A power grid model data problem classification method and system based on a graph convolutional neural network

Abstract

The application discloses a kind of power grid model data problem classification method and system based on graph convolutional neural network belongs to neural network technical field.The power grid model data problem classification method includes: the equipment near the unqualified point of measurement is positioned, and subgraph is constructed;And filter out the parameter problem node in subgraph as the training sample of problem classification;Based on the training sample of problem classification, obtain the graph convolutional neural network problem classification model trained in advance;Based on the model trained in advance, the power grid model data is filtered, and different types of problem data are obtained, the model trained in advance is used to comprehensively filter the power grid model data, different types of problem data can be quickly and accurately obtained, and centralized and efficient positioning of the problem is realized.The application combines a large amount of state estimation operation data and operation and maintenance experience, based on graph neural network technology, can improve the data management efficiency of dispatch automation personnel, and reduce labor cost.

Description

Technical Field

[0001] This invention belongs to the field of neural network technology and relates to a method and system for classifying power grid model data problems based on graph convolutional neural networks. Background Technology

[0002] As the power grid expands, so too does its computational scale, increasing the difficulty of model data governance. Low-voltage side model data often suffers from numerous issues (primarily topology, measurements, and parameters), failing to meet the requirements of subsequent applications. Currently, the analysis of unqualified state estimation points relies mainly on the results of state estimation model data verification and bus balance verification. These results are scattered and poorly integrated, requiring further manual analysis and reasoning of the relevant model data. This results in low efficiency and accuracy in problem localization, and some problems (such as parameter issues) are difficult to locate quickly and accurately through manual analysis. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for classifying problems in power grid model data based on graph convolutional neural networks, aiming to solve the problems in the prior art where the analysis results of unqualified points in state estimation are relatively scattered, the integration is not high, and the efficiency and accuracy of problem localization are low.

[0004] To achieve the above objectives, the present invention employs the following technical solution:

[0005] This invention proposes a classification method for power grid model data problems based on graph convolutional neural networks, comprising:

[0006] Obtain the power grid model data to be filtered;

[0007] The power grid model data to be screened is input into a pre-trained graph convolutional neural network problem classification model to obtain data of different types of problems;

[0008] The pre-trained graph convolutional neural network problem classification model is obtained through the following steps:

[0009] A subgraph is constructed based on measurement non-compliance points in the power grid model data, and a graph convolutional neural network problem classification model is built.

[0010] Problem nodes in the power grid model data are labeled with their types, and parameter problem nodes in the subgraphs are selected as training samples for problem classification.

[0011] The graph convolutional neural network problem classification model is trained using training samples based on problem classification to obtain a pre-trained graph convolutional neural network problem classification model.

[0012] Preferably, the construction of the sub-graph based on measurement non-conformity points in the power grid model data specifically involves:

[0013] Collect power grid model data, filter out data problems in the power grid model data, and identify measurement non-conformities; aggregate the non-conformities, perform a topology search starting from the non-conformity equipment, set the number of layers n_st between adjacent substations, retain all equipment in adjacent n_st layer substations to form a subgraph; save the subgraph and equipment parameter measurement information to a file; simultaneously traverse all non-conformities until all non-conformities have been traversed, forming graph structure data for neural network training;

[0014] The power grid model data includes: steady-state parameters of power grid equipment, topological connections during equipment operation, and measurement data; the unqualified points in state estimation include: topological problems, measurement problems, and parameter problems in the power grid model data.

[0015] Preferably, the construction of the graph convolutional neural network problem classification model specifically involves:

[0016] The input dimension of the graph convolutional neural network (GNN) problem classification model is: node feature vector dimension × number of nodes; the output vector dimension is: node classification dimension × number of nodes. The model includes one graph embedding layer, three convolutional layers and pooling layers, three fully connected layers, an activation function, and a classifier. The graph embedding layer uses the feature representations of nodes and edges in the graph to capture the relationships and attributes in the graph structure data, converting the graph structure data into vectors of other dimensions, and using these vectors as input to subsequent convolutional layers. The convolutional layers are SAGEConv, stacking multiple SAGEConv graph convolutional modules to construct a multi-layer graph neural network that generates vector representations of nodes. Pooling layers reduce the number of nodes in the final fully connected layer. In the fully connected layer, each node is connected to all nodes in the previous layer, used to synthesize the features extracted earlier. The activation function is the ReLU function. The classifier is a SoftMax classifier, which is a model that predicts discrete output probabilities based on input features.

