Method for positioning packet loss fault of communication network of intelligent substation

Abstract

The invention relates to the technical field of digital information transmission, in particular to a method for positioning a packet loss fault of an intelligent substation communication network, which comprises the following steps of: S1, acquiring real-time transceiving data and state information of each node in the intelligent substation communication network, and constructing a space-time dynamic state map; s2, predicting a packet loss high-risk area by combining historical state information based on a time-space relationship between map modeling nodes, and generating a dynamic risk map; and S3, inputting the risk map into a fault positioning algorithm, fusing node state features and prediction area information, and accurately identifying and outputting a packet loss fault source position. According to the method, high-precision, low-delay and high-interpretability positioning and early warning of the packet loss fault in the communication network of the intelligent substation are realized by constructing the space-time dynamic state map and fusing map neural network prediction and propagation consistency analysis.

Description

Technical Field

[0001] This invention relates to the field of digital information transmission technology, and in particular to a method for locating packet loss faults in the communication network of an intelligent substation. Background Technology

[0002] With the development of smart grids, the stability and reliability of communication networks in smart substations have become a key foundation for ensuring the safe operation of power systems. Communication networks undertake core tasks such as information transmission between devices, status monitoring, and dispatch control. Any data transmission failure, especially packet loss, may cause control anomalies or even system cascading failures. Therefore, how to efficiently and accurately identify the source of packet loss in the communication network is a core technical challenge in the operation and maintenance of smart substations.

[0003] Existing fault location methods for communication networks mainly rely on static rule configuration, simple indicator alarms, or traditional statistical analysis methods, which are difficult to effectively cope with complex scenarios such as multiple nodes, high concurrency, and dynamic topology changes in smart substations. On the one hand, they lack systematic modeling of the spatiotemporal state changes of nodes, making it difficult to capture early features in the fault evolution process. On the other hand, existing methods often ignore the reuse value of historical event patterns, resulting in low fault identification accuracy, high response delay, and lack of interpretability of results, which is not conducive to timely intervention by engineering operation and maintenance personnel. Summary of the Invention

[0004] This invention provides a method for locating packet loss faults in the communication network of intelligent substations. By constructing a spatiotemporal dynamic state map, integrating a graph neural network model with historical event alignment analysis, it accurately predicts potential packet loss risk areas. Furthermore, by combining a propagation consistency verification mechanism, it achieves high-precision identification of fault source nodes. This method improves the foresight of fault warnings, the accuracy of location results, and the interpretability, providing an efficient and reliable fault diagnosis solution for intelligent substations.

[0005] A method for locating packet loss faults in the communication network of an intelligent substation includes the following steps: S1 collects real-time data packet sending and receiving information from multiple nodes in the intelligent substation communication network, and obtains the status information of each node, including latency, bandwidth, and traffic load, to construct a spatiotemporal dynamic status map of the network nodes. S2, Based on the spatiotemporal dynamic state map, model the spatiotemporal relationship between each node, predict the spatiotemporal region where packet loss faults occur, and generate a dynamic risk map of packet loss by combining historical state information. S3. Input the dynamic risk map into the fault location algorithm. Through multi-dimensional information fusion, combine the state characteristics of each node and the predicted fault occurrence area to locate the source of packet loss fault and output the location result.

[0006] Optionally, S1 includes: S11 collects real-time data packet sending and receiving information of each communication node in the smart substation through the communication monitoring system, including the data packet transmission timestamp, sending frequency, receiving frequency, packet loss situation and data transmission stability of each node; S12: Obtain the status information of each node, including the node's latency, bandwidth, and traffic load, and perform spatiotemporal correlation processing on the status information to construct a spatiotemporal dynamic status map of the network nodes.

