A coal mine user power supply path risk identification method based on a knowledge graph

Abstract

The application discloses a coal mine user power supply path risk identification method based on a knowledge graph, which comprises the following steps: collecting coal mine enterprise power supply related data, constructing a directed coal mine enterprise power supply graph and generating a power supply graph basic representation; generating a transfer supply script sequence according to dispatching and disposal records and binding a power supply path; determining a risk trigger node and injecting a power failure disturbance token to form an initial disturbance power supply graph; synchronously updating node representation and power failure disturbance representation through GraphSAGE; generating branch final representation according to the transfer supply script sequence; obtaining power supply path risk identification results and constructing a backtracking query vector according to the branch final representation; and generating a risk backtracking chain and a risk traceability result through a backtracking pointer GraphSAGE. The application realizes accurate identification of coal mine power supply risks, transfer supply analysis and traceability of risk sources.

Description

Technical Field

[0001] This invention relates to the field of power safety risk identification technology, and in particular to a method for identifying power supply path risks for coal mine users based on knowledge graphs. Background Technology

[0002] Coal mining enterprises are important electricity users, and the reliability of their power supply directly affects safe production processes such as underground ventilation, drainage, hoisting, and gas extraction. Current methods for identifying power supply risks in coal mining enterprises typically rely on grid topology, equipment status, fault records, and dispatching experience to analyze the main power supply path, backup power supply path, and related equipment to determine whether the coal mining enterprise faces a risk of power outage.

[0003] Existing methods often focus on static topology or the state of a single device, making it difficult to describe the propagation of the impact of power loss along the power supply direction after a power outage occurs. For scenarios such as main power path failure, backup path maintenance, and abnormal switch operation, existing methods can usually only provide the risk level, failing to simultaneously reflect the transmission of risk disturbances in the power supply diagram.

[0004] Existing methods do not adequately utilize dispatch and handling records. Dispatch operation tickets and power transfer recovery processes are typically only used for manual analysis and fail to be transformed into power transfer script sequences that drive state changes in the graph model. This makes it difficult for the model to determine the recovery capability under different power transfer branches. Existing risk tracing methods rely heavily on rule matching or manual tracing, lacking a mechanism for recursively selecting risk source nodes along the reverse power supply direction, making it difficult to form a risk backtracking chain corresponding to the disturbance propagation process.

[0005] Therefore, how to provide a knowledge graph-based method for identifying risks in the power supply paths of coal mine users is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a knowledge graph-based method for identifying power supply path risks in coal mines. This invention utilizes a power supply knowledge graph, power outage disturbance tokens, power transfer script branches, and backtracking pointer mechanisms to describe the process of identifying and tracing the source of power supply path risks in coal mines. It has the advantages of accurate risk identification, analyzable power transfer process, and traceable risk sources.

[0007] A method for identifying power supply path risks for coal mine users based on a knowledge graph, according to an embodiment of the present invention, includes the following steps: Collect relevant power supply data from coal mining enterprises, construct power supply maps for coal mining enterprises, and generate basic representations of the power supply maps; Based on the basic representation of the power supply diagram and the dispatch and handling records, a power transfer script sequence is generated, and the correspondence between the power transfer script sequence and the power supply path in the power supply diagram of the coal mine enterprise is established. Based on the power supply data of coal mining enterprises, risk trigger nodes are identified. The node representation of the risk trigger nodes is expanded to include the initial disturbance representation containing the power outage disturbance token. The power outage disturbance token participates in the subsequent neighborhood update as an independent hidden state, forming an initial disturbance power supply graph. The initial disturbance power supply graph is input into the GraphSAGE model that integrates disturbance propagation. The node representation and the power loss disturbance representation are updated synchronously in the node update function. The power loss disturbance token is propagated along the power supply direction. The perturbation propagation process is branched based on the transfer script sequence. The transfer branch state is embedded into the GraphSAGE node update function. The node representation generates the final branch representation under the combined action of the transfer script sequence and the power failure perturbation token. Based on the final representation of the branches, generate the risk identification result of the power supply path of the coal mining enterprise, and construct the backtracking query vector when the risk identification result of the power supply path of the coal mining enterprise reaches the backtracking trigger condition. Input the backtracking query vector into the GraphSAGE model, recursively select risk source nodes along the reverse power supply direction, use the selection result of each time as the input for the next backtracking, generate a risk backtracking chain, and generate the risk source tracing result of the power supply path of the coal mine enterprise based on the risk backtracking chain and the disturbance propagation process.

[0008] Optionally, constructing the power supply diagram for coal mining enterprises specifically includes: After collecting power supply-related data of coal mining enterprises, the data is normalized by object, converting power supply objects into graph nodes, converting the power supply business relationships between power supply objects into directed edges, and determining the start and end nodes of the directed edges by the power supply direction to obtain a power supply graph of coal mining enterprises with directional attributes. Write node attribute codes to graph nodes and edge relationship codes to directed edges. Then, matrix the node attribute codes and edge relationship codes according to the topological connection relationship of the power supply graph of the coal mine enterprise to generate the basic representation of the power supply graph.

[0009] Optionally, the step of generating the power transfer script sequence based on the power supply diagram and scheduling records specifically includes: The dispatch and handling record is defined as a record of the dispatch operation process for power supply anomalies in coal mining enterprises. The dispatch and handling record records the fault handling object, the power supply restoration object, the dispatch operation content and the operation execution sequence. Based on the basic representation of the power supply diagram, determine the corresponding diagram nodes of the fault handling objects and power supply restoration objects in the dispatching and handling records in the power supply diagram of the coal mining enterprise; The scheduling and handling records are sorted according to the order of operation execution. Each scheduling operation is converted into a script instruction. The consecutive script instructions are connected according to the order of execution to generate a transfer script sequence. Based on the corresponding graph nodes involved in the script instructions, determine the directed edges and power supply paths associated with the script instructions in the power supply graph of the coal mine enterprise; The power transfer script sequence is bound to the power supply path identifier to form a correspondence between the power transfer script sequence and the power supply path in the power supply diagram of the coal mine enterprise.

[0010] Optionally, the determination of risk triggering nodes based on coal mine enterprise power supply-related data specifically includes: Identify the power supply objects that have experienced power supply anomalies from the relevant power supply data of coal mining enterprises, match the power supply objects that have experienced power supply anomalies with the graph nodes in the power supply diagram of coal mining enterprises, and determine the matched graph nodes as risk trigger nodes. Read the node representation of the risk-triggered node in the basic representation of the power supply graph, construct a power outage disturbance token with the same dimension as the node representation, the power outage disturbance token is a hidden state vector that participates in the GraphSAGE neighborhood update, and the power outage disturbance token records the propagation state of the power outage impact caused by the power supply anomaly of the coal mine enterprise in the power supply graph of the coal mine enterprise. Initialize the power outage disturbance token corresponding to the risk-triggered node as a non-zero hidden state vector, and initialize the power outage disturbance token corresponding to the non-risk-triggered node as a zero-value hidden state vector. Connect the node representation of the risk-triggered node with the corresponding power failure disturbance token to generate the initial disturbance representation of the risk-triggered node; connect the node representation of the non-risk-triggered node with the corresponding zero-value hidden state vector to generate the initial disturbance representation of the non-risk-triggered node. Keeping the directed edge relationships in the power supply diagram of the coal mine enterprise unchanged, the power supply diagram of the coal mine enterprise containing the initial representation of the disturbance of each graph node is determined as the initial disturbance power supply diagram.

