A method for predicting power distribution network electrothermal stress and related devices

Abstract

The application discloses a power distribution network electric thermal stress prediction method and related devices, and the method comprises the following steps: constructing a directed graph of the power distribution network according to the topological structure, node characteristics and edge characteristics of the power distribution network; and obtaining the final node stress sequence and edge stress sequence at each time in the future period by using a stress prediction model according to the directed graph, and the node stress sequence and edge stress sequence at each time in the current period. The application obtains high-order characteristics of nodes and edges based on the directed graph, obtains the initial node stress sequence and edge stress sequence at each time in the future period based on the node stress sequence and edge stress sequence at each time in the current period, fuses the high-order characteristics and the initial stress sequence, and obtains the final stress sequence at each time in the future period, so that the whole process of stress accumulation is predicted, and a reliable basis is provided for the prediction of the service life and failure time of the power distribution network.

Description

Technical Field

[0001] This invention relates to a method and related apparatus for predicting the electrothermal stress of a power distribution network, belonging to the field of integrated circuit design and electronic design automation (EDA). Background Technology

[0002] In power distribution networks, due to high current density and the presence of local hot spots, electrothermal stress accumulates continuously in the interconnect structure, leading to interconnect aging and system failure. Therefore, to accurately predict the lifespan and failure time of power distribution networks, it is necessary to predict the electrothermal stress over multiple future time steps. In recent years, with the application of machine learning in EDA, some deep learning-based electrothermal stress prediction methods have emerged. By learning about interconnect structure and feature information, they achieve rapid prediction of electrothermal stress at a single time step. However, these methods are limited to predicting instantaneous stress and cannot predict the entire process of stress accumulation. Summary of the Invention

[0003] This invention provides a method and related apparatus for predicting the electrothermal stress of a power distribution network, which solves the problems disclosed in the background art.

[0004] According to one aspect of this application, a method for predicting the electrothermal stress of a power distribution network is provided, comprising:

[0005] Based on the topology, node characteristics, and edge characteristics of the power distribution network, a directed graph of the power distribution network is constructed; where nodes are the points where wires intersect in the power distribution network, and edges are the wires connecting adjacent nodes.

[0006] Based on the directed graph and the node stress sequences and edge stress sequences at each moment in the current time period, a stress prediction model is used to obtain the final node stress sequences and edge stress sequences at each moment in the future time period. The current time period is the time interval from the initial moment to the current moment, and the future time period is the time interval from the next moment to the future moment. In the stress prediction model, based on the directed graph, the higher-order features of the nodes and edges are obtained. Based on the node stress sequences and edge stress sequences at each moment in the current time period, the initial node stress sequences and edge stress sequences at each moment in the future time period are obtained respectively. The initial node stress sequences and the higher-order features of the nodes are fused, and the final node stress sequence is obtained based on the fusion result. The initial edge stress sequences and the higher-order features of the edges are fused, and the final edge stress sequence is obtained based on the fusion result. The higher-order features of node A represent the influence of edges and other nodes on node A, and the higher-order features of edge B represent the influence of nodes and other edges on edge B.

[0007] Furthermore, node features include node temperature, and edge features include edge current density, discrete temperature distribution, length, and width.

[0008] Furthermore, the stress prediction model includes a graph attention network, a Transformer network, a gated fusion layer, and a linear transformation layer;

[0009] Graph attention networks are used to obtain higher-order features of nodes and edges;

[0010] Transformer networks are used to obtain the initial node stress sequence and edge stress sequence at each time point in the future time period;

[0011] The gated fusion layer is used to fuse the initial node stress sequence and the higher-order features of the nodes, and to fuse the initial edge stress sequence and the higher-order features of the edges.

[0012] The linear transformation layer is used to obtain the final nodal stress sequence and the final edge stress sequence based on the fusion result.

