A micro-grid group fault diagnosis method and device based on multi-source data fusion

Abstract

The application provides a micro-grid group fault diagnosis method and device based on multi-source data fusion, comprising the following steps: determining the sub-micro-grid and node inclusion relationship of the micro-grid group according to the node time sequence load matrix when the micro-grid group is normally operated; determining the sub-micro-grid connection relationship change of the micro-grid group according to the node time sequence load matrix when the micro-grid group is close to failure; determining the micro-grid group topology characteristics based on the sub-micro-grid and node inclusion relationship and the sub-micro-grid connection relationship change; determining the micro-grid group fault diagnosis result according to the MPNN network model, the node recording wave current when the micro-grid group is in failure and the micro-grid group topology characteristics; the MPNN network model is trained by using the micro-grid group fault sample data set and is used for realizing the micro-grid group fault positioning and fault type identification under different micro-grid group topologies and different node state matrices. The micro-grid group topology is accurately identified through the micro-grid group topology based on multi-source data fusion, and the efficiency of the micro-grid group fault diagnosis is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of power system fault detection technology, and in particular to a method and apparatus for diagnosing faults in microgrid groups using multi-source data fusion. Background Technology

[0002] A microgrid cluster consists of multiple interconnected microgrids connected by power distribution lines. Compared to a single microgrid, a microgrid cluster offers higher power supply reliability, scalability, and energy efficiency, making it a promising candidate for applications at the end of urban power distribution networks and in remote areas. Fault detection within a microgrid cluster is fundamental to ensuring its power supply reliability. Rapid and accurate microgrid cluster topology identification can effectively support fault diagnosis and provide more accurate location information for microgrid cluster operation and maintenance.

[0003] Currently, there are two main types of microgrid topology identification methods. One type uses traditional power system flow analysis and mathematical tools to deduce node correlations. These methods often require active injection of disturbance signals and have poor adaptability to multi-source networks. The other type is a data-driven approach. This type uses artificial intelligence to deduce the microgrid topology, but it is limited to obtaining precise node connections and ignores the impact of distributed energy sources in microgrid topology identification. This type of method is often only applicable to specific topology identification scenarios. For example, methods using historical smart meter status data cannot handle mesh topologies, and methods based on graph convolutional networks do not consider the fluctuations of distributed energy sources. Therefore, existing microgrid topology identification methods struggle to accurately identify microgrid group topologies.

[0004] Therefore, a method for identifying the topology of microgrid groups is needed to support fault diagnosis of microgrid groups. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a method and apparatus for fault diagnosis of microgrid groups based on multi-source data fusion. By accurately identifying the topology of microgrid groups through multi-source data fusion, the efficiency of fault diagnosis of microgrid groups is greatly improved.

[0006] In a first aspect, the present invention provides a method for fault diagnosis of microgrid groups by multi-source data fusion, the method comprising:

[0007] Based on the node time-series load matrix during normal operation of the microgrid group, determine the sub-microgrids and node inclusion relationships of the microgrid group;

[0008] Based on the node time-series load matrix when the microgrid group is nearing a fault, determine the changes in the sub-microgrid connection relationships of the microgrid group;

[0009] Based on the sub-microgrid and node inclusion relationships and the changes in sub-microgrid connection relationships, the topological characteristics of the microgrid group are determined.

[0010] Based on the MPNN network model, the node waveform currents during microgrid group faults, and the topological characteristics of the microgrid group, the fault diagnosis results of the microgrid group are determined.

[0011] The MPNN network model is trained using a labeled microgrid fault sample dataset and is used to locate and identify fault types in microgrids with different topologies and node state matrices.

[0012] Secondly, the present invention provides a microgrid group fault diagnosis device that integrates multi-source data fusion, the device comprising:

[0013] The first determining module is used to determine the sub-microgrids and node inclusion relationships of the microgrid group based on the node time-series load matrix during normal operation of the microgrid group;

[0014] The second determining module is used to determine the changes in the sub-microgrid connection relationships of the microgrid group based on the node time-series load matrix when the microgrid group is near a fault.

[0015] The third determining module is used to determine the topological characteristics of the microgrid group based on the sub-microgrid and node inclusion relationship and the changes in the sub-microgrid connection relationship;

[0016] The fault diagnosis module is used to determine the fault diagnosis result of the microgrid group based on the pre-stored MPNN network model, the node waveform current when the microgrid group is faulty, and the topological characteristics of the microgrid group.

[0017] The MPNN network model is trained using a labeled microgrid fault sample dataset and is used to locate and identify fault types in microgrids with different topologies and node state matrices.

[0018] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the microgrid group fault diagnosis method of multi-source data fusion as described in the first aspect.

[0019] Fourthly, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the microgrid group fault diagnosis method of multi-source data fusion as described in the first aspect.

[0020] This invention provides a method and apparatus for fault diagnosis of microgrid groups based on multi-source data fusion, comprising: determining the sub-microgrids and node inclusion relationships of the microgrid group based on the node time-series load matrix during normal operation; determining the changes in sub-microgrid connection relationships based on the node time-series load matrix when the microgrid group is nearing a fault; determining the microgrid group topological characteristics based on the sub-microgrid and node inclusion relationships and the changes in sub-microgrid connection relationships; and determining the fault diagnosis result of the microgrid group based on the MPNN network model, the node recorded current during a microgrid group fault, and the microgrid group topological characteristics. The MPNN network model is trained using a labeled microgrid group fault sample dataset and is used to achieve fault location and fault type identification of microgrid groups under different topologies and node state matrices. This invention significantly improves the efficiency of microgrid group fault diagnosis through accurate microgrid group topology identification via multi-source data fusion. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the microgrid group fault diagnosis method based on multi-source data fusion provided by the present invention.

