A method and device for establishing a power grid digital twin system based on a graph neural network

Abstract

The application discloses a power grid digital twin system establishment method and device based on a graph neural network, wherein the method comprises the following steps: calculating a prediction result of a digital twin model of each device in a device set contained in a target power grid system; determining a device relationship between each device in the device set and other devices in the device set; defining a device interaction graph based on the prediction result of the digital twin model and the device relationship; obtaining a new representation of each device through a deep graph convolutional neural network model based on the device interaction graph; and merging the new representations of all devices in the device set to obtain an overall state representation of the device set. The application uses the prediction result of the digital twin model of the device, and then adjusts the prediction result according to the relationship between the devices, so as to obtain an overall prediction result.

Description

Technical Field

[0001] This invention relates to the field of software engineering technology, and more specifically, to a method and apparatus for establishing a digital twin system for power grids based on graph neural networks. Background Technology

[0002] For modeling digital twin systems of different sizes, the existing approach is to model the entirety of the devices contained in the target system using digital twin modeling techniques.

[0003] However, the problem is that existing technologies require remodeling for physical systems with varying numbers of devices, which is time-consuming. Furthermore, existing technologies cannot utilize existing digital twin models of equipment, but nonlinear relationships are prevalent among power grid devices. Therefore, a simple set of devices cannot constitute a complete digital twin model of the substation. Typical nonlinear relationships include:

[0004] Generators and Loads: The voltage and frequency of a generator can be affected by changes in the load connected to the power grid. For example, when the load increases, the generator's output frequency may decrease without proper control. This is because the additional load absorbs more mechanical power, thus slowing down the generator's rotation.

[0005] Transformers and Loads: The output voltage of a transformer is affected by its input voltage and the load connected to its output. This relationship is non-linear because the transformer's behavior is affected by the saturation of the magnetic material, which is more pronounced under high loads or high input voltages.

[0006] Power electronic equipment: For example, devices such as HVDC (High Voltage Direct Current Transmission) and FACTS (Flexible AC Transmission Systems) behave nonlinearly. These devices typically use power electronic converters (such as inverters and rectifiers) to control the power flow, and these converters behave nonlinearly.

[0007] Protecting equipment and power grids: For example, the behavior of circuit breakers and relays is affected by power grid conditions such as voltage and current. When a fault occurs in the power grid, these devices interrupt the power flow to protect other parts of the grid. This interruption behavior is non-linear.

[0008] Therefore, a technology is needed to enable in-depth mining and accurate modeling of relationships between devices based on existing digital twin models of devices. Summary of the Invention

[0009] The present invention provides a method and apparatus for establishing a power grid digital twin system based on graph neural networks, in order to solve the problem of how to accurately model the relationship between devices based on graph neural networks and existing equipment digital twin models.

[0010] To address the above problems, this invention provides a method for establishing a power grid digital twin system based on graph neural networks, the method comprising:

[0011] Calculate the prediction results of the digital twin model for each device in the set of devices included in the target power grid system;

[0012] Determine the device relationships between each device in the device set and the other devices in the device set;

[0013] Based on the prediction results of the digital twin model and the device relationships, a device interaction diagram is defined;

[0014] Based on the device interaction graph, a new representation of each device is obtained through a depth graph convolutional neural network model;

[0015] The new representations of all devices in the device set are merged to obtain the overall state representation of the device set.

[0016] Preferably, the step of obtaining a new representation of each device based on the device interaction graph using a depth graph convolutional neural network model includes:

[0017] The depth graph convolutional neural network model uses an activation function at each layer to perform graph convolution operations on the device interaction graph to obtain a new representation for each device.

[0018] Preferably, the method further includes training the depth map convolutional neural network model:

[0019] Set the training objective for the depth map convolutional neural network model;

[0020] After setting the training objective, set the objective function:

[0021] Let y be the actual fault, y hat The fault is predicted by the depth map convolutional neural network model, and the objective function is set as follows: Where ||•||2 is the L2 norm, and ||·|| F λ is the Frobenius norm, and λ is the regularization parameter.

