Power distribution network minute-level cooperative fault recovery method oriented to network construction type inverter power supply

Abstract

The invention relates to the technical field of power grid dispatching, in particular to a power distribution network minute-level cooperative fault recovery method for a network construction type inverter power supply. When a power grid fault is triggered, a load transfer area and an island key area are extracted, and the island key area is a power grid power failure area which is not covered by the load transfer area; a distributed cooperative power distribution solution is configured for the load transfer area, a network-following cooperative power supply solution is configured for the island key area, analysis is carried out through a graph neural network cost analyzer, and recovery duration and loss electric energy are generated; and when the recovery duration is smaller than or equal to the duration threshold value and the loss electric energy is smaller than or equal to the loss threshold value, controlling the distributed power supply to supply power to the load transfer area through the distributed cooperative power distribution electrolysis, and controlling the network construction type inverter power supply and the network following type inverter power supply to supply power to the island key area through the network construction and network following cooperative power supply electrolysis. And the toughness and the power supply reliability of the power distribution network under the fault condition are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid dispatching, and in particular to a minute-level collaborative fault recovery method for a distribution network of a grid-forming inverter power supply. Background Art

[0002] Currently, in the field of distribution network fault recovery, traditional meshed control strategies mostly rely on highly reliable communication networks and intelligent terminals. Once communication is interrupted or equipment fails, the recovery strategy will fail. Furthermore, the computing models used lack real-time performance and cannot meet minute-level recovery requirements. Furthermore, traditional solutions only configure different power supply types for different regions, without considering the refined parameters of power supply control. This leads to insufficient control accuracy and poor adaptability when dealing with high-penetration distributed power sources and dynamic reconfiguration scenarios. Some solutions can only handle specific types of faults and cannot fully cover complex fault scenarios. Technical issues exist, such as low speed, accuracy, and cost-effectiveness of grid fault recovery. Summary of the Invention

[0003] The present invention aims to solve the technical problems of low speed, accuracy and economy of power grid fault recovery in the prior art by providing a minute-level coordinated fault recovery method for distribution networks of grid-forming inverter power supplies.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: In the first aspect, the present invention provides a minute-level collaborative fault recovery method for distribution networks of grid-type inverter-type power supplies, comprising: when a grid fault is triggered, extracting load transfer areas and island key areas, wherein the island key area is a grid outage area not covered by the load transfer area; configuring a distributed collaborative distribution solution for the load transfer area, configuring a grid-following collaborative power solution for the island key area, and generating recovery time and power loss through analysis by a graph neural network cost analyzer; when the recovery time is less than or equal to a time threshold, and when the power loss is less than or equal to a loss threshold, controlling the distributed power supply to supply power to the load transfer area through the distributed collaborative distribution solution, and controlling the grid-following inverter power supply and the grid-following inverter power supply to supply power to the island key area through the grid-following collaborative power solution.

[0005] Optionally, when a power grid fault is triggered, the load transfer area and the island key area are extracted, including: when the grid connection point voltage is less than the voltage threshold, it is regarded as a power grid fault trigger, and the power grid blackout area is received; based on the line topology of the power grid blackout area, the distributed power sources connected to the power grid blackout area are extracted; the distributed power sources are traversed to perform flow calculations to determine the load transfer area of ​​the power grid blackout area; the area not covered by the load transfer area in the power grid blackout area is extracted and set as the initial island key area; the initial island key area is traversed to perform source-load balance analysis to obtain the island key area.

[0006] Optionally, when the grid connection point voltage is less than the voltage threshold and before receiving the grid blackout area, it includes: adjusting the current overload capacity to a preset multiple of the rated current within k seconds, where 0≤k≤2.

[0007] Among them, a distributed collaborative distribution solution is configured for the load transfer area, and a structure-based network collaborative power solution is configured for the island key area, and the recovery time and loss electric energy are generated through analysis by a graph neural network cost analyzer, including: matching the graph neural network cost analyzer from a set of graph neural network cost analyzers according to the power grid fault area, wherein the graph neural network cost analyzer has a label that identifies all line topologies of the power grid fault area; based on the load transfer area line topology, compared with all line topologies, the graph neural network cost analyzer is separated to obtain the load transfer area cost analyzer, and the distributed collaborative distribution solution is processed to generate a first recovery time and a first loss electric energy; based on the island key area line topology, compared with all line topologies, the graph neural network cost analyzer is separated to obtain the island key area cost analyzer, and the structure-based network collaborative power solution is processed to generate a second recovery time and a second loss electric energy; the larger value of the first recovery time and the second recovery time is set as the recovery time; the larger value of the first loss electric energy and the second loss electric energy is set as the loss electric energy.

[0008] Among them, based on the line topology of the load transfer area, compared with the total line topology, the graph neural network cost analyzer is separated to obtain the load transfer area cost analyzer, including: extracting a first load transfer area from the load transfer area, wherein the first load transfer area has an accessible distributed power source number identifier; according to the accessible distributed power source number identifier, with the power demand of the first load transfer area before the fault as the target, the first load transfer area is powered on and a coordinated power distribution solution for the first load transfer area is obtained; according to the accessible distributed power source number identifier and the first load transfer area, the line topology of the first load transfer area is extracted from the line topology map of the power outage area of ​​the power grid; based on the line topology of the first load transfer area, compared with the total line topology, the graph neural network cost analyzer is separated to obtain the load transfer area cost analyzer.

[0009] The steps for building the graph neural network cost analyzer set include: Step 1: Enumerate and decompose the target area line topology into PQ unit line topologies to obtain several PQ unit line topologies, where the initial value of P is equal to 1, the initial value of Q is equal to 1, and the PQ unit line topology includes Q power nodes and P load nodes. P and Q are both integers, P≤the total number of load nodes, and Q≤the total number of power nodes. Step 2: Traverse the PQ unit line topologies, use the power distribution location as the input node and the load distribution location as the output node, perform graph neural network simulation, and construct the PQ unit line graph neural network architecture; Step 3: Using the power supply amount as input data and the power supply duration and power loss as output data, train the neural network architectures of the plurality of PQ unit circuit diagrams to generate a plurality of PQ unit circuit cost analyzers; Step 4: When P < the total number of load nodes, P increases by one and returns to step 1 to execute the loop: When Q is less than the total number of power nodes, Q is increased by one and the loop returns to step 1. When Q = the total number of power nodes and P < the total number of load nodes, P is increased by one and the loop returns to step 1. Step 5: When P = the total number of load nodes and Q < the total number of power nodes, Q is increased by one and the loop returns to step 1. Step 6: When Q = the total number of power nodes and P = the total number of load nodes, output all PQ unit line cost analyzers, store them in association with the target area, and add them to the graph neural network cost analyzer.