[0013] Furthermore, in the graph attention network, a node attention aggregator is used for node feature updating, with the following formula:

[0014] ;

[0015] In the formula, σ is the Sigmoid activation function. This represents the updated feature of the i-th node. Let represent the features of the i-th node, the j-th node, and the edge (i,j) before the update, respectively. (i,j) represents the edge between the i-th and j-th nodes, and || denotes the cascading operation. is the node attention coefficient, used to quantify the degree of influence of the neighbor node features of the i-th node on the features of the i-th node;

[0016] Edge feature updates are performed using a bidirectional edge aggregator, with the following formula:

[0017] ;

[0018] In the formula, The features of the updated edge (i,j) These are the learnable weights and biases, respectively.

[0019] Furthermore, in the Transformer network, a learnable linear projection is used to adaptively map the encoder's output to the decoder's temporal domain, while simultaneously generating synchronous positional codes as cues.

[0020] Furthermore, the fusion formula for the gated fusion layer is as follows:

[0021] ;

[0022] In the formula, H represents the fusion result, and g represents the gating weight. σ is the Sigmoid activation function. These are the learnable weights and biases, H represents element-wise matrix multiplication, || represents concatenation, and H represents the concatenation operation. s For the initial nodal stress sequence or the initial edge stress sequence, H T These are the higher-order features of nodes or edges.

[0023] Furthermore, the formula for the linear transformation layer is:

[0024] ;

[0025] In the formula, Here is the final predicted stress sequence, and H is the fusion result. These are the learnable weights and biases, respectively.

[0026] According to another aspect of this application, a device for predicting the electrothermal stress of a power distribution network is provided, comprising:

[0027] The construction module constructs a directed graph of the power distribution network based on its topology, node characteristics, and edge characteristics; where nodes are the points where wires intersect in the power distribution network, and edges are the wires connecting adjacent nodes.

[0028] The prediction module, based on the directed graph and the node stress sequences and edge stress sequences at each moment within the current time period, employs a stress prediction model to obtain the final node stress sequences and edge stress sequences at each moment within the future time period. Here, the current time period is the time interval from the initial moment to the current moment, and the future time period is the time interval from the next moment to the future moment. In the stress prediction model, based on the directed graph, higher-order features of nodes and edges are obtained. Based on the node stress sequences and edge stress sequences at each moment within the current time period, the initial node stress sequences and edge stress sequences at each moment within the future time period are obtained respectively. The initial node stress sequences and the higher-order features of the nodes are fused, and the final node stress sequence is obtained based on the fusion result. The initial edge stress sequences and the higher-order features of the edges are fused, and the final edge stress sequence is obtained based on the fusion result. The higher-order features of node A represent the influence of edges and other nodes on node A, and the higher-order features of edge B represent the influence of nodes and other edges on edge B.

[0029] According to another aspect of this application, a computer-readable storage medium is provided that stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform a method for predicting the electrothermal stress of a power distribution network.

[0030] According to another aspect of this application, a computer device is provided, including one or more processors and one or more memories, wherein one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing a method for predicting the electrothermal stress of a power distribution network.

[0031] The beneficial effects achieved by this invention are as follows: Based on a directed graph, this invention obtains the higher-order features of nodes and edges. Based on the node stress sequence and edge stress sequence at each moment in the current time period, it obtains the initial node stress sequence and edge stress sequence at each moment in the future time period. By fusing the higher-order features and the initial stress sequence, the final stress sequence at each moment in the future time period is obtained, realizing the prediction of the entire process of stress accumulation and providing a reliable basis for predicting the service life and failure time of power distribution networks. Attached Figure Description

[0032] Figure 1 A flowchart of a method for predicting electrothermal stress in power distribution networks;

[0033] Figure 2 This is a schematic diagram of the stress prediction model;

[0034] Figure 3 Block diagram of a device for predicting the electrothermal stress of a power distribution network. Detailed Implementation

[0035] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this application or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0036] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0037] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0038] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0039] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0040] It should be noted that similar symbols and letters in the following figures represent similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0041] Graph Attention Network (GAT) is a specialized neural network designed specifically for processing graph-structured data. It employs an attention mechanism to compute relationships between nodes, thereby better capturing structural information within the graph. GAT's main advantage lies in its ability to automatically learn these relationships without manual pre-setting. In GAT, each node's state update considers the states of its neighbors. GAT calculates the attention weights between a node and its neighbors and then updates the node's state based on these weights. This mechanism allows GAT to better capture the structural information within the graph. Currently, GAT is widely used in various scenarios requiring the processing of graph-structured data, such as social network analysis, knowledge graphs, and recommender systems.

