Power distribution network switch state identification method, system and device based on graph neural network, and medium
By constructing a power distribution network topology and node features, and using a graph neural network model to identify switch states, the problems of high cost and susceptibility to interference in traditional methods are solved, achieving high-precision and low-cost switch state identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-21
- Publication Date
- 2026-04-07
AI Technical Summary
In existing technologies, the identification of switch status in power distribution networks relies on expensive communication devices and is susceptible to channel interference, leading to topology errors, and the accuracy and reliability of sensors are difficult to guarantee.
A graph neural network-based approach is used to construct a power distribution network topology. The graph neural network model is trained using node electrical measurement data, and the switch status is identified through end-to-end learning, reducing the dependence on hardware.
It achieves high-precision and robust switch state recognition, reduces costs, improves recognition accuracy, has tolerance to noise and incomplete data, and reduces reliance on sensors.
Smart Images
Figure CN121808475A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of power automation and artificial intelligence technology, and in particular to a method, system, device and medium for identifying the switch status of a power distribution network based on graph neural networks. Background Technology
[0002] As the final link in the power system facing users, the accurate sensing of the distribution network's operational status is crucial to ensuring power supply reliability and power quality. The status ("open" or "closed") of switches (such as circuit breakers, sectionalizing switches, and tie switches) directly determines the network topology, thus affecting power flow distribution and system security. Various sensors in the distribution network need to effectively sense the system's operational status under different scenarios, placing high demands on sensor hardware in terms of accuracy and reliability. However, under actual field operating conditions, sensors cannot always maintain reliable operation due to factors such as equipment age and environmental conditions. Therefore, reducing reliance on sensor sensing and improving the robustness of the overall system sensing has become a necessary condition for ensuring the efficient and reliable operation of the distribution network.
[0003] Traditional switch status identification methods primarily rely on communication signals uploaded by remote terminal units (RTUs) or feeder terminal units (FTUs) installed on the switches. However, this method has significant drawbacks. First, configuring communication equipment for all switches is costly. Second, communication signals may be lost or erroneous due to channel interference or equipment failure, leading to a discrepancy between the topology acquired by the master system and the actual topology—a phenomenon known as "topology error." Traditional discrimination methods heavily depend on sensor accuracy and have high computational complexity. Summary of the Invention
[0004] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a method for identifying the switch status of a power distribution network based on a graph neural network to solve the above problems.
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for identifying the switch status of a power distribution network based on a graph neural network, comprising: constructing a power distribution network topology diagram structure according to the physical connection relationship of the power distribution network; Acquire electrical measurement data of nodes and use it as node characteristics; A graph neural network model is constructed. The distribution network topology and node features are input into the pre-trained graph neural network model, and the state probabilities of each edge in the distribution network topology are output to identify the switch state.
[0006] As a preferred embodiment of the distribution network switch state recognition method based on graph neural networks described in this invention, the construction of the distribution network topology graph structure includes: Based on the physical connection relationship of the power distribution network, construct the corresponding topology graph structure G=(V, E), where V represents the set of nodes, each node represents an electrical connection point or device, and E represents the set of edges, each edge represents a switch or connection line.
[0007] As a preferred embodiment of the power distribution network switch status identification method based on graph neural networks described in this invention, the acquisition of electrical measurement data of nodes includes: Obtain real-time or historical electrical measurement data of each node in the power distribution network topology and construct a node feature matrix X; The beneficial effects of this preferred scheme are that it can reflect the actual operating state of the network at a specific moment, which is the core basis for the model to make judgments, supports end-to-end learning, and enhances robustness when combined with graph structure.
[0008] As a preferred embodiment of the distribution network switch state recognition method based on graph neural networks described in this invention, the construction of the graph neural network model includes: The graph neural network model includes an encoder and a decoder; The encoder learns and outputs the node embedding representation Z based on the input distribution network topology structure G and node feature matrix X; The decoder calculates the existence probability of each edge in the power distribution network topology based on the embedded representation Z to determine the switch state; The advantages of this preferred solution are that it captures complex spatial relationships and electrical couplings, can efficiently simulate the propagation and mutual influence of electrical quantities in the distribution network, and can achieve high-precision and robust identification by integrating graph structure and node characteristics, thereby reducing the dependence on hardware.
[0009] As a preferred embodiment of the power distribution network switch state recognition method based on graph neural networks described in this invention, the method for determining switch states includes: The graph neural network model is trained using a historical dataset with known switch state labels to optimize model parameters; The real-time topology and node features of the distribution network to be identified are input into the trained graph neural network model, and the status identification results of each switch are output.