[0010] Optionally, the minute-level collaborative fault recovery method for distribution networks of grid-type inverter-type power sources also includes: when the recovery time is greater than a time threshold, or / and when the lost power is greater than a loss threshold, updating the distributed collaborative distribution solution and the grid-type collaborative power solution, and executing a cycle.

[0011] The minute-level coordinated fault recovery method for distribution networks with grid-connected inverter power supplies also includes: When the preset number of cycles fails to converge, a recovery power supply scheme in which the power loss is less than or equal to the loss threshold is extracted; when the recovery power supply scheme is not empty, the scheme with the shortest recovery time among the recovery power supply schemes is extracted for power supply; when the recovery power supply scheme is empty, the normalized recovery time of each power supply scheme is weighted based on a predefined time weight to obtain a plurality of first weighted features; based on a predefined loss weight, the normalized power loss of each power supply scheme is weighted to obtain a plurality of second weighted features; the one-to-one corresponding plurality of first weighted features and the plurality of second weighted features are weighted to obtain a plurality of power supply scheme fitnesses; the power supply scheme corresponding to the minimum value of the fitnesses of the plurality of power supply schemes is extracted for power supply.

[0012] Among them, the duration weight ∈ [0.8, 1], and the loss weight ∈ [0, 0.2].

[0013] The PQ unit line topology includes Q power supply nodes and P load nodes, and further includes: the P load nodes can be powered by the Q power supply nodes.

[0014] By implementing the present invention, when a power grid fault is triggered, the load transfer area and the key island area can be extracted. The key island area is the power grid outage area not covered by the load transfer area, which can achieve accurate division of the power outage area, lay the foundation for the subsequent targeted power supply plan, avoid the waste of power supply resources and the blindness of power supply strategy, and ensure that each area can obtain the appropriate power supply method. By implementing the present invention, it is possible to configure a distributed collaborative power distribution solution for the load transfer area, and a grid-based collaborative power supply solution for the key island area. Through analysis by a graph neural network cost analyzer, the recovery time and power loss are generated. By configuring differentiated power supply solutions, the advantages of different power sources are fully utilized (distributed power sources are suitable for transfer areas, and grid-based and grid-based inverter power sources are collaboratively suitable for key island areas). The graph neural network cost analyzer can be used to efficiently and accurately analyze the recovery time and power loss, providing a key basis for subsequent judgment on the feasibility of the solution, ensuring the scientific nature and timeliness of the analysis. By implementing the present invention, it can be achieved that when the recovery time is less than or equal to the time threshold, and when the lost electric energy is less than or equal to the loss threshold, the distributed collaborative distribution solution is used to control the distributed power supply to supply power to the load transfer area, and the grid-following collaborative power supply solution is used to control the grid-following inverter power supply and the grid-following inverter power supply to supply power to the key area of ​​the island, ensuring that the power supply plan can be restored within the specified time and the electric energy loss is within an acceptable range, thereby ensuring the timeliness and economy of power supply and effectively improving the reliability and efficiency of power grid fault recovery.

[0015] In summary, by implementing the present invention, the speed, accuracy and economy of distribution network fault recovery can be improved, and the resilience and power supply reliability of the distribution network in the event of a fault can be enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of a method for minute-level coordinated fault recovery of a distribution network for a grid-connected inverter power supply provided by the present invention; Figure 2 A schematic diagram of a line topology diagram of a power outage area; Figure 3 This is a schematic diagram of the key island area in the line topology diagram of the power grid blackout area. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0019] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0020] Example 1, as Figure 1 As shown, an embodiment of the present invention provides a minute-level coordinated fault recovery method for a distribution network of a grid-type inverter power supply, including: S100: When a power grid fault is triggered, extracting a load transfer area and an island key area, wherein the island key area is a power grid blackout area not covered by the load transfer area; S200: configuring a distributed collaborative power distribution solution for the load transfer area, configuring a structure-grid collaborative power supply solution for the island key area, and generating restoration time and power loss through a graph neural network cost analyzer; S300: When the recovery time is less than or equal to the time threshold, and when the lost electric energy is less than or equal to the loss threshold, the distributed collaborative distribution system is used to control the distributed power source to supply power to the load transfer area, and the grid-type inverter power source and the grid-type inverter power source are controlled to supply power to the key area of ​​the island through the grid-based collaborative power supply system.

[0021] In step S100 of the embodiment of the present application, when a power grid fault is triggered, the load transfer area and the island key area are extracted, including: When the grid connection point voltage is less than the voltage threshold, it is regarded as a grid fault trigger, and the grid blackout area is received; Extracting distributed power sources connected to the power outage area based on a line topology map of the power grid; Traversing the distributed power sources to perform power flow calculations and determine the load transfer area in the power grid blackout area; Extract the area not covered by the load transfer area in the power grid blackout area and set it as the initial island key area; The initial island key area is traversed to perform source-load balance analysis to obtain the island key area.

[0022] In the embodiment of this application, the core purpose is to accurately define the load transfer area and the key island area after the power grid failure, and provide a clear regional division basis for the configuration of the subsequent differentiated power supply plan. Specifically, by identifying the fault triggering conditions and locating the power outage area, combining the distribution of distributed power sources and the flow calculation to determine the range of load that can be transferred, and then screening the key island areas that require special coordinated power supply through source-load balance analysis, it ensures that resource scheduling is more targeted when the fault is restored and avoids the blindness of the power supply strategy. First, it is necessary to set the fault trigger condition, that is, the voltage threshold. When the grid connection point voltage is less than the voltage threshold, it is considered as a grid fault trigger and the grid blackout area is received. For example, the voltage threshold can be set as "grid connection point phase voltage U PCC-PP <0.85pu" or "grid-connected point phase voltage U PC-PG <0.9pu", where pu is the per-unit value, is used to unify voltage measurement standards. This determination is directly related to rapid response in the early stages of a fault, ensuring that the fault is caught in a timely manner.