[0017] Preferably, the step of labeling the problematic nodes in the power grid model data and selecting the parameter problem nodes in the subgraph as training samples for problem classification specifically involves:

[0018] Problem node types include normal nodes, topology problem nodes, measurement problem nodes, and parameter problem nodes. Type labeling of problem nodes in the power grid model data includes manual labeling and automatic labeling. Manual labeling is based on on-site operation and maintenance experience, labeling problem nodes in the power grid model to generate training samples; manually labeled information is saved to a tag file. Automatic labeling involves searching by substation; if a substation and its surrounding substations have no non-compliant points, the model data at that substation is automatically modified, changing normal model data to abnormal data, generating a model data problem. Modifying model data includes deleting some equipment, changing equipment connection points (generating topology problems), modifying equipment measurements or measurement quality points (generating measurement problems), and modifying equipment parameters (generating parameter problems). The original parameters and modified problem parameters are recorded; the original parameters serve as the initial values ​​for subsequent physical information neural network training. State estimation calculations are performed on the modified cross-sections; if deviation problems occur near the modification point, that cross-section is used as a problem classification training sample, and sub-graph files and tag files are automatically saved.

[0019] Preferably, the training samples based on problem classification are used to train the graph convolutional neural network problem classification model to obtain a pre-trained graph convolutional neural network problem classification model, specifically as follows:

[0020] The Adam optimizer is used to optimize the parameters of the graph convolutional neural network (GNN) problem classification model; the CrossEntropyLoss function is used to calculate the deviation between the output of the GNN problem classification model and the label information, specifically:

[0021]

[0022] Where N represents the number of samples, y i This represents the true label of the i-th sample. This represents the predicted probability of the i-th sample;

[0023] The average loss is obtained by dividing the total deviation of the acquired test data by the total number of test data; training stops when the average loss is less than a preset threshold.

[0024] This invention proposes a classification system for power grid model data problems based on graph convolutional neural networks, comprising:

[0025] The data acquisition module is used to acquire the power grid model data to be filtered.

[0026] The data processing module is used to input the power grid model data to be screened into a pre-trained graph convolutional neural network problem classification model to obtain different types of problem data;

[0027] The pre-trained graph convolutional neural network problem classification model is obtained through the following steps:

[0028] A subgraph is constructed based on measurement non-compliance points in the power grid model data, and a graph convolutional neural network problem classification model is built.

[0029] Problem nodes in the power grid model data are labeled with their types, and parameter problem nodes in the subgraphs are selected as training samples for problem classification.

[0030] The graph convolutional neural network problem classification model is trained using training samples based on problem classification to obtain a pre-trained graph convolutional neural network problem classification model.

[0031] Preferably, the construction of the graph convolutional neural network problem classification model specifically involves:

[0032] The input dimension of the graph convolutional neural network (GNN) problem classification model is: node feature vector dimension × number of nodes; the output vector dimension is: node classification dimension × number of nodes. The model includes one graph embedding layer, three convolutional layers and pooling layers, three fully connected layers, an activation function, and a classifier. The graph embedding layer uses the feature representations of nodes and edges in the graph to capture the relationships and attributes in the graph structure data, converting the graph structure data into vectors of other dimensions, and using these vectors as input to subsequent convolutional layers. The convolutional layers are SAGEConv, stacking multiple SAGEConv graph convolutional modules to construct a multi-layer graph neural network that generates vector representations of nodes. Pooling layers reduce the number of nodes in the final fully connected layer. In the fully connected layer, each node is connected to all nodes in the previous layer, used to synthesize the features extracted earlier. The activation function is the ReLU function. The classifier is a SoftMax classifier, which is a model that predicts discrete output probabilities based on input features.