[0007] Optionally, S11 includes: S111, through data monitoring units deployed on core switching equipment or border routing nodes of the communication network, monitors the incoming and outgoing data flows of each communication node, logs the sending and receiving operations of each data packet, and adds a local timestamp to each record. and ,in, For the first The local sending timestamp of each data packet. For the first The local timestamp of each received data packet; S112, during a fixed monitoring cycle Within the node, the number of data packets sent and received by each node is counted. S113, calculate the packet loss rate by comparing the expected data packets to be received with the actual data packets received. ; S114 defines the volatility of data transmission based on the degree of variation in round-trip delay of multiple data packets. This reflects the stability of communication.

[0008] Optionally, S12 includes: S121, through a time synchronization mechanism, local timestamps are added to the same data packet at both the sending and receiving ends, the one-way transmission delay of data packets between nodes is calculated, and the average transmission delay of multiple data packets is calculated to obtain the average delay within the current monitoring window. ; S122 estimates the current node's communication bandwidth by monitoring the number of data bytes successfully transmitted per unit time. ; S123, the network load is reflected by the sum of the total traffic received and sent by the computing node per unit time. ; S124, the average latency of each node Communication bandwidth Network load Perform unified timestamp alignment and map node IDs to their corresponding spatial locations in the topology. Define the spacetime state vector as ; S125, the state vectors of all nodes at different time points. Aggregation, organized chronologically and according to network topology, constructs a global spatiotemporal dynamic state map. .

[0009] Optionally, S2 includes: S21. Based on the constructed spatiotemporal dynamic state map, extract the state evolution sequence of each communication node in a continuous time period. Use the temporal graph convolutional network (T-GCN) model as the spatiotemporal dependency model between nodes in the communication network. By identifying the change trend of node state in the time dimension and the propagation path in the spatial dimension, detect evolution patterns similar to historical packet loss events and mark them as potential abnormal areas. S22, aligns and analyzes the real-time status information and historical status information of potential abnormal areas to identify key nodes and time periods with packet loss precursor characteristics. Based on the identification results, a dynamic risk map reflecting high-risk areas of packet loss failure is generated to indicate the time and location of possible future failures.

[0010] Optionally, S21 includes: S211, extract each communication node in the sliding time window from the constructed spatiotemporal dynamic state map in chronological order. The multidimensional state information sequence within constitutes the state evolution sequence. ; S212, Based on the topology of the communication network, construct the spatial adjacency matrix between nodes. This is used to reflect the connection relationships and influence weights between nodes; S213, using state evolution sequences and adjacency matrix Construct a Temporal Graph Convolutional Network (T-GCN) model; S214, based on the constructed Temporal Graph Convolutional Network (T-GCN) model, predicts the state change trend of each node at future time steps and calculates the error residual between the predicted value and the actual observed value. The error residuals are then matched with a database of historical packet loss events to calculate temporal similarity. ,like Then the current node is considered Nodes exhibiting a trend highly similar to historical packet loss evolution patterns are marked as potential anomalies. The similarity threshold; S215, the spatial location of the identified potential abnormal nodes at the current moment. and timestamp These are marked as potential anomalous areas.

[0011] Optionally, S22 includes: S221, Align the real-time status information (node ​​latency, bandwidth, traffic load) of the current potentially abnormal region with the historical status information, assuming the current timestamp is... Historical status information uses time windows Align; S222: Based on the historical aligned state information, a regression model is used to calculate the regression prediction value of the historical aligned state information. Combined with the real-time collected state information, the prediction residual of each node in different time periods is calculated. ; S223, based on the calculated prediction residuals of each node at different time periods. Build a dynamic risk map When the prediction residual of the node Exceeding the residual threshold At that time, the node was considered to be in a potentially high-risk state.

[0012] Optionally, S3 includes: S31 integrates the dynamic risk map with the status characteristics (latency, bandwidth, traffic load and residual value) of each node for analysis, calculates the fault risk score of each node, and filters out a set of suspected faulty nodes. S32 performs spatial and temporal consistency verification on the set of suspected faulty nodes, determines whether it matches the fault propagation path and high-risk area, identifies the source node of the packet loss fault, and outputs the location result.

