A power distribution network dynamic reconfiguration method and device for fast failure recovery

Abstract

The application provides a power distribution network dynamic reconstruction method and device for fault rapid recovery, and relates to the technical field of intelligent power distribution network fault recovery and topology control. The method is a power distribution network fault rapid recovery method for the operation level, which combines an attention mechanism graph neural network and model predictive control (MPC), starts from the current operation state, realizes dynamic topology reconstruction of the power distribution network under fault through graph structure modeling, candidate reconstruction topology structure scheme generation, rolling optimization scheduling and closed-loop feedback control, realizes efficient dynamic recovery of the fault area, solves the limitation of response lag and poor recovery path quality of the traditional static reconstruction scheme, and significantly improves the self-healing ability and recovery efficiency of the power grid system.

Description

Technical Field

[0001] This application relates to the field of smart distribution network fault recovery and topology control technology, and in particular to a method and apparatus for dynamic reconfiguration of distribution networks for rapid fault recovery. Background Technology

[0002] In traditional distribution network operation, reconfiguration strategies are one of the key means to cope with abnormal conditions such as line faults, load migration, and voltage over-limits. However, existing technologies still have significant shortcomings in several aspects, specifically in the following areas:

[0003] 1. Existing reconfiguration strategies are slow to respond, and topology adjustments rely on static rules or manual decisions. Currently, most distribution network reconfiguration methods still rely on fixed sets of candidate topologies or manually preset "reconfiguration tables." When a fault occurs, switching operations are triggered by looking up the table or rules. This approach can cope with single-branch faults or minor local disturbances, but in scenarios with multiple faults, abrupt topology changes, or drastic load changes, the lack of real-time assessment and adaptive capabilities leads to severe response lag, easily resulting in critical loads not being restored in time or prolonged power outages in localized areas.

[0004] 2. Static optimization methods struggle to handle the dynamics and uncertainties of real-world operation. Traditional reconfiguration methods typically rely on static optimization modeling, such as generating multiple schemes based on power flow calculations and then evaluating their merits. However, these schemes are constructed based on pre-fault or pre-defined scenarios. In actual operation, parameters such as node voltage, current load, and DG (Distributed Generation) output all change in real time. Once the environment shifts, these static schemes often become suboptimal or even infeasible, failing to meet the self-healing control requirements of the distribution network for "monitoring, controlling, and repairing simultaneously."

[0005] 3. Combinatorial explosion in multi-point failure scenarios: Traditional methods suffer from low computational efficiency. In large-scale complex networks, after a multi-point failure, it is necessary to simultaneously consider the interruption of multiple branches, feasible combinations of switches, and the matching of DG support capabilities, resulting in an exponentially growing combination space. Taking the IEEE 123-node system as an example, there are more than a hundred controllable switches and DG points. Traditional optimization methods based on enumeration or mixed integer programming are difficult to find feasible solutions in a short time, severely limiting computational efficiency and failing to meet the "second-level" control response requirements of practical deployment scenarios.

[0006] 4. DG and load regulation capabilities have not been effectively integrated into reconfiguration schemes. In recent years, a large number of DG units (such as photovoltaic, wind power, hydropower, and micro gas turbines) have been connected to the distribution network, possessing a certain degree of output regulation capability. However, existing reconfiguration strategies mostly focus on switching "structural paths," failing to consider the flexibility of DG as an important input variable for optimization decisions. Simultaneously, the load-side response capabilities (such as load priority, interruptibility, and dynamic recovery willingness) have not been modeled or involved in regulation, resulting in overly simplistic reconfiguration strategies and scheduling strategies that fail to fully unleash the system's flexibility potential.

[0007] 5. Lack of joint strategies integrating AI (Artificial Intelligence) and optimization control, resulting in insufficient intelligence. Although intelligent algorithms such as genetic algorithms and particle swarm optimization have been attempted for reconstruction scheme search in recent years, most remain at the offline solution level, lacking integration with real-time state awareness, system operation constraints, and predictive repair capabilities. Summary of the Invention

[0008] In view of the above problems, this application is proposed to provide a dynamic reconfiguration method, device, electronic equipment, and storage medium for distribution networks oriented towards rapid fault recovery, which overcomes or at least partially solves the above problems. By introducing artificial intelligence and optimized control mechanisms, a dynamic, rolling, and adaptive reconfiguration decision-making method is realized to solve the technical bottlenecks of traditional methods, such as response lag, non-optimal reconfiguration paths, and low recovery efficiency, thereby improving the rapid recovery capability of the distribution network under sudden faults and the self-healing performance of the power grid system. The technical solution is as follows:

[0009] Firstly, a dynamic reconfiguration method for distribution networks oriented towards rapid fault recovery is provided, the method comprising:

[0010] After a fault occurs in the distribution network, obtain the current network structure, operating status, and available resources;

[0011] Based on the current network structure, operating status, and available resources, a graph modeling method is adopted to represent the distribution network as a graph structure with attributes. The node set in the graph structure includes buses, load nodes, and distributed generation (DG) nodes; the edge set includes distribution lines, switches, and tie lines; the attribute characteristics of each node in the node set include one or more of the following: voltage, current, load power, connected DG output, DG type, and power supply status; the attribute characteristics of each edge in the edge set include one or more of the following: branch impedance, maximum allowable current, switch status, and whether it is faulty.

[0012] Input the attributed graph structure into a pre-trained attention mechanism graph neural network, and output multiple candidate reconfigured topologies that satisfy the physical and structural constraints of the power distribution network;

[0013] With the dual objectives of maximizing load recovery and minimizing reconfiguration cost, a predictive optimization model in the rolling time domain is constructed. Model predictive control (MPC) is used to make multi-time-step optimization decisions on multiple candidate reconfiguration topologies, apply preset constraints, and use a mathematical optimization solver to calculate the optimal reconfiguration path sequence. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operation status is updated in real time, thereby updating the attributed graph structure.

[0014] By utilizing the updated attributed graph structure, the next round of input is a pre-trained attention-based graph neural network, which outputs multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decision-making is performed on the multiple candidate reconfiguration topologies using MPC, and preset constraints are applied. The optimal reconfiguration path sequence is calculated using a mathematical optimization solver. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The next round of input is then performed on the pre-trained attention-based graph neural network, and so on, until the preset termination condition is met, thus realizing the dynamic reconfiguration of the distribution network.

[0015] In one possible implementation, the attributed graph structure is as follows: , ,in, For a set of nodes, For edge set, For nodes The attributes and characteristics, For the edge The attribute characteristics, the edges here Also known as a branch road ;

[0016] (1)

[0017] In equation (1), For nodes The voltage; For nodes The load power; For nodes The output power of the connected DG; For nodes DG type; For nodes The power supply status is indicated by 1 for normal operation and 0 for power failure.

[0018] (2)

[0019] In equation (2), For the edge Branch impedance; For the edge Maximum allowable current; For the edge The switch status is 1 for closed and 0 for open; For the edge A flag indicating whether a fault exists;

[0020] The connectivity relationships in a graph structure are determined by an adjacency matrix. express:

[0021] (3)

[0022] This adjacency matrix will serve as the core structural input in the pre-trained attention mechanism graph neural network propagation mechanism, guiding information transmission within the graph.