[0023] Figure 2 This is an architectural design diagram of the message passing mechanism provided by the present invention;

[0024] Figure 3 This is a schematic diagram of the fault diagnosis framework of the microgrid group fault diagnosis method based on multi-source data fusion provided by the present invention;

[0025] Figure 4 This is a schematic diagram of the ADMM algorithm update process provided by the present invention;

[0026] Figure 5 This is a schematic diagram of a microgrid group topology structure and four topology variations provided by the present invention;

[0027] Figure 6 This is a schematic diagram of the inverse covariance matrix corresponding to the first topology provided by the present invention;

[0028] Figure 7 This is a schematic diagram of the inverse covariance matrix corresponding to the second topology provided by the present invention;

[0029] Figure 8This is a schematic diagram of the inverse covariance matrix corresponding to the third topology provided by the present invention;

[0030] Figure 9 This is a schematic diagram of the inverse covariance matrix corresponding to the fourth topology provided by the present invention;

[0031] Figure 10 This is a schematic diagram of the structure of the microgrid group fault diagnosis device with multi-source data fusion provided by the present invention;

[0032] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention;

[0033] Figure label:

[0034] 1110: Processor; 1120: Communication interface; 1130: Memory; 1140: Communication bus. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0036] The following is combined with Figures 1-11 This invention describes a method and apparatus for fault diagnosis of microgrid groups based on multi-source data fusion.

[0037] Firstly, this invention provides a method for fault diagnosis of microgrid groups using multi-source data fusion, such as... Figure 1 As shown, the method includes:

[0038] S11. Based on the node time-series load matrix during normal operation of the microgrid group, determine the sub-microgrids and node inclusion relationships of the microgrid group;

[0039] S12. Based on the node time-series load matrix when the microgrid group is nearing a fault, determine the changes in the sub-microgrid connection relationship of the microgrid group;

[0040] A typical AC microgrid group consists of a low-voltage feeder connecting all sub-microgrids. This feeder has no distributed energy sources or loads. Distributed energy sources and loads form sub-microgrids and are connected to a common junction point (CJ) among them. Therefore, microgrid groups are generally radial, and the connections between distributed energy sources and load nodes within a single sub-microgrid are relatively simple. Simultaneously, lines in a microgrid are short, load power consumption is relatively low, and nodes within the same sub-microgrid are electrically close and difficult to distinguish. Simplifying microgrid topology identification involves determining the connection relationships between sub-microgrids and identifying the inclusion relationship between each electrical node and the sub-microgrid under normal operating conditions.

[0041] In other words, the microgrid topology during a fault can be obtained by knowing the connection relationships of the sub-microgrids and identifying the inclusion relationships between each electrical node and the sub-microgrid under normal operating conditions.

[0042] S13. Determine the microgrid group topology characteristics based on the sub-microgrid and node inclusion relationship and the changes in the sub-microgrid connection relationship;

[0043] Multi-source data fusion and multimodal data fusion both involve integrating data from different types or sources, but they differ in practical applications. Multimodal data fusion refers to integrating data from different senses, such as processing audio and video data together, typically aiming to improve the expressiveness or accuracy of the data and provide more complete information. Multi-source data fusion, on the other hand, refers to integrating data from different sources, such as data collected from various sensors or devices, like GPS, weather observations, and comments on social media. The goal of multi-source data fusion is to obtain more comprehensive information, analyzing data from different perspectives to arrive at more accurate conclusions.

[0044] This invention adds a pre-processing algorithm based on multi-source data fusion for topology identification in fault diagnosis. The topology identification based on multi-source data fusion includes: First, obtaining the topological characteristics of the microgrid group's operating state using node power data during normal grid-connected operation; Second, obtaining the changes in the topological characteristics of the microgrid group's operating state using node power data before the fault; Third, fusing the results of the first and second steps to obtain the topological characteristics of the microgrid group at the time of the fault. The application of this pre-processing algorithm avoids the current problem where the lack of suitable microgrid group topology identification technology means that microgrid group topology characteristics can only be generated manually from electrical wiring diagrams using on-site survey and measurement data. This avoids the problem of delayed fault diagnosis caused by the lag in manually generated microgrid group topology characteristics, and also solves the problem of low fault diagnosis accuracy due to the inaccuracy of manually generated microgrid group topology characteristics.

[0045] S14. Based on the MPNN network model, the node waveform current during microgrid group faults, and the topological characteristics of the microgrid group, determine the fault diagnosis results of the microgrid group.

[0046] The MPNN network model is trained using a labeled microgrid fault sample dataset and is used to locate and identify fault types in microgrids with different topologies and node state matrices.

[0047] Specifically, the construction process of the MPNN network model includes the following:

[0048] (I) Building the MPNN network architecture:

[0049] The input data for fault detection in microgrid clusters is typically provided by waveform recording devices or measurement devices. In other words, fault detection in microgrid clusters requires determining the presence and type of fault based on short-time timing measurement data of components. This is a classification problem based on component operating status; however, data loss during measurement can negatively impact the accuracy of the detection.

[0050] Meanwhile, the short-circuit capacity of distributed energy sources in microgrids is relatively small. Therefore, compared with the large power grid, the short-circuit fault current capability provided by microgrids is very low. When a microgrid group is operating in grid-connected mode, the fault current is much larger than when it is operating in islanded mode. Furthermore, the topology of a microgrid group is relatively complex, and grid-connected and islanded operation of sub-microgrids may coexist. Relying solely on short-time time-series measurement data of components places excessive demands on the fault detection model of the microgrid group.

[0051] To address this issue, microgrid topology information is incorporated into the fault detection modeling of the microgrid group, as shown below.

[0052] P = f(X, A; θ)

[0053] Where X is the timing measurement matrix of electrical nodes in the microgrid group, and some data in the measurement matrix may be empty; A is the topology matrix of the microgrid group, which changes with the topology of the microgrid group; f is the nonlinear mapping model from timing measurement data and topology matrix to component operating state probability, θ is the model parameter; and P is the component operating anomaly probability matrix.

[0054] In deep learning, various neural networks can be used to directly construct high-dimensional mapping models, enabling feature extraction from various data driven by data. While discriminative neural networks in supervised learning possess powerful classification capabilities, most neural network models handle Euclidean data structures. Many practical tasks utilize non-Euclidean data structures, such as protein structure models and social network models. To address non-Euclidean data, Graph Neural Networks (GNNs) were proposed. The input features of vertices and the graph topology in the non-Euclidean data feature graph are fed into the GNN. The features contained in each vertex are propagated along the edges to other vertices, and convergence is achieved through iterative iteration. Graph Convolutional Neural Networks (GCNs) are a variant of GNNs and are the most famous and successful type. GCNs directly perform convolutions on the adjacency matrix of the graph, resulting in high computational efficiency and strong scalability.