[0013] Optionally, S31 includes: S311, based on the dynamic risk map Mark the set of nodes that are currently within the potential risk area. For each node Extract its current state features This includes current latency, bandwidth, traffic load, and prediction residuals; S312, Based on the extracted state features, a weighted linear combination model is used to construct a node failure risk scoring function and calculate the failure risk score. ; S313, Set the fault risk scoring threshold Nodes whose fault risk scores exceed the fault risk score threshold are selected to form a set of suspected fault nodes. .

[0014] Optionally, S32 includes: S321, in the set of suspected faulty nodes In the process, the geospatial location of each node is extracted. and the timestamp when its fault risk score exceeds the fault risk score threshold Constructing spatiotemporal propagation trajectories ; S322, using the linear least squares method to determine the spatiotemporal propagation trajectory Propagation is fitted in both spatial and temporal directions by fitting the propagation direction vector. Calculate the residual from each suspected fault node to the starting point of the propagation path, determine whether there is a consistent propagation path, and deduce the fault source node. S323, Set propagation residual threshold The node with the smallest residual from the propagation path and the earliest time was selected as the source node of the packet loss fault. .

[0015] The beneficial effects of this invention are: This invention, by constructing a spatiotemporal dynamic state map of network nodes that integrates communication state characteristics, can comprehensively depict the state change characteristics of each node in the communication network of a smart substation at different times, enhance the multi-dimensional perception of node transmission behavior, provide a solid foundation for subsequent packet loss fault risk modeling and propagation trend identification, and improve the ability to accurately extract communication anomalies.

[0016] This invention introduces a temporal graph convolutional network to model the node state evolution sequence and combines it with dynamic time warping distance matching of historical packet loss evolution patterns. This enables the prediction of the spatiotemporal evolution trend of packet loss faults and the construction of a dynamic risk map to identify high-risk areas and potential abnormal nodes in advance, thereby improving the early warning capability and dynamic tracking accuracy of communication faults.

[0017] This invention integrates multidimensional state features with a dynamic risk map to construct a node failure risk scoring model. By combining linear least squares fitting and propagation consistency verification methods, it can effectively identify the direction of fault propagation and locate the source node, significantly improving the accuracy and robustness of fault location and reducing the probability of false alarms and false negatives. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the positioning method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the high-risk area prediction process for packet loss failure according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] like Figures 1-2 As shown, a method for locating packet loss faults in the communication network of a smart substation includes the following steps: S1 collects real-time data packet sending and receiving information from multiple nodes in the intelligent substation communication network, and obtains the status information of each node, including latency, bandwidth, and traffic load, to construct a spatiotemporal dynamic status map of the network nodes. S2, based on the spatiotemporal dynamic state map, models the spatiotemporal relationship between each node, predicts the spatiotemporal region where packet loss faults occur, and generates a dynamic risk map of packet loss by combining historical state information. S3 inputs the dynamic risk map into the fault location algorithm. Through multi-dimensional information fusion, it integrates the state characteristics of each node and the predicted fault occurrence area to locate the source of packet loss fault and output the location result.

[0022] S1 includes: S11 collects real-time data packet sending and receiving information of each communication node in the smart substation through the communication monitoring system, including the data packet transmission timestamp, sending frequency, receiving frequency, packet loss situation and data transmission stability of each node; S12: Obtain the status information of each node, including the node's latency, bandwidth, and traffic load, and perform spatiotemporal correlation processing on the status information to construct a spatiotemporal dynamic status map of the network nodes.

