Power transmission line environment risk assessment method fusing expert knowledge

Abstract

The invention relates to the field of power transmission line risk hidden danger monitoring, and provides a power transmission line environment risk assessment method fusing expert knowledge, which comprises the following steps: collecting a multi-source disaster-causing rule and carrying out symbol coding; embedding the knowledge graph to generate a rule vector, and constructing a rule factor topological graph; the graph neural network extracts high-order interaction features to generate a composite rule; calculating a rule confidence coefficient in combination with an expert knowledge base and a time sequence attenuation factor algorithm; and fusing the rule confidence and real-time monitoring data, constructing an evidence space, calculating a risk probability, and generating a risk assessment report containing a traceability path and an evidence cluster. According to the method, the heterogeneous rule is projected into the vector space through knowledge graph embedding, the graph neural network is driven to aggregate the high-order features, the cross-domain composite rule is generated, the novel disaster chain recognition efficiency is improved, and the response delay is reduced; the semantic similarity of the distance quantification rule is edited, the confidence coefficient is corrected in real time by adopting a coupling exponential attenuation function, and the confidence error is reduced through dynamic adjustment under the extreme working condition.

Description

Technical Field

[0001] This invention relates to the field of monitoring risks and hazards in power transmission lines, and in particular to a method for assessing environmental risks of power transmission lines that integrates expert knowledge. Background Technology

[0002] Environmental risk assessment of transmission lines has formed a multi-source information collaborative analysis framework, and its technological evolution can be divided into three stages: the early stage mainly relied on meteorological parameter threshold models and static geographic information system overlay analysis;

[0003] In the medium term, machine learning algorithms are introduced to establish single-factor risk mapping;

[0004] Current technological frontiers focus on federated knowledge fusion frameworks and cross-modal deep feature extraction. By integrating heterogeneous data sources such as satellite remote sensing, ground-based monitoring stations, and numerical weather prediction, the accuracy of identifying single risk factors such as wildfires, icing, and lightning strikes is improved. At the knowledge representation level, the improved model based on DS evidence theory achieves probabilistic fusion of multi-sensor data, while symbolic rule expert systems serve as the core carrier for logical decision-making.

[0005] Current transmission line environmental risk assessment technology faces two major bottlenecks. On the one hand, the ability to model dynamic disaster chains is severely lacking. The system is limited by predefined rule combinations and cannot construct a transmission topology network between multi-source rules from meteorology, geology, and equipment. This leads to delays in identifying novel cascading paths, and the black-box nature of deep learning results in low coverage of cross-domain rule interactions. On the other hand, the confidence decay mechanism is decoupled from environmental parameters. Rule credibility depends on the initial static allocation of expert weights. The fixed decay coefficient does not integrate real-time variables such as sudden wind speed changes and equipment aging. The confidence bias caused by extreme scenarios ultimately leads to the failure of key rules and a surge in false alarm rates.

[0006] Therefore, this invention proposes a method for assessing the environmental risks of transmission lines that integrates expert knowledge, in order to solve the problems existing in the prior art. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for assessing the environmental risks of transmission lines that integrates expert knowledge, thereby resolving the problems mentioned in the background section.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] Firstly, this application provides a method for environmental risk assessment of transmission lines that integrates expert knowledge, comprising the following steps:

[0010] S1. Collect multi-source disaster-causing rules and obtain rule symbol encoding through preprocessing;

[0011] S2. Input rule symbol encoding, embed the output rule vector through knowledge graph, and construct a rule factor topology graph;

[0012] S3. Input the rule factor topology graph into the graph neural network to extract high-order interaction features and generate composite rules;

[0013] S4. Load the expert knowledge base dataset based on composite rules to obtain the initial confidence level, and generate the rule confidence level through the time-series decay factor algorithm;

[0014] S5. Integrate rule confidence with real-time data of transmission lines, calculate risk probability through multi-source evidence, and generate risk assessment report through decision tracing.

[0015] Preferably, the multi-source disaster-causing rules include case library evidence, meteorological disaster rules, geological disaster rules, power grid equipment status rules, and system operation characteristic rules.

