Broadband oscillation signal positioning method and device based on composite intelligent network driving

Abstract

The invention relates to the technical field of electronics, and discloses a broadband oscillation signal positioning method and device based on composite intelligent network driving, and the method comprises the steps: inputting to-be-repaired graph structure data into a graph repair network, enabling the graph repair network to output the target feature information of each node after repair, and enabling the graph repair network to be constructed according to a target generative adversarial network; performing feature extraction on the target feature information of each node to obtain a plurality of signal features of the corresponding node; and positioning the broadband oscillation signal in the power system based on the plurality of signal characteristics of each node. According to the method provided by the invention, the to-be-repaired graph structure data is subjected to anomaly repair through the graph repair network, the target feature information of each node is obtained, the quality of the feature data of each node is effectively improved, signal feature extraction is performed based on the target feature information of each node, and the broadband oscillation signal in the power system is positioned based on the extraction result. And the accuracy of the broadband oscillation signal positioning result in the power system is ensured.

Description

Technical Field

[0001] The present invention relates to the field of electronic technology, and in particular to a method and device for locating a broadband oscillation signal based on a composite intelligent network drive. Background Art

[0002] Broadband oscillations in power systems are a frequent and potentially harmful phenomenon, and their identification and location are crucial for ensuring stable grid operation. Traditional methods suffer from insufficient recognition accuracy and weak interference immunity when faced with complex oscillation patterns and high-noise environments. To overcome these challenges, a growing number of studies have explored combining complex intelligent networks with advanced signal processing techniques to improve the accuracy and efficiency of oscillation signal location.

[0003] In related technologies, the widely used Phasor Measurement Unit (PMU) can provide high-frequency power system data. Spectral analysis of the phase signals collected by the PMU can be used to locate broadband oscillations in the power system based on the results of spectral analysis. However, in practical applications, the signals collected by the PMU are subject to data loss and noise interference. Spectral analysis based on the signals collected by the PMU cannot accurately locate broadband oscillation signals in the power system. Summary of the Invention

[0004] In view of this, the present invention provides a method and device for locating broadband oscillation signals based on a composite intelligent network drive to solve the problem in the related art that the signals collected by the PMU have data missing and noise infection, and the broadband oscillation signals in the power system cannot be accurately located by performing spectrum analysis on the signals collected by the PMU.

[0005] In the first aspect, the present invention provides a method for locating a broadband oscillation signal based on a composite intelligent network drive, the method comprising: obtaining a pre-constructed graph repair network and graph structure data to be repaired of a power system, the graph structure data to be repaired including feature information of multiple nodes and edge feature information representing the association relationship between the nodes, the multiple nodes corresponding one-to-one to multiple measurement points in the power system, the feature information of each node being used to represent the voltage and current phasor data of the corresponding measurement point, and the edge feature information being used to represent the electrical distance between the nodes, the graph repair network being constructed based on a target generative adversarial network, and being used to perform abnormal repair on the voltage and current phase data of each node in the graph structure data to be repaired; inputting the graph structure data to be repaired into the graph repair network so that the graph repair network outputs the target feature information of each node after repair; performing feature extraction on the target feature information of each node to obtain multiple signal features of the corresponding node; and locating the broadband oscillation signal in the power system based on the multiple signal features of each node.

[0006] The present invention provides a composite intelligent network-driven broadband oscillation signal location method. The method involves inputting graph structure data to be repaired into a graph repair network, which is constructed based on a target generative adversarial network, so that the network outputs target feature information for each node after repair. Feature extraction is performed on the target feature information for each node to obtain multiple signal features for the corresponding node. Based on these multiple signal features for each node, broadband oscillation signals in the power system are located. The method provides an improved quality of each node's feature data by performing anomaly repair on the graph structure data to be repaired using the graph repair network. Signal feature extraction is performed based on the target feature information for each node. The extraction results are used to locate broadband oscillation signals in the power system, ensuring the accuracy of broadband oscillation signal location results in the power system.

[0007] In an optional embodiment, the graph repair network is constructed by the following steps: obtaining a target generative adversarial network and a target dataset; training the target generative adversarial network using the target dataset until the accuracy of the target generative adversarial network meets preset conditions, thereby obtaining the graph repair network.

[0008] The method provided in this optional implementation manner utilizes a target data set to train a target generative adversarial network to obtain a graph repair network, which is capable of accurately repairing the graph structure data to be repaired.

[0009] In an optional embodiment, the target generative adversarial network includes a generator and a discriminator;

[0010] The generator includes an input layer, a multi-scale input module, and a graph attention network module; the input layer is used to introduce random noise into the acquired first graph structure data to obtain second graph structure data; the multi-scale input module is used to repair the feature information of each node in the input second graph structure data at different spatial scales to obtain third graph structure data; the graph attention network module is used to repair the feature information of each node in the third graph structure data and output fourth graph structure data;

[0011] The discriminator includes an input convolution layer, a multi-resolution block, a downsampling layer, a residual connection layer and an output layer. The input convolution layer is used to extract features from the input data to obtain a first node feature matrix, and the input data is the first graph structure data or the fourth graph structure data; the multi-resolution block includes multiple convolution layers, and the multi-resolution block uses multiple convolution layers to extract features from the first node feature matrix to obtain a second node feature matrix; the downsampling layer is used to calculate the scores of multiple nodes in the second node feature matrix, and based on the scores of each node, a preset number of target nodes are determined from the multiple nodes, and a third node feature matrix is constructed based on each target node. The fourth node feature matrix is determined based on the scores of each target node and the third node feature matrix; the residual connection layer is used to perform a residual connection between the fourth node feature matrix and the second node feature matrix to obtain a fifth node feature matrix; the output layer is used to calculate the probability that the input data is real data based on the fifth node feature matrix.