[0023] When a fault is triggered, the system receives the overall scope information of the power grid outage area. This information is determined based on the real-time monitoring data such as node topology and load distribution uploaded before the distribution network fault, providing the basic boundary for subsequent area division.

[0024] Furthermore, it is necessary to extract the distributed power sources connected to the power outage area based on the line topology map of the power grid. This step requires the line topology map of the power grid outage area, which is presented in the form of a "graph theory model", as shown in the attached figure. Figure 2 As shown in the figure, each load branch, external system, and line bifurcation is a node. Lines with tie switches, section switches, and protective switches are disconnectable edges, represented by dashed lines. The remaining lines are normal edges, represented by solid lines. By traversing all nodes in the outage area in the line topology diagram, nodes connected to other distributed generation (DGs) are selected. The location, capacity, and operating status of the DGs connected to the outage area are determined based on pre-fault data. These DGs will be used for subsequent load transfer.

[0025] Furthermore, it is necessary to determine the load transfer area based on the flow calculation and construct a simplified flow calculation model for the distribution network. The input parameters of the model include: the line length and impedance parameters between each node; the load power connected to each node, that is, active power and reactive power; and the real-time output of the distributed power source.

[0026] Taking the non-fault nodes adjacent to the power outage area, such as the nodes in the load transfer node set as the starting point, input their voltage measurement values, such as the normal range of 0.85-1.15pu, and calculate the voltage backward node by node.

[0027] If the node voltage remains within the normal range during the calculation process, the calculation range is expanded across disconnectable edges such as tie switches and section switches. If the node voltage in a certain area exceeds the threshold, such as below 0.85 pu or above 1.15 pu, the expansion is stopped.

[0028] When there are distributed power sources, the voltage support capability is stronger and the transfer range is larger. Therefore, it is necessary to combine the support capability of distributed power sources to ultimately determine the maximum range of load transfer, that is, the load transfer area, which is set as the initial island key area.

[0029] The initial island key area is the portion of the power outage area that is not covered by the load transfer area, that is, the area where power cannot be restored through external systems or distributed power supply transfer. When extracting the initial island key area, it is necessary to compare the line topology of the power grid power outage area and remove the coverage of the load transfer area from the total range of the power grid power outage area. The remaining area is the initial island key area. For example, if the power outage area includes nodes 1-30, and the load transfer area covers nodes 1-10, then the initial island key area is nodes 11-30, and the specific adjustment needs to be based on the actual topology.

[0030] Finally, the islanding key area needs to be determined based on the source-load balance analysis. The source-load balance analysis is to match the power output and load demand in the initial islanding key area to determine whether the power supply in the area can meet the power demand of the load, that is, the active and reactive power balance.

[0031] The analysis logic is that if the source and load in the initial island key area are balanced, that is, the power output ≥ load demand, then this area is the final island key area, which needs to rely on the internal grid-type and grid-following inverter-type power supplies to coordinate and support power supply.

[0032] If the source and load are unbalanced, that is, the power output is less than the load demand, the regional boundary needs to be further expanded to include the surrounding power sources and repeat the above source-load balance analysis until the key island area that meets the conditions is determined, such as the attached Figure 3 The area formed by midpoints 10, 12, and 29.

[0033] If the initial islanding critical areas cannot meet the source-load balance after analysis, then there is no islanding critical area in the area, indicating that power supply can be fully restored through load transfer.

[0034] In step S100 of the embodiment of the present application, when the grid connection point voltage is less than the voltage threshold and before receiving the grid blackout area, it includes: adjusting the current overload capacity to a preset multiple of the rated current within k seconds, where 0≤k≤2.

[0035] In the embodiment of the present application, the purpose of setting up this mechanism is to provide immediate voltage support for the power outage area within a very short time after the power grid fault is triggered, such as within 0-2 seconds, by enhancing the current output capacity of the grid-forming inverter power supply, thereby avoiding the grid-forming inverter power supply from being disconnected from the grid due to voltage abnormalities, and at the same time buying time for subsequent load transfer areas and island key area extraction work, laying a stable initial power supply foundation.

[0036] Specifically, when it is detected that the grid connection point voltage is less than the voltage threshold, that is, the grid fault is triggered, before receiving information about the grid power outage area, the grid-connected inverter power supply will quickly adjust its current overload capacity within k seconds, 0≤k≤2, usually from the moment the fault occurs to 2 seconds, and adjust the upper and lower limits of the d-axis and q-axis current reference values ​​from the conventional 1.2 times the rated current to a preset 2 times the rated current, wherein the d-axis and q-axis current reference values ​​are key control parameters obtained based on the three-phase AC "Park transformation" and are used to describe the two components of the inverter power supply output current in the rotating coordinate system. The d-axis current, that is, the direct-axis current, is mainly related to active power and determines the amount of active power output by the power supply to the grid. The q-axis current, that is, the quadrature-axis current, is mainly related to reactive power and affects the power supply's ability to support the grid voltage, such as voltage amplitude regulation. In the fault recovery scenario of this application, the core role of the d-axis and q-axis current reference values ​​is to accurately control the output characteristics of the grid-connected inverter power supply.

[0037] Through this dynamic current limit adjustment, the grid-connected inverter power supply can output a larger current, thereby providing stronger voltage support for the power grid, maintaining the voltage in the power outage area at a certain level, and preventing the grid-connected inverter power supply from being disconnected from the grid due to voltage drops or fluctuations.

[0038] In step S200 of the embodiment of the present application, a distributed collaborative power distribution solution is configured for the load transfer area, and a structure-grid collaborative power supply solution is configured for the island key area. The graph neural network cost analyzer is used to analyze and generate the recovery time and power loss, including: Matching a graph neural network cost analyzer from a set of graph neural network cost analyzers according to the grid fault area, wherein the graph neural network cost analyzer has a label identifying the topology of all lines in the grid fault area; Based on the load transfer area line topology, compared with the overall line topology, the graph neural network cost analyzer is separated to obtain a load transfer area cost analyzer, and the distributed collaborative power distribution solution is processed to generate a first restoration time and a first power loss; Based on the line topology of the island key area, compared with the topology of all lines, the graph neural network cost analyzer is separated to obtain the island key area cost analyzer, and the structure-grid collaborative power solution is processed to generate the second recovery time and the second power loss; The larger value of the first recovery time and the second recovery time is set as the recovery time; The larger value of the first power loss and the second power loss is set as the power loss.