[0023] In one possible implementation, the single-layer propagation update formula for the attention mechanism graph neural network is as follows:

[0024] (4)

[0025] In equation (4), For nodes In the Hidden representation of layers; For nodes The set of adjacent nodes; For nodes In the Hidden representation of layers; The weight matrix is ​​a learnable weight matrix; Attention weights are used to measure neighboring nodes. right The importance of; It is a non-linear activation function;

[0026] Attention weight Calculated using the following formula:

[0027] (5)

[0028] In equation (5), the nodes are... and nodes Input feature vector and Through linear transformation matrix Mapping to a new feature space yields and Subsequently, the two vectors were concatenated into a single entity. Then, along with a learnable weight vector. The inner product operation is performed to generate a score, which is then processed by the LeakyReLU activation function and further transformed by an exponential function to form an unnormalized attention score. Finally, the attention score is generated by processing the nodes. All neighboring nodes The scores are normalized to obtain the normalized attention weights. .

[0029] In one possible implementation, a set of multiple candidate topology reconfigurations for:

[0030] (6)

[0031] In equation (6), For the first k There are candidate topology reconstructions, with a total of . K The candidate topologies are selected based on their physical and structural constraints, satisfying the physical and structural constraints of the distribution network. Candidate reconfiguration topologies are generated through multi-class prediction, graph decoder, or incremental reconfiguration. Each candidate reconfiguration topology is accompanied by a feasibility score.

[0032] (7)

[0033] In equation (7), For the first k The relative scores of the candidate topology reconstructions.

[0034] In one possible implementation, with the dual objectives of maximizing load recovery and minimizing reconfiguration cost, the following predictive optimization model in the rolling time domain is constructed:

[0035] (8)

[0036] In equation (8), Indicates time Until time All switching actions within the rolling time domain; This means that within this rolling time domain, the load recovery amount is maximized and the reconstruction cost is minimized through reconstruction operations. Represents a node In time Is there power? For nodes Load weight; For time switch Control actions; This is a penalty coefficient used to constrain the topology switching frequency; The set of nodes participating in the reconstruction; This is the prediction step size for MPC;

[0037] The preset constraints are as follows:

[0038] 1) The distribution network must maintain a radial structure, and a virtual flow model is used for constraint modeling:

[0039] (9)

[0040] In equation (9), For time The topological edge flow variable indicates whether a branch is selected;

[0041] 2) Power flow constraints, using an approximate DC power flow model:

[0042] (10)

[0043] In equation (10), at time , and They are nodes and The voltage; branch road The trend; branch road impedance;

[0044] Branch capacity is limited to:

[0045] (11)

[0046] In equation (11), branch road The maximum allowable current, also known as the maximum capacity;

[0047] 3) Node voltage constraints:

[0048] (12)

[0049] In equation (12), and These are the minimum and maximum values ​​of the voltage, respectively.

[0050] 4) Control action restrictions to prevent frequent switching:

[0051] (13)

[0052] The implementation logic of MPC is that it only executes [the necessary steps] in each round of optimization. The corresponding switching operations update the power distribution network operating status in real time, thereby updating the attributed graph structure.

[0053] In one possible implementation, the current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. Using the updated attributed graph structure, a pre-trained attention-based graph neural network is input for the next round, outputting multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decision-making is performed on these candidate topologies using MPC, imposing preset constraints. A mathematical optimization solver is used to calculate the optimal reconfiguration path sequence, and the current step control operation is executed according to the optimal reconfiguration path sequence, while the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The process continues with the next round of inputting the pre-trained attention-based graph neural network, and so on, until a preset termination condition is met, achieving dynamic reconfiguration of the distribution network, including:

[0054] Execute the current step control operation according to the optimal reconstruction path sequence, and adjust the state of operable switches in the network. The action set is denoted as:

[0055] (14)

[0056] In equation (14), Indicates the number of controllable switches; Indicates switch In time The operational status, among which , Indicates closure. Indicates disconnection;

[0057] After the control action is executed, the operating status of the distribution network is updated from the operating status. Entering a new operating state The following information was collected:

[0058] Node voltage Node power DG's power output status Branch current The actual topology connection status;

[0059] The collected data forms the new state vector of the current network:

[0060] (15)

[0061] Based on the newly collected status Reconstruct the attribute features of nodes and edges, update the attributed graph structure, and obtain... The next round input is a pre-trained attention mechanism graph neural network, which outputs multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decisions are made on these candidate topologies using MPC, with pre-defined constraints applied. A mathematical optimization solver is used to calculate the optimal reconfiguration path sequence. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The next round input is then processed through the pre-trained attention mechanism graph neural network, and so on, until any of the following termination conditions are met:

[0062] 1) Power has been restored to all recoverable loads:

[0063] (16)

[0064] In equation (16), Represents a node In time t Is there power?

[0065] 2) There are no remaining feasible topologies or device resources to limit;

[0066] At this point, dynamic reconfiguration of the distribution network is achieved, the restoration task is determined to be completed, and the system switches to normal operation mode or standby mode.

[0067] In one possible implementation, after achieving dynamic reconfiguration of the distribution network, the method further includes:

[0068] Select the following key indicators to evaluate fault recovery capability:

[0069] 1) Average load recovery rate (ALRR) is defined as:

[0070] (17)

[0071] In equation (17), For nodes In time t The recovery power; For nodes Total recovery amount; ALRR means the recovery amount in total recovery. The ratio of average recovery load to total load during each recovery phase;

[0072] 2) Mean Recovery Time (ART), defined as:

[0073] (18)

[0074] In equation (18), This is the initial set of power-depleted nodes; For nodes The time it takes for power to be restored The time of the fault occurrence;

[0075] 3) Reconfiguration cost RCost, priced per unit of switch operation:

[0076] (19)

[0077] In equation (19), For switch The unit operating cost Indicates switch In time The operating status, Indicates closure. This indicates a disconnection.

[0078] Secondly, a dynamic reconfiguration device for distribution networks oriented towards rapid fault recovery is provided, the device comprising:

[0079] The graph structure construction unit is used to obtain the current network structure, operating status, and available resources after a fault occurs in the distribution network. Based on the current network structure, operating status, and available resources, a graph structure modeling method is used to represent the distribution network as a graph structure with attributes. The node set in the graph structure includes buses, load nodes, and distributed generation (DG) nodes; the edge set in the graph structure includes distribution lines, switches, and tie lines; the attribute characteristics of each node in the node set include one or more of the following: voltage, current, load power, connected DG output, DG type, and power supply status; the attribute characteristics of each edge in the edge set include one or more of the following: branch impedance, maximum allowable current, switch status, and whether it is faulty.

[0080] The candidate reconfiguration scheme generation unit is used to input the attributed graph structure into a pre-trained attention mechanism graph neural network and output multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network.

[0081] The reconfiguration optimization scheduling unit is used to construct a predictive optimization model in the rolling time domain with the dual objectives of maximizing load recovery and minimizing reconfiguration cost. It uses model predictive control (MPC) to make multi-time-step optimization decisions on multiple candidate reconfiguration topologies, applies preset constraints, and uses a mathematical optimization solver to calculate the optimal reconfiguration path sequence. It then executes the current step control operation according to the optimal reconfiguration path sequence and updates the distribution network operating status in real time, thereby updating the attributed graph structure.

