Power grid power flow generalization calculation method, system and device based on dynamic subgraph incremental reasoning and medium

Abstract

The present application relates to a power grid flow generalization calculation method, system, device and medium based on dynamic subgraph incremental reasoning, the method comprising: constructing a power grid graph structure and extracting the disturbance influence intensity of each node; according to the disturbance influence intensity, dividing the activated nodes, boundary nodes and frozen nodes and constructing a dynamic activated subgraph; constructing a physical admittance-aware edge gate propagation weight inside the dynamic activated subgraph, and performing local state reasoning on the nodes inside the activated subgraph; performing state decoupling on the activated area and the frozen area and completing preliminary reconstruction of the whole network; performing consistency correction on the boundary nodes and outputting the whole network power flow result. The present application realizes fast and accurate estimation of the whole network power flow state after disturbance without retraining the model for the new topology scene, by dynamically identifying the disturbance range under physical constraints, performing physical consistent local reasoning on the disturbed subgraph, splicing the local reasoning result with the stable area historical state, and performing boundary consistency correction.

Description

Technical Field

[0001] This invention relates to the field of power grid power flow calculation technology, and in particular to a power grid power flow generalization calculation method, system, device and medium based on dynamic subgraph incremental reasoning. Background Technology

[0002] Power flow calculation is a fundamental step in power system operation analysis, safety verification, fault assessment, and dispatch decision-making. Traditional power flow calculations typically employ numerical iterative methods such as the Newton-Raphson method and PQ decomposition to solve for a given network topology, node injected power, and branch parameters. However, in scenarios involving large-scale power grids, frequent topology changes, and rapid online analysis, these methods often require re-performing the entire network solution after each topology or operational change, resulting in significant repetitive computational overhead and making it difficult to balance computational accuracy and real-time performance.

[0003] In recent years, with the development of graph neural networks in non-Euclidean topology data modeling, existing technologies have begun to represent power grids as graph structures and utilize the connectivity and physical properties between nodes to predict power flow conditions. These existing technologies typically first construct power grid components such as buses and branches as nodes or edges in the graph. Then, parameters such as active power injection, reactive power injection, voltage magnitude, phase angle, and branch impedance and admittance are used as input features. Neighborhood feature aggregation is completed through graph convolution or message passing mechanisms, thereby establishing a mapping relationship between the power grid graph and the power flow state.

[0004] However, since power flow redistribution after grid faults or switching is essentially governed by electrical coupling relationships, and existing technologies mostly rely on unified network-wide solutions or unified graph-wide propagation for power flow calculation, they often suffer from insufficient adaptability, the need for additional retraining, or degraded generalization performance when facing dynamic changes in local topology. Specifically, when local disturbances occur in the power grid, such as line disconnections, line switching, switching actions, or local equipment shutdowns, existing technologies typically still need to re-execute the complete power flow solution or full graph neural network propagation based on the updated network model, failing to distinguish between disturbed and undisturbed areas. Therefore, in large-scale online rapid analysis scenarios of power grids, problems such as high overhead from repeated solutions and long response times easily arise. In addition, in existing graph neural network methods, the neighborhood propagation process often revolves around fixed connections or unified neighborhood aggregation, failing to fully utilize branch physical parameters to control the strength of local state propagation. Therefore, in scenarios of local topology abrupt changes, problems such as irrelevant information diffusion, distortion of boundary region states, or insufficient representation of local features easily occur. Summary of the Invention

[0005] (a) Technical problems to be solved In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a power flow generalization calculation method, system, device and medium based on dynamic subgraph incremental reasoning, which solves the technical problems of insufficient adaptability to new topology structures, serious reasoning redundancy, lack of physical parameter guidance for local propagation and difficulty in quickly restoring the power flow state of the whole network after disturbance in the scenario of dynamic change of power grid topology.

[0006] (II) Technical Solution To achieve the above objectives, the main technical solutions adopted by the present invention include: In a first aspect, embodiments of the present invention provide a power grid power flow generalization calculation method based on dynamic subgraph incremental reasoning, comprising: The acquired power grid data before and after the disturbance are used to construct a power grid graph structure, and the disturbance source's influence on each node in the power grid graph structure is extracted. Based on the distribution information of the intensity of disturbance, all nodes are divided into active nodes, boundary nodes and frozen nodes, and a dynamic activation subgraph is constructed based on the set of branches between active nodes and boundary nodes. After constructing the edge-gated propagation weights of physical admittance perception within the dynamically activated subgraph, local state reasoning is performed on the nodes within the activated subgraph based on the boundary reference information of the frozen node region. The prediction results of the dynamic activation subgraph obtained by local state reasoning are then spliced ​​with the historical states of the frozen node regions to obtain the preliminary reconstruction results of the entire network. Perform consistency correction on the boundary nodes, write the corrected boundary node states back to the preliminary reconstruction results of the entire network, and output the power flow distribution results of the entire network after the disturbance.

[0007] Optionally, constructing a power grid diagram structure from the acquired power grid data before and after the disturbance includes: Obtain the steady-state operating results of the power grid to be analyzed before the disturbance; Using a graph structure that includes a set of nodes and a set of branches, the steady-state operation results are constructed into a power grid graph structure before the disturbance. Based on the changes in the power grid topology caused by the disturbance source, the connection relationships or state parameters of the corresponding branches in the power grid diagram before the disturbance are updated to obtain the power grid diagram after the disturbance. Each node's attributes include node active power injection, node reactive power injection, node voltage amplitude, node voltage phase angle, node type identifier, and node feature vector representing the historical state or historical implicit representation of the previous steady state. Each branch's attributes include branch resistance, branch reactance, branch admittance, transformer turns ratio, and branch feature vector representing the branch's operating state.