[0055] Graph Convolutional Networks (GCNs) are mainly divided into two types: spectral domain-based and spatial domain-based. In spectral domain-based graph convolution methods, filters from graph signal processing are used for denoising. The eigenvectors of the symmetric normalized Laplacian matrix are used to perform Fourier transforms on the node features, and convolution kernels are used to transform the features in orthogonal space. Finally, an inverse Fourier transform is performed. The convolution kernel parameters in this method are those trained on the model, while the Fourier transform parameters are topologically dependent on the input graph and cannot be optimized. Furthermore, when the graph's topology changes, this method requires recalculating the Laplacian matrix, resulting in high computational costs.

[0056] In contrast, spatial domain-based graph convolution methods define graph convolution operations through information propagation, offering advantages such as high computational efficiency, strong flexibility, and good generalization. MPNN, a type of spatial domain-based graph convolution, is similar to traditional CNNs in image convolution. Spatial domain graph convolution defines the spatial relationships between nodes, utilizing information from each node and its neighbors to update the node representation. Furthermore, spatial domain-based graph convolution propagates node information along the edges of the graph during convolution, ensuring that each node's features are related to its neighbors, resulting in higher computational efficiency and stronger generalization. Therefore, spatial domain-based graph convolution methods have a wider range of applications.

[0057] Therefore, the fault detection modeling of this invention adopts the MPNN network architecture and decomposes the solution of the microgrid group fault detection problem into three steps: (1) local node feature extraction; (2) neighboring node feature fusion and node state prediction; (3) fault line and fault type inference.

[0058] First, in order to extract features from the time-series measurement data of each electrical node in the microgrid cluster, a node network layer including a CNN network and a multi-layer perceptron (MLP) is designed.

[0059] The electrical node timing measurement data I consists of the line current near the feeder side, that is:

[0060]

[0061] In the above formula, the subscripts a, b, and c represent the three phases respectively, and NT is the timing measurement length.

[0062] The time-series measurement data of all electrical nodes constitute the 3D measurement matrix X of the microgrid electrical nodes. If some measurement data is lost, the lost part is set to 0.

[0063] The X input node network layer extracts the local node state features of X, as shown in the following formula:

[0064] Z1 = C(X; θ1)

[0065] In the above formula, Let Nv be the electrical node state feature matrix, Nv be the dimension of the extracted features, C be the node network layer, and θ1 be the parameters of the node network layer.

[0066] The node network layer is a pre-processing local information extraction network that handles the input node state data. In practical applications, the node network layer can be appropriately adjusted, but it should be noted that it needs to be robust to data quality issues such as random data loss. To prevent excessively large parameters during training, the node network layer should be relatively simple.

[0067] Next, drawing an analogy to the propagation process of fault impact, a message passing layer (MPNN layer) based on the message passing mechanism is designed;

[0068] Similar to CNNs, MPNNs form node feature vectors by iteratively aggregating and passing information from neighboring nodes. Multi-layer graph convolution operations can be viewed as multiple message propagations on the graph. The MPNN constructed in this invention is a fully sampled GraphSAGE (Graph Sample and aggregate) network with a total of four layers. Figure 2 An example of an architecture design diagram for a message passing mechanism is provided. (For example...) Figure 2 As shown, the specific calculation process for each message transmission is as follows:

[0069]

[0070] In the above formula, Let be the node feature vector of node i after the (l+1)th convolution, aggreagte is the message aggregation function, N(i) is the set of neighboring nodes of node i, σ is the nonlinear transformation function, W is the weight matrix, concat means concatenation, and norm means L2 norm normalization.

[0071] If we add the weight vector of each edge between the nodes, Transformed into the following formula:

[0072]

[0073] In the above formula, e ji The weight vector represents the edge between node j and node i. The input of the topological features defines the neighboring nodes of each node. For example, the adjacency matrix of the microgrid group topology represents the adjacency relationship between vertices. Inputting the adjacency matrix into MPNN can obtain all N(i).

[0074] As can be seen from the calculation process, in each message propagation cycle, the node features generated by aggregating the node features obtained by all neighboring nodes of node i in the (l-1)th message transmission are concatenated with the node features obtained by node i in the (l-1)th message transmission and subjected to nonlinear transformation to generate the first node feature of node i in the lth message transmission; the L2 norm normalized result of the first node feature of node i in the lth message transmission is the node feature of node i in the lth message transmission.

[0075] The MPNN described above uses pooling aggregation to aggregate messages from nodes:

[0076]

[0077] In the above formula, max represents taking the maximum value of each feature, and b is the bias parameter. In this aggregation method, all neighboring nodes first undergo an MLP to extract features, then a non-linear transformation is performed, and finally max pooling is used to form the feature information of the nodes.

[0078] The expression for the message passing layer can be simplified to:

[0079] Z2 = S(Z1, A; θ2)

[0080] In the above formula, S represents the constructed message passing layer, and θ2 represents the message passing parameters. Let Nm be the probability matrix of the electrical node states, and let Nm be the total number of predicted node states.

[0081] The proposed MPNN's structure and operating mechanism are strongly coupled with the electrical characteristics of microgrid clusters. In particular, by incorporating the impact of topology changes into the model's learning through message passing, the model learns the dependencies between electrical node state features, adapting to the complex operating conditions of microgrid clusters. Therefore, after multiple layers of message passing, the node state feature output vector in the MPNN can reflect the level of electrical anomalies.

[0082] Then, design the label mapping layer;

[0083] Integrate all node state feature vectors into a fully connected network for classification, and output the node anomaly probability P and the fault type label Y.

[0084] (P,Y)=e(Z2)

[0085] In the above formula, node label Y is the label corresponding to the highest probability output after the node is labeled, and e is the label mapping.