[0011] Optionally, inputting the initial perturbation power supply graph into the GraphSAGE model for fusion perturbation propagation specifically includes: Read the initial disturbance representation of each graph node from the initial disturbance power supply graph, and split the initial disturbance representation into node representation and power failure disturbance token; For the current graph node, read the upstream neighbor nodes along the power supply direction of the directed edge, perform GraphSAGE neighborhood aggregation on the node representation of the upstream neighbor nodes, and perform same-direction aggregation on the power outage disturbance tokens of the upstream neighbor nodes. The power outage disturbance token of the current graph node is superimposed with the power outage disturbance token after being aggregated in the same direction. The superposition result is then normalized to obtain the updated power outage disturbance representation of the current graph node. The node representation of the current graph node, the GraphSAGE neighborhood aggregation result, and the updated power outage perturbation representation are input into the node update function. Activation processing is performed on the output of the node update function to obtain the updated node representation of the current graph node. The above updates are performed sequentially on the graph nodes in the power supply diagram of the coal mining enterprise according to the power supply direction. The power outage disturbance token of the risk-triggered node is written to the downstream graph node along the directed edge until the updated node representation and the updated power outage disturbance representation of each graph node are obtained.

[0012] Optionally, the branching evolution of the perturbation propagation process based on the transfer script sequence specifically includes: Read the power transfer script sequence and the updated node representations and updated power outage disturbance representations of each graph node, and determine the power transfer operation steps according to the script instruction sequence in the power transfer script sequence; Based on the power supply path and the node to be operated corresponding to the script instruction, the disturbance propagation process is copied to form a transfer branch corresponding to the transfer script sequence; In each transfer operation step, the node state of the operation object node in the transfer branch is changed according to the script instructions, and the state of the directed edge connected to the operation object node is adjusted synchronously. The adjusted node state, directed edge state, updated power outage disturbance representation, and script instruction encoding are written together into the power transfer branch state; The power transfer branch state, the node representation of the current graph node, the GraphSAGE neighborhood aggregation result, and the updated power outage disturbance representation are input together into the GraphSAGE node update function. Activation processing is performed on the output of the node update function to obtain the branch node representation under the current power transfer operation step. The graph nodes in the transfer branch are continuously updated according to the script instructions until the transfer script sequence is completed. The branch node representation and the branch power failure disturbance representation obtained after the transfer script sequence is completed are determined as the final branch representation.

[0013] Optionally, constructing a backtracking query vector when the risk identification result of the power supply path of a coal mine enterprise reaches the backtracking trigger condition specifically includes: The branch node representation and branch power outage disturbance representation corresponding to the coal mine enterprise node are read from the final branch representation. The branch node representation and branch power outage disturbance representation are concatenated on the feature dimension and then input into the risk discrimination layer to obtain the power supply path risk value corresponding to the transfer branch. Based on the correspondence between the power transfer script sequence and the power supply path, the risk value of the power supply path is written into the corresponding power supply path to form the power supply path risk identification result of the coal mine enterprise. When the risk value of the power supply path reaches the preset risk threshold, the risk identification result of the power supply path of the coal mining enterprise will be determined as the trigger condition for backtracking. When the backtracking trigger condition is met, the coal mine enterprise node representation, branch power failure disturbance representation, and power supply path risk value corresponding to the power supply path that has met the backtracking trigger condition are read. The read results are concatenated and linear transformation is performed to construct the backtracking query vector.

[0014] Optionally, inputting the backtracking query vector into the backtracking pointer GraphSAGE model specifically includes: The coal mining enterprise node in the power supply path that meets the backtracking trigger condition is determined as the initial backtracking node, and the backtracking query vector is used as the initial backtracking input. At the current backtracking node, read the upstream graph node along the reverse power supply direction, and input the node representation and power failure disturbance representation of the upstream graph node together with the current backtracking input into the GraphSAGE model to obtain the backtracking pointer value corresponding to the upstream graph node; The upstream graph node with the largest backtracking pointer value is selected as the candidate node of the current risk source. The candidate node of the current risk source is written into the risk backtracking chain. The node representation and power failure disturbance representation of the candidate node of the current risk source are merged into the current backtracking input to generate the next backtracking input. The current risk source candidate node is used as the next backtracking node. The reverse reading, pointer calculation, node writing and backtracking input update are repeatedly executed until the current backtracking node has no upstream graph node or the maximum backtracking pointer value is lower than the backtracking stop threshold. The last candidate node of the risk source written in the risk backtracking chain is determined as the risk source node, and the risk source tracing result of the power supply path of the coal mine enterprise is generated based on the risk backtracking chain and the disturbance propagation process.

[0015] The beneficial effects of this invention are: (1) This invention constructs a power supply diagram of coal mine enterprises with power supply direction, and converts the power supply related data of coal mine enterprises into graph nodes, directed edges and power supply graph basic representation, so that the power supply relationship between the main power supply path, backup path, coal mine enterprise nodes and key load nodes can be expressed in a unified graph structure, which is conducive to improving the data consistency of power supply path risk identification and path association analysis capability of coal mine users.

[0016] (2) The present invention introduces a power outage disturbance token in the GraphSAGE model, so that the power outage disturbance token participates in the neighborhood update along the power supply direction, so that the power outage impact caused by the power supply anomaly of the coal mine can be propagated step by step in the power supply graph; combined with the transfer script sequence, the disturbance propagation process is branched and evolved, so that the power supply recovery process under different transfer branches can participate in risk judgment, which is conducive to improving the ability to identify main power supply failure, backup anomaly and transfer failure scenarios.