[0010] As a preferred embodiment of the distribution network switch state recognition method based on graph neural network described in this invention, the training of the graph neural network model includes: The graph neural network model training uses the binary cross-entropy loss function as the optimization objective, and the loss function is expressed as: in,E Represents the set of edges that represent actual closed switches. This represents the set of edges where the switch is open, obtained through negative sampling. Representing an edge The true status label, This represents the probability that the edge is in a closed state, as predicted by the model.
[0011] As a preferred embodiment of the distribution network switch state recognition method based on graph neural networks described in this invention, the distribution network topology diagram structure includes: The nodes include bus nodes, load nodes, and distributed power generation nodes, and the edges represent switches, lines, or transformers.
[0012] Secondly, the present invention provides a power distribution network switch status identification system based on graph neural networks, comprising: The building module is used to construct the power distribution network topology structure based on the physical connection relationship of the power distribution network; The acquisition module is used to acquire electrical measurement data of the nodes and use it as node characteristics; The identification module is used to construct a graph neural network model. It inputs the distribution network topology structure and node features into the pre-trained graph neural network model and outputs the state probability of each edge in the distribution network topology to identify the switch state.
[0013] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the method for identifying the switch status of a power distribution network based on a graph neural network.
[0014] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method for identifying the switch status of a power distribution network based on a graph neural network.
[0015] Compared with existing technologies, the advantages of this invention are as follows: This invention can reduce the dependence on hardware, eliminating the need to equip all switches with expensive communication devices, and can achieve state recognition using only existing measurement data, thus saving costs; it has strong robustness, as the graph neural network can compensate for missing or abnormal node measurements through neighbor information, and has a natural tolerance to incomplete data and noise; it has high recognition accuracy, as the model can capture the complex electrical coupling relationships and nonlinear characteristics between nodes in the distribution network, significantly improving recognition accuracy compared to traditional methods, achieving end-to-end learning, eliminating the need for manually designing complex feature extraction rules, and the model can automatically learn and discriminate features from the data, making it highly versatile. Attached Figure Description
[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a schematic diagram of the overall process of a power distribution network switch status identification method based on graph neural networks according to an embodiment of the present invention.
[0018] Figure 2 This is a diagram illustrating the switch state recognition process based on a graph neural network, as described in an embodiment of the present invention. Detailed Implementation
[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0020] Reference Figures 1-2 As an embodiment of the present invention, a method for identifying the switch status of a power distribution network based on a graph neural network is provided, comprising: S101, Construct the power distribution network topology structure based on the physical connection relationship of the power distribution network; S102, acquire the electrical measurement data of the node and use it as node characteristics; S103. Construct a graph neural network model. Input the distribution network topology diagram structure and node features into the pre-trained graph neural network model, and output the state probability of each edge in the distribution network topology diagram to identify the switch state.
[0021] It should be noted that the switch state recognition process based on graph neural networks in this invention is as follows: Figure 1 , Figure 2 As shown, the current data feature distribution and topology of the current power distribution network are first collected based on the input. The graph neural network performs deep discrimination on the collected data to obtain the switch status.
[0022] In a preferred embodiment, constructing the distribution network topology includes: Based on the physical connection relationship of the power distribution network, construct the corresponding topology graph structure G=(V, E), where V represents the set of nodes, each node represents an electrical connection point or device, and E represents the set of edges, each edge represents a switch or connection line.
[0023] In a preferred embodiment, nodes include electrical equipment such as bus nodes, load nodes, and distributed power supply nodes, and edges represent switches, lines, or transformers.
[0024] Specifically, the busbars, loads, and distributed power sources in the distribution network are abstracted as graph nodes, and the switches (such as CB1, S1, S2, S3) and connecting lines are abstracted as graph edges.
[0025] In a preferred embodiment, acquiring the electrical measurement data of the node includes: Obtain real-time or historical electrical measurement data of each node in the power distribution network topology and construct a node feature matrix X.
[0026] Specifically, the characteristics of a node include one or more of the following: voltage amplitude, voltage phase angle, active power, reactive power, and current amplitude. Real-time measurement data of each node (bus) is obtained from the distribution automation system or advanced measurement system, such as voltage amplitude V, phase angle θ, injected active power P, and reactive power Q. The data is preprocessed to form a node feature matrix X, where each row of the node feature matrix corresponds to a node and each column corresponds to a feature. If some features are unavailable (such as phase angle), they will be set to 0 by default or interpolated to ensure the integrity of the feature matrix.