[0008] Optionally, the intensity of the disturbance source's impact on each node in the power grid diagram structure is extracted, including: Based on the power grid structure before and after the disturbance, branches with disconnected lines, switched lines, or changed device states are identified and these branches are taken as disturbance sources. Based on the active power flow, reactive power flow, and admittance of the branch to which the disturbance source belongs before disconnection, obtain the disturbance source strength term; The equivalent electrical distance between all nodes in the network and the disturbance source is obtained by traversing the network. The equivalent electrical distance is the minimum value of the sum of the electrical weights of the branches on all candidate paths from the disturbance source to the node. The intensity of the disturbance source is calculated by exponentially decaying the disturbance source intensity term based on the equivalent electrical distance, thus obtaining the disturbance influence intensity of the disturbance source on each node.

[0009] Optionally, based on the distribution information of the disturbance influence intensity, all nodes are divided into active nodes, boundary nodes, and frozen nodes, and a dynamic activation subgraph is constructed based on the set of branches between active nodes and boundary nodes, including: Based on the distribution information of the disturbance impact intensity of all network nodes, the initially set node partitioning threshold or the node partitioning threshold set in the last disturbance is dynamically adjusted to the first node partitioning threshold and the second node partitioning threshold, with the first node partitioning threshold being greater than the second node partitioning threshold. Nodes whose disturbance intensity is not less than the first node partitioning threshold are classified as active nodes; Nodes whose disturbance influence intensity is not less than the second node partitioning threshold and less than the first node partitioning threshold are classified as boundary nodes. Nodes whose disturbance intensity is less than the second node partitioning threshold are classified as frozen nodes; Based on the active nodes and boundary nodes, and combined with the branch sets where the active nodes and boundary nodes are located, a dynamic activation subgraph is constructed, and the interface relationship between adjacent nodes in the boundary node and frozen node regions is preserved in the dynamic activation subgraph.

[0010] Optionally, after constructing the physical admittance-aware edge-gated propagation weights within the dynamically activated subgraph, local state inference is performed on the nodes within the activated subgraph based on the boundary reference information of the frozen node region, including: Within the dynamically activated subgraph, the electrical physical parameters and operating status parameters of each branch are mapped to the side-gated propagation weight of that branch, so that the strength of the role of different branches in message transmission is adjusted according to the actual electrical characteristics. During the node state update process, the node's own historical state characteristics before the disturbance, the propagation characteristics of adjacent nodes in the dynamically activated subgraph, the interface reference information transmitted from the frozen node region at the subgraph boundary, and the edge gating propagation weights corresponding to the branches are fused together to update the feature representations of activated nodes and boundary nodes. After multi-layer local propagation, the updated node features are mapped to local power flow state prediction results. The local power flow state prediction results include one or more of the following: node voltage magnitude, node voltage phase angle, branch active power flow, and branch reactive power flow.

[0011] Optionally, the predicted results of the dynamic activation subgraph obtained from local state inference are spliced ​​with the historical states of the frozen node regions to obtain the preliminary reconstruction results of the entire network, including: Obtain the prediction results of the nodes inside the dynamically activated subgraph after local state inference, as well as the historical state of the frozen node region before the disturbance; Based on the unique identifier of nodes in the power grid diagram structure, the prediction results of active nodes and boundary nodes are spliced ​​with the historical states of frozen nodes to obtain the preliminary reconstruction results of the entire network. Among them, fixed-point splicing involves writing local prediction results to the positions belonging to active nodes and boundary nodes, and directly writing historical steady-state values ​​to the positions belonging to frozen nodes, so as to transform the unified recalculation of the entire network into local incremental updates.

[0012] Optionally, consistency correction is performed on the boundary nodes, and the corrected boundary node states are written back to the preliminary reconstruction results of the entire network. The output of the perturbed power flow distribution results of the entire network includes: For each boundary node, construct a boundary compensation term between it and its adjacent nodes in the frozen node region; The local prediction results of the boundary nodes, the historical states of adjacent nodes in the frozen node region, and the boundary compensation term are fused to obtain the corrected boundary node states. The corrected boundary node states are written back to the corresponding boundary node positions in the preliminary reconstruction results of the entire network to obtain the final power flow distribution results of the entire network after disturbance.

[0013] Secondly, this embodiment also proposes a power grid power flow generalization calculation system based on dynamic subgraph incremental reasoning, including: The power grid diagram construction module is used to construct a power grid diagram structure from the acquired power grid data before and after the disturbance, and to extract the disturbance source's influence on each node in the power grid diagram structure. The dynamic subgraph partitioning module is used to divide all nodes into active nodes, boundary nodes, and frozen nodes based on the distribution information of the intensity of disturbance, and to construct a dynamic active subgraph based on the set of branches between active nodes and boundary nodes. The local state inference module is used to perform local state inference on nodes inside the active subgraph after constructing the edge-gated propagation weights of physical admittance perception inside the dynamically activated subgraph and based on the boundary reference information of the frozen node region. The fixed-point stitching and reconstruction module is used to stitch the prediction results of the dynamic activation subgraph obtained by local state reasoning with the historical state of the frozen node region to obtain the preliminary reconstruction results of the whole network. The boundary correction output module is used to perform consistency correction on the boundary nodes, write back the corrected boundary node states to the preliminary reconstruction results of the entire network, and output the power flow distribution results of the entire network after the disturbance.

[0014] Thirdly, this embodiment also proposes an electronic device, comprising: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, which are then executed by at least one processor to enable the at least one processor to perform the above-described generalized calculation method for power flow based on dynamic subgraph incremental reasoning.

[0015] Fourthly, this embodiment also proposes a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described generalized calculation method for power flow based on dynamic subgraph incremental reasoning.

