Distributed load transfer path searching method, device, equipment, medium and product

Abstract

The application discloses a kind of distributed load transfer path search method, device, equipment, medium and product, it is related to load transfer technical field, the method includes: when load transfer demand triggers, the storage relationship mapping table between distributed knowledge graph and each distributed node in target distribution network is obtained;The atlas is divided into several subgraphs according to power supply partition and voltage level, each subgraph is stored in corresponding node, and the same node has the subgraph of different voltage levels in the same partition;Determine starting subgraph, terminal subgraph and target distributed node according to storage relationship mapping table;In each target node, hierarchical path expansion search is executed, to generate candidate path;Candidate path is spliced across partition and safety check, and the optimal load transfer path is determined by comprehensive evaluation and sorting, the present application can solve the technical problems of low search efficiency caused by dependence on traditional graph traversal algorithm in existing distributed load transfer technology.

Description

Technical Field

[0001] This invention relates to the field of load transfer technology, and in particular to a method, apparatus, equipment, medium and product for searching distributed load transfer paths. Background Technology

[0002] With the continuous expansion of distribution network scale and the large-scale integration of distributed power sources, the power grid topology is becoming increasingly complex. When a distribution network fails or requires planned maintenance, the dispatching system needs to search for feasible power supply paths from a massive number of possible power transfer schemes within a very short time to quickly restore power supply to non-faulty sections. Therefore, how to efficiently and quickly search for load transfer paths in large-scale and complex distribution networks has become a key problem that urgently needs to be solved in the field of power system dispatching.

[0003] Currently, load transfer path search methods based on knowledge graphs or graph databases have been proposed and are gradually being applied. However, existing technologies all rely on traditional graph traversal algorithms (such as Dijkstra's algorithm, breadth-first search, depth-first search, etc.) to perform global path searches on the overall knowledge graph or graph model. These algorithms need to traverse all adjacency relationships of the current node each time a path is expanded. As the number of nodes in the distribution network increases, the search space grows exponentially, leading to a sharp increase in the computational overhead of path search. Although existing methods use graph parallel computing technology to distribute the computational tasks across multiple nodes, they are essentially still distributed deployments of traditional graph traversal algorithms and do not fundamentally change the global traversal search mode. That is, each computing node still needs to perform a complete graph traversal operation on its own subgraph. Therefore, existing technologies lack specialized indexing mechanisms designed for distributed storage structures, and the efficiency bottlenecks of neighbor expansion and path localization have not been effectively overcome, thus failing to fully leverage the performance advantages of distributed architectures. Summary of the Invention

[0004] This invention provides a method, apparatus, device, medium, and product for searching distributed load transfer paths, which can solve the technical problem of low search efficiency caused by the reliance on traditional graph traversal algorithms in existing distributed load transfer technologies.

[0005] In a first aspect, embodiments of the present invention provide a distributed load transfer path search method, including: When load transfer demand is triggered, the storage relationship mapping table between the distributed knowledge graph and distributed nodes in the target distribution network is obtained; wherein, the distributed knowledge graph contains several subgraphs; the subgraphs are divided according to power supply zones and voltage levels; the subgraphs are stored inside the distributed nodes and each distributed node contains subgraphs of different voltage levels within the same power supply zone; Based on the voltage levels and power supply zones of the starting and target grid nodes with current load transfer demand, and in conjunction with the storage relationship mapping table, the starting subgraph, the ending subgraph, and the target distributed node are determined. Based on the starting and ending subgraphs, within each target distributed node, a hierarchical path expansion search is performed on the corresponding stored subgraph to obtain several first candidate paths. Perform a cross-partition splicing operation on the first candidate path to obtain several second candidate paths, and filter the second candidate paths through a preset security verification mechanism to obtain several third candidate paths; The third candidate path is comprehensively evaluated and ranked to obtain the optimal load transfer path for the target distribution network.

[0006] This invention, through obtaining a storage relationship mapping table, enables rapid location of each subgraph on which distributed node is stored when load transfer demand occurs. By segmenting the vast global distribution network knowledge graph along two dimensions—power supply zone and voltage level—several smaller subgraphs are formed. Power supply zone segmentation limits the search scope geographically / structurally to relevant areas, while voltage level segmentation is used because load transfer paths in the distribution network typically occur within the same voltage level. This combination significantly reduces invalid traversals during the search process. Furthermore, each distributed node is responsible for storing subgraphs for all voltage levels within a power supply zone. The graph significantly reduces cross-node communication overhead. By defining the starting subgraph, ending subgraph, and target distributed node, it ensures that the search is not performed on irrelevant subgraphs and nodes, precisely limiting the search scope to the necessary areas and avoiding invalid searches. Furthermore, by having each node search only its own small-scale subgraph, rather than the global graph, the computational complexity of a single search is greatly reduced. By splicing and filtering candidate paths, it ensures that the final output path is safe and feasible in actual power grid operation, achieving a balance between efficient search and safety and reliability. Through comprehensive evaluation and ranking, it ensures the optimal quality of the output results, improving efficiency without compromising path quality.

[0007] Compared with existing technologies, this invention addresses the efficiency bottleneck of traditional global graph traversal due to its large search space and high computational redundancy. First, it constructs a distributed subgraph by introducing a two-dimensional partitioning strategy of "power supply partition + voltage level," decoupling the originally massive global topology into multiple small-scale subgraphs related only to local physical regions and specific voltage levels, resulting in an exponential decrease in the amount of data per search. Based on this, a storage relationship mapping table is used for precise positioning, avoiding blind polling on irrelevant distributed nodes and subgraphs. Then, within each node, a hierarchical expansion of the subgraph within a defined range is performed, transforming the traditional centralized serial traversal into a distributed parallel local search, significantly shortening the search time. Next, a cross-partition splicing mechanism repairs path fragments caused by divide-and-conquer with minimal communication overhead, ensuring that efficiency improvements do not come at the cost of losing feasible solutions. Finally, combined with security checks and comprehensive evaluation, the improved output results still possess practical engineering usability. Thus, the overall solution achieves a comprehensive benefit of significantly reducing the search space, lowering communication burden, increasing parallelism, and ensuring the completeness and security of the solution set, fundamentally solving the technical problem of low search efficiency in existing technologies.

[0008] In some preferred embodiments of the first aspect, the starting subgraph, the ending subgraph, and the target distributed node are determined based on the voltage levels and power supply zones of the starting and target grid nodes where load transfer demand currently exists, in conjunction with the storage relationship mapping table. Specifically: Based on the voltage level and power supply zone of the starting power grid node, and in conjunction with the storage relationship mapping table, the starting subgraph and the starting target distributed node of the starting power grid node are determined. Based on the voltage level and power supply zone of the target power grid node, and in conjunction with the storage relationship mapping table, the endpoint subgraph and endpoint target distributed node of the target power grid node are determined; When the starting target distributed node and the ending target distributed node are not the same distributed node, the system determines whether there is a cross-partition connection path between the power supply partition where the starting subgraph is located and the power supply partition where the target subgraph is located, based on the pre-stored cross-partition connection relationship mapping table. If there is, the distributed node of the intermediate subgraph traversed by the cross-partition connection path is determined as the intermediate target distributed node according to the stored relationship mapping table.