[0027] In an alternative implementation, the node feature matrix can also be constructed by generating node features using the topological relationships of the power distribution network, and calculating the graph theory features of each node, such as degree centrality, proximity centrality, eigenvector centrality, etc., which can capture the structural importance of nodes in the network, thereby assisting in the identification of switch states, reflecting the structural role of nodes, and enhancing the model's sensitivity to topological changes. When constructing node features, in addition to electrical measurements, the degree of the node (i.e., the number of connected edges) can be added as an additional feature.
[0028] In another alternative implementation, the node feature matrix can also be constructed using time series statistical features. Time series statistical features are extracted from historical electrical measurement data as node features, such as calculating the mean, variance, maximum, minimum, and trend coefficient of parameters such as voltage and power. Historical data can be analyzed through a sliding window, which can capture the dynamic behavior patterns of nodes and is robust to noise and transient anomalies. Specifically, for each node, the mean and standard deviation of the voltage amplitude time series over a period of time are calculated and used as the node feature input model.
[0029] In a preferred embodiment, constructing a graph neural network model includes: Graph neural network models include encoders and decoders; The encoder learns and outputs the node embedding representation Z based on the input distribution network topology structure G and node feature matrix X; The decoder calculates the existence probability of each edge in the power distribution network topology based on the embedded representation Z to determine the switch state.
[0030] Specifically, the graph neural network model aggregates neighbor node information through a message passing mechanism and learns the deep representation of nodes in the topology. Finally, the decoder outputs the state probabilities of each edge in the graph to achieve accurate discrimination of the "on" or "off" state of the switch.
[0031] Graph neural networks use graph attention networks as their basic units; the encoder aggregates and updates node features through multi-layer message passing, and the message passing process is represented as follows: in, Indicates the first Layer nodes Feature representation, Represents a node The set of neighboring nodes, It is an aggregate function. For update functions.
[0032] The decoder uses either an inner product decoder or a neural network decoder, by computing node pairs. The correlation between the embedding representations is used to obtain the corresponding edges. The state probability is calculated using the following formula: in, and Let i and j be the embedding representations of nodes i and j, respectively. σ For the sigmoid function, For the edge The predicted state, express The transpose of .
[0033] In a preferred embodiment, the graph neural network model training uses the binary cross-entropy loss function as the optimization objective, and the loss function is expressed as: in, E Represents the set of edges that represent actual closed switches. This represents the set of edges where the switch is open, obtained through negative sampling. Representing an edge The true status label, This represents the probability that the edge is in a closed state, as predicted by the model.
[0034] In this embodiment, based on the node feature structure, the default unconfigured structural feature is set to 0. A two-layer GCN structure is used, and the loss function is a binary cross-entropy function. The collected data is input, and the switch state is used as a label for identification to obtain the switch state, which can be used for business scenarios such as network reconstruction and fault diagnosis.
[0035] In an alternative implementation, the graph neural network model can also be a graph sampling and aggregation (GraphSAGE) model, which generates node embeddings by sampling and aggregating the features of neighboring nodes. The node representation is iteratively updated by sampling a fixed number of neighbors and using a learnable aggregation function. For each node, the encoder samples, aggregates, and updates at each layer. A fixed number of nodes are randomly sampled from the node's neighbors. The aggregation function is used to aggregate the feature vectors of the sampled neighboring nodes in the previous layer into a single vector. The current features of the node are concatenated or combined with the aggregated neighboring features. A new feature vector for the node is obtained through a fully connected layer. The decoder can use a decoder such as an inner product to predict edge states.
[0036] In another alternative implementation, the graph neural network model can also be a graph attention network model, which introduces an attention mechanism. Through a learnable mechanism, each node adaptively assigns different weights (attention coefficients) to its different neighbor nodes, making the model pay more attention to the neighbors that are more important for judging the current node's state. The encoder uses multiple GAT layers. In each layer, the update of node features depends on the weighted sum of the features of its neighbor nodes. The weights are calculated from the node features. The decoder uses an inner product decoder or similar method to predict edge states.
[0037] It should be noted that this invention reduces reliance on hardware, eliminating the need to equip all switches with expensive communication devices, and achieving state recognition using only existing measurement data, thus saving costs. It also exhibits strong robustness; the graph neural network can compensate for missing or abnormal node measurements using neighbor information, and has a natural tolerance for incomplete data and noise. Furthermore, it boasts high recognition accuracy; the model can capture the complex electrical coupling relationships and nonlinear characteristics between nodes in the distribution network, significantly improving recognition accuracy compared to traditional methods. It achieves end-to-end learning, eliminating the need for manually designing complex feature extraction rules; the model can automatically learn and discriminate features from the data, demonstrating strong versatility.