[0017] (3) This invention recursively selects risk source nodes along the reverse power supply direction using the GraphSAGE backtracking pointer model, and uses the selection result of each time as the input for the next backtracking to generate a risk backtracking chain. This enables the risk identification results to be combined with the disturbance propagation process, which is beneficial for locating risk source nodes, restoring the risk propagation path, and improving the interpretability of the risk tracing results of the power supply path of coal mining enterprises. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a knowledge graph-based method for identifying power supply path risks for coal mine users, as proposed in this invention. Figure 2 This is a schematic diagram illustrating the generation of a power supply script sequence for a knowledge graph-based method for identifying power supply path risks in coal mines, as proposed in this invention. Figure 3 This diagram illustrates the risk identification and backtracking process of a knowledge graph-based method for identifying power supply path risks in coal mines, as proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-3 A knowledge graph-based method for identifying power supply path risks for coal mine users includes the following steps: Collect relevant power supply data from coal mining enterprises, construct power supply maps for coal mining enterprises, and generate basic representations of the power supply maps; Based on the basic representation of the power supply diagram and the dispatch and handling records, a power transfer script sequence is generated, and the correspondence between the power transfer script sequence and the power supply path in the power supply diagram of the coal mine enterprise is established. Based on the power supply data of coal mining enterprises, risk trigger nodes are identified. The node representation of the risk trigger nodes is expanded to include the initial disturbance representation containing the power outage disturbance token. The power outage disturbance token participates in the subsequent neighborhood update as an independent hidden state, forming an initial disturbance power supply graph. The initial disturbance power supply graph is input into the GraphSAGE model that integrates disturbance propagation. The node representation and the power loss disturbance representation are updated synchronously in the node update function. The power loss disturbance token is propagated along the power supply direction. The perturbation propagation process is branched based on the transfer script sequence. The transfer branch state is embedded into the GraphSAGE node update function. The node representation generates the final branch representation under the combined action of the transfer script sequence and the power failure perturbation token. Based on the final representation of the branches, generate the risk identification result of the power supply path of the coal mining enterprise, and construct the backtracking query vector when the risk identification result of the power supply path of the coal mining enterprise reaches the backtracking trigger condition. Input the backtracking query vector into the GraphSAGE model, recursively select risk source nodes along the reverse power supply direction, use the selection result of each time as the input for the next backtracking, generate a risk backtracking chain, and generate the risk source tracing result of the power supply path of the coal mine enterprise based on the risk backtracking chain and the disturbance propagation process.

[0021] In this embodiment, constructing the power supply diagram for a coal mining enterprise specifically includes: After collecting power supply-related data from coal mining enterprises, object normalization processing is performed on the data. Power supply objects are converted into graph nodes, and the power supply business relationships between power supply objects are converted into directed edges. The start and end nodes of the directed edges are determined by the power supply direction, resulting in a coal mining enterprise power supply graph with directional attributes. The power supply-related data of coal mining enterprises is used to at least characterize the power supply subordination, connection, and operational status relationships between coal mining enterprises and upstream power supply objects. Object normalization processing includes unifying the name, number, voltage level, affiliated station, affiliated line, and connected coal mining enterprise for the same power supply object. Records with the same equipment number are merged into one power supply object. Records lacking equipment numbers but with a name similarity of not less than 0.90 and consistent voltage level and affiliated station are merged into one power supply object. This process is then used to normalize each power supply object. A unique object identifier is generated for each power supply object. When a power supply object is converted into a graph node, a unique object identifier corresponds to a graph node in the power supply graph of the coal mine enterprise. A graph node must contain at least a node identifier, a node type, the identifier of the coal mine enterprise to which it belongs, the identifier of the power supply path to which it belongs, and node attribute fields. The node type is determined according to the power supply level in the power supply path of the coal mine enterprise. The power supply level is recorded sequentially from the power source side to the load side, so that the graph node can represent the power source side object, intermediate power supply object, and load side object in the power supply path of the coal mine enterprise. When the power supply business relationship is converted into a directed edge, a candidate edge is generated based on the connection record between two power supply objects. The upstream and downstream positions of the power supply objects in the power supply path are used to determine whether the candidate edge is retained. When the two graph nodes corresponding to the candidate edge have a direct power supply connection, a backup power supply connection, a transfer connection, or an access connection, the candidate edge is written into the power supply graph of the coal mine enterprise. Node attribute codes are written to graph nodes, and edge relationship codes are written to directed edges. The node attribute codes and edge relationship codes are matrixed according to the topological connection relationship of the power supply graph of the coal mine enterprise to generate the basic representation of the power supply graph. The node attribute code is obtained by concatenating the node type code and the node state code. The node type code adopts one-hot encoding, and the node state code is generated by numerical normalization. The edge relationship code consists of the edge relationship type code and the direction code. The edge relationship type code is used to distinguish the power supply business relationship corresponding to the directed edge, and the direction code is used to record the power supply direction of the directed edge from the starting node to the ending node. A node feature matrix is ​​generated according to the node arrangement order in the power supply graph of the coal mine enterprise. Each row in the node feature matrix corresponds to the node attribute code of a graph node. An edge relationship matrix is ​​generated according to the directed edge arrangement order in the power supply graph of the coal mine enterprise. The elements in the edge relationship matrix correspond to the elements in the adjacency matrix that indicate the existence of directed edges.

[0022] In this embodiment, generating the power transfer script sequence based on the power supply diagram basic representation and scheduling records specifically includes: The dispatch and handling record is defined as a record of the dispatch operation process for power supply anomalies in coal mining enterprises. The dispatch and handling record records the fault handling object, the power supply restoration object, the dispatch operation content, and the operation execution sequence. The fault handling object refers to the power supply object that is isolated, switched, or restored during the handling process. The power supply restoration object refers to the target object that needs to have its power restored after the handling is completed. The dispatch operation content refers to the state change action applied to the fault handling object. The operation execution sequence is determined based on the time information in the dispatch and handling record. Based on the basic representation of the power supply diagram, the corresponding graph nodes of the fault handling objects and power restoration objects in the dispatching and handling records are determined in the power supply diagram of the coal mine enterprise. When determining the corresponding graph nodes, the object identifier in the dispatching and handling records is matched with the graph node identifier in the power supply diagram of the coal mine enterprise. If no matching result is obtained, the object name, the power supply area, and the voltage level are combined for matching, and the name similarity threshold for combined matching is set to 0.90. When multiple candidate graph nodes are obtained, the candidate graph node with the shortest directed path length to the power restoration object is selected as the corresponding graph node. The scheduling records are sorted according to the order of operation execution. Each scheduling operation is converted into a script instruction, and consecutive script instructions are connected according to their execution order to generate a transfer script sequence. When generating the transfer script sequence, each sorted scheduling operation is converted into a script instruction containing the operation object, operation action, execution sequence number, and post-execution status. The operation action uses the same action name as the scheduling record, and the execution sequence number is set to increment according to the order of operation execution. A sequential connection relationship is established between adjacent script instructions. Based on the corresponding graph nodes involved in the script instructions, the directed edges and power supply paths associated with the script instructions are determined in the power supply graph of the coal mine enterprise. Taking the graph node corresponding to the operation object in the script instructions as the search starting point, a directed path is searched in the power supply graph of the coal mine enterprise along the power supply direction to the graph node corresponding to the power restoration object. The directed path found and passing through the graph node corresponding to the operation object is determined as the power supply path associated with the script instructions. When the script instructions are used to restore power supply, the associated power supply path is the directed path that supplies power to the power restoration object after the transfer. When the script instructions are used to isolate faults, the associated power supply path is the directed path that includes the fault handling object before the handling. The power transfer script sequence is bound to the power supply path identifier to form a correspondence between the power transfer script sequence and the power supply path in the power supply diagram of the coal mine enterprise.