[0022] Training stops when the depth map convolutional neural network model meets the training objective.

[0023] Preferably, the method further includes setting the parameters of the depth map convolutional neural network model:

[0024] The learning rate of the depth graph convolutional neural network model is 0.001, the regularization parameter λ is 0.001, and the depth L is 3 layers.

[0025] Preferably, the method further includes defining graph convolutional layers for the deep graph convolutional neural network model, wherein the graph convolution operation of each graph convolutional layer l is represented as: H l+1 =σ(D -0.5 ∧D -0.5 H l W l H l For the node features of layer l, H l+1 For the node features of layer l+1, W l Let be the weight matrix of the l-th layer, σ be the activation function, and D be the degree matrix.

[0026] Preferably, the feature representation of each device in the device set is a feature matrix X∈R. n×d Where n is the number of devices, d is the feature dimension of the devices, X is the feature matrix, and R is the feature of the devices;

[0027] Define the climate model as f c (x c The climate model is f. c (x c The output of f is added to the features to generate an enhanced feature matrix X′ = [X; f c (x c )).

[0028] Preferably, determining the device relationship between each device in the device set and other devices in the device set includes:

[0029] The device relationships are represented as an adjacency matrix A∈R n×n The adjacency matrix A is a weighted matrix, and the elements in the adjacency matrix A represent the connection strength between devices.

[0030] According to another aspect of the present invention, the present invention provides an apparatus for establishing a digital twin system for a power grid based on a graph neural network, the apparatus comprising:

[0031] The initial unit is used to calculate the prediction results of the digital twin model of each device in the set of devices contained in the target power grid system; and to determine the device relationships between each device in the set of devices and other devices in the set of devices.

[0032] A definition unit is used to define a device interaction diagram based on the prediction results of the digital twin model and the device relationships;

[0033] The acquisition unit is used to acquire a new representation of each device based on the device interaction graph using a depth graph convolutional neural network model;

[0034] The result unit is used to merge the new representations of all devices in the device set to obtain the overall state representation of the device set.

[0035] Preferably, the acquisition unit is configured to acquire a new representation of each device based on the device interaction graph using a depth graph convolutional neural network model, and is further configured to:

[0036] The depth graph convolutional neural network model uses an activation function at each layer to perform graph convolution operations on the device interaction graph to obtain a new representation for each device.

[0037] Preferably, the acquisition unit is further configured to train the depth map convolutional neural network model:

[0038] Set the training objective for the depth map convolutional neural network model;

[0039] After setting the training objective, set the objective function:

[0040] Let y be the actual fault, y hat The fault is predicted by the depth map convolutional neural network model, and the objective function is set as follows: Where ||•||2 is the L2 norm, and ||·|| F λ is the Frobenius norm, and λ is the regularization parameter.

[0041] Training stops when the depth map convolutional neural network model meets the training objective.

[0042] Preferably, the acquisition unit is further configured to set the parameters of the depth map convolutional neural network model:

[0043] The learning rate of the depth graph convolutional neural network model is 0.001, the regularization parameter λ is 0.001, and the depth L is 3 layers.

[0044] Preferably, the acquisition unit is further configured to define graph convolutional layers for the deep graph convolutional neural network model, wherein the graph convolution operation of each graph convolutional layer l is represented as: H l+1 =σ(D -0.5 ∧D -0.5 H l W l H l For the node features of layer l, H l+1 For the node features of layer l+1, W l Let be the weight matrix of the l-th layer, σ be the activation function, and D be the degree matrix.

[0045] Preferably, the initial unit is further configured to represent the features of each device in the device set as a feature matrix X∈R. n×d Where n is the number of devices, d is the feature dimension of the devices, X is the feature matrix, and R is the feature of the devices;

[0046] Define the climate model as f c (x c The climate model is f. c (x c The output of f is added to the features to generate an enhanced feature matrix X′ = [X; f c (x c )).

