in A is the adjacency matrix, which is used to represent the distribution network topology graph N The connection relationship between nodes and edges, X is the node feature matrix. When the topology of the distribution network changes, the adjacency matrix is A Changes to A’ ; The node feature matrix X is a sequence composed of 24 types of statistical features of each node, expressed as: , X ∈ R N×24 ,in x i Indicates the node i Class statistical properties, R is a real number, N Indicates the number of nodes. Each node has a label indicating its status. 0 indicates a normal state and 1 indicates a fault state. The 24 statistical features include: Central tendency statistical characteristics of the time domain, including: mean, median, lower quartile and upper quartile; Dispersion statistics in the time domain, including minimum, maximum, interquartile range, standard deviation, root mean square, and square root amplitude; The statistical characteristics of the distribution shape in the time domain include: kurtosis, skewness, shape factor, gap coefficient, and crest factor; The centroid frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation characteristics in the frequency domain; Power spectrum entropy, singular value entropy, energy entropy and permutation entropy characteristics in time-frequency domain.

[0008] The FS module assigns high weights to the features related to each node label in the distribution network and low weights to the remaining features. It performs weighted combination by multiplying the scalar with the features, and uses the Softmax function to regularize the parameters for feature screening.

[0009] The input of the FS module is the node feature matrix X , and then feature fusion is performed after passing through two network layers, where the two network layers respectively include a linear transformation layer, a BatchNorm layer, a SumPool layer, a ReLu layer, and a fully connected layer.

[0010] The structure of the FSGATv2 model is as follows: input, FS module, 3 GAL layers, fully connected layer, output, where the input is G =( A , X ), the output is the fault status of each node. Each GAL layer is composed of a linear transformation layer, a dynamic attention mechanism and a fully connected layer. The dynamic attention mechanism is a multi-head attention mechanism that integrates the two steps of attention weight calculation and weighted feature aggregation.

[0011] In step S3, the singular values ​​and Jaccard coefficients of the distribution network topology adjacency matrix are calculated to measure the topological similarity, and the Mahalanobis distance between distribution network samples is used as the clustering criterion to classify the distribution network topology.

[0012] In step S5, the maximum mean difference (MMD) is used to measure the distribution distance between the source domain and the target domain samples, thereby measuring the data distribution difference between the source domain and the target domain.

[0013] The expression of the optimized dynamic loss function in step S5 is as follows: ; Where: J is the total loss function, J c is the cross entropy loss function of the classifier on the source domain dataset, J d is the MMD distance between the source domain and target domain datasets, α To optimize the parameters.

[0014] The expression of Jaccard coefficient is as follows: ; Where: s is the Jaccard coefficient,F 00 express A is 0 and A’ The number of 0s in F 01 express A is 0 and A’ The number of 1s in F 10 express A is 1 and A’ The number of 0s in F 11 express A is 1 and A’ The number of 1s in .

[0015] By the adjacency matrix A ∈ R N×N conduct The singular value decomposition of the adjacency matrix is ​​obtained by ,in yes A rank, is an orthogonal matrix.

[0016] The beneficial effects of the present invention compared with the prior art are as follows: 1) The present invention uses 24 types of statistical features of transient zero-sequence current as initial features, and improves the sensitivity of the model to the fault characteristics of the distribution network containing DG through the feature selection module. The GATv2 network uses a dynamic graph attention mechanism to adaptively adjust the attention coefficients between the topological nodes of the distribution network, and improves the adaptability to changes in the relationship between the same topological nodes.

[0017] 2) The present invention classifies the distribution network topology based on topological similarity, determines the actual distribution network topology category, and improves the topological generalization ability of the fault location model to be applicable to distribution networks with completely different topological structures.

