Broadband data oscillation positioning method and system based on depth map neural network

Abstract

The invention discloses a broadband data oscillation positioning method and system based on a depth map neural network, and relates to the technical field of power system safety and stability analysis, and the method comprises the steps: abstracting a power grid into an undirected graph, constructing a power grid topological graph, and extracting features; performing feature fusion through residual connection and a feature pyramid, adding a full connection layer, and constructing a depth map neural network model; and inputting data, carrying out model training, and carrying out broadband oscillation positioning by using the trained model. According to the method, residual connection and pyramid feature fusion are carried out, so that the deep training capability and the feature learning capability of the graph neural network are effectively improved, and multi-scale and multi-level graph feature learning is realized; by constructing the depth map neural network model, deep features of a power grid topological structure can be learned, more accurate broadband oscillation positioning is realized, and better effects are achieved in the aspects of improving positioning efficiency, reducing positioning misjudgment, reducing information loss and improving broadband oscillation precision.

Description

Technical Field

[0001] The present invention relates to the technical field of power system safety and stability analysis, and specifically to a method and system for locating oscillations of broadband data based on a deep graph neural network. Background Art

[0002] In recent years, graph neural networks have developed rapidly, and they can effectively learn non-Euclidean data. In the field of power systems, some literature applies message passing neural networks to the problem of transient stability assessment of power systems; some literature applies graph convolutional neural networks to the problem of cascading fault location in power systems, and obtains better classification results by constructing the fault location problem as a multi-classification task. Spatiotemporal graph neural networks introduce time series features on the basis of graph neural networks, and can simultaneously obtain the association of feature data in the time domain and the space domain. It is suitable for scenarios where node features change over time. Some literature proposes a multi-task transient stability assessment model based on recurrent graph convolutional neural networks by aggregating graph convolutional units and long short-term memory neural units; some literature applies spatiotemporal graph convolutional neural networks to the field of power system load forecasting to achieve spatiotemporal load forecasting. On this basis, using graph neural networks to analyze power grid topology to achieve oscillation location has inherent advantages.

[0003] In terms of locating the broadband oscillation source of the power grid, there are some problems with the traditional forced oscillation source positioning method based on physical mechanisms. Traditional methods such as the energy method, the traveling wave method and the modal analysis method have some limitations and defects under specific conditions, such as difficulty in accurate positioning in a specific power grid environment, the influence of complex power grid topology and misjudgment when the frequency is equal. Traditional machine learning methods require feature selection, which may lead to information loss and low efficiency. In addition, the deep learning method lacks full utilization of the power grid topology and cannot effectively reflect the characteristics of the power grid spatial information. Therefore, the present invention proposes a new method, that is, using a deep graph neural network in combination with power grid measurement data and power grid topology information to achieve power grid broadband oscillation positioning. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing technology has the problems of low positioning efficiency, information loss, misjudgment under the influence of complex power grid topology, and how to more accurately locate the broadband oscillation source.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for locating broadband data oscillations based on a deep graph neural network, comprising abstracting the power grid into an undirected graph, constructing a power grid topology map and extracting features; fusing features through residual connections and feature pyramids, adding a fully connected layer, and constructing a deep graph neural network model; inputting data, performing model training, and using the trained model to locate broadband oscillations.

[0007] As a preferred scheme of the broadband data oscillation positioning method based on deep graph neural network described in the present invention, wherein: the abstraction of the power grid into an undirected graph includes representing the bus nodes in the power grid as nodes V in the graph, and the electrical lines connecting the bus bars as lines between the nodes in the graph.

[0008] As a preferred solution of the broadband data oscillation positioning method based on deep graph neural network described in the present invention, wherein: the construction of the power grid topology map and the extraction of features include obtaining the adjacency matrix A of the power grid topology structure map, and the node feature matrix of the power grid node voltage, current, active and reactive power composition map is expressed as:

[0009]

[0010] Among them, X represents the node feature matrix, N represents the number of nodes, and d represents the dimension of the input feature.