[0082] The control execution and feedback unit utilizes the updated attributed graph structure to input a pre-trained attention mechanism graph neural network for the next round, outputting multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. MPC is used to perform multi-time-step optimization decisions on these candidate reconfiguration topologies, applying preset constraints. A mathematical optimization solver is used to calculate the optimal reconfiguration path sequence, and the current step control operation is executed according to the optimal reconfiguration path sequence. The distribution network operating status is updated in real time, and the attributed graph structure is updated again. The next round of input to the pre-trained attention mechanism graph neural network continues, and so on, until the preset termination condition is met, thus achieving dynamic reconfiguration of the distribution network.

[0083] Thirdly, an electronic device is provided, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the dynamic reconfiguration method for rapid fault recovery of the distribution network as described in any of the preceding claims.

[0084] Fourthly, a storage medium is provided that stores a computer program, wherein the computer program is configured to execute the dynamic reconfiguration method for rapid fault recovery of the distribution network as described above during runtime.

[0085] By utilizing the above technical solutions, the dynamic reconfiguration method, device, electronic equipment, and storage medium for distribution networks oriented towards rapid fault recovery provided in this application embodiment are a distribution network fault rapid recovery method oriented towards the operation level. It integrates attention mechanism graph neural network and MPC (Model Predictive Control). Starting from the current operating state, it realizes dynamic topology reconfiguration of the distribution network under fault through graph structure modeling, candidate reconfiguration topology generation, rolling optimization scheduling, and closed-loop feedback control. It achieves efficient dynamic recovery of the fault area, solves the limitations of traditional static reconfiguration schemes such as response lag and poor recovery path quality, and significantly improves the self-healing capability and recovery efficiency of the power grid system. Attached Figure Description

[0086] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0087] Figure 1 A flowchart of the dynamic reconfiguration method for distribution networks oriented towards rapid fault recovery provided in an embodiment of this application is shown;

[0088] Figure 2 A schematic diagram of an IEEE 123-node distribution network provided in an embodiment of this application is shown;

[0089] Figure 3 This application illustrates the DG node information provided in an embodiment of the present application;

[0090] Figure 4 The BESS node information provided in the embodiments of this application is shown;

[0091] Figure 5 The typical daily load time-series curves provided in the embodiments of this application are shown;

[0092] Figure 6 The following is a typical daily power generation time-series curve of distributed photovoltaic power provided in an embodiment of this application;

[0093] Figure 7 The following is a typical daily power generation time-series curve of distributed wind power provided in an embodiment of this application;

[0094] Figure 8 This application provides a description of key indicators for evaluating fault recovery capabilities, as illustrated in its embodiments.

[0095] Figure 9 A list of methods provided in the embodiments of this application is shown;

[0096] Figure 10 The key indicator values ​​of the various methods provided in the embodiments of this application are shown;

[0097] Figure 11 This paper shows a structural diagram of a power distribution network dynamic reconfiguration device for rapid fault recovery provided in an embodiment of this application.

[0098] Figure 12 A structural diagram of a distribution network dynamic reconfiguration device for rapid fault recovery provided in another embodiment of this application is shown;

[0099] Figure 13 A structural diagram of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0100] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0101] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such use can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the term "comprising" and its variations should be interpreted as open-ended terms meaning "including but not limited to."

[0102] To address the aforementioned technical problems, embodiments of this application provide a dynamic reconfiguration method for distribution networks aimed at rapid fault recovery, such as... Figure 1 As shown, the dynamic reconfiguration method for distribution networks oriented towards rapid fault recovery may include the following steps S101 to S105:

[0103] Step S101: After a fault occurs in the distribution network, obtain the current network structure, operating status, and available resources.

[0104] Step S102: Based on the current network structure, operating status, and available resources, a graph structure modeling method is used to represent the distribution network as a graph structure with attributes. The node set in the graph structure includes busbars, load nodes, and distributed generation (DG) nodes; the edge set in the graph structure includes distribution lines, switches, and tie lines; the attribute characteristics of each node in the node set include one or more of the following: voltage, current, load power, connected DG output, DG type, and power supply status; the attribute characteristics of each edge in the edge set include one or more of the following: branch impedance, maximum allowable current, switch status, and whether it is faulty.

[0105] Step S103: Input the attributed graph structure into a pre-trained attention mechanism graph neural network, and output multiple candidate reconfigured topologies that satisfy the physical and structural constraints of the power distribution network.

[0106] In this step, sample training data can be pre-constructed. The sample training data may include the attributed graph structure of the sample distribution network and one or more sample candidate reconstructed topologies. The initial attention mechanism graph neural network is trained using the sample training data to obtain the trained attention mechanism graph neural network.

[0107] Step S104: With the dual objectives of maximizing load recovery and minimizing reconfiguration cost, a predictive optimization model in the rolling time domain is constructed. Model predictive control (MPC) is used to make multi-time-step optimization decisions on multiple candidate reconfiguration topologies, and preset constraints are applied. The optimal reconfiguration path sequence is calculated using a mathematical optimization solver. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operation status is updated in real time, thereby updating the attributed graph structure.

[0108] Step S105: Using the updated attributed graph structure, input the pre-trained attention mechanism graph neural network into the next round, outputting multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. MPC is used to perform multi-time-step optimization decisions on the multiple candidate reconfiguration topologies, apply preset constraints, use a mathematical optimization solver to calculate the optimal reconfiguration path sequence, execute the current step control operation according to the optimal reconfiguration path sequence, and update the distribution network operating status in real time, thereby updating the attributed graph structure, and continuing to input the pre-trained attention mechanism graph neural network into the next round, and so on, until the preset termination condition is met, realizing dynamic reconfiguration of the distribution network.

[0109] This embodiment presents a method for rapid recovery of distribution network faults at the operational level. It integrates attention mechanism graph neural network and MPC. Starting from the current operating state, it realizes dynamic topology reconstruction of the distribution network under fault through graph structure modeling, candidate reconfiguration topology generation, rolling optimization scheduling and closed-loop feedback control. This achieves efficient dynamic recovery of the fault area, solves the limitations of traditional static reconfiguration schemes such as response lag and poor recovery path quality, and significantly improves the self-healing capability and recovery efficiency of the power grid system.

[0110] This application provides a possible implementation method, wherein the attributed graph structure mentioned in step S102 above is... This allows for the subsequent use of a pre-trained attention-based graph neural network for candidate topology generation and rolling optimization control.

[0111] The entire graph structure is formally represented as follows: ,in, For a set of nodes, For edge set, For nodes The attributes and characteristics, For the edge The attribute characteristics, the edges here Also known as a branch road ;

[0112] (1)

[0113] In equation (1), For nodes The voltage; For nodes The load power; For nodes The output power of the connected DG; For nodes DG types, such as photovoltaic, wind power, hydropower, micro gas turbine units, etc.; For nodes The power supply status is indicated by 1 for normal operation and 0 for power failure.

[0114] (2)

[0115] In equation (2), For the edge Branch impedance; For the edge Maximum allowable current; For the edge The switch status is 1 for closed and 0 for open; For the edge A flag indicating whether a fault exists;

[0116] The connectivity relationships in a graph structure are determined by an adjacency matrix. express:

[0117] (3)

[0118] This adjacency matrix will serve as the core structural input in the pre-trained attention mechanism graph neural network propagation mechanism, guiding information transmission within the graph.