[0033] Preferably, the training samples based on problem classification are used to train the graph convolutional neural network problem classification model to obtain a pre-trained graph convolutional neural network problem classification model, specifically as follows:

[0034] The Adam optimizer is used to optimize the neural network parameters; the CrossEntropyLoss function is used to calculate the deviation between the neural network output and the label information, specifically:

[0035]

[0036] Where N represents the number of samples, y i This represents the true label of the i-th sample. This represents the predicted probability of the i-th sample;

[0037] The average loss is obtained by dividing the total deviation of the acquired test data by the total number of test data; training stops when the average loss is less than a preset threshold.

[0038] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a power grid model data problem classification method based on a graph convolutional neural network.

[0039] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a power grid model data problem classification method based on a graph convolutional neural network.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] This invention proposes a problem classification method for power grid model data based on graph convolutional neural networks. By locating problems in equipment near measurement non-compliance points, a subgraph is constructed for neural network training. This integrates previously scattered non-compliance points into one or more subgraphs, enabling the neural network to capture the potential relationships and contextual information between these points. Parameter problem nodes in the subgraphs are selected as training samples for problem classification. Accurate type labeling helps provide a clear learning objective for the neural network, improving model accuracy. Based on the problem classification training samples, a pre-trained graph convolutional neural network problem classification model is obtained. Optimizing the model with a large number of training samples allows for more accurate identification and classification of problems. The pre-trained graph convolutional neural network problem classification model is used to filter power grid model data, obtaining different types of problem data. This comprehensive filtering of power grid model data using the pre-trained model enables rapid and accurate acquisition of different types of problem data, achieving centralized and efficient problem location. Therefore, the power grid model data problem classification method proposed in this invention can solve the problems existing in the prior art.

[0042] This invention proposes a power grid model data problem classification system based on graph convolutional neural networks. By dividing the system into a data acquisition module and a data processing module, it acquires different types of problem data and ultimately achieves power grid model data problem classification. The modular approach ensures that each module is independent, facilitating unified management of all modules. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart illustrating the classification method for power grid model data problems based on graph convolutional neural networks according to the present invention.

[0045] Figure 2 This is a schematic diagram of the model data processing flow of the present invention.

[0046] Figure 3 This is a single-line diagram of the 5-node system of the present invention.

[0047] Figure 4 This is a schematic diagram of the 5-node system structure of the present invention.

[0048] Figure 5 This is a schematic diagram of the 5-node features and node labels of the present invention.

[0049] Figure 6 This is a diagram of the power grid model data problem classification system based on graph convolutional neural networks of the present invention.

[0050] Figure 7 This is a schematic diagram of the structure of an electronic device according to the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0052] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0053] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0054] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0055] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0056] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0057] The present invention will now be described in further detail with reference to the accompanying drawings:

[0058] This invention proposes a classification method for power grid model data problems based on graph convolutional neural networks, such as... Figure 1 As shown, it includes:

[0059] S1. Obtain the power grid model data to be filtered;

[0060] S2. Input the power grid model data to be screened into a pre-trained graph convolutional neural network problem classification model to obtain data of different types of problems;

[0061] The pre-trained graph convolutional neural network problem classification model is obtained through the following steps:

[0062] S201. Construct a subgraph based on measurement non-conformity points in power grid model data, and build a graph convolutional neural network problem classification model.

[0063] The construction of the sub-graph based on measurement non-compliance points in the power grid model data is specifically as follows:

[0064] Collect power grid model data, filter out data problems in the power grid model data, and identify measurement non-conformities; aggregate the non-conformities, perform a topology search starting from the non-conformity equipment, set the number of layers n_st between adjacent substations, retain all equipment in adjacent n_st layer substations to form a subgraph; save the subgraph and equipment parameter measurement information to a file; simultaneously traverse all non-conformities until all non-conformities have been traversed, forming graph structure data for neural network training;

[0065] The power grid model data includes: steady-state parameters of power grid equipment, topological connections during equipment operation, and measurement data; the unqualified points in state estimation include: topological problems, measurement problems, and parameter problems in the power grid model data.