[0011] As a preferred solution of the broadband data oscillation positioning method based on deep graph neural network described in the present invention, the residual connection and feature pyramid include residual connection by introducing cross-layer connection in the network and feature pyramid by fusing feature graphs of different scales; residual connection is to introduce cross-layer information flow, and residual learning is realized by directly adding the output of the previous layer to the input of the next layer. The formula of residual connection is expressed as:

[0012] Y c =H(x)+x

[0013] Among them, Y c Represents residual connection, x represents the input of the previous layer, and H(x) represents the input of the next layer. The feature pyramid accurately detects and recognizes objects of different sizes through multi-scale feature maps. The fusion formula of the feature pyramid is expressed as:

[0014] P i =U i (F i )+D i (P i+1 )

[0015] Among them, P i represents the i-th layer feature map of the pyramid, U irepresents the upsampling operation, F i Represents the feature maps of different layers, D i Represents a fusion operation.

[0016] As a preferred solution of the broadband data oscillation positioning method based on deep graph neural network described in the present invention, wherein: the feature fusion includes realizing multi-feature fusion of graph neural network through residual connection and feature pyramid to realize deep graph neural network; realizing deep graph neural network includes introducing residual block, introducing k graph convolution operations of different scales in the residual block of the lth layer, and the feature matrix of the kth layer is recorded as It is expressed as:

[0017]

[0018] Among them, ReLU represents the activation function, A represents the graph adjacency matrix, and D represents the degree matrix. represents the inverse square root of the degree matrix, Represents the kth weight matrix to be learned in the lth layer; introduce the pyramid feature matrix, concatenate the feature matrices of all scales, and record it as Z (l) , the shape is N×(d×K), expressed as:

[0019]

[0020] The pyramid feature matrix Z (l) Connected to the input node feature matrix and passed through a learned weight matrix Perform linear mapping to obtain the residual connection feature skip (l) , expressed as:

[0021]

[0022] The information of the residual connection is added to the output of the current layer, and the node feature H is updated through a graph convolution operation. (l+1) , expressed as:

[0023]

[0024] Among them, W (l+1) Represents the weight matrix to be learned in the l+1th layer, the residual connection feature skip (l) .

[0025] As a preferred solution of the broadband data oscillation positioning method based on deep graph neural network described in the present invention, the adding of the fully connected layer includes adding a fully connected layer F before the final output, which is expressed as:

[0026] F=Concat(ΣH (l) )

[0027] Among them, Concat represents the connection layer, ∑H (l) Represents all feature layers.

[0028] As a preferred solution of the broadband data oscillation positioning method based on deep graph neural network described in the present invention, the broadband oscillation positioning includes using softmax to achieve oscillation positioning, which is expressed as:

[0029] Y=Softmax(H (L) W f )

[0030] Among them, H (L) is the node feature matrix of the last layer, W f is the weight matrix of the fully connected layer.

[0031] Another object of the present invention is to provide a broadband data oscillation positioning system based on a deep graph neural network, which can accurately locate the broadband oscillation source by constructing and training a deep graph neural network model, thereby solving the problem of low positioning accuracy of the current energy method and traveling wave method.

[0032] As a preferred solution of the broadband data oscillation positioning system based on deep graph neural network described in the present invention, it includes: a feature extraction module, a feature fusion module, and a positioning training module; the feature extraction module is used to abstract the power grid into an undirected graph, construct a power grid topology map and extract features; the feature fusion module is used to perform feature fusion through residual connection and feature pyramid, increase the fully connected layer, and construct a deep graph neural network model; the positioning training module is used to input data, perform model training, and use the trained model to perform broadband oscillation positioning.

[0033] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a step of a broadband data oscillation positioning method based on a deep graph neural network.

[0034] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a broadband data oscillation positioning method based on a deep graph neural network.

[0035] Beneficial effects of the present invention: The broadband data oscillation positioning method based on deep graph neural network provided by the present invention enhances cross-layer information transmission and gradient feedback by introducing residual connections in the graph convolutional network, improves the model depth training effect, and avoids overfitting; by introducing graph convolution operations of different scales in each residual block, pyramid features are constructed, richer spatial information is extracted, and network performance and generalization ability are improved; by fusing residual connections and feature pyramid features, the deep training ability and feature learning ability of the graph neural network are effectively improved, and multi-scale and multi-level graph feature learning is realized; by constructing a deep graph neural network model, the deep characteristics of the power grid topology structure can be learned, and more accurate broadband oscillation positioning is achieved. The present invention achieves better results in improving positioning efficiency, reducing positioning misjudgment, reducing information loss, and improving broadband oscillation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0037] Figure 1 An overall flow chart of a broadband data oscillation positioning method based on a deep graph neural network provided for the first embodiment of the present invention.