[0023] S11 includes: S111, through data monitoring units deployed on core switching equipment or border routing nodes of the communication network, monitors the incoming and outgoing data flows of each communication node, logs the sending and receiving operations of each data packet, and adds a local timestamp to each record. and ,in, For the first The local sending timestamp of each data packet. For the first The local timestamp of each received data packet; S112, during a fixed monitoring cycle Within this, the number of data packets sent and received by each node is counted, and represented as follows: ; ; in, For sending frequency, For receiving frequency, To monitor the number of data packets sent by this node within a monitoring period, To monitor the number of data packets received by the node within the monitoring period; S113, calculate the packet loss rate by comparing the expected data packets to be received with the actual data packets received. Let the number of data packets a node should receive within a certain monitoring period be... The actual received is The packet loss rate is then expressed as: ; S114 defines the volatility of data transmission based on the degree of variation in round-trip delay of multiple data packets. To reflect communication stability, it is represented as: ; in, For the first The round-trip delay of a single data packet. For all Average latency of data packets This represents the total number of data packets.

[0024] S12 includes: S121, through a time synchronization mechanism, local timestamps are added to the same data packet at both the sending and receiving ends, the one-way transmission delay of data packets between nodes is calculated, and the average transmission delay of multiple data packets is calculated to obtain the average delay within the current monitoring window. , is represented as: ; in, For the first The transmission delay of each data packet, This is the timestamp of the data packet's transmission. This is the timestamp of when the data packet was received; S122 estimates the current node's communication bandwidth by monitoring the number of data bytes successfully transmitted per unit time. , is represented as: ; in, The effective bandwidth of the node during the monitoring period. For the first The size of a successfully transmitted data packet. To monitor the number of successfully transmitted data packets within a monitoring period, For monitoring time windows; S123, the network load is reflected by the sum of the total traffic received and sent by the computing node per unit time. , is represented as: ; in, For the node's traffic load, The amount of data sent per unit of time. The amount of data received per unit of time. For load statistics time window; S124, the average latency of each node Communication bandwidth Network load Perform unified timestamp alignment and map node IDs to their corresponding spatial locations in the topology. Define the spacetime state vector as , is represented as: ; in, This is the current timestamp; S125, the state vectors of all nodes at different time points. Aggregation, organized chronologically and according to network topology, constructs a global spatiotemporal dynamic state map. , is represented as: ; in, The total number of nodes in the network. For the first Each node in time The state vector on, .

[0025] S2 includes: S21. Based on the constructed spatiotemporal dynamic state map, extract the state evolution sequence of each communication node in a continuous time period. Use the temporal graph convolutional network (T-GCN) model as the spatiotemporal dependency model between nodes in the communication network. By identifying the change trend of node state in the time dimension and the propagation path in the spatial dimension, detect evolution patterns similar to historical packet loss events and mark them as potential abnormal areas. S22, aligns and analyzes the real-time status information and historical status information of potential abnormal areas to identify key nodes and time periods with packet loss precursor characteristics. Based on the identification results, a dynamic risk map reflecting high-risk areas of packet loss failure is generated to indicate the time and location of possible future failures.

[0026] S21 includes: S211, extract each communication node in the sliding time window from the constructed spatiotemporal dynamic state map in chronological order. The multidimensional state information sequence within constitutes the state evolution sequence. , is represented as: ; in, They are nodes In time The state vector (latency, bandwidth, traffic load); S212, Based on the topology of the communication network, construct the spatial adjacency matrix between nodes. This is used to reflect the connection relationships and influence weights between nodes, and is expressed as: ; in, For nodes and The strength of the connection between them Weights are defined based on link signal quality or communication frequency; S213, using state evolution sequences and adjacency matrix The Temporal Graph Convolutional Network (T-GCN) model is constructed as follows: ; in, For the first Layered graph convolution output, To add a self-loop adjacency matrix, This is the corresponding degree matrix. , For nodes The diagonal elements of the degree matrix, These are the elements of the adjacency matrix. For the first Layered graph convolution output, It is the ReLU activation function. For the first Trainable weights of the layer; S214, based on the constructed Temporal Graph Convolutional Network (T-GCN) model, predicts the state change trend of each node at future time steps and calculates the error residual between the predicted value and the actual observed value. The error residuals are then matched with a database of historical packet loss events to calculate temporal similarity. ,like Then the current node is considered Nodes exhibiting a trend highly similar to historical packet loss evolution patterns are marked as potential anomalies. The similarity threshold is expressed as: ; in, To use the temporal graph convolutional network model for the first time... The node state vector output by the layer represents the first... Each node in time The predicted state, This is the actual observed node state vector; ; in, The dynamic time-normalized distance represents the current node. error residual The first in the historical packet loss event database A template for packet loss evolution Temporal similarity between them For DTW dynamic alignment path, For the first Error residuals at each time point For the first in the historical template The residuals at each point in time; Similarity threshold Based on the adaptive learning method, it is expressed as: ; in, For the first The similarity threshold after the next iteration. In the first The similarity threshold at the next iteration For learning rate, The error residual of the current node Average DTW distance to all patterns in the historical packet loss evolution pattern library The dynamic time-normalized distance between them The average DTW distance for all packet loss evolution patterns in the historical packet loss evolution pattern library; S215, the spatial location of the identified potential abnormal nodes at the current moment. and timestamp These are marked as potential anomalous areas.