[0009] This invention, through matching a storage mapping table based on the dual attributes of voltage level and power supply zone of the starting and target nodes, allows the system to instantly anchor the physical nodes of the starting and ending subgraphs without traversing the entire network. Furthermore, when the starting and ending nodes do not belong to the same distributed node, this invention actively queries the cross-zone connection relationship mapping table to determine the existence of a connection channel, accurately selecting the intermediate target distributed nodes that truly participate in the search. This invention precisely reduces the global blind search to a localized, targeted search covering only necessary partitions and nodes. This avoids wasting computational resources on partitions lacking connectivity while ensuring no potential cross-zone power transfer channels are overlooked, thus significantly compressing the search domain in the early stages of the search and laying a decisive foundation for subsequent efficient path expansion.

[0010] In some preferred embodiments of the first aspect, when intermediate target distributed nodes exist, the step of performing a hierarchical path expansion search on the corresponding stored subgraph within each target distributed node to obtain several first candidate paths is as follows: Within the initial target distributed node, starting from the initial power grid node in the initial subgraph, a path expansion search is performed on the corresponding subgraph to obtain several first local paths. Based on the exit boundary nodes of the preceding local paths and the cross-partition connection mapping table, the next intermediate target distributed node is determined. The next intermediate target distributed node then performs path expansion search on the corresponding subgraph, starting from each of the exit boundary nodes, until each intermediate target distributed node has completed the path expansion search, resulting in several second local paths. The preceding local paths include either the first local path or the second local path in the previous intermediate target distributed node. Inside the final target distributed node, starting from the exit boundary node of the second local path in the last intermediate target distributed node that has completed the path expansion search, a path expansion search is performed on the corresponding subgraph until the target power grid node in the final subgraph is reached, resulting in several third local paths. The first local path, the second local path, and the third local path are combined to obtain the first candidate path.

[0011] This invention, through its embodiments, forces the determination of the next intermediate node based on the exit boundary node of the preceding local path after completing the first local path search locally at the starting node. This forms a strict chain-triggered propagation mechanism, ensuring clear and backtrack-free cross-partition search paths. Simultaneously, each target distributed node independently performs expansion only within its internally stored subgraph, divided by partition and voltage. This cleverly decouples the complex global path planning across multiple partitions into multiple independent local subtasks that do not interfere with each other. Each node can perform parallel computations and only exchanges necessary data at the boundaries. This invention significantly suppresses redundant communication and repeated traversals in cross-partition searches, significantly improves the search convergence speed in multi-partition cascade scenarios, and ensures topological continuity of local segments during splicing through precise connection of boundary nodes.

[0012] In some preferred embodiments of the first aspect, a cross-partition concatenation operation is performed on the first candidate path to obtain several second candidate paths, specifically: Based on the cross-partition connection relationship mapping table, a cross-partition concatenation operation is performed on the first candidate path to obtain the second candidate path.

[0013] This invention enables all local path segments to be quickly and accurately stitched together according to predetermined physical connections by directly reusing the cross-partition connection mapping table as a splicing bridge.

[0014] In some preferred embodiments of the first aspect, a preset security verification mechanism is used to filter the second candidate paths to obtain several third candidate paths, specifically: The second candidate paths are initially screened using a preset DC power flow model to obtain a preset number of paths to be verified. Obtain the node parameters of each node in the path to be verified, the line parameters of the lines traversed, and the real-time operating data of the boundary nodes. Based on the node parameters, line parameters, and real-time operating data, calculate the system power flow distribution of each path to be verified after load transfer. The power flow distribution of the system is subjected to security checks according to the preset security operation constraints, and the path to be checked that passes the security check is determined as the third candidate path.

[0015] This invention first employs a computationally insignificant DC power flow model for coarse screening, quickly eliminating paths that clearly violate fundamental power transmission constraints and reducing the number of paths for fine-tuning to a preset range. Subsequently, only the few remaining paths to be verified are analyzed using real-time operational data for precise AC power flow calculations. This invention cleverly balances the conflict between computational speed and accuracy while ensuring absolute accuracy in the final safety verification, avoiding wasting valuable computing power on obviously infeasible paths. This ensures that the entire load transfer decision-making process does not experience bottlenecks during the verification phase, maintaining efficient response throughout the entire chain from search to decision.

[0016] In some preferred embodiments of the first aspect, the third candidate path is comprehensively evaluated and ranked to obtain the optimal load transfer path for the target distribution network, specifically as follows: Based on the line parameters, node parameters, and dynamic response attributes of the third candidate path, an evaluation index is calculated for each of the third candidate paths; wherein, the evaluation index includes path network loss, number of switching operations, load balancing degree, and distributed resource adjustment cost; The evaluation indicators are weighted and summed according to preset weight coefficients to obtain a comprehensive score for each of the third candidate paths; The third candidate paths are sorted in descending order of their comprehensive scores, and the third candidate path with the highest comprehensive score is determined as the optimal load transfer path for the target distribution network.

[0017] This invention introduces a multi-dimensional quantitative index system encompassing network losses, switching operation counts, load balancing, and regulation costs, and integrates these into a comprehensive score using preset weights. This transforms the abstract concept of path superiority into calculable numerical values, achieving complete automation of the optimization process and eliminating the time delay caused by repeated manual comparisons. Simultaneously, through a weighted summation mechanism, the selection of the optimal path flexibly adapts to the current operational preferences of the power grid (e.g., prioritizing switching speed in emergency situations and network losses during economical operation), ensuring that the ultimately recommended optimal path not only searches quickly but also achieves the highest overall efficiency in actual implementation, thus completing a closed loop from efficient search to efficient decision-making.

[0018] Secondly, embodiments of the present invention provide a distributed load transfer path search device, including a knowledge graph acquisition module, a path expansion search module, a candidate path filtering module, and an optimal path acquisition module, wherein... The knowledge graph acquisition module is used to acquire a storage relationship mapping table between the distributed knowledge graph and distributed nodes in the target distribution network when load transfer demand is triggered; wherein, the distributed knowledge graph contains several subgraphs; the subgraphs are divided according to power supply zones and voltage levels; the subgraphs are stored inside the distributed nodes and each distributed node has subgraphs of different voltage levels within the same power supply zone. The path extension search module is used to determine the starting subgraph, the ending subgraph, and the target distributed node based on the voltage level and power supply zone of the starting grid node and the target grid node with current load transfer demand, in conjunction with the storage relationship mapping table; based on the starting subgraph and the ending subgraph, a hierarchical path extension search is performed on the corresponding stored subgraph within each target distributed node to obtain several first candidate paths; The candidate path filtering module is used to perform a cross-partition splicing operation on the first candidate path to obtain a number of second candidate paths, and to filter the second candidate paths respectively through a preset security verification mechanism to obtain a number of third candidate paths. The optimal path acquisition module is used to comprehensively evaluate and rank the third candidate paths to obtain the optimal load transfer path of the target distribution network.

[0019] This invention employs a knowledge graph acquisition module to obtain a storage relationship mapping table. This enables rapid location of each subgraph on which distributed node it is stored when load transfer demand occurs. The massive global distribution network knowledge graph is segmented along two dimensions: power supply zone and voltage level, forming several smaller subgraphs. Power supply zone segmentation limits the search scope geographically / structurally to relevant areas, while voltage level segmentation is used because load transfer paths in the distribution network typically occur within the same voltage level. This combination significantly reduces invalid traversals during the search process. Furthermore, each distributed node is responsible for storing subgraphs for all voltage levels within a single power supply zone, significantly reducing cross-node traversal. The system minimizes communication overhead. A path expansion search module determines the starting subgraph, ending subgraph, and target distributed nodes, ensuring the search doesn't occur on irrelevant subgraphs and nodes, precisely limiting the search scope to the necessary areas and avoiding ineffective searches. Furthermore, by having each node search only its own stored small-scale subgraph, rather than the global graph, the computational complexity of a single search is significantly reduced. A candidate path filtering module concatenates and filters candidate paths, ensuring the final output path is safe and feasible in actual power grid operation, achieving a balance between efficient search and safety. Finally, an optimal path acquisition module comprehensively evaluates and ranks the paths, ensuring optimal output quality and improving efficiency without compromising path quality.