[0119] This application provides a possible implementation method. After a fault occurs in the distribution network or equipment goes offline, the original radial structure of the distribution network is destroyed, and some load nodes lose power. In order to quickly generate feasible reconfiguration paths in a complex and ever-changing operating environment and improve the recovery speed of the power grid system, this embodiment introduces a graph neural network (GNN) structure, specifically a graph attention network (GAT). This allows for deep fusion modeling of the topology and operating state in the graph structure model, automatically generating a set of feasible "candidate reconfiguration schemes".

[0120] This embodiment employs an attention-based graph neural network. The single-layer propagation update formula for the attention-based graph neural network mentioned in step S103 above is as follows:

[0121] (4)

[0122] In equation (4), For nodes In the Hidden representation of layers; For nodes The set of adjacent nodes; For nodes In the Hidden representation of layers; The weight matrix is ​​a learnable weight matrix; Attention weights are used to measure neighboring nodes. right The importance of; It is a non-linear activation function;

[0123] Attention weight Calculated using the following formula:

[0124] (5)

[0125] In equation (5), the nodes are... and nodes Input feature vector and Through linear transformation matrix Mapping to a new feature space yields and Subsequently, the two vectors were concatenated into a single entity. Then, along with a learnable weight vector. The inner product operation is performed to generate a score, which is then processed by the LeakyReLU activation function and further transformed by an exponential function to form an unnormalized attention score. Finally, the attention score is generated by processing the nodes. All neighboring nodes The scores are normalized to obtain the normalized attention weights. .

[0126] This application provides a possible implementation method, which is a set of multiple candidate reconstructed topologies. for:

[0127] (6)

[0128] In equation (6), For the first k There are candidate topology reconstructions, with a total of . K The candidate topologies are selected based on their physical and structural constraints, satisfying the physical and structural constraints of the distribution network. Candidate reconfiguration topologies are generated through multi-class prediction, graph decoder, or incremental reconfiguration. Each candidate reconfiguration topology is accompanied by a feasibility score.

[0129] (7)

[0130] In equation (7), For the first k The relative scores of the candidate topology reconstructions.

[0131] Finally, GAT outputs the following information for MPC to use:

[0132] 1) A set of multiple candidate topology reconstructions ;

[0133] 2) The scoring vector for each topology ;

[0134] 3) Each topology meets the structural and electrical feasibility requirements and can be used as input for the optimizer in the next step.

[0135] This application provides a possible implementation method in which a set of multiple candidate reconstructed topologies output by GAT is obtained. Subsequently, this embodiment employs MPC to perform multi-time-step optimization decisions on multiple candidate reconfiguration topologies, dynamically selecting the optimal topology adjustment path within a rolling time window, thereby improving recovery efficiency and ensuring the operational safety of the power grid system.

[0136] MPC at each time step Execute the following logic:

[0137] 1) Predicting the future based on the current network state The evolution of the power grid system;

[0138] 2) For each candidate reconstructed topology generated by GAT Perform scrolling optimization;

[0139] 3) Select the optimal topology and execute the action. ;

[0140] 4) Implement control, provide feedback on the power grid system status, and proceed to the next round.

[0141] Step S104 above aims to maximize load recovery and minimize reconstruction cost, and constructs the following predictive optimization model in the rolling time domain:

[0142] (8)

[0143] In equation (8), Indicates time Until time All switching actions within the rolling time domain; This means that within this rolling time domain, the load recovery amount is maximized and the reconstruction cost is minimized through reconstruction operations. Represents a node In time Is there power? For nodes Load weight; For time switch Control actions; This is a penalty coefficient used to constrain the topology switching frequency; The set of nodes participating in the reconstruction; This is the prediction step size for MPC;

[0144] The preset constraints are as follows:

[0145] 1) The distribution network must maintain a radial structure, and a virtual flow model is used for constraint modeling:

[0146] (9)

[0147] In equation (9), For time The topological edge flow variable indicates whether a branch is selected;

[0148] 2) Power flow constraints, using an approximate DC power flow model:

[0149] (10)

[0150] In equation (10), at time , and They are nodes and The voltage; branch road The trend; branch road impedance;

[0151] Branch capacity is limited to:

[0152] (11)

[0153] In equation (11), branch road The maximum allowable current, also known as the maximum capacity;

[0154] 3) Node voltage constraints:

[0155] (12)

[0156] In equation (12), and These are the minimum and maximum values ​​of the voltage, respectively.

[0157] 4) Control action restrictions to prevent frequent switching:

[0158] (13)

[0159] The implementation logic of MPC is that it only executes [the necessary steps] in each round of optimization. The corresponding switching operations update the power distribution network's operating status in real time, thereby updating the attributed graph structure and forming a "prediction-execution-feedback" closed loop.

[0160] This application provides a possible implementation method. Step S104 executes the current step control operation according to the optimal reconfiguration path sequence and updates the distribution network operating status in real time, thereby updating the attributed graph structure. Step S105 uses the updated attributed graph structure to input a pre-trained attention mechanism graph neural network for the next round, outputting multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. MPC is used to perform multi-time-step optimization decisions on the multiple candidate reconfiguration topologies, applying preset constraints. A mathematical optimization solver is used to calculate the optimal reconfiguration path sequence. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The next round of input to the pre-trained attention mechanism graph neural network continues, and so on, until a preset termination condition is met, realizing dynamic reconfiguration of the distribution network. Specifically, this may include:

[0161] Execute the current step control operation according to the optimal reconstruction path sequence, and adjust the state of operable switches in the network. The action set is denoted as:

[0162] (14)

[0163] In equation (14), Indicates the number of controllable switches; Indicates switch In time The operational status, among which , Indicates closure. Indicates disconnection;

[0164] Here, the power grid system will send instructions to the site through automated devices (such as remote terminal units) to execute the switching actions of this group of switches.

[0165] After the control action is executed, the operating status of the distribution network is updated from the operating status. Entering a new operating state The following information was collected:

[0166] Node voltage Node power DG's power output status Branch current The actual topology connection status;

[0167] The collected data forms the new state vector of the current network:

[0168] (15)

[0169] Based on the newly collected status Reconstruct the attribute features of nodes and edges, update the attributed graph structure, and obtain... The next round input is a pre-trained attention mechanism graph neural network, which outputs multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decisions are made on these candidate topologies using MPC, with pre-defined constraints applied. A mathematical optimization solver is used to calculate the optimal reconfiguration path sequence. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The next round input is then processed through the pre-trained attention mechanism graph neural network, and so on, until any of the following termination conditions are met:

[0170] 1) Power has been restored to all recoverable loads:

[0171] (16)

[0172] In equation (16), Represents a node In time t Is there power?

[0173] 2) There are no remaining feasible topologies or device resources to limit;

[0174] At this point, dynamic reconfiguration of the distribution network is achieved, the restoration task is determined to be completed, and the system switches to normal operation mode or standby mode.

[0175] In the above embodiment, based on the newly collected status... Reconstruct the attribute features of nodes and edges, update the attributed graph structure, and obtain... ; Generate candidate reconstruction schemes again, i.e., multiple candidate reconstruction topologies; re-optimize and predict the reconstruction path for the future time period t+1 to t+N+1; This closed loop forms the following information link:

[0176]

[0177] The power distribution network operation status is updated in real time, which in turn updates the attributed graph structure and continues to input the pre-trained attention mechanism graph neural network into the next round, and so on, until the termination condition is met.