[0039] In the embodiment of the present application, a graph neural network cost analyzer is matched to the fault area to ensure that the subsequent power supply plan analysis for the load transfer area and the key island area is accurate and targeted. By matching the graph neural network cost analyzer corresponding to the topology of all lines in the fault area of ​​the power grid, a reliable model foundation is provided for analyzing the recovery time and power loss of distributed collaborative distribution solutions and grid-connected collaborative power supply solutions, ensuring that the analysis results can truly reflect the actual situation of the fault area and support the optimization decision of the subsequent power supply plan.

[0040] In step S200 of the embodiment of the present application, the steps of constructing a graph neural network cost analyzer set include: Step 1: Enumerate and decompose the target area line topology into PQ unit line topologies to obtain several PQ unit line topologies, where the initial value of P is equal to 1, the initial value of Q is equal to 1, and the PQ unit line topology includes Q power nodes and P load nodes. P and Q are both integers, P≤the total number of load nodes, and Q≤the total number of power nodes. Step 2: Traverse the PQ unit line topologies, use the power distribution location as the input node and the load distribution location as the output node, perform graph neural network simulation, and construct the PQ unit line graph neural network architecture; Step 3: Using the power supply amount as input data and the power supply duration and power loss as output data, train the neural network architectures of the plurality of PQ unit circuit diagrams to generate a plurality of PQ unit circuit cost analyzers; Step 4: When P < the total number of load nodes, P increases by one and returns to step 1 to execute the loop: When Q is less than the total number of power nodes, Q is increased by one and the loop returns to step 1. When Q = the total number of power nodes and P < the total number of load nodes, P is increased by one and the loop returns to step 1. Step 5: When P = the total number of load nodes and Q < the total number of power nodes, Q is increased by one and the loop returns to step 1. Step 6: When Q = the total number of power nodes and P = the total number of load nodes, output all PQ unit line cost analyzers, store them in association with the target area, and add them to the graph neural network cost analyzer.

[0041] In an embodiment of the present application, by enumerating all combinations of power nodes (Q) and load nodes (P) and training the corresponding graph neural network cost analyzer, efficient analysis of power supply schemes under any power-load configuration is achieved, providing computing power support for minute-level fault recovery.

[0042] First, in step 1, the target substation's line topology needs to be enumerated and decomposed into PQ unit line topologies. This means taking the target substation's complete line topology as the basis and decomposing it into several "PQ unit line topologies." Here, P represents the number of load nodes, with an initial value of 1, and Q represents the number of power nodes, with an initial value of 1. P ≤ total number of load nodes, and Q ≤ total number of power nodes. Each PQ unit topology must satisfy the basic logic that P load nodes can be powered by Q power nodes, ensuring that the unit topology is practically feasible for power supply.

[0043] Then, in step 2, we need to build the PQ unit circuit diagram neural network architecture. We need to traverse all the disassembled PQ unit topologies. The “power distribution location” in each unit is used as the input node, and the “load distribution location” is used as the output node.

[0044] The input nodes, or power distribution locations, are the core input data for graph neural networks. For example, the power distribution location is represented by the coordinate information in the node features, which the graph neural network uses as a starting point to analyze the power supply range.

[0045] The output nodes, or load distribution locations, are characterized by their characteristics, which are the analysis targets of the graph neural network. The graph neural network uses input node features and edge features (line attributes) to calculate whether the power supply can cover the load, the required time and losses, and ultimately outputs a result corresponding to the load distribution location.

[0046] Through simulation training of a graph neural network (GNN), a neural network architecture corresponding to the topology of each PQ unit is constructed. The characteristics of GNN enable it to accurately capture the topological relationships between nodes, laying the structural foundation for subsequent cost analysis.

[0047] In step 3, the PQ unit line cost analyzer can be built using a gated graph neural network. This model can effectively handle the topology and timing characteristics of the power system and accurately capture the complex interaction between power nodes and load nodes. Among them, the features of the unit line cost analyzer include node features and edge features.

[0048] The node features of the unit line cost analyzer include power source node features and load node features. Power source node features include power source type, such as distributed power source, grid-forming type, and grid-following type; capacity (kW), maximum overload capacity, such as the 2x rated current mentioned in the previous step; and real-time output (pu), which measures the power supply. Load node features include load type, such as residential, industrial, and commercial; power demand (kW); and priority, such as primary load and secondary load. The unified feature dimension is 64, ensuring consistency across all node feature dimensions.

[0049] The edge features of the unit line cost analyzer include line impedance (R, X), length, rated current, and switch status, such as closed / open. The edge feature dimension is 32.

[0050] The PQ unit line cost analyzer is structured as a four-layer GGNN to capture local and global topological information of the power network. The GRU unit has a hidden dimension of 128 to enhance time series modeling capabilities. The output structure is a fully connected layer with output dimensions of 64, 32, and 2. The last two layers of dimensional features correspond to power supply duration prediction and power loss prediction, respectively.

[0051] Input data is the core variable used by the Unit Line Cost Analyzer for calculations, specifically "power supply capacity." This refers to the actual power output of the power supply, such as the real-time output of distributed power sources and the output power of grid-connected power sources. This is a dynamically changing operating parameter that directly reflects the power supply's current power supply capacity.

[0052] The output data is the result of the unit line cost analyzer's calculations, specifically "power supply duration" and "power loss." These data represent the time required to restore load power through the power supply solution and the power loss during line transmission during the restoration process. These are key indicators of the power supply solution's efficiency.

[0053] For the PQ unit circuit cost analyzer, the learning rate was set to 0.001, with adaptive adjustment using the Adam optimizer. The batch size was 64. Regularization used an L2 regularization coefficient of 0.0001 and a dropout rate of 0.2. The hidden layer activation function was LeakyReLU with α = 0.01. The output layer had no activation function. The number of training epochs was 200.

