A Transmission Line Reliability Prediction Method Based on Multimodal Cognitive Network

Abstract

The present invention discloses a method for predicting the reliability of power transmission lines based on a multimodal cognitive network. The method first obtains multi-source heterogeneous data including equipment records, operation and maintenance logs, historical faults, meteorological and geographic information, extracts equipment-type, fault-type and environmental-type entities and their attributes, and constructs a unified three-layer meta-model through semantic similarity fusion. Subsequently, the association and strength between various entities are identified to construct a multimodal cognitive network. On this basis, the real-time status and degradation information of the equipment are extracted to establish a dynamic failure rate prediction model driven by health indicators; further, environmental factors are integrated to construct a line-level overall reliability prediction model to achieve a comprehensive assessment of the basic failure risk and environmental risk of the power transmission line. This method has the advantages of high data fusion, strong modeling accuracy and fine prediction granularity, and provides effective support for the precise operation and maintenance and risk warning of power grid equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent operation and maintenance of power transmission lines, and specifically relates to a power transmission line reliability prediction method based on a multimodal cognitive network. Background Art

[0002] As a core component of the power system, transmission lines are mainly composed of poles, towers, conductors, insulators, crossarms, drain wires, hardware and other components. They are characterized by a wide distribution range, a large number of equipment, and a complex operating environment. According to statistics, about 75% of unplanned power outages in the power grid are caused by transmission line faults, and the causes of the faults are mostly the accumulation of small defects and the synergistic effect of external environmental risks. For example, the gradual accumulation of dirt on the surface of insulators may cause flashover, and the coupling effect of tower corrosion and strong wind loads can lead to structural instability. Sudden environmental events such as lightning strikes and icing will increase the risk of cascading failures. However, existing operation and maintenance technologies have the following bottlenecks:

[0003] The multi-source data generated during operations and maintenance, such as equipment records, inspection logs, and fault reports, is highly heterogeneous, with structured data, semi-structured data, and unstructured text coexisting and exhibiting significant differences in temporal and spatial scales. Traditional methods rely on manual experience to screen key information, making it difficult to analyze the correlation between equipment status, environmental risks, and historical faults. This results in low data value density and weak decision-making support capabilities.

[0004] Existing reliability assessments are often based on periodic offline testing or single sensor data, using static thresholds to determine equipment health. These assessments lack the ability to dynamically model equipment degradation trends, and are even more incapable of quantifying the time-varying impact of the external environment on failure rates. Grid transformation and maintenance strategies typically use entire lines as units, relying on qualitative judgments from manual inspections. This can easily lead to two types of problems: over-maintenance, where healthy equipment is unnecessarily replaced, resulting in wasted resources; and delayed maintenance, where hidden defects or real-time risks of environmentally sensitive equipment are ignored, leading to escalated failures. Existing methods for quantifying environmental risks rely on historical statistics, without dynamically linking them to the real-time status and topology of the equipment. This results in coarse-grained risk warnings and poor timeliness.

[0005] In the prior art, Chinese patent CN117131783A discloses a transmission line risk prediction model, method, and system based on multimodal learning. The present invention proposes a method for constructing a transmission line risk prediction model based on multimodal learning. First, a lightning intensity analysis module, a distance warning module, and a transmission line risk analysis module are separately constructed. The lightning intensity analysis module predicts the lightning intensity at the next time point, the distance warning module predicts the lightning strike distance at the next time point, and the transmission line risk analysis module predicts the risk level of a lightning strike on the transmission line based on the lightning intensity at the next time point, the distance warning level, and the transmission line performance.

[0006] However, this technical solution still has the following shortcomings: First, although this method improves the prediction performance through multimodal features, the lightning intensity analysis is only based on time series modeling, which fails to fully consider the spatial distribution characteristics of lightning activities and the dynamic evolution of their propagation to neighboring areas. It lacks the characterization of potential spatial correlation risks between multiple lines, and it is difficult to meet the prediction needs of cross-regional lightning impacts in complex terrain areas. Secondly, the risk assessment module is implemented using a clustering method. Although it can achieve rapid classification, the clustering model is sensitive to data distribution and initial parameters, and has weak generalization capabilities. Especially when facing multi-source heterogeneous data or extreme environmental changes, it is prone to prediction deviations. In addition, each sub-module adopts a staged training strategy, and there is a lack of end-to-end collaborative optimization mechanism between modules, which leads to problems of accuracy loss and error accumulation in the overall risk prediction.

[0007] Therefore, how to build a reliability prediction model that integrates multi-source heterogeneous data and dynamically correlates equipment-environment-fault factors, achieve the transition from "local static assessment" to "global dynamic prediction", accurately identify weak links in the line and generate risk-adaptive operation and maintenance strategies has become a key problem in improving the reliability of power supply and the economy of operation and maintenance. Summary of the Invention

[0008] The purpose of the present invention is to solve the above problems and provide a transmission line reliability prediction method by introducing a multimodal cognitive network to construct a dynamic prediction model for transmission line operation reliability, calculate the overall reliability of the line, identify the weak links in line operation, and improve the digital operation and maintenance management level and reliability of the transmission line.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] The present invention provides a transmission line reliability prediction method based on a multimodal cognitive network, comprising the following steps:

[0011] Acquire multi-source heterogeneous data on transmission lines, including equipment inventory data, operation and maintenance logs, historical fault records, meteorological data, and geographic information data;

[0012] Extracting initial entities and attribute information of equipment, fault and environment classes based on the multi-source heterogeneous data;

[0013] Calculate and filter the similarity of the extracted entities, merge semantically similar entities, build a three-layer metamodel framework including equipment class, fault class, and environment class, and calculate the association strength between entities;

[0014] Based on the three-layer metamodel framework and the association strength between entities, a multimodal cognitive network is constructed;

[0015] Utilizing the multimodal cognitive network, extracting real-time status parameters and degradation information of the equipment, and constructing an equipment-level dynamic failure rate prediction model based on health indicators;

[0016] Integrate the equipment-level dynamic failure rate model with environmental factors to calculate the line-level basic failure risk and environmental risk, and build an overall reliability prediction model for transmission lines;

[0017] The reliability and risk of the transmission line are predicted based on the overall reliability prediction model of the transmission line.