[0042] The Transformer network is a neural network architecture based on self-attention mechanisms, primarily used for processing sequential data, and is widely used in Natural Language Processing (NLP). The core idea of ​​the Transformer network is to utilize self-attention to capture the relationships between elements in a sequence, thereby avoiding the limitations of traditional methods. The problem of handling long sequence dependencies is difficult, and its main components include encoders, decoders, multi-head self-attention mechanisms, positional encoding, feedforward neural networks, and masking mechanisms.

[0043] The embodiments of this application are based on The Transformer network provides a method for predicting the electrothermal stress in power distribution networks, aiming to... A stress prediction model is constructed using a Transformer network to predict the electrothermal stress in a power distribution network. This prediction method can be executed by a prediction device, which can be a terminal device or a server. The terminal device can include, but is not limited to, mobile phones, computers, smart wearable devices, smart vehicle devices, etc., as described in this application embodiment. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, big data, and artificial intelligence platforms, etc., as described in this application embodiment. Optionally, the prediction method can also be executed collaboratively by multiple electronic devices with computing power; for ease of explanation, subsequent embodiments will be described as being executed by a prediction device.

[0044] See Figure 1 , Figure 1 This is a flowchart illustrating a method for predicting the electrothermal stress of a power distribution network according to an embodiment of this application. The prediction method can be executed by a prediction device and may include at least the following steps:

[0045] Step 1: Construct a directed graph of the power distribution network based on its topology, node characteristics, and edge characteristics; where nodes are the points where wires intersect in the power distribution network, and edges are the wires connecting adjacent nodes.

[0046] It should be noted that the topology of the power distribution network remains unchanged during the evolution of electrothermal stress. Therefore, a static directed graph can be used to represent the interconnection space topology of the power distribution network, which can be denoted as G=(V,E); where V is the set of nodes, E is the set of edges with current directions, and |V|=N. v Represents the number of nodes, |E|=N e This indicates the number of edges.

[0047] By further embedding spatial features into nodes and edges, a directed graph of the power distribution network can be obtained; among which, spatial features mainly include node features and edge features.

[0048] Node features include node temperature, and the features embedded in a node can be represented as:

[0049] ;

[0050] In the formula, v i T represents the feature of the i-th node. i This represents the temperature of the i-th node;

[0051] Edge features include edge current density, discrete temperature distribution, length, and width. The features of embedded edges can be represented as follows:

[0052] ;

[0053] In the formula, (i,j) represents the edge between the i-th node and the j-th node, e i,j As a feature of edge (i,j), J i,j T i,j L i,j and W i,j Let be the current density, discrete temperature distribution, length, and width of edge (i,j), respectively.

[0054] Step 2: Based on the directed graph and the node stress sequences and edge stress sequences at each moment in the current time period, a stress prediction model is used to obtain the final node stress sequences and edge stress sequences at each moment in the future time period. The current time period is the time interval from the initial moment to the current moment, and the future time period is the time interval from the next moment to the future moment. In the stress prediction model, based on the directed graph, the higher-order features of the nodes and edges are obtained. Based on the node stress sequences and edge stress sequences at each moment in the current time period, the initial node stress sequences and edge stress sequences at each moment in the future time period are obtained respectively. The initial node stress sequences and the higher-order features of the nodes are fused, and the final node stress sequence is obtained based on the fusion result. The initial edge stress sequences and the higher-order features of the edges are fused, and the final edge stress sequence is obtained based on the fusion result. The higher-order features of node A represent the influence of edges and other nodes on node A, and the higher-order features of edge B represent the influence of nodes and other edges on edge B.