[0054] The following data in the PQ unit line cost analyzer is historical fault data, extracted from historical distribution network fault records. This data includes fault location, power source distribution, load data, restoration time, and lost energy. A digital twin model can also be constructed based on the actual distribution network topology, using Monte Carlo simulation to generate fault scenarios, such as faults at different locations and power output fluctuations under different weather conditions. Combining historical data with simulation data ensures that the sample covers a wide range of possible fault types and topologies.

[0055] For each PQ unit line topology, such as 1 power supply-1 load, 1 power supply-2 loads, etc., at least 10,000 training samples are generated. When the mean absolute error (MAE) on the validation set decreases by less than 0.01% for 10 consecutive rounds, training is stopped and the PQ unit line cost analyzer is obtained. The power supply amount is input to the PQ unit line cost analyzer of the corresponding PQ unit line to predict the output power supply duration and power loss. Using the same method, several PQ unit line cost analyzers are generated. For example, consider a PQ unit line topology with one source node (Q=1) and two load nodes (P=2). Assume the source node features: a grid-connected inverter power source with a capacity of 1000 kW, a maximum overload capability of 2 times the rated current, located at the start of line A, and supplying 800 kW. Load node features: Load 1, a residential load with a power demand of 300 kW, located 1 km from the source node; Load 2, an industrial load with a power demand of 400 kW, located 2 km from the source node. Edge features: a line impedance of 0.05 Ω / km and a closed switch state.

[0056] The PQ unit line cost analyzer with the label "1 power node, 2 load nodes" is matched from the graph neural network cost analyzer. The power supply capacity of 800 kilowatts, node features such as power capacity and load demand, and edge features such as line impedance and length are input into the PQ unit line cost analyzer. The PQ unit line cost analyzer, through graph neural network analysis, outputs a recovery time of 15 minutes and an energy loss of 12 kilowatt-hours.

[0057] The core purpose of steps 4 to 6 in the embodiment of the present application is to enumerate all possible PQ unit line topology combinations in the target substation by cyclically increasing the number of power nodes (Q) and the number of load nodes (P), ensuring that the constructed graph neural network cost analyzer set can cover all power and load configuration scenarios in the substation. By exhaustively combining, any possible power-load combination is avoided, thereby ensuring that in the event of a power grid failure, the analyzer corresponding to the actual fault scenario can always be matched from the set, laying the foundation for accurately analyzing the power supply plan in step S200.

[0058] Assume that the initial conditions are: the initial value of P is 1, the initial value of Q is 1, and P ≤ the total number of load nodes, Q ≤ the total number of power nodes.

[0059] When P is less than the total number of load nodes, increase P by 1. For example, if P changes from 1 to 2, return to step 1 and re-disassemble the PQ unit line topology, that is, generate a topology of Q power nodes and new P load nodes, and repeat the subsequent training steps.

[0060] When P reaches the total number of load nodes, if Q is less than the total number of power nodes, increase Q by 1. For example, if Q changes from 1 to 2, return to step 1 and re-disassemble the topology to generate a topology with "new Q power nodes and P = total number of load nodes and load nodes", and repeat the training steps.

[0061] When Q reaches the total number of power supply nodes and P is less than the total number of load nodes, increase P by 1 and return to step 1 to disassemble the topology, for example, Q = the total number of power supply nodes and P changes from 2 to 3.

[0062] When P reaches the total number of load nodes and Q reaches the total number of power nodes, the loop stops, and all trained PQ unit line cost analyzers are associated with the target substation and stored, and included in the graph neural network cost analyzer set.

[0063] For example, if the target area has 2 power nodes, i.e. total Q = 2, and 3 load nodes, i.e. total P = 3, the loop process is: P=1, Q=1→P=2, Q=1→P=3, Q=1→Q=2, P=1→Q=2, P=2→Q=2, P=3→The loop ends and 6 types of PQ unit analyzers are generated.

[0064] In step S200 of the embodiment of the present application, based on the load transfer area line topology, the graph neural network cost analyzer is separated and compared with the overall line topology to obtain a load transfer area cost analyzer, including: Extracting a first load transfer area from the load transfer area, wherein the first load transfer area has an accessible distributed power source position identifier; According to the accessible distributed power source identifier, the power supply configuration is performed on the first load transfer area with the power demand of the first load transfer area before the fault as the target, and a coordinated power distribution solution for the first load transfer area is obtained; Extracting a first load transfer area line topology from a power grid blackout area line topology map according to the accessible distributed power source identifier and the first load transfer area; Based on the first load transfer area line topology and compared with all line topologies, the graph neural network cost analyzer is separated to obtain a load transfer area cost analyzer.

[0065] In the embodiment of the present application, the core purpose of the above steps is to separate a cost analyzer specifically suitable for the load transfer area from the matched graph neural network cost analyzer, ensuring that the distributed collaborative power distribution analysis of the load transfer area is targeted and accurate. By extracting areas with accessible distributed power sources, configuring power supply plans, and separating the load transfer area cost analyzer of the corresponding topology, a dedicated tool is provided for calculating the recovery time and power loss of the area, ensuring that the analysis results are consistent with the actual power supply needs and topological characteristics of the load transfer area, and supporting the optimization decision of subsequent power supply plans.

[0066] First, we need to extract the first load transfer area. From this identified load transfer area, we filter out the sub-areas with "accessible distributed power generation tag identifiers," which are the first load transfer areas. "Accessible distributed power generation tag identifiers" indicate that the area has clear distributed power generation access points, such as distributed power generation interfaces numbered D1 and D2, which clearly indicate that power can be supplied by distributed power generation.

[0067] Then, it is necessary to configure the first load transfer area collaborative power distribution solution for the first load transfer area based on the bit number identifier of the accessible distributed power source and the power demand of the first load transfer area before the fault.

[0068] Specifically, based on the bit number identification of the accessible distributed power source, with the "power demand in the area before the fault" as the goal, for example, the area required 1000kW of electricity before the fault, the output distribution and access method of the distributed power source can be configured, such as scheduling the D1 power source to provide 600kW and the D2 power source to provide 400kW, to form a power supply plan for the sub-area, namely the first load transfer area collaborative distribution solution, as a sub-plan of the distributed collaborative distribution solution.