[0020] Thirdly, embodiments of the present invention provide a terminal device, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform the operation of the distributed load transfer path search method as described in any of the above.

[0021] Fourthly, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device or apparatus where the computer-readable storage medium is located to perform the distributed load transfer path search method as described in any one of the preceding claims.

[0022] Fifthly, embodiments of the present invention provide a computer program product, including a computer program or instructions, wherein the computer program or instructions execute the distributed load transfer path search method as described in any one of the preceding claims.

[0023] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of a distributed load transfer path search method provided in an embodiment of the present invention; Figure 2 This is a structural diagram of a distributed load transfer path search device provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example 1: like Figure 1 As shown, a distributed load transfer path search method provided by an embodiment of the present invention includes the following steps: S101, When load transfer demand is triggered, obtain the storage relationship mapping table between the distributed knowledge graph and distributed nodes in the target distribution network; wherein, the distributed knowledge graph contains several subgraphs; the subgraphs are divided according to power supply zones and voltage levels; the subgraphs are stored inside the distributed nodes and each distributed node has subgraphs of different voltage levels within the same power supply zone. In one specific embodiment, the construction process of the distributed knowledge graph is as follows: The first issue to address when constructing a distributed knowledge graph is the data source and standardization. Power system operation data is scattered across multiple independent information systems, including Supervisory Control and Data Acquisition (SCADA), Geographic Information System (GIS), Automatic Meter Reading (AMR), Advanced Metering Management System (AMI), Energy Management System (EMS), Distribution Management System (DMS), and Equipment Asset Management System. These systems have different data formats, encoding standards, and data granularities, making them unsuitable for direct use in knowledge graph construction. Therefore, data extraction is required from at least two of the aforementioned data sources. The extracted raw data is then subjected to preprocessing operations such as data cleaning, formatting, normalization, and entity alignment. The cleaning stage removes duplicate data, corrects erroneous data, and fills in missing data; the formatting stage converts data from different sources into a predefined standard data format; the normalization stage maps data with different dimensions to a unified numerical range; and the entity alignment stage unifies heterogeneous descriptions of the same power equipment from different data sources to the same globally unique identifier. After the above preprocessing, a standardized dataset is obtained.

[0027] Furthermore, based on the standardized dataset, an original knowledge graph is instantiated and constructed. Specifically, physical devices in the power system are abstracted as nodes in the knowledge graph. These physical devices include substations, buses, circuit breakers, disconnectors, transformers, feeders, distribution transformers, smart meters, distributed power sources, and loads. Electrical connections in the power system are abstracted as edges in the knowledge graph, including transmission connections, distribution connections, feeder connections, and user access connections. Simultaneously, device attribute information is attached as node attributes to the corresponding nodes, including device type, rated capacity, rated voltage, commissioning date, and geographical location. Connection attribute information is attached as edge attributes to the corresponding edges, including line impedance parameters, rated current carrying capacity, and line length. Through the definition and instantiation of nodes, edges, and their attributes, an original knowledge graph reflecting the complete topology and device attributes of the power system is constructed.

[0028] Furthermore, after the original knowledge graph is constructed, it needs to be decomposed into subgraph structures suitable for distributed storage and computation. The basis for this division is the inherent hierarchical characteristic of voltage levels in power systems. That is, equipment at different voltage levels has relative independence in physical structure and operation management. Using voltage level as the dimension for graph partitioning maximizes the preservation of topological cohesion within subgraphs while reducing the number of cross-subgraph edge connections. Specifically, the voltage level hierarchical strategy is first determined, dividing the power grid into five levels from high to low voltage: backbone network layer, high-voltage distribution network layer, medium-voltage distribution network layer, low-voltage distribution network layer, and user-side layer. The knowledge graph is divided into three layers: the backbone network layer, encompassing 500kV and 220kV voltage levels, including inter-regional interconnection lines and corresponding substation nodes; the high-voltage distribution network layer, encompassing 110kV and 35kV voltage levels, including the urban backbone distribution network structure and corresponding nodes; the medium-voltage distribution network layer, encompassing 10kV voltage level, including feeder-level structure and corresponding nodes; the low-voltage distribution network layer, encompassing 0.4kV voltage level, including transformer substation structure and corresponding nodes; and the user-side layer, including smart meters and distributed power generation nodes. After determining the layering strategy, all nodes in the original knowledge graph are traversed, and the voltage level attribute of each node is read. Nodes are then assigned to their corresponding layer subgraphs according to their voltage level.

[0029] Furthermore, edges connecting nodes at different voltage levels, such as transformer branches connecting a 220kV substation and a 110kV substation, or step-down branches connecting a 10kV feeder and a 0.4kV distribution transformer, require retaining association information in each relevant subgraph. Specifically, the edge's identifier and the global identifier of the opposite node are stored in the subgraphs containing the two endpoints of the edge. A separate cross-level connection mapping table is also established, storing complete records of all cross-level edges, including the edge's globally unique identifier, the subgraph identifiers and node identifiers of both endpoints, and the edge's electrical parameters. This mapping table enables rapid location of cross-level connections during subsequent cross-subgraph path searches, maintaining topological connectivity between subgraphs.

[0030] After subgraph partitioning, each subgraph is allocated to different nodes in the distributed computing cluster according to its voltage level. The distributed computing cluster contains multiple computing nodes, each with independent CPU cores, memory resources, and storage resources. Allocation can employ a consistent hashing algorithm to store each subgraph completely on a single computing node; for larger subgraphs, internal subgraph partitioning can be used to divide subgraphs at the same level and store them across multiple computing nodes. The computing nodes communicate with each other via a high-speed network to support cross-node data exchange during subsequent distributed path search.

[0031] In a preferred embodiment, to further improve the efficiency of subsequent path search, a hierarchical index structure can be built within each distributed node. This process is performed independently within each distributed node, without the need for cross-node communication, thus exhibiting good parallel scalability. Specifically, the process begins by extracting the data attributes of all nodes and edges within the subgraph of each distributed node. Then, a popularity assessment is performed based on business importance and query frequency. The business importance score comprehensively considers indicators such as device failure frequency, power threshold proximity, and device criticality. Query frequency is calculated by counting the number of times each node and edge is queried within a preset historical time period. The business importance score and access popularity value are weighted and summed to obtain a comprehensive popularity score for each node and edge. A popularity threshold is then set for the comprehensive popularity score. Nodes and edges with scores greater than or equal to the threshold are classified as high-frequency access core nodes (edges), while those with scores less than the threshold are classified as low-frequency edge nodes (edges). Finally, a hierarchical inverted index is built based on the classification results. A first-level index is created for high-frequency access core nodes (edges), stored in the fast storage media (such as SSDs or memory) of the distributed nodes. A second-level index is created for low-frequency edge nodes (edges), stored in ordinary storage media (such as HDDs). Both the first-level and second-level indexes use the data attribute values ​​of the nodes and edges as the index keys and the corresponding node and edge identifiers as the index content. When performing a query, the first-level index is retrieved first, and the second-level index is retrieved only when the first-level index is not found or the search scope needs to be expanded. After the construction of the hierarchical inverted index described above, each distributed node contains complete subgraph data and a corresponding hierarchical index structure, thus forming an optimized distributed knowledge graph that can be directly used for load balancing and path search.