[0047] Preferably, the initial unit is configured to, in order to determine the device relationship between each device in the device set and other devices in the device set, further be configured to:

[0048] The device relationships are represented as an adjacency matrix A∈R n×n The adjacency matrix A is a weighted matrix, and the elements in the adjacency matrix A represent the connection strength between devices.

[0049] According to another aspect of the present invention, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program for executing a method for establishing a power grid digital twin system based on a graph neural network.

[0050] According to another aspect of the present invention, the present invention provides an electronic device, characterized in that the electronic device comprises: a processor and a memory; wherein,

[0051] The memory is used to store the processor-executable instructions;

[0052] The processor is configured to read the executable instructions from the memory and execute the instructions to implement a method for establishing a power grid digital twin system based on a graph neural network.

[0053] This invention provides a method and apparatus for establishing a digital twin system for a power grid based on graph neural networks. The method includes: calculating the prediction result of the digital twin model of each device in a set of devices included in the target power grid system; determining the device relationships between each device and other devices in the set; defining a device interaction graph based on the prediction results of the digital twin model and the device relationships; obtaining a new representation of each device using a deep graph convolutional neural network model based on the device interaction graph; and merging the new representations of all devices in the set to obtain the overall state representation of the set. This invention first uses the prediction results of the digital twin models of the devices, and then adjusts these prediction results according to the relationships between the devices to obtain the overall prediction result. The overall prediction result of this invention is obtained by considering the relationships between devices based on the prediction results of each device, and can be used as the prediction result of the overall state of the substation. This invention avoids the complex and time-consuming global training process, providing a more efficient and practical method for modeling digital twin systems of large-scale device sets. Attached Figure Description

[0054] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures:

[0055] Figure 1 A flowchart illustrating a method for establishing a power grid digital twin system based on a preferred embodiment of the present invention; and

[0056] Figure 2 This is a structural diagram of a device for establishing a power grid digital twin system based on a preferred embodiment of the present invention. Detailed Implementation

[0057] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0058] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0059] Figure 1This is a flowchart illustrating a method for establishing a power grid digital twin system based on a preferred embodiment of the present invention. The invention uses a 10kV substation as the research object, which includes equipment such as Schneider SM6 ring main units, Mvnex circuit breaker cabinets, Schneider-EV12s high-voltage circuit breakers, ABB-SFG high-voltage load switches, LZZBJ9-10 current transformers, and ABB-140C integrated protection devices. Each piece of equipment already has a corresponding digital twin model. Without needing to retrain the overall digital twin model of the substation, the digital twin system of the substation is constructed by modeling the relationships between the equipment.

[0060] like Figure 1 As shown, this invention provides a method for establishing a digital twin system for a power grid based on a graph neural network, the method comprising:

[0061] Step 101: Calculate the prediction results of the digital twin model for each device in the set of devices included in the target power grid system;

[0062] Preferably, the feature representation of each device in the device set is a feature matrix X∈R. n×d Where n is the number of devices, d is the feature dimension of the devices, X is the feature matrix, and R is the feature of the devices;

[0063] Define the climate model as f c (x c ), and set the climate model to f c (x c The output of f is added to the features to generate an enhanced feature matrix X′ = [X; f c (x c )).

[0064] The digital twin model of the prior art device of this invention can be expressed as a function f i (x i The effective range includes a temperature range T = [0, 50] degrees Celsius and a load range L = [0, 100] kilowatts. This model can also accurately capture the dynamic behavior and nonlinear characteristics of the equipment. Where x... i This is the input to the device, including real-time and historical data. Assuming we have n = 7 devices, each with a feature dimension of d = 100, we can represent the features of all devices as a feature matrix X ∈ R. n ×d .