[0069] Next, based on the accessible distributed generation unit identifier and the first load transfer area, the line topology of the first load transfer area must be extracted from the line topology map of the power grid's blackout area. Specifically, the line topology of the sub-area, including internal nodes, branch connection relationships, and the locations of distributed generation unit access points, is extracted from the overall line topology map of the blackout area, combining the accessible distributed generation unit identifier and the scope of the first load transfer area.

[0070] Finally, based on the first load transfer area's line topology and comparing it with the overall line topology, the graph neural network cost analyzer is separated to obtain a load transfer area cost analyzer. Specifically, the extracted first load transfer area line topology is compared with the overall line topology of the substation. From the matched graph neural network cost analyzers, a graph neural network cost analyzer that only contains the topology of this sub-area is separated, i.e., the load transfer area cost analyzer. This graph neural network cost analyzer analyzes only the topology and power supply configuration of the first load transfer area.

[0071] By merging and integrating the coordinated distribution solutions of each load transfer area, we can obtain the distributed coordinated distribution solution. The distributed coordinated distribution solution is an overall power supply plan covering the entire load transfer area, including the coordinated distribution solution of the first load transfer area, and also covers the coordinated distribution solutions of other sub-areas within the load transfer area.

[0072] Furthermore, a load transfer area cost analyzer is required to process the distributed collaborative power distribution solution to generate a first restoration duration and a first power loss. Specifically, the distributed collaborative power distribution solution configured for a load transfer area is input into the isolated load transfer area cost analyzer. Based on the line topology and power supply plan for that area, the analyzer calculates the first restoration duration required for power restoration in that area and the first power loss during the restoration process.

[0073] Then, it is necessary to refer to the logic of the weight-separated load transfer area cost analyzer, compare the line topology of the island key area with the entire line topology, and separate the "island key area cost analyzer" that is only applicable to the island key area topology from the graph neural network cost analyzer. The grid-forming and grid-following collaborative power supply solution configured for the island key area, that is, the collaborative output plan of the grid-forming and grid-following inverter power sources, is input into the island key area cost to calculate the second recovery time and the second power loss of the area.

[0074] Compare the first recovery time and the second recovery time, and use the larger value as the recovery time for the entire fault recovery process to ensure that both areas are fully recovered.

[0075] Compare the first power loss and the second power loss, and take the larger value as the overall "power loss" to ensure that the total loss is within an acceptable range.

[0076] Suppose that after a distribution network failure, a load transfer area has been designated: DGs D1 (500kW) and D2 (300kW) are accessible, with a pre-fault power demand of 600kW. The critical island area, encompassing important hospitals and government agencies, requires coordinated power supply from grid-connected power source G1 (1000kW) and grid-connected power source G2 (800kW). The pre-fault power demand was 1200kW.

[0077] The distributed collaborative power distribution solution provides 400kW for the dispatching D1 power supply and 200kW for the D2 power supply, which are supplied through the line L1-L2-L3. The load transfer regional cost analyzer calculates that the first recovery time is 8 minutes and the first loss of electricity is 5.2kWh.

[0078] The grid-coordinated power supply solution for the island's critical area is for power source G1 to provide 700kW and power source G2 to provide 500kW, delivered via lines L4-L5-L6. The island's critical area cost analyzer calculated a second-order recovery time of 12 minutes and a second-order energy loss of 8.7kWh.

[0079] The final recovery time is max(8 minutes, 12 minutes) = 12 minutes; the power loss is max(5.2 kWh, 8.7 kWh) = 8.7 kWh.

[0080] The above steps analyze the power supply plans of the two areas in a targeted manner and take the stricter and larger value of the two as the overall evaluation standard to ensure that the fault recovery plan can simultaneously meet the recovery requirements of both areas, avoid affecting the overall power supply reliability due to non-compliance in one area, and provide a unified quantitative basis for judging whether the plan is feasible.

[0081] In step S300 of the embodiment of the present application, when the recovery time is less than or equal to the time threshold, and when the lost electric energy is less than or equal to the loss threshold, the distributed collaborative distribution solution is used to control the distributed power source to supply power to the load transfer area, and the grid-type inverter power source and the grid-type inverter power source are controlled to supply power to the key area of ​​the island through the grid-following collaborative power supply solution.

[0082] This step is to perform corresponding power supply control on the load transfer area and island key area, under the premise that the power supply plan meets the recovery time and power loss threshold. The duration threshold and loss threshold are set with "minute-level recovery" as the core goal. For example, the duration threshold refers to the industry standard for distribution network fault recovery and the support capability of the grid-type power supply, for example, it can be set to ≤60 minutes; the loss threshold combines the line transmission efficiency, power output limit and economic cost to ensure that the loss does not affect the system stability and meets the economic requirements, for example, it can be set to ≤50kWh, while being compatible with the fluctuation range under extreme working conditions, balancing the recovery speed and energy efficiency.

[0083] In this step, following the example in the above steps, if the recovery time is 12 minutes, ≤ 60 minutes, and the power loss is 8.7 kWh, ≤ 50 kWh, then in the load transfer area, according to the distributed collaborative distribution solution, control the distributed power supply D1 to output 400 kW and D2 to output 200 kW, close the connecting switch L1-2, and supply power to the area; in the key area of ​​the island, according to the structure and grid collaborative power supply solution, control the structure and grid type power supply G1 to output 700 kW and the grid type power supply G2 to output 500 kW, close the section switch L4-5, and supply power to the area.

[0084] Eventually, power was restored to both areas, completing fault recovery.

[0085] In step S300 of the embodiment of the present application, when the recovery time is greater than the time threshold, or / and when the lost electric energy is greater than the loss threshold, the distributed collaborative distribution solution and the structure-grid collaborative power solution are updated and a cycle is executed.