[0178] This application provides a possible implementation method in which, after achieving dynamic reconfiguration of the distribution network, the method further includes:

[0179] Select the following key indicators to evaluate fault recovery capability:

[0180] 1) Average load recovery rate (ALRR) is defined as:

[0181] (17)

[0182] In equation (17), For nodes In time t The recovery power; For nodes Total recovery amount; ALRR means the recovery amount in total recovery. The ratio of average recovery load to total load during each recovery phase;

[0183] 2) Mean Recovery Time (ART), defined as:

[0184] (18)

[0185] In equation (18), This is the initial set of power-depleted nodes; For nodes The time it takes for power to be restored The time of the fault occurrence;

[0186] 3) Reconfiguration cost RCost, priced per unit of switch operation:

[0187] (19)

[0188] In equation (19), For switch The unit operating cost Indicates switch In time The operating status, Indicates closure. This indicates a disconnection.

[0189] The above introduces Figure 1 The embodiments shown have various implementation methods for each stage. The following will further explain the dynamic reconfiguration method for distribution networks oriented towards rapid fault recovery in this application through specific embodiments.

[0190] In a specific embodiment, the following is selected: Figure 2 A simulation platform was built for the IEEE 123-node distribution network shown, and recovery and scheduling tests were conducted under scenarios of multi-point faults and high-proportion distributed generation (DG) access. The test system deployed 9 DGs (including photovoltaic and wind power) and 5 BESS (Battery Energy Storage Systems), detailed as follows: Figure 3 and Figure 4 As shown, multiple faults are set in the main branch lines to create a large-scale power outage area.

[0191] Load and DG output, such as Figure 5 , Figure 6 and Figure 7 As shown, multiple candidate reconstruction topologies are generated by GAT trained on historical multi-scenario data, and reconstruction scheduling is implemented by the MPC module in a rolling optimization manner. The simulation sets the control period to 15 seconds and the prediction step size to 3.

[0192] To comprehensively evaluate fault recovery capabilities, the following were selected: Figure 8 The key metrics shown are: Average Load Recovery Rate (ALRR), Average Restoration Time (ART), and Reconfiguration Cost (RCost).

[0193] To comprehensively evaluate the practical effect of the "Fast Fault Reconfiguration Method for Distribution Networks Based on Joint Control of GAT and MPC" proposed in this invention, this embodiment selects four mainstream typical recovery strategies as comparison methods, such as... Figure 9 As shown, the details are as follows:

[0194] 1. Heuristic Reconfiguration (HR): This method uses fixed rules (such as minimum number of switch operations, shortest path recovery, etc.) to perform single-step topology switching. The operation logic is simple and the response speed is fast, but it cannot dynamically adapt to changes in network status after a fault, resulting in low recovery efficiency and load utilization.

[0195] 2. Topology Scoring (TR): This method evaluates and ranks the topology of all possible reconstruction paths, selecting the path with the best score for reconstruction. While it incorporates some optimization ideas, it lacks predictive consideration of temporal dynamic evolution and DG / BESS fluctuations.

[0196] 3. Mathematical Programming-R (MILP-R): This method constructs a mixed-integer linear programming model to optimize the topology with the goal of maximizing load recovery. Although the solution is globally optimal, its computational complexity is high, making it difficult to meet the scheduling requirements of multi-stage, fast-response operations.

[0197] 4. Data-driven approach (DQN-RL): This approach utilizes deep reinforcement learning methods, such as DQN (Deep Q-Network), to train recovery strategies. It has strong generalization ability and real-time performance, but training depends on a large number of samples and it is difficult to ensure that the electrical physical constraints are strictly met, which poses a risk of unstable recovery paths.

[0198] Based on the comparison of the above five methods in terms of the three core performance indicators of Average Load Recovery Rate (ALRR), Average Recovery Time (ART), and Reconfiguration Cost (RCost), see [link to relevant documentation]. Figure 10 It can be clearly seen that the dynamic topology reconstruction method based on the joint control of attention mechanism graph neural network and model predictive control (GAT+MPC) proposed in this invention has significant advantages in many aspects.

[0199] First, in terms of average load recovery rate, the method of this invention can generate high-quality candidate reconfiguration paths based on the current system state, and achieve rapid power supply recovery to grid fault areas through MPC dynamic scheduling of distributed generation (DG) and energy storage system (BESS), with a recovery rate of 96.2%, which is higher than all other methods, especially compared with heuristic methods, it is more than 14.8% higher.

[0200] Secondly, regarding average recovery time, although the mathematical programming method (MILP-R) theoretically finds the optimal path, its complexity limits the scheduling frequency in practical applications. The method of this invention, through short-time rolling optimization, reduces the recovery time to 4.2 steps while maintaining high computational efficiency, second only to MILP-R and far superior to heuristic and topology scoring methods.

[0201] Finally, regarding the cost of control operations, thanks to the attention mechanism graph neural network's ability to learn the "minimum action set" and the MPC's fine control over constraints, this method requires only 13 switching operations, significantly lower than the other four methods, effectively reducing the physical cost and execution complexity of power grid system reconfiguration.

[0202] The dynamic topology reconfiguration method based on attention-based graph neural networks and model predictive control proposed in this invention exhibits significant advantages in complex fault scenarios. This method integrates graph structure learning and rolling optimization scheduling mechanisms, enabling rapid, efficient, and adaptive recovery of faulty regions while ensuring physical feasibility. Compared to traditional heuristic methods and static scoring strategies, this method boasts higher load recovery rates, shorter recovery times, and lower control operation costs. Furthermore, compared to mathematical programming and reinforcement learning methods, it offers superior computational efficiency, more comprehensive constraint handling, and stronger generalization capabilities. It can broadly adapt to the operational needs of multi-scenario, multi-source, and heterogeneous distribution networks, demonstrating excellent engineering practicality and intelligent control capabilities.

[0203] Compared with existing technologies, the fast fault recovery method for distribution networks oriented towards the operation level provided by this invention has the following beneficial effects:

[0204] 1. This invention integrates attention-based graph neural networks (GAT) with model predictive control (MPC) strategies. By sensing the current network operating state and intelligently generating candidate topology reconfiguration schemes that meet operational constraints, it achieves efficient dynamic recovery of faulty regions. This method overcomes the limitations of traditional static reconfiguration schemes, such as slow response and poor recovery path quality, significantly improving the self-healing capability and recovery efficiency of the power grid system.

[0205] 2. Starting from the operation control level, this invention constructs a dynamic recovery control link that includes "candidate generation - rolling optimization - control execution - closed-loop feedback" throughout the entire process. It comprehensively considers various operational constraints and control objectives such as load recovery priority, voltage compliance, operation cost, and number of switching operations, effectively ensuring the feasibility and economic efficiency of the recovery path.

[0206] Through numerical verification on a typical IEEE 123-node system, the results show that the method proposed in this invention outperforms traditional heuristic reconfiguration, static scoring, reinforcement learning, and mathematical programming methods in terms of average load recovery rate, average recovery time, and reconfiguration cost (also known as operational cost). It can effectively improve the operation and recovery capability of the distribution network and is suitable for active distribution system recovery scenarios under high-penetration renewable energy and distributed power source access. It has good engineering adaptability and promotion value.