[0055] It should be noted that the temporal characteristics of nodes and edges can be represented by the node stress sequence and edge stress sequence at each time step. Specifically, the node stress sequence is the sequence composed of the stresses of all nodes at the same time step, and the edge stress sequence is the sequence composed of the stresses of all edges at the same time step. These sequences can capture the temporal characteristics of stress accumulation on the power distribution network. Assuming the initial time step is 1 and the current time step is p, then the time characteristic S up to the current time step can be expressed as:

[0056] ;

[0057] In the formula, This represents the nodal stress sequence at time t. This represents the edge stress sequence at time t.

[0058] It should be noted that the node stresses in the node stress sequence can be represented by scalar stress values; the edge stress sequence is obtained by discrete sampling. The discrete sampling points are set at equal intervals in the interconnection segments of the power distribution network. Specifically, within each interconnection segment, five equally spaced discrete sampling points are selected for stress feature sampling. This reduces the amount of computation while retaining the key stress gradient information of the edge stress, thereby improving the efficiency of subsequent model training and prediction.

[0059] In some embodiments, see Figure 2 The stress prediction model can include at least a graph attention network, a Transformer network, a gated fusion layer, and a linear transformation layer. The graph attention network is used to obtain the high-order features of nodes and edges. The Transformer network is used to obtain the initial node stress sequence and edge stress sequence at each time point in the future time period. The gated fusion layer is used to fuse the initial node stress sequence and the high-order features of nodes, and to fuse the initial edge stress sequence and the high-order features of edges. The linear transformation layer is used to obtain the final node stress sequence and the final edge stress sequence based on the fusion result.

[0060] It should be noted that the graph attention mechanism of the graph attention network is used to capture spatially dependent features in the power distribution network, while taking into account the interactive updates between node features and edge features.

[0061] In graph attention networks, node feature updates are performed using a node attention aggregator. This aggregator aggregates features from neighboring nodes and connecting edges to generate updated features for the target node. Node attention coefficients quantify the influence of neighboring node features on the target node's features, and weighted aggregation of node features is achieved through a learnable weight vector. Edge feature updates are implemented using a bidirectional edge aggregator, which aggregates topological features from connected node pairs to generate updated edge features. In this way, the final node and edge features generated by the graph attention network can be viewed as a comprehensive encoding of the spatial features of the power distribution network.

[0062] Suppose that the node attention aggregator aggregates the features of two adjacent nodes and the features of the edges connecting the two nodes to generate updated features for the target node, which can be represented as:

[0063] ;

[0064] In the formula, σ is the Sigmoid activation function. This represents the updated feature of the i-th node. Let represent the features of the i-th node, the j-th node, and the edge (i,j) before the update, respectively. || denotes a cascading operation. The node attention coefficients are obtained by normalizing the concatenated vector of node features and edge features after mapping to a learnable weight vector:

[0065] ;

[0066] In the formula, LeakyReLU is the activation function of the leakage rectifier linear unit, and the superscript T indicates transpose. Let i and k represent the features of the k-th node and the edge (i,k) before the update, respectively. This represents the learnable weight vector.

[0067] For edge feature updates, a bidirectional edge aggregator is used to aggregate the topological features from the connected node pairs:

[0068] ;

[0069] In the formula, The features of the updated edge (i,j) These are the learnable weights and biases, respectively.

[0070] It should be noted that the Transformer network adopts an encoder-decoder framework, which models the temporal evolution of the stress sequence. It enhances the input sequence by adding sinusoidal position encoding to preserve the temporal order. The encoder part captures the inter-step dependency of the control electrothermal stress accumulation through a multi-head self-attention mechanism.