[0066] The construction of the graph convolutional neural network problem classification model is specifically as follows:

[0067] The input dimension of the graph convolutional neural network (GNN) problem classification model is: node feature vector dimension × number of nodes; the output vector dimension is: node classification dimension × number of nodes. The model includes one graph embedding layer, three convolutional layers and pooling layers, three fully connected layers, an activation function, and a classifier. The graph embedding layer uses the feature representations of nodes and edges in the graph to capture the relationships and attributes in the graph structure data, converting the graph structure data into vectors of other dimensions, and using these vectors as input to subsequent convolutional layers. The convolutional layers are SAGEConv, stacking multiple SAGEConv graph convolutional modules to construct a multi-layer graph neural network that generates vector representations of nodes. Pooling layers reduce the number of nodes in the final fully connected layer. In the fully connected layer, each node is connected to all nodes in the previous layer, used to synthesize the features extracted earlier. The activation function is the ReLU function. The classifier is a SoftMax classifier, which is a model that predicts discrete output probabilities based on input features.

[0068] S202. Type labeling of problem nodes in the power grid model data, and selection of parameter problem nodes in the subgraph as training samples for problem classification.

[0069] The process of labeling problem nodes in the power grid model data and selecting parameter problem nodes in the subgraph as training samples for problem classification involves:

[0070] Problem node types include normal nodes, topology problem nodes, measurement problem nodes, and parameter problem nodes. Type labeling of problem nodes in the power grid model data includes manual labeling and automatic labeling. Manual labeling is based on on-site operation and maintenance experience, labeling problem nodes in the power grid model to generate training samples; manually labeled information is saved to a tag file. Automatic labeling involves searching by substation; if a substation and its surrounding substations have no non-compliant points, the model data at that substation is automatically modified, changing normal model data to abnormal data, generating a model data problem. Modifying model data includes deleting some equipment, changing equipment connection points (generating topology problems), modifying equipment measurements or measurement quality points (generating measurement problems), and modifying equipment parameters (generating parameter problems). The original parameters and modified problem parameters are recorded; the original parameters serve as the initial values ​​for subsequent physical information neural network training. State estimation calculations are performed on the modified cross-sections; if deviation problems occur near the modification point, that cross-section is used as a problem classification training sample, and sub-graph files and tag files are automatically saved.

[0071] S203. Train the graph convolutional neural network problem classification model based on the training samples of problem classification to obtain the pre-trained graph convolutional neural network problem classification model;

[0072] The training samples based on question classification are used to train the graph convolutional neural network question classification model to obtain a pre-trained graph convolutional neural network question classification model, specifically as follows:

[0073] The Adam optimizer is used to optimize the parameters of the graph convolutional neural network (GNN) problem classification model; the CrossEntropyLoss function is used to calculate the deviation between the output of the GNN problem classification model and the label information, specifically:

[0074]

[0075] Where N represents the number of samples, y i This represents the true label of the i-th sample. This represents the predicted probability of the i-th sample;

[0076] The average loss is obtained by dividing the total deviation of the acquired test data by the total number of test data; training stops when the average loss is less than a preset threshold.

[0077] The following describes in detail a method for classifying power grid model data problems based on graph convolutional neural networks. This includes: diagnosing problems near equipment located at points where state estimation measurements are substandard (points where measured values ​​deviate significantly from estimated values). This requires local network topology and parameter information near the substandard points. A subgraph is formed by searching outwards from the substandard points at a given maximum depth, creating a subgraph for neural network training. Based on field operation and maintenance experience and automatic problem generation, problem nodes in the power grid model are labeled with their types, generating training samples. A graph neural network model for intelligent classification of power grid model data problems is established. Supervised learning is used to train the graph neural network model, classifying nodes into four categories: normal nodes, topology problem nodes, measurement problem nodes, and parameter problem nodes.

[0078] See Figure 2 , Figure 3 , Figure 4 and Figure 5 State estimation only diagnoses problems for devices near measurement failure points (points where the measured value deviates significantly from the estimated value). Therefore, it only requires the local network topology, parameters, and measurement information near the failure point. Centered on the state estimation failure point, it searches outward at a given depth to form graph structure data for neural network training.

[0079] Node characteristics include device ID, voltage level ID, device parameters, device measurements, and measurement estimates. Among them, device parameters include resistance, reactance, susceptance, tap type, rated voltage, and rated capacity, while device measurements include active power, reactive power, voltage, and tap position.

[0080] Node labels are divided into four categories: normal nodes, topology problem nodes, measurement problem nodes, and parameter problem nodes.