[0065] This invention defines a climate model, which can be expressed as f c (x c ), which is normalized through a preprocessing step to ensure that it is on the same dimension as the device characteristics, where xc This is a climate-related input. This invention can add the output of a climate model to the device's features, forming an enhanced feature matrix X′=[X;f c (x c )).

[0066] Step 102: Determine the device relationships between each device in the device set and the other devices in the device set;

[0067] Preferably, determining the device relationships between each device in the device set and other devices in the device set includes:

[0068] Device relationships are represented as an adjacency matrix A∈R n×n The adjacency matrix A is a weighted matrix, and the elements in the adjacency matrix A represent the connection strength between devices.

[0069] The relationships between devices in this invention are represented as an adjacency matrix A∈R n×n The adjacency matrix A here is a weighted matrix, where A ij This represents the connection strength between device i and device j. This value can be set according to the actual operation data. For devices that are not related, the corresponding item in the matrix is ​​0.

[0070] Step 103: Based on the prediction results of the digital twin model and the device relationships, define the device interaction diagram;

[0071] This invention defines a device interaction graph. The device interaction graph of this invention can be represented as a graph G = (V, E), which is a dynamic graph whose topology may adjust with changes in device status or time. The specific adjustments are determined through analysis of historical data, where V represents the set of devices and E represents the set of relationships between devices. In this graph, the characteristics of nodes are given by the feature matrix X, and the connections between nodes are defined by the adjacency matrix A.

[0072] Step 104: Based on the device interaction graph, obtain a new representation for each device using a depth graph convolutional neural network model;

[0073] Preferably, based on the device interaction graph, a new representation of each device is obtained through a depth graph convolutional neural network model, including:

[0074] The deep graph convolutional neural network model uses activation functions in each layer to perform graph convolution operations on the device interaction graph to obtain a new representation for each device.

[0075] Preferably, the method further includes training a depth map convolutional neural network model:

[0076] Set the training objective for the depth map convolutional neural network model;

[0077] After setting the training objective, set the objective function:

[0078] Let y be the actual fault, y hat The fault is predicted by the depth map convolutional neural network model. The objective function is set as follows: Where ||·||2 is the L2 norm, ||·|| F λ is the Frobenius norm, and λ is the regularization parameter.

[0079] Training stops when the depth graph convolutional neural network model meets the training objective.

[0080] Preferably, the method further includes setting the parameters of the depth map convolutional neural network model:

[0081] The learning rate of the deep graph convolutional neural network model is 0.001, the regularization parameter λ is 0.001, and the depth L is 3 layers.

[0082] Preferably, it further includes defining graph convolutional layers for the deep graph convolutional neural network model, wherein the graph convolution operation of each graph convolutional layer l is represented as: H l+1 =σ(D -0.5 ∧D -0.5 H l W l H l For the node features of layer l, H l+1 For the node features of layer l+1, W l Let be the weight matrix of the l-th layer, σ be the activation function, and D be the degree matrix.

[0083] This invention defines a depth graph convolutional neural network model. In this model, a series of graph convolutional layers need to be defined. The graph convolution operation of each layer l can be represented as H. l+1 =σ(D -0.5 ∧D -0.5 H l W l Here, we assume that graph G = (V, E) is undirected. If the graph is directed in a practical application, this will be handled by symmetrizing the adjacency matrix A. Furthermore, H... l W is the node feature of layer l. l σ is the weight matrix of the l-th layer, and σ is the activation function, such as the ReLU function. Here, D is the degree matrix, and its diagonal elements are... ii It is the degree of node i. To simplify the model, this invention assumes that the weight matrix W of each layer is... l The dimensions are all d×d, meaning that each layer maintains the same feature dimension.

[0084] This invention trains a depth map convolutional network model. The training objective of this invention is to minimize the difference between the predicted fault and the actual fault. Let y be the actual fault, y hat The fault is predicted by the model, and the objective function of this invention can be expressed as: Where ||·||2 is the L2 norm, ||·|| F λ is the Frobenius norm, and λ is the regularization parameter. Furthermore, the datasets used for model training, validation, and testing are generated from real-world operational data collected over a year. There is no temporal overlap between these three datasets to ensure the model's generalization ability.