[0018] 3) This paper establishes a universal and transferable FSGATv2 distribution network fault location model, migrates the source domain location model to the target domain, uses the t-SNE algorithm to visualize the case test results, provides interpretable analysis for the algorithm, and improves the adaptability of the deep learning-based fault location method under unlabeled and few sample conditions. In addition, field experiments were conducted on distribution networks under different operating conditions to test and verify the effectiveness of the method of this paper. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings: Figure 1It is a structural diagram of the distribution network fault location model based on FSGATv2 and topology similarity; Figure 2 It is a structural diagram of the distribution network fault location model based on FSGATv2; Figure 3 It is a structural diagram of the feature selection module; Figure 4 is a flow chart of the method of the present invention; Figure 5 This is a diagram showing the topology classification result of the distribution network in an embodiment of the present invention; Figure 6 This is a topological structure diagram of the second type of central distribution network system c in an embodiment of the present invention; Figure 7 This is a comparison result diagram of fault location of each model in the embodiment of the present invention; Figure 8 This is a comparison result diagram of fault location under different permeabilities in an embodiment of the present invention; Fig. 9 This is a comparison result diagram of the clustering distance and positioning accuracy of each distribution network and the distribution network system c in an embodiment of the present invention; Fig.10 This is a comparison result diagram of the positioning accuracy of each distribution network under two schemes in an embodiment of the present invention; Fig.11 This is a diagram showing the performance comparison results of the models in the embodiments of the present invention; Fig.12 This is a comparison result diagram of different fault location migration solutions in an embodiment of the present invention; Fig.13 It is a visualization diagram showing the classification results of different schemes in the embodiment of the present invention; Fig.14 This is a scene diagram of a 10kV distribution network grounding fault experiment in an embodiment of the present invention; Fig.15 Built using Matlab / Simulink Fig.14 Experimental simulation model diagram of the site diagram shown. DETAILED DESCRIPTION

[0020] like Figures 1 to 15 As shown, the present invention provides a distribution network fault location method based on FSGATv2 and topology similarity, and constructs a fault location model. The fault location model is based on the GATv2 network, integrates the feature selection (Feature Selection, FS) module and the transfer learning model, and can realize the three functions of distribution network topology classification, fault feature extraction and fault location. The existing technology involved in the model and its improvement are described in detail below.

[0021] 1. Distribution network topology classification From the perspective of graph theory, the power network has a natural graph structure. The distribution network is mostly radial. Without considering the electrical information of the internal components, it can be abstracted as a graph consisting of nodes and edges. The nodes are composed of electrical nodes of the distribution network, and the edges are composed of overhead lines or cable lines. Accurate topological structure is the basis for fault location. In order to improve the economy and safety of the power grid system, the distribution network may be reconstructed during operation, resulting in changes in the number of nodes and edges. Economic reconstruction does not have faulty lines, and the number of nodes remains unchanged before and after reconstruction, while faulty reconstruction may cut off some lines and nodes. The network topology will change under both reconstruction methods.

[0022] Assume that the graph structure of the given distribution network is G =( A , X ),in A is the adjacency matrix, which is used to represent the distribution network topology graph N The connection relationship between nodes and edges, A ∈ R N×N , R is a real number, X is the node feature matrix. When the topology of the distribution network changes, the adjacency matrix is A Changes to A’ .

[0023] Node feature matrix X In the method, transient zero-sequence current is used to indicate whether the node has a fault. The present invention uses μPMU as a measuring device to collect zero-sequence current of the line, and extracts 24 types of statistical features of zero-sequence current with rich fault characteristics in multiple domains, namely, central trend statistical features (mean, median, lower quartile and upper quartile) in the time domain, dispersion statistical features (minimum, maximum, interquartile range, standard deviation, root mean square and square root amplitude) and distribution shape statistical features (kurtosis, skewness, shape factor, gap coefficient, crest factor) in the frequency domain, centroid frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation in the frequency domain, power spectrum entropy, singular value entropy, energy entropy and permutation entropy features in the time-frequency domain. The sequence composed of 24 types of statistical features of each node is As the feature input of the model, X ∈ R N×24 ,in x i Indicates the node i Class statistical properties, N Represents the number of nodes. Each node has a label indicating its status. 0 indicates a normal state and 1 indicates a fault state. The output of the model is the state judgment result of the node, which is used to locate the fault.

[0024] Because node features propagate through multiple network layers in the model, the present invention improves the sensitivity of the GATv2 network to fault features by embedding the FS module when inputting features. The FS module assigns high weights to the features related to each node label in the distribution network and low weights to the remaining features. By multiplying the scalar with the feature for weighted combination, the Softmax function is used to regularize the parameters and perform feature screening to reduce the interference of noise features, thereby improving the accuracy of node classification. The structure of the feature selection module is as follows: Figure 3 shown.