[0032] S102, based on the voltage level and power supply zone of the starting grid node and the target grid node with current load transfer demand, and in conjunction with the storage relationship mapping table, determine the starting subgraph, the ending subgraph, and the target distributed node; based on the starting subgraph and the ending subgraph, within each target distributed node, perform a hierarchical path expansion search on the corresponding stored subgraph to obtain several first candidate paths; In this embodiment, based on the voltage levels and power supply zones of the starting and target grid nodes with current load transfer demand, and in conjunction with the stored relationship mapping table, the starting subgraph, the ending subgraph, and the target distributed node are determined. Specifically: based on the voltage level and power supply zone of the starting grid node, and in conjunction with the stored relationship mapping table, the starting subgraph and the starting target distributed node are determined; based on the voltage level and power supply zone of the target grid node, and in conjunction with the stored relationship mapping table, the ending subgraph and the ending target distributed node are determined; when the starting target distributed node and the ending target distributed node are not the same distributed node, based on the pre-stored cross-zone connection relationship mapping table, it is determined whether there is a cross-zone connection path between the power supply zone where the starting subgraph is located and the power supply zone where the target subgraph is located; if so, based on the stored relationship mapping table, the distributed node of the intermediate subgraph traversed by the cross-zone connection path is determined as the intermediate target distributed node.

[0033] In one specific embodiment, based on the voltage level and power supply zone of the starting and target grid nodes with current load transfer demand, and in conjunction with the storage relationship mapping table, the starting subgraph, ending subgraph, and target distributed node are determined. Specifically, after the load transfer demand is triggered, determining the target distributed node and its corresponding subgraph is the starting point of the entire search process. This process relies on the specific information of the starting and target grid nodes carried in the load transfer demand (each node has its own unique global identifier in the power system), and this identifier is associated with the node's voltage level, power supply zone, and other attribute information during the knowledge graph construction phase. After obtaining the identifiers of the starting and target nodes, the storage relationship mapping table is first queried based on the identifier of the starting node. This mapping table is generated after the distributed knowledge graph is constructed, and its core function is to record the correspondence between each grid node and its subgraph, as well as the distributed nodes stored in that subgraph. Through this mapping table, it can be determined which voltage level subgraph the starting node is assigned to, which power supply zone the subgraph belongs to, and which node in the distributed computing cluster the subgraph is specifically stored on. By performing the same operation once on the target node, the endpoint subgraph and the endpoint target distributed node can be determined.

[0034] Furthermore, after determining the starting and target distributed nodes, it is necessary to determine whether they are the same node. If they are the same, it means that the starting node and the target node are exactly located within the subgraph governed by the same distributed node. In this case, the path search only needs to be completed within that single node, without involving cross-node communication or cross-partition splicing. If they are different, it is necessary to further determine whether there is a cross-partition connection path available for load transfer between the power supply partition where the starting subgraph is located and the power supply partition where the target subgraph is located. This determination is based on a pre-stored cross-partition connection relationship mapping table, which is specifically used to record the electrical connection relationships formed between power supply partitions through tie lines, cross-voltage level transformer branches, etc. The mapping table stores complete information for each cross-partition connection path, including the identifiers of the intermediate subgraphs traversed by the path, the identifiers of the distributed nodes where each intermediate subgraph is located, and the electrical parameters of each tie line segment on the path. By querying this mapping table, if there are one or more cross-partition connection paths between the starting partition and the target partition, the distributed nodes where the intermediate subgraphs traversed by each connection path are located are determined one by one as intermediate target distributed nodes according to the stored relationship mapping table.

[0035] It should be noted that there may be multiple cross-partition connection paths between the starting partition and the target partition, or there may be no connection paths at all. For the former, a parallel processing strategy can be adopted, where the sequence of intermediate nodes involved in each connection path is treated as an independent search branch. Each branch executes subsequent path expansion searches in parallel on its respective distributed nodes. Finally, the search results of all branches are aggregated for unified safety verification and comprehensive evaluation to ensure that feasible solutions are not missed due to premature path selection. For the latter, when the cross-partition connection mapping table confirms that there are no connection paths between the starting partition and the target partition, the system directly determines that there is no feasible path for the current load transfer demand under the current grid topology, terminates the subsequent search process, returns an empty result, and sends a message indicating that there is no executable path to the dispatch center, awaiting manual intervention or reassessment of the load transfer demand. The introduction of this exception handling mechanism effectively avoids the waste of computing resources and response delays caused by continuing to execute invalid searches when there are no connected paths.

[0036] In this embodiment, when intermediate target distributed nodes exist, the hierarchical path expansion search is performed on the corresponding stored subgraph within each target distributed node to obtain several first candidate paths. Specifically, within the starting target distributed node, starting from the starting power grid node in the starting subgraph, a path expansion search is performed on the corresponding subgraph to obtain several first local paths. Based on the exit boundary node of the preceding local path and combined with the cross-partition connection relationship mapping table, the next intermediate target distributed node is determined, so that the next intermediate target distributed node, starting from each of the exit boundary nodes, performs a path expansion search on the corresponding subgraph until each intermediate target distributed node completes the path expansion search, obtaining several second local paths. The preceding local path includes the first local path or the second local path in the previous intermediate target distributed node. Within the destination target distributed node, starting from the exit boundary node of the second local path in the last intermediate target distributed node that has completed the path expansion search, a path expansion search is performed on the corresponding subgraph until the target power grid node in the destination subgraph is reached, obtaining several third local paths. The first local paths, second local paths, and third local paths are summarized to obtain the first candidate paths.

[0037] In one specific embodiment, when intermediate target distributed nodes exist, the hierarchical path expansion search is performed on the corresponding stored subgraph within each target distributed node to obtain several first candidate paths. Specifically, after determining the starting subgraph, the ending subgraph, and all intermediate target distributed nodes, the parallel path expansion search phase within each node is entered. The execution order of this phase is as follows: starting from the starting target distributed node, the search proceeds sequentially towards the endpoint along the node sequence determined by the cross-regional connection path. Specifically, firstly, within the starting target distributed node, taking the starting grid node in the starting subgraph as the search starting point, a path expansion search is performed on the subgraph stored by that node (using a breadth-first or depth-first traversal strategy during the search). Starting from the starting node, the search expands outwards along the edges inside the subgraph, recording the currently traversed node sequence and edge sequence at each step, until the search reaches the boundary node of the subgraph. The boundary node referred to here is a node connected to cross-regional connection lines or cross-voltage level transformer branches, and these nodes are clearly recorded in the cross-regional connection relationship mapping table. Each complete path from the starting node to a certain exit boundary node obtained by searching within the initial subgraph is saved as a first local path.

[0038] Furthermore, after the first local path search is completed, the next search direction needs to be determined based on the exit boundary nodes of these local paths. Specifically, the exit boundary node identifier of each first local path is extracted, and combined with the cross-partition connection relationship mapping table, the node information of the counterpart node of the cross-partition connection line connected to the exit boundary node is queried. The node information includes the identifier of the subgraph where the counterpart node is located, the identifier of the distributed node where the counterpart node is located, and the specific node identifier of the counterpart node within the subgraph. After retrieving the counterpart information, the next intermediate target distributed node to be searched can be determined. Subsequently, within the intermediate target distributed node, starting from the counterpart node corresponding to each exit boundary node, a path expansion search is performed on the subgraph stored by the node. The search strategy is the same as the search at the starting node, and it is also expanded to the boundary node of the subgraph. Each path obtained is saved as a second local path. After that, based on the exit boundary nodes of the second local paths in the current batch, the cross-partition connection relationship mapping table is queried again to determine the next intermediate target distributed node and perform the search. This process is repeated iteratively until the path expansion search of each intermediate target distributed node is completed. In this process, the preceding local path includes the first local path or the second local path generated in the previous intermediate target distributed node. The search starting point of the current batch is determined by the exit boundary node of the preceding local path after cross-partition mapping transformation.