[0086] When the preset number of cycles fails to converge, extracting a power supply recovery plan in which the power loss is less than or equal to the power loss threshold; When the power supply restoration plan is not empty, extract the plan with the shortest restoration time from the power supply restoration plans to supply power; When the power restoration plan is empty, weighting the normalized value of the restoration duration of each power restoration plan based on a predefined duration weight to obtain a plurality of first weighted features; Based on predefined loss weights, weighting the normalized value of the power loss of each power supply scheme to obtain multiple second weighted features; Weighting the one-to-one corresponding multiple first weighted features and the multiple second weighted features to obtain multiple power supply scheme adaptabilities; A power supply scheme corresponding to a minimum value of the adaptability of the multiple power supply schemes is extracted for power supply.

[0087] Among them, the duration weight ∈ [0.8, 1], and the loss weight ∈ [0, 0.2].

[0088] In the embodiment of the present application, the purpose of the above steps is to gradually approach the optimal solution that meets the threshold by iteratively updating the power supply plan when the recovery time or the power loss exceeds the threshold.

[0089] For example, assuming that the initial power supply plan for the load transfer area and the key island area does not meet the threshold, that is, the above-mentioned duration threshold of 60 minutes and the loss threshold of 50kWh, the particle swarm algorithm can be used to update the distributed collaborative distribution solution and the grid-based collaborative power supply solution, setting the number of particles to 5 and the number of iterations to 3.

[0090] Assume the following initial solution parameters: Distributed generation outputs (D1) in the load transfer area are 300 kW and 200 kW, respectively. The grid-forming / following generation outputs in the key island area are G1 = 500 kW and G2 = 400 kW. Using the method described above, the restoration time is calculated to be 70 minutes, with a power loss of 55 kWh.

[0091] First, particle encoding is required. Each particle represents a set of power output schemes (D1, D2, G1, G2). The initial particle group is generated, such as: particle 1 (350, 250, 550, 450); particle 2 (400, 200, 600, 500); particle 3 (320, 280, 520, 480); particle 4 (380, 220, 580, 420); particle 5 (420, 180, 620, 520).

[0092] Next, the fitness of each particle group needs to be evaluated. This can be calculated by combining the recovery time and energy loss. The recovery time and energy loss values ​​can be obtained from the parameters represented by the particles using the graph neural network cost analyzer described above. These values ​​are expressed in minutes and kilowatt-hours, and are not described here. The calculated fitness values ​​are as follows: Particle 1 has a recovery time of 65, an energy loss of 52, and a particle fitness of 117; Particle 2 has a recovery time of 58, an energy loss of 49, and a particle fitness of 107, making it the optimal individual; Particle 3 has a recovery time of 68, an energy loss of 53, and a particle fitness of 121; Particle 4 has a recovery time of 62 minutes, an energy loss of 50, and a particle fitness of 112; Particle 5 has a recovery time of 60, an energy loss of 51, and a particle fitness of 111.

[0093] By comparing the fitness of particles, we can find that the global optimal particle is particle 2, which guides other particles to approach it.

[0094] After the second iteration, particle 2 is updated to (410, 210, 610, 510). As a result, the recovery time is 55 minutes, the power consumption is 48 kWh, and the particle fitness is 103.

[0095] After the third iteration, particle 2's recovery time was 55 minutes (≤60) and its energy loss was 48 kWh (≤50), meeting the threshold. The updated power supply plan is: load transfer area: D1 = 410 kW, D2 = 210 kW, distributed coordinated distribution and control power supply; key island area: G1 = 610 kW, G2 = 510 kW, structured and grid coordinated power supply and control power supply.

[0096] Assuming that convergence has not occurred after three iterations, that is, no power supply plan simultaneously satisfies the conditions of duration ≤ 60 minutes and loss ≤ 50 kWh, then a power recovery plan with the power loss less than or equal to the loss threshold is extracted, that is, a plan with less power loss is given priority for power supply.

[0097] When the power supply recovery plan is not empty, the plan with the shortest recovery time in the power supply recovery plan is extracted for power supply. That is, when there are multiple power supply recovery plans, the plan with the shortest recovery time in the power supply recovery plan is preferentially selected for power supply to shorten the power failure time as much as possible.

[0098] When the power supply recovery plan is empty, the normalized value of the recovery time of each power supply plan is weighted based on the predefined time weight to obtain multiple first weighted features; based on the predefined loss weight, the normalized value of the loss of electric energy of each power supply plan is weighted to obtain multiple second weighted features; the one-to-one corresponding multiple first weighted features and the multiple second weighted features are weighted to obtain multiple power supply plan adaptabilities; the power supply plan corresponding to the minimum value of the adaptabilities of the multiple power supply plans is extracted for power supply.

[0099] Specifically, the recovery time of multiple power supply schemes is normalized first, and the normalized recovery time value = (current recovery time value - minimum recovery time value) / (maximum recovery time value - minimum recovery time value). Then, the power loss of multiple power supply schemes is normalized, and the normalized power loss value = (current power loss value - minimum power loss value) / (maximum power loss value - minimum power loss value).

[0100] Assume that in a multi-group power supply solution, the maximum recovery time is 70 minutes, the minimum recovery time is 55 minutes, and the recovery time threshold is 60 minutes; the maximum power loss is 55 kWh, the minimum power loss is 51 kWh, and the power loss threshold is 50 kWh.

[0101] Assume there are three power supply plans. If the recovery time for Plan A is 65 minutes, the normalized recovery time is (65-55) / (70-55)=10 / 15≈0.67. Similarly, if the recovery time for Plan B is 58 minutes, the normalized recovery time is 0.2; if the recovery time for Plan C is 70 minutes, the normalized recovery time is 1.0.

[0102] If the energy loss in Plan A is 53kWh, the normalized energy loss value is (53-51) / (55-51)=2 / 4=0.5. Similarly, if the energy loss in Plan B is 55kWh, the normalized energy loss value is 1.0, and if the energy loss in Plan C is 51kWh, the normalized energy loss value is 0.0.

[0103] Based on the urgency of the region's electricity demand, we can set the duration weight to 0.9 and the loss weight to 0.1. Therefore, the fitness of Plan A is 0.67 × 0.9 + 0.5 × 0.1 = 0.603 + 0.05 = 0.653; the fitness of Plan B is 0.2 × 0.9 + 1.0 × 0.1 = 0.18 + 0.1 = 0.28; and the fitness of Plan C is 1.0 × 0.9 + 0.0 × 0.1 = 0.9.

[0104] Then, solution B with the smallest weighted feature is selected for power supply.

