Electric power communication network fault intelligent processing method based on knowledge graph

Abstract

The invention requests to protect a power communication network fault intelligent processing method based on a knowledge graph, and the method comprises the steps: firstly employing ALBERT and XLNet to cooperatively extract deep semantic features of a professional network management alarm log and an equipment ledger, employing a heterogeneous graph attention network (HGAT) to carry out deep correlation mining among multi-dimensional data, and breaking a data barrier in the communication field; then, constructing a three-dimensional dynamic knowledge graph comprising a physical layer, a logic layer and a service layer, and quantitatively calculating the potential influence of the communication link fault on the power grid service based on the network topological relation; and finally, aiming at an N-X communication network fault scene, taking a communication equipment alarm state and a network topology connection state as a state space, and utilizing a Q-learning deep reinforcement learning algorithm to automatically generate an optimal service roundabout routing strategy. According to the method, a passive alarm monitoring mode can be converted into an active intelligent research and judgment mode, the fault handling period is remarkably shortened, and the efficiency bottleneck of traditional manual analysis is broken through.

Description

A knowledge graph-based intelligent fault handling method for power communication networks Technical Field

[0001] This invention relates to the fields of power communication operation and maintenance and artificial intelligence technology. Specifically, it relates to a method for cross-disciplinary data fusion, comprehensive fault assessment, and intelligent detour decision-making for power communication networks in NX scenarios, utilizing natural language processing, graph neural networks, and deep reinforcement learning technologies. Background Technology

[0002] With the accelerated construction of new power systems, the reliability requirements of power communication networks, as a supporting platform for the safe and stable operation of the power grid, are increasing. Currently, they mainly rely on the SDH / MSTP transmission system to carry critical services such as relay protection and stability control. However, existing communication management models and systems show obvious limitations when facing increasingly complex network environments and multi-dimensional data.

[0003] Specifically, communication operations and maintenance involve multiple independent systems such as transmission network management, online fiber optic cable monitoring, and environmental monitoring. This results in significant data silos between the unstructured alarm logs of professional network management, the structured data from fiber optic cable monitoring, and network topology data, making it difficult to uncover the implicit correlation between the physical state of fiber optic cables and the logical alarms of transmission equipment. In actual operation, fault diagnosis heavily relies on the manual experience of dispatchers, requiring cross-system comparison of a large amount of alarm information for tracing the source, which is not only inefficient but also prone to misjudgment.

[0004] In existing technologies (such as the patent application with application number CN120567660A), some studies have attempted to use graph neural networks (GNNs) to model fault text and alarm data in power communication networks to achieve automatic fault location. However, such conventional techniques still face the following significant technical bottlenecks when dealing with the complex needs of new power systems:

[0005] First, there is a lack of cross-layer, three-dimensional modeling capabilities. Most existing technologies are based on flat knowledge graphs or single-layer network topology analysis, which makes it difficult to capture the complex cross-layer transmission relationship between "optical cable physical layer interruption" and "business logic layer channel alarm", resulting in limited root cause location accuracy when facing concurrent optical cable and equipment failures.

[0006] Second, there is a lack of closed-loop handling capabilities encompassing "perception and decision-making." Most existing technologies stop at "fault detection" and "fault location," meaning they can only output the fault point but cannot provide subsequent handling strategies. When an NX (e.g., N-1 or N-2) fault occurs in the network, causing the primary route to be interrupted, existing technologies cannot automatically generate service rerouting and recovery strategies based on real-time changes in the network's topology. Schedulers still need to manually consult maps to formulate rerouting plans, which is insufficient to meet the urgent needs for minute-level fault response and high service reliability.

[0007] To address the shortcomings of the existing technologies, this invention proposes a knowledge graph-based intelligent fault handling method for power communication networks, which addresses these shortcomings through the following targeted approaches:

[0008] First, addressing the issue of "insufficient three-dimensional modeling capabilities": This invention constructs a three-dimensional dynamic knowledge graph that includes "physical layer - logical layer - business layer", and uses Heterogeneous Graph Attention Network (HGAT) to perform multimodal data fusion, clearly mapping the cross-level transmission path of faults and achieving accurate location of the root cause of faults.