[0207] It should be noted that the sequence numbers of the steps in the above embodiments do not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. In practical applications, all the above possible implementation methods can be arbitrarily combined in a combined manner to form possible embodiments of this application, which will not be described in detail here.

[0208] Based on the distribution network dynamic reconfiguration method for rapid fault recovery provided in the above embodiments, and based on the same inventive concept, this application also provides a distribution network dynamic reconfiguration device for rapid fault recovery.

[0209] Figure 11 This is a structural diagram of the dynamic reconfiguration device for distribution networks designed for rapid fault recovery provided in an embodiment of this application. Figure 11 As shown, the distribution network dynamic reconfiguration device for rapid fault recovery may specifically include a graph structure construction unit 1110, a candidate reconfiguration scheme generation unit 1120, a reconfiguration optimization scheduling unit 1130, and a control execution and feedback unit 1140.

[0210] The graph structure construction unit 1110 is used to obtain the current network structure, operating status, and available resources after a fault occurs in the distribution network. Based on the current network structure, operating status, and available resources, a graph structure modeling method is used to represent the distribution network as a graph structure with attributes. The node set in the graph structure includes buses, load nodes, and distributed generation (DG) nodes; the edge set in the graph structure includes distribution lines, switches, and tie lines; the attribute characteristics of each node in the node set include one or more of the following: voltage, current, load power, connected DG output, DG type, and power supply status; the attribute characteristics of each edge in the edge set include one or more of the following: branch impedance, maximum allowable current, switch status, and whether it is faulty.

[0211] The candidate reconfiguration scheme generation unit 1120 is used to input the attributed graph structure into a pre-trained attention mechanism graph neural network and output multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network.

[0212] The reconfiguration optimization scheduling unit 1130 is used to construct a predictive optimization model in the rolling time domain with the dual objectives of maximizing load recovery and minimizing reconfiguration cost. It uses model predictive control (MPC) to make multi-time-step optimization decisions on multiple candidate reconfiguration topologies, applies preset constraints, and uses a mathematical optimization solver to calculate the optimal reconfiguration path sequence. It then executes the current step control operation according to the optimal reconfiguration path sequence and updates the distribution network operating status in real time, thereby updating the attributed graph structure.

[0213] The control execution and feedback unit 1140 is used to utilize the updated attributed graph structure to input a pre-trained attention mechanism graph neural network for the next round, output multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network, use MPC to perform multi-time-step optimization decisions on the multiple candidate reconfiguration topologies, apply preset constraints, use a mathematical optimization solver to calculate the optimal reconfiguration path sequence, execute the current step control operation according to the optimal reconfiguration path sequence, update the distribution network operating status in real time, and then update the attributed graph structure to continue the next round of inputting the pre-trained attention mechanism graph neural network, and so on, until the preset termination condition is met, thereby realizing the dynamic reconfiguration of the distribution network.

[0214] This application provides a possible implementation method, where the attributed graph structure is as follows: , ,in, For a set of nodes, For edge set, For nodes The attributes and characteristics, For the edge The attribute characteristics, the edges here Also known as a branch road ;

[0215] (1)

[0216] In equation (1), For nodes The voltage; For nodes The load power; For nodes The output power of the connected DG; For nodes DG type; For nodes The power supply status is indicated by 1 for normal operation and 0 for power failure.

[0217] (2)

[0218] In equation (2), For the edge Branch impedance; For the edge Maximum allowable current; For the edge The switch status is 1 for closed and 0 for open; For the edge A flag indicating whether a fault exists;

[0219] The connectivity relationships in a graph structure are determined by an adjacency matrix. express:

[0220] (3)

[0221] This adjacency matrix will serve as the core structural input in the pre-trained attention mechanism graph neural network propagation mechanism, guiding information transmission within the graph.

[0222] This application provides a possible implementation method, and the single-layer propagation update formula of the attention mechanism graph neural network is as follows:

[0223] (4)

[0224] In equation (4), For nodes In the Hidden representation of layers; For nodes The set of adjacent nodes; For nodes In the Hidden representation of layers; The weight matrix is ​​a learnable weight matrix; Attention weights are used to measure neighboring nodes. right The importance of; It is a non-linear activation function;

[0225] Attention weight Calculated using the following formula:

[0226] (5)

[0227] In equation (5), the nodes are... and nodes Input feature vector and Through linear transformation matrix Mapping to a new feature space yields and Subsequently, the two vectors were concatenated into a single entity. Then, along with a learnable weight vector. The inner product operation is performed to generate a score, which is then processed by the LeakyReLU activation function and further transformed by an exponential function to form an unnormalized attention score. Finally, the attention score is generated by processing the nodes. All neighboring nodes The scores are normalized to obtain the normalized attention weights. .

[0228] This application provides a possible implementation method, which is a set of multiple candidate reconstructed topologies. for:

[0229] (6)

[0230] In equation (6), For the first k There are candidate topology reconstructions, with a total of . K The candidate topologies are selected based on their physical and structural constraints, satisfying the physical and structural constraints of the distribution network. Candidate reconfiguration topologies are generated through multi-class prediction, graph decoder, or incremental reconfiguration. Each candidate reconfiguration topology is accompanied by a feasibility score.

[0231] (7)

[0232] In equation (7), For the first k The relative scores of the candidate topology reconstructions.

[0233] This application provides a possible implementation method that aims to maximize load recovery and minimize reconstruction cost, constructing the following prediction optimization model in the rolling time domain:

[0234] (8)

[0235] In equation (8), Indicates time Until time All switching actions within the rolling time domain; This means that within this rolling time domain, the load recovery amount is maximized and the reconstruction cost is minimized through reconstruction operations. Represents a node In time Is there power? For nodes Load weight; For time switch Control actions; This is a penalty coefficient used to constrain the topology switching frequency; The set of nodes participating in the reconstruction; This is the prediction step size for MPC;

[0236] The preset constraints are as follows:

[0237] 1) The distribution network must maintain a radial structure, and a virtual flow model is used for constraint modeling:

[0238] (9)

[0239] In equation (9), For time The topological edge flow variable indicates whether a branch is selected;

[0240] 2) Power flow constraints, using an approximate DC power flow model:

[0241] (10)

[0242] In equation (10), at time , and They are nodes and The voltage; branch road The trend; branch road impedance;

[0243] Branch capacity is limited to:

[0244] (11)

[0245] In equation (11), branch road The maximum allowable current, also known as the maximum capacity;

[0246] 3) Node voltage constraints:

[0247] (12)

[0248] In equation (12), and These are the minimum and maximum values ​​of the voltage, respectively.

[0249] 4) Control action restrictions to prevent frequent switching:

[0250] (13)

[0251] The implementation logic of MPC is that it only executes [the necessary steps] in each round of optimization. The corresponding switching operations update the power distribution network operating status in real time, thereby updating the attributed graph structure.