[0012] In an optional embodiment, the step of performing feature extraction on the target feature information of each node to obtain multiple signal features of the corresponding node includes: obtaining the modal number and penalty parameter; performing variational modal decomposition on the target feature information of each node using the modal number and penalty parameter to obtain multiple modal components of the corresponding node; processing each modal component of each node using a preset window function to obtain the corresponding target modal component; and performing spectral analysis on each target modal component of each node to obtain multiple signal features of the corresponding node.

[0013] In an optional embodiment, a spectral analysis is performed on each target modal component of each node to obtain multiple signal features of the corresponding node, including: performing Fourier transform on different target modal components of each node to obtain target spectra corresponding to different target modal components of the corresponding node; and determining multiple signal features of the corresponding node based on the target spectra corresponding to different target modal components of each node.

[0014] In an optional embodiment, multiple signal features of the corresponding node are determined based on the target spectra corresponding to different target modal components of each node, including: determining the spectral features of each target modal component of the corresponding node based on the different target modal components of each node; solving a pre-constructed signal model based on the spectral features of each target modal component to obtain signal parameters of the corresponding target modal component; and determining multiple signal features of the corresponding node using the signal parameters of the different target modal components of each node.

[0015] In the second aspect, the present invention provides a broadband oscillation signal positioning device driven by a composite intelligent network, which includes: an acquisition module for acquiring a pre-constructed graph repair network and graph structure data to be repaired of the power system, the graph structure data to be repaired includes feature information of multiple nodes and edge feature information representing the association relationship between the nodes, multiple nodes correspond one-to-one to multiple measurement points in the power system, the feature information of each node is used to represent the voltage and current phasor data of the corresponding measurement point, and the edge feature information is used to represent the electrical distance between the nodes. The graph repair network is constructed based on a target-generated adversarial network, and is used to perform abnormal repair on the voltage and current phase data of each node in the graph structure data to be repaired; a repair module is used to input the graph structure data to be repaired into the graph repair network, so that the graph repair network outputs the target feature information of each node after repair; an extraction module is used to perform feature extraction on the target feature information of each node to obtain multiple signal features of the corresponding node; a positioning module is used to locate the broadband oscillation signal in the power system based on the multiple signal features of each node.

[0016] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to execute the wide-band oscillation signal positioning method based on composite intelligent network driving according to the above-mentioned first aspect or any corresponding embodiment thereof.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the wide-band oscillation signal positioning method based on composite intelligent network driving according to the above-mentioned first aspect or any corresponding embodiment thereof.

[0018] In a fifth aspect, the present invention provides a computer program product comprising computer instructions for causing a computer to execute the wide-band oscillation signal positioning method based on composite intelligent network driving according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 1 is a flow chart of a method for locating a broadband oscillation signal based on a composite intelligent network driver according to an embodiment of the present invention;

[0021] Figure 2 is a flow chart of another method for locating a broadband oscillation signal based on a composite intelligent network driver according to an embodiment of the present invention;

[0022] Figure 3 is a structural block diagram of a broadband oscillation signal positioning device based on a composite intelligent network drive according to an embodiment of the present invention;

[0023] Figure 4 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0025] In related technologies, the widely used Phasor Measurement Unit (PMU) can provide high-frequency power system data. Spectral analysis of the phase signals collected by the PMU can be used to locate broadband oscillations in the power system based on the results of spectral analysis. However, in practical applications, the signals collected by the PMU are subject to data loss and noise interference. Spectral analysis based on the signals collected by the PMU cannot accurately locate broadband oscillation signals in the power system.

[0026] In view of this, the embodiments of the present application provide a method for locating broadband oscillation signals based on a composite intelligent network driver, which can be applied to a server to achieve broadband oscillation signal location in power systems. The method provided in the embodiments of the present application uses a graph repair network to perform anomaly repair on the graph structure data to be repaired, obtaining target feature information for each node, effectively improving the quality of each node's feature data. Signal feature extraction is performed based on the target feature information of each node, and based on the extraction results, broadband oscillation signals in the power system are located, ensuring the accuracy of the broadband oscillation signal location results in the power system.

[0027] According to an embodiment of the present invention, an embodiment of a wide-band oscillation signal positioning method based on a composite intelligent network driver is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] In this embodiment, a broadband oscillation signal positioning method based on a composite intelligent network driver is provided, which can be used for the above-mentioned server. Figure 1 FIG. 1 is a flow chart of a method for locating a broadband oscillation signal based on a composite intelligent network driver according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0029] Step S101: obtain a pre-built graph repair network and the graph structure data to be repaired of the power system. The graph structure data to be repaired includes feature information of multiple nodes and edge feature information representing the association relationship between the nodes. The multiple nodes correspond one-to-one to multiple measurement points in the power system. The feature information of each node is used to represent the voltage and current phasor data of the corresponding measurement point, and the edge feature information is used to represent the electrical distance between the nodes. The graph repair network is constructed based on the target generative adversarial network and is used to perform abnormal repair on the voltage and current phase data of each node in the graph structure data to be repaired.

[0030] For example, a power system is pre-configured with multiple measurement points. Phase measurement units (PMUs) collect voltage and current phasor data from each measurement point in the power system. The voltage and current phasor data collected by the PMUs is time-series data. This data is converted into a graph data structure G to be repaired. The repair graph data structure G contains multiple nodes and edges representing the relationships between nodes. G can represent the relationships and spatial relationships between nodes in the power system, allowing for better repair of missing and abnormal raw data and capturing the potential characteristics of the nodes. Each PMU measurement point in the power system serves as a node in the repair graph data structure G. A node can be represented by x. The characteristics of node x include the voltage and current phasor data of that measurement point. Importantly, the data for a node is actually composed of current and voltage phasor data at multiple time points. For a specific node, its voltage and current phasor values at different time points constitute the node's characteristic information. Edges between nodes are constructed based on the physical topology or electrical connectivity of the PMU deployment locations. Edges are represented by V. The edge characteristics of these edges can represent the electrical distance between nodes. This method of converting time series data into a graph data structure helps to use graph theory to analyze the complex relationships in the power system. By considering the correlation between nodes, missing values can be estimated and filled or abnormal data can be corrected more accurately. With the help of the graph structure, patterns or relationships between nodes can be better identified, thereby revealing system behavior characteristics that may not be easy to observe directly from the original time series data.