[0023] In this embodiment, determining the risk triggering node based on coal mine enterprise power supply-related data specifically includes: From the power supply data of coal mining enterprises, identify the power supply objects experiencing power supply anomalies. Match these anomalies with nodes in the coal mining enterprise's power supply diagram, and designate the matched nodes as risk trigger nodes. Match the power supply object's name or number with nodes in the coal mining enterprise's power supply diagram. If the anomaly record only contains the coal mining enterprise's name, select the node closest to the coal mining enterprise's node and located in the upstream power supply direction from the corresponding power supply path as the risk trigger node. If the anomaly record contains multiple power supply objects, designate all nodes matched for each power supply object as risk trigger nodes. Read the node representation of the risk-triggered node in the basic power supply graph representation, construct a power outage disturbance token with the same dimension as the node representation, the power outage disturbance token is a hidden state vector participating in the GraphSAGE neighborhood update, the power outage disturbance token records the propagation state of the power outage impact caused by the power supply anomaly of the coal mine enterprise in the coal mine enterprise power supply graph; the node representation is the vectorized representation of the corresponding graph node in the basic power supply graph representation, the vector length of the power outage disturbance token is the same as the vector length of the node representation; the power outage disturbance token is synchronously input into the node update function with the node representation in the GraphSAGE neighborhood update; The power outage disturbance token corresponding to the risk-triggered node is initialized as a non-zero hidden state vector, and the power outage disturbance token corresponding to the non-risk-triggered node is initialized as a zero-value hidden state vector. When initializing the power outage disturbance token corresponding to the risk-triggered node, the power supply anomaly records of the risk-triggered node within a preset historical statistical period are extracted from the power supply related data of the coal mining enterprise. The frequency of anomaly occurrence, duration of anomaly, and scope of anomaly impact of the risk-triggered node are calculated. The frequency of anomaly occurrence, duration of anomaly, and scope of anomaly impact are normalized respectively to obtain the anomaly statistical vector of the risk-triggered node. The anomaly statistical vector is mapped to a hidden state vector with the same dimension as the node representation. The mapped hidden state vector is used as the power outage disturbance token corresponding to the risk-triggered node. When there is no power supply anomaly record of the risk-triggered node within the preset historical statistical period, the power outage disturbance token is generated using the anomaly statistical mean of graph nodes with the same node type as the risk-triggered node. The node representation of a risk-triggered node is connected to its corresponding power outage disturbance token to generate the initial disturbance representation of the risk-triggered node. The node representation of a non-risk-triggered node is connected to its corresponding zero-value hidden state vector to generate the initial disturbance representation of the non-risk-triggered node. The connection between the node representation and the power outage disturbance token is done by vector concatenation, with the concatenation direction being the feature dimension direction. The same concatenation method is used to generate the initial disturbance representation of non-risk-triggered nodes, so that all nodes in the power supply diagram of the coal mine enterprise have a consistent initial disturbance representation dimension. Keeping the directed edge relationships in the power supply diagram of the coal mine enterprise unchanged, the power supply diagram of the coal mine enterprise containing the initial disturbance representation of each graph node is determined as the initial disturbance power supply diagram; the initial disturbance power supply diagram adopts the node arrangement order, directed edge direction and edge relationship encoding in the power supply diagram of the coal mine enterprise; the node characteristics in the initial disturbance power supply diagram are composed of the initial disturbance representation of each graph node, and the edge relationships in the initial disturbance power supply diagram still adopt the edge relationship encoding in the basic representation of the power supply diagram.

[0024] In this embodiment, inputting the initial disturbance power supply graph into the GraphSAGE model for fusion disturbance propagation specifically includes: Read the initial disturbance representation of each graph node from the initial disturbance power supply graph, and split the initial disturbance representation into node representation and power failure disturbance token; For the current graph node, read the upstream neighbor nodes along the power supply direction of the directed edge, perform GraphSAGE neighborhood aggregation on the node representations of the upstream neighbor nodes, and perform same-direction aggregation on the power outage disturbance tokens of the upstream neighbor nodes; the upstream neighbor nodes are graph nodes that supply power to the current graph node through directed edges; the GraphSAGE neighborhood aggregation uses the mean aggregator in the GraphSAGE model; the neighborhood aggregation result is obtained by averaging the node representations of the upstream neighbor nodes by dimension, and the same-direction aggregation result is obtained by averaging the power outage disturbance tokens of the upstream neighbor nodes according to the same node order and the same aggregation method; The power outage disturbance token of the current graph node is superimposed with the power outage disturbance token aggregated in the same direction. The superposition result is then normalized to obtain the updated power outage disturbance representation of the current graph node. The superposition is performed by dimension-by-dimensional addition, and the normalization is performed by Sigmoid normalization. The Sigmoid function is applied to each dimension in the dimension-by-dimensional addition result. The updated power outage disturbance representation is used to represent the degree of upstream power outage impact received by the current graph node. The node representation of the current graph node, the GraphSAGE neighborhood aggregation result, and the updated power outage perturbation representation are all input into the node update function. Activation processing is then applied to the output of the node update function to obtain the updated node representation of the current graph node. The node update function employs a linear transformation function, whose input is the concatenation of the node representation of the current graph node, the GraphSAGE neighborhood aggregation result, and the updated power outage perturbation representation along the feature dimension. The output dimension of the linear transformation function is consistent with the node representation dimension of the current graph node, and the ReLU activation function is used for activation processing. The updated power outage perturbation representation is then concatenated and fed into the linear transformation function. The above updates are performed sequentially on the graph nodes in the power supply diagram of the coal mine enterprise according to the power supply direction. The power outage disturbance token of the risk-triggered node is written to the downstream graph node along the directed edge until the updated node representation and the updated power outage disturbance representation of each graph node are obtained. The update order of the graph nodes is determined according to the directed path order from the power source side to the load side in the power supply diagram of the coal mine enterprise. When there are multiple directed paths in the power supply diagram of the coal mine enterprise, the updates are performed in parallel on the graph nodes located at the same power supply level. When a graph node receives power outage disturbance tokens from multiple upstream neighbor nodes at the same time, the power outage disturbance tokens of multiple upstream neighbor nodes are first aggregated in the same direction, and then written to the current graph node. After the update is completed, the updated node representation and the updated power outage disturbance representation of each graph node are saved as the output of the disturbance propagation process.