[0016] (III) Beneficial Effects The beneficial effects of this invention are as follows: The power flow generalization calculation method based on dynamic subgraph incremental reasoning of this invention, by adopting a mechanism that dynamically identifies the disturbance range based on the strength of electrical coupling and performs local state deduction with physical admittance sensing, decoupling of the active region from the frozen state, and boundary consistency correction only within the dynamically activated subgraph, can transform the complex recalculation of the entire power flow into local incremental calculation within the disturbed dynamic subgraph range, significantly reducing the number of nodes and edges involved in the reasoning, and lowering computational complexity and time overhead. Simultaneously, by introducing edge-gated propagation weights guided by branch electrical physical parameters, it can enhance the modeling capability of the dominant channels for power flow redistribution, reduce the problem of feature oversmoothing and information mis-diffusion caused by local topological abrupt changes, thereby improving the accuracy of local state deduction under unknown topology change conditions. Furthermore, this invention does not require retraining the model for each new fault topology, nor does it require re-executing the complete network solution, thus achieving rapid estimation of the power flow state of the entire network after disturbance. Therefore, it has good adaptability to unknown topologies and engineering application value, and is particularly suitable for large-scale power grid online analysis and rapid post-fault assessment scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a power grid power flow generalization calculation method based on dynamic subgraph incremental reasoning, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of dynamic subgraph boundary delineation based on physical impedance potential energy attenuation according to an embodiment of the present invention. Figure 3 This is a schematic diagram of a physical admittance sensing local extrapolation mechanism provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the state decoupling and boundary consistency correction of the active and frozen regions according to an embodiment of the present invention. Figure 5 This is a schematic diagram of a boundary node interface provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of a local incremental inference process in a power grid topology disturbance scenario provided by an embodiment of the present invention. Detailed Implementation

[0018] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figures 1 to 6 As shown in the embodiment of the present invention, a power flow generalization calculation method based on dynamic subgraph incremental reasoning is proposed, comprising: constructing a power grid graph structure from the acquired power grid data before and after the disturbance, and extracting the disturbance influence intensity of the disturbance source on each node in the power grid graph structure; dividing all nodes into active nodes, boundary nodes, and frozen nodes according to the distribution information of the disturbance influence intensity, and constructing a dynamic active subgraph based on the branch set between active nodes and boundary nodes; after constructing the physical admittance-aware edge-gated propagation weights inside the dynamic active subgraph, performing local state reasoning on the nodes inside the active subgraph based on the boundary reference information of the frozen node region; performing fixed-point splicing of the prediction result of the dynamic active subgraph obtained by the local state reasoning with the historical state of the frozen node region to obtain the preliminary reconstruction result of the entire network; performing consistency correction on the boundary nodes, writing the corrected boundary node state back to the preliminary reconstruction result of the entire network, and outputting the power flow distribution result of the entire network after the disturbance.

[0020] This embodiment employs a network-wide recovery mechanism that dynamically identifies the affected area based on the strength of electrical coupling and performs physical admittance-sensing local state deduction, decoupling of the active region from the frozen state, and boundary consistency correction only within the dynamically activated subgraph. Compared to existing technologies, it transforms the complex recalculation of the entire network power flow into local incremental calculations within the affected dynamic subgraph, significantly reducing the number of nodes and edges involved in the inference, thus lowering computational complexity and time overhead. Simultaneously, by introducing edge-gated propagation weights guided by branch electrical physical parameters, it enhances the modeling capability for the dominant channels of power flow redistribution, reducing feature oversmoothing and information mis-diffusion problems caused by local topological abrupt changes, thereby improving the accuracy of local state deduction under unknown topology change conditions. Furthermore, this embodiment does not require retraining the model for each new fault topology or re-performing the complete network solution, enabling rapid estimation of the network power flow state after disturbance. Therefore, it has good adaptability to unknown topologies and engineering application value, and is particularly suitable for large-scale power grid online analysis and rapid post-fault assessment scenarios.

[0021] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0022] Specifically, refer to Figure 1 As shown, the power flow generalization calculation method based on dynamic subgraph incremental reasoning proposed in this embodiment may include the following steps S100 to S500: S100. Construct a power grid diagram structure from the power grid data before and after the disturbance, and extract the disturbance source's influence on each node in the power grid diagram structure.

[0023] In this embodiment, by constructing the power grid data before and after the disturbance into a power grid graph structure, a unified graph data input is formed for subsequent disturbance range identification and local state deduction. The specific steps may include the following sub-steps S111 to S113: S111. Obtain the steady-state operation results of the power grid to be analyzed before the disturbance.

[0024] S112. Using a graph structure that includes a set of nodes and a set of branches, the steady-state operation results are constructed into a power grid graph structure before the disturbance.

[0025] Furthermore, the power grid to be analyzed is represented as a graph structure G=(V,E), where V is the set of nodes and E is the set of branches. For any node i∈V, the node feature vector is constructed as follows: xi =[P i Q i U i ,θ i ,τ i ,h i ]; In the formula, P i Indicates active power injection at the node, Q i Indicates reactive power injection at the node, U i θ represents the node voltage magnitude. i τ represents the phase angle of the node voltage. i h represents the node type identifier. i It represents the historical state or historical latent representation of the previous steady state.

[0026] For any branch e ij ∈E, construct the branch feature vector: z ij =[r ij ,x ij ,y ij ,t ij ,s ij ]; In the formula, r ij Indicates the branch resistance, x ij Indicates branch reactance, y ij t represents the branch admittance. ij Indicates the transformer turns ratio, s ij This indicates the operating status of the branch circuit, which is preferably marked as either on or off.

[0027] S113. Based on the changes in the power grid topology caused by the disturbance source, update the connection relationships or state parameters of the corresponding branches in the power grid diagram structure before the disturbance to obtain the power grid diagram structure after the disturbance.

[0028] Furthermore, when a line disconnection, line switching, switch action, or transformer status change is detected, the connection relationship or status parameters of the corresponding branch are updated to obtain the disturbed power grid diagram G'=(V,E').

[0029] In this embodiment, after obtaining the disturbed power grid diagram, the first step is not to perform unified inference across the entire network, but rather to identify the actual impact range of the current topology disturbance, i.e., to calculate the disturbance impact intensity at each node based on the disturbance source and electrical distance. Specifically, the step of extracting the disturbance impact intensity at each node may include the following sub-steps S121 to S124: S121. Based on the power grid diagram structure before and after the disturbance, obtain the branches where the line is disconnected, the line is switched, or the device state changes, and take the branch as the disturbance source.

[0030] S122. Obtain the disturbance source strength term based on the active power flow, reactive power flow, and admittance of the branch to which the disturbance source belongs before disconnection.