[0009] Second, addressing the issue of "lack of closed-loop handling capabilities": This invention introduces deep reinforcement learning (Q-learning) technology to construct an intelligent detour decision-making model. This model, based on fault location, can further automatically calculate the optimal detour route according to the real-time status of the entire network, achieving a leap from "passive location" to "active self-healing," effectively solving the business continuity assurance problem in NX scenarios. Summary of the Invention

[0010] This invention aims to address the problems of insufficient data fusion, low analysis efficiency, and lack of automated decision support in the existing technologies. It proposes a knowledge graph-based intelligent fault handling method for power communication networks. The technical solution of this invention is as follows:

[0011] A knowledge graph-based intelligent fault handling method for power communication networks includes the following steps:

[0012] Step S1: Data Acquisition and Multidimensional Feature Fusion: Collect professional network management alarm logs, equipment ledgers, and optical cable monitoring data from the power communication network; use a hybrid algorithm model based on pre-trained language models and graph neural networks to process the above data, extract semantic features, and mine the implicit relationships between data to form a unified communication resource feature vector.

[0013] Step S2: Construction of a three-dimensional dynamic knowledge graph: Based on a unified communication resource feature vector, a three-dimensional dynamic knowledge graph containing physical, logical, and business layers is constructed using graph database technology; the three-dimensional dynamic knowledge graph maps communication resource entities and their topological dependencies.

[0014] Step S3: Fault Judgment and Impact Analysis: When a communication alarm is detected, the fault node is located in the three-dimensional dynamic knowledge graph, and correlation analysis is performed based on the topological relationship of the graph to identify the root cause of the fault and quantify the impact range of the fault on the reliability of communication services.

[0015] Step S4: Intelligent detour decision-making in the NX scenario: For the NX communication network failure scenario (i.e., the scenario in which N-1 single point failure or N-2 multiple failures occur in the power communication network, resulting in the interruption of service routing), construct an intelligent detour decision-making model based on deep reinforcement learning; take the current alarm status of communication equipment and the network topology connection status as the model input, automatically generate the optimal service detour routing strategy and output it.

[0016] Furthermore, in step S1, the hybrid algorithm model based on pre-trained language models and graph neural networks specifically includes:

[0017] We use ALBERT (Lightweight Bidirectional Transformer Encoder Representation Model) and XLNet (Generalized Autoregressive Pretrained Language Model) in collaboration to extract semantic features from unstructured text data in professional network management alarm logs and equipment ledgers.

[0018] The extracted semantic features are fused with the structured features of the optical cable monitoring data using a Heterogeneous Graph Attention Network (HGAT). HGAT is a graph neural network that can handle multiple types of nodes and edge relationships. It calculates the association weights between different modalities of data through an attention mechanism to generate a multi-dimensional fused node embedding vector.

[0019] Furthermore, in step S2, the specific hierarchical definition of the three-dimensional dynamic knowledge graph is as follows:

[0020] Physical layer: includes communication sites, equipment rooms, optical cable segments, optical path entities and their interconnections;

[0021] Logical layer: includes transmission devices, boards, ports, multiplex sections, logical links, and their topology connections;

[0022] The service layer includes the relay protection channels, stability control channels, and dispatch data network channels carried by the relays, as well as their service routing relationships.

[0023] The graph establishes a cross-layer mapping relationship where physical layer optical cable interruptions lead to logical layer link alarms, which in turn affect the availability of service layer channels.

[0024] Furthermore, the NX communication network failure scenario specifically refers to:

[0025] In power communication networks, if N-1 single device or link failures or N-2 multiple failures occur, the original primary route becomes unavailable, and it is necessary to find a backup route that meets the requirements of latency and reliability.

[0026] Furthermore, in step S3, the scope of the quantification of the impact of faults on the reliability of communication services specifically includes:

[0027] The betweenness centrality and degree centrality of faulty nodes in a three-dimensional dynamic knowledge graph are calculated using complex network theory.