[0016] Preferably, the preprocessing in S1 includes data cleaning, outlier handling, time series alignment, and format normalization; based on the preprocessed multi-source disaster-causing rules, rule symbol encoding is obtained through feature symbol mapping; the rule symbol encoding includes feature state symbols, logical operators, and disaster conclusion symbols.

[0017] Preferably, S2 specifically includes:

[0018] S201. Based on rule symbol encoding, rule triples are constructed by decomposition, and an entity vector mapping dictionary is generated by assigning random initial vectors.

[0019] S202. Extract the conditional entity vectors from the entity vector mapping dictionary, perform an arithmetic mean operation to obtain the fusion vector, and generate regular vectors through principal component analysis for dimensionality reduction.

[0020] S203. Load all rule vectors into memory to construct a rule vector space, and generate the degree of association using the cosine similarity algorithm to obtain the rule similarity matrix;

[0021] S204. Obtain the set of directed edges based on the core connections in the rule similarity matrix, and construct the rule factor topology graph by creating a node set of topology graph nodes.

[0022] Preferably, S3 specifically includes:

[0023] S301. Load the topology graph data in the rule factor topology graph and construct a graph neural network through the node feature update function;

[0024] S302. Extract the regular nodes in the graph neural network and perform neighbor feature aggregation to generate a high-order interaction feature vector;

[0025] S303. The high-order interaction feature vector is averaged and pooled to generate community feature vector. The high-order vector of the bridge node is obtained by identifying cross-community connection nodes, and composite rules are generated by fusing community feature vectors.

[0026] Preferably, S4 specifically includes:

[0027] S401. Read the expert knowledge base dataset to obtain the stored expert review records, and use the edit distance algorithm to perform similarity matching based on the composite rules to obtain the matching benchmark value;

[0028] S402. Calculate the rule time decay difference based on expert review records, and generate rule confidence through an exponential decay function.

[0029] Preferably, the expert review record in S4 includes a regular expression, an initial confidence level, and a last update time field.

[0030] Preferably, S5 specifically includes:

[0031] S501. Combine the rule confidence level with the evidence from the case library to form a multi-source evidence set, and construct the geometric structure of the evidence space according to the distribution of evidence types.

[0032] S502. Calculate the distance between evidence points in the spatial geometric structure to form evidence clusters, and generate a risk probability surface by extracting the core evidence clusters.

[0033] S503. Calculate the risk probability by integrating the center trajectory based on the risk probability surface, and obtain the source address by marking the coordinate index of the original evidence in space.

[0034] S504. Write the risk probability, traceability address, and evidence cluster into a dynamic data structure to generate a risk assessment report.

[0035] Secondly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when executed by the processor, the computer program causes the processor to perform the steps of the transmission line environmental risk assessment method of this application that integrates expert knowledge.

[0036] Thirdly, this application also provides a computer-readable storage medium storing instructions thereon, which, when executed by a processor, enable the implementation of a transmission line environmental risk assessment method incorporating expert knowledge.

[0037] The present invention discloses a method for environmental risk assessment of transmission lines that integrates expert knowledge, which has the following beneficial effects.

[0038] This invention includes: collecting multi-source disaster-causing rules and encoding them with symbols; generating rule vectors based on knowledge graph embedding and constructing a rule factor topology graph; inputting the rule factor topology graph into a graph neural network to extract high-order interaction features and generate composite rules; generating rule confidence scores by combining an expert knowledge base and a time-series decay factor algorithm; fusing rule confidence scores with real-time monitoring data of transmission lines to construct an evidence space, calculate risk probabilities, and generate a risk assessment report containing source tracing paths and evidence clusters. This invention projects heterogeneous rules into a vector space through knowledge graph embedding, driving a graph neural network to aggregate high-order features and generate cross-domain composite rules, thereby improving the efficiency of identifying new disaster chains and reducing response latency; it quantifies the semantic similarity of rules through edit distance, uses a coupling exponential decay function to correct confidence scores in real time, and dynamically adjusts the confidence error under extreme conditions. Attached Figure Description

[0039] Figure 1 This is an overall flowchart of a transmission line environmental risk assessment method that integrates expert knowledge, as described in this application.