[0031] In an embodiment of the present application, the graph structure data to be repaired may also include an adjacency matrix A, which represents the connection relationship of the graph, that is, the connection relationship of each PMU measurement point, and its value is the electrical distance between the connected nodes. The adjacency matrix can clearly show the connection status of the edges between the nodes in the graph structure in the form of a mathematical matrix. The graph repair network is constructed based on the target generative adversarial network. In an embodiment of the present application, the target generative adversarial network is trained and the network parameters of the target generative adversarial network are continuously optimized to obtain the graph repair network. The target generative adversarial network is obtained by improving the generator and discriminator of the traditional generative adversarial network.

[0032] Step S102: inputting the graph structure data to be repaired into the graph repair network, so that the graph repair network outputs target feature information of each node after repair.

[0033] For example, the graph repair network repairs defective data in the graph structure data to be repaired, and ultimately generates high-quality repaired data.

[0034] Step S103: extract the target feature information of each node to obtain multiple signal features of the corresponding node.

[0035] For example, in an embodiment of the present application, a preset feature extraction algorithm may be used to extract target feature information of each node to obtain multiple signal features of the corresponding node.

[0036] Step S104 : locating the broadband oscillation signal in the power system based on multiple signal characteristics of each node.

[0037] For example, in the embodiments of the present application, multiple signal features can describe characteristics of a node signal, such as frequency, attenuation coefficient, amplitude, and phase, and are used to identify and distinguish multi-frequency oscillation components, detect the stability and characteristics of the oscillation, and provide a basis for locating the oscillation source. The embodiments of the present application do not limit the specific method for locating broadband oscillation signals, and those skilled in the art can determine this as needed.

[0038] This embodiment provides a method for locating broadband oscillation signals based on a composite intelligent network. The method inputs the graph structure data to be repaired into a graph repair network, which is constructed based on a target generative adversarial network, so that the network outputs target feature information for each node after repair. Feature extraction is performed on the target feature information for each node to obtain multiple signal features for the corresponding node. Based on these multiple signal features, broadband oscillation signals in the power system are located. The graph repair network repairs abnormalities in the graph structure data to be repaired, obtaining target feature information for each node. This effectively improves the quality of each node's feature data. Signal feature extraction is performed based on the target feature information for each node. Based on the extracted results, broadband oscillation signals in the power system are located, ensuring the accuracy of broadband oscillation signal location results in the power system.

[0039] In this embodiment, a broadband oscillation signal positioning method based on a composite intelligent network driver is provided, which can be used for the above-mentioned server. Figure 2 FIG. 1 is a flow chart of a method for locating a broadband oscillation signal based on a composite intelligent network driver according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0040] Step S201: Obtain a pre-built graph repair network and the graph structure data to be repaired for the power system. The graph structure data to be repaired includes feature information of multiple nodes and edge feature information representing the relationship between nodes. The multiple nodes correspond one-to-one to multiple measurement points in the power system. The feature information of each node is used to represent the voltage and current phasor data of the corresponding measurement point, and the edge feature information is used to represent the electrical distance between nodes. The graph repair network is constructed based on the target generative adversarial network and is used to repair anomalies in the voltage and current phase data of each node in the graph structure data to be repaired. For details, please refer to Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0041] Step S202: inputting the graph structure data to be repaired into the graph repair network, so that the graph repair network outputs target feature information of each node after repair.

[0042] In some optional implementations, the graph repair network is constructed by the following steps:

[0043] Step a1: Obtain a target generative adversarial network and a target dataset. For example, in the embodiment of the present application, the target dataset can be constructed graph structure data extracted from PMU data as the real data Xreal in training.

[0044] Specifically, the target generative adversarial network consists of a generator and a discriminator. For example, the generator (G) is used to input random noise (such as a Gaussian distribution) and output simulated "fake data." The discriminator (D) outputs a probability value (0-1) based on the input real data or the fake data output by the generator, determining whether the data is "real" or "fake."

[0045] The generator includes an input layer, a multi-scale input module and a graph attention network module; the input layer is used to introduce random noise into the acquired first graph structure data to obtain second graph structure data; the multi-scale input module is used to repair the feature information of each node in the input second graph structure data at different spatial scales to obtain third graph structure data, and the graph attention network module is used to repair the feature information of each node in the third graph structure data and output fourth graph structure data.

[0046] Exemplarily, the first graph structure data may include but is not limited to PMU graph structure data with missing data. In an embodiment of the present application, the input layer of the generator is used to introduce a random noise vector into the first graph structure data. The random noise vector is used to introduce random data, thereby enhancing the diversity and generalization ability of the generative adversarial network model. Input data N (N is a set containing the features of all nodes), and for each node, a random noise vector z is generated, usually extracted from a standard normal distribution. Before splicing, the noise vector z is nonlinearly transformed by a simple fully connected neural network to match the dimension of the node features. The random noise vector z is added to the node features. The first graph structure data with original defective data is combined with random noise as the input of the generator to generate complete data.

[0047] The main purpose of the multi-scale input module is to process the input nodes at different spatial scales, enhancing the ability to extract local information to obtain a richer and more comprehensive representation of node voltage and current vector data. By combining multiple convolutional layers and sampling operations, it is possible to extract information from details to the global picture. The data processing process of the multi-scale input module can be shown as follows (1):

[0048] Gx=G scalr1 (x)+up(G scale2 (down(x)))+up(up(G scale3 (down(down(x))))) (1)

[0049] Among them, down represents the downsampling operation, up represents the upsampling operation, G scale1 , G scale2 and G scale3 They represent convolutional layers of different scales, Gx represents the third graph structure data, and x represents the second graph structure data.