[0028] Traverse all business layer entities that pass through the fault node in the graph, identify the list of interrupted services, and generate a fault impact assessment report based on the importance level of the services.

[0029] Furthermore, step S4 constructs an intelligent detour decision-making model based on deep reinforcement learning, which is essentially an automated decision-making agent with self-learning capabilities. Unlike traditional static routing rule tables, it learns, through continuous interaction (trial and error and feedback) with the communication network environment, which routing switching action will yield the greatest service reliability benefit under different fault combinations (States), thereby achieving adaptive decision-making in complex fault scenarios. Specifically, this includes:

[0030] State space definition: It is composed of the device alarm vectors of the entire communication network and the network topology adjacency matrix, and is used to characterize the fault distribution and connectivity of the network at the current moment;

[0031] Action space definition: Corresponds to the set of optional alternative routes for services affected by the failure. The selection of an action represents switching a specific service to the specified backup route;

[0032] Reward function design: The reward value is calculated based on the success rate of business recovery, the number of hops of the detour route, and the risk of the route; if the detour route is successfully established and the number of hops is the fewest, a positive reward is given.

[0033] A power communication network fault intelligent processing system implementing any one of the methods, comprising:

[0034] Multi-source data access module: used to connect to professional network management systems, optical cable monitoring systems and environmental monitoring systems to acquire multi-source heterogeneous data;

[0035] The knowledge graph construction and management module is used to store and manage 3D dynamic knowledge graphs and update the graph status based on real-time data.

[0036] Intelligent judgment and decision-making module: used to run hybrid algorithm model and Q-learning algorithm, and perform fault location and detour route calculation;

[0037] Human-computer interaction module: used to display fault assessment results, panoramic topology view and recommended detour routing strategies.

[0038] A computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the knowledge graph-based intelligent fault handling method for power communication networks as described in any one of the claims.

[0039] The advantages and beneficial effects of this invention are as follows:

[0040] This invention overcomes the technical bottlenecks of conventional single-modal analysis, achieving deep fusion of multi-source heterogeneous data. Existing conventional methods typically only process single network management alarm texts or only structured monitoring data, making it difficult to discover hidden fault correlations across modalities. This invention is not a simple algorithm superposition, but innovatively utilizes the HGAT heterogeneous graph attention mechanism to solve the problem of the difficulty in quantifying and correlated "unstructured text semantics" and "structured physical parameters" in the same vector space, thereby greatly improving the accuracy of complex fault assessment.

[0041] A three-dimensional "physical-logical-service" map was constructed, solving the challenge of cross-layer fault propagation analysis. Conventional techniques often analyze based on a flat network topology, ignoring the cascading impact of physical fiber optic cable interruptions on upper-layer services. The three-dimensional dynamic map constructed in this invention can clearly map the complete propagation chain of "physical fiber break - logical alarm - service interruption," which is a non-obvious system-level innovation in existing communication management systems, enabling precise location of fault root causes and quantitative assessment of service impact.

[0042] This represents a qualitative leap from "passive fault location" to "proactive self-healing." Existing technologies (such as CN120567660A) mostly stop at fault location. This invention, targeting complex NX scenarios, introduces a Q-learning reinforcement learning algorithm, transforming the traditional "manual map lookup + experience-based decision-making" model into an "AI-automatic calculation + strategy recommendation" model. This closed-loop processing mechanism, integrating "perception and decision-making," can automatically generate optimal business detour strategies, significantly compressing the fault handling cycle from hours to minutes, providing technical support that conventional methods cannot achieve to ensure the continuity of core power grid services. Attached Figure Description

[0043] Figure 1 is a general flowchart of the method provided in the embodiment of the present invention;

[0044] Figure 2 is a diagram of the multi-source heterogeneous data fusion algorithm model architecture in an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of the hierarchical structure of the three-dimensional dynamic knowledge graph in an embodiment of the present invention;

[0046] Figure 4 is a schematic diagram of the intelligent detour decision-making logic based on Q-learning in an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of the present invention.