[0105] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0106] Those skilled in the art will appreciate that embodiments of the present invention may provide methods, systems, or computer program products. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0107] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0108] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0110] Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts become known.

[0111] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A minute-level collaborative fault recovery method for distribution networks of grid-connected inverter power supplies, characterized in that: include: When a power grid fault is triggered, the load transfer area and the island key area are extracted. The island key area is the power grid outage area not covered by the load transfer area. A distributed collaborative power distribution solution is configured for the load transfer area, and a network-based collaborative power supply solution is configured for the key island area. The restoration time and power loss are generated through a graph neural network cost analyzer. When the recovery time is less than or equal to the time threshold, and when the lost electric energy is less than or equal to the loss threshold, the distributed collaborative distribution solution is used to control the distributed power source to supply power to the load transfer area, and the grid-following collaborative power supply solution is used to control the grid-following inverter power source and the grid-following inverter power source to supply power to the key area of ​​the island.

2. The method according to claim 1, wherein When a grid fault is triggered, load transfer areas and island key areas are extracted, including: When the grid connection point voltage is less than the voltage threshold, it is regarded as a grid fault trigger, and the grid blackout area is received; Extracting distributed power sources connected to the power outage area based on a line topology map of the power grid; Traversing the distributed power sources to perform power flow calculations and determine the load transfer area in the power grid blackout area; Extract the area not covered by the load transfer area in the power grid blackout area and set it as the initial island key area; The initial island key area is traversed to perform source-load balance analysis to obtain the island key area.

3. The method according to claim 2, wherein When the voltage at the grid connection point is less than the voltage threshold, before receiving the power grid blackout area, it includes: adjusting the current overload capacity to a preset multiple of the rated current within k seconds, where 0≤k≤2.

4. The method according to claim 1, wherein A distributed collaborative power distribution solution is configured for the load transfer area, and a structure-grid collaborative power supply solution is configured for the key island area. The graph neural network cost analyzer is used to analyze and generate the recovery time and power loss, including: Matching a graph neural network cost analyzer from a set of graph neural network cost analyzers according to the grid fault area, wherein the graph neural network cost analyzer has a label identifying the topology of all lines in the grid fault area; Based on the load transfer area line topology, compared with the overall line topology, the graph neural network cost analyzer is separated to obtain a load transfer area cost analyzer, and the distributed collaborative power distribution solution is processed to generate a first restoration time and a first power loss; Based on the line topology of the island key area, compared with the topology of all lines, the graph neural network cost analyzer is separated to obtain the island key area cost analyzer, and the structure-grid collaborative power solution is processed to generate the second recovery time and the second power loss; The larger value of the first recovery time and the second recovery time is set as the recovery time; The larger value of the first power loss and the second power loss is set as the power loss.

5. The method according to claim 4, wherein Based on the load transfer area line topology, the graph neural network cost analyzer is separated and compared with the overall line topology to obtain a load transfer area cost analyzer, including: Extracting a first load transfer area from the load transfer area, wherein the first load transfer area has an accessible distributed power source position identifier; According to the accessible distributed power source identifier, the power supply configuration is performed on the first load transfer area with the power demand of the first load transfer area before the fault as the target, and a coordinated power distribution solution for the first load transfer area is obtained; Extracting a first load transfer area line topology from a power grid blackout area line topology map according to the accessible distributed power source identifier and the first load transfer area; Based on the first load transfer area line topology and compared with all line topologies, the graph neural network cost analyzer is separated to obtain a load transfer area cost analyzer.

6. The method according to claim 4, wherein The steps to build the graph neural network cost analyzer set include: Step 1: Enumerate and decompose the target area line topology into PQ unit line topologies to obtain several PQ unit line topologies, where the initial value of P is equal to 1, the initial value of Q is equal to 1, and the PQ unit line topology includes Q power nodes and P load nodes. P and Q are both integers, P≤the total number of load nodes, and Q≤the total number of power nodes. Step 2: Traverse the PQ unit line topologies, use the power distribution location as the input node and the load distribution location as the output node, perform graph neural network simulation, and construct the PQ unit line graph neural network architecture; Step 3: Using the power supply amount as input data and the power supply duration and power loss as output data, train the neural network architectures of the plurality of PQ unit circuit diagrams to generate a plurality of PQ unit circuit cost analyzers; Step 4: When P < the total number of load nodes, P increases by one and returns to step 1 to execute the loop: When Q is less than the total number of power nodes, Q is increased by one and the loop returns to step 1. When Q = the total number of power nodes and P < the total number of load nodes, P is increased by one and the loop returns to step 1. Step 5: When P = the total number of load nodes and Q < the total number of power nodes, Q is increased by one and the loop returns to step 1. Step 6: When Q = the total number of power nodes and P = the total number of load nodes, output all PQ unit line cost analyzers, store them in association with the target area, and add them to the graph neural network cost analyzer.

7. The method according to claim 1, wherein Also includes: When the recovery time is greater than the time threshold, or / and when the lost electric energy is greater than the loss threshold, the distributed collaborative power distribution solution and the structure-grid collaborative power supply solution are updated and a cycle is executed.

8. The method according to claim 7, wherein Also includes: When the preset number of cycles fails to converge, extracting a power supply recovery plan in which the power loss is less than or equal to the power loss threshold; When the power supply restoration plan is not empty, extract the plan with the shortest restoration time from the power supply restoration plans to supply power; When the power restoration plan is empty, weighting the normalized value of the restoration duration of each power restoration plan based on a predefined duration weight to obtain a plurality of first weighted features; Based on predefined loss weights, weighting the normalized value of the power loss of each power supply scheme to obtain multiple second weighted features; Weighting the one-to-one corresponding multiple first weighted features and the multiple second weighted features to obtain multiple power supply scheme adaptabilities; A power supply scheme corresponding to a minimum value of the adaptability of the multiple power supply schemes is extracted for power supply.

9. The method according to claim 8, wherein The duration weight ∈ [0.8, 1], and the loss weight ∈ [0, 0.2].

10. The method according to claim 6, wherein The PQ unit line topology includes Q power supply nodes and P load nodes, and further includes: the P load nodes can be powered by the Q power supply nodes.

Images