[0025] For distribution network topology G =( A , X ), considering the separate mapping relationship between labels and features in the classification problem, the set of node features is , x i , i ∈{1,2,…, d} represents the node i Class statistical features, in this embodiment d The value can be 24, indicating that there are 24 types of statistical features. l k , k ∈{0,1,…, m} represents the proportion of each type of statistical feature of each node, that is, the importance of the feature. In this embodiment, m The value is 23. is the node feature matrix X No. i The scalar value that multiplies the components is initialized before model training begins. , and normalized using the softmax function , the normalized With features x i Multiply to get weighted combination features , which is expressed as follows: (1); (2); In the above formula: c is a constant, and in this embodiment, its value is 100. Its function is to prevent Tends to 0. is the Hadamard product of the corresponding elements. Since the FS module needs gradients when updating parameters, soft feature selection should be used, so that during training, the weights of features related to node labels are close to 1, and the weights corresponding to other features are close to 0, and the influence of noise features on positioning results becomes smaller.

[0026] The specific implementation process of the FS module is as follows: Input node feature matrix X After that, the output features are obtained through weighted combination , is a vector connector. As a new feature input to the output feature of the first network layer . Then Input to the second network layer and concatenate the output results to obtain the node features after passing through the FS module X new .

[0027] FSGATv2 is a dynamic graph attention GATv2 network (Feature Selection graphattention network, FSGATv2) embedded in the FS module. The distribution network fault location model based on FSGATv2 is as follows Figure 2 As shown in the figure, the distribution network topology corresponds to the input graph adjacency matrix, and the node characteristic matrix of the line zero-sequence current is X As the initial features of each node, the FS module obtains the same number of features as the input layer nodes X new The two network layers respectively contain a linear transformation layer, a BatchNorm layer, a SumPool layer, a ReLu layer, and a fully connected layer. Then X new and the adjacency matrix A The input is sent to three graph attention layers (GAL) for node feature aggregation. The GAL layer consists of a linear transformation layer, a dynamic attention mechanism, and a fully connected layer. X new Mapped to a higher dimensional representation. The dynamic attention mechanism is a multi-head attention mechanism that integrates the two steps of attention weight calculation and weighted feature aggregation. Figure 2 The feature aggregation process uses a multi-head attention operation mode with 4 heads. The number of attention heads of the first two layers of GAL in the FSGATv2 model of the present invention is 3, and the number of heads of the last layer is 6, which is used for classification tasks. Through multiple experiments, the learning rate is adjusted to 0.01, the batch size batchsize=8, and the number of iterations is 50. The fully connected layer completes the node classification and finally outputs whether each node is in a fault state. The algorithm flow of FSGATv2 is shown in Table 1, where I ∈ R d Represents a vector of all 1s.

[0028]

[0029] Table 1.

[0030] 2. Transfer learning theory based on distribution network topology similarity The high penetration rate of distributed power sources connected to the distribution network makes the network topology complex and changes, which brings challenges to the fault location problem. If a distribution network fault location model that is applicable to completely different topologies, universal and transferable is established, this problem can be solved. When the distribution network fault location algorithm performs transfer learning on different distribution networks, the migration effect is related to the similarity of the distribution network topology, because the spectral radius of the distribution network is proportional to the number of nodes, which only reflects the size of the distribution network, and the norm reflects the connection of the distribution network topology, which cannot fully measure the topological similarity. Based on this, the present invention uses singular values ​​and Jaccard coefficients as indicators to measure topological similarity, classifies the distribution network topology, and improves the migration ability of the fault location model.

[0031] Assume that there is an orthogonal matrix , the adjacency matrix of the distribution network topology graph is A ∈ R N×N Perform singular value decomposition as shown in equation (2) to obtain equation (3). yes A The singular value sequence of yes A rank.

[0032] (2); (3).

[0033] Distribution network topology adjacency matrix A and A’ The Jaccard coefficient is shown in formula (4); (4).

[0034] because A and A’ There are only binary data 0 and 1. The simple match coefficient (SMC) is obtained from formula (4), as shown in formula (5).

[0035] (5); Where: F 00 express A is 0 and A’ The number of 0s in F 01 express A is 0 and A’ The number of 1s in F10 express A is 1 and A’ The number of 0s in F 11 express A is 1 and A’ In general, the adjacency matrix is ​​sparse, which leads to class imbalance. To solve this problem, we ignore the numerator F 00 , the final Jaccard coefficient is shown in formula (6).

[0036] (6).