[0027] S22 includes: S221, Align the real-time status information (node ​​latency, bandwidth, traffic load) of the current potentially abnormal region with the historical status information, assuming the current timestamp is... Historical status information uses time windows Alignment is represented as: ; in, This provides real-time status information for the current potentially abnormal area. For historical state information, At the current time point Aligned status information; S222: Based on the historical aligned state information, a regression model is used to calculate the regression prediction value of the historical aligned state information. Combined with the real-time collected state information, the prediction residual of each node in different time periods is calculated. , is represented as: ; ; ; in, For nodes in front A sequence of concatenated aligned state vectors within each time step. For nodes In time The predicted state vector, This is the weight matrix of the regression model. It is the bias vector; S223, based on the calculated prediction residuals of each node at different time periods. Build a dynamic risk map When the prediction residual of the node Exceeding the residual threshold When a node is considered to be in a potentially high-risk state, it is represented as: ; in, For indicator functions, if the current node Predicted residuals Exceeding the residual threshold The system returns 1, indicating that the node is a high-risk node. For nodes The spatial location, including the geographic coordinates or network topology location of the node; Residual threshold The settings are based on real-time dynamic updates of the sliding window, and are represented as follows: ; in, In time The residual threshold is dynamically updated in real time. For sliding windows The mean of the internal residuals, For sliding windows Standard deviation of internal residuals This is the deviation adjustment coefficient. This represents the length of the sliding window.

[0028] S3 includes: S31 integrates the dynamic risk map with the status characteristics (latency, bandwidth, traffic load and residual value) of each node for analysis, calculates the fault risk score of each node, and filters out a set of suspected faulty nodes. S32 performs spatial and temporal consistency verification on the set of suspected faulty nodes, determines whether it matches the fault propagation path and high-risk area, identifies the source node of the packet loss fault, and outputs the location result.

[0029] S31 includes: S311, based on the dynamic risk map Mark the set of nodes that are currently within the potential risk area. For each node Extract its current state features This includes current latency, bandwidth, traffic load, and prediction residuals; S312, Based on the extracted state features, a weighted linear combination model is used to construct a node failure risk scoring function and calculate the failure risk score. , is represented as: ; in, For nodes The current latency, For nodes bandwidth, For nodes network traffic load, For nodes The predicted residuals For normalization function, , , , These are the weighting coefficients corresponding to each feature dimension in the risk score; S313, Set the fault risk scoring threshold Nodes whose fault risk scores exceed the fault risk score threshold are selected to form a set of suspected fault nodes. , is represented as: ; Fault risk scoring threshold The setting is expressed as: ; in, For the current moment Dynamically updated fault scoring thresholds In the sliding window Internal score mean In the sliding window Internal rating standard deviation This is the deviation control coefficient. This represents the length of the sliding window.