[0031] Specifically, the disconnected line e f As a disturbance source, obtain the active power flow of the disturbance source branch before disconnection. Unproductive Tide and branch admittance amplitude First, the apparent power flow intensity Q of the disturbance source branch before disconnection is calculated based on the active and reactive power flow. f : ; Then, the disturbance source strength term is constructed by combining the apparent power flow strength and the branch admittance strength: ; In the formula, S base Let represent the baseline capacity, y represent the average value of the admittance of all branches in the network, and ε represent a minimal constant to prevent the denominator from being zero. and These represent the weighting coefficients corresponding to the apparent power flow intensity and admittance intensity, respectively. Using this calculation method, the disturbance source intensity term Q... f It also reflects the power flow load borne by the disturbance source branch before disconnection and the electrical coupling strength between the branch and the power grid.

[0032] S123. Traverse the network to obtain the equivalent electrical distance between all nodes and the disturbance source. The equivalent electrical distance is the minimum value of the sum of the electrical weights of the branches on all candidate paths from the disturbance source to the node.

[0033] Furthermore, to demonstrate that the propagation of disturbances in the power grid does not depend on the simple topology hop count, but rather on the combined effects of branch impedance, admittance, and turns ratio, this embodiment first calculates the equivalent electrical distance from the disturbance source to each node. For node i, its equivalent electrical distance is: ; In the formula, w represents the set of candidate paths from the disturbance source to node i. e This represents the electrical weight of a branch in the candidate path set. The electrical weight is: ; In the formula, ε is a minimal constant to prevent the denominator from being zero, and r e x is the branch resistance. e For branch reactance, τ e This refers to the transformer turns ratio or equivalent turns ratio system. This electrical weight is used to measure the equivalent electrical distance of disturbance propagation along the branch. The smaller the weight, the stronger the electrical coupling of the branch, and the easier it is for the disturbance to propagate along the branch.

[0034] S124. Based on the equivalent electrical distance, perform exponential decay calculation on the disturbance source strength term to obtain the disturbance influence intensity of the disturbance source on each node.

[0035] Furthermore, the formula for calculating the intensity of the disturbance effect is: ; In the formula, α is the attenuation coefficient.

[0036] S200. Based on the distribution information of the intensity of disturbance, all nodes are divided into active nodes, boundary nodes and frozen nodes, and a dynamic active subgraph is constructed based on the set of branches between active nodes and boundary nodes.

[0037] In this embodiment, to limit subsequent calculations to the truly disturbed local area and avoid repeatedly performing inference on undisturbed areas, the amount of redundant calculations across the entire network is reduced. (Reference) Figure 2 The process of identifying the disturbance range and defining the dynamic subgraph boundary based on physical impedance potential energy attenuation, as shown, first calculates the equivalent electrical distance from each node to the disturbance source based on the physical parameters of the disturbance source branch and related branches. Then, based on the disturbance influence intensity, all nodes in the network are divided into active nodes, boundary nodes, and frozen nodes, and a dynamic active subgraph and boundary interface set are constructed accordingly. Specifically, this step may include the following substeps S210 to S230: S210. Based on the distribution information of the disturbance impact intensity of all network nodes, dynamically adjust the initially set node division threshold or the node division threshold set in the last disturbance to the first node division threshold and the second node division threshold, wherein the first node division threshold is greater than the second node division threshold.

[0038] Furthermore, the initial node partitioning threshold can be adaptively determined by the mean and standard deviation of the network-wide disturbance influence intensity, or it can be pre-defined manually. This embodiment is based on two dynamically adjusted node partitioning thresholds: the first node partitioning threshold θ a The second node splitting threshold θ b Nodes can be divided into three categories: when node P i ≥θ a When this happens, the node is assigned to the set of active nodes V. a When θ b ≤P i <θ a If so, then the node is assigned to the boundary node set V. b When P i <θ b If the node is in this state, then it is assigned to the frozen node set V. f .

[0039] To further explain, based on the distribution information of the disturbance influence intensity of all network nodes corresponding to the current disturbance source, the threshold θ for dividing the first node is dynamically determined. a The second node splitting threshold θ b The first node partitioning threshold θ a Greater than the second node partitioning threshold θ b ; where the first node's partitioning threshold θ a The second node splitting threshold θ b The subgraph is dynamically generated based on the statistical distribution of the current disturbance intensity, or it is smoothly updated based on the statistical distribution of the current disturbance intensity and the node partitioning threshold corresponding to the previous disturbance. This enables the boundary of the dynamically activated subgraph to adapt to changes in the location of the disturbance source, the electrical parameters of the branch, and the disturbance propagation intensity.

[0040] ; ; ; In the formula, This indicates the intensity of the disturbance effect at node i. This represents the average intensity of disturbances affecting all nodes in the network. This represents the standard deviation of the disturbance's influence. and This is the threshold adjustment coefficient, and .

[0041] S220a. Nodes whose disturbance influence intensity is not less than the first node division threshold are classified as active nodes.

[0042] S220b: Nodes whose disturbance influence intensity is not less than the second node partitioning threshold and less than the first node partitioning threshold are classified as boundary nodes.

[0043] S220c: Nodes whose disturbance impact intensity is less than the second node division threshold are classified as frozen nodes.

[0044] S230. Based on the active node and the boundary node, and combined with the branch set where the active node and the boundary node are located, construct a dynamic activation subgraph, and retain the interface relationship between adjacent nodes in the boundary node and frozen node region in the dynamic activation subgraph.

[0045] Furthermore, the set of activated nodes V is obtained. a Boundary node set V b and the set of frozen nodes V f And combine the set of branches E corresponding to the active nodes and the boundary nodes. a This forms a dynamic activation subgraph. At the same time, the interface relationship between the boundary node and its three neighboring nodes is preserved for subsequent boundary consistency processing.

[0046] S300. After constructing the edge-gated propagation weights of physical admittance perception within the dynamic activation subgraph, local state reasoning is performed on the nodes within the activation subgraph based on the boundary reference information of the frozen node region.