[0085] This invention sets the hyperparameters, employing the Adam optimizer with a learning rate of 0.001 and a regularization parameter λ of 0.001. The depth of the graph convolutional network is set to L = 3 layers.

[0086] Step 105: Merge the new representations of all devices in the device set to obtain the overall state representation of the device set.

[0087] This invention assembles a complete digital twin model and uses a model of relationships between devices to adjust the output of the digital twin model of the devices.

[0088] For each device i, first calculate the output y of the device's digital twin model. i =f i (x i Then, a new representation for each device is obtained using a graph neural network (GNN), which takes into account the relationships between devices. The new representation for device i can be obtained by the following formula: h i =GNN(y i ,A) where A is the adjacency matrix of the relationships between devices.

[0089] This invention assembles a comprehensive digital twin model of a substation. This invention provides new representations h for all equipment. i The state representation of the merged, integrated substation is h = ∑ i h i The model first uses the prediction results from the digital twin models of the equipment, and then adjusts these prediction results according to the relationships between the equipment to obtain the overall prediction result. This overall prediction result is obtained by considering the relationships between the equipment based on the prediction results of each individual device, and can be used as the prediction result of the overall state of the substation.

[0090] This invention provides a method for modeling relationships between physical devices. It employs a deep graph convolutional neural network, using activation functions such as ReLU to perform graph convolution operations at each layer. The model aims to minimize the L2 norm loss between predicted and actual faults, while also considering the Frobenius norm of the weight matrix to achieve model regularization, thereby enabling effective network training.

[0091] This invention provides a method for assembling a larger-scale digital twin system based on existing digital twin models of physical equipment and the relationships between these equipment. Building upon existing digital twin models of physical equipment, this invention uses graph neural networks to accurately represent and model the interaction relationships between equipment. Based on this, the invention adjusts the outputs of the digital twin models of each equipment and merges these new representations to form the overall state representation of the substation. This method can effectively assemble the overall digital twin system without retraining on a larger set of equipment, thus forming a larger-scale digital twin system that includes the relationships between equipment.

[0092] The digital twin system assembly method proposed in this invention has the following advantages:

[0093] (1) Mining and Modeling Relationships Between Devices: The method proposed in this invention can achieve in-depth mining and accurate modeling of relationships between devices based on existing digital twin models. Specifically, a device interaction graph is first constructed, which reveals the relationships between devices through an adjacency matrix. Subsequently, the output of the digital twin model of each device is adjusted using a graph neural network (GNN) to generate a new device representation that takes into account the relationships between devices. This process not only reveals the relationships between devices but also mines potential interactions between devices that significantly affect device performance, thus improving the accuracy of the model.

[0094] (2) Achieving Digital Twin System Modeling for Larger Equipment Sets: The method proposed in this invention can quickly and effectively model the corresponding digital twin system based on existing digital twin models and inter-equipment relationship models without requiring retraining for a larger equipment set. In this process, the new representations of all equipment are merged to form a unified substation state representation. This overall prediction result, based on the prediction results of each individual device and considering the relationships between devices, can be used as the prediction result for the overall substation state. This approach avoids the complex and time-consuming global training process, providing a more efficient and practical method for modeling digital twin systems with large-scale equipment sets.

[0095] Figure 2This is a structural diagram of a device for establishing a power grid digital twin system based on a preferred embodiment of the present invention.

[0096] like Figure 2 As shown, the present invention provides a device 200 for establishing a digital twin system for a power grid based on a graph neural network. The device includes:

[0097] The initial unit 201 is used to calculate the prediction results of the digital twin model of each device in the set of devices included in the target power grid system; and to determine the device relationship between each device in the set of devices and other devices in the set of devices.