[0050] The above formula (1) combines multi-scale features in a multi-scale input module. The role of downsampling is to reduce the spatial resolution of the input data so that the model can capture larger-scale context information. The role of upsampling is to increase the spatial resolution of the data to restore it to the original scale. First, x passes through the convolutional layer G scale1 Processing and extracting feature information of the initial scale. This process can capture the basic features of the node data.

[0051] Secondly, the input data x is scaled down by downsampling operation down(x), which is implemented by average pooling to capture more local details. The downsampled data down(x) is passed through the second convolutional layer G scale2 Processing is performed to extract features at a smaller scale. The extracted features are processed by upsampling operation up(G scale2 (down(x))), upsampling is achieved by max pooling to capture more global information and restore to the same resolution as the initial scale.

[0052] Again, the input data x is downsampled twice, down(down(x)), to further reduce the scale and extract more context information. The downsampled data is passed through the third convolutional layer G scale3 Processing, extracting the features at the minimum scale. The extracted features are subjected to two upsampling operations up(up(G scale3 (down(down(x))))), restore to the same resolution as the initial scale.

[0053] Finally, features G from three different scales scale1 (x), up(G scale2 (down(x))) and up(up(G scale3 (down(down(x))))) performs element-by-element fusion to form an output containing multi-scale information. Defective data is repaired by capturing multi-scale local features within each node. The final output is Gx.

[0054] The purpose of using the Graph Attention Network module is to achieve efficient processing of graph-structured data by combining the advantages of the attention mechanism and graph neural networks. This module not only dynamically adjusts the weights between nodes but also captures the complex relationships between nodes and edges in the graph, improving the accuracy and robustness of data restoration. The Graph Attention Network module consists of the following four parts:

[0055] (1) Graph convolution: The feature map from the multi-scale input module is input into the graph neural network. By aggregating the feature information of the node and its neighboring nodes, the feature representation of each node is updated. It can capture the local structure of the graph and understand the relationship between different nodes. Using this information, it can repair incomplete node features caused by defective data.

[0056] Update the features of each node using the following formula (2):

[0057]

[0058] Among them, the characteristics of node i Through the features of its neighbor node j According to the normalization factor c ij The weighted sum is obtained, where Gx i ∈Gx,W (l) is the weight matrix of the current layer l, σ is the activation function, and N(i) represents the neighborhood set of node i (i.e., the set of nodes connected to node i).

[0059] (2) Attention mechanism: The graph attention mechanism is used in the graph convolution process to adaptively assign different weights to each node and its neighboring nodes, ensuring that important node features are more strongly represented, thereby better repairing the missing parts.

[0060] The attention weight between node i and its neighbor node j is calculated by the following formula (3):

[0061]

[0062] Among them, a ij is the attention weight between node i and its neighbor node j, a T is a trainable attention vector, || represents the feature concatenation operation, [WGx i ||WGx j ] represents the transformed node feature WGx i Concatenated with WGx, W is the weight matrix, N(i) represents the neighborhood set of node i (that is, the set of nodes connected to node i); LeakyReLU represents an activation function that allows negative inputs to have smaller gradients.

[0063] By combining graph convolution and attention mechanisms, an enhanced node feature graph is output, which represents graph data with more accurate features. Based on the multi-scale input in the previous step, the defective data is repaired by further utilizing the features of adjacent nodes and considering the edges.

[0064] The discriminator includes an input convolution layer, a multi-resolution block, a downsampling layer, a residual connection layer and an output layer. The input convolution layer is used to extract features from the input data to obtain a first node feature matrix, and the input data is the first graph structure data or the fourth graph structure data; the multi-resolution block includes multiple convolution layers, and the multi-resolution block uses multiple convolution layers to extract features from the first node feature matrix to obtain a second node feature matrix; the downsampling layer is used to calculate the scores of multiple nodes in the second node feature matrix, and based on the scores of each node, a preset number of target nodes are determined from the multiple nodes, and a third node feature matrix is constructed based on each target node. The fourth node feature matrix is determined based on the scores of each target node and the third node feature matrix; the residual connection layer is used to perform a residual connection between the fourth node feature matrix and the second node feature matrix to obtain a fifth node feature matrix; the output layer is used to calculate the probability that the input data is real data based on the fifth node feature matrix.

[0065] Exemplarily, in an embodiment of the present application, the first graph structure data will obtain repaired graph structure data after passing through the generator, and the task of the discriminator is to judge the authenticity of the graph structure data repaired by the generator. The input data of the generator includes two parts: real data and generated data. The first graph structure data refers to real data, and the real data represents the collected graph structure data containing real (but may be partially missing) features. These data serve as positive samples of the discriminator to help the discriminator identify the distribution of real data. The generated data represents the repaired graph data output after processing by the generator. These data serve as negative samples of the discriminator to help the discriminator learn how to distinguish between generated data and real data.

[0066] The input convolution layer performs preliminary feature extraction on the input data (real data or generated data) through a convolution layer. This convolution layer performs a convolution operation Conv2D and a ReLU activation function on each node of the graph data to capture the basic features of the input data.

[0067] The multi-resolution block is an improved convolution module that extracts features of different resolutions. The multi-resolution block consists of multiple convolutional layers with different convolution kernel sizes, which are set according to the dimension of the node features (such as 3x3, 5x5, 7x7, etc.) to extract features of different scales. In each multi-resolution block, the outputs of different convolutional layers are fused through a pixel-by-pixel addition operation and then passed through a ReLU activation function. In order to prevent gradient disappearance and model overfitting, residual connections can be used between different multi-resolution blocks. These connections make it easier for information to flow in the network, improving training efficiency and model performance. Residual connections also help the model learn more complex features because the original features and the convolutional features can be jointly involved in decision-making. The first node feature matrix is output after the convolution processing of the multi-resolution block to obtain the second node feature matrix. The second node feature matrix can be shown as follows (4):

[0068] X'=MultiConv(X) (4)

[0069] Among them, X' represents the second node feature matrix, MultiConv represents the multi-convolution operation, and X represents the first node feature matrix.