[0037] In actual situations, distribution networks with different topological structures have different fault characteristics, and the data distribution of training sets and test sets is quite different. However, for the fault location model based on neural networks, the training set and test set are derived from the same topological structure. In the absence of historical fault data or incomplete fault data, the model has strong limitations. To address this problem, the present invention proposes a transfer learning model based on topological similarity. The model does not need to comprehensively collect the fault data of distribution networks of various topologies, and migrates the FSGATv2 fault location model to different distribution networks to quickly and accurately complete the fault location task.

[0038] Transfer learning refers to using the similarities between data, tasks, or models to transfer D s Model applied to target domain D t A learning process. For distribution network fault location tasks with different topologies, D s is a distribution network with an existing topology, whose labels are available. D t is a distribution network with a new topology and no labels are available. Transfer learning is defined as: given a source domain dataset and target domain dataset , x and y are the input data and output data of samples in the two domains respectively, using D s Data Learning D t The mapping function on , so that f exist D t has the smallest prediction error (using l The definition is shown in formula (7).

[0039] (7).

[0040] Where: f * is the minimum prediction error, For expectation.

[0041] Most of the existing fault location models based on transfer learning are trained through historical data or massive simulation data, and then the data under the same learning task is used to predict or classify in the trained network. This process assumes that the data distribution of the source domain data and the target domain is the same. However, in actual application, the distribution of the new structure distribution network in the target domain is different from that of the trained source domain data. It is necessary to use domain adaptive transfer learning methods to make the data distribution of the source domain and the target domain closer to improve the fault location accuracy. For the problem that the task is the same but the distribution of the source domain and target domain data in the feature space is inconsistent, domain adaptation solves the optimal projection matrix and projects the data of the two domains into the new feature space. The present invention uses the maximum mean discrepancy (MMD) to measure the distribution distance between the source domain and the target domain samples, so that the distribution distance of different data is shortened.

[0042] For the source domain dataset and target domain dataset , the inner product form of the kernel function is used to perform nonlinear mapping on the Hilbert space, as shown in formula (8).

[0043] (8); Where: n s and n t are the sample sizes of the source domain and the target domain, respectively. is a nonlinear mapping function in the reconstruction kernel Hilbert space, mapping each sample to a The associated Hilbert space H , we get formula (9).

[0044] (9).

[0045] The Gaussian kernel function shown in formula (10) is selected to calculate the inter-domain MMD.

[0046] (10).

[0047] Where: is the width of the kernel function and determines the degree of smoothness. The domain adaptive transfer learning method embeds the MMD measurement method as a regularization method into the network training process, optimizes the supervision criterion by training the network parameters, reduces the difference in the distribution of feature data between the source domain and the target domain, and thus improves the transfer effect.

[0048] 3. Distribution network fault location model based on FSGATv2 and topology similarity In order to achieve high-precision and strong robustness in locating distribution network faults and solve the problem of small sample sizes when the model changes from simulation to practical application, this paper proposes a distribution network fault location model based on FSGATv2 and topological similarity, such as Figure 1 As shown, the fault location model is used to locate the distribution network fault, which mainly includes the following steps: S1: Build the FSGATv2 model; and collect the zero-sequence current data of the actual distribution network as the target domain data set; S2: Build a simulation model with the same operating conditions as the actual distribution network, simulate various faults (different fault lines, different fault locations, different fault grounding resistances, different fault initial phase angles, etc.) and different numbers of μPMUs, collect and extract the node feature matrix of zero-sequence current , , input to the FS module for feature selection to generate a labeled source domain dataset.

[0049] S3: Calculate the singular values ​​and Jaccard coefficient of the distribution network topology adjacency matrix to measure the topological similarity, and use the Mahalanobis distance between distribution network samples as the clustering criterion to classify the distribution network topology.

[0050] S4: For the new topology distribution network, firstly perform category judgment according to step S3. Then select a distribution network with a clear topology structure, complete historical data, and complete fault scenario in the corresponding category as the central distribution network, and use the FSGATv2 model to extract the transferable features of the central distribution network, guide the classifier to output the correct prediction target, and obtain the cross entropy loss function of the classifier on the source domain data set. J c ; S5: Define the MMD distance between the source domain and target domain datasets J d ,According to the topological structure and data characteristics of the new distribution network in the target domain, the data distribution difference between the source domain and the target domain is measured, as shown in formula (11).

[0051] (11); Where: and is a sample of n and The weight of M is the total number of categories in the dataset. Because linear loss is used Learning weights will cause the weights to converge to zero quickly. Therefore, the total loss function needs to be JDynamic optimization is performed and the optimization result is shown in formula (12).

[0052] (12); Where: α is the optimization parameter. The optimized dynamic loss function is smooth and differentiable, which solves the problem that the weights converge to zero quickly. J Optimize and adaptively transfer the knowledge learned by the model from the source domain data to the target domain data, and use the test set to complete fault location in the target domain.

[0053] 4. Experimental verification and analysis 4.1 Case Study Design Since the signal characteristics of the distribution network with the neutral point grounded through the arc suppression coil are weak when a single-phase grounding fault occurs, this embodiment uses Matlab / Simulink software to build 12 distribution network models with different scales, diverse topologies and grounded through arc suppression coils as shown in Table 2.

[0054]

[0055] Table 2 Topological structures of various distribution networks.

[0056] The positive-sequence and zero-sequence parameters of the overhead line and cable line are shown in equations (13) and (14).

[0057] (13); (14).

[0058] Calculate the first six singular values ​​and Jaccard coefficients of the adjacency matrix of each distribution network, and obtain the bubble matrix diagram of the first six singular values ​​and Jaccard coefficient of the adjacency matrix. Figure 5 shown.

[0059] The distribution network is classified by the first six singular values ​​of the adjacency matrix and the Jaccard coefficient matrix: a, b are the first category, c, f, i, k, l are the second category, e, g, h, j are the third category, and d is the fourth category. Figure 6The second type of central distribution network system c shown in the figure adopts the neutral point grounding method through the arc suppression coil, the compensation degree is 8%, the arc suppression coil inductance L=0.3885H, there are three feeders, 26 nodes and 23 branches, of which L10, L12, L14-L23 are overhead lines, and the rest are cable lines. Each distributed power source is connected to the feeder through an isolation transformer. Nodes 1, 6, and 17 are connected to photovoltaic power with a capacity of 3MVA, and nodes 10 and 20 are connected to wind power with a capacity of 2MVA. The measurement device μPMU is installed at the feeder outlet and before the terminal nodes 6, 3, 9, 14, 17, 20 and 23 to obtain fault data.

[0060] Each branch is set to have a single-phase grounding fault at 0.8s, and the protection is reliably activated to clear the fault after 0.2s. Data sampling starts 0.1s before the fault occurs and ends 0.2s after the fault is cleared. 300 sampling data points are taken from each branch as samples, and a total of 6900 sets of data from 23 branches constitute the data sample set. In order to solve the problem of sample imbalance, the ratio of fault data to non-fault data is 1:1 after undersampling the data, and then the training set, validation set and test set are divided according to the ratio of 8:1:1, of which the training set has 5520 samples, and the validation set and test set have 690 samples each.

[0061] 4.2 Comparative Analysis of Source Domain Test Results This embodiment adopts F 1 The score is used to evaluate the fault location model results, as shown in formula (15).

[0062] (15); Where: True Positive T 1 Indicates samples that are actually positive and predicted to be positive; false positive T 2 Indicates samples that are actually negative but predicted to be positive; false negative T 3 Indicates samples that are actually positive but predicted to be negative; F 1 The score takes into account the accuracy of the classification model p And recall r , evaluate the classification effect of the positioning model from multiple dimensions.

[0063] 4.2.1. Analysis of the validity of the source domain model The FSGCN (GCN network model with embedded FS module), FSGAT (GAT network model with embedded FS module) and FSGATv2 models are used to locate the single-phase ground fault of the distribution network system c, and the influence of different transition resistances, fault initial phase angles and fault distances on the model accuracy, as well as the robustness of the model under the influence of noise and data missing, are verified. Data missing refers to the situation where some node measurement values ​​are discarded and the node characteristic values ​​are replaced by 0 for positioning. The positioning accuracy under different conditions is shown in Table 3.

[0064]

[0065] Table 3 Comparison of source domain fault location model results under different conditions.

[0066] The results in Table 3 show that the performance of the FSGATv2 model proposed in the present invention is not affected by the initial fault phase angle and the fault location, and the highest accuracy can reach 99.98%, which has strong adaptability to measurement data noise.

[0067] 4.2.2 Analysis of the validity of the source domain model In order to verify the effectiveness of the FSGATv2 model of the present invention on the source domain dataset, the fault localization accuracy and loss function of FSGATv2 are compared with those of GCN, GAT, FSGCN and FSGAT models. Figure 7 shown.

[0068] Figure 7 The results show that the fault location model based on FSGAT converges in 256 rounds with an accuracy of 97.2%, while the FSGATv2 fault location model converges in 124 rounds with an accuracy of 99.98%, indicating that the FSGATv2 model not only accelerates the convergence speed of the model, but also improves the fault location accuracy. Because the penetration rate of distributed power sources affects the selection of fault features, it affects the fault location accuracy. In order to verify the fault location capabilities of each model under different DG penetration operating scenarios, a comparative experiment was conducted in the source domain. The results are as follows: Figure 8 shown.

[0069] Figure 8 The results show that various fault location models can complete the location task in the initial fault scenario, but as the DG penetration rate increases, the GCN and GAT models lack the ability to adapt to new scenario data and cannot complete the fault location task. The FSGCN and FSGAT models with embedded FS modules have better positioning accuracy than the GCN and GAT models, while the FSGATv2 model of the present invention is not affected by the DG penetration rate, has strong model adaptability and topology generalization capabilities, and the positioning accuracy remains at 99.97%.

[0070] 4.3 Comparative analysis of target domain test results In the process from simulation to actual application, the topological structure, grounding method changes and small sample problems in the field of distribution network fault diagnosis still exist. It is time-consuming and labor-intensive to collect fault data for the actual distribution network and rely on a large amount of simulation data for training and diagnosis. Therefore, the target domain considers a variety of fault conditions. First, the distribution network system c data collection method is used to generate fault scenarios as the target domain data set; then the source domain FSGATv2 model parameters are migrated to preliminarily build the target domain model; finally, the unlabeled target domain data is used to optimize the dynamic loss function of the target domain model MMD processing to complete the construction of the target domain model, so as to locate the distribution network fault.

[0071] 4.3.1 Analysis of the effect of target domain fault location migration In order to verify the migration effect of the FSGATv2 fault location model on different topological structures, the pre-trained model of the distribution network system c is migrated to distribution networks with different structures. The Mahalanobis distance is used as the clustering distance, and the comparison results of the clustering distance and positioning accuracy between each distribution network and the distribution network system c are obtained as shown in the following figure. Fig. 9 shown.

[0072] Fig. 9 The results show that the FSGATv2 fault location model of the central distribution network system c is transferable to the other 11 distribution networks, and the positioning accuracy is inversely proportional to the clustering distance of the distribution network system c, that is, the larger the clustering distance, the worse the migration effect. Because the clustering distance depends on the topological similarity of the model, for the new distribution network, a topological similarity analysis is first performed, and migration between distribution networks with high topological similarity will obtain better positioning results. In order to analyze the migration effect of the target domain fault location, 4000 sets of fault data of distribution networks a, d, f, and h are collected, and the FSGATv2 fault location model is used to conduct direct training and domain adaptive transfer learning comparative experiments. The two schemes have the same test set, and the fault location accuracy comparison results are shown in the figure. Fig.10 shown.

[0073] Fig.10 The results show that transfer learning has the ability to apply the FSGATv2 fault location model to distribution networks with different structures, and the retrained model after migration can locate faults more quickly and accurately. In order to verify the impact of the amount of target domain test data on the performance of the FSGATv2 fault location model, 2500 groups of fault samples were collected for the topology of a 48-node 22 μPMU distribution network, and a direct training and domain adaptive transfer learning comparison experiment was conducted. The localization effect of different amounts of unlabeled target domain data was verified. The results are shown in Figure 2. Fig.11 shown.

[0074] Fig.11The results show that with the increase of data volume, the highest accuracy of FSGATv2 fault location model is 89.94%, 94.88%, 97.89% and 99.96% respectively. And with the increase of data volume, the location accuracy is improved and the convergence speed is gradually accelerated.

[0075] In order to verify the effectiveness of the fault location model of the present invention, a comparative experiment was conducted using the domain adaptive migration method and the fine-tuning migration method. The fine-tuning migration method freezes all GAL layers and only trains the parameters of the fully connected layer to ensure the optimal deployment of the new structure distribution network. The comparison results of various schemes are shown in Figure 2. Fig.12 shown.

[0076] Fig.12 The results show that although the fine-tuning method can be used to optimize the deployment, freeze the hidden layer, and fine-tune the parameters of the fully connected layer to improve the applicability of the model, the FSGATv2 fault localization model fusion domain adaptive transfer learning method of the present invention has higher generalization ability on the target domain data. Fig.13 The classification results on the target domain of the model are visualized using the t-distributed stochastic neighbor embedding (t-SNE) algorithm, where the target domain data is blue and the source domain data is red.

[0077] Although the GATv2 fault localization model has certain classification and clustering capabilities, its effect is not significant. On this basis, the fine-tuning scheme improves the limited generalization ability, and the FSGATv2 fault localization model fusion domain adaptive transfer learning method of the present invention can achieve better classification effect in the target domain. Taking GATv2 as the basic model, the effect comparison of different fault localization schemes in the target domain is shown in Table 4.

[0078]

[0079] Table 4 Comparison of fault localization results of different schemes in the target domain.

[0080] It can be concluded from Table 4 that after multiple iterations, the positioning accuracy of Scheme 1 in the target domain is still poor. The reason is that the network topology and fault data characteristics of the target domain and the source domain are quite different, resulting in low generalization ability of the model in the target domain. Scheme 2 adds a feature selection module to the source domain model to accelerate the model convergence speed, but the positioning accuracy in the target domain is still not significantly improved. Schemes 3 and 4 fine-tune the hyperparameters of the fully connected layer. Compared with Schemes 1 and 2, although the positioning accuracy is improved, the effect is still very poor. Schemes 5 and 6 accelerate the model convergence speed and improve the generalization of the model by embedding the FS module and domain adaptive transfer learning method, and achieve reliable fault location.

[0081] 4.3.3 Effectiveness analysis under different conditions In the actual distribution network operation environment, the traditional fault location method is easily affected by the neutral point grounding method and the asymmetry of the line distribution parameters. To this end, this embodiment still considers the non-effective grounding method such as the neutral point through the arc suppression coil and the asymmetry of the line for the distribution network of 48 nodes and 22 μPMUs, generates the target domain data set, and verifies the effectiveness of the method of the present invention. The fault location accuracy of various situations is shown in Tables 5 and 6.

[0082]

[0083] Table 5 Fault location accuracy of different neutral point grounding methods.

[0084]

[0085] Table 6 Fault location results for asymmetric lines.

[0086] It can be seen from Table 5 and Table 6 that the method of the present invention can still accurately locate under different neutral point grounding modes and line asymmetry, and is not affected by the unbalanced parameters of the distribution line, and has good generalization ability.

[0087] 4.4 Experimental Verification Considering the differences between the simulation test and the actual distribution network, in order to better simulate the fault location scenario of single-phase grounding fault in the actual distribution network, the following Fig.14 In the 10kV distribution network grounding fault diagnosis experiment shown in the figure, a total of 4800 fault samples were collected in the experiment, and the samples included different DG grid-connected modes and changes in the DG grid-connected power at the same point, so as to verify the effectiveness of the method of the present invention.

[0088] Build with Matlab / Simulink Fig.14 The field experiment simulation model of the line is a mixture of overhead lines and cable lines, such as Fig.15 shown.

[0089] The simulation model is used for data collection and model training, and the actual data collected by the field experiment is used to verify the reliability of the method of the present invention. The photovoltaic power generation adopts a reversible grid-connected system, and the fault location results under various fault conditions are shown in Table 7. In the table, the fault type AG is the A-phase grounding fault, BG is the B-phase grounding fault, and CG is the C-phase grounding fault.

[0090]

[0091] Table 7 Fault location results under field experimental conditions.

[0092] The results in Table 7 show that the proposed method can accurately locate faults in actual distribution networks. The experiment verifies the effectiveness of the proposed algorithm for fault location problems in distribution networks with few samples and topology changes.

[0093] Aiming at the distribution network containing DG, the present invention establishes a new distribution network fault location method based on topological similarity and FSGATv2 domain adaptive transfer learning, and verifies it using distribution networks with different structures and fault diagnosis experiments. The FS module eliminates the disadvantages of the traditional model that the input features are single and highly dependent on input data. In the case of different topological structures and non-effective grounding methods, accurate fault location can still be achieved. It has strong robustness and practicality, and provides a new idea for the study of the few samples and topology change problems of distribution network fault location.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distribution network fault location method based on FSGATv2 and topology similarity, characterized by: The following steps are involved: S1: Use graph structure to represent the actual distribution network topology, collect line zero-sequence current data as the target domain data set through μPMU, and build FSGATv2 model; FSGATv2 model is based on GATv2 network, and FS module for feature screening is embedded in GATv2 network; S2: Build a simulation model with the same operating conditions as the actual distribution network, simulate various faults and different numbers of μPMUs, collect and extract the statistical features of zero-sequence current, input them into the FS module for feature selection, and generate a labeled source domain data set; S3: Classify distribution network topologies based on topological similarity; S4: For the new topology distribution network, firstly perform category judgment according to step S3; then select a distribution network with a clear topology structure, complete historical data, and complete fault scenario in the corresponding category as the central distribution network, and use the FSGATv2 model to extract the transferable features of the central distribution network, guide the classifier to output the correct prediction target, and obtain the cross entropy loss function of the classifier on the source domain data set; S5: A domain adaptive transfer learning method with optimized dynamic loss function is used to measure the data distribution difference between the source domain and the target domain. The knowledge learned by the FSGATv2 model from the source domain data is adaptively transferred to the target domain data to complete fault location in the target domain.

2. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 1, characterized in that: The graph structure of the actual distribution network topology in step S1 is represented as: G =( A , X ),in A is the adjacency matrix, which is used to represent the distribution network topology graph N The connection relationship between nodes and edges, X is the node feature matrix. When the topology of the distribution network changes, the adjacency matrix is A Changes to A’ ; The node feature matrix X is a sequence composed of 24 types of statistical features of each node, expressed as: , X ∈ R N×24 ,in x i Indicates the node i Class statistical properties, R is a real number, N Indicates the number of nodes. Each node has a label indicating its status. 0 indicates a normal state and 1 indicates a fault state. The 24 statistical features include: Central tendency statistical characteristics of the time domain, including: mean, median, lower quartile and upper quartile; Dispersion statistics in the time domain, including minimum, maximum, interquartile range, standard deviation, root mean square, and square root amplitude; The statistical characteristics of the distribution shape in the time domain include: kurtosis, skewness, shape factor, gap coefficient, and crest factor; The centroid frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation characteristics in the frequency domain; Power spectrum entropy, singular value entropy, energy entropy and permutation entropy characteristics in time-frequency domain.

3. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 2, characterized in that: The FS module assigns high weights to the features related to each node label in the distribution network and low weights to the remaining features. It performs weighted combination by multiplying the scalar with the features, and uses the Softmax function to regularize the parameters for feature screening.

4. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 3, characterized in that: The input of the FS module is the node feature matrix X , and then feature fusion is performed after passing through two network layers, where the two network layers respectively include a linear transformation layer, a BatchNorm layer, a SumPool layer, a ReLu layer, and a fully connected layer.

5. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 4, characterized in that: The structure of the FSGATv2 model is as follows: input, FS module, 3 GAL layers, fully connected layer, output, where the input is G =( A , X ), the output is the fault status of each node. Each GAL layer is composed of a linear transformation layer, a dynamic attention mechanism and a fully connected layer. The dynamic attention mechanism is a multi-head attention mechanism that integrates the two steps of attention weight calculation and weighted feature aggregation.

6. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 2, characterized in that: In step S3, the singular values ​​and Jaccard coefficients of the distribution network topology adjacency matrix are calculated to measure the topological similarity, and the Mahalanobis distance between distribution network samples is used as the clustering criterion to classify the distribution network topology.

7. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 1, characterized in that: In step S5, the maximum mean difference (MMD) is used to measure the distribution distance between the source domain and the target domain samples, thereby measuring the data distribution difference between the source domain and the target domain.

8. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 7, characterized in that: The expression of the optimized dynamic loss function in step S5 is as follows: ； Where: J is the total loss function, J c is the cross entropy loss function of the classifier on the source domain dataset, J d is the MMD distance between the source domain and target domain datasets, α To optimize the parameters.

9. The method for locating distribution network faults based on FSGATv2 and topology similarity according to claim 6, characterized in that: The expression of Jaccard coefficient is as follows: ； Where: s is the Jaccard coefficient, F 00 express A is 0 and A’ The number of 0s in F 01 express A is 0 and A’ The number of 1s in F 10 express A is 1 and A’ The number of 0s in F 11 express A is 1 and A’ The number of 1s in .

10. A distribution network fault location method based on FSGATv2 and topology similarity according to claim 6, characterized in that: By the adjacency matrix A ∈ R N×N conduct The singular value decomposition of the adjacency matrix is ​​obtained by ,in yes A rank, is an orthogonal matrix.

Images