[0047] In this embodiment, after completing the dynamic activation subgraph segmentation, local state deduction is performed only within the dynamic activation subgraph. Simultaneously, to improve the consistency between the local deduction process and the physical laws of the power grid, this embodiment explicitly incorporates electrical physical parameters and operating state parameters into the neighbor node information aggregation process during graph convolution or graph message passing. This allows the strength of different branches' roles in message passing to be adjusted according to actual electrical characteristics, thereby avoiding the error propagation problem caused by the unified propagation based solely on static connection relationships in traditional graph models. Specifically, refer to... Figure 3 As shown, step S300 may include the following sub-steps S310 to S330: S310. Within the dynamically activated subgraph, the electrical physical parameters and operating status parameters of each branch are mapped to the side-gating propagation weight of that branch, so that the strength of the role of different branches in message transmission is adjusted according to the actual electrical characteristics.

[0048] Furthermore, during graph convolution or graph message passing, key physical parameters such as branch resistance, reactance, admittance, transformer turns ratio, and switching state are explicitly introduced into the neighbor node information aggregation process. That is, branches are no longer simply treated as ordinary connecting edges, but are assigned propagation gating weights reflecting their electrical coupling capabilities, thereby controlling the contribution of different neighbor information in local propagation. The gating weight of the branch between node i and node j can be expressed as: ; In the formula, r ij Indicates the branch resistance, x ij Indicates branch reactance, y ij t represents the branch admittance. ij Indicates the transformer turns ratio, s ij Indicates the operating status of the branch line. This represents a parameter mapping function. This represents the normalized activation function.

[0049] S320. During the node state update process, the node's own historical state characteristics before the disturbance, the propagation characteristics of adjacent nodes in the dynamically activated subgraph, the interface reference information transmitted from the frozen node region at the subgraph boundary, and the edge gating propagation weights corresponding to the branches are fused together to update the feature representations of the activated nodes and boundary nodes.

[0050] Furthermore, during the node state update process, the current layer features of a node not only depend on its own historical state features before the perturbation, but also simultaneously integrate the propagation features of adjacent nodes within the activated subgraph, the interface reference information transmitted from the frozen region at the subgraph boundary, and the edge-gated propagation weights corresponding to the branches. The local update process of node i can be represented as: ; In the formula, Indicates that node i is at the 1st rank. l Layer feature representation, W represents the set of neighbors of node i in the dynamic activation subgraph. s and W n Let W represent the self-loop transformation matrix and the neighbor transformation matrix, respectively. b Let c represent the boundary reference mapping matrix. i This represents the boundary reference information transmitted from the frozen region. This represents a non-linear activation function.

[0051] S330. After multi-layer local propagation, the updated node features are mapped to local power flow state prediction results. The local power flow state prediction results include one or more of the following: node voltage amplitude, node voltage phase angle, branch active power flow, and branch reactive power flow.

[0052] Furthermore, after L layers of local propagation, the final hidden representations of the activated nodes and boundary nodes are obtained. The local power flow state prediction results are obtained through output mapping: ; In the formula, f out (·) is the output mapping function, which can be a linear mapping, a multilayer perceptron, or other differentiable mapping structures. It is used to map the node implicit representation after multilayer local propagation to the local power flow state prediction result. It may include node voltage magnitude, node voltage phase angle, and, if necessary, one or more of branch active power flow and branch reactive power flow.

[0053] To further explain, in multi-layer local propagation, node features are reweighted and propagated along the low equivalent impedance dominant channel, thereby enhancing the modeling ability of the dominant path of power flow redistribution and suppressing the mis-diffusion of irrelevant neighborhood information after local topological abrupt changes. Figure 3 The implementation process of physical admittance sensing local extrapolation in this embodiment can be illustrated. Inside the dynamically activated subgraph, edge-level gating weights are constructed based on branch resistance, reactance, admittance, transformer ratio, and switching state. Combined with the boundary reference information transmitted from the frozen region, reweighted local state propagation is performed on the activated nodes and boundary nodes to obtain the local power flow state prediction result.

[0054] S400. The prediction results of the dynamic activation subgraph obtained by local state reasoning are spliced ​​with the historical states of the frozen node regions to obtain the preliminary reconstruction results of the entire network.

[0055] In this embodiment, after obtaining the local inference results of the dynamically activated subgraph, the local results are restored to the overall network power flow results through an activation and freeze state decoupling mechanism. This transforms the problem of uniformly updating all nodes across the network into incremental updates only performed on nodes within the dynamic subgraph, while other nodes directly reuse their historical states, thereby significantly reducing the inference scope and redundant computation overhead. (Reference) Figure 4 As shown, specific step S400 may include the following sub-steps S410 to S420: S410. Obtain the prediction results of the nodes inside the dynamically activated subgraph after local state inference, as well as the historical state of the frozen node region before the disturbance.

[0056] S420. Based on the unique identifier information of the node in the power grid diagram structure, the prediction results of the active nodes and boundary nodes are spliced ​​with the historical states of the frozen nodes to obtain the preliminary reconstruction results of the entire network.

[0057] Furthermore, the prediction results for the activated region and the boundary region are denoted as... The historical state of the frozen area is recorded as Then the initial reconstruction of the entire network state can be expressed as: ; In the formula, This refers to a fixed-point splicing operation based on the unique identifier information of the node number, topology index, or network-wide mapping relationship. Specifically, it writes local prediction results to the positions belonging to active nodes and boundary nodes, and directly writes historical steady-state values ​​to the positions belonging to frozen nodes, so as to transform the unified recalculation of the entire network into local incremental updates.

[0058] S500 performs consistency correction on the boundary nodes, writes the corrected boundary node states back to the initial reconstruction results of the entire network, and outputs the power flow distribution results of the entire network after the disturbance.