[0048] The technical solution of the present invention to solve the above-mentioned technical problems is:

[0049] A knowledge graph-based intelligent fault handling method for power communication networks includes the following steps:

[0050] Step S1: Data Acquisition and Multidimensional Feature Fusion Based on Hybrid Model

[0051] As shown in Figure 1, this step aims to break down the "data silos" in communication operations and maintenance and establish a unified data foundation.

[0052] First, multi-source heterogeneous data is collected from the professional network management system (NMS), optical cable online monitoring system, and environmental monitoring system of the power communication network through northbound interfaces (such as Corba and SNMP) and database interfaces. The collected data includes: unstructured network management alarm logs (such as R_LOS, MS_AIS, B1 error codes, etc.), semi-structured equipment ledgers (manufacturer, model, board information), structured optical cable monitoring data (optical power, attenuation value, distance), and environmental data (equipment room temperature, power supply status).

[0053] Next, in order to understand the deep connections between these heterogeneous data, this invention constructs an ALBERT-XLNet-HGAT hybrid algorithm model for processing:

[0054] (1) Semantic feature extraction: The ALBERT (A Lite BERT) model is used to process massive alarm logs, and its parameter sharing mechanism is used to quickly extract short text keyword features such as "signal loss" and "frame synchronization failure"; at the same time, the XLNet model is used to process long texts in the device ledger and alarm descriptions to capture the logical dependencies between devices (such as "the failure of the light board at station A caused an alarm at station B"). The vectors extracted from both are concatenated to form a text semantic feature vector.

[0055] (2) Heterogeneous Association Mining: Multimodal fusion is performed using a Heterogeneous Graph Attention Network (HGAT). The aforementioned textual semantic features are used as node attributes, and the physical connections in the optical cable monitoring data are used as edge attributes. HGAT automatically learns the weights between different modalities through an attention mechanism. For example, when it detects that "a sudden increase in attenuation in a certain optical cable segment" and "bit error rate alarm" on the network management side are highly correlated in time and logical location, it assigns a very high weight to this association, thereby generating a unified communication resource feature vector that integrates physical status and logical alarms.

[0056] Step S2: Construct a three-dimensional dynamic knowledge graph of "physical-logical-business".

[0057] As shown in Figure 3, based on the feature vectors generated in step S1, a hierarchical and three-dimensional knowledge graph is constructed using a graph database (such as Neo4j) to accurately map a panoramic view of the communication network.

[0058] (1) Physical Layer: Nodes include entities such as communication sites, equipment rooms, optical cable segments, junction boxes, and optical distribution frames (ODF); edge relationships describe physical connections and geographical location relationships.

[0059] (2) Logical Layer: Nodes include transmission network elements (NE), slots, ports, multiplex sections, high / low order channels (VC4 / VC12), etc.; edge relationships describe logical bearer relationships and specify which segment of optical cable the multiplex section carries in the physical layer.

[0060] (3) Business Layer: Nodes include relay protection channels, stability control channels, dispatch data network channels, dispatch telephones, etc.; edge relationships describe business routes and clarify which time slot resources of the logic layer are occupied by the business channels.

[0061] The system will dynamically update the status attributes of nodes in the graph (such as "normal", "alarm", "interruption") based on the data collected in real time.

[0062] Step S3: Fault Assessment and Impact Analysis

[0063] When an anomaly occurs in the communication network, the system executes the following judgment logic in the three-dimensional dynamic knowledge graph:

[0064] (1) Fault location: Real-time alarm information is mapped to specific nodes in the graph, and the fault source is located by combining the topology relationship. For example, if a node of a certain optical cable segment in the physical layer turns red (interruption), the system immediately identifies that node as the root cause.

[0065] (2) Cross-layer correlation analysis: Using the graph traversal algorithm of the graph, trace upwards along the path of "physical layer → logical layer → service layer". For example: if the optical cable in the physical layer is interrupted, the related multiplex section and VC4 channel in the logical layer will become unavailable, which in turn will cause the "xx line main protection" service carried in the service layer to be interrupted.