[0025] In this embodiment, the branching evolution of the disturbance propagation process based on the transfer script sequence specifically includes: Read the power transfer script sequence and the updated node representations and updated power outage disturbance representations of each graph node, and determine the power transfer operation steps according to the script instruction sequence in the power transfer script sequence; The disturbance propagation process is copied according to the power supply path and operation object node corresponding to the script instruction, forming a transfer branch corresponding to the transfer script sequence. When copying the disturbance propagation process, the copied objects are the updated node representations of each graph node, the updated power outage disturbance representations of each graph node, and the directed edge relationships in the power supply graph of the coal mine enterprise. Each transfer script sequence generates a transfer branch, and the initial state of the transfer branch is consistent with the copied disturbance propagation process, with the script sequence identifier as the transfer branch identifier. In each power transfer operation step, the node state of the operation object node within the power transfer branch is changed according to the script instructions, and the state of the directed edges connected to the operation object node is adjusted synchronously. The operation object node is the graph node corresponding to the operation object recorded in the script instructions. The node state is updated according to the state after execution in the script instructions. When the script instructions change the power supply object corresponding to the operation object node from available to unavailable, the state of the directed edges connected to and passing through the operation object node is updated to disconnected. When the script instructions change the power supply object corresponding to the operation object node from unavailable to available, the state of the directed edges connected to the operation object node and satisfying the power supply direction is updated to connected. The adjusted node state, directed edge state, updated power outage disturbance representation, and script instruction code are written together into the power transfer branch state; the power transfer branch state is used to record the changes in the power supply diagram after the current power transfer operation step is executed; the script instruction code is generated by the execution sequence number of the script instruction, the identification of the operation object node, and the status code after execution; the power transfer branch state is gradually overwritten and updated according to the script instruction sequence, so that each power transfer operation step corresponds to a power transfer branch state; The power transfer branch state, the node representation of the current graph node, the GraphSAGE neighborhood aggregation result, and the updated power outage perturbation representation are all input into the GraphSAGE node update function. Activation processing is performed on the output of the node update function to obtain the branch node representation under the current power transfer operation step. The input of the GraphSAGE node update function is obtained by concatenating the node representation of the current graph node, the GraphSAGE neighborhood aggregation result, the updated power outage perturbation representation, and the power transfer branch state along the feature dimension. When the power transfer branch state indicates that the power supply path of the node to be operated on has been disconnected, GraphSAGE neighborhood aggregation is only performed on the upstream neighbor nodes corresponding to the directed edges in the connected state. When the power transfer branch state indicates that the backup power supply path has been connected, GraphSAGE neighborhood aggregation includes the upstream neighbor nodes on the backup power supply path in the aggregation scope. The graph nodes in the power transfer branch are continuously updated according to the script instructions until the power transfer script sequence is completed. The branch node representation and branch power loss disturbance representation obtained after the power transfer script sequence is completed are determined as the final branch representation. After each power transfer operation step is completed, the branch node representation and branch power loss disturbance representation under the current power transfer operation step are used as the input of the next power transfer operation step. When all script instructions in the power transfer script sequence are completed, the power transfer branch update is stopped, and the branch node representation and branch power loss disturbance representation corresponding to the last power transfer operation step are used together as the final branch representation.

[0026] In this embodiment, constructing a backtracking query vector when the risk identification result of the power supply path of a coal mine enterprise reaches the backtracking trigger condition specifically includes: The branch node representation and branch power outage disturbance representation corresponding to the coal mine enterprise node are read from the final branch representation. The branch node representation and branch power outage disturbance representation are concatenated on the feature dimension and then input into the risk discrimination layer to obtain the power supply path risk value corresponding to the transfer branch. The risk discrimination layer is composed of a fully connected layer and a Sigmoid function. After the branch node representation and branch power outage disturbance representation are concatenated, they are first input into the fully connected layer to obtain the intermediate risk discrimination value. The Sigmoid function is then performed on the intermediate risk discrimination value. The closer the power supply path risk value is to 1, the higher the risk of the power supply path under the corresponding transfer branch. According to the correspondence between the power transfer script sequence and the power supply path, the power supply path risk value is written into the corresponding power supply path to form the power supply path risk identification result of the coal mine enterprise. When writing the power supply path, the power supply path risk value, the power transfer branch identifier and the corresponding coal mine enterprise node identifier are written into the risk identification record using the power supply path identifier as the index. When the same power supply path corresponds to multiple power transfer branches, the power supply path risk values ​​corresponding to all power transfer branches are written into the same power supply path identifier, and the power supply path risk identification result of the coal mine enterprise retains the branch source. When the power supply path risk value reaches the preset risk threshold, the power supply path risk identification result of the coal mine enterprise is determined to meet the backtracking trigger condition. The preset risk threshold is determined based on the statistical distribution of the power supply path risk values ​​in the training samples. The lower quartile of the power supply path risk value corresponding to the high-risk sample in the training samples is taken as the preset risk threshold. When there are no labeled high-risk samples, the 75th percentile of all power supply path risk values ​​in the training samples is taken as the preset risk threshold. When the power supply path risk value is greater than or equal to the preset risk threshold, it is determined that the power supply path risk identification result of the coal mine enterprise has met the backtracking trigger condition. When the backtracking trigger condition is met, the representation of the coal mine enterprise node, the branch power outage disturbance representation, and the power supply path risk value corresponding to the power supply path that meets the backtracking trigger condition are read. The read results are concatenated and a linear transformation is performed to construct a backtracking query vector. When constructing the backtracking query vector, the representation of the coal mine enterprise node adopts the branch node representation of the coal mine enterprise node in the power supply path that meets the backtracking trigger condition, and the branch power outage disturbance representation adopts the branch power outage disturbance representation of the same coal mine enterprise node under the same transfer branch. The representation of the coal mine enterprise node, the branch power outage disturbance representation, and the power supply path risk value are concatenated according to the feature dimension.