[0086] Finally, design the fault location layer;

[0087] g(Z2,A)=g(P,U(Y,A))

[0088] In the above formula, U is the line mapping, and g is the logic function for locating and inferring the faulty line / area and the fault type.

[0089] The MPNN network architecture, through supervised learning, transforms the optimization process of predicting node label probabilities into a parameter optimization problem of the neural network, obtaining the optimal parameters θ1 and θ2, i.e.:

[0090]

[0091] Based on the requirements of fault localization, a loss function for parameter optimization is constructed, and parameters θ1 and θ2 are updated using a gradient descent training algorithm. For multi-class classification problems, the cross-entropy loss function (CELoss) is generally chosen, as shown in the following equation:

[0092]

[0093] In the above formula, p′ i p represents the predicted probability. i This represents the label probability. Since only two nodes are labeled as faulty in the case of a single line failure, the number of faulty labels is extremely imbalanced compared to the number of normal labels. This data imbalance makes it difficult for the model to learn fault features; therefore, a weight w negatively correlated with the number of labels is introduced into CELows. i This will improve the model's learning performance on fault label data.

[0094] In summary, the MPNN network architecture designed in this invention is a projection of the physical entities of a microgrid cluster onto the neural network model space, which helps to improve the system's fault location performance.

[0095] (II) Generation of labeled microgrid cluster fault sample dataset:

[0096] First, topological feature design;

[0097] To adapt to the varying operating states and topologies of microgrid clusters, a topology feature specifically designed for MPNN is presented. Power networks naturally possess a graph structure, allowing the various power nodes and lines within a microgrid cluster to be abstracted as a graph composed of nodes and edges. For a simple microgrid cluster topology graph structure, node 0 is the bus, node 4 is energy storage, nodes 5 and 7 are photovoltaic systems, and nodes 2, 5, and 6, and nodes 3, 7, and 8 each form two sub-microgrids.

[0098] The topological features of a graph need to express the electrical connections between nodes in a power system and characterize the strength of these connections, thus affecting the transmission of information features on the graph. In microgrids, line lengths are relatively short, and line impedance has a smaller impact. While node admittance matrices or node impedance matrices could be used as topological features, they would significantly increase the computational complexity of graph convolution. Therefore, adjacency matrices are used as the topological features of the graph.

[0099] In MPNN, the direction of feature transfer is determined by topological features. The direction of energy transfer between electrical nodes in a microgrid is not always bidirectional. Using an undirected graph adjacency matrix as a topological feature would lose information about the direction of power flow in the power system. Therefore, it is optimized from the perspective of energy flow direction. In this simple microgrid, when the output of the photovoltaic node is low, energy flows unidirectionally from node 0 to node 1. The energy flow between nodes 5 and 2, and between nodes 7 and 3 in the two sub-microgrids, is also unidirectional, with energy output from the photovoltaic system. The energy flow direction between nodes 4 and 1 is determined by the charging and discharging state of the energy storage. When the energy storage is in a charging state and the microgrid is operating in grid-connected mode, its adjacency matrix A and the improved adjacency matrix A′ are shown below.

[0100]

[0101] When the microgrid group is disconnected from the grid and the sub-microgrids operate in islanded mode, the energy storage discharges, and nodes 3 and 1 disconnect. The adjacency matrix A and the improved adjacency matrix A′ of the microgrid group are shown below:

[0102]

[0103] When the microgrid group topology changes or its operation mode changes, the topology feature input of the model can be changed according to the topology feature design method mentioned above.

[0104] Secondly, fault sample generation;

[0105] Fault sample data is generated using an electromagnetic transient simulation model of a microgrid group. During each line fault, the operation of a fault recorder is simulated. If the line overcurrent amplitude exceeds 1.2 times the normal value, the three-phase current waveform data for one cycle after the trigger is retained, and an electrical node measurement matrix X is generated. The electrical node measurement matrix X, the fault type, and the microgrid group topology are considered as a single fault sample.

[0106] The fault types include single-phase ground fault, two-phase ground fault, two-phase short circuit, and three-phase short circuit, as shown in Table 1. For each line in the microgrid group, 10 fault types were randomly simulated, and batch simulations were used to generate the first microgrid group fault sample dataset.

[0107] Table 1

[0108]

[0109] To ensure the feature extraction capability of the node network, the convolutional kernel module adopts GoogleNet with multiple convolutional kernels merged in parallel, and two convolutional networks are superimposed in the node network.

[0110] Finally, an enhanced dataset was constructed based on the first microgrid fault sample dataset. The measurement matrix in the first microgrid fault sample dataset consists of the three-phase current waveforms of the nodes. Incomplete waveform data was generated by randomly losing some data points from the complete current waveform data. By randomly losing 20% ​​and 50% of each current waveform from the two original datasets respectively, two incomplete waveform datasets were formed. These two incomplete waveform datasets were then merged with the first microgrid fault sample dataset to form an enhanced dataset with three times the number of samples. Finally, the enhanced dataset was used as the microgrid fault sample dataset.

[0111] (III) The MPNN network architecture is trained using a labeled microgrid group fault sample dataset to obtain the MPNN network model.

[0112] This invention provides a microgrid group fault diagnosis method based on multi-source data fusion. By accurately identifying the microgrid group topology through multi-source data fusion, the efficiency of microgrid group fault diagnosis is greatly improved.

[0113] Specifically, Figure 3 A schematic diagram illustrating a microgrid group fault diagnosis framework is provided, such as... Figure 3 As shown, both S11 and S12 are implemented using the graph Lasso algorithm. Further, S11 includes:

[0114] S11.3: Using the graph Lasso algorithm, estimate the inverse covariance matrix of the node time-series load matrix when the microgrid group is operating normally, and denote it as the first inverse covariance matrix; S11.2: Filter the first inverse covariance matrix; S11.3: Analyze the filtered first inverse covariance matrix to obtain the sub-microgrids and node inclusion relationships of the microgrid group;

[0115] S12 includes:

[0116] S11.1: Using the graph Lasso algorithm, estimate the inverse covariance matrix of the node time-series load matrix when the microgrid group is nearing a fault, and denote it as the second inverse covariance matrix; S11.2: Filter the second inverse covariance matrix; S11.3: Analyze the filtered second inverse covariance matrix to obtain the changes in the sub-microgrid connection relationship of the microgrid group.