[0252] This application embodiment provides a possible implementation, wherein the reconfiguration optimization scheduling unit 1130 is further configured to:

[0253] Execute the current step control operation according to the optimal reconstruction path sequence, and adjust the state of operable switches in the network. The action set is denoted as:

[0254] (14)

[0255] In equation (14), Indicates the number of controllable switches; Indicates switch In time The operational status, among which , Indicates closure. Indicates disconnection;

[0256] After the control action is executed, the operating status of the distribution network is updated from the operating status. Entering a new operating state The following information was collected:

[0257] Node voltage Node power DG's power output status Branch current The actual topology connection status;

[0258] The collected data forms the new state vector of the current network:

[0259] (15)

[0260] Based on the newly collected status Reconstruct the attribute features of nodes and edges, update the attributed graph structure, and obtain... ;

[0261] The regulation execution and feedback unit 1140 is further configured to: utilize the updated attributed graph structure The pre-trained attention mechanism graph neural network is input into the next round, and the output consists of multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decision-making is performed on these candidate topologies using MPC, with preset constraints applied. The optimal reconfiguration path sequence is calculated using a mathematical optimization solver. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The process continues until any of the following termination conditions are met:

[0262] 1) Power has been restored to all recoverable loads:

[0263] (16)

[0264] In equation (16), Represents a node In time t Is there power?

[0265] 2) There are no remaining feasible topologies or device resources to limit;

[0266] At this point, dynamic reconfiguration of the distribution network is achieved, the restoration task is determined to be completed, and the system switches to normal operation mode or standby mode.

[0267] This application provides a possible implementation method after achieving dynamic reconfiguration of the distribution network, such as... Figure 12 As shown above, Figure 11 The demonstrated device may also include an evaluation unit 1210 for selecting the following key indicators to evaluate fault recovery capability:

[0268] 1) Average load recovery rate (ALRR) is defined as:

[0269] (17)

[0270] In equation (17), For nodes In time t The recovery power; For nodes Total recovery amount; ALRR means the recovery amount in total recovery. The ratio of average recovery load to total load during each recovery phase;

[0271] 2) Mean Recovery Time (ART), defined as:

[0272] (18)

[0273] In equation (18), This is the initial set of power-depleted nodes; For nodes The time it takes for power to be restored The time of the fault occurrence;

[0274] 3) Reconfiguration cost RCost, priced per unit of switch operation:

[0275] (19)

[0276] In equation (19), For switch The unit operating cost Indicates switch In time The operating status, Indicates closure. This indicates a disconnection.

[0277] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the dynamic reconfiguration method for power distribution networks oriented towards rapid fault recovery of any of the above embodiments.

[0278] In an exemplary embodiment, an electronic device is provided, such as Figure 13 As shown, Figure 13 The illustrated electronic device 1300 includes a processor 1301 and a memory 1303. The processor 1301 and the memory 1303 are connected, for example, via a bus 1302. Optionally, the electronic device 1300 may also include a transceiver 1304. It should be noted that in practical applications, the transceiver 1304 is not limited to one type, and the structure of this electronic device 1300 does not constitute a limitation on the embodiments of this application.

[0279] Processor 1301 may be a CPU (Central Processing Unit), GPU (Graphics Processing Unit), DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0280] Bus 1302 may include a pathway for transmitting information between the aforementioned components. Bus 1302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 1302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 13 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0281] The memory 1303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0282] The memory 1303 is used to store computer program code that executes the scheme of this application, and its execution is controlled by the processor 1301. The processor 1301 is used to execute the computer program code stored in the memory 1303 to implement the content shown in the foregoing method embodiments.

[0283] Among them, electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 13 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0284] Based on the same inventive concept, this application also provides a storage medium storing a computer program, wherein the computer program is configured to execute the dynamic reconfiguration method for rapid fault recovery of the distribution network according to any of the above embodiments when running.

[0285] Those skilled in the art will clearly understand that the specific working process of the systems, devices, and modules described above can be referred to the corresponding process in the foregoing method embodiments. For the sake of brevity, it will not be repeated here.

[0286] Those skilled in the art will understand that the technical solution of this application, or all or part of it, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several program instructions to cause an electronic device (e.g., a personal computer, server, or network device) to execute all or part of the steps of the methods described in the embodiments of this application when running the program instructions. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0287] Alternatively, all or part of the steps of the foregoing method embodiments can be implemented by hardware (such as electronic devices like personal computers, servers, or network devices) associated with program instructions. The program instructions can be stored in a computer-readable storage medium. When the program instructions are executed by the processor of the electronic device, the electronic device executes all or part of the steps of the methods described in the embodiments of this application.

[0288] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that within the spirit and principles of this application, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the corresponding technical solutions to leave the protection scope of this application.

Claims

1. A dynamic reconfiguration method for distribution networks oriented towards rapid fault recovery, characterized in that, The method includes: After a fault occurs in the distribution network, obtain the current network structure, operating status, and available resources; Based on the current network structure, operating status, and available resources, a graph modeling method is adopted to represent the distribution network as a graph structure with attributes. The node set in the graph structure includes buses, load nodes, and distributed generation (DG) nodes; the edge set includes distribution lines, switches, and tie lines; the attribute characteristics of each node in the node set include one or more of the following: voltage, current, load power, connected DG output, DG type, and power supply status; the attribute characteristics of each edge in the edge set include one or more of the following: branch impedance, maximum allowable current, switch status, and whether it is faulty. Input the attributed graph structure into a pre-trained attention mechanism graph neural network, and output multiple candidate reconfigured topologies that satisfy the physical and structural constraints of the power distribution network; With the dual objectives of maximizing load recovery and minimizing reconfiguration cost, a predictive optimization model in the rolling time domain is constructed. Model predictive control (MPC) is used to make multi-time-step optimization decisions on multiple candidate reconfiguration topologies, apply preset constraints, and use a mathematical optimization solver to calculate the optimal reconfiguration path sequence. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operation status is updated in real time, thereby updating the attributed graph structure. By utilizing the updated attributed graph structure, the next round of input is a pre-trained attention-based graph neural network, which outputs multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decision-making is performed on the multiple candidate reconfiguration topologies using MPC, and preset constraints are applied. The optimal reconfiguration path sequence is calculated using a mathematical optimization solver. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The next round of input is then performed on the pre-trained attention-based graph neural network, and so on, until the preset termination condition is met, thus realizing the dynamic reconfiguration of the distribution network.

2. The method according to claim 1, characterized in that, Attributed graph structures are , ,in, For a set of nodes, For edge set, For nodes The attributes and characteristics, For the edge The attribute characteristics, the edges here Also known as a branch road ; （1） In equation (1), For nodes The voltage; For nodes The load power; For nodes The output power of the connected DG; For nodes DG type; For nodes The power supply status is indicated by 1 for normal operation and 0 for power failure. （2） In equation (2), For the edge Branch impedance; For the edge Maximum allowable current; For the edge The switch status is 1 for closed and 0 for open; For the edge A flag indicating whether a fault exists; The connectivity relationships in a graph structure are determined by an adjacency matrix. express: （3） This adjacency matrix will serve as the core structural input in the pre-trained attention mechanism graph neural network propagation mechanism, guiding information transmission within the graph.

3. The method according to claim 2, characterized in that, The single-layer propagation update formula for the attention mechanism graph neural network is as follows: （4） In equation (4), For nodes In the Hidden representation of layers; For nodes The set of adjacent nodes; For nodes In the Hidden representation of layers; The weight matrix is ​​a learnable weight matrix; Attention weights are used to measure neighboring nodes. right The importance of; It is a non-linear activation function; Attention weight Calculated using the following formula: （5） In equation (5), the nodes are... and nodes Input feature vector and Through linear transformation matrix Mapping to a new feature space yields and ; The two vectors are then concatenated into a single unit. Then, along with a learnable weight vector. The inner product operation is performed to generate a score, which is then processed by the LeakyReLU activation function and further transformed by an exponential function to form an unnormalized attention score. Finally, the attention score is generated by processing the nodes. All neighboring nodes The scores are normalized to obtain the normalized attention weights. .