[0040] Figure 2 This is a flowchart of the construction of the rule factor topology graph in this application.

[0041] Figure 3 This is a flowchart of the process for generating composite rules in this application.

[0042] Figure 4 This is a flowchart of the process for generating a risk assessment report for this application.

[0043] Figure 5 This is a schematic diagram of the computer device structure of this application. Detailed Implementation

[0044] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0045] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0046] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0047] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0048] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. For those skilled in the art to which this application pertains, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0049] As disclosed in the background section, current transmission line environmental risk assessment technology faces two major bottlenecks. On the one hand, the ability to model dynamic disaster chains is severely lacking. The system is limited by predefined rule combinations and cannot construct a transmission topology network between multi-source rules from meteorology, geology, and equipment. This leads to delays in identifying novel cascading paths, and the black-box nature of deep learning results in low coverage of cross-domain rule interactions. On the other hand, the confidence decay mechanism is decoupled from environmental parameters. Rule credibility depends on the initial static allocation of expert weights. The fixed decay coefficient does not integrate real-time variables such as sudden wind speed changes and equipment aging. The confidence bias caused by extreme scenarios ultimately leads to the failure of key rules and a surge in false alarm rates.

[0050] To address the aforementioned problems, this invention proposes a method for assessing the environmental risks of transmission lines that integrates expert knowledge. This method includes the following steps:

[0051] S1. Collect multi-source disaster-causing rules and obtain rule symbol encoding through preprocessing;

[0052] S2. Input rule symbol encoding, embed the output rule vector through knowledge graph, and construct a rule factor topology graph;

[0053] S3. Input the rule factor topology graph into the graph neural network to extract high-order interaction features and generate composite rules;

[0054] S4. Load the expert knowledge base dataset based on composite rules to obtain the initial confidence level, and generate the rule confidence level through the time-series decay factor algorithm;

[0055] S5. Integrate rule confidence with real-time data of transmission lines, calculate risk probability through multi-source evidence, and generate risk assessment report through decision tracing.

[0056] The technical solution provided by this invention projects heterogeneous rules into a vector space through knowledge graph embedding, drives graph neural networks to aggregate high-order features, and generates cross-domain composite rules, thereby improving the efficiency of new disaster chain identification and reducing response latency. By editing the distance to quantify the semantic similarity of the rules, the confidence level is corrected in real time using a coupling exponential decay function, and the confidence error is reduced through dynamic adjustment under extreme conditions.

[0057] The above plan will be explained in detail below.

[0058] Please see the appendix Figure 1 , Figure 1 This is a schematic flowchart illustrating the main steps of a transmission line environmental risk assessment method integrating expert knowledge, as described in this invention. Figure 1 As shown, an embodiment of the present invention provides a method for assessing the environmental risk of transmission lines that integrates expert knowledge, which mainly includes the following steps:

[0059] S1. Collect multi-source disaster-causing rules and obtain rule symbol codes through preprocessing. The purpose of S1 is to standardize and structure the disaster-causing rules described in natural language and transform them into a machine-readable symbol code format to support subsequent map construction and modeling analysis.

[0060] Preferably, in this embodiment, the multi-source disaster-causing rules include case library evidence, meteorological disaster rules, geological disaster rules, power grid equipment status rules, and system operation characteristic rules;

[0061] It should be noted that a case library of evidence is constructed by collecting complete data records of historical power grid fault events. This serves as an empirical rule to provide decision-making references for similar scenarios in the three-dimensional evidence space, supporting the generation of risk probability surfaces from multi-source evidence clusters. Meteorological disaster rules refer to logical criteria for determining environmental risks based on meteorological element thresholds (e.g., "wind speed > 20 m / s triggers wind deviation risk"), which quantifies the probability of physical damage to transmission lines caused by climatic factors such as typhoons and lightning. Geological disaster rules are early warning conditions set based on geological activity parameters (e.g., "daily displacement > 5 mm and rainfall > 50 mm → landslide risk"), which predicts tower structure failures caused by geological activities such as landslides and subsidence. Power grid equipment status rules are status assessment criteria formulated based on real-time equipment monitoring data (e.g., "ice thickness > 10 mm and wind speed > 15 m / s → conductor galloping alarm"), enabling early warning of equipment mechanical failures. The system operation characteristic rules focus on the safety boundaries of power grid operation parameters (such as "line load rate > 85% and thunderstorm probability > 60% → tripping risk") to ensure the controllability of power grid stability risks.