[0117] Furthermore, S11.1 / S12.1 includes:

[0118] S11.1.1 / S12.1.1 Based on the graph Lasso algorithm, construct the formula expression for maximizing the inverse covariance matrix of the node time-series load matrix when the microgrid group is operating normally / the node time-series load matrix when the microgrid group is nearing a fault;

[0119] S11.1.2 / S12.1.2 rewrites the formula expression into a first expression suitable for solving using the ADMM algorithm, and solves the first expression using the ADMM algorithm to obtain the estimated value of the inverse covariance matrix of the node time-series load matrix when the microgrid group is operating normally / the node time-series load matrix when the microgrid group is nearing a fault.

[0120] Graph Lasso is a sparse modeling method for graph and network data. By estimating the sparse precision matrix of given graph and node features, it uses the L1 norm as a regularization term to make the weights of some edges become 0, selectively deleting some edges, thereby reducing graph complexity and improving modeling accuracy, and identifying structure from the input data.

[0121] This study uses Lasso graph analysis for microgrid topology identification, aiming to obtain the optimal accuracy matrix. Node correlation information is derived from input node features to determine node connectivity and thus the microgrid's topological characteristics. Nodes in a power system interact with each other, and power changes between adjacent nodes in a power grid typically exhibit correlation. Furthermore, current microgrid electromagnetic transient simulations involve stochastic changes in distributed energy resources. Therefore, during the electromagnetic transient simulation of microgrid operation, node power is affected by these stochastic processes. A change in the load of one node usually leads to changes in the loads of its neighboring nodes. By analyzing the correlation of these changes between nodes, the node's domain can be determined, thereby obtaining the microgrid's topological characteristics.

[0122] In fault diagnosis using multi-source data fusion, topological features need to be obtained through topology identification. Therefore, the microgrid group topology identification problem is modeled as a topology feature generation problem. This problem takes a node load matrix as input and outputs topology features, i.e.:

[0123]

[0124] In the above formula, The node load matrix is ​​generated from the historical time-series load data of each node. N T N represents the length of the historical time-series load data. V* θ represents the number of nodes from which load data is available; f represents the mapping model from node load data to microgrid topology characteristics; θ represents the model parameters; and A represents the microgrid topology characteristics.

[0125] Assuming the data points follow a Gaussian distribution, and there are potential correlations between the features, an undirected graphical model can be used to estimate these relationships. In this model, each feature corresponds to a node, and the relationships between nodes represent the correlation between features. For a Gaussian distribution, the elements in the inverse covariance matrix ∑⁻¹ indicate whether there is a correlation between nodes. Non-zero elements in this matrix correspond to nodes with a relationship, while elements with a value of 0 indicate conditional independence between a pair of variables.

[0126] Given n samples following a Gaussian distribution, the traditional method for estimating the inverse covariance matrix ∑-1 is based on the principle of maximizing the log-likelihood function, which is:

[0127] logdetΣ -1 -tr(SΣ -1 )

[0128] Where det represents the determinant and tr represents the trace of the matrix. This represents the node time-series load matrix when the microgrid group is operating normally / the node time-series load matrix when the microgrid group is nearing a fault.

[0129] Let Θ = Σ -1 Maximum likelihood estimation Taking an L1 penalty for Θ, the penalty log-likelihood function of the data is:

[0130] logdetΘ-tr(SΘ)-ρ‖Θ‖1

[0131] Where ρ represents the penalty parameter.

[0132] The problem of estimating sparse graphs by applying Lasso penalty is solved by implementing a simple graph Lasso algorithm using the coordinate descent process of Lasso. The goal is to maximize Θ, which can be expressed as:

[0133]

[0134] Maximizing the penalized log-likelihood function makes the model in The maximum value can be estimated from the data.

[0135] Compared to other sparse modeling algorithms, the graph Lasso algorithm is relatively simple, easy to understand, and easy to implement. Even with a small sample size or when some variables do not satisfy the sparsity assumption, the graph Lasso algorithm can still provide reasonable results to a certain extent. However, the graph Lasso method is computationally expensive, requiring multiple iterations, which can be extremely time-consuming with large-scale data. Furthermore, parameter selection significantly impacts the algorithm's stability and accuracy; therefore, the selection of parameters for L1 regularization requires empirical and experimental verification.

[0136] Alternating Direction Method of Multipliers (ADMM) and Graph Lasso are commonly used algorithms in high-dimensional data analysis and sparse model selection. Specifically, Graph Lasso is a method based on L1 regularization to infer relationships between variables and perform feature selection or dimensionality reduction. ADMM is an iterative solver used to solve optimization problems on high-dimensional data. Because the iterative process of the ADMM algorithm can be decomposed into several independent steps, parallel algorithms can be used to accelerate computation. Compared to other model selection algorithms, ADMM converges faster and yields results in a shorter time. Furthermore, ADMM can still perform relatively accurate modeling even when the data does not fully satisfy the sparsity assumption, exhibiting good robustness. Since ADMM can be used to solve problems with regularization terms, this invention applies ADMM to the solution of Graph Lasso.

[0137] The basic idea of ​​the ADMM algorithm is to construct a new optimization problem by adding Lagrange multipliers and then iteratively solve the original problem. Assume the original problem is:

[0138]

[0139] stAx+Bz=c

[0140] Where f and g are convex functions, x and z are variables, A and B are matrices, c is a fixed parameter in the domain, and the goal is to find the minimum value of the problem.

[0141] To solve this problem, we perform a Lagrange transform and define the Lagrange multiplier μ:

[0142]

[0143] Here, ρ is the penalty parameter, and since regularization techniques are used, ρ>0.

[0144] For the Lagrange dual function, the following definition holds:

[0145]

[0146] That is, with μ fixed as the input, the lower bound of the function obtained under unconstrained conditions is Q(μ).