[0038] Figure 2 A WECC 179 node diagram of a broadband data oscillation positioning method based on a deep graph neural network is provided for the first embodiment of the present invention.

[0039] Figure 3 Comparison results of the model accuracy of a broadband data oscillation positioning method based on a deep graph neural network provided in the second embodiment of the present invention.

[0040] Figure 4 An overall flow chart of a broadband data oscillation positioning system based on a deep graph neural network is provided for the third embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0042] Example 1, reference Figure 1-Figure 2 , as an embodiment of the present invention, provides a broadband data oscillation positioning method based on a deep graph neural network, comprising:

[0043] S1: Abstract the power grid into an undirected graph, construct the power grid topology map and extract features.

[0044] Furthermore, a graph can be represented as G = (V, E), where V represents the graph nodes and E represents the connections between nodes. The topological connection relationship of the graph can be represented by the node adjacency matrix A. The power grid is essentially a graph structure such as Figure 2 As shown, the power grid is abstracted as an undirected graph, the bus nodes in the power grid are represented as nodes V in the graph, the electrical lines connecting the busbars are represented as lines between the nodes in the graph, and the voltage, current and other characteristics of each bus node can be regarded as features on the node.

[0045] It should be noted that constructing the power grid topology graph and extracting features includes obtaining the adjacency matrix A of the power grid topology graph, and forming the node feature matrix of the power grid node voltage, current, active and reactive power graph, which is expressed as:

[0046]

[0047] Among them, X represents the node feature matrix, N represents the number of nodes, and d represents the dimension of the input feature.

[0048] S2: Perform feature fusion through residual connection and feature pyramid, add fully connected layers, and build a deep graph neural network model.

[0049] Furthermore, residual connection and feature pyramid include residual connection by introducing cross-layer connection in the network and feature pyramid by fusing feature maps of different scales; residual connection introduces cross-layer information flow, and realizes residual learning by directly adding the output of the previous layer to the input of the next layer. The formula of residual connection is expressed as:

[0050] Y c =H(x)+x

[0051] Among them, Y c Represents residual connection, x represents the input of the previous layer, and H(x) represents the input of the next layer. The feature pyramid accurately detects and recognizes objects of different sizes through multi-scale feature maps. The fusion formula of the feature pyramid is expressed as:

[0052] P i =U i (F i )+D i (P i+1 )

[0053] Among them, P i represents the i-th layer feature map of the pyramid, U i represents the upsampling operation, F i Represents the feature maps of different layers, D i Represents a fusion operation.

[0054] It should be noted that feature fusion includes realizing multi-feature fusion of graph neural network through residual connection and feature pyramid to realize deep graph neural network; realizing deep graph neural network includes introducing residual block, introducing k graph convolution operations of different scales in the residual block of the lth layer, and the feature matrix of the kth layer is recorded as It is expressed as:

[0055]

[0056] Among them, ReLU represents the activation function, A represents the graph adjacency matrix, and D represents the degree matrix. is defined as a diagonal matrix, represents the inverse square root of the degree matrix, Represents the kth weight matrix to be learned in the lth layer; introduce the pyramid feature matrix, concatenate the feature matrices of all scales, and record it as Z (l) , the shape is N×(d×K), expressed as:

[0057]

[0058] The pyramid feature matrix Z (l) Connected to the input node feature matrix and passed through a learned weight matrix Perform linear mapping to obtain the residual connection feature skip (l) , expressed as:

[0059]

[0060] The information of the residual connection is added to the output of the current layer, and the node feature H is updated through a graph convolution operation. (l+1) , expressed as:

[0061]

[0062] Among them, W (l+1) Represents the weight matrix to be learned in the l+1th layer, the residual connection feature skip (l) .

[0063] It should also be noted that adding a fully connected layer includes adding a fully connected layer F before the final output, expressed as:

[0064] F=Concat(∑H (l) )

[0065] Among them, Concat represents the connection layer, ∑H (l) Represents all feature layers.

[0066] It should also be noted that in order to realize deep graph neural networks, the more graph convolution layers are needed, the better. Deeper graph convolutions can obtain deeper receptive eyes, but the actual situation is not the case. Simply increasing the number of graph convolution layers will not only fail to obtain deeper receptive eyes, but will also cause the model to enter an overfitting state, thereby causing the accuracy of the graph neural network model to decrease. However, if there are no deeper receptive eyes, for the positioning of broadband oscillations in the power grid, the graph neural network cannot accurately locate a certain node, and the effect achieved may only be to locate a certain area.