[0018] Furthermore, the extraction of initial entities and attribute information of equipment, fault, and environment classes based on the multi-source heterogeneous data specifically includes:

[0019] Extract equipment entity information from equipment ledger data and build equipment entity collection ,in Indicates the Equipment type entities include wires, towers and other types of equipment. Each entity has attributes including equipment service time. , health status indicators , topology information , shape parameter β, scale parameter η, equipment normal operation start time τ 0 , where health status indicators Real-time detection values ​​of health indicators such as current, voltage, and temperature of each device ;

[0020] Extract fault entity information from historical fault records and operation and maintenance logs to build a fault entity set ,in Indicates the j Class faults, the fault class entities include lightning strikes, icing, equipment aging and other fault types, and the set attributes include historical fault frequency , device information related to the fault ;

[0021] Extract environmental entity information from meteorological data and geographic information data and construct an environmental entity set ,in Indicates the k Environmental factors, such as slope, wind speed, lightning intensity, land use type, etc., are included in the environmental entities. The set attributes include the real-time monitoring values ​​of each environmental factor. 、 t Number of environmental events that occurred during the period .

[0022] Furthermore, similarity calculation and screening are performed on the extracted entities, and entities with similar semantics are merged to construct a three-layer metamodel framework including equipment class, fault class and environment class, which specifically includes:

[0023] The FastText model is used to vectorize the text descriptions of equipment, fault, and environment entities to generate entity semantic vectors. ,d=300 is the entity semantic vector dimension, which is used to enhance the semantic representation ability;

[0024] The similarity between entity semantic vectors of entities in the same category is calculated based on the Jaccard similarity coefficient. The calculation formula of the Jaccard similarity coefficient is:

[0025]

[0026] Where: A 、 B Respectively represent the semantic vectors of the entities to be matched;

[0027] Setting similarity threshold , when Jaccard(A, B)≥ When , A and B are determined to be semantically similar entities and merged into a unified entity node;

[0028] The merged equipment class, fault class and environment class entities are structurally divided according to entity attributes and their categories, and a three-layer metamodel framework including equipment class, fault class and environment class is constructed.

[0029] Furthermore, the calculation of the association strength between entities specifically includes:

[0030] Based on the device, fault, and environment entity sets, the association strength between the three types of entities is calculated. The calculation formula for the association strength is:

[0031]

[0032] in, Represents a device entity , fault entity Environmental entities The strength of the correlation between the three Respectively represent device entities , fault entity Environmental entities The cumulative number of occurrences identified and recorded in multi-source heterogeneous data, Respectively represent device entities Fault Entity , equipment entities Environmental entities , Environmental Entities Fault Entity The number of times the same data appears together in multi-source heterogeneous data; Represents a device entity Fault Entity Environmental entities The number of times a term appears in the same data in multi-source heterogeneous data.

[0033] Furthermore, the multimodal cognitive network is constructed based on the three-layer metamodel framework and the association strength between entities, specifically including:

[0034] The equipment, fault and environment entities in the three-layer metamodel framework are mapped to nodes in the multimodal cognitive network, and the node attributes are entity attributes;

[0035] Based on the calculated association relationship and association strength between entities, edges between nodes are constructed, and the attribute of the edge is the association strength between the node entities;

[0036] Neo4j graph database is used for structured storage of multimodal cognitive networks, and the relationships between nodes and edges are organized in the form of a graph structure.

[0037] Furthermore, the multimodal cognitive network is used to extract the real-time status parameters and degradation information of the equipment and to construct a device-level dynamic failure rate prediction model based on health indicators, specifically including:

[0038] The improved Weibull-Markov chain algorithm is used to quantify the impact of equipment aging and external environmental factors on the failure rate, and the equipment aging correction coefficient and the external environment correction coefficient are obtained.

[0039] According to the real-time detection value of the device , using health status indicators Evaluate the health status of the equipment, health index The calculation formula is:

[0040]

[0041] in, For the i Indicators at time t The detection values ​​include real-time current, voltage, temperature and other health indicators of the device; For the i The weight of each health indicator is determined by the hierarchical analysis method; N is the number of real-time detection values, 、 is the threshold range of the real-time monitoring value;

[0042] Based on the equipment, fault and environment entities and their attribute information extracted from the multimodal cognitive network, a device-level time-varying failure rate prediction model is constructed by combining the equipment aging correction coefficient, external environment correction coefficient and health index.

[0043] Furthermore, the equipment aging correction coefficient formula is:

[0044]

[0045] in, For device entities The equipment aging correction factor at the current time t is, 、 Equipment entities The shape and scale parameters of is the current time, For device entities Normal operation start time;

[0046] The calculation formula of the external environment correction coefficient is:

[0047]

[0048] in, Represents a device entity External environment correction factor, Represents a device entity The number of occurrences of the associated k-th environmental entity, For all entities with devices The total number of occurrences of the associated environmental entity, is the total number of environmental entities considered, Represents environmental entities The impact weight on the equipment is determined based on expert knowledge.