[0081] Problem node labeling includes both manual and automatic labeling. Manual labeling, based on on-site operation and maintenance experience, involves labeling problematic nodes in the power grid model to generate training samples. Manually labeled information is saved to a tag file, formatted as shown in Table 1.

[0082] Table 1

[0083] Node identifier Problem Type Problem Node 1 Number Parameter issues Problem Node 2 Number Measurement problems … …

[0084] Automatic labeling is performed as follows: If a plant and its surrounding plants have no non-conforming points, the model data at that plant is automatically modified, changing normal model data to abnormal data, thus generating model data problems. This modification includes deleting some equipment, changing equipment connection points (generating topology problems), modifying equipment measurements or measurement quality points (generating measurement problems), and modifying equipment parameters (generating parameter problems). The original parameters and the modified problem parameters are recorded, with the original parameters serving as initial values ​​for subsequent physical information neural network training. State estimation calculations are performed on the modified cross-sections. If a large deviation occurs near the modification point, that cross-section is used as a problem classification training sample, and sub-map files and label files are automatically saved. This automatic sample generation addresses the problem of insufficient actual samples.

[0085] The following is a graph convolutional neural network classification model suitable for the node classification task of graphs:

[0086] The node feature vector dimension is multiplied by the number of nodes; the output vector dimension is multiplied by the node classification dimension. The graph convolutional neural network problem classification model includes a graph embedding layer, three convolutional layers and pooling layers, three fully connected layers, an activation function, and a classifier. The graph embedding layer uses the feature representations of nodes and edges on the graph to capture the relationships and attributes in the graph structure data, converting the graph structure data into vectors of other dimensions, and using these vectors as inputs to subsequent convolutional layers. The convolutional layer is SAGEConv, which stacks multiple SAGEConv graph convolutional modules to construct a multi-layer graph neural network to generate vector representations of nodes. The pooling layer reduces the number of nodes in the final fully connected layer. In the fully connected layer, each node is connected to all nodes in the previous layer to synthesize the features extracted earlier. The activation function is the ReLU function. The classifier is a SoftMax classifier, which is a model that predicts discrete output probabilities based on input features.

[0087] SAGEConv is a convolutional operation used in graph neural networks (GNNs). It originates from an algorithm called GraphSAGE, which was designed for efficient node feature learning on graph data. The core idea of ​​SAGEConv is to update the feature representation of each node by aggregating information from its neighbors.

[0088] Multiple SAGEConv graph convolutional modules are stacked to construct a multi-layer graph neural network (more layers mean more neighbor information can be accessed) to generate vector representations of nodes. Based on experience, a particular failure is highly correlated with the model parameters of devices within a radius of less than three branches; therefore, three graph convolutions are used here.

[0089] A fully connected layer is a layer in which every node is connected to all nodes in the previous layer, used to synthesize the features extracted earlier.

[0090] To train deep neural networks, an activation function (ReLU) is needed. It looks and behaves like a linear function, but is actually a non-linear function, allowing the learning of complex relationships in the data. This function must also provide more sensitive activation and input, avoiding saturation.

[0091] The SoftMax classifier is a model that predicts discrete output probabilities based on input features and is suitable for multi-class prediction problems.

[0092] Based on the training samples used for problem classification, a graph convolutional neural network (GNN) problem classification model is trained to obtain a pre-trained GNN problem classification model. The Adam optimizer is used to optimize the parameters of the GNN problem classification model; the CrossEntropyLoss function is used to calculate the deviation between the output of the GNN problem classification model and the label information, specifically:

[0093]

[0094] Where N represents the number of samples, y i This represents the true label of the i-th sample. This represents the predicted probability of the i-th sample;

[0095] To improve performance, the training set is processed in batches, with each batch containing 64 sets of test data. Each set of data is simultaneously input into the neural network for parallel computation. The output results and labels are used to calculate the loss, and finally the average loss of the 64 sets of data is calculated. Training stops when the average loss value is less than the threshold value (which can be equal to 10 based on experience).

[0096] Example 2

[0097] See Figure 6 This invention proposes a classification system for power grid model data problems based on graph convolutional neural networks, comprising:

[0098] The data acquisition module is used to acquire the power grid model data to be filtered.