[0070] The downsampling layer uses graph pooling to reduce the resolution of the input graph structure and consider the overall characteristics of the graph structure from a deeper level. Consider using node selection pooling (Top-k Pooling) to implement this. First, calculate the importance score of the node and use a learnable function to calculate the score s of each node as shown below (5):

[0071] s=X′p(5)

[0072] Where X′ represents the second node feature matrix, s represents the score vectors of different nodes in the second node feature matrix, and p represents the learnable weight vector.

[0073] Based on the calculated score vector s, the k nodes with the highest scores are selected for retention. The pooling ratio r is a hyperparameter that represents the proportion of nodes retained after each pooling. N is the total number of nodes, so k = r × N. Here, k is the number of nodes after downsampling. The k nodes with the highest scores are extracted from the second node feature matrix X′ to form a new subgraph. At this time, the node feature matrix is updated to the third node feature matrix X″. For the retained k nodes, the third node feature matrix X″ is normalized according to the score vector of each node to obtain the fourth feature matrix, as shown in the following formula (6):

[0074] X″′=X″⊙tanh(s′) (6)

[0075] Where X″′ represents the fourth characteristic matrix, tanh is the hyperbolic tangent function, and s′ is the score vector of the selected k nodes.

[0076] The residual connection layer performs a residual connection on the output of the multi-resolution block and the data obtained by downsampling. Before the connection, because the downsampling data will have a reduced dimension, the dimension is first adjusted to the same as X' by appropriate padding. In the embodiment of the present application, the fourth node feature matrix is residually connected with the second node feature matrix by the following formula (7):

[0077] X res =X′+X′′′ (7)

[0078] Among them, X res Represents the fifth node feature matrix after residual connection.

[0079] The output layer of the discriminator is a fully connected layer. The feature matrix of the fifth node is passed to the last fully connected layer, and a scalar D(X) is output. This value is converted into a probability between 0 and 1 through the Sigmoid function. D(X) is calculated by the following formula (8):

[0080] D(X)=σ(f out (X res ))(8)

[0081] Among them, σ represents the Sigmoid activation function, which maps the output value to the [0,1] interval, indicating the probability; f out represents the last fully connected layer (output layer) of the discriminator, which converts the input feature into a scalar value; X res The residual feature map or intermediate feature representation after processing by all previous layers of the discriminator (such as convolutional layers, residual blocks, etc.) is used as the input of the output layer.

[0082] The output of the discriminator, D(X), is a scalar that represents the probability that the input graph data is real data. If D(X) ≈ 1, the discriminator considers the input to be real graph structure data; if D(X) ≈ 0, the input is considered to be generated graph structure data.

[0083] Step a2: Use the target data set to train the target generative adversarial network until the accuracy of the target generative adversarial network meets the preset conditions, thereby obtaining the graph restoration network.

[0084] For example, in an embodiment of the present application, before training the target generative adversarial network, it is necessary to construct a loss function. The loss function should combine GAN loss, consistency loss and cycle consistency loss to ensure the rationality and consistency of data filling.

[0085] GAN loss is the core loss used to train the generator G and the discriminator D, so that the generator can generate more realistic data and the discriminator can effectively distinguish between real data and generated data. GAN loss is shown in the following formula (9):

[0086] L GAN =E x～pdata(xreal) [logD(x real )]+E z～pz(z) [log(1-D(G(x real ,z)))] (9)

[0087] Among them, D(x real ) is the discriminator output for the input data x real Judgment of G(x real ,z) is the filling data generated by the generator, and the input is the signal x real and random noise vector z, E x～pdata(xreal) For the real data x real The expected value, E z～pz(z) is the expected value of the random noise vector z.

[0088] The consistency loss is shown in the following formula (10):

[0089]

[0090] Among them, L consistency represents the consistency loss, and ||G(x,z)-x||1 is the L1 norm between the generated data and the real data.

[0091] The cycle consistency loss is shown in the following formula (11):

[0092]

[0093] Among them, L cycle represents the cycle consistency loss value, and G(G(x,z),z)-x||1 is the L1 norm between the two generated data and the real data.

[0094] The total loss function is shown in the following formula (12):

[0095] L total =L GAN +λ1L consistency +λ2L cycle (12)

[0096] Among them, λ1 and λ2 are weight factors used to balance the impact of different loss terms.

[0097] By constructing a loss function that includes GAN loss, consistency loss, and cycle consistency loss, the generator is not only trained to generate realistic graph data, but also to keep the structure and features of the graph highly consistent with the original data.

[0098] When training a target generative adversarial network using a target dataset, adversarial training forces the generator and discriminator to compete with each other. The generator attempts to generate more realistic data to deceive the discriminator, while the discriminator attempts to better distinguish between real data and generated data. This competition drives the two networks to continuously optimize, ultimately generating high-quality inpainted data. Adversarial training involves the following steps:

[0099] 1. Initialization parameters: Parameter initialization of generator G and discriminator D: First, randomly initialize the weight parameters of generator G and discriminator D. Usually, uniform distribution or Gaussian distribution is used for initialization to ensure that the model parameters have appropriate initial values at the beginning of training.

[0100] 2. Prepare training data: Get real data: Extract the constructed graph structure data from the PMU data as the real data X in training real .

[0101] 3. Cycle training: the real data X real Input into the discriminator D, calculate the output D(X real ). Use the generator G to output the generated data X=G(Z, X real ). Input the generated data X into the discriminator D and calculate the output D(X) of the discriminator for the generated data.

[0102] Repeat the above training steps until the preset convergence condition is reached (that is, the constructed total loss function tends to be stable, or the maximum number of training rounds is reached).

[0103] By introducing multiple processing modules in the generator and discriminator, and combining multiple loss functions, the improved GAN network has higher accuracy and robustness when processing defect data repair of broadband oscillation signals in power systems. Ultimately, the generator is used to generate repaired defect raw data for subsequent analysis.