[0039] Furthermore, after all intermediate target distributed nodes have completed their searches, a final search is performed within the final target distributed node. At this point, the exit boundary node of the second local path output by the last intermediate target distributed node to complete the path expansion search is used as the starting point. After transformation through the cross-partition connection relationship mapping table, the entry node on the final subgraph side is obtained. A path expansion search is then performed on the final subgraph starting from this entry node. The termination condition for the search is no longer reaching the boundary node, but rather reaching the target power grid node in the final subgraph. Each path from the entry node to the target power grid node is saved as a third local path. After completing all the above searches, the first local paths, the second local paths generated by each intermediate node, and the third local paths are summarized and spliced ​​according to the connection relationships recorded in the cross-partition connection relationship mapping table. Specifically, the exit boundary node of the preceding local path is connected to the entry node of the next local path through a cross-partition tie line, forming a complete path from the starting power grid node to the target power grid node. All the spliced ​​complete paths are the first candidate paths.

[0040] It should be noted that this invention introduces a dual pruning strategy during the search process: the first is topology pruning, which records the set of nodes already visited by the current path in real time during the search process, and immediately terminates further expansion in that direction when an expansion encounters a visited node to avoid loops; the second is electrical constraint pruning, which reads the rated current carrying capacity and current load rate of each edge before expanding it. If the edge has no remaining capacity for additional load to pass through under the current operating state, the edge is removed from the expansion queue and not included in the candidate path.

[0041] It's important to note the issue of inconsistent search progress among branches in multi-branch parallel search. Since the subgraph sizes stored by each intermediate target distributed node differ, the search time for each branch can vary significantly. Using a serial waiting approach would limit the overall response time to the slowest branch. To address this, an asynchronous parallel coordination mechanism can be employed. Each intermediate target distributed node, after completing its local path search, immediately sends its search results to the coordinating node. The coordinating node, upon receiving the search result for each branch, triggers the search task for the next node without waiting for other branches to complete. This achieves pipelined parallel progress for each branch, effectively shortening the overall search time.

[0042] S103, perform a cross-partition splicing operation on the first candidate path to obtain a number of second candidate paths, and filter the second candidate paths respectively through a preset security verification mechanism to obtain a number of third candidate paths; In this embodiment, a cross-partition concatenation operation is performed on the first candidate path to obtain several second candidate paths. Specifically, according to the cross-partition connection relationship mapping table, a cross-partition concatenation operation is performed on the first candidate path to obtain second candidate paths.

[0043] In one specific embodiment, a cross-partition splicing operation is performed on the first candidate path to obtain several second candidate paths. Specifically, after completing the path expansion search within each distributed node, each node obtains several local path fragments belonging to its locality, including a first local path within the starting subgraph, second local paths within each intermediate subgraph, and a third local path within the ending subgraph. These local path fragments are stored independently on different distributed nodes and have not yet formed a complete path from the starting grid node to the target grid node. The purpose of the cross-partition splicing operation is to connect these local path fragments, which are scattered across different distributed nodes, end to end according to their actual connection relationship in the grid topology, and merge them to generate several complete candidate paths.

[0044] The core basis of the splicing operation is a pre-stored cross-regional connection relationship mapping table. This mapping table was established during the subgraph partitioning stage of the distributed knowledge graph, and it fully records the electrical connection relationships formed between power supply zones through cross-regional tie lines, cross-voltage level transformer branches, etc. For each cross-regional connection, the mapping table stores the subgraph identifiers of its two ends, the node identifiers of the two end nodes within their respective subgraphs, and the electrical parameters of the connection relationship (such as line impedance, rated current carrying capacity, etc.). During splicing, the system extracts the exit boundary node identifier of each first local path, uses this identifier as the key to query the cross-regional connection relationship mapping table, and obtains the information of the peer node connected to the exit boundary node, including the subgraph identifier of the peer node and its node identifier within the subgraph. Then, it searches for the local path starting from the peer node in the second local path set, connects the exit boundary node of the first local path to the entry node of the second local path through the cross-regional tie line, and completes one cross-regional splicing. Repeat the above process, sequentially connecting the first local path, the second local paths generated by each intermediate node, and the third local path according to the connection relationships recorded in the mapping table, ultimately forming a complete path from the starting power grid node to the target power grid node. All the completed paths are the second candidate paths.

[0045] Preferably, path conflicts and duplicate splicing may occur during the splicing process. This is because the same exit boundary node may connect to multiple cross-partition connection lines, and the same entry node may be reached by multiple preceding local paths. If left uncontrolled, the same complete path may be spliced ​​multiple times, increasing unnecessary computational overhead. To address this issue, this invention generates a path fingerprint for each spliced ​​complete path during the splicing process (the hash value is obtained by sequentially splicing the identifiers of all nodes traversed by the path). Whenever a new path is spliced, its path fingerprint is first calculated and compared with the saved path fingerprint set. If the fingerprint already exists, the path is discarded; if the fingerprint does not exist, it is added to the second candidate path set, and the fingerprint set is updated.

[0046] Preferably, a connectivity check is performed immediately after each splicing operation: This checks whether the node identifiers on both sides of the splicing point have valid records in the cross-partition connection mapping table, and simultaneously checks whether the voltage levels of the nodes on both sides are consistent or meet the transformer step-up / step-down matching relationship (e.g., the 10kV side can only be spliced ​​to the 10kV side, or spliced ​​to different voltage level sides through a clearly recorded transformer branch). Only when all the above checks pass is the spliced ​​path formally included in the second candidate path set; if the checks fail, the splicing is marked as invalid and the path is discarded to avoid incorrectly merging topologically disconnected path segments into the subsequent safety check stage, thus avoiding a waste of computational resources.

[0047] In this embodiment, a preset safety verification mechanism is used to screen the second candidate paths to obtain several third candidate paths. Specifically, the second candidate paths are initially screened using a preset DC power flow model to obtain a preset number of paths to be verified. The node parameters of each node in the path to be verified, the line parameters of the lines they pass through, and the real-time operating data of the boundary nodes are obtained. Based on the node parameters, line parameters, and real-time operating data, the system power flow distribution of each path to be verified after load transfer is calculated. The system power flow distribution is verified according to preset safety operating constraints, and the paths to be verified that pass the safety verification are determined as third candidate paths.

[0048] In one specific embodiment, a preset security verification mechanism is used to screen the second candidate paths to obtain several third candidate paths. Specifically, the security verification is performed step-by-step in two levels. The first level is a preliminary screening based on the DC power flow model. The DC power flow model is a linear approximation of the AC power flow. Its core assumption is to ignore line resistance and the coupling relationship between reactive power and voltage amplitude, simplifying the power flow calculation to a set of linear equations related only to the active power injected into the nodes and the line reactance. Although this approximation sacrifices some calculation accuracy, it brings extremely high calculation efficiency. Especially for scenarios containing hundreds or thousands of second candidate paths, iterative calculation of AC power flow one by one would consume a lot of time, while DC power flow can complete the rapid evaluation of all candidate paths in milliseconds. In the specific implementation, the system obtains the current power grid operating status data (including the active power injected into each node, the initial values ​​of voltage amplitude and phase angle of each node, etc.). For each second candidate path, the system simulates the power grid operating status after the load to be transferred from the starting node is transferred to the target node according to the path, and calculates the approximate value of the active power distribution on each line through the DC power flow model. Then, the approximate power values ​​of each line are compared with their rated current carrying capacity. Paths that would cause any line to exceed its power limit after the transfer are eliminated, while paths with relatively balanced power distribution are retained.