[0030] S32 includes: S321, in the set of suspected faulty nodes In the process, the geospatial location of each node is extracted. and the timestamp when its fault risk score exceeds the fault risk score threshold Constructing spatiotemporal propagation trajectories , is represented as: ; S322, using the linear least squares method to determine the spatiotemporal propagation trajectory Propagation is fitted in both spatial and temporal directions by fitting the propagation direction vector. Calculate the residual from each suspected fault node to the starting point of the propagation path, determine whether a consistent propagation path exists, and deduce the fault source node, represented as: ; in, For nodes The spatial residuals of the fitted propagation trajectory The location of the initial risk propagation source node. High-risk timestamps for initial propagation nodes; ; in, The spatial location of the node relative to the initial high-risk node The matrix composed of offset vectors, A column vector consisting of the time difference between the high failure score timestamp of each high-risk node and the initial node timestamp; S323, Set propagation residual threshold The node with the smallest residual from the propagation path and the earliest time was selected as the source node of the packet loss fault. , is represented as: ; Propagation residual threshold Represented as: ; in, For the current moment The propagation residual threshold, The propagation residual threshold from the previous time step. As a smoothing factor, This represents the average propagation residual of high-risk nodes that align with the current propagation trend.

[0031] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0032] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for locating packet loss faults in a smart substation communication network, characterized in that, Includes the following steps: S1 collects real-time data packet sending and receiving information from multiple nodes in the intelligent substation communication network, and obtains the status information of each node, including latency, bandwidth, and traffic load, to construct a spatiotemporal dynamic status map of the network nodes. S2, Based on the spatiotemporal dynamic state map, model the spatiotemporal relationship between each node, predict the spatiotemporal region where packet loss faults occur, and generate a dynamic risk map of packet loss by combining historical state information. S3. Input the dynamic risk map into the fault location algorithm. Through multi-dimensional information fusion, combine the state characteristics of each node and the predicted fault occurrence area to locate the source of packet loss fault and output the location result.

2. The method for locating packet loss faults in a smart substation communication network according to claim 1, characterized in that, S1 includes: S11 collects real-time data packet sending and receiving information of each communication node in the smart substation through the communication monitoring system, including the data packet transmission timestamp, sending frequency, receiving frequency, packet loss situation and data transmission stability of each node; S12: Obtain the status information of each node, including the node's latency, bandwidth, and traffic load, and perform spatiotemporal correlation processing on the status information to construct a spatiotemporal dynamic status map of the network nodes.

3. The method for locating packet loss faults in a smart substation communication network according to claim 2, characterized in that, S11 includes: S111, through data monitoring units deployed on core switching equipment or border routing nodes of the communication network, monitors the incoming and outgoing data flows of each communication node, logs the sending and receiving operations of each data packet, and adds a local timestamp to each record. and ,in, For the first The local sending timestamp of each data packet. For the first The local timestamp of each received data packet; S112, during a fixed monitoring cycle Within the node, the number of data packets sent and received by each node is counted. S113, calculate the packet loss rate by comparing the expected data packets to be received with the actual data packets received. ; S114 defines the volatility of data transmission based on the degree of variation in round-trip delay of multiple data packets. This reflects the stability of communication.

4. The method for locating packet loss faults in a smart substation communication network according to claim 3, characterized in that, S12 includes: S121, through a time synchronization mechanism, local timestamps are added to the same data packet at both the sending and receiving ends, the one-way transmission delay of data packets between nodes is calculated, and the average transmission delay of multiple data packets is calculated to obtain the average delay within the current monitoring window. ; S122 estimates the current node's communication bandwidth by monitoring the number of data bytes successfully transmitted per unit time. ; S123, the network load is reflected by the sum of the total traffic received and sent by the computing node per unit time. ; S124, the average latency of each node Communication bandwidth Network load Perform unified timestamp alignment and map node IDs to their corresponding spatial locations in the topology. Define the spacetime state vector as ; S125, the state vectors of all nodes at different time points. Aggregation, organized chronologically and according to network topology, constructs a global spatiotemporal dynamic state map. .