[0027] In this embodiment, inputting the backtracking query vector into the backtracking pointer GraphSAGE model specifically includes: The coal mining enterprise node in the power supply path that meets the backtracking trigger condition is determined as the initial backtracking node, and the backtracking query vector is used as the initial backtracking input. The initial backtracking node is the coal mining enterprise node corresponding to the power supply path that meets the backtracking trigger condition. The initial backtracking input is the backtracking query vector. The vector dimension of the backtracking query vector is consistent with the input dimension of the backtracking pointer GraphSAGE model. When the dimensions are inconsistent, the backtracking query vector is mapped to the input dimension of the backtracking pointer GraphSAGE model through linear transformation. At the current backtracking node, upstream graph nodes are read along the reverse power supply direction. The node representation and power outage disturbance representation of the upstream graph nodes are input together with the current backtracking input into the GraphSAGE backtracking pointer model to obtain the backtracking pointer value corresponding to the upstream graph node. The reverse power supply direction is opposite to the direction of the directed edges in the power supply diagram of the coal mine enterprise. At the current backtracking node, upstream graph nodes that can directly supply power to the current backtracking node through a directed edge are read, and each upstream graph node is taken as a candidate backtracking node. The GraphSAGE backtracking pointer model concatenates the node representation, power outage disturbance representation and current backtracking input of the candidate backtracking nodes and then performs a linear transformation and Sigmoid function calculation to obtain the backtracking pointer value. The upstream graph node with the largest backtracking pointer value is selected as the candidate node for the current risk source. This candidate node is written into the risk backtracking chain. The node representation and power outage disturbance representation of this candidate node are merged into the current backtracking input to generate the next backtracking input. The risk backtracking chain records the candidate nodes for the risk source according to the backtracking selection order. After each selection, the node identifier, backtracking pointer value, and corresponding edge direction of this candidate node are written into the risk backtracking chain. The next backtracking input is obtained by concatenating the current backtracking input, the node representation of this candidate node, and the power outage disturbance representation of this candidate node along the feature dimension, followed by a linear transformation, so that the next backtracking input carries the selection result of the previous one. The current risk source candidate node is used as the next backtracking node. The reverse read, pointer calculation, node writing, and backtracking input update are repeatedly executed until the current backtracking node has no upstream graph nodes or the maximum backtracking pointer value is lower than the backtracking stop threshold. The backtracking stop threshold is determined based on the statistical distribution of the backtracking pointer values ​​corresponding to the risk source nodes in the training samples. The lower quartile of the backtracking pointer value corresponding to the labeled risk source node is taken as the backtracking stop threshold. When there is no labeled risk source node, the 75th percentile of the backtracking pointer values ​​corresponding to all candidate backtracking nodes in the training phase is taken as the backtracking stop threshold. To avoid repeated backtracking, risk source candidate nodes that have already been written in the same risk backtracking chain will no longer be used as subsequent candidate backtracking nodes. The last candidate node of the risk source written in the risk backtracking chain is determined as the risk source node. The risk source tracing result of the power supply path of the coal mine enterprise is generated based on the risk backtracking chain and the disturbance propagation process. The risk source node is the last candidate node of the risk source corresponding to the termination of the risk backtracking chain. The risk source tracing result of the power supply path of the coal mine enterprise shall record at least the risk source node, the risk backtracking chain, the power supply path that meets the backtracking trigger condition, and the propagation node that overlaps with the risk backtracking chain during the disturbance propagation process. When there are common graph nodes between the disturbance propagation process and the risk backtracking chain, the common graph node shall be written as the risk propagation confirmation node into the risk source tracing result of the power supply path of the coal mine enterprise.

[0028] The training data is divided into two categories: risk identification training data and backtracking training data.

[0029] The risk identification training data is used to train the GraphSAGE model for fusion disturbance propagation and the risk discrimination layer. The risk identification training data consists of a coal mine enterprise power supply map, a basic representation of the power supply map, an initial disturbance power supply map, a power transfer script sequence, a branch final representation, and power supply path risk labels. The power supply path risk labels are generated based on historical coal mine enterprise power supply anomaly handling results. When a power supply anomaly causes a power outage, loss of power to critical loads, power transfer failure, or recovery time exceeding a set time threshold, the corresponding power supply path is marked as high-risk. When a power supply anomaly does not cause a power outage but there is a power transfer delay, insufficient backup power supply capacity, or disturbance propagation to coal mine enterprise nodes, the corresponding power supply path is marked as medium-risk. When a power supply anomaly is successfully isolated or absorbed by the power transfer script sequence and does not affect the coal mine enterprise's critical loads, the corresponding power supply path is marked as low-risk. This is primarily done manually. During training, the initial perturbation power supply graph is input into the GraphSAGE model that integrates perturbation propagation, and the final branch representation is generated based on the power transfer script sequence. The power supply path risk value is output through the risk discrimination layer, and the model parameters are updated using the difference between the power supply path risk value and the power supply path risk label.

[0030] The backtracking training data is used to train the GraphSAGE backtracking pointer model. This training data consists of power supply paths that meet the backtracking trigger conditions, backtracking query vectors, historical risk source nodes, historical risk propagation paths, and disturbance propagation processes. Historical risk source nodes are determined based on fault analysis records, dispatching and handling records, and maintenance records after power supply anomalies in coal mining enterprises. Historical risk propagation paths are formed based on the location of the power supply anomaly, the direction of power supply, the location of power transfer failure, and the affected coal mining enterprise nodes. During training, the backtracking query vector is used as the initial backtracking input, and the coal mining enterprise nodes are used as the initial backtracking nodes. The GraphSAGE backtracking pointer model calculates the backtracking pointer values ​​of candidate upstream graph nodes sequentially along the reverse power supply direction. Graph nodes in the corresponding order of the historical risk propagation path are used as supervision targets, enabling the GraphSAGE backtracking pointer model to learn to select the correct risk source candidate nodes in each backtracking until a risk backtracking chain consistent with the historical risk propagation path is generated.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to a power supply path risk identification and tracing scenario in a coal mining enterprise. In this scenario, the coal mining enterprise has a main power supply path, a backup power supply path, and an emergency power supply path, with clear upstream and downstream power supply directions between the power supply objects. The safe production load of the coal mining enterprise has high requirements for power supply continuity. When the main power supply path experiences power supply anomalies, the backup path becomes unavailable, the switch status is abnormal, or the dispatching and transfer process is not fully executed, critical loads may be affected. Traditional processing methods usually rely on manually checking topology diagrams, dispatching records, and equipment status. While this can identify some obvious anomalies, it is difficult to simultaneously analyze the propagation of the impact of power outages, changes in transfer paths, and the location of risk sources, easily leading to problems such as delayed risk identification and unclear tracing results.

[0032] In this scenario, power supply-related data from coal mining enterprises is collected, and the power supply objects are transformed into graph nodes in the coal mining enterprise power supply graph. The power supply business relationships between the power supply objects are transformed into directed edges with power supply directions. Through this processing method, the primary power supply path, backup path, coal mining enterprise nodes, and critical load nodes can be represented in the same power supply graph, and the subsequent model can directly analyze along the power supply direction. To enable the power supply graph to be processed by the GraphSAGE model, the attribute information of the graph nodes is written into the node representation, and the relationship information of the directed edges is written into the edge relationship representation, forming the basic representation of the power supply graph.

[0033] A power transfer script sequence is generated based on the dispatch and handling records. The dispatch and handling records describe the handling process after a power supply anomaly occurs in a coal mine, including the fault handling object, the power restoration object, the dispatch operation content, and the operation execution sequence. After converting the dispatch operation content into script instructions, a power transfer script sequence is formed according to the execution order. This sequence is then bound to the power supply path in the coal mine's power supply diagram. The dispatch and handling process is no longer just manually recorded, but transformed into a model input that can drive changes in the power supply diagram's state.