[0049] Furthermore, the device-level time-varying failure rate prediction model is:

[0050]

[0051] in, is the device entity at the current time t The dynamic failure rate, is the baseline failure rate, i.e. the initial failure rate without degradation or environmental influences, is a set of fault entities, is a collection of environment entities, 、 Equipment entities With fault entity , equipment entity and environmental entities The strength of association, For in time t Internal fault entity The number of occurrences, For in time t The total number of occurrences of all fault entities in Environmental Entity Real-time monitoring value at time t.

[0052] Furthermore, the device-level dynamic failure rate model is integrated with environmental factors to calculate the line-level basic failure risk and environmental risk, and to construct an overall reliability prediction model for the transmission line, specifically including:

[0053] Based on the topological association between device entities in the multimodal cognitive network, the dynamic failure rate of each device entity is calculated according to the critical weight of the device in the topological structure. Aggregate into line-level basic fault risks , the calculation formula is:

[0054]

[0055] in, is the device entity at the current time t The dynamic failure rate, For device entities The critical weight in the line structure is determined based on the equipment redundancy and load distribution ratio parameters, m is the total number of equipment entities in the transmission line, is the line-level basic fault risk at the current time t;

[0056] Leverage the strength of associations between devices and environmental entities and the weight factors of each environmental factor , build a device-level environmental risk model:

[0057]

[0058] in, For device entities Equipment-level environmental risks, Represents environmental entities The weighting factor of the equipment is determined based on expert knowledge. Represents a device entity and environmental entities The strength of the association between Represents environmental entities The real-time monitoring value at the current time t, p is the total number of environmental entities;

[0059] Line-level infrastructure failure risk Equipment-level environmental risks The overall reliability prediction model of the transmission line is constructed by fusion, and its comprehensive risk expression is:

[0060]

[0061] in, For the moment t The overall comprehensive risk of the transmission line is represented by m, where m is the total number of equipment entities.

[0062] Furthermore, the weight factors of the environmental factors are Determined by entropy weight method, the formula is:

[0063]

[0064]

[0065] in, Environmental Entity The information entropy of Environmental Entity At the moment The real-time monitoring value, M is the number of samples, that is, the time segment or the total number of monitoring points used for statistical analysis, Environmental Entity The sum of all sample values ​​of .

[0066] Compared with the prior art, the present invention has the following advantages:

[0067] (1) The present invention introduces a multimodal cognitive network, integrates multi-source heterogeneous data such as fault records and equipment ledgers to construct a multimodal cognitive network for transmission lines, analyzes the failure principles and mutual correlations of various components, and then constructs a transmission line equipment failure rate prediction model and an operation reliability dynamic prediction model based on the multimodal cognitive network, conducts a comprehensive analysis of the whole, and provides support for the safe and stable operation of the power grid; it can be widely used in the daily operation and maintenance management of transmission lines.

[0068] (2) The present invention achieves unified modeling of multi-source heterogeneous data in power transmission lines by constructing a knowledge graph structure that integrates three layers of multimodal entities: equipment, fault, and environment. The graph is based on structured equipment ledger data, unstructured inspection text, and image information. It uses entity recognition and relationship extraction methods to establish semantic connections between entities, such as hierarchy, association, and causality, thereby breaking the traditional operation and maintenance data silo pattern and constructing a power transmission line cognitive network with semantic relevance and temporal logic. This design effectively solves the technical bottlenecks of traditional methods that make it difficult to integrate different data modalities and to analyze the association between equipment status and environmental risks, and provides a unified data foundation and semantic support for subsequent risk modeling.

[0069] (3) This invention proposes an improved Weibull-Markov chain combined modeling method to dynamically model the health status evolution process of key equipment in transmission lines. Compared with the traditional static threshold judgment method, this method combines the statistical distribution law of equipment life and the time series characteristics of state transition. It can reflect the gradual degradation process of equipment from healthy to failed in real time, and introduces environmental variables such as temperature, humidity, icing, and lightning strike frequency as regulating factors for state transition, thereby dynamically adjusting the fault probability estimate. This modeling technology effectively improves the early warning capability of potential equipment failures and avoids the problem of periodic detection mechanism missing hidden defects.

[0070] (4) The present invention constructs a comprehensive risk assessment method for power transmission lines based on a graph network topology structure, performs weighted fusion of equipment health status nodes, fault event nodes, and environmental risk nodes, and forms a risk propagation graph with both spatial and semantic connectivity attributes. During the line risk quantification process, the system comprehensively calculates the risk level of the entire line and the key fault propagation path based on the status value, connection strength, and impact path weight of each node. This technical solution effectively overcomes the limitations of traditional evaluation units based on equipment or lines and the lack of multi-factor fusion, improves the granularity and accuracy of risk assessment, and realizes a "point-line-surface" linkage intelligent early warning.

[0071] (5) By introducing a graph-driven fault deduction mechanism, the present invention can automatically trace the potential causes and related environmental factors after identifying abnormal equipment conditions, thereby achieving the automated construction of the fault causal chain. Combined with the dynamic risk assessment results, the system can predict the trend of risk events that may occur within a certain time window in the future and generate intervention recommendations. This mechanism avoids the lag and subjectivity of traditional reliance on manual review, and enhances the foresight and scientific nature of the operation and maintenance strategy.