[0104] Step S203: extract the target feature information of each node to obtain multiple signal features of the corresponding node.

[0105] Specifically, the above step S203 includes:

[0106] Step S2031, obtaining the modal number and penalty parameter.

[0107] For example, in the embodiments of the present application, the modal number K and the penalty parameter α can be calculated using the particle swarm optimization (PSO) algorithm. This is an optimization algorithm based on swarm intelligence that optimizes complex objective functions by simulating the foraging process of a flock of birds. The goal is to find the optimal VMD parameter combination (modal number K and penalty parameter α). The steps are as follows:

[0108] (1) Initialize the particle swarm: Initialize multiple particles. The dimension of each particle represents the parameters of VMD. A particle consists of the modal number K and the penalty parameter α. Therefore, each particle is a two-dimensional vector. The dimension here refers to the dimension of the particle vector.

[0109] (2) Evaluate particle fitness: For each particle, perform VMD decomposition with the current K and α and calculate the fitness function, such as the reconstruction error or the orthogonality index between modes.

[0110] (3) Update particle speed and position: Update the speed and position of each particle based on its current speed, position, personal best historical position, and global best position. The position of the particle represents the current VMD parameter value, i.e., the current value of K and α, and the speed of the particle represents the parameter update step. In each iteration, the VMD modal components generated based on the current parameters will be extracted with key features (such as center frequency overlap, energy entropy distribution, and modal bandwidth), and the decomposition quality will be quantified by the preset fitness function. If abnormal features such as modal overlap or energy dispersion are detected, the system will trigger dynamic repair (such as merging overlapping modes or eliminating low-energy components) and recalculate the fitness value. PSO adjusts the particle speed and position by comparing the repaired feature fitness with the historical optimal: the parameter combination with higher fitness will guide the particle to move in a better search direction. At the same time, the gradient feedback mechanism maps the feature difference into a correction term in the parameter space, further refining the adjustment step. This closed-loop association not only improves the goal-oriented nature of parameter optimization, but also avoids invalid search through real-time feature repair, ultimately achieving a synergistic improvement in VMD decomposition accuracy and efficiency.

[0111] (4) Iterative optimization: Repeat the evaluation and update steps until the maximum number of iterations is reached or the fitness function converges, and the optimized modal number K and penalty parameter α are obtained.

[0112] The particle velocity is updated by the following equation (13), and the particle position is updated by the following equation (14):

[0113] v i (t+1)=w·v i (t)+c1·r1·(p i -x i (t))+c2·r2·(gx i(t))x i (t+1)=x i (t)+v i (t+1) (13)

[0114] Among them, v i (t) represents the velocity of particle i at the tth iteration, x i (t) represents the position of particle i, p i is the historical best position of particle i, g is the global best position, w, c1, c2 are the control parameters of the algorithm, and r1 and r2 are random numbers.

[0115]

[0116] in, represents the historical position of particle i at the tth iteration, that is, the parameter value before the update, represents the velocity of particle i at the t+1th iteration, which is calculated by formula (13) and determines the direction and step size of parameter update. It represents the position of particle i at the t+1th iteration and the updated value of the current parameters (such as the combination of the modal number K and the penalty factor α of VMD).

[0117] Step S2032: performing variational modal decomposition on the target feature information of each node using the modal number and penalty parameter to obtain multiple modal components of the corresponding node.

[0118] For example, in the embodiment of the present application, the target feature information of each node is subjected to variational modal decomposition using the optimized modal number K and penalty parameter α to obtain multiple modal component IMFs of the corresponding node, each modal component representing a frequency component in the signal. These modal functions are used for subsequent spectrum analysis. The modal component IMFs are calculated using the following formula (15):

[0119] IMFs=VMD(Preprocessed Signal) (15)

[0120] Among them, VMD stands for Variational Mode Decomposition; Preprocessed signal refers to the preprocessed signal, which in the embodiment of the present application refers to the target feature information of the node.

[0121] Step S2033: Process each modal component of each node using a preset window function to obtain a corresponding target modal component.

[0122] Exemplarily, in an embodiment of the present application, the preset window function may include but is not limited to a Hamming window and a Hamming window, and a suitable window function is applied to the modal signal obtained by VMD decomposition to reduce edge effects and improve the accuracy of spectrum estimation.

[0123] Step S2034 , performing spectrum analysis on each target modal component of each node to obtain multiple signal features of the corresponding node.

[0124] In some optional implementations, the above step S2034 includes:

[0125] Step b1: Perform Fourier transform on different target modal components of each node to obtain target frequency spectra corresponding to different target modal components of the corresponding node.

[0126] For example, in the present embodiment, a Fast Fourier Transform (FFT) is performed on the different target modal components of each node to obtain their spectral representation. The spectrum is analyzed to identify key frequency components and oscillation modes in the signal, and spectral resolution and stability can be improved through overlapping weighted averaging.

[0127] Step b2: determining a plurality of signal features of the corresponding node based on the target frequency spectra corresponding to the different target modal components of each node.

[0128] Specifically, the above step b2 includes:

[0129] Step b21: determining the frequency spectrum characteristics of each target modal component of the corresponding node based on the different target modal components of each node.

[0130] Illustratively, in the embodiment of the present application, according to the target spectra corresponding to different target modal components of each node, the spectrum characteristics of each target modal component can be obtained.

[0131] Step b22: solving the pre-built signal model based on the frequency spectrum characteristics of each target modal component to obtain the signal parameters corresponding to the target modal component.

[0132] For example, in an embodiment of the present application, after the signal spectrum feature extraction is completed, the signal is modeled and parameter estimation is performed. Based on the Prony algorithm, the multimodal signal obtained after VMD decomposition and the results obtained by spectrum analysis are used to model the signal. The Prony algorithm is a mathematical method for analyzing and fitting signals. The Prony algorithm can use the frequency characteristics of the modal signals obtained by these decompositions and analyses to perform accurate signal modeling. The Prony algorithm is a linear prediction method for signal analysis that can estimate parameters such as the attenuation index, frequency, amplitude, and phase of the signal, and is an important tool for signal modeling.