[0099] The data processing module is used to input the power grid model data to be screened into a pre-trained graph convolutional neural network problem classification model to obtain different types of problem data;

[0100] The pre-trained graph convolutional neural network problem classification model is obtained through the following steps:

[0101] A subgraph is constructed based on measurement non-compliance points in the power grid model data, and a graph convolutional neural network problem classification model is built.

[0102] The construction of the sub-graph based on measurement non-compliance points in the power grid model data is specifically as follows:

[0103] Collect power grid model data, filter out data problems in the power grid model data, and identify measurement non-conformities; aggregate the non-conformities, perform a topology search starting from the non-conformity equipment, set the number of layers n_st between adjacent substations, retain all equipment in adjacent n_st layer substations to form a subgraph; save the subgraph and equipment parameter measurement information to a file; simultaneously traverse all non-conformities until all non-conformities have been traversed, forming graph structure data for neural network training;

[0104] The power grid model data includes: steady-state parameters of power grid equipment, topological connections during equipment operation, and measurement data; the unqualified points in state estimation include: topological problems, measurement problems, and parameter problems in the power grid model data.

[0105] The construction of the graph convolutional neural network problem classification model is specifically as follows:

[0106] The input dimension of the graph convolutional neural network (GNN) problem classification model is: node feature vector dimension × number of nodes; the output vector dimension is: node classification dimension × number of nodes. The model includes one graph embedding layer, three convolutional layers and pooling layers, three fully connected layers, an activation function, and a classifier. The graph embedding layer uses the feature representations of nodes and edges in the graph to capture the relationships and attributes in the graph structure data, converting the graph structure data into vectors of other dimensions, and using these vectors as input to subsequent convolutional layers. The convolutional layers are SAGEConv, stacking multiple SAGEConv graph convolutional modules to construct a multi-layer graph neural network that generates vector representations of nodes. Pooling layers reduce the number of nodes in the final fully connected layer. In the fully connected layer, each node is connected to all nodes in the previous layer, used to synthesize the features extracted earlier. The activation function is the ReLU function. The classifier is a SoftMax classifier, which is a model that predicts discrete output probabilities based on input features.

[0107] Problem nodes in the power grid model data are labeled with their types, and parameter problem nodes in the subgraphs are selected as training samples for problem classification.

[0108] The process of labeling problem nodes in the power grid model data and selecting parameter problem nodes in the subgraph as training samples for problem classification involves:

[0109] Problem node types include normal nodes, topology problem nodes, measurement problem nodes, and parameter problem nodes. Type labeling of problem nodes in the power grid model data includes manual labeling and automatic labeling. Manual labeling is based on on-site operation and maintenance experience, labeling problem nodes in the power grid model to generate training samples; manually labeled information is saved to a tag file. Automatic labeling involves searching by substation; if a substation and its surrounding substations have no non-compliant points, the model data at that substation is automatically modified, changing normal model data to abnormal data, generating a model data problem. Modifying model data includes deleting some equipment, changing equipment connection points (generating topology problems), modifying equipment measurements or measurement quality points (generating measurement problems), and modifying equipment parameters (generating parameter problems). The original parameters and modified problem parameters are recorded; the original parameters serve as the initial values ​​for subsequent physical information neural network training. State estimation calculations are performed on the modified cross-sections; if deviation problems occur near the modification point, that cross-section is used as a problem classification training sample, and sub-graph files and tag files are automatically saved.

[0110] The graph convolutional neural network problem classification model is trained based on the training samples of problem classification to obtain a pre-trained graph convolutional neural network problem classification model.

[0111] The training samples based on question classification are used to train the graph convolutional neural network question classification model to obtain a pre-trained graph convolutional neural network question classification model, specifically as follows:

[0112] The Adam optimizer is used to optimize the parameters of the graph convolutional neural network (GNN) problem classification model; the CrossEntropyLoss function is used to calculate the deviation between the output of the GNN problem classification model and the label information, specifically:

[0113]

[0114] Where N represents the number of samples, y i This represents the true label of the i-th sample. This represents the predicted probability of the i-th sample;

[0115] The average loss is obtained by dividing the total deviation of the acquired test data by the total number of test data; training stops when the average loss is less than a preset threshold.