[0147] Therefore, the original problem can be solved equivalently by the following formula:

[0148]

[0149] stAx+Bz=c

[0150] In the iterations of the ADMM algorithm, one variable is updated each time:

[0151]

[0152]

[0153] u (k) =u (k-1) +ρ(Ax (k) +Bz (k) -c)

[0154] Where k = 1, 2, 3, ... represents the number of update iterations. Figure 4 A schematic diagram illustrating the ADMM algorithm update process is provided.

[0155] Therefore, for the Lasso problem of the present invention, it is equivalent to:

[0156]

[0157] Θ = Y

[0158] Wherein, Θ represents the inverse covariance matrix of the node time-series load matrix when the microgrid is operating normally / the node time-series load matrix when the microgrid is near a fault. Let represent the node time-series load matrix of the microgrid group during normal operation / the node time-series load matrix of the microgrid group when nearing a fault, where n represents the number of nodes in the microgrid group, det represents the determinant, tr represents the trace of the matrix, ρ represents the penalty parameter, and ||.|1 represents the L1 norm. F Describing the F-norm, Let Θ be positive semidefinite, Z be the auxiliary parameter, μ be the Lagrange multiplier, and <.> be the inner product.

[0159] The solution can be obtained by applying the ADMM algorithm to update the rules.

[0160] Furthermore, S11.1.2 / S12.1.2 includes:

[0161] Elements in the first inverse covariance matrix / second inverse covariance matrix that are below a first preset threshold are set to 0, and elements that are above the first preset threshold are set to 1.

[0162] For example, the first preset threshold is 1.

[0163] The first preset threshold needs to be selected based on the on-site working conditions.

[0164] Specifically, S14 includes:

[0165] S14.1: Generate the first node state matrix based on the node waveform current recorded during the fault of the microgrid group;

[0166] S14.2 Substitute the first node state matrix into the MPNN network model to obtain the fault type and fault line / fault area of ​​the microgrid group.

[0167] Steps S14.1 to S14.2 above are equivalent to:

[0168] A first node state matrix is ​​generated based on the node waveform current recorded during the fault of the microgrid group.

[0169] The first node state matrix is ​​substituted into the node network layer of the MPNN network model for feature extraction to obtain the node state feature matrix.

[0170] The node state feature matrix and the microgrid group topology features are input into the message passing layer of the MPNN network model to obtain the node state probability matrix output by the message passing layer.

[0171] The node state probability matrix is ​​input into the label mapping layer of the MPNN network model to obtain the node anomaly probability and node fault type label output by the label mapping layer; the microgrid topology features, the node anomaly probability, and the node fault type label are input into the fault location layer of the MPNN network model to obtain the fault type and fault line / fault area of ​​the microgrid output by the fault location layer.

[0172] The node state probability matrix is ​​the result obtained by the message passing layer through multi-layer message passing on the node state feature matrix based on the microgrid group topology features;

[0173] For message passing at layer l, the node features generated by aggregating the node features obtained by all neighboring nodes of any node in the (l-1)th message passing are concatenated with the node features obtained by any node in the (l-1)th message passing and then subjected to nonlinear transformation to generate the first node feature of any node in the lth message passing.

[0174] The L2 norm normalized result of the first node feature of any node in the l-th message transmission is the node feature of any node in the l-th message transmission.

[0175] The following example supports this invention:

[0176] For a microgrid group topology, since the energy storage power source is directly connected to the sub-microgrid and there is no line information between it and the grid connection port, the energy storage node is ignored. The microgrid group topology and four topology variations after simplifying the energy storage node are as follows: Figure 5 As shown.

[0177] Based on the simulation parameters shown in Table 2, 100 sets of node power change data during normal grid-connected operation of the microgrid group were generated using the microgrid group electromagnetic transient simulation model. The power data was then input into the graph Lasso algorithm to obtain the following results: Figure 6 The inverse covariance matrix representing the node correlation is shown in Topology 1. Figure 6As shown, the second row of the filtered inverse covariance matrix represents the correlation between node 2 and other nodes. Among the leaf nodes, only nodes 5 to 9 are strongly correlated with node 2, and these nodes represent all nodes in sub-microgrid 1, which uses node 2 as its grid connection point. The correlation between the other two sub-microgrid nodes is similar: nodes 10 to 20 are only strongly correlated with node 3, and nodes 21 to 29 are only strongly correlated with node 4. Simultaneously, node 1, the grid connection point of the microgrid group, is strongly correlated with the grid connection points of the three sub-microgrids (nodes 2, 3, and 4), and nodes 2, 3, and 4 are strongly correlated in pairs, indicating that all three sub-microgrids are operating in grid-connected mode. Ignoring the interconnections of the leaf nodes, the node inclusion relationships of each sub-microgrid during normal grid-connected operation can be obtained using the Lasso graph.

[0178] Table 2

[0179]

[0180] Batch simulations were performed using the simulation parameters shown in Table 2, yielding 100 sets of power variation data for microgrid nodes under different topologies. The node power data was then input into the graph Lasso algorithm to obtain the inverse covariance matrix representing node correlation under different topologies.

[0181] The inverse covariance matrix representing node correlation in Topology 2 is as follows: Figure 7 As shown. By Figure 7 It can be seen that the filtered inverse covariance matrix shows that node 1 has no strong correlation with the three sub-microgrid grid connection ports. Only the three grid connection port nodes are correlated with each other, and the distributed energy nodes 9, 19, 20, 28, and 29 are autocorrelated. This indicates that the microgrid group is in an off-grid operation state at this time.

[0182] The inverse covariance matrix representing node correlation in topology 3 is as follows: Figure 8 As shown. By Figure 8 The filtered inverse covariance matrix shows that nodes 1 and 3 are no longer strongly correlated, while nodes 1 remains strongly correlated with nodes 2 and 4. Meanwhile, nodes 2 and 4 are strongly correlated with each other but not strongly correlated with node 3. This indicates that sub-micronet 2 is now disconnected, meaning node 3 is no longer connected to node 1.