[0072] (6) The present invention supports the system's self-learning and knowledge updating through a graph-level knowledge evolution mechanism. When new equipment types, new fault forms, or new environmental interference factors are introduced, the system can automatically update the graph structure and state assessment model through incremental learning, maintaining the timeliness and robustness of the knowledge base. This solves the existing system's "static rules - fixed models - delayed response" problem and improves its responsiveness and adaptability to emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 is a flow chart of the method of the present invention;

[0074] Figure 2 The following is the distribution diagram of lightning strikes on towers with different slopes;

[0075] Figure 3 The distribution diagram of lightning strikes on towers with different slopes;

[0076] Figure 4 The following is the distribution diagram of lightning strikes on towers of different heights;

[0077] Figure 5 The distribution map of lightning strikes on towers located on different types of land;

[0078] Figure 6 The figure shows the distribution of lightning strikes on towers with different river network densities. DETAILED DESCRIPTION

[0079] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts should fall within the scope of protection of the present invention.

[0080] Example 1:

[0081] This embodiment relates to a method for dynamic prediction of transmission line reliability based on a multimodal cognitive network. Figure 1 As shown, the method includes the following steps:

[0082] Step 1: Collect multi-source heterogeneous data of transmission lines and build a multimodal cognitive network for transmission lines.

[0083] Taking the actual data of a local power transmission network as an example, a multimodal cognitive network of power transmission lines was constructed. This multimodal cognitive network covers 122 lines, including 10,350 nodes and 15,678 relationships.

[0084] Acquire multi-source heterogeneous data on transmission lines, including equipment inventory data, operation and maintenance logs, historical fault records, meteorological data, and geographic information data;

[0085] Based on the multi-source heterogeneous data, initial entities and attribute information of equipment, fault, and environment are extracted, specifically including:

[0086] Extract equipment entity information from equipment ledger data and build equipment entity collection ,in Indicates the Equipment type entities include wires, towers and other types of equipment. Each entity has attributes including equipment service time. , health status indicators , topology information , shape parameter β, scale parameter η, equipment normal operation start time τ 0 , where health status indicators Real-time detection values ​​of health indicators such as current, voltage, and temperature of each device ;

[0087] Extract fault entity information from historical fault records and operation and maintenance logs to build a fault entity set ,in Indicates the j Class faults, the fault class entities include lightning strikes, icing, equipment aging and other fault types, and the set attributes include historical fault frequency , device information related to the fault ;

[0088] Extract environmental entity information from meteorological data and geographic information data and construct an environmental entity set ,in Indicates the k Environmental factors, such as slope, wind speed, lightning intensity, land use type, etc., are included in the environmental entities. The set attributes include the real-time monitoring values ​​of each environmental factor. 、 t Number of environmental events that occurred during the period .

[0089] Calculate and filter the similarity of the extracted entities, merge semantically similar entities, build a three-layer metamodel framework including equipment class, fault class, and environment class, and identify the association relationship and association strength between entities, including:

[0090] The FastText model is used to vectorize the text descriptions of equipment, fault, and environment entities to generate entity semantic vectors. ,d=300 is the entity semantic vector dimension, which is used to enhance the semantic representation ability;

[0091] The similarity between the semantic vectors of various entities is calculated based on the Jaccard similarity coefficient. The calculation formula of the Jaccard similarity coefficient is:

[0092]

[0093] Where: A 、 B Respectively represent the semantic vectors of the entities to be matched;

[0094] Setting similarity threshold , when Jaccard(A, B)≥ When , A and B are determined to be semantically similar entities and merged into a unified entity node;

[0095] The merged equipment class, fault class and environment class entities are structurally divided according to entity attributes and their categories, and a three-layer metamodel framework including equipment class, fault class and environment class is constructed.

[0096] Based on the three-layer metamodel framework and the association relationships and association strengths between entities, a multimodal cognitive network is constructed;

[0097] Based on the equipment, fault, and environment entity sets, the association relationships between entities are identified, and the association strength between the three types of entities is calculated. The calculation formula for the association strength is:

[0098]

[0099] in, Represents a device entity , fault entity Environmental entities The strength of the correlation between the three Respectively represent device entities , fault entity Environmental entities The cumulative number of occurrences identified and recorded in multi-source heterogeneous data, Respectively represent device entities Fault Entity , equipment entities Environmental entities , Environmental Entities Fault Entity The number of times the same data appears together in multi-source heterogeneous data; Represents a device entity Fault Entity Environmental entities The number of times a term appears in the same data in multi-source heterogeneous data.

[0100] The equipment, fault and environment entities in the three-layer metamodel framework are mapped to nodes in the multimodal cognitive network, and the node attributes are entity attributes;

[0101] Based on the calculated association relationship and association strength between entities, edges between nodes are constructed, and the attribute of the edge is the association strength between the node entities;

[0102] Neo4j graph database is used for structured storage of multimodal cognitive networks, and the relationships between nodes and edges are organized in the form of a graph structure.

[0103] Step one mainly solves the problems of difficulty in integrating multi-source heterogeneous data of transmission lines, inconsistent semantics, and insufficient association modeling. When faced with multi-source data such as equipment ledgers, fault records, operation and maintenance logs, meteorological information, and geographic data, traditional methods are often difficult to manage in a unified manner due to inconsistent data structures and diverse description methods, resulting in serious information fragmentation and redundancy, affecting the accuracy and efficiency of subsequent analysis and fault diagnosis. Through the method of the present invention, first, equipment-type, fault-type, and environment-type entities and their attribute information are systematically extracted from multi-source data, and the entity text semantics are vectorized in combination with the FastText model. The Jaccard similarity coefficient is used to perform similarity calculation and entity merging, thereby solving the redundancy and ambiguity problems caused by inconsistent semantic expressions and improving the accuracy of entity recognition and fusion. Furthermore, a method for calculating the association strength based on the co-occurrence frequency of entities is proposed to quantify the coupling relationship between the three types of entities: equipment, faults, and environment, effectively establish a multi-factor influence chain, and realize the association modeling and deep mining of complex fault scenarios. Ultimately, the three types of entities were mapped as nodes in a multimodal cognitive network, and the strong relationships between entities were constructed as graph edges, enabling structured and semantic knowledge representation. The entire network is efficiently stored and queried in a Neo4j graph database, offering excellent scalability and visualization capabilities. This effectively supports applications such as knowledge reasoning, anomaly identification, and risk warning in intelligent operations and maintenance systems, enhancing the system's cognitive capabilities and operational intelligence.