[0062] Preferably, in this embodiment, preprocessing includes data cleaning, outlier handling, time series alignment, and format normalization.

[0063] It should be noted that data cleaning refers to removing invalid records and redundant information (such as deleting empty rows when sensor communication is interrupted), which is used to ensure the basic quality of the data; outlier handling corrects observations that exceed reasonable ranges (such as smoothing sudden jumps in temperature and humidity sensor values ​​through median filtering), which is used to eliminate noise interference caused by false alarms from equipment; time series alignment unifies the time base of multi-source data (such as aligning 1-hour sampling of meteorological data with 5-minute sampling of equipment monitoring through interpolation), which solves the problem of time scale fragmentation; format normalization transforms heterogeneous data into a structured form (such as text log "wind speed: 12.3m / s" → numerical field 12.3), which realizes a unified data interface that can be parsed by machines.

[0064] Based on the preprocessed multi-source disaster-causing rules, the rule symbol encoding is obtained through feature symbol mapping;

[0065] It should be noted that the preprocessed multi-source disaster-causing rules are used as input data. Continuous monitoring values ​​in the multi-source disaster-causing rules are discretized and converted into feature state symbols (e.g., wind speed monitoring values ​​greater than 20 m / s are mapped to the high wind speed symbol "WS_H"); logical relation operators are uniformly converted into standardized logical operators (e.g., the "AND" relation in natural language is mapped to the logical AND operator "&"); risk type identifiers are directly mapped to disaster conclusion symbols (e.g., "wildfire risk" is mapped to "FIRE_RISK"). The rule symbol encoding output is obtained, and the feature state symbols, logical operators, and disaster conclusion symbols are combined to form a complete encoding structure (e.g., the input rule "wind speed > 20 m / s and humidity < 30% → wildfire risk" is mapped to the symbol encoding "(WS_H&HUM_L) → FIRE_RISK"). The feature symbol mapping process ensures the accurate conversion of natural language rules into machine-parseable semantic units, providing a structured expression foundation for subsequent knowledge graph embedding.

[0066] The rule symbol encoding includes feature state symbols, logical operators, and disaster conclusion symbols.

[0067] It should be noted that feature state symbols are discretized classification identifiers for continuous monitoring values, serving to standardize the expression of environmental characteristics; logical operators are conditional relational symbols connecting multiple feature states, functioning to construct machine-readable composite conditional structures; disaster conclusion symbols represent semantic identifiers of risk types, with the core value of accurately outputting risk assessment conclusions. Feature state symbols collaboratively form machine-parseable rule logic chains, providing structured semantic units for knowledge graph embedding, achieving lossless transformation from natural language experience to computer instructions, and supporting accurate decision-making in subsequent risk assessment models.

[0068] S2. Input rule symbol encoding, embed the output rule vector through knowledge graph, and construct a rule factor topology graph. The purpose of S2 is to transform the symbolic rules extracted in step S1 into rule vectors that can participate in mathematical operations, and construct a "rule factor topology graph" based on the semantic similarity between rules. That is, a directed graph structure with rules as nodes and the correlation strength between rules as edges, which provides a foundation for subsequent modeling and interaction of high-order relationships between rules based on graph neural networks (GNN).

[0069] Preferably, in this embodiment, such as Figure 2 As shown, S2 specifically includes:

[0070] S201. Based on rule symbol encoding, rule triples are constructed by decomposition, and an entity vector mapping dictionary is generated by assigning random initial vectors.

[0071] Specifically, in this embodiment, the condition entity vector, logical relation predicate, and conclusion entity vector are obtained by parsing the rule symbol encoding. By separating the logical operators, a list containing all unique entity names is generated. A 64-dimensional floating-point vector is initialized using a standard normal distribution. A key-value pair mapping structure between entity names and 64-dimensional vectors is established, and finally, an entity vector mapping dictionary is output.