[0034] During risk identification, the power supply objects experiencing power supply anomalies in coal mining enterprises are determined from relevant power supply data, and the corresponding graph nodes are identified as risk trigger nodes. Then, a power outage disturbance token is added to the node representation of the risk trigger node to form an initial disturbance power supply graph. The power outage disturbance token participates in GraphSAGE neighborhood updates as an independent hidden state, recording the propagation status of the power outage impact caused by the coal mining enterprise's power supply anomaly in the power supply graph. For non-risk trigger nodes, a zero-value disturbance state is written, enabling the model to distinguish between the anomaly origin and normal power supply nodes.

[0035] After inputting the initial disturbance power supply graph into the GraphSAGE model that integrates disturbance propagation, the model reads upstream neighbor nodes along the power supply direction, performs neighborhood aggregation on the node representations, and performs same-direction aggregation on power outage disturbance tokens. The updated power outage disturbance representations participate in the node update function, ensuring that the node representations reflect not only power supply topology neighborhood information but also the impact of upstream power outages. Through this processing, coal mine enterprise nodes and critical load nodes can obtain node representations related to the propagation of power supply anomalies.

[0036] During the power transfer analysis, the disturbance propagation process is branched and evolved according to the power transfer script sequence. Each power transfer script sequence corresponds to a power transfer branch, and the script instructions change the node state and directed edge state within the branch. The power transfer branch state is embedded in the GraphSAGE node update function, enabling the model to determine whether the power outage disturbance is isolated, weakened, or continues to propagate to the coal mine enterprise nodes under different power transfer paths. After the script execution is complete, the final branch representation is obtained, based on which the risk identification result of the power supply path for the coal mine enterprise is generated.

[0037] When the risk identification results of the power supply path of a coal mining enterprise reach the backtracking trigger condition, a backtracking query vector is constructed and input into the backtracking pointer GraphSAGE model. The backtracking pointer GraphSAGE model starts from the coal mining enterprise node, reads upstream graph nodes along the reverse power supply direction, calculates the backtracking pointer value of candidate nodes, and writes the node with the largest pointer value into the risk backtracking chain. Each written risk source candidate node serves as part of the input for the next backtracking, driving the model to continue recursively backtracking upstream until the stopping condition is met. The model generates a risk backtracking chain, and combined with the disturbance propagation process, obtains the risk source tracing results of the coal mining enterprise's power supply path.

[0038] Table 1: Comparison of Risk Identification Results for Power Supply Paths of Coal Mine Users

[0039] As shown in Table 1, the method of this invention exhibits superior overall performance in the scenario of power supply path risk identification for coal mine users. The risk identification accuracy reaches 94.2%, an improvement of 15.6 percentage points compared to the 78.6% of the traditional static rule method, and an improvement of 8.3 percentage points compared to the 85.9% of the single GraphSAGE method. This indicates that after introducing the power outage disturbance token, the model can more fully express the impact of power supply anomalies propagating along the power supply direction. The overall F1 score reaches 93.4%, and the false negative rate decreases to 6.1%, indicating that this invention has good stability in identifying high-risk power supply paths and can reduce the number of high-risk paths that are not identified. The risk tracing accuracy reaches 90.7%, higher than the 78.6% of the single GraphSAGE method, indicating that the backtracking pointer mechanism can effectively locate the risk source node. The critical load impact identification accuracy reaches 93.5%, indicating that the evolution of the transfer script branch can more accurately reflect the impact of power supply anomalies on critical loads. The average identification time is 3.21 seconds, which is higher than the comparative methods, but still meets the needs of coal mine enterprises for rapid identification of power supply risks.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying power supply path risks for coal mine users based on knowledge graphs, characterized in that, Includes the following steps: Collect relevant power supply data from coal mining enterprises, construct power supply maps for coal mining enterprises, and generate basic representations of the power supply maps; Based on the basic representation of the power supply diagram and the dispatch and handling records, a power transfer script sequence is generated, and the correspondence between the power transfer script sequence and the power supply path in the power supply diagram of the coal mine enterprise is established. Based on the power supply data of coal mining enterprises, risk trigger nodes are identified. The node representation of the risk trigger nodes is expanded to include the initial disturbance representation containing the power outage disturbance token. The power outage disturbance token participates in the subsequent neighborhood update as an independent hidden state, forming an initial disturbance power supply graph. The initial disturbance power supply graph is input into the GraphSAGE model that integrates disturbance propagation. The node representation and the power loss disturbance representation are updated synchronously in the node update function. The power loss disturbance token is propagated along the power supply direction. The perturbation propagation process is branched based on the transfer script sequence. The transfer branch state is embedded into the GraphSAGE node update function. The node representation generates the final branch representation under the combined action of the transfer script sequence and the power failure perturbation token. Based on the final representation of the branches, generate the risk identification result of the power supply path of the coal mining enterprise, and construct the backtracking query vector when the risk identification result of the power supply path of the coal mining enterprise reaches the backtracking trigger condition. Input the backtracking query vector into the GraphSAGE model, recursively select risk source nodes along the reverse power supply direction, use the selection result of each time as the input for the next backtracking, generate a risk backtracking chain, and generate the risk source tracing result of the power supply path of the coal mine enterprise based on the risk backtracking chain and the disturbance propagation process.

2. The method for identifying power supply path risks for coal mine users based on knowledge graphs according to claim 1, characterized in that, The construction of the power supply diagram for coal mining enterprises specifically includes: After collecting power supply-related data of coal mining enterprises, the data is normalized by object, converting power supply objects into graph nodes, converting the power supply business relationships between power supply objects into directed edges, and determining the start and end nodes of the directed edges by the power supply direction to obtain a power supply graph of coal mining enterprises with directional attributes. Write node attribute codes to graph nodes and edge relationship codes to directed edges. Then, matrix the node attribute codes and edge relationship codes according to the topological connection relationship of the power supply graph of the coal mine enterprise to generate the basic representation of the power supply graph.

3. The method for identifying power supply path risks for coal mine users based on knowledge graphs according to claim 2, characterized in that, The specific steps of generating the power transfer script sequence based on the power supply diagram and scheduling records include: The dispatch and handling record is defined as a record of the dispatch operation process for power supply anomalies in coal mining enterprises. The dispatch and handling record records the fault handling object, the power supply restoration object, the dispatch operation content and the operation execution sequence. Based on the basic representation of the power supply diagram, determine the corresponding diagram nodes of the fault handling objects and power supply restoration objects in the dispatching and handling records in the power supply diagram of the coal mining enterprise; The scheduling and handling records are sorted according to the order of operation execution. Each scheduling operation is converted into a script instruction. The consecutive script instructions are connected according to the order of execution to generate a transfer script sequence. Based on the corresponding graph nodes involved in the script instructions, determine the directed edges and power supply paths associated with the script instructions in the power supply graph of the coal mine enterprise; The power transfer script sequence is bound to the power supply path identifier to form a correspondence between the power transfer script sequence and the power supply path in the power supply diagram of the coal mine enterprise.