[0071] To achieve asymmetry between the input and output lengths of the Transformer network, in some embodiments, a learnable linear projection is used to adaptively map the encoder output to the temporal domain of the decoder, while generating synchronous positional codes as cues. This enables effective prediction of input-output sequences with asymmetric lengths. The decoder combines these projected features with the target-aware positional codes through a cross-attention mechanism, thereby achieving regression prediction of the evolution of node / edge stresses over multiple future time steps.

[0072] It should be noted that the stress distribution in the interconnect exhibits temporal dependence on its historical evolution and spatial interaction with neighboring conductors at any time step. To adaptively fuse the spatiotemporal representation, a gated fusion layer is used in the stress prediction model to fuse the initial stress sequence with higher-order features (i.e., spatial features), which can be expressed by the formula:

[0073] ;

[0074] In the formula, H represents the fusion result, and g represents the gating weight. σ is the Sigmoid activation function. These are the learnable weights and biases, H represents element-wise multiplication of matrices. s For the initial nodal stress sequence or the initial edge stress sequence, H T These are the higher-order features of nodes or edges.

[0075] It should be noted that, to further realize the mapping from the fused high-dimensional feature representation space to the actual stress value space, a linear transformation layer is introduced after the gated fusion layer. A learnable linear projection is used to map the fused features to the predicted stress values ​​at each time step. This process can be expressed by the following formula:

[0076] ;

[0077] In the formula, Here is the final predicted stress sequence, and H is the fusion result. These are the learnable weights and biases, respectively.

[0078] It should be noted that the stress prediction model described above needs to be trained before it can be used. The training process can be as follows:

[0079] S1) Introduce random perturbations into the power distribution network design parameters of multiple different circuits and use commercial physical design tools to generate a dataset covering a variety of power distribution network structures.

[0080] It should be noted that random disturbances in the power distribution network design parameters mainly affect the line width, spacing, and supply voltage of each layer of the power distribution network, and satisfy the minimum line width and minimum spacing constraints of the process design.

[0081] S2) Commercial finite element simulation tools are used to perform electrothermal migration coupling simulation on the power distribution network structure, extract the stress distribution data of the corresponding structure, and construct a stress distribution dataset.

[0082] It should be noted that when using commercial simulation tools to obtain the stress distribution of the power distribution network, the influence of the steady-state temperature field on electrothermal migration is considered. The steady-state temperature distribution of the power distribution network structure is calculated using commercial simulation tools such as the finite element method, and this is used as the input condition for the coupled simulation of electrothermal migration.

[0083] S3) Construct a training dataset suitable for stress prediction models based on datasets extracted from commercial tools.

[0084] S4) The stress prediction model is trained using the training dataset.

[0085] The above method is based on a directed graph and uses a graph attention network to obtain the higher-order features of nodes and edges, i.e., the spatial features of nodes and edges, which are also the spatial dependencies between interconnected units in the power distribution network. Based on the node stress sequence and edge stress sequence at each time point in the current time period, a Transformer network is used to obtain the initial node stress sequence and edge stress sequence at each time point in the future time period, i.e., the temporal features, which are also the temporal dependencies of the sequence. The higher-order features and the initial stress sequence are fused to obtain the final stress sequence at each time point in the future time period, realizing the prediction of the entire process of stress accumulation and providing a reliable basis for predicting the service life and failure time of the power distribution network.

[0086] See Figure 3 , Figure 3 This is a block diagram of a device for predicting the electrothermal stress of a power distribution network according to an embodiment of this application. The embodiment is a virtual device that can be loaded and executed by a computer device, which may include the aforementioned prediction device. Figure 3 The apparatus may include a construction module and a prediction module, which, when executing the above-described method for predicting the electrothermal stress of a power distribution network, can:

[0087] The construction module constructs a directed graph of the power distribution network based on its topology, node characteristics, and edge characteristics; where nodes are the points where wires intersect in the power distribution network, and edges are the wires connecting adjacent nodes.

[0088] It should be noted that in the construction module, node features include node temperature, and edge features include edge current density, discrete temperature distribution, length, and width.