[0059] In this embodiment, considering that the boundary nodes are electrically coupled to both the active and frozen regions, direct splicing may result in state discontinuities or local distortions at the interface. Therefore, this embodiment further performs consistency correction on the boundary nodes during the global reconstruction phase. After completing the boundary consistency correction, the correction results of the boundary nodes are written back to the overall network state vector to obtain the final power flow distribution result after the disturbance. Specifically, refer to... Figure 4 As shown, step S500 may include S510 to S530: S510. For a boundary node, construct a boundary compensation term between it and its adjacent nodes in the frozen node region.

[0060] Furthermore, for any boundary node i, obtain the set of frozen nodes adjacent to it. And based on the electrical weights of the branches between boundary node i and frozen node j w ij Calculate the boundary reference weights of frozen node j relative to the boundary nodes: ; In the formula, κ represents the temperature regulation coefficient, and ε is a minimal constant to prevent the denominator from being zero. (Electrical weight) w ij The smaller the value, the stronger the electrical coupling between boundary node i and frozen node j, and the greater the reference value of the historical state of frozen node j for boundary node i.

[0061] Based on the aforementioned boundary reference weights, the historical states of adjacent nodes in the frozen node region are weighted and aggregated to obtain the historical reference states transmitted from the frozen region to boundary node i. : ; in, T represents the historical state of node j before the disturbance. ij This represents the interface mapping matrix constructed based on the branch resistance, reactance, admittance, turns ratio, and operating state between boundary node i and frozen node j; when the state variables of the boundary node and the frozen node have the same dimension, T ij Available unit mapping.

[0062] Therefore, in this embodiment, the difference between the historical reference state of the frozen region transmitted to boundary node i and the local prediction result of boundary node i can be used as the boundary compensation term: ; In the formula, This represents the local prediction result obtained by reasoning about the local state of the boundary node i through the dynamically activated subgraph. This represents the boundary compensation term for boundary node i.

[0063] S520. The local prediction results of the boundary nodes, the historical states of adjacent nodes in the frozen node region, and the boundary compensation terms are fused to obtain the corrected boundary node states.

[0064] Furthermore, the local prediction results of the boundary nodes, the historical states of adjacent nodes in the frozen node region, and the boundary compensation term are fused to obtain a smoother interface state that better conforms to physical constraints. This interface state exhibits boundary consistency. A schematic diagram of the boundary node interface is shown below. Figure 5As shown. The corrected boundary node state. It can be represented as: ; In the formula, This represents the local prediction result obtained by reasoning about the local state of the boundary node i through the dynamically activated subgraph. This represents the boundary compensation term constructed based on the historical states of adjacent nodes in the frozen node region. , representing the fusion coefficient.

[0065] Because the boundary compensation term in step S510 satisfies: ; Therefore, the above correction process can also be equivalently expressed as: ; in, This represents the historical reference state obtained in step S510 by weighted aggregation of the historical states of adjacent nodes in the frozen node region. Through the above fusion method, the corrected boundary node state simultaneously includes the local prediction results of the dynamic activated subgraph and the historical state reference information of the frozen node region, thereby improving the boundary consistency between the activated region and the frozen region.

[0066] S530. Write the corrected boundary node state back to the corresponding boundary node position in the preliminary reconstruction result of the whole network to obtain the final power flow distribution result of the whole network after disturbance.

[0067] Furthermore, after completing the boundary consistency correction, the correction results of the boundary nodes are written back to the entire network state vector to obtain the final network power flow result: ; In the formula, This indicates a write-back update operation performed on the boundary node location. The overall network power flow results include, but are not limited to, one or more of the following: node voltage magnitude, node voltage phase angle, branch active power flow, and branch reactive power flow.

[0068] To further explain, through the above steps, this embodiment achieves rapid recovery of the power flow state of the entire network after disturbance without repeatedly solving the entire network. Figure 5 This embodiment illustrates the process of state decoupling, fixed-point splicing, and boundary consistency correction between the active and frozen regions. Specifically, the local prediction results of the active and boundary regions are first spliced ​​with the historical steady-state state of the frozen region to form a preliminary reconstruction result of the entire network. Then, considering the characteristic that the boundary nodes are electrically coupled to both the active and frozen regions, consistency compensation and fusion processing are performed on the boundary interfaces to finally obtain the perturbed power flow distribution result of the entire network.

[0069] In one specific embodiment, such as Figure 6 As shown, taking the scenario of a transmission network experiencing a line N-1 disconnection as an example, the disconnected branch is first considered as the disturbance source. A post-disturbance network graph is constructed based on the steady-state results before the disturbance, the post-disturbance topology, and the physical parameters of nodes and branches. Then, based on the equivalent electrical distance between the disturbance source and each node and the intensity of the disturbance, all nodes in the network are classified into three categories: activated, boundary, and frozen, constructing a dynamic activation subgraph. Within this dynamic activation subgraph, local state propagation using physical admittance sensing is performed, and boundary reference information is constructed by combining the historical states of the frozen regions. Finally, the local prediction results are spliced ​​with the historical states of the frozen regions at fixed points, and consistency correction is performed at the boundary interface to obtain the post-disturbance network power flow results. Compared with existing technologies that re-perform power flow calculations or full-graph neural network propagation after each topology change, this embodiment significantly reduces the number of nodes and branches involved in the calculation, improves online analysis efficiency, and maintains high result accuracy and adaptability to unknown topology disturbances.

[0070] Furthermore, this embodiment also proposes a power grid power flow generalization calculation system based on dynamic subgraph incremental reasoning, including: The power grid diagram construction module is used to construct a power grid diagram structure from the acquired power grid data before and after the disturbance, and to extract the disturbance source's influence on each node in the power grid diagram structure.

[0071] The dynamic subgraph partitioning module is used to divide all nodes into active nodes, boundary nodes, and frozen nodes based on the distribution information of the intensity of disturbance influence, and to construct a dynamic active subgraph based on the set of branches between active nodes and boundary nodes.

[0072] The local state inference module is used to perform local state inference on nodes inside the active subgraph after constructing the edge-gated propagation weights of physical admittance perception within the dynamically activated subgraph and based on the boundary reference information of the frozen node region.

[0073] The fixed-point stitching and reconstruction module is used to stitch the predicted results of the dynamic activation subgraph obtained from local state reasoning with the historical states of the frozen node region to obtain the preliminary reconstruction results of the entire network.