[0038] The above is an illustrative scheme of a distribution network switch status identification method based on graph neural networks according to this embodiment. It should be noted that the technical solution of this distribution network switch status identification system based on graph neural networks belongs to the same concept as the technical solution of the distribution network switch status identification method based on graph neural networks described above. Details not described in detail in the technical solution of the distribution network switch status identification system based on graph neural networks in this embodiment can be found in the description of the technical solution of the distribution network switch status identification method based on graph neural networks described above.
[0039] This embodiment provides a power distribution network switch status identification system based on graph neural networks, including: The building module is used to construct the power distribution network topology structure based on the physical connection relationship of the power distribution network; The acquisition module is used to acquire electrical measurement data of the nodes and use it as node characteristics; The identification module is used to build a graph neural network model. It inputs the distribution network topology and node features into the pre-trained graph neural network model and outputs the state probability of each edge in the distribution network topology to identify the switch state.
[0040] This embodiment also provides a computer device suitable for identifying the switch status of a power distribution network based on a graph neural network, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a power distribution network switch state recognition method based on a graph neural network as proposed in the above embodiments.
[0041] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a power distribution network switch state recognition method based on a graph neural network as proposed in the above embodiment.
[0042] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for identifying the switch status of a power distribution network based on a graph neural network proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0043] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0044] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for identifying the switch status of a power distribution network based on a graph neural network, characterized in that, include: Based on the physical connections of the power distribution network, construct the power distribution network topology. Acquire electrical measurement data of nodes and use it as node characteristics; A graph neural network model is constructed. The distribution network topology and node features are input into the pre-trained graph neural network model, and the state probabilities of each edge in the distribution network topology are output to identify the switch state.
2. The method for identifying the switch status of a power distribution network based on a graph neural network as described in claim 1, characterized in that, Constructing the distribution network topology includes: Based on the physical connection relationship of the power distribution network, construct the corresponding topology graph structure G=(V, E), where V represents the set of nodes, each node represents an electrical connection point or device, and E represents the set of edges, each edge represents a switch or connection line.
3. The method for identifying the switch status of a power distribution network based on a graph neural network as described in claim 1, characterized in that, The electrical measurement data of the nodes includes: Obtain real-time or historical electrical measurement data of each node in the power distribution network topology and construct a node feature matrix X.
4. The method for identifying the switch status of a power distribution network based on a graph neural network as described in claim 1, characterized in that, Building a graph neural network model includes: The graph neural network model includes an encoder and a decoder; The encoder learns and outputs the node embedding representation Z based on the input distribution network topology structure G and node feature matrix X; The decoder calculates the existence probability of each edge in the power distribution network topology based on the embedded representation Z to determine the switch state.
5. The method for identifying the switch status of a power distribution network based on a graph neural network as described in claim 4, characterized in that, Determining the switch status includes: The graph neural network model is trained using a historical dataset with known switch state labels to optimize model parameters; The real-time topology and node features of the distribution network to be identified are input into the trained graph neural network model, and the status identification results of each switch are output.
6. The method for identifying the switch status of a power distribution network based on a graph neural network as described in claim 5, characterized in that, Training the graph neural network model includes: The graph neural network model training uses the binary cross-entropy loss function as the optimization objective, and the loss function is expressed as: in, E Represents the set of edges that represent actual closed switches. This represents the set of edges where the switch is open, obtained through negative sampling. Representing an edge The true status label, This represents the probability that the edge is in a closed state, as predicted by the model.
7. The method for identifying the switch status of a power distribution network based on a graph neural network as described in claim 2, characterized in that, The distribution network topology structure includes: The nodes include bus nodes, load nodes, and distributed power generation nodes, and the edges represent switches, lines, or transformers.
8. A power distribution network switch status identification system based on graph neural networks, employing the power distribution network switch status identification method based on graph neural networks as described in any one of claims 1 to 7, characterized in that, include: The building module is used to construct the power distribution network topology structure based on the physical connection relationship of the power distribution network; The acquisition module is used to acquire electrical measurement data of the nodes and use it as node characteristics; The identification module is used to construct a graph neural network model. It inputs the distribution network topology structure and node features into the pre-trained graph neural network model and outputs the state probability of each edge in the distribution network topology to identify the switch state.
9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the distribution network switch state recognition method based on graph neural network as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, which, when executed by a processor, implement the steps of the power distribution network switch state recognition method based on a graph neural network as described in any one of claims 1 to 7.