[0089] The prediction module, based on the directed graph and the node stress sequences and edge stress sequences at each moment within the current time period, employs a stress prediction model to obtain the final node stress sequences and edge stress sequences at each moment within the future time period. Here, the current time period is the time interval from the initial moment to the current moment, and the future time period is the time interval from the next moment to the future moment. In the stress prediction model, based on the directed graph, higher-order features of nodes and edges are obtained. Based on the node stress sequences and edge stress sequences at each moment within the current time period, the initial node stress sequences and edge stress sequences at each moment within the future time period are obtained respectively. The initial node stress sequences and the higher-order features of the nodes are fused, and the final node stress sequence is obtained based on the fusion result. The initial edge stress sequences and the higher-order features of the edges are fused, and the final edge stress sequence is obtained based on the fusion result. The higher-order features of node A represent the influence of edges and other nodes on node A, and the higher-order features of edge B represent the influence of nodes and other edges on edge B.

[0090] It should be noted that the stress prediction model in the prediction module includes a graph attention network, a Transformer network, a gated fusion layer, and a linear transformation layer. The graph attention network is used to obtain the high-order features of nodes and edges. The Transformer network is used to obtain the initial node stress sequence and edge stress sequence at each time point in the future time period. The gated fusion layer is used to fuse the initial node stress sequence and the high-order features of nodes, and to fuse the initial edge stress sequence and the high-order features of edges. The linear transformation layer is used to obtain the final node stress sequence and the final edge stress sequence based on the fusion result.

[0091] In the prediction module, the fusion formula for the gated fusion layer is:

[0092] ;

[0093] In the formula, H represents the fusion result, and g represents the gating weight. σ is the Sigmoid activation function. These are the learnable weights and biases, H represents element-wise matrix multiplication, || is the concatenation operation, and H s For the initial nodal stress sequence or the initial edge stress sequence, H T These are the higher-order features of nodes or edges.

[0094] In the prediction module, the formula for the linear transformation layer is:

[0095] ;

[0096] In the formula, Here is the final predicted stress sequence, and H is the fusion result. These are the learnable weights and biases, respectively.

[0097] The aforementioned device is based on a directed graph and utilizes a graph attention network to obtain the higher-order features of nodes and edges, i.e., the spatial features of nodes and edges, which also represent the spatial dependencies between interconnected units in the power distribution network. Based on the node stress sequence and edge stress sequence at each moment in the current time period, a Transformer network is used to obtain the initial node stress sequence and edge stress sequence at each moment in the future time period, i.e., the temporal features, which also represent the temporal dependencies of the sequence. By fusing the higher-order features and the initial stress sequence, the final stress sequence at each moment in the future time period is obtained, realizing the prediction of the entire process of stress accumulation and providing a reliable basis for predicting the service life and failure time of the power distribution network.

[0098] This application also relates to a computer-readable storage medium that stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform a method for predicting the electrothermal stress of a power distribution network.

[0099] This application also relates to a computer device including one or more processors and one or more memories, wherein one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing a method for predicting the electrothermal stress of a power distribution network.

[0100] 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.

[0101] 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.

[0102] 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.

[0103] 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 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0104] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for predicting the electrothermal stress of a power distribution network, characterized in that, include: Based on the topology, node characteristics, and edge characteristics of the power distribution network, a directed graph of the power distribution network is constructed; where nodes are the points where wires intersect in the power distribution network, and edges are the wires connecting adjacent nodes. Based on the directed graph and the node stress sequences and edge stress sequences at each moment in the current time period, a stress prediction model is used to obtain the final node stress sequences and edge stress sequences at each moment in the future time period. The current time period is the time interval from the initial moment to the current moment, and the future time period is the time interval from the next moment to the future moment. In the stress prediction model, based on the directed graph, the higher-order features of the nodes and edges are obtained. Based on the node stress sequences and edge stress sequences at each moment in the current time period, the initial node stress sequences and edge stress sequences at each moment in the future time period are obtained respectively. The initial node stress sequences and the higher-order features of the nodes are fused, and the final node stress sequence is obtained based on the fusion result. The initial edge stress sequences and the higher-order features of the edges are fused, and the final edge stress sequence is obtained based on the fusion result. The higher-order features of node A represent the influence of edges and other nodes on node A, and the higher-order features of edge B represent the influence of nodes and other edges on edge B.