[0116] Example 3

[0117] Please see Figure 7As shown, the present invention also provides an electronic device 100 for a power grid model data problem classification method based on graph convolutional neural networks; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0118] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the power grid model data problem classification method based on graph convolutional neural networks described in Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0119] The at least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor. The processor 102 is the control center of the electronic device 100, connecting various parts of the electronic device 100 via various interfaces and lines.

[0120] The memory 101 in the electronic device 100 stores multiple instructions to implement a power grid model data problem classification method based on graph convolutional neural networks, and the processor 102 can execute the multiple instructions to achieve the following:

[0121] Obtain the power grid model data to be filtered;

[0122] The power grid model data to be screened is input into a pre-trained graph convolutional neural network problem classification model to obtain data of different types of problems;

[0123] The pre-trained graph convolutional neural network problem classification model is obtained through the following steps:

[0124] A subgraph is constructed based on measurement non-compliance points in the power grid model data, and a graph convolutional neural network problem classification model is built.

[0125] Problem nodes in the power grid model data are labeled with their types, and parameter problem nodes in the subgraphs are selected as training samples for problem classification.

[0126] The graph convolutional neural network problem classification model is trained using training samples based on problem classification to obtain a pre-trained graph convolutional neural network problem classification model.

[0127] Example 4

[0128] If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0129] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0130] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A classification method for power grid model data problems based on graph convolutional neural networks, characterized in that, include: Obtain the power grid model data to be filtered; The power grid model data to be screened is input into a pre-trained graph convolutional neural network problem classification model to obtain data on different types of problems; The pre-trained graph convolutional neural network problem classification model is obtained through the following steps: A subgraph is constructed based on measurement non-compliance points in the power grid model data, and a graph convolutional neural network problem classification model is built. Problem nodes in the power grid model data are labeled with their types, and parameter problem nodes in the subgraphs are selected as training samples for problem classification. The graph convolutional neural network problem classification model is trained based on the training samples of problem classification to obtain a pre-trained graph convolutional neural network problem classification model. The construction of the sub-graph based on measurement non-compliance points in the power grid model data is specifically as follows: Data from the power grid model is collected, and data problems are filtered out to identify measurement non-compliance points. These non-compliance points are then aggregated, and a topology search is performed starting from the non-compliance equipment, setting the layer number for adjacent substations. Preserve adjacent All equipment in the substation is used to form a subgraph; the subgraph and equipment parameter measurement information are saved to a file; at the same time, all non-conforming points are traversed until all non-conforming points have been traversed, forming graph structure data for neural network training; wherein, the power grid model data includes: steady-state parameters of power grid equipment, topological connection relationships of equipment during operation, and measurement data; state estimation non-conforming points include: topology problems, measurement problems, and parameter problems of power grid model data; The graph convolutional neural network problem classification model includes a graph embedding layer, three convolutional layers and pooling layers, three fully connected layers, as well as an activation function and a classifier. The convolutional layers are SAGEConv, which stacks multiple SAGEConv graph convolutional modules to construct a multi-layer graph neural network that generates vector representations of nodes. The pooling layers reduce the number of nodes in the final fully connected layer. Node characteristics include device ID, voltage level ID, device parameters, device measurements, and measurement estimates. Among them, device parameters include resistance, reactance, susceptance, tap type, rated voltage, and rated capacity. Device measurements include active power, reactive power, voltage, and tap position.

2. The classification method for power grid model data problems based on graph convolutional neural networks according to claim 1, characterized in that, The construction of the graph convolutional neural network problem classification model is specifically as follows: The input dimension of the graph convolutional neural network problem classification model is: node feature vector dimension × number of nodes; the output vector dimension is: node classification dimension × number of nodes; the graph embedding layer uses the feature representation of nodes and edges on the graph to capture the relationships and attributes in the graph structure data, converts the graph structure data into vectors of other dimensions, and uses these vectors as the input of the subsequent convolutional layers; the fully connected layer connects each node to all nodes in the previous layer, and is used to synthesize the features extracted earlier; The activation function is the ReLU function; the classifier is the SoftMax classifier, which is a model that predicts the discrete output probability based on the input features.