4. The method according to claim 2, characterized in that, A set of multiple candidate topology reconstructions for: （6） In equation (6), For the first k There are candidate topology reconstructions, with a total of . K The candidate topologies are selected based on their physical and structural constraints, satisfying the physical and structural constraints of the distribution network. Candidate reconfiguration topologies are generated through multi-class prediction, graph decoder, or incremental reconfiguration. Each candidate reconfiguration topology is accompanied by a feasibility score. （7） In equation (7), For the first k The relative scores of the candidate topology reconstructions.

5. The method according to claim 2, characterized in that, With the dual objectives of maximizing load recovery and minimizing reconfiguration cost, the following predictive optimization model in the rolling time domain is constructed: （8） In equation (8), Indicates time Until time All switching actions within the rolling time domain; This means that within this rolling time domain, the load recovery amount is maximized and the reconstruction cost is minimized through reconstruction operations. Represents a node In time Is there power? For nodes Load weight; For time switch Control actions; This is a penalty coefficient used to constrain the topology switching frequency; The set of nodes participating in the reconstruction; This is the prediction step size for MPC; The preset constraints are as follows: 1) The distribution network must maintain a radial structure, and a virtual flow model is used for constraint modeling: （9） In equation (9), For time The topological edge flow variable indicates whether a branch is selected; 2) Power flow constraints, using an approximate DC power flow model: （10） In equation (10), at time , and They are nodes and The voltage; branch road The trend; branch road impedance; Branch capacity is limited to: （11） In equation (11), branch road The maximum allowable current, also known as the maximum capacity; 3) Node voltage constraints: （12） In equation (12), and These are the minimum and maximum values ​​of the voltage, respectively. 4) Control action restrictions to prevent frequent switching: （13） The implementation logic of MPC is that it only executes [the necessary steps] in each round of optimization. The corresponding switching operations update the power distribution network operating status in real time, thereby updating the attributed graph structure.

6. The method according to claim 5, characterized in that, The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. Using the updated attributed graph structure, the next round input is a pre-trained attention-based graph neural network, outputting multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decision-making is performed on these candidate topologies using MPC, imposing preset constraints. The optimal reconfiguration path sequence is calculated using a mathematical optimization solver, and the current step control operation is executed according to the optimal reconfiguration path sequence, with real-time updates to the distribution network operating status, thereby updating the attributed graph structure. The next round input is then performed to the pre-trained attention-based graph neural network, and so on, until a preset termination condition is met, achieving dynamic reconfiguration of the distribution network, including: Execute the current step control operation according to the optimal reconstruction path sequence, and adjust the state of operable switches in the network. The action set is denoted as: （14） In equation (14), Indicates the number of controllable switches; Indicates switch In time The operational status, among which , Indicates closure. Indicates disconnection; After the control action is executed, the operating status of the distribution network is updated from the operating status. Entering a new operating state The following information was collected: Node voltage Node power DG's power output status Branch current The actual topology connection status; The collected data forms the new state vector of the current network: （15） Based on the newly collected status Reconstruct the attribute features of nodes and edges, update the attributed graph structure, and obtain... The next round input is a pre-trained attention mechanism graph neural network, which outputs multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. Multi-time-step optimization decisions are made on these candidate topologies using MPC, with pre-defined constraints applied. A mathematical optimization solver is used to calculate the optimal reconfiguration path sequence. The current step control operation is executed according to the optimal reconfiguration path sequence, and the distribution network operating status is updated in real time, thereby updating the attributed graph structure. The next round input is then processed through the pre-trained attention mechanism graph neural network, and so on, until any of the following termination conditions are met: 1) Power has been restored to all recoverable loads: （16） In equation (16), Represents a node In time t Is there power? 2) There are no remaining feasible topologies or device resources to limit; At this point, dynamic reconfiguration of the distribution network is achieved, the restoration task is determined to be completed, and the system switches to normal operation mode or standby mode.

7. The method according to claim 6, characterized in that, After achieving dynamic reconfiguration of the distribution network, the method further includes: Select the following key indicators to evaluate fault recovery capability: 1) Average load recovery rate (ALRR) is defined as: （17） In equation (17), For nodes In time t The recovery power; For nodes Total recovery amount; ALRR means the recovery amount in the total recovery amount. The ratio of average recovery load to total load during each recovery phase; 2) Mean Recovery Time (ART), defined as: （18） In equation (18), This is the initial set of power-depleted nodes; For nodes The time it takes for power to be restored The time of the fault occurrence; 3) Reconfiguration cost RCost, priced per unit of switch operation: （19） In equation (19), For switch The unit operating cost Indicates switch In time The operating status, Indicates closure. This indicates a disconnection.

8. A dynamic reconfiguration device for distribution networks oriented towards rapid fault recovery, characterized in that, The device includes: The graph structure construction unit is used to obtain the current network structure, operating status, and available resources after a fault occurs in the distribution network. Based on the current network structure, operating status, and available resources, a graph structure modeling method is used to represent the distribution network as a graph structure with attributes. The node set in the graph structure includes buses, load nodes, and distributed generation (DG) nodes; the edge set in the graph structure includes distribution lines, switches, and tie lines; the attribute characteristics of each node in the node set include one or more of the following: voltage, current, load power, connected DG output, DG type, and power supply status; the attribute characteristics of each edge in the edge set include one or more of the following: branch impedance, maximum allowable current, switch status, and whether it is faulty. The candidate reconfiguration scheme generation unit is used to input the attributed graph structure into a pre-trained attention mechanism graph neural network and output multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. The reconfiguration optimization scheduling unit is used to construct a predictive optimization model in the rolling time domain with the dual objectives of maximizing load recovery and minimizing reconfiguration cost. It uses model predictive control (MPC) to make multi-time-step optimization decisions on multiple candidate reconfiguration topologies, applies preset constraints, and uses a mathematical optimization solver to calculate the optimal reconfiguration path sequence. It then executes the current step control operation according to the optimal reconfiguration path sequence and updates the distribution network operating status in real time, thereby updating the attributed graph structure. The control execution and feedback unit utilizes the updated attributed graph structure to input a pre-trained attention mechanism graph neural network for the next round, outputting multiple candidate reconfiguration topologies that satisfy the physical and structural constraints of the distribution network. MPC is used to perform multi-time-step optimization decisions on these candidate reconfiguration topologies, applying preset constraints. A mathematical optimization solver is used to calculate the optimal reconfiguration path sequence, and the current step control operation is executed according to the optimal reconfiguration path sequence. The distribution network operating status is updated in real time, and the attributed graph structure is updated again. The next round of input to the pre-trained attention mechanism graph neural network continues, and so on, until the preset termination condition is met, thus achieving dynamic reconfiguration of the distribution network.

9. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the dynamic reconfiguration method for distribution networks oriented towards rapid fault recovery as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute, at runtime, the distribution network dynamic reconfiguration method for rapid fault recovery as described in any one of claims 1 to 7.

Images