[0072] It should be noted that rule triples include conditional entity vectors, logical relation predicates, and conclusion entity vectors; their core function is to transform natural language rules into machine-computable logical units and establish causal relationship mappings between entities. In a rule triple, the conditional entity vector represents the state of the disaster-causing factor, such as the wind speed symbol "WS_H"; the logical relation predicate defines the interaction between entities, such as "→"; and the conclusion entity vector identifies the risk type, such as "FIRE_RISK". Rule triples achieve precise semantic parsing through structured separation, ensuring the effectiveness of vector relationship learning in knowledge graph embedding. As a fundamental unit of knowledge representation, rule triples support subsequent vector space mapping and graph neural network topology construction.

[0073] S202. Extract the conditional entity vectors from the entity vector mapping dictionary, perform an arithmetic mean operation to obtain the fusion vector, and generate regular vectors through principal component analysis for dimensionality reduction.

[0074] Specifically, in this embodiment, the rule symbol encoding for the current processing is determined. Before processing, the uniqueness of each rule symbol encoding should be confirmed, and it corresponds to a disaster-causing rule. Based on the current rule symbol encoding, all associated conditional entity vectors (such as the vectors corresponding to "TEM_H" and "HUM_L") are extracted from the entity vector mapping dictionary. An arithmetic mean is performed on all conditional entity vectors to calculate the average value of each dimension's coordinates, generating a fusion vector. Principal component analysis is applied to process the 64-dimensional fusion vector, calculating the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvectors corresponding to the first 32 largest eigenvalues ​​are selected to construct a projection matrix. Finally, through the 64-dimensional fusion vector and the projection matrix, the dimensionality-reduced rule vector is output.

[0075] S203. Load all rule vectors into memory to construct a rule vector space, and generate the degree of association using the cosine similarity algorithm to obtain the rule similarity matrix;

[0076] Specifically, in this embodiment, the generated rule vectors are loaded into an ordered memory storage structure. The rule vector set contains all rule symbol codes and their corresponding 32-dimensional vectors. Cosine similarity calculations are performed on each pair of rule vectors in the rule vector space to generate a complete rule similarity matrix. The specific operation process is as follows: To achieve efficient similarity calculation, all rule symbol codes are centrally organized, a rule symbol code list is created, and the rule vectors corresponding to the indices are extracted to construct a two-dimensional array matrix. The row and column vectors of the matrix are traversed to calculate cosine similarity. When the rule vector space matrix has a size of N×32, an N×N dimensional rule similarity matrix is ​​output. The matrix elements are positioned, and the similarity values ​​between the rule symbol codes are stored. For example, the rule vector space contains rule ID_001 vector [0.12, 0.24, ..., 0.17] and rule ID_002 vector [0.09, -0.11, ..., 0.03]. The cosine similarity calculation result 0.86 is stored in the first row and second column of the rule similarity matrix. Finally, the rule similarity matrix is ​​constructed. The expression for calculating cosine similarity is:

[0077]

[0078] Where s is the cosine similarity. Let be the feature vector of the i-th rule in the rule vector space. Let be the feature vector of the j-th rule in the rule vector space. vector The Euclidean norm, vector The Euclidean norm.

[0079] S204. Obtain the set of directed edges based on the core connections in the rule similarity matrix, and construct the rule factor topology graph by creating a node set of topology graph nodes.

[0080] Specifically, in this embodiment, the similarity sequence corresponding to each rule in the rule similarity matrix is ​​processed, and a connection is established for the target rule with high similarity; a set of directed edges containing the symbol codes of the source rule and the target rule is generated; topology graph nodes are created based on the rule symbol code list, and each node is bound to rule attribute data to form a node set; the set of directed edges and the set of nodes are input into the graph building engine to generate a rule factor topology graph. The similarity sequence of the current rule is obtained by traversing the row vectors of the rule similarity matrix; the high-value values ​​and corresponding rule symbol codes in the sequence are filtered; the rule symbol codes are traversed, and a node object is created and attributes are bound for each rule symbol code; all nodes form a node set; the set of directed edges and the set of nodes are loaded into the graph building engine; and the engine interface is called to generate a rule factor topology graph containing nodes and edges.