[0104] Step 2: Using the constructed multimodal cognitive network, a time-varying failure rate prediction model for power transmission equipment is constructed to obtain prediction results. The specific steps include:

[0105] The improved Weibull-Markov chain algorithm is used to quantify the dynamic impact weights of key factors such as equipment aging, lightning strikes, and strong winds. The dynamic weights are corrected and the calculation formula for the equipment aging correction coefficient is:

[0106]

[0107] in, For device entities The equipment aging correction factor at the current time t is, 、 Equipment entities The shape and scale parameters of is the current time, For device entities Normal operation start time;

[0108] The calculation formula of the external environment correction coefficient is:

[0109]

[0110] in, Represents a device entity External environment correction factor, Represents a device entity The number of occurrences of the associated k-th environmental entity, For all entities with devices The total number of occurrences of the associated environmental entity, is the total number of environmental entities considered, Represents environmental entities The impact weight on the equipment is determined based on expert knowledge.

[0111] The weights are shown in Table 1. They represent transmission lines, hardware, insulators, iron parts, towers, and guy wires respectively. They respectively represent the weights of factors affecting equipment aging, lightning strikes, strong winds, external foreign objects, tree obstacles and other faults.

[0112] Table 1 Weights of failure influencing factors

[0113]

[0114] (2) Extract equipment health indicators through multimodal cognitive networks, use hierarchical analysis to determine indicator weights, and calculate the health index, specifically:

[0115] The calculation formula for the equipment health index (HI) is:

[0116]

[0117] in, For the i Indicators at time t The detection values ​​include real-time current, voltage, temperature and other health indicators of the device; For the i The weight of each health indicator is determined by the hierarchical analysis method; N is the number of real-time detection values, 、 is the threshold range of the real-time monitoring value;

[0118] The predicted equipment failure rate based on failure influencing factors is compared with the historical average failure rate, as shown in Table 2.

[0119] Table 2 Comparison of failure rates of different equipment

[0120]

[0121] As can be seen from the table, the failure rates of some equipment vary greatly, such as transmission lines. This may be because lightning strikes and windy and rainy weather affect the real-time failure rate of the equipment, while the historical average failure rate cannot easily reflect this difference.

[0122] (3) Based on the historical equipment failure data and real-time status parameters in the multimodal cognitive network, combined with the dynamic weight correction coefficient and health index, a device-level time-varying failure rate prediction model is constructed:

[0123]

[0124] in, is the device entity at the current time t The dynamic failure rate, is the baseline failure rate, i.e. the initial failure rate without degradation or environmental influences, is a set of fault entities, is a collection of environment entities, 、 Equipment entities With fault entity , equipment entity and environmental entities The strength of association, For in time t Internal fault entity The number of occurrences, For in time t The total number of occurrences of all fault entities in Environmental Entity Real-time monitoring value at time t.

[0125] The multimodal cognitive network query data was input into the health index evaluation model to calculate the health index and real-time failure rate of each transmission equipment, as shown in Table 3. Compared with the failure rate and historical average failure rate of the fault factor correction model, the improved failure rate prediction model based on the health index is more able to reflect the differences between different transmission equipment.

[0126] Table 3. Real-time failure rate prediction model for power transmission equipment based on HI

[0127]

[0128] Step 2 primarily addresses the difficulty of traditional transmission equipment failure rate assessment models in effectively capturing the combined effects of equipment state changes, external environmental interference, and failure factors. Previous methods typically estimated equipment failure rates based on static parameters or empirical models, ignoring the time-varying nature of equipment operating conditions and the complex coupling relationships between multiple influencing factors. This resulted in limited prediction accuracy and made it difficult to meet the requirements for dynamic perception and accurate assessment of equipment health in actual transmission systems.

[0129] This invention, through the construction of a multimodal cognitive network, organizes three types of entities—equipment, faults, and environment—and their correlation strengths into a structured knowledge graph, providing a data foundation and semantic support for subsequent dynamic modeling. On this basis, an improved Weibull-Markov chain algorithm is used to model the equipment aging process. The aging correction coefficient is calculated using the equipment's shape and scale parameters to accurately quantify the impact of equipment operating time on its health status. Furthermore, an external environmental correction coefficient model is constructed by combining environmental factors extracted from multi-source heterogeneous data. Expert knowledge is introduced to set environmental weights, thereby quantitatively expressing the contribution of factors such as lightning strikes, strong winds, and tree obstructions to equipment risk.

[0130] By integrating equipment operating parameters such as real-time current, voltage, and temperature, a health index model was constructed. The analytic hierarchy process was introduced to assign appropriate weights to each indicator, making the health assessment process scientific and explainable. Furthermore, by integrating historical failure data, the health index, and dynamic correction factors, a device-level, time-varying failure rate prediction model was constructed. This model dynamically reflects the health level and potential failure risks of equipment at different time points.