[0067] To this end, feature fusion is introduced to further optimize the structure of graph neural networks and achieve the purpose of expanding the depth of graph neural networks. Feature fusion can merge feature maps of multiple convolutional layers or multiple branches. Feature fusion can improve the performance and generalization ability of the network. By fusing feature information at different levels and scales, graph features can be better captured, making the network have good discrimination and robustness.

[0068] S3: Input data, perform model training, and use the trained model to perform broadband oscillation positioning.

[0069] Furthermore, broadband oscillation localization includes using softmax to achieve oscillation localization, which is expressed as:

[0070] Y=Softmax(H (L) W f )

[0071] Among them, H (L) is the node feature matrix of the last layer, W f is the weight matrix of the fully connected layer.

[0072] Example 2, reference Figure 3 , as an embodiment of the present invention, a wideband data oscillation positioning method based on deep graph neural network is provided. In order to verify the test accuracy of the deep graph neural network model of the present invention, the application performance evaluation and comparative verification of the disturbance source positioning model are carried out based on the WECC179 node standard system provided by the IEEE oscillation source positioning working group.

[0073] First, the WECC179 node standard system consists of 179 busbars, 29 generators and multiple branches. In order to be closer to the actual synchronous phasor measurement data, three different data sets were set up in the experiment. Data set A is complete and noise-free, data set B is added with 45dB white noise, and data set C is partially missing data.

[0074] Secondly, the three data sets are trained in the graph convolutional network GCN, the graph attention network GAT, the graph isomorphism network GIN and the deep graph neural network TGCN of the present invention respectively. The training data is divided into 70% training set, 15% validation set and 15% test set ratios according to each database. The final test set accuracy is shown in Table 1.

[0075] Table 1 Performance indicators of the positioning model on dataset A

[0076] Model Dataset A Dataset B Dataset C GCN 0.851 0.867 0.848 GAT 0.873 0.891 0.872 GIN 0.911 0.943 0.933 TGCN 0.981 0.986 0.979

[0077] It can be seen from the data in the table that the graph convolutional network GCN achieves accuracies of 0.851, 0.867, and 0.848 on datasets A, B, and C respectively; the graph attention network GAT achieves accuracies of 0.873, 0.891, and 0.872 on datasets A, B, and C respectively; the graph isomorphism network GIN achieves accuracies of 0.911, 0.943, and 0.933 on datasets A, B, and C respectively; the deep graph neural network TGCN of the model of the present invention achieves accuracies of 0.981, 0.986, and 0.979 on datasets A, B, and C respectively.

[0078] GCN is one of the most basic graph neural networks. It performs convolution operations based on the features of adjacent nodes. GAT uses the attention mechanism to weight neighbor nodes, effectively improving the ability to distinguish the importance of different nodes. In terms of accuracy, GAT is slightly better than GCN, while GIN has a stronger ability to capture the complex relationships between power grid nodes than GCN and GAT.

[0079] Finally, TGCN performs best on all data sets, and has higher test accuracy than GIN. This shows that TGCN introduces residual connection and pyramid feature fusion into graph neural network, which can further promote cross-layer transmission of information, improve the deep training ability of the network, and greatly improve the accuracy of broadband oscillation source positioning.

[0080] like Figure 2 It can be clearly seen that the positioning accuracy of the positioning model based on TGCN on the three data sets is higher than that of other positioning models. From Table 1, it can be seen that TGCN performs well on multiple data sets. In contrast, GCN, as a basic model, is suitable for simpler graph structures, but has limited performance in complex scenarios. Although GAT and GIN have been improved, they are still not enough to surpass the multi-feature fusion advantage of TGCN. Therefore, the present invention has made great progress in solving the positioning of broadband oscillation sources.

[0081] Example 3, reference Figure 4, as an embodiment of the present invention, provides a broadband data oscillation positioning system based on deep graph neural network, including a feature extraction module, a feature fusion module, and a positioning training module.

[0082] The feature extraction module is used to abstract the power grid into an undirected graph, construct a power grid topology map and extract features; the feature fusion module is used to fuse features through residual connections and feature pyramids, add a fully connected layer, and build a deep graph neural network model; the positioning training module is used to input data, perform model training, and use the trained model to perform broadband oscillation positioning.