[0098] Preferably, the initial unit 201 is further configured to represent the features of each device in the device set as a feature matrix X∈R. n×d Where n is the number of devices, d is the feature dimension of the devices, X is the feature matrix, and R is the feature of the devices;

[0099] Define the climate model as f c (x c ), and set the climate model to f c (x c The output of f is added to the features to generate an enhanced feature matrix X′ = [X; f c (x c )).

[0100] Preferably, the initial unit 201 is used to determine the device relationship between each device in the device set and other devices in the device set, and is further used to:

[0101] Device relationships are represented as an adjacency matrix A∈R n×n The adjacency matrix A is a weighted matrix, and the elements in the adjacency matrix A represent the connection strength between devices.

[0102] Define unit 202, which is used to define the device interaction diagram based on the prediction results of the digital twin model and the device relationships;

[0103] The acquisition unit 203 is used to acquire a new representation of each device based on the device interaction graph using a depth graph convolutional neural network model.

[0104] Preferably, the acquisition unit 203 is used to acquire a new representation of each device based on the device interaction graph using a depth graph convolutional neural network model, and is also used to:

[0105] The deep graph convolutional neural network model uses activation functions in each layer to perform graph convolution operations on the device interaction graph to obtain a new representation for each device.

[0106] Preferably, the acquisition unit 203 is also used to train the depth map convolutional neural network model:

[0107] Set the training objective for the depth map convolutional neural network model;

[0108] After setting the training objective, set the objective function:

[0109] Let y be the actual fault, y hat The fault is predicted by the depth map convolutional neural network model. The objective function is set as follows: Where ||·||2 is the L2 norm, ||·|| F λ is the Frobenius norm, and λ is the regularization parameter.

[0110] Training stops when the depth graph convolutional neural network model meets the training objective.

[0111] Preferably, the acquisition unit 203 is also used to set the parameters of the depth map convolutional neural network model:

[0112] The learning rate of the deep graph convolutional neural network model is 0.001, the regularization parameter λ is 0.001, and the depth L is 3 layers.

[0113] Preferably, the acquisition unit 203 is further configured to define graph convolutional layers for the deep graph convolutional neural network model, wherein the graph convolution operation of each graph convolutional layer l is represented as: H l+1 =σ(D -0.5 ∧D -0.5 H l W l H l For the node features of layer l, H l+1 For the node features of layer l+1, W l Let be the weight matrix of the l-th layer, σ be the activation function, and D be the degree matrix.

[0114] Result unit 204 is used to merge the new representations of all devices in the device set to obtain the overall state representation of the device set.

[0115] The present invention provides a power grid digital twin system establishment device 200 based on graph neural network, which corresponds to the power grid digital twin system establishment method based on graph neural network provided by the present invention, and will not be described in detail here.

[0116] The present invention provides a computer-readable storage medium storing a computer program for executing a method for establishing a power grid digital twin system based on a graph neural network.

[0117] This invention provides an electronic device, which includes: a processor and a memory; wherein,

[0118] Memory, used to store processor-executable instructions;

[0119] A processor for reading executable instructions from memory and executing the instructions to implement a power grid digital twin system establishment device based on graph neural networks.

[0120] 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 implemented 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. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0121] 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.

[0122] 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.

[0123] 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.

[0124] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0125] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0126] The invention has been described with reference to a few embodiments. However, as will be known to those skilled in the art, and as defined in the appended claims, other embodiments besides those disclosed above fall equivalently within the scope of the invention.

[0127] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the art, unless otherwise expressly defined herein. All references to “a / the / the [device, component, etc.]” ​​are openly interpreted as at least one instance of said device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed unless explicitly stated otherwise.