[0183] The inverse covariance matrix representing node correlation in topology 4 is as follows: Figure 9 As shown. By Figure 9 It can be seen from the filtered inverse covariance matrix under the operating state of Topology 4 that node 1 is not strongly correlated with the grid connection ports of the three sub-microgrids, and node 3 also has no correlation information with nodes 2 and 4. This indicates that the microgrid group is offline at this time, and sub-microgrids 1 and 2 are connected to the grid, while sub-microgrid 3 operates as an island. At this time, nodes 0 and 1 are disconnected, and nodes 1 and 4 are disconnected.

[0184] The results above show that when the microgrid group is operating normally, the graph Lasso method can identify the node composition of each sub-microgrid and establish an initial node association matrix. When the operating state of the microgrid group changes or different topologies exist, the graph Lasso method can accurately identify the current operating state of the microgrid group and the changes in the grid connection node relationships caused by the topology changes. Based on the parameters of the inverse covariance matrix, the topological features required for the MPNN network model can be directly generated. Since the ADMM algorithm has a fast convergence speed, the generation time of a single inverse covariance matrix is ​​less than 2 seconds, which can meet the requirement of rapidly generating topological features for fault diagnosis.

[0185] Secondly, a fault diagnosis device for microgrid groups based on multi-source data fusion is described below. The fault diagnosis device for microgrid groups based on multi-source data fusion described below can be referred to in correspondence with the fault diagnosis method for microgrid groups based on multi-source data fusion described above. Figure 10 A schematic diagram of a microgrid group fault diagnosis device based on multi-source data fusion is shown in the example. Figure 10 As shown, the device includes:

[0186] The first determining module 21 is used to determine the sub-microgrids and node inclusion relationships of the microgrid group based on the node time-series load matrix during normal operation. The second determining module 22 is used to determine the changes in sub-microgrid connection relationships of the microgrid group based on the node time-series load matrix when the microgrid group is nearing a fault. The third determining module 23 is used to determine the topological characteristics of the microgrid group based on the sub-microgrid and node inclusion relationships and the changes in sub-microgrid connection relationships. The fault diagnosis module 24 is used to determine the fault diagnosis results of the microgrid group based on the pre-stored MPNN network model, the node waveform current recorded during a fault in the microgrid group, and the topological characteristics of the microgrid group. The MPNN network model is trained using a labeled microgrid group fault sample dataset and is used to realize the fault location and fault type identification of the microgrid group under different microgrid group topologies and different node state matrices.

[0187] This invention provides a microgrid group fault diagnosis device based on multi-source data fusion. By accurately identifying the microgrid group topology through multi-source data fusion, the efficiency of microgrid group fault diagnosis is greatly improved.

[0188] Thirdly, Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11As shown, the electronic device may include: a processor 1110, a communications interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call logic instructions in the memory 1130 to execute a microgrid group fault diagnosis method based on multi-source data fusion. This method includes: determining the sub-microgrids and node inclusion relationships of the microgrid group based on the node time-series load matrix during normal operation; determining the changes in sub-microgrid connection relationships based on the node time-series load matrix when the microgrid group is nearing a fault; determining the microgrid group topology characteristics based on the sub-microgrid and node inclusion relationships and the changes in sub-microgrid connection relationships; and determining the fault diagnosis result of the microgrid group based on the MPNN network model, the node waveform currents during a microgrid group fault, and the microgrid group topology characteristics. The MPNN network model is trained using a labeled microgrid group fault sample dataset and is used to achieve fault location and fault type identification for microgrid groups with different topologies and node state matrices.

[0189] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0190] Fourthly, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multi-source data fusion microgrid group fault diagnosis method provided by the above methods. The method includes: determining the sub-microgrids and node inclusion relationships of the microgrid group based on the node time-series load matrix during normal operation; determining the changes in sub-microgrid connection relationships based on the node time-series load matrix when the microgrid group is nearing a fault; determining the microgrid group topology characteristics based on the sub-microgrid and node inclusion relationships and the changes in sub-microgrid connection relationships; and determining the fault diagnosis result of the microgrid group based on the MPNN network model, the node waveform current during a microgrid group fault, and the microgrid group topology characteristics. The MPNN network model is trained using a labeled microgrid group fault sample dataset and is used to realize microgrid group fault location and fault type identification under different microgrid group topologies and different node state matrices.

[0191] Fifthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a microgrid group fault diagnosis method based on multi-source data fusion provided by the methods described above. This method includes: determining the sub-microgrids and node inclusion relationships of the microgrid group based on the node time-series load matrix during normal operation; determining the changes in sub-microgrid connection relationships based on the node time-series load matrix when the microgrid group is nearing a fault; determining the microgrid group topological characteristics based on the sub-microgrid and node inclusion relationships and the changes in sub-microgrid connection relationships; and determining the fault diagnosis result of the microgrid group based on an MPNN network model, node waveform currents during a microgrid group fault, and the microgrid group topological characteristics. The MPNN network model is trained using a labeled microgrid group fault sample dataset and is used to achieve microgrid group fault location and fault type identification under different microgrid group topologies and different node state matrices.

[0192] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0193] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for fault diagnosis of microgrid groups using multi-source data fusion, characterized in that, The method includes: Based on the node time-series load matrix during normal operation of the microgrid group, determine the sub-microgrids and node inclusion relationships of the microgrid group; Based on the node time-series load matrix when the microgrid group is nearing a fault, determine the changes in the sub-microgrid connection relationships of the microgrid group; Based on the sub-microgrid and node inclusion relationships and the changes in sub-microgrid connection relationships, the topological characteristics of the microgrid group are determined. Based on the MPNN network model, the node waveform currents during microgrid group faults, and the topological characteristics of the microgrid group, the fault diagnosis results of the microgrid group are determined. The MPNN network model is trained using a labeled microgrid fault sample dataset and is used to locate and identify fault types in microgrids under different microgrid topologies and different node state matrices. The process of determining the fault diagnosis results of the microgrid group based on the MPNN network model, the node waveform currents during microgrid group faults, and the topological characteristics of the microgrid group includes: A first node state matrix is ​​generated based on the node waveform current recorded during the fault of the microgrid group. The first node state matrix is ​​substituted into the node network layer of the MPNN network model for feature extraction to obtain the node state feature matrix. The node state feature matrix and the microgrid group topology features are input into the message passing layer of the MPNN network model to obtain the node state probability matrix output by the message passing layer. The node state probability matrix is ​​input into the label mapping layer of the MPNN network model to obtain the node anomaly probability and node fault type label output by the label mapping layer; the microgrid topology features, the node anomaly probability, and the node fault type label are input into the fault location layer of the MPNN network model to obtain the fault type and fault line / fault area of ​​the microgrid output by the fault location layer.