[0131] Compared to traditional average models, this approach achieves more fine-grained, personalized predictions, accurately distinguishing transmission equipment of different types and states, effectively supporting the optimal allocation of O&M resources and the early triggering of risk warnings. Its significant advantage lies in its integration of static structural characteristics with dynamic operational information, comprehensively considering both historical and real-time factors, and fully demonstrating the timeliness, comprehensiveness, and intelligence of transmission equipment health assessments.

[0132] Step 3: Use the obtained multimodal cognitive network to predict the lightning strike risk of the transmission line and obtain the prediction results. The specific steps include:

[0133] (1) Through multimodal cognitive network analysis, slope, aspect, land use type, river network density and tower height were selected as the main influencing factors affecting the lightning strike of transmission lines.

[0134] (2) Determine the weight of the impact factor through the entropy weight method to accurately judge the impact of specific variables on the overall evaluation. The formula for determining the weight of the impact factor using the entropy weight method is:

[0135]

[0136]

[0137] in, Environmental Entity The information entropy of Environmental Entity At the moment The real-time monitoring value, M is the number of samples, that is, the time segment or the total number of monitoring points used for statistical analysis, Environmental Entity The sum of all sample values ​​of .

[0138] (3) Based on the association strength in multimodal cognitive networks , calculate the environmental risk, the calculation formula for environmental risk is:

[0139]

[0140] in, For device entities Equipment-level environmental risks, Represents environmental entities The weighting factor of the equipment is determined based on expert knowledge. Represents a device entity and environmental entities The strength of the association between Represents environmental entities The real-time monitoring value at the current time t, p is the total number of environmental entities;

[0141] (4) Based on the device topology association in the multimodal cognitive network, the real-time failure rate at the device level is aggregated into the basic failure risk at the line level. The basic failure risk at the line level is dynamically coupled with the environmental risk to generate an overall lightning risk warning model for the line.

[0142] The formula for aggregating the device-level real-time failure rate into the line-level basic failure risk is:

[0143]

[0144] in, is the device entity at the current time t The dynamic failure rate, For device entities The critical weight in the line structure is determined based on the equipment redundancy and load distribution ratio parameters, m is the total number of equipment entities in the transmission line, is the line-level basic fault risk at the current time t;

[0145] Dynamically couple line-level foundation fault risks with environmental risk factors to build a comprehensive lightning strike risk prediction model:

[0146]

[0147] in, For the moment t The overall comprehensive risk of the transmission line is represented by m, where m is the total number of equipment entities.

[0148] (5) Analyze the risk factors in the examples

[0149] Correlation analysis between lightning strike and tower slope Figure 2 As shown in the figure, as the slope increases, the proportion of lightning strikes also gradually increases. Therefore, it can be concluded that the slope of the tower is positively correlated with the probability of being struck by lightning.

[0150] Correlation analysis between lightning strike and tower slope Figure 3 As shown in the figure, a multimodal cognitive network was used to query slope data and calculate the proportion of towers with different slopes that were struck by lightning. Towers with a southeasterly slope had a significantly higher lightning strike rate, approximately 0.051%. Towers with easterly and southerly slopes were next highest, with a lightning trip rate of approximately 0.048%. Flatland had the lowest lightning trip rate, at approximately 0.019%. Summer is a high-incidence period for lightning-triggered power grid accidents. In summer, winds predominate in my country's southeastern coastal areas, mainly from the east and southeast. Therefore, mountainous and hilly areas facing due east or southeast are windward, which facilitates updrafts and produces severe convective weather, making lightning damage more likely.

[0151] Correlation analysis between lightning strike and tower height Figure 4 As shown in the figure, tower height data was queried through the equipment inventory network to statistically calculate the relationship between different tower heights and tower lightning strikes. As tower height increases, the probability of being struck by lightning also increases. This may be because the increased inductance of the tower body increases the lightning overvoltage, while the lightning protection measures of the transmission lines are limited.

[0152] Correlation analysis between lightning strike and land type Figure 5 As shown, the percentage of towers tripped due to lightning strikes was calculated for each land use type. Forested areas have the highest probability of being struck by lightning, followed by wetlands, farmland, and grassland. The remaining land types have a lower probability of being struck by lightning. Forested areas are often located at high altitudes and experience high humidity, making them more susceptible to lightning strikes.

[0153] Correlation analysis between lightning strike and river network density Figure 6 As shown in the figure, by calculating the river network density within the grid, we can generate a river network density distribution map for the target area. A higher river network density indicates a greater probability of lightning strikes on towers. On the one hand, a higher river network density indicates a larger proportion of river area within the region, resulting in better moisture conditions in the air and, therefore, a higher probability of lightning strikes. On the other hand, rivers have a much higher electrical conductivity than flat land, making them more susceptible to lightning strikes.

[0154] (6) Evaluate the risk level of the embodiment, obtain the transmission line lightning risk prediction results, and compare them with the actual data to verify the effectiveness of this solution.

[0155] According to the above analysis, the influencing factors are ranked according to the degree of correlation with the probability of lightning strike on the line, which are river network density, tower height, slope, slope direction and land type, and their numbers are defined as 、 、 、 、 , establish the lightning strike correlation matrix, as shown in Table 4.

[0156] Table 4 Correlation matrix of lightning impact factors

[0157]

[0158] Through calculation, the maximum eigenvalue of the correlation matrix of lightning impact factors is 5.05, and the consistency ratio is 0.019 < 0.10, which means that the consistency of the matrix is ​​reasonable. Its weight is determined by entropy weight theory.