4. The method for identifying power supply path risks for coal mine users based on knowledge graphs according to claim 3, characterized in that, The determination of risk triggering nodes based on coal mine enterprise power supply-related data specifically includes: Identify the power supply objects that have experienced power supply anomalies from the relevant power supply data of coal mining enterprises, match the power supply objects that have experienced power supply anomalies with the graph nodes in the power supply diagram of coal mining enterprises, and determine the matched graph nodes as risk trigger nodes. Read the node representation of the risk-triggered node in the basic representation of the power supply graph, construct a power outage disturbance token with the same dimension as the node representation, the power outage disturbance token is a hidden state vector that participates in the GraphSAGE neighborhood update, and the power outage disturbance token records the propagation state of the power outage impact caused by the power supply anomaly of the coal mine enterprise in the power supply graph of the coal mine enterprise. Initialize the power outage disturbance token corresponding to the risk-triggered node as a non-zero hidden state vector, and initialize the power outage disturbance token corresponding to the non-risk-triggered node as a zero-value hidden state vector. Connect the node representation of the risk-triggered node with the corresponding power failure disturbance token to generate the initial disturbance representation of the risk-triggered node; connect the node representation of the non-risk-triggered node with the corresponding zero-value hidden state vector to generate the initial disturbance representation of the non-risk-triggered node. Keeping the directed edge relationships in the power supply diagram of the coal mine enterprise unchanged, the power supply diagram of the coal mine enterprise containing the initial representation of the disturbance of each graph node is determined as the initial disturbance power supply diagram.

5. The method for identifying power supply path risks for coal mine users based on knowledge graphs according to claim 4, characterized in that, The specific steps of inputting the initial perturbation power supply graph into the GraphSAGE model for fusion perturbation propagation include: Read the initial disturbance representation of each graph node from the initial disturbance power supply graph, and split the initial disturbance representation into node representation and power failure disturbance token; For the current graph node, read the upstream neighbor nodes along the power supply direction of the directed edge, perform GraphSAGE neighborhood aggregation on the node representation of the upstream neighbor nodes, and perform same-direction aggregation on the power outage disturbance tokens of the upstream neighbor nodes. The power outage disturbance token of the current graph node is superimposed with the power outage disturbance token after being aggregated in the same direction. The superposition result is then normalized to obtain the updated power outage disturbance representation of the current graph node. The node representation of the current graph node, the GraphSAGE neighborhood aggregation result, and the updated power outage perturbation representation are input into the node update function. Activation processing is performed on the output of the node update function to obtain the updated node representation of the current graph node. The above updates are performed sequentially on the graph nodes in the power supply diagram of the coal mining enterprise according to the power supply direction. The power outage disturbance token of the risk-triggered node is written to the downstream graph node along the directed edge until the updated node representation and the updated power outage disturbance representation of each graph node are obtained.

6. The method for identifying power supply path risks for coal mine users based on knowledge graphs according to claim 5, characterized in that, The specific steps of performing branched evolution on the disturbance propagation process based on the transferred script sequence include: Read the power transfer script sequence and the updated node representations and updated power outage disturbance representations of each graph node, and determine the power transfer operation steps according to the script instruction sequence in the power transfer script sequence; Based on the power supply path and the node to be operated corresponding to the script instruction, the disturbance propagation process is copied to form a transfer branch corresponding to the transfer script sequence; In each transfer operation step, the node state of the operation object node in the transfer branch is changed according to the script instructions, and the state of the directed edge connected to the operation object node is adjusted synchronously. The adjusted node state, directed edge state, updated power outage disturbance representation, and script instruction encoding are written together into the power transfer branch state; The power transfer branch state, the node representation of the current graph node, the GraphSAGE neighborhood aggregation result, and the updated power outage disturbance representation are input together into the GraphSAGE node update function. Activation processing is performed on the output of the node update function to obtain the branch node representation under the current power transfer operation step. The graph nodes in the transfer branch are continuously updated according to the script instructions until the transfer script sequence is completed. The branch node representation and the branch power failure disturbance representation obtained after the transfer script sequence is completed are determined as the final branch representation.

7. The method for identifying power supply path risks for coal mine users based on knowledge graphs according to claim 6, characterized in that, The construction of the backtracking query vector when the risk identification result of the power supply path of a coal mining enterprise reaches the backtracking trigger condition specifically includes: The branch node representation and branch power outage disturbance representation corresponding to the coal mine enterprise node are read from the final branch representation. The branch node representation and branch power outage disturbance representation are concatenated on the feature dimension and then input into the risk discrimination layer to obtain the power supply path risk value corresponding to the transfer branch. Based on the correspondence between the power transfer script sequence and the power supply path, the risk value of the power supply path is written into the corresponding power supply path to form the power supply path risk identification result of the coal mine enterprise. When the risk value of the power supply path reaches the preset risk threshold, the risk identification result of the power supply path of the coal mining enterprise will be determined as the trigger condition for backtracking. When the backtracking trigger condition is met, the coal mine enterprise node representation, branch power failure disturbance representation, and power supply path risk value corresponding to the power supply path that has met the backtracking trigger condition are read. The read results are concatenated and linear transformation is performed to construct the backtracking query vector.

8. The method for identifying power supply path risks for coal mine users based on knowledge graphs according to claim 7, characterized in that, The specific steps of inputting the backtracking query vector into the backtracking pointer GraphSAGE model include: The coal mining enterprise node in the power supply path that meets the backtracking trigger condition is determined as the initial backtracking node, and the backtracking query vector is used as the initial backtracking input. At the current backtracking node, read the upstream graph node along the reverse power supply direction, and input the node representation and power failure disturbance representation of the upstream graph node together with the current backtracking input into the GraphSAGE model to obtain the backtracking pointer value corresponding to the upstream graph node; The upstream graph node with the largest backtracking pointer value is selected as the candidate node of the current risk source. The candidate node of the current risk source is written into the risk backtracking chain. The node representation and power failure disturbance representation of the candidate node of the current risk source are merged into the current backtracking input to generate the next backtracking input. The current risk source candidate node is used as the next backtracking node. The reverse reading, pointer calculation, node writing and backtracking input update are repeatedly executed until the current backtracking node has no upstream graph node or the maximum backtracking pointer value is lower than the backtracking stop threshold. The last candidate node of the risk source written in the risk backtracking chain is determined as the risk source node, and the risk source tracing result of the power supply path of the coal mine enterprise is generated based on the risk backtracking chain and the disturbance propagation process.

Images