[0074] The boundary correction output module is used to perform consistency correction on the boundary nodes, write back the corrected boundary node states to the preliminary reconstruction results of the entire network, and output the power flow distribution results of the entire network after the disturbance.

[0075] Furthermore, this embodiment also proposes an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the above-described generalized calculation method for power flow based on dynamic subgraph incremental reasoning.

[0076] Finally, this embodiment also proposes a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described generalized calculation method for power flow based on dynamic subgraph incremental reasoning.

[0077] In summary, this embodiment proposes a power flow generalization calculation method, system, device, and medium based on dynamic subgraph incremental reasoning. It takes the steady-state operation results of the power grid before the disturbance, the topology of the power grid after the disturbance, and the physical parameters of nodes and branches as basic inputs. It unfolds around the processing link of disturbance range identification, local state inference, and global result reconstruction, transforming the traditional unified power flow recalculation oriented towards the whole network into a local incremental reasoning process oriented towards the disturbed dynamic subgraph. Its core lies in two aspects: First, using branches that experience disconnection, switching, or state changes as disturbance sources, and combining physical parameters such as branch resistance, reactance, turns ratio, and admittance, a characterization method is constructed to show the attenuation of disturbance influence intensity with electrical distance. Based on this, all network nodes are divided into active nodes, boundary nodes, and frozen nodes, forming a disturbed dynamic subgraph. This transforms the process of recalculating nodes that truly require recalculation from empirical local pruning to a dynamic determination process based on the strength of electrical coupling, improving the targeting of disturbed area identification. Second, within the active subgraph, the branch electrical physical parameters and operating state parameters are mapped to edge-level gating weights, enabling the propagation effect of different branches in node state updates to be recalculated based on the actual strength of electrical coupling. Weighted control is used to enhance the modeling ability of the dominant channels for power flow redistribution and suppress the mis-propagation of irrelevant information after local topological mutations. Third, incremental updates are performed only on active nodes and boundary nodes within the subgraph, while the historical steady-state characteristics of unperturbed or weakly perturbed frozen regions are directly preserved, transforming the original whole-network recalculation process into a local incremental calculation process, significantly compressing the range of nodes and branches involved in inference. Fourth, a global reconstruction method is established between the update results of the active region and the historical state of the frozen region. Through fixed-point splicing and boundary consistency compensation and fusion processing, the potential state discontinuity problem at the interface is corrected, and the power flow distribution of the whole network after disturbance is restored without repeating the solution for the whole network.

[0078] Due to the adoption of the above technical solutions, this embodiment has the following beneficial effects compared with the prior art: First, the dynamic subgraph boundary defined by the disturbance influence characterization based on physical impedance potential energy attenuation is more consistent with the actual disturbance propagation law, which can reduce the misinclusion of irrelevant nodes in the reasoning process and improve the accuracy of disturbed area identification and the pertinence of local solution; Second, through the edge-level gating weight propagation mechanism of physical admittance sensing, the modeling ability of the dominant path of power flow redistribution can be strengthened, reducing the problems of feature oversmoothing, information mis-diffusion and boundary region state estimation distortion caused by local topological changes, and improving the accuracy of local state inference under unknown topological change conditions; Third, through activation, freezing The state decoupling mechanism transforms the complex recalculation of the entire network power flow into local incremental calculations within the scope of the disturbed dynamic subgraph. This significantly reduces the number of nodes and edges involved in inference, lowering computational complexity and time overhead. It is particularly suitable for large-scale power grid real-time analysis, streaming topology change processing, and rapid safety verification scenarios. Finally, through fixed-point splicing and boundary consistency correction mechanisms, it can achieve rapid estimation of the power flow state of the entire network after disturbance without retraining the model for each new fault topology or re-executing the complete network solution. It has good adaptability to unknown topologies and engineering application value, making it suitable for large-scale power grid online analysis and rapid post-fault assessment scenarios.

[0079] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0080] 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.

[0081] 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, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0082] It should be noted that in the description of this invention, the word "a" or "an" preceding a component does not exclude the existence of multiple such components. This invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. The use of terms such as first, second, third, etc., is merely for convenience and does not indicate any order. These terms can be understood as part of the component names.

[0083] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0084] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning of the basic inventive concept, can make other changes and modifications to these embodiments.

[0085] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of the invention.

Claims

1. A generalized calculation method for power grid power flow based on dynamic subgraph incremental reasoning, characterized in that, include: The acquired power grid data before and after the disturbance are used to construct a power grid graph structure, and the disturbance source's influence on each node in the power grid graph structure is extracted. Based on the distribution information of the intensity of disturbance, all nodes are divided into active nodes, boundary nodes and frozen nodes, and a dynamic activation subgraph is constructed based on the set of branches between active nodes and boundary nodes. After constructing the edge-gated propagation weights of physical admittance perception within the dynamically activated subgraph, local state reasoning is performed on the nodes within the activated subgraph based on the boundary reference information of the frozen node region. The prediction results of the dynamic activation subgraph obtained by local state reasoning are then spliced ​​with the historical states of the frozen node regions to obtain the preliminary reconstruction results of the entire network. Perform consistency correction on the boundary nodes, write the corrected boundary node states back to the preliminary reconstruction results of the entire network, and output the power flow distribution results of the entire network after the disturbance.

2. The method as described in claim 1, characterized in that, The process of constructing a power grid diagram from the acquired power grid data before and after the disturbance includes: Obtain the steady-state operating results of the power grid to be analyzed before the disturbance; Using a graph structure that includes a set of nodes and a set of branches, the steady-state operation results are constructed into a power grid graph structure before the disturbance. Based on the changes in the power grid topology caused by the disturbance source, the connection relationships or state parameters of the corresponding branches in the power grid diagram before the disturbance are updated to obtain the power grid diagram after the disturbance. Each node's attributes include node active power injection, node reactive power injection, node voltage amplitude, node voltage phase angle, node type identifier, and node feature vector representing the historical state or historical implicit representation of the previous steady state. Each branch's attributes include branch resistance, branch reactance, branch admittance, transformer turns ratio, and branch feature vector representing the branch's operating state.