[0081] S3. Input the rule factor topology graph into the graph neural network to extract high-order interaction features and generate composite rules. The purpose of S3 is to feed the rule factor topology graph constructed in S2 into the graph neural network (GNN) to learn the complex influence relationships between rules, and thus generate "composite rules". "Composite rules" are new rules generated by integrating multiple similar rules and strongly correlated rules, which can express richer disaster mechanisms or risk warning conditions.

[0082] Preferably, in this embodiment, such as Figure 3 As shown, S3 specifically includes:

[0083] S301. Load the topology graph data in the rule factor topology graph and construct a graph neural network through the node feature update function;

[0084] Specifically, in this embodiment, the node set data structure and directed edge set data structure contained in the loading rule factor topology graph are used; a node feature update function is defined based on the topology structure, and the neighbor node feature mean aggregation operation is performed and a ReLU activation function nonlinear transformation is applied. A hierarchical progressive mechanism is adopted when constructing the multi-layer graph neural network framework. First, the input layer is initialized, loading the initial feature vectors from the node set to the bottom layer of the network; in this embodiment, the neural network can use GCN or GraphSAGE; then, neighborhood feature propagation is performed, with the first layer aggregating the features of direct neighbor nodes and generating intermediate feature vectors; next, higher-order feature extraction is performed, with the second layer aggregating the features of one-hop neighbor nodes based on the output of the previous layer to generate a deep representation vector; finally, the output layer generates the final node feature vector as a higher-order interaction feature vector. This completes the construction of the graph neural network.

[0085] S302. Extract the regular nodes in the graph neural network and perform neighbor feature aggregation to generate a high-order interaction feature vector;

[0086] Specifically, in this embodiment, the rule nodes contained in the graph neural network framework are extracted; each rule node identifier is traversed; the set of direct neighbor nodes of the rule node in the graph neural network framework is retrieved; a neighbor node feature aggregation operation is performed, and the feature vectors corresponding to the direct neighbor nodes are aggregated using the mean calculation method; an intermediate aggregated feature vector of the current rule node is generated. The aggregation range is further expanded; based on the first-layer intermediate aggregation result, the set of first-layer neighbor nodes of the rule node is retrieved; the first-layer neighbor node feature vectors are aggregated using the mean calculation method to generate a higher-order aggregated feature vector. The multi-layer aggregation outputs are integrated, and the direct neighbor aggregated feature vector is concatenated with the first-layer neighbor aggregated feature vector to form a higher-order interactive feature vector. After completing the traversal processing of all rule nodes, a set data structure containing all rule node identifiers and their corresponding higher-order interactive feature vectors is output, ultimately forming a complete higher-order interactive feature vector.

[0087] S303. The high-order interaction feature vector is averaged and pooled to generate community feature vector. The high-order vector of the bridge node is obtained by identifying cross-community connection nodes, and composite rules are generated by fusing community feature vectors.

[0088] Specifically, in this embodiment, the high-order interaction feature vector set is subjected to community mean pooling. Each community identifier in the community partitioning structure is traversed, and the high-order interaction feature vectors corresponding to all rule nodes within the community are extracted. An arithmetic mean operation is then performed to generate community feature vectors. Cross-community connection nodes are located, and the connection structure of the rule factor topology graph is analyzed. Rule nodes that simultaneously connect two or more communities are identified as bridge nodes. The high-order interaction feature vectors corresponding to the bridge nodes are extracted as high-order vectors of the bridge nodes. Multi-community features are fused by performing dimensional concatenation operations between the high-order vectors of the bridge nodes and the feature vectors of the associated communities to form a fused feature matrix. The fused feature matrix is ​​then processed through a fully connected neural network to output composite rules.