[0049] It should be noted that the preset number refers to the upper limit of the number of paths retained after the initial screening. Its value is preset according to the real-time computing capability of the system and the requirements of the scheduling procedure. If the number of paths after DC power flow screening is less than the preset number, all of them are retained and enter the next stage. If it is more than the preset number, they are sorted according to the power distribution balance of each path (usually measured by the standard deviation of the load rate of each line), and the top few paths with the most balanced load rate are retained as paths to be verified and enter the second level of precise verification.

[0050] Furthermore, the second level is a precise security verification based on the AC power flow model. For each path to be verified that passes the initial screening, the system extracts the node parameters (including node voltage level, node rated capacity, node current load value, etc.) of all nodes traversed by the path and the line parameters (including the resistive and reactive components of the line impedance, the line rated current carrying capacity, line length, etc.) from the distributed knowledge graph. Simultaneously, the system acquires real-time operating data of boundary nodes through a SCADA system or a wide-area measurement system. The voltage amplitude, phase angle, and active and reactive power injection of these nodes are key boundary conditions for power flow calculation. Based on the above parameters and real-time data, using the current grid operating state as the initial profile, the system simulates the grid operating state after implementing the load transfer path. Iterative AC power flow calculations are performed using the Newton-Raphson method or the fast decoupling method to obtain the system power flow distribution after implementing the path, including the voltage amplitude of each node, the voltage phase angle of each node, and the active and reactive power transmitted by each line.

[0051] Furthermore, after obtaining the power flow distribution, each item is verified according to the preset safe operation constraints. These constraints mainly include two aspects: voltage constraints, verifying whether the voltage amplitude of each node in the system is between 0.9 and 1.1 times the rated voltage; any node whose voltage exceeds this range is deemed not to meet the voltage safety constraint. Line capacity constraints, verifying whether the transmission power of each line in the system does not exceed the rated transmission power limit of that line; any line experiencing overload is deemed not to meet the capacity safety constraint. For each path to be verified, only when its corresponding power flow distribution simultaneously meets all the above safe operation constraints is the path deemed to have passed the safety verification, formally designated as the third candidate path, and enters the subsequent comprehensive evaluation stage.

[0052] S104, comprehensively evaluate and rank the third candidate paths to obtain the optimal load transfer path for the target distribution network.

[0053] In this embodiment, the third candidate paths are comprehensively evaluated and ranked to obtain the optimal load transfer path for the target distribution network. Specifically, the evaluation index of each third candidate path is calculated based on its line parameters, node parameters, and dynamic response attributes. The evaluation index includes path network loss, number of switching operations, load balancing degree, and distributed resource adjustment cost. The evaluation indexes are weighted and summed according to preset weight coefficients to obtain a comprehensive score for each third candidate path. The third candidate paths are ranked in descending order of their comprehensive scores, and the third candidate path with the highest comprehensive score is determined as the optimal load transfer path for the target distribution network.

[0054] In one specific embodiment, the third candidate paths are comprehensively evaluated and ranked to obtain the optimal load transfer path for the target distribution network. Specifically, four evaluation indicators are calculated based on the line parameters, node parameters, and dynamic response attributes of each third candidate path. The first evaluation indicator is path network loss. Path network loss refers to the power loss that occurs when load current flows through each segment of the line and transformer along a candidate path after load transfer is implemented. The specific calculation method is as follows: the impedance parameters (resistive and reactive components) of all lines traversed by the path are extracted, as well as the load current value carried by each segment of the path after transfer. For each line segment, the formula is used... Calculate the active power loss of this segment, and then sum the active power losses of all lines along the path to obtain the total network loss value of the path, in kilowatts. The smaller the network loss, the higher the energy transmission efficiency of the path.

[0055] The second evaluation metric is the number of switching operations. Load transfer relies on the opening and closing operations of various switching devices (including circuit breakers and disconnectors) along the path. Specifically, it requires disconnecting certain switches at the starting node to cut off the load, while simultaneously closing certain switches at the target node to connect the load. Switches in the middle of the path may need to be switched according to specific operational strategies. The number of switching operations refers to the total number of times the switching devices need to be operated when implementing the load transfer strategy corresponding to this candidate path. Each switch operation involves mechanical movement of the equipment and the issuance of corresponding control commands. A higher number of operations not only means a longer time required to execute the transfer but also an increased probability of equipment failure and the risk of operational errors. Therefore, the fewer the number of switching operations, the higher the superiority of the path at the execution level.

[0056] The third evaluation indicator is load balance. Load balance measures the uniformity of load distribution among the main equipment in each segment of the line and each power supply zone along a candidate path after load transfer is implemented. The specific calculation method is as follows: extract the load rate of all lines along the path after transfer (i.e., the ratio of the transferred power to the rated current carrying capacity of the line), and simultaneously extract the load rate of the main transformers along the path after transfer. Calculate the standard deviation of all these load rate values. The smaller the standard deviation, the more evenly the load is distributed among the segments of the path and among the equipment, without any imbalance where some equipment has an excessively high load rate while others have an excessively low load rate. Higher load balance means a more stable power grid operation and a lower risk of local equipment overload.

[0057] The fourth evaluation indicator is the distributed resource regulation cost. In modern distribution networks, a large number of distributed power sources (such as photovoltaic power generation, wind power generation, and energy storage devices) are connected to the grid. During load transfer, it may be necessary to coordinate and adjust the output of these distributed resources to cooperate with the transfer operation. Specifically, when implementing load transfer according to a candidate path, in order to maintain system power balance and voltage stability, it may be necessary to adjust the active and reactive power outputs of each distributed power source in the area traversed by the path, or to call upon energy storage devices for charging and discharging operations. The distributed resource regulation cost refers to the economic quantification of the distributed resource regulation costs incurred to cooperate with the load transfer of this path, including the power generation revenue loss corresponding to the output adjustment of distributed power sources, the depreciation cost of the charge and discharge cycle life of energy storage devices, and the demand-side response compensation costs that may be involved. This cost indicator is calculated based on the regulation cost curves of each distributed power source and energy storage device, as well as the regulation depth and duration required for the implementation of this path. The lower the distributed resource regulation cost, the lower the dependence of the path on distributed resource scheduling, and the smaller the economic burden of implementation.

[0058] After calculating the four evaluation indicators, each indicator needs to be normalized, mapping the values ​​of different dimensions to the same numerical range (e.g., between 0 and 1). Then, a weighted sum is performed according to preset weighting coefficients. The magnitude of the weighting coefficients reflects the different levels of importance that dispatching personnel place on different evaluation indicators. For example, if the current period is peak electricity consumption, reducing network losses and ensuring load balancing will have a higher priority than reducing the number of switching operations; if equipment maintenance is frequent or the switching equipment is in poor condition, the weight of the number of switching operations should be appropriately increased. The specific values ​​of each weighting coefficient are preset by dispatching personnel based on actual operating conditions and management objectives, and can be dynamically adjusted during system operation. After weighted summation, a comprehensive score is obtained for each third candidate path. The higher the comprehensive score, the higher the overall superiority of the path after considering multiple dimensions such as economy, execution efficiency, operational quality, and adjustment costs. Finally, all third candidate paths are sorted in descending order of comprehensive score, and the third candidate path with the highest comprehensive score is determined as the optimal load transfer path under the current load transfer demand, as the final output result.