[0066] (3) Impact Quantification: Based on the results of the graph analysis, the number and types of affected services are counted, and the impact range of the fault on the reliability of the communication service is quantified according to the importance level of the communication service (e.g., the priority of relay protection service is higher than that of administrative telephone), and a fault assessment report is generated.

[0067] Step S4: Intelligent Detour Decision-Making in NX Scenarios

[0068] For NX communication network failure scenarios (i.e., N-1 single point of failure or N-2 multiple failures in the network, resulting in the unavailability of the primary route), this invention constructs an intelligent detour decision model based on Q-learning deep reinforcement learning to automatically calculate the optimal detour route.

[0069] (1) State space:

[0070] Define the state space S = {M_alarm, G_topo}.

[0071] Where M_alarm is the real-time alarm status vector of the device collected by the current communication network management system (for example, 0 indicates normal and 1 indicates alarm); G_topo is the logical topology connectivity matrix of the current network (faulty links are marked as disconnected in the matrix and the weight is set to infinity).

[0072] (2) Action Space:

[0073] Defined as a set of route switching actions. In the SDH / MSTP network architecture, this involves switching affected services from a faulty route to a backup route. The backup route must meet the time slot continuity constraints of the SDH system.

[0074] (3) Reward function:

[0075] Establish a reward mechanism to optimize decision-making. The reward function is expressed as:

[0076] R = r1 × Success - r2 × Hops - r3 × Risk

[0077] in:

[0078] Success represents a business connectivity reward (a positive reward is given if the business successfully restores connectivity).

[0079] Hops represent the number of hops in the detour route (the fewer the hops, the lower the latency and the less points are deducted);

[0080] Risk represents the path risk value (if the detour route passes through a high-risk link, such as old equipment or degraded optical cable, the higher the Risk value, the more negative the penalty).

[0081] r1, r2, and r3 are the weighting coefficients of the three indicators mentioned above.

[0082] (4) Decision output:

[0083] The agent queries the trained Q-value table based on the current fault state S, selects the action with the largest Q-value as the optimal business detour routing strategy, and outputs it to the operation and maintenance scheduling personnel for confirmation or automatic execution.

[0084] In this embodiment: Figure 1 is a schematic diagram of the overall process of the method provided in this embodiment of the invention. Figure 2 is a schematic diagram of the multi-source heterogeneous data fusion algorithm architecture based on the ALBERT-XLNet-HGAT hybrid model in this embodiment of the invention. Figure 3 is a schematic diagram of the hierarchical structure of the "physical-logical-business" three-dimensional dynamic knowledge graph constructed in this embodiment of the invention. Figure 4 is a schematic diagram of the intelligent detour decision-making logic based on Q-learning deep reinforcement learning in this embodiment of the invention.

Claims

1. A knowledge graph-based intelligent fault handling method for power communication networks, characterized in that, The process includes the following steps: Step S1: Data Acquisition and Multidimensional Feature Fusion: Collect professional network management alarm logs, equipment ledgers, and optical cable monitoring data from the power communication network; process the above data using a hybrid algorithm model based on a pre-trained language model and a graph neural network, extract semantic features, and mine implicit relationships between data to form a unified communication resource feature vector; Step S2: Three-Dimensional Dynamic Knowledge Graph Construction: Based on the unified communication resource feature vector, construct a three-dimensional dynamic knowledge graph containing physical, logical, and business layers using graph database technology; the three-dimensional dynamic knowledge graph maps communication resource entities and their topological dependencies; Step S3: Fault Judgment and Impact Analysis: When a communication alarm is detected, the fault node is located in the three-dimensional dynamic knowledge graph, and correlation analysis is performed based on the topological relationship of the graph to identify the root cause of the fault and quantify the impact range of the fault on the reliability of communication services; Step S4: Intelligent Detour Decision in NX Scenarios: For NX communication network fault scenarios, i.e., scenarios in which N-1 single-point faults or N-2 multiple faults occur in the power communication network, resulting in service routing interruption, an intelligent detour decision model based on deep reinforcement learning is constructed; taking the current alarm status of communication equipment and the network topology connection status as the model input, the optimal service detour routing strategy is automatically generated and output.