3. The method as described in claim 1, characterized in that, The intensity of the disturbance source's impact on each node in the power grid diagram structure is extracted, including: Based on the power grid structure before and after the disturbance, branches with disconnected lines, switched lines, or changed device states are identified and these branches are taken as disturbance sources. Based on the active power flow, reactive power flow, and admittance of the branch to which the disturbance source belongs before disconnection, obtain the disturbance source strength term; The equivalent electrical distance between all nodes in the network and the disturbance source is obtained by traversing the network. The equivalent electrical distance is the minimum value of the sum of the electrical weights of the branches on all candidate paths from the disturbance source to the node. The intensity of the disturbance source is calculated by exponentially decaying the disturbance source intensity term based on the equivalent electrical distance, thus obtaining the disturbance influence intensity of the disturbance source on each node.

4. The method as described in claim 1, characterized in that, Based on the distribution information of the disturbance influence intensity, all nodes are divided into active nodes, boundary nodes, and frozen nodes. A dynamic activation subgraph is constructed based on the set of branches between active nodes and boundary nodes, including: Based on the distribution information of the disturbance impact intensity of all network nodes, the initially set node partitioning threshold or the node partitioning threshold set in the last disturbance is dynamically adjusted to the first node partitioning threshold and the second node partitioning threshold, with the first node partitioning threshold being greater than the second node partitioning threshold. Nodes whose disturbance intensity is not less than the first node partitioning threshold are classified as active nodes; Nodes whose disturbance influence intensity is not less than the second node partitioning threshold and less than the first node partitioning threshold are classified as boundary nodes. Nodes whose disturbance intensity is less than the second node partitioning threshold are classified as frozen nodes; Based on the active nodes and boundary nodes, and combined with the branch sets where the active nodes and boundary nodes are located, a dynamic activation subgraph is constructed, and the interface relationship between adjacent nodes in the boundary node and frozen node regions is preserved in the dynamic activation subgraph.

5. The method as described in claim 1, characterized in that, After constructing edge-gated propagation weights for physical admittance sensing within the dynamically activated subgraph, local state inference is performed on the nodes within the activated subgraph based on the boundary reference information of the frozen node region, including: Within the dynamically activated subgraph, the electrical physical parameters and operating status parameters of each branch are mapped to the side-gated propagation weight of that branch, so that the strength of the role of different branches in message transmission is adjusted according to the actual electrical characteristics. During the node state update process, the node's own historical state characteristics before the disturbance, the propagation characteristics of adjacent nodes in the dynamically activated subgraph, the interface reference information transmitted from the frozen node region at the subgraph boundary, and the edge gating propagation weights corresponding to the branches are fused together to update the feature representations of activated nodes and boundary nodes. After multi-layer local propagation, the updated node features are mapped to local power flow state prediction results. The local power flow state prediction results include one or more of the following: node voltage magnitude, node voltage phase angle, branch active power flow, and branch reactive power flow.

6. The method as described in claim 1, characterized in that, The predicted results of the dynamic activation subgraph obtained from local state inference are then spliced ​​with the historical states of the frozen node regions to obtain the preliminary reconstruction results of the entire network, including: Obtain the prediction results of the nodes inside the dynamically activated subgraph after local state inference, as well as the historical state of the frozen node region before the disturbance; Based on the unique identifier of nodes in the power grid diagram structure, the prediction results of active nodes and boundary nodes are spliced ​​with the historical states of frozen nodes to obtain the preliminary reconstruction results of the entire network. Among them, fixed-point splicing involves writing local prediction results to the positions belonging to active nodes and boundary nodes, and directly writing historical steady-state values ​​to the positions belonging to frozen nodes, so as to transform the unified recalculation of the entire network into local incremental updates.

7. The method as described in claim 1, characterized in that, Perform consistency correction on the boundary nodes, write the corrected boundary node states back to the initial reconstruction results of the entire network, and output the perturbed power flow distribution results of the entire network, including: For each boundary node, construct a boundary compensation term between it and its adjacent nodes in the frozen node region; The local prediction results of the boundary nodes, the historical states of adjacent nodes in the frozen node region, and the boundary compensation term are fused to obtain the corrected boundary node states. The corrected boundary node states are written back to the corresponding boundary node positions in the preliminary reconstruction results of the entire network to obtain the final power flow distribution results of the entire network after disturbance.

8. A power grid power flow generalization calculation system based on dynamic subgraph incremental reasoning, characterized in that, include: The power grid diagram construction module is used to construct a power grid diagram structure from the acquired power grid data before and after the disturbance, and to extract the disturbance source's influence on each node in the power grid diagram structure. The dynamic subgraph partitioning module is used to divide all nodes into active nodes, boundary nodes, and frozen nodes based on the distribution information of the intensity of disturbance, and to construct a dynamic active subgraph based on the set of branches between active nodes and boundary nodes. The local state inference module is used to perform local state inference on nodes inside the active subgraph after constructing the edge-gated propagation weights of physical admittance perception inside the dynamically activated subgraph and based on the boundary reference information of the frozen node region. The fixed-point stitching and reconstruction module is used to stitch the prediction results of the dynamic activation subgraph obtained by local state reasoning with the historical state of the frozen node region to obtain the preliminary reconstruction results of the whole network. The boundary correction output module is used to perform consistency correction on the boundary nodes, write back the corrected boundary node states to the preliminary reconstruction results of the entire network, and output the power flow distribution results of the entire network after the disturbance.

9. An electronic device, characterized in that, include: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, which are executed by at least one processor to enable the at least one processor to perform a power flow generalization calculation method based on dynamic subgraph incremental reasoning as described in any one of claims 1-7.

10. A computer-readable storage medium storing computer-executable instructions thereon, characterized in that, When the computer-executable instructions are executed by the processor, they implement the power flow generalization calculation method based on dynamic subgraph incremental reasoning as described in any one of claims 1-7.

Images