[0059] This invention, through obtaining a storage relationship mapping table, enables rapid location of each subgraph on which distributed node is stored when load transfer demand occurs. By segmenting the vast global distribution network knowledge graph along two dimensions—power supply zone and voltage level—several smaller subgraphs are formed. Power supply zone segmentation limits the search scope geographically / structurally to relevant areas, while voltage level segmentation is used because load transfer paths in the distribution network typically occur within the same voltage level. This combination significantly reduces invalid traversals during the search process. Furthermore, each distributed node is responsible for storing subgraphs for all voltage levels within a power supply zone. The graph significantly reduces cross-node communication overhead. By defining the starting subgraph, ending subgraph, and target distributed node, it ensures that the search is not performed on irrelevant subgraphs and nodes, precisely limiting the search scope to the necessary areas and avoiding invalid searches. Furthermore, by having each node search only its own small-scale subgraph, rather than the global graph, the computational complexity of a single search is greatly reduced. By splicing and filtering candidate paths, it ensures that the final output path is safe and feasible in actual power grid operation, achieving a balance between efficient search and safety and reliability. Through comprehensive evaluation and ranking, it ensures the optimal quality of the output results, improving efficiency without compromising path quality.

[0060] Compared with existing technologies, this invention addresses the efficiency bottleneck of traditional global graph traversal due to its large search space and high computational redundancy. First, it constructs a distributed subgraph by introducing a two-dimensional partitioning strategy of "power supply partition + voltage level," decoupling the originally massive global topology into multiple small-scale subgraphs related only to local physical regions and specific voltage levels, resulting in an exponential decrease in the amount of data per search. Based on this, a storage relationship mapping table is used for precise positioning, avoiding blind polling on irrelevant distributed nodes and subgraphs. Then, within each node, a hierarchical expansion of the subgraph within a defined range is performed, transforming the traditional centralized serial traversal into a distributed parallel local search, significantly shortening the search time. Next, a cross-partition splicing mechanism repairs path fragments caused by divide-and-conquer with minimal communication overhead, ensuring that efficiency improvements do not come at the cost of losing feasible solutions. Finally, combined with security checks and comprehensive evaluation, the improved output results still possess practical engineering usability. Thus, the overall solution achieves a comprehensive benefit of significantly reducing the search space, lowering communication burden, increasing parallelism, and ensuring the completeness and security of the solution set, fundamentally solving the technical problem of low search efficiency in existing technologies.

[0061] Secondly, embodiments of the present invention provide a distributed load transfer path search device, including a knowledge graph acquisition module 201, a path expansion search module 202, a candidate path filtering module 203, and an optimal path acquisition module 204, wherein... When the load transfer demand is triggered, the knowledge graph acquisition module 201 acquires the storage relationship mapping table between the distributed knowledge graph and distributed nodes in the target distribution network; wherein, the distributed knowledge graph contains several subgraphs; the subgraphs are divided according to power supply zones and voltage levels; the subgraphs are stored inside the distributed nodes and each distributed node has subgraphs of different voltage levels within the same power supply zone. The path extension search module 202 is used to determine the starting subgraph, the ending subgraph, and the target distributed node based on the voltage level and power supply zone of the starting grid node and the target grid node with current load transfer demand, in conjunction with the storage relationship mapping table; based on the starting subgraph and the ending subgraph, a hierarchical path extension search is performed on the corresponding stored subgraph within each target distributed node to obtain several first candidate paths; In this embodiment, the path extension search module 202 determines the starting subgraph, the ending subgraph, and the target distributed node based on the voltage level and power supply zone of the starting grid node and the target grid node with current load transfer demand, in conjunction with the stored relationship mapping table. Specifically, the path extension search module 202 determines the starting subgraph and the starting target distributed node of the starting grid node based on the voltage level and power supply zone of the starting grid node, in conjunction with the stored relationship mapping table; it determines the ending subgraph and the ending target distributed node of the target grid node based on the voltage level and power supply zone of the target grid node, in conjunction with the stored relationship mapping table; when the starting target distributed node and the ending target distributed node are not the same distributed node, it determines whether there is a cross-zone connection path between the power supply zone of the starting subgraph and the power supply zone of the target subgraph based on the pre-stored cross-zone connection relationship mapping table; if so, it determines the distributed node of the intermediate subgraph traversed by the cross-zone connection path as the intermediate target distributed node based on the stored relationship mapping table.

[0062] The candidate path filtering module 203 is used to perform a cross-partition splicing operation on the first candidate path to obtain a number of second candidate paths, and to filter the second candidate paths respectively through a preset security verification mechanism to obtain a number of third candidate paths. In this embodiment, the candidate path filtering module 203 performs a cross-partition splicing operation on the first candidate path to obtain several second candidate paths. Specifically, the candidate path filtering module 203 performs a cross-partition splicing operation on the first candidate path according to the cross-partition connection relationship mapping table to obtain second candidate paths.

[0063] In this embodiment, the candidate path screening module 203 filters the second candidate paths through a preset security verification mechanism to obtain several third candidate paths. Specifically, the candidate path screening module 203 performs preliminary screening on each of the second candidate paths through a preset DC power flow model to obtain a preset number of paths to be verified; obtains the node parameters of each node in the path to be verified, the line parameters of the lines it passes through, and the real-time operating data of the boundary nodes, and calculates the system power flow distribution of each path to be verified after load transfer based on the node parameters, line parameters, and real-time operating data; performs security verification on the system power flow distribution according to preset security operating constraints, and determines the path to be verified that passes the security verification as the third candidate path.

[0064] The optimal path acquisition module 204 is used to comprehensively evaluate and rank the third candidate paths to obtain the optimal load transfer path of the target distribution network.

[0065] In this embodiment, the optimal path acquisition module 204 comprehensively evaluates and sorts the third candidate paths to obtain the optimal load transfer path of the target distribution network. Specifically, the optimal path acquisition module 204 calculates the evaluation index of each third candidate path based on the line parameters, node parameters, and dynamic response attributes of the third candidate paths. The evaluation index includes path network loss, number of switching operations, load balancing degree, and distributed resource adjustment cost. The evaluation index is weighted and summed according to a preset weight coefficient to obtain a comprehensive score for each third candidate path. The third candidate paths are sorted in descending order of the comprehensive score, and the third candidate path with the highest comprehensive score is determined as the optimal load transfer path of the target distribution network.

[0066] For a more detailed explanation of the working principle and procedures of this embodiment, please refer to the relevant description in Embodiment 1.