2. The intelligent fault handling method for power communication networks based on knowledge graphs according to claim 1, characterized in that, In step S1, the hybrid algorithm model based on pre-trained language model and graph neural network specifically includes: using the ALBERT lightweight bidirectional Transformer encoder representation model and the XLNet generalized autoregressive pre-trained language model to extract semantic features from unstructured text data in professional network management alarm logs and equipment ledgers; using the heterogeneous graph attention network HGAT to fuse the extracted semantic features with the structured features of optical cable monitoring data; HGAT is a graph neural network that can handle multiple types of nodes and edge relationships. It calculates the association weights between different modal data through the attention mechanism to generate multi-dimensional fused node embedding vectors.

3. The intelligent fault handling method for power communication networks based on knowledge graphs according to claim 1, characterized in that, In step S2, the specific hierarchical definitions of the three-dimensional dynamic knowledge graph are as follows: Physical layer: includes communication sites, equipment rooms, optical cable segments, optical path entities and their connection relationships; Logical layer: includes transmission equipment, boards, ports, multiplex segments, logical links and their topology connection relationships. The service layer includes the relay protection channel, stability control channel, and dispatch data network channel, as well as their service routing relationships. The diagram establishes a cross-layer mapping relationship where physical layer optical cable interruption leads to logical layer link alarms, which in turn affects the availability of service layer channels.

4. The intelligent fault handling method for power communication networks based on knowledge graphs according to claim 1, characterized in that, Step S4 constructs an intelligent detour decision-making model based on deep reinforcement learning. Essentially, it is an automated decision-making agent with self-learning capabilities. Unlike traditional static routing rule tables, it learns, through continuous interaction with the communication network environment (trial and error, feedback), which routing action (Action) will yield the greatest service reliability reward under different fault combinations (State). This enables adaptive decision-making in complex fault scenarios. The specific construction process includes: State space definition: composed of the device alarm vectors and network topology adjacency matrix of the entire communication network, used to represent the fault distribution and connectivity of the network at the current moment; Action space definition: corresponding to the set of optional detour routes for services affected by faults, the selection of an action represents switching a specific service to a designated backup route; Reward function design: the reward value is calculated based on the success rate of service recovery, the number of hops of the detour route, and the risk of the route; if the detour route is successfully established with the fewest hops, a positive reward is given.

5. The intelligent fault handling method for power communication networks based on knowledge graphs according to claim 4, characterized in that, The NX communication network failure scenario specifically refers to the scenario in which an N-1 single device or link failure or an N-2 multiple failure occurs in the power communication network, resulting in the unavailability of the original primary route and the need to find a backup route that meets the requirements of latency and reliability.

6. The intelligent fault handling method for power communication networks based on knowledge graphs according to claim 1, characterized in that, In step S3, quantifying the impact of faults on the reliability of communication services specifically includes: using complex network theory to calculate the betweenness centrality and degree centrality of the fault node in the three-dimensional dynamic knowledge graph; traversing all service layer entities that pass through the fault node in the graph, identifying the list of interrupted services, and generating a fault impact assessment report based on the importance level of the services.

7. A fault intelligent processing system for implementing the method of any one of claims 1 to 6, characterized in that, include: Multi-source data access module: used to connect to professional network management systems, optical cable monitoring systems and environmental monitoring systems to acquire multi-source heterogeneous data; The knowledge graph construction and management module is used to store and manage the 3D dynamic knowledge graph and update the graph status according to real-time data; the intelligent judgment and decision-making module is used to run the hybrid algorithm model and Q-learning algorithm to perform fault location and detour routing calculation; the human-computer interaction module is used to display fault judgment results, panoramic topology view and recommended detour routing strategies.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the knowledge graph-based intelligent fault handling method for power communication networks as described in any one of claims 1 to 6.

Images