[0089] like Figure 2 As shown, S4 loads the expert knowledge base dataset based on the composite rule to obtain the initial confidence level, and generates the rule confidence level through the time decay factor algorithm. The goal of S4 is to compare the "composite rule" generated in S3 with the existing rules in the "expert knowledge base dataset", determine the similarity by editing distance, find the most similar expert rule, and combine the expert's "confidence score" and "update time" to dynamically generate the confidence level (confidence level) of the current composite rule using the "time decay mechanism".

[0090] Preferably, in this embodiment, S4 specifically includes:

[0091] S401. Read the expert knowledge base dataset to obtain the stored expert review records, and use the edit distance algorithm to perform similarity matching based on the composite rules to obtain the matching benchmark value;

[0092] Specifically, in this embodiment, the stored expert review records in the expert knowledge base dataset are traversed, and the content of the rule expression field in each stored expert review record is extracted; the composite rule is used as the base rule expression, and the edit distance algorithm is used to calculate the string similarity value between the base rule expression and the rule expression field content in the stored expert review record; the stored expert review record corresponding to the highest similarity value is filtered by the similarity value sorting operation, and the initial confidence field value of the stored expert review record corresponding to the highest similarity value is extracted as the matching benchmark value;

[0093] It should be noted that the expert knowledge base dataset is derived from power technology regulations, power expert experience documents, and accident analysis reports. During construction, natural language rules are parsed into machine-readable symbolic expressions. Experts assign initial confidence levels and mark the last update timestamp. Failure rules are updated through expert review, resulting in an expert knowledge base dataset containing a rule expression field, an initial confidence level field, and a last update time field.

[0094] Preferably, in this embodiment, the expert review record includes a regular expression, an initial confidence level, and a last update time field;

[0095] It should be noted that the rule expression stores a formalized string description of the risk judgment logic statement defined by the expert for subsequent similarity matching operations; the initial confidence score stores a decimal value between 0 and 1 assigned by the expert to the rule expression, used as a reference benchmark for confidence score generation; the last update time field stores the timestamp of the expert's most recent review and revision of the rule, accurate to the second, used to calculate the time decay difference and determine the confidence score timeliness weight; the rule expression provides a semantic carrier for the rule, supporting the edit distance algorithm to quantify similarity and realize expert knowledge transfer; the initial confidence score carries the expert's experience and authority value to support the dynamic adjustment calculation of subsequent confidence scores; the last update time field records the rule lifecycle status and controls the confidence score decay rate.

[0096] S402. Calculate the rule time decay difference based on expert review records, and generate rule confidence through an exponential decay function.

[0097] Specifically, in this embodiment, the last update time field of the expert review record is read to extract the last update timestamp value of the rule; the current system timestamp value is obtained, and the decay difference between the last update timestamp value and the current system timestamp value is calculated. The timestamp difference is converted into a day interval value through a conversion factor; the confidence level is dynamically calculated using an exponential decay function, with the natural constant e as the base and the product of the input decay coefficient and the day interval value as the exponent to generate the decay factor value; the initial confidence level field value of the expert review record is extracted as the calculation benchmark value; the product of the initial confidence level field value and the decay factor value is used as the output to finally generate the rule confidence level.

[0098] S5 integrates rule confidence levels with real-time data from transmission lines, calculates risk probabilities using multi-source evidence, and generates a risk assessment report through decision tracing. The purpose of S5 is to construct an "evidence space geometric model," integrate multi-source evidence to calculate risk probabilities, and generate a risk assessment report containing tracing paths, thereby achieving a quantifiable, verifiable, and traceable closed loop for risk assessment conclusions.

[0099] Preferably, in this embodiment, S5 specifically includes:

[0100] S501. Combine the rule confidence level with the evidence from the case library to form a multi-source evidence set, and construct the geometric structure of the evidence space according to the distribution of evidence types.

[0101] Specifically, in this embodiment, the loaded rule confidence entity data and case library evidence records form a multi-source evidence set; the rule confidence includes composite rule symbol encoding, confidence value, and benchmark template name fields; the case library evidence records store historical case feature vectors. Evidence type distribution is defined: rule confidence is categorized as expert knowledge evidence; case library evidence is categorized as historical experience evidence; real-time monitoring data is categorized as real-time