[0133] The pre-built signal model can be expressed as follows:

[0134]

[0135] Among them, A k is the amplitude of the modal signal, σ k is the attenuation exponent of the modal signal, ω k is the frequency of the modal signal, j is the imaginary unit, e(t) is the noise term or modeling error, and K is the order of the signal model, which is determined by frequency domain analysis and information criteria (such as AIC and BIC).

[0136] When solving the Prony model parameters, an L2 regularization term (ridge regression) is added to prevent overfitting and improve stability. The solved Prony model parameters are used as the signal features of the corresponding target modal components.

[0137] Step b23: using the signal parameters of different target modal components of each node to determine multiple signal features of the corresponding node.

[0138] For example, in the embodiments of the present application, after completing the Prony algorithm, a set of accurate signal parameters is obtained. These parameters describe the signal's characteristics, such as frequency, attenuation coefficient, amplitude, and phase. These parameters are used to identify and distinguish multi-frequency oscillation components, detect the stability and characteristics of the oscillation, and provide a basis for locating the oscillation source.

[0139] Step S204: Locate the broadband oscillation signal in the power system based on multiple signal characteristics of each node. Figure 1 Step S104 of the illustrated embodiment will not be described in detail here.

[0140] The method provided in the embodiment of the present application integrates generative adversarial networks and super-resolution technology to improve the generative adversarial network (GAN) to achieve the repair of defective data and signal enhancement, thereby improving data quality. Combining variational mode decomposition (VMD) and particle swarm optimization (PSO), efficient signal decomposition and parameter optimization are achieved, and the accuracy of modal decomposition is improved. An improved Prony algorithm is used in combination with multimodal signal modeling. The Prony algorithm is applied to each mode after VMD decomposition to obtain its attenuation characteristics and frequency components. The Prony model results of all modes are combined to analyze the global oscillation mode, and regularization is introduced to limit excessive fluctuations in parameters, thereby improving the accuracy of broadband oscillation detection and achieving the purpose of improving the ability to analyze complex oscillation signals.

[0141] In this embodiment, a broadband oscillation signal positioning device based on a composite intelligent network driver is also provided. This device is used to implement the above-mentioned embodiments and preferred embodiments, and the details already described will not be repeated. As used below, the term "module" can mean a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0142] This embodiment provides a broadband oscillation signal positioning device based on a composite intelligent network drive, such as Figure 3 Shown, including:

[0143] Acquisition module 301 is used to acquire a pre-constructed graph repair network and graph structure data to be repaired of the power system. The graph structure data to be repaired includes feature information of multiple nodes and edge feature information representing the relationship between the nodes. The multiple nodes correspond one-to-one to multiple measurement points in the power system. The feature information of each node is used to represent the voltage and current phasor data of the corresponding measurement point, and the edge feature information is used to represent the electrical distance between the nodes. The graph repair network is constructed based on the target generative adversarial network and is used to repair anomalies in the voltage and current phase data of each node in the graph structure data to be repaired.

[0144] The repair module 302 is used to input the graph structure data to be repaired into the graph repair network, so that the graph repair network outputs the target feature information of each node after repair;

[0145] The extraction module 303 is used to extract the target feature information of each node to obtain multiple signal features of the corresponding node;

[0146] The positioning module 304 is configured to locate the broadband oscillation signal in the power system based on multiple signal characteristics of each node.

[0147] In some optional implementations, the graph repair network is constructed by the following steps:

[0148] Obtain the target generative adversarial network and target dataset;

[0149] The target generative adversarial network is trained using the target dataset until the accuracy of the target generative adversarial network meets the preset conditions, and a graph restoration network is obtained.

[0150] In some optional embodiments, the target generative adversarial network includes a generator and a discriminator;

[0151] The generator includes an input layer, a multi-scale input module, and a graph attention network module; the input layer is used to introduce random noise into the acquired first graph structure data to obtain second graph structure data; the multi-scale input module is used to repair the feature information of each node in the input second graph structure data at different spatial scales to obtain third graph structure data; the graph attention network module is used to repair the feature information of each node in the third graph structure data and output fourth graph structure data;

[0152] The discriminator includes an input convolution layer, a multi-resolution block, a downsampling layer, a residual connection layer and an output layer. The input convolution layer is used to extract features from the input data to obtain a first node feature matrix, and the input data is the first graph structure data or the fourth graph structure data; the multi-resolution block includes multiple convolution layers, and the multi-resolution block uses multiple convolution layers to extract features from the first node feature matrix to obtain a second node feature matrix; the downsampling layer is used to calculate the scores of multiple nodes in the second node feature matrix, and based on the scores of each node, a preset number of target nodes are determined from the multiple nodes, and a third node feature matrix is constructed based on each target node. The fourth node feature matrix is determined based on the scores of each target node and the third node feature matrix; the residual connection layer is used to perform a residual connection between the fourth node feature matrix and the second node feature matrix to obtain a fifth node feature matrix; the output layer is used to calculate the probability that the input data is real data based on the fifth node feature matrix.

[0153] In some optional implementations, the extraction module 303 includes:

[0154] Get submodule to get the number of modes and penalty parameters;

[0155] The decomposition submodule is used to perform variational modal decomposition on the target feature information of each node using the modal number and penalty parameter to obtain multiple modal components of the corresponding node;

[0156] The processing submodule is used to process each modal component of each node using a preset window function to obtain the corresponding target modal component;

[0157] The analysis submodule is used to perform spectrum analysis on each target modal component of each node to obtain multiple signal features of the corresponding node.

[0158] In some optional embodiments, the analysis submodule includes:

[0159] A transform unit is used to perform Fourier transform on different target modal components of each node to obtain target spectra corresponding to different target modal components of the corresponding node;

[0160] The determination unit is used to determine multiple signal features of the corresponding node based on the target frequency spectra corresponding to different target modal components of each node.