[0083] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0084] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0085] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0086] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0087] 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 preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A broadband data oscillation positioning method based on deep graph neural network, characterized in that: include: Abstract the power grid into an undirected graph, construct the power grid topology graph and extract features; Through residual connection and feature pyramid, feature fusion is performed, fully connected layers are added, and a deep graph neural network model is constructed; Input data, perform model training, and use the trained model to locate broadband oscillations.

2. The broadband data oscillation positioning method based on deep graph neural network according to claim 1, characterized in that: The abstraction of the power grid into an undirected graph includes representing busbar nodes in the power grid as nodes V in the graph, and representing electrical lines connecting busbars as lines between nodes in the graph.

3. The broadband data oscillation positioning method based on deep graph neural network according to claim 2, characterized in that: The construction of the power grid topology diagram and the extraction of features include obtaining an adjacency matrix A of the power grid topology diagram, and forming a node feature matrix of the power grid node voltage, current, active power and reactive power diagram, which is expressed as: Among them, X represents the node feature matrix, N represents the number of nodes, and d represents the dimension of the input feature.

4. The broadband data oscillation positioning method based on deep graph neural network according to claim 3, characterized in that: The residual connection and feature pyramid include residual connection by introducing cross-layer connection in the network and feature pyramid by fusing feature maps of different scales; Residual connection introduces cross-layer information flow, and realizes residual learning by directly adding the output of the previous layer to the input of the next layer. The formula of residual connection is expressed as: Y c =H(x)+x Among them, Y c represents the residual connection, x represents the input of the previous layer, and H(x) represents the input of the next layer; The feature pyramid accurately detects and recognizes objects of different sizes through multi-scale feature maps. The fusion formula of the feature pyramid is expressed as: P i =U i (F i )+D i (P i+1 ) Among them, P i represents the i-th layer feature map of the pyramid, U i represents the upsampling operation, F i Represents the feature maps of different layers, D i Represents a fusion operation.

5. The broadband data oscillation positioning method based on deep graph neural network according to claim 4, characterized in that: The feature fusion includes realizing multi-feature fusion of graph neural network through residual connection and feature pyramid to realize deep graph neural network; The implementation of deep graph neural network includes introducing residual blocks, introducing k graph convolution operations of different scales in the residual blocks of the lth layer, and the feature matrix of the kth layer is recorded as k=1,2,…,k, expressed as: Among them, ReLU represents the activation function, A represents the graph adjacency matrix, and D represents the degree matrix. represents the inverse square root of the degree matrix, represents the kth weight matrix to be learned in the lth layer; Introduce the pyramid feature matrix and concatenate the feature matrices of all scales, denoted as Z (l) , the shape is N×(d×K), expressed as: The pyramid feature matrix Z (l) Connected to the input node feature matrix and passed through a learned weight matrix Perform linear mapping to obtain the residual connection feature skip (l) , expressed as: The information of the residual connection is added to the output of the current layer, and the node feature H is updated through a graph convolution operation. (l +1) , expressed as: Among them, W (l+1) Represents the weight matrix to be learned in the l+1th layer, the residual connection feature skip (l) .

6. The broadband data oscillation positioning method based on deep graph neural network according to claim 5, characterized in that: The addition of a fully connected layer includes adding a fully connected layer F before the final output, which is expressed as: F=Concat(∑H (l) ) Among them, Concat represents the connection layer, ΣH (l) Represents all feature layers.

7. The broadband data oscillation positioning method based on deep graph neural network according to claim 6, characterized in that: The broadband oscillation positioning includes using softmax to achieve oscillation positioning, which is expressed as: Y =Softmax(H (L) W f ) Among them, H (L) is the node feature matrix of the last layer, W f is the weight matrix of the fully connected layer.

8. A system using the broadband data oscillation positioning method based on a deep graph neural network as claimed in any one of claims 1 to 7, characterized in that: Including feature extraction module, feature fusion module, positioning training module; The feature extraction module is used to abstract the power grid into an undirected graph, construct a power grid topology graph and extract features; The feature fusion module is used to perform feature fusion through residual connection and feature pyramid, add a fully connected layer, and build a deep graph neural network model; The positioning training module is used to input data, perform model training, and use the trained model to perform broadband oscillation positioning.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the broadband data oscillation positioning method based on deep graph neural network described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the broadband data oscillation positioning method based on deep graph neural network described in any one of claims 1 to 7 are implemented.

Images