[0067] This invention employs a knowledge graph acquisition module 201 to obtain a storage relationship mapping table. This enables rapid location of each subgraph on which distributed node it is stored when load transfer demand occurs. The massive global distribution network knowledge graph is segmented along two dimensions: power supply zone and voltage level, forming several smaller subgraphs. Power supply zone segmentation limits the search scope geographically / structurally to relevant areas, while voltage level segmentation is used because load transfer paths in the distribution network typically occur within the same voltage level. This combination significantly reduces invalid traversals during the search process. Furthermore, each distributed node is responsible for storing subgraphs for all voltage levels within a power supply zone, significantly reducing cross-node communication. Overhead; The path expansion search module 202 determines the starting subgraph, ending subgraph, and target distributed node, ensuring that the search is not performed on irrelevant subgraphs and nodes, precisely limiting the search scope to the necessary area and avoiding invalid searches. Then, by having each node search only its own stored small-scale subgraph, rather than the global graph, the computational complexity of a single search is greatly reduced. The candidate path filtering module 203 splices and filters candidate paths to ensure that the final output path is safe and feasible in actual power grid operation, thus achieving a balance between efficient search and safety and reliability. The optimal path acquisition module 204 comprehensively evaluates and sorts the paths, ensuring the optimal quality of the output results, so as to improve efficiency without reducing path quality.

[0068] Example 3: This embodiment provides a terminal device, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction that causes the processor to perform the operation of the distributed load transfer path search method as described in any of the above.

[0069] Example 4: This invention provides a computer-readable storage medium including a stored computer program, wherein the computer program, when running, controls the device or apparatus containing the computer-readable storage medium to perform the distributed load transfer path search method as described in any of the preceding claims.

[0070] Example 5: This invention provides a computer program product, including a computer program or instructions, wherein the computer program or instructions execute the distributed load transfer path search method as described in any of the above embodiments.

[0071] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0072] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for searching distributed load transfer paths, characterized in that, include: When load transfer demand is triggered, the storage relationship mapping table between the distributed knowledge graph and distributed nodes in the target distribution network is obtained; wherein, the distributed knowledge graph contains several subgraphs; the subgraphs are divided according to power supply zones and voltage levels; the subgraphs are stored inside the distributed nodes and each distributed node contains subgraphs of different voltage levels within the same power supply zone; Based on the voltage levels and power supply zones of the starting and target grid nodes with current load transfer demand, and in conjunction with the storage relationship mapping table, the starting subgraph, the ending subgraph, and the target distributed node are determined. Based on the starting and ending subgraphs, within each target distributed node, a hierarchical path expansion search is performed on the corresponding stored subgraph to obtain several first candidate paths. Perform a cross-partition splicing operation on the first candidate path to obtain several second candidate paths, and filter the second candidate paths through a preset security verification mechanism to obtain several third candidate paths; The third candidate path is comprehensively evaluated and ranked to obtain the optimal load transfer path for the target distribution network.

2. The distributed load transfer path search method as described in claim 1, characterized in that, Based on the voltage levels and power supply zones of the starting and target grid nodes with current load transfer demand, and in conjunction with the aforementioned storage relationship mapping table, the starting subgraph, ending subgraph, and target distributed nodes are determined as follows: Based on the voltage level and power supply zone of the starting power grid node, and in conjunction with the storage relationship mapping table, the starting subgraph and the starting target distributed node of the starting power grid node are determined. Based on the voltage level and power supply zone of the target power grid node, and in conjunction with the storage relationship mapping table, the endpoint subgraph and endpoint target distributed node of the target power grid node are determined; When the starting target distributed node and the ending target distributed node are not the same distributed node, the pre-stored cross-partition connection relationship mapping table is used to determine whether there is a cross-partition connection path between the power supply partition where the starting subgraph is located and the power supply partition where the target subgraph is located. If it exists, then according to the storage relationship mapping table, the distributed node where the intermediate subgraph traversed by the cross-partition connection path is located is determined as the intermediate target distributed node.

3. The distributed load transfer path search method as described in claim 2, characterized in that, When intermediate target distributed nodes exist, within each target distributed node, a hierarchical path expansion search is performed on the corresponding stored subgraph to obtain several first candidate paths, specifically as follows: Within the initial target distributed node, starting from the initial power grid node in the initial subgraph, a path expansion search is performed on the corresponding subgraph to obtain several first local paths. Based on the exit boundary nodes of the preceding local paths and the cross-partition connection mapping table, the next intermediate target distributed node is determined. The next intermediate target distributed node then performs path expansion search on the corresponding subgraph, starting from each of the exit boundary nodes, until each intermediate target distributed node has completed the path expansion search, resulting in several second local paths. The preceding local paths include either the first local path or the second local path in the previous intermediate target distributed node. Inside the final target distributed node, starting from the exit boundary node of the second local path in the last intermediate target distributed node that has completed the path expansion search, a path expansion search is performed on the corresponding subgraph until the target power grid node in the final subgraph is reached, resulting in several third local paths. The first local path, the second local path, and the third local path are combined to obtain the first candidate path.

4. The distributed load transfer path search method as described in claim 3, characterized in that, Perform a cross-partition concatenation operation on the first candidate path to obtain several second candidate paths, specifically: Based on the cross-partition connection relationship mapping table, a cross-partition concatenation operation is performed on the first candidate path to obtain the second candidate path.

5. The distributed load transfer path search method as described in claim 4, characterized in that, By using a preset security verification mechanism, the second candidate paths are filtered to obtain several third candidate paths, specifically: The second candidate paths are initially screened using a preset DC power flow model to obtain a preset number of paths to be verified. Obtain the node parameters of each node in the path to be verified, the line parameters of the lines traversed, and the real-time operating data of the boundary nodes. Based on the node parameters, line parameters, and real-time operating data, calculate the system power flow distribution of each path to be verified after load transfer. The power flow distribution of the system is subjected to security checks according to the preset security operation constraints, and the path to be checked that passes the security check is determined as the third candidate path.

6. The distributed load transfer path search method as described in claim 5, characterized in that, The third candidate path is comprehensively evaluated and ranked to obtain the optimal load transfer path for the target distribution network, specifically as follows: Based on the line parameters, node parameters, and dynamic response attributes of the third candidate path, an evaluation index is calculated for each of the third candidate paths; wherein, the evaluation index includes path network loss, number of switching operations, load balancing degree, and distributed resource adjustment cost; The evaluation indicators are weighted and summed according to preset weight coefficients to obtain a comprehensive score for each of the third candidate paths; The third candidate paths are sorted in descending order of their comprehensive scores, and the third candidate path with the highest comprehensive score is determined as the optimal load transfer path for the target distribution network.

7. A distributed load transfer path search device, characterized in that, It includes a knowledge graph acquisition module, a path expansion and search module, a candidate path filtering module, and an optimal path acquisition module. The knowledge graph acquisition module is used to acquire a storage relationship mapping table between the distributed knowledge graph and distributed nodes in the target distribution network when load transfer demand is triggered; wherein, the distributed knowledge graph contains several subgraphs; the subgraphs are divided according to power supply zones and voltage levels; the subgraphs are stored inside the distributed nodes and each distributed node has subgraphs of different voltage levels within the same power supply zone. The path extension search module is used to determine the starting subgraph, the ending subgraph, and the target distributed node based on the voltage level and power supply zone of the starting grid node and the target grid node with current load transfer demand, in conjunction with the storage relationship mapping table; based on the starting subgraph and the ending subgraph, a hierarchical path extension search is performed on the corresponding stored subgraph within each target distributed node to obtain several first candidate paths; The candidate path filtering module is used to perform a cross-partition splicing operation on the first candidate path to obtain a number of second candidate paths, and to filter the second candidate paths respectively through a preset security verification mechanism to obtain a number of third candidate paths. The optimal path acquisition module is used to comprehensively evaluate and rank the third candidate paths to obtain the optimal load transfer path of the target distribution network.

8. A terminal device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation of the distributed load transfer path search method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device or apparatus containing the computer-readable storage medium to perform the distributed load transfer path search method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instructions perform the distributed load transfer path search method as described in any one of claims 1 to 6.

Images