3. The classification method for power grid model data problems based on graph convolutional neural networks according to claim 1, characterized in that, The process of labeling problem nodes in the power grid model data and selecting parameter problem nodes in the subgraph as training samples for problem classification involves: Problem node types include normal nodes, topology problem nodes, measurement problem nodes, and parameter problem nodes. Type labeling of problem nodes in the power grid model data includes manual labeling and automatic labeling. Manual labeling is based on on-site operation and maintenance experience, labeling problem nodes in the power grid model to generate training samples; manually labeled information is saved to a tag file. Automatic labeling involves searching by substation; if a substation and its surrounding substations have no non-compliant points, the model data at that substation is automatically modified, changing normal model data to abnormal data, generating a model data problem. Modifying model data includes deleting some equipment, changing equipment connection points (generating topology problems), modifying equipment measurements or measurement quality points (generating measurement problems), and modifying equipment parameters (generating parameter problems). The original parameters and modified problem parameters are recorded; the original parameters serve as the initial values ​​for subsequent physical information neural network training. State estimation calculations are performed on the modified cross-sections; if deviation problems occur near the modification point, that cross-section is used as a problem classification training sample, and sub-graph files and tag files are automatically saved.

4. The classification method for power grid model data problems based on graph convolutional neural networks according to claim 1, characterized in that, The training samples based on problem classification are used to train the graph convolutional neural network problem classification model to obtain a pre-trained graph convolutional neural network problem classification model, specifically as follows: The Adam optimizer is used to optimize the parameters of the graph convolutional neural network (GNN) problem classification model; the CrossEntropyLoss function is used to calculate the deviation between the output of the GNN problem classification model and the label information, specifically: Where N represents the number of samples, Indicates the first The true label of each sample Indicates the first The predicted probability of each sample; The average loss is obtained by dividing the total deviation of the acquired test data by the total number of test data; training stops when the average loss is less than a preset threshold.

5. A classification system for power grid model data problems based on graph convolutional neural networks, characterized in that, The power grid model data classification method based on graph convolutional neural networks as described in claim 1 includes: The data acquisition module is used to acquire the power grid model data to be filtered. The data processing module is used to input the power grid model data to be screened into a pre-trained graph convolutional neural network problem classification model to obtain different types of problem data; The pre-trained graph convolutional neural network problem classification model is obtained through the following steps: A subgraph is constructed based on measurement non-compliance points in the power grid model data, and a graph convolutional neural network problem classification model is built. Problem nodes in the power grid model data are labeled with their types, and parameter problem nodes in the subgraphs are selected as training samples for problem classification. The graph convolutional neural network problem classification model is trained using training samples based on problem classification to obtain a pre-trained graph convolutional neural network problem classification model.

6. The power grid model data classification system based on graph convolutional neural networks according to claim 5, characterized in that, The construction of the graph convolutional neural network problem classification model is specifically as follows: The input dimension of the graph convolutional neural network problem classification model is: node feature vector dimension × number of nodes; the output vector dimension is: node classification dimension × number of nodes; the graph embedding layer uses the feature representation of nodes and edges on the graph to capture the relationships and attributes in the graph structure data, converts the graph structure data into vectors of other dimensions, and uses these vectors as the input of the subsequent convolutional layers; the fully connected layer connects each node to all nodes in the previous layer, and is used to synthesize the features extracted earlier; The activation function is the ReLU function; the classifier is the SoftMax classifier, which is a model that predicts the discrete output probability based on the input features.

7. The power grid model data classification system based on graph convolutional neural networks according to claim 5, characterized in that, The training samples based on problem classification are used to train the graph convolutional neural network problem classification model to obtain a pre-trained graph convolutional neural network problem classification model, specifically as follows: The Adam optimizer is used to optimize the neural network parameters; the CrossEntropyLoss function is used to calculate the deviation between the neural network output and the label information, specifically: Where N represents the number of samples, Indicates the first The true label of each sample Indicates the first The predicted probability of each sample; The average loss is obtained by dividing the total deviation of the acquired test data by the total number of test data; training stops when the average loss is less than a preset threshold.

8. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the power grid model data problem classification method based on graph convolutional neural networks as described in any one of claims 1 to 4.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the power grid model data problem classification method based on graph convolutional neural networks as described in any one of claims 1 to 4.

Images