[0159] Finally, the final evaluation results are shown in Table 5. As can be seen from the table, the probability of this transmission line suffering moderate lightning damage is the highest, at 0.37. The probabilities of suffering very high and very low lightning damage are relatively small, at 0.04 and 0.06, respectively, with an uncertainty of 0. Therefore, based on the overall evaluation results, the probability of this transmission line suffering lightning damage is moderate.

[0160] Table 5 Transmission line lightning damage risk assessment results

[0161]

[0162] Statistics were collected for all transmission lines in this embodiment and compared with actual data, as shown in Table 6. As can be seen from the table, the predicted results are generally consistent with the actual situation. The number of lightning damages decreases significantly with increasing lightning damage risk levels, and the number of lightning damages at different risk levels is also generally consistent. Therefore, the method proposed in this solution can effectively predict the probability of transmission lines being struck by lightning.

[0163] Table 6 Comparison of lightning damage prediction and actual data

[0164]

[0165] In summary, by applying the dynamic prediction method for transmission line reliability based on multimodal cognitive networks proposed in the present invention, the weak links of the lines can be accurately identified, and the efficiency of operation and maintenance decision-making and the reliability of power supply of the power grid can be improved.

[0166] Example 2:

[0167] This embodiment provides a transmission line reliability and risk prediction system based on a multimodal cognitive network, including:

[0168] The data acquisition module is used to obtain structured equipment ledger data, sensor monitoring data, as well as unstructured inspection images, text logs, and environmental monitoring data from transmission line-related equipment, enabling comprehensive collection of multi-source heterogeneous data.

[0169] The knowledge graph construction module is used to preprocess the collected data, identify entities, extract and fuse relationships, and construct a multimodal cognitive graph including equipment entities, fault entities, and environment entities, forming a graph structure with semantic associations and temporal logic;

[0170] The semantic fusion and reasoning module is used to achieve semantic aggregation between different modal data based on word vector models and graph neural network technology, explore potential relationships such as causality, common causes, and evolution between entities, and support knowledge reasoning based on rules or probabilities;

[0171] The equipment status assessment module is used to dynamically evaluate the health status of key equipment based on the improved Weibull-Markov model, and to quantitatively predict equipment degradation trends by combining operating history and environmental factors;

[0172] The risk prediction module is used to combine the entity status information, graph topology and historical evolution path in the cognitive graph structure to build a risk propagation network and realize quantitative assessment and prediction of potential risks;

[0173] A decision-making support module integrates equipment status assessment results with risk prediction results to form a comprehensive line-level reliability analysis report, providing maintenance scheduling recommendations, resource optimization solutions, and operation and maintenance intervention priority rankings;

[0174] The knowledge evolution module is used to achieve incremental updates of the graph structure and evaluation model based on newly accessed data, support the continuous evolution of the knowledge system, and improve the system's adaptability and intelligence level when facing new equipment, new failure modes and emergencies.

[0175] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0176] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A transmission line reliability prediction method based on a multimodal cognitive network, characterized in that: The following steps are involved: Acquire multi-source heterogeneous data on transmission lines, including equipment inventory data, operation and maintenance logs, historical fault records, meteorological data, and geographic information data; Extracting initial entities and attribute information of equipment, fault and environment classes based on the multi-source heterogeneous data; Calculate and filter the similarity of the extracted entities, merge semantically similar entities, build a three-layer metamodel framework including equipment class, fault class, and environment class, and calculate the association strength between entities; Constructing a multimodal cognitive network based on the three-layer metamodel framework and the association strength between entities; Utilizing the multimodal cognitive network, extracting real-time status parameters and degradation information of the equipment, and constructing an equipment-level dynamic failure rate prediction model based on health indicators; Integrate the equipment-level dynamic failure rate model with environmental factors to calculate the line-level basic failure risk and environmental risk, and build an overall reliability prediction model for transmission lines; Predicting the reliability and risk of a transmission line based on the overall reliability prediction model of the transmission line; The above mentioned method integrates the equipment-level dynamic failure rate model with environmental factors, calculates the line-level basic failure risk and environmental risk, and constructs an overall reliability prediction model for the transmission line, specifically including: Based on the topological association between device entities in the multimodal cognitive network, the dynamic failure rate of each device entity is calculated according to the critical weight of the device in the topological structure. Aggregate into line-level basic fault risks , the calculation formula is: in, is the device entity at the current time t The dynamic failure rate, For device entities The critical weight in the line structure is determined based on the equipment redundancy and load distribution ratio parameters, m is the total number of equipment entities in the transmission line, is the line-level basic fault risk at the current time t; Leverage the strength of associations between devices and environmental entities and the weight factors of each environmental factor , build a device-level environmental risk model: in, For device entities Equipment-level environmental risks, Represents environmental entities The weighting factor of the equipment is determined based on expert knowledge. Represents a device entity and environmental entities The strength of the association between Represents environmental entities The real-time monitoring value at the current time t, p is the total number of environmental entities; Line-level infrastructure failure risk Equipment-level environmental risks The overall reliability prediction model of the transmission line is constructed by fusion, and its comprehensive risk expression is: in, For the moment t The overall comprehensive risk of the transmission line is represented by m, where m is the total number of equipment entities.