Claims

1. A method for establishing a digital twin system for a power grid based on a graph neural network, the method comprising: Calculate the prediction results of the digital twin model for each device in the set of devices included in the target power grid system; Determine the device relationships between each device in the device set and other devices in the device set; represent the features of each device in the device set as a feature matrix X∈ Where n is the number of devices, d is the feature dimension of the devices, X is the feature matrix, and R is the feature of the devices; Define the climate model as The climate model is The output is added to the features to generate an enhanced feature matrix. ]; The input is climate-related; the device relationships are represented as an adjacency matrix A∈ The adjacency matrix A is a weighted matrix, and the elements in the adjacency matrix A represent the connection strength between devices; Based on the prediction results of the digital twin model and the device relationships, a device interaction graph is defined. The device interaction graph is represented as a graph G=(V,E), which is a dynamic graph whose topology adjusts with changes in device status or time. The specific adjustments are determined through analysis of historical data, where V represents the set of devices and E represents the set of relationships between devices. Based on the device interaction graph, a new representation of each device is obtained using a deep graph convolutional neural network model, including: The depth graph convolutional neural network model uses an activation function at each layer to perform graph convolution operations on the device interaction graph to obtain a new representation for each device; Summing and merging the new representations of all devices in the device set yields the overall state representation of the device set.

2. The method according to claim 1 further includes training the depth map convolutional neural network model: Set the training objective for the depth map convolutional neural network model; After setting the training objective, set the objective function: Let y be the actual fault. The fault is predicted by the depth map convolutional neural network model, and the objective function is set as follows: ,in It is the L2 norm. λ is the Frobenius norm, and λ is the regularization parameter. Training stops when the depth map convolutional neural network model meets the training objective.

3. The method according to claim 2 further includes setting the parameters of the depth map convolutional neural network model: The learning rate of the depth graph convolutional neural network model is 0.001, the regularization parameter λ is 0.001, and the depth L is 3 layers.

4. A device for establishing a digital twin system for a power grid based on a graph neural network, the device comprising: The initial unit is used to calculate the prediction results of the digital twin model of each device in the set of devices contained in the target power grid system; Determine the device relationships between each device in the device set and other devices in the device set; represent the features of each device in the device set as a feature matrix X∈ Where n is the number of devices, d is the feature dimension of the devices, X is the feature matrix, and R is the feature of the devices; Define the climate model as The climate model is The output is added to the features to generate an enhanced feature matrix. ]; Climate-related inputs; The device relationships are represented as an adjacency matrix A∈ The adjacency matrix A is a weighted matrix, and the elements in the adjacency matrix A represent the connection strength between devices; A definition unit is used to define a device interaction graph based on the prediction results of the digital twin model and the device relationships. The device interaction graph is represented as a graph G=(V,E), which is a dynamic graph whose topology will be adjusted as the device status or time changes. The specific adjustment is determined by the analysis of historical data, where V represents the set of devices and E represents the set of relationships between devices. The acquisition unit is configured to acquire a new representation of each device based on the device interaction graph using a depth graph convolutional neural network model, and is also configured to: The depth graph convolutional neural network model uses an activation function at each layer to perform graph convolution operations on the device interaction graph to obtain a new representation for each device; The result unit is used to sum and merge the new representations of all devices in the device set to obtain the overall state representation of the device set.

5. The apparatus according to claim 4, wherein the acquisition unit is further configured to train the depth map convolutional neural network model: Set the training objective for the depth map convolutional neural network model; After setting the training objective, set the objective function: Let y be the actual fault. The fault is predicted by the depth map convolutional neural network model, and the objective function is set as follows: ,in It is the L2 norm. λ is the Frobenius norm, and λ is the regularization parameter. Training stops when the depth map convolutional neural network model meets the training objective.

6. The apparatus according to claim 5, wherein the acquisition unit is further configured to set the parameters of the depth map convolutional neural network model: The learning rate of the depth graph convolutional neural network model is 0.001, the regularization parameter λ is 0.001, and the depth L is 3 layers.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program for performing the method according to any one of claims 1-3.

8. An electronic device, characterized in that, The electronic device includes: a processor and a memory; wherein, The memory is used to store the processor-executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method of any one of claims 1-3.

Images