2. The method according to claim 1, characterized in that, Node features include node temperature, and edge features include edge current density, discrete temperature distribution, length, and width.

3. The method according to claim 1, characterized in that, The stress prediction model includes a graph attention network, a Transformer network, a gated fusion layer, and a linear transformation layer; Graph attention networks are used to obtain higher-order features of nodes and edges; Transformer networks are used to obtain the initial node stress sequence and edge stress sequence at each time point in the future time period; The gated fusion layer is used to fuse the initial node stress sequence and the higher-order features of the nodes, and to fuse the initial edge stress sequence and the higher-order features of the edges. The linear transformation layer is used to obtain the final nodal stress sequence and the final edge stress sequence based on the fusion result.

4. The method according to claim 3, characterized in that, In graph attention networks, node feature updates are performed using a node attention aggregator, with the following formula: ； In the formula, σ is the Sigmoid activation function. This represents the updated feature of the i-th node. Let represent the features of the i-th node, the j-th node, and the edge (i,j) before the update, respectively. (i,j) represents the edge between the i-th and j-th nodes, and || denotes the cascading operation. is the node attention coefficient, used to quantify the degree of influence of the features of the neighboring nodes of the i-th node on the features of the i-th node; Edge feature updates are performed using a bidirectional edge aggregator, with the following formula: ； In the formula, The features of the updated edge (i,j) These are the learnable weights and biases, respectively.

5. The method according to claim 3, characterized in that, In the Transformer network, a learnable linear projection is used to adaptively map the encoder's output to the decoder's temporal domain, while simultaneously generating synchronous positional codes as cues.

6. The method according to claim 3, characterized in that, The fusion formula for the gated fusion layer is: ； In the formula, H represents the fusion result, and g represents the gating weight. σ is the Sigmoid activation function. These are the learnable weights and biases, H represents element-wise matrix multiplication, || represents concatenation, and H represents the concatenation operation. s For the initial nodal stress sequence or the initial edge stress sequence, H T These are the higher-order features of nodes or edges.

7. The method according to claim 3, characterized in that, The formula for the linear transformation layer is: ； In the formula, Here is the final predicted stress sequence, and H is the fusion result. These are the learnable weights and biases, respectively.

8. A device for predicting the electrothermal stress of a power distribution network, characterized in that, include: The construction module constructs a directed graph of the power distribution network based on its topology, node characteristics, and edge characteristics; where nodes are the points where wires intersect in the power distribution network, and edges are the wires connecting adjacent nodes. The prediction module, based on the directed graph and the node stress sequences and edge stress sequences at each moment within the current time period, employs a stress prediction model to obtain the final node stress sequences and edge stress sequences at each moment within the future time period. Here, the current time period is the time interval from the initial moment to the current moment, and the future time period is the time interval from the next moment to the future moment. In the stress prediction model, based on the directed graph, higher-order features of nodes and edges are obtained. Based on the node stress sequences and edge stress sequences at each moment within the current time period, the initial node stress sequences and edge stress sequences at each moment within the future time period are obtained respectively. The initial node stress sequences and the higher-order features of the nodes are fused, and the final node stress sequence is obtained based on the fusion result. The initial edge stress sequences and the higher-order features of the edges are fused, and the final edge stress sequence is obtained based on the fusion result. The higher-order features of node A represent the influence of edges and other nodes on node A, and the higher-order features of edge B represent the influence of nodes and other edges on edge B.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method of any one of claims 1 to 7.

10. A computer device, characterized in that, include: One or more processors and one or more memories, one or more programs stored in one or more memories and configured to be executed by one or more processors, the one or more programs including instructions for performing the method of any one of claims 1 to 7.

Images