2. The method for predicting power transmission line reliability based on a multimodal cognitive network according to claim 1, characterized in that: The extracting of initial entities and attribute information of equipment, fault, and environment classes based on the multi-source heterogeneous data specifically includes: Extract equipment entity information from equipment ledger data and build equipment entity collection ,in Indicates the Equipment class entities include conductors and towers, and each entity has attributes including equipment service time. , health status indicators , topology information , shape parameter β, scale parameter η, equipment normal operation start time τ 0 , where health status indicators Real-time detection values ​​of current, voltage and temperature of each device ; Extract fault entity information from historical fault records and operation and maintenance logs to build a fault entity set ,in Indicates the j Type of fault, the fault type entities include lightning strike, icing, equipment aging, and the set attributes include historical fault frequency , device information related to the fault ; Extract environmental entity information from meteorological data and geographic information data and construct an environmental entity set ,in Indicates the k Environmental factors, the environmental entities include slope, wind speed, lightning intensity, land use type, and the set attributes include real-time monitoring values ​​of each environmental factor 、 t Number of environmental events that occurred during the period .

3. The method for predicting power transmission line reliability based on a multimodal cognitive network according to claim 1, characterized in that: The similarity calculation and screening of the extracted entities are performed, and entities with similar semantics are merged to construct a three-layer metamodel framework including equipment class, fault class and environment class, which specifically includes: The FastText model is used to vectorize the text descriptions of equipment, fault, and environment entities to generate entity semantic vectors. ,d=300 is the entity semantic vector dimension, which is used to enhance the semantic representation ability; The similarity between entity semantic vectors of entities in the same category is calculated based on the Jaccard similarity coefficient. The calculation formula of the Jaccard similarity coefficient is: Where: A 、 B Respectively represent the semantic vectors of the entities to be matched; Setting similarity threshold , when Jaccard(A, B)≥ When , A and B are determined to be semantically similar entities and merged into a unified entity node; The merged equipment class, fault class and environment class entities are structurally divided according to entity attributes and their categories, and a three-layer metamodel framework including equipment class, fault class and environment class is constructed.

4. The method for predicting transmission line reliability based on a multimodal cognitive network according to claim 1, characterized in that: The calculation of the association strength between entities specifically includes: Based on the device, fault, and environment entity sets, the association strength between the three types of entities is calculated. The calculation formula for the association strength is: in, Represents a device entity , fault entity Environmental entities The strength of the correlation between the three Respectively represent device entities , fault entity Environmental entities The cumulative number of occurrences identified and recorded in multi-source heterogeneous data, Respectively represent device entities Fault Entity , equipment entities Environmental entities , Environmental Entities Fault Entity The number of times the same data appears together in multi-source heterogeneous data; Represents a device entity Fault Entity Environmental entities The number of times a term appears in the same data in multi-source heterogeneous data.

5. The method for predicting power transmission line reliability based on a multimodal cognitive network according to claim 1, characterized in that: The multimodal cognitive network is constructed based on the three-layer metamodel framework and the association strength between entities, specifically including: The equipment, fault and environment entities in the three-layer metamodel framework are mapped to nodes in the multimodal cognitive network, and the node attributes are entity attributes; Based on the calculated association relationship and association strength between entities, edges between nodes are constructed, and the attribute of the edge is the association strength between the node entities; Neo4j graph database is used for structured storage of multimodal cognitive networks, and the relationships between nodes and edges are organized in the form of a graph structure.

6. The method for predicting power transmission line reliability based on a multimodal cognitive network according to claim 1, characterized in that: The multimodal cognitive network is used to extract the real-time status parameters and degradation information of the equipment and to construct a device-level dynamic failure rate prediction model based on health indicators, specifically including: The improved Weibull-Markov chain algorithm is used to quantify the impact of equipment aging and external environmental factors on the failure rate, and the equipment aging correction coefficient and the external environment correction coefficient are obtained. According to the real-time detection value of the device , using health status indicators Evaluate the health status of the equipment, health index The calculation formula is: in, For the i Indicators at time t The detection values ​​include the real-time current, voltage and temperature of the equipment; For the i The weight of each health indicator is determined by the hierarchical analysis method; N is the number of real-time detection values, 、 is the threshold range of the real-time monitoring value; Based on the equipment, fault and environment entities and their attribute information extracted from the multimodal cognitive network, a device-level time-varying failure rate prediction model is constructed by combining the equipment aging correction coefficient, external environment correction coefficient and health index.

7. The method for predicting power transmission line reliability based on a multimodal cognitive network according to claim 6, characterized in that: The equipment aging correction factor formula is: in, For device entities The equipment aging correction factor at the current time t is, 、 Equipment entities The shape and scale parameters of is the current time, For device entities Normal operation start time; The calculation formula of the external environment correction coefficient is: in, Represents a device entity External environment correction factor, Represents a device entity The number of occurrences of the associated k-th environmental entity, For all entities with devices The total number of occurrences of the associated environmental entity, is the total number of environmental entities considered, Represents environmental entities The impact weight on the equipment is determined based on expert knowledge.

8. The method for predicting transmission line reliability based on a multimodal cognitive network according to claim 6, characterized in that: The device-level time-varying failure rate prediction model is: in, is the device entity at the current time t The dynamic failure rate, is the baseline failure rate, i.e. the initial failure rate without degradation or environmental influences, is a set of fault entities, is a collection of environment entities, 、 Equipment entities With fault entity , equipment entity and environmental entities The strength of association, For in time t Internal fault entity The number of occurrences, For in time t The total number of occurrences of all fault entities in Environmental Entity Real-time monitoring value at time t.

9. The method for predicting power transmission line reliability based on a multimodal cognitive network according to claim 1, characterized in that: The weighting factors of the environmental factors Determined by entropy weight method, the formula is: in, Environmental Entity The information entropy of Environmental Entity At the moment The real-time monitoring value, M is the number of samples, that is, the time segment or the total number of monitoring points used for statistical analysis, Environmental Entity The sum of all sample values ​​of .

Images