[0161] In some optional implementations, the determining unit includes:

[0162] A first determining subunit is configured to determine the frequency spectrum characteristics of each target modal component of the corresponding node based on different target modal components of each node;

[0163] A solving subunit, configured to solve a pre-built signal model based on the frequency spectrum characteristics of each target modal component to obtain signal parameters corresponding to the target modal component;

[0164] The second determining subunit is used to determine multiple signal features of the corresponding node using signal parameters of different target modal components of each node.

[0165] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0166] The broadband oscillation signal positioning device based on composite intelligent network drive in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0167] The embodiment of the present invention also provides a computer device having the above Figure 3 The broadband oscillation signal positioning device shown is based on a composite intelligent network drive.

[0168] See also Figure 4 , Figure 4 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 4 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of a GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 4 A processor 10 is taken as an example.

[0169] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0170] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0171] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0172] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0173] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0174] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0175] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0176] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A broadband oscillation signal positioning method based on composite intelligent network drive, characterized in that: The method comprises: Obtaining a pre-built graph repair network and graph structure data to be repaired for the power system. The graph structure data to be repaired includes feature information of multiple nodes and edge feature information representing the association relationship between the nodes. The multiple nodes correspond one-to-one to multiple measurement points in the power system. The feature information of each node is used to represent the voltage and current phasor data of the corresponding measurement point, and the edge feature information is used to represent the electrical distance between the nodes. The graph repair network is constructed based on a target generative adversarial network and is used to repair anomalies in the voltage and current phase data of each node in the graph structure data to be repaired. Inputting the graph structure data to be repaired into the graph repair network so that the graph repair network outputs target feature information of each node after repair; Extract the target feature information of each node to obtain multiple signal features of the corresponding node; The broadband oscillation signal in the power system is located based on multiple signal characteristics of each node.

2. The method according to claim 1, characterized in that The graph repair network is constructed through the following steps: Obtain the target generative adversarial network and target dataset; The target generative adversarial network is trained using the target dataset until the accuracy of the target generative adversarial network meets a preset condition, thereby obtaining the graph restoration network.

3. The method according to claim 2, characterized in that The target generative adversarial network includes a generator and a discriminator; The generator includes an input layer, a multi-scale input module, and a graph attention network module; the input layer is used to introduce random noise into the acquired first graph structure data to obtain second graph structure data; the multi-scale input module is used to repair the feature information of each node in the input second graph structure data at different spatial scales to obtain third graph structure data; the graph attention network module is used to repair the feature information of each node in the third graph structure data and output fourth graph structure data; The discriminator includes an input convolution layer, a multi-resolution block, a downsampling layer, a residual connection layer, and an output layer. The input convolution layer is used to extract features from input data to obtain a first node feature matrix, where the input data is the first graph structure data or the fourth graph structure data. The multi-resolution block includes multiple convolution layers, and the multi-resolution block uses the multiple convolution layers to extract features from the first node feature matrix to obtain a second node feature matrix. The downsampling layer is used to calculate scores of multiple nodes in the second node feature matrix, determine a preset number of target nodes from the multiple nodes based on the scores of each node, construct a third node feature matrix based on each target node, and determine a fourth node feature matrix based on the scores of each target node and the third node feature matrix; The residual connection layer is used to perform a residual connection on the fourth node feature matrix and the second node feature matrix to obtain a fifth node feature matrix; The output layer is used to calculate the probability that the input data is true data according to the fifth node feature matrix.

4. The method according to claim 3, characterized in that The step of extracting the target feature information of each node to obtain multiple signal features of the corresponding node includes: Get the number of modes and penalty parameters; Performing variational modal decomposition on the target feature information of each node using the modal number and penalty parameter to obtain multiple modal components of the corresponding node; Using a preset window function to process each modal component of each node to obtain a corresponding target modal component; The spectrum analysis is performed on each target modal component of each node to obtain multiple signal features of the corresponding node.

5. The method according to claim 4, characterized in that The spectrum analysis of each target modal component of each node is performed to obtain multiple signal features of the corresponding node, including: Perform Fourier transform on different target modal components of each node to obtain the target spectra corresponding to different target modal components of the corresponding node; Based on the target frequency spectra corresponding to different target modal components of each node, multiple signal features of the corresponding node are determined.

6. The method according to claim 5, characterized in that Based on the target spectra corresponding to the different target modal components of each node, multiple signal features of the corresponding node are determined, including: Determine the frequency spectrum characteristics of each target modal component of the corresponding node based on different target modal components of each node; Solve the pre-built signal model based on the spectral characteristics of each target modal component to obtain the signal parameters of the corresponding target modal component; The signal parameters of different target modal components of each node are used to determine multiple signal features of the corresponding node.

7. A broadband oscillation signal positioning device based on a composite intelligent network drive, characterized in that: The device comprises: An acquisition module is configured to acquire a pre-built graph repair network and graph structure data to be repaired for the power system. The graph structure data to be repaired includes feature information of multiple nodes and edge feature information representing the association relationship between the nodes. The multiple nodes correspond one-to-one to multiple measurement points in the power system. The feature information of each node is used to represent the voltage and current phasor data of the corresponding measurement point, and the edge feature information is used to represent the electrical distance between the nodes. The graph repair network is constructed based on a target generative adversarial network and is used to perform anomaly repair on the voltage and current phase data of each node in the graph structure data to be repaired. A repair module, configured to input the graph structure data to be repaired into the graph repair network, so that the graph repair network outputs target feature information of each node after repair; The extraction module is used to extract the target feature information of each node and obtain multiple signal features of the corresponding node; The positioning module is used to locate the broadband oscillation signal in the power system based on multiple signal characteristics of each node.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the broadband oscillation signal positioning method based on composite intelligent network driving according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the broadband oscillation signal positioning method based on composite intelligent network driving according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the broadband oscillation signal positioning method based on composite intelligent network driving according to any one of claims 1 to 6.

Images