5. The method for locating packet loss faults in a smart substation communication network according to claim 4, characterized in that, S2 includes: S21. Based on the constructed spatiotemporal dynamic state map, extract the state evolution sequence of each communication node in a continuous time period. Use the temporal graph convolutional network model as the spatiotemporal dependency model between nodes in the communication network. By identifying the change trend of node state in the time dimension and the propagation path in the spatial dimension, detect evolution patterns similar to historical packet loss events and mark them as potential abnormal areas. S22 aligns and analyzes the real-time status information and historical status information of potential abnormal areas to identify key nodes and time periods with packet loss precursor characteristics. Based on the identification results, a dynamic risk map reflecting high-risk areas of packet loss failure is generated.

6. The method for locating packet loss faults in a smart substation communication network according to claim 5, characterized in that, S21 includes: S211, extract each communication node in the sliding time window from the constructed spatiotemporal dynamic state map in chronological order. The multidimensional state information sequence within constitutes the state evolution sequence. ; S212, Based on the topology of the communication network, construct the spatial adjacency matrix between nodes. This is used to reflect the connection relationships and influence weights between nodes; S213, using state evolution sequences and adjacency matrix Construct a temporal graph convolutional network model; S214, based on the constructed temporal graph convolutional network model, predicts the state change trend of each node at future time steps and calculates the error residual between the predicted value and the actual observed value. The error residuals are then matched with a database of historical packet loss events to calculate temporal similarity. ,like Then the current node is considered Nodes exhibiting a trend highly similar to historical packet loss evolution patterns are marked as potential anomalies. The similarity threshold; S215, the spatial location of the identified potential abnormal nodes at the current moment. and timestamp These are marked as potential anomalous areas.

7. The method for locating packet loss faults in a smart substation communication network according to claim 6, characterized in that, S22 includes: S221, Align the real-time status information of the current potential anomaly region with the historical status information, assuming the current timestamp is... Historical status information uses time windows Align; S222: Based on the historical aligned state information, a regression model is used to calculate the regression prediction value of the historical aligned state information. Combined with the real-time collected state information, the prediction residual of each node in different time periods is calculated. ; S223, based on the calculated prediction residuals of each node at different time periods. Build a dynamic risk map When the prediction residual of the node Exceeding the residual threshold At that time, the node was considered to be in a potentially high-risk state.

8. The method for locating packet loss faults in a smart substation communication network according to claim 7, characterized in that, S3 includes: S31, integrate the dynamic risk map with the status characteristics of each node for analysis, calculate the fault risk score of each node, and screen out a set of suspected fault nodes; S32 verifies the spatial and temporal consistency of the suspected fault node set, determines whether it matches the fault propagation path and high-risk area, identifies the source node of the packet loss fault, and outputs the location result.

9. The method for locating packet loss faults in a smart substation communication network according to claim 8, characterized in that, S31 includes: S311, based on the dynamic risk map Mark the set of nodes that are currently within the potential risk area. For each node Extract its current state features This includes current latency, bandwidth, traffic load, and prediction residuals; S312, Based on the extracted state features, a weighted linear combination model is used to construct a node failure risk scoring function and calculate the failure risk score. ; S313, Set the fault risk scoring threshold Nodes whose fault risk scores exceed the fault risk score threshold are selected to form a set of suspected fault nodes. .

10. A method for locating packet loss faults in a smart substation communication network according to claim 9, characterized in that, S32 includes: S321, in the set of suspected faulty nodes In the process, the geospatial location of each node is extracted. and the timestamp when its fault risk score exceeds the fault risk score threshold Constructing spatiotemporal propagation trajectories ; S322, using the linear least squares method to determine the spatiotemporal propagation trajectory Propagation is fitted in both spatial and temporal directions by fitting the propagation direction vector. Calculate the residual from each suspected fault node to the starting point of the propagation path, determine whether there is a consistent propagation path, and deduce the fault source node. S323, Set propagation residual threshold The node with the smallest residual from the propagation path and the earliest time was selected as the source node of the packet loss fault. .

Images