2. The microgrid group fault diagnosis method based on multi-source data fusion according to claim 1, characterized in that, The step of determining the sub-microgrids and node inclusion relationships of a microgrid group based on the node time-series load matrix during normal operation includes: Using the graph Lasso algorithm, the inverse covariance matrix of the node time-series load matrix during normal operation of the microgrid group is estimated and denoted as the first inverse covariance matrix; Filter the first inverse covariance matrix; By analyzing the first inverse covariance matrix after filtering, the sub-microgrids and node inclusion relationships of the microgrid group are obtained; The step of determining the changes in sub-microgrid connectivity of a microgrid group based on the node time-series load matrix at the time of a near-fault includes: Using the graph Lasso algorithm, the inverse covariance matrix of the node time-series load matrix of the microgrid group near fault is estimated and denoted as the second inverse covariance matrix. Filter the second inverse covariance matrix; By analyzing the filtered second inverse covariance matrix, the changes in the sub-microgrid connection relationships of the microgrid group are obtained.

3. The microgrid group fault diagnosis method based on multi-source data fusion according to claim 2, characterized in that, Using the graph Lasso algorithm, the inverse covariance matrix of the node time-series load matrix during normal operation of the microgrid group / the node time-series load matrix nearing a fault in the microgrid group is estimated, including: Based on the graph Lasso algorithm, a formula expression is constructed to maximize the inverse covariance matrix of the node time-series load matrix when the microgrid group is operating normally / the node time-series load matrix when the microgrid group is near a fault. The formula expression is rewritten into a first expression suitable for solving using the ADMM algorithm, and the first expression is solved using the ADMM algorithm to obtain the estimated value of the inverse covariance matrix of the node time-series load matrix when the microgrid group is operating normally / the node time-series load matrix when the microgrid group is nearing a fault.

4. The microgrid group fault diagnosis method based on multi-source data fusion according to claim 3, characterized in that, The formula expression is as follows: ， The first expression is as follows: ， ， in, express The estimated value, This represents the inverse covariance matrix of the node time-series load matrix during normal operation of the microgrid group / the node time-series load matrix nearing a fault in the microgrid group. , This represents the node time-series load matrix during normal operation of the microgrid group / the node time-series load matrix when the microgrid group is nearing a fault. This indicates the number of nodes in the microgrid group. Represents a determinant. Represents the trace of a matrix. Indicates the penalty parameter. Describing the L1 norm, Describing the F-norm, express It is positive semidefinite. Indicates auxiliary parameters, Represents the Lagrange multipliers. This indicates the inner product.

5. The microgrid group fault diagnosis method based on multi-source data fusion according to any one of claims 2 to 4, characterized in that, Filtering the first inverse covariance matrix / the second inverse covariance matrix includes: Elements in the first inverse covariance matrix / second inverse covariance matrix that are below a first preset threshold are set to 0, and elements that are above the first preset threshold are set to 1.

6. The microgrid group fault diagnosis method based on multi-source data fusion according to claim 1, characterized in that, The node state probability matrix is ​​the result of the message passing layer performing multi-layer message passing on the node state feature matrix based on the microgrid group topology features; Among them, for the first l Layer message passing, all neighboring nodes of any node at the 1st layer l The node features generated by aggregating the node features obtained during the -1st message transmission, and the node features of any node in the -1st message transmission, are compared with the node features obtained during the second message transmission. l The node features obtained during the -1st message transmission are concatenated and subjected to a nonlinear transformation to generate the node at the -1st message transmission. l The characteristics of the first node during the next message transmission; The node in the first l The L2 norm normalization result of the first node's feature during the second message transmission is the value of any node in the first... l Node characteristics during message passing.

7. A microgrid group fault diagnosis device based on multi-source data fusion, characterized in that, The device includes: The first determining module is used to determine the sub-microgrids and node inclusion relationships of the microgrid group based on the node time-series load matrix during normal operation of the microgrid group; The second determining module is used to determine the changes in the sub-microgrid connection relationships of the microgrid group based on the node time-series load matrix when the microgrid group is near a fault. The third determining module is used to determine the topological characteristics of the microgrid group based on the sub-microgrid and node inclusion relationship and the changes in the sub-microgrid connection relationship; The fault diagnosis module is used to determine the fault diagnosis result of the microgrid group based on the pre-stored MPNN network model, the node waveform current when the microgrid group is faulty, and the topological characteristics of the microgrid group. The MPNN network model is trained using a labeled microgrid fault sample dataset and is used to locate and identify fault types in microgrids under different microgrid topologies and different node state matrices. The process of determining the fault diagnosis results of the microgrid group based on the MPNN network model, the node waveform currents during microgrid group faults, and the topological characteristics of the microgrid group includes: A first node state matrix is ​​generated based on the node waveform current recorded during the fault of the microgrid group. The first node state matrix is ​​substituted into the node network layer of the MPNN network model for feature extraction to obtain the node state feature matrix. The node state feature matrix and the microgrid group topology features are input into the message passing layer of the MPNN network model to obtain the node state probability matrix output by the message passing layer. The node state probability matrix is ​​input into the label mapping layer of the MPNN network model to obtain the node anomaly probability and node fault type label output by the label mapping layer; the microgrid topology features, the node anomaly probability, and the node fault type label are input into the fault location layer of the MPNN network model to obtain the fault type and fault line / fault area of ​​the microgrid output by the fault location layer.

8. An electronic 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 program, it implements the microgrid group fault diagnosis method based on multi-source data fusion as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the microgrid group fault diagnosis method based on multi-source data fusion as described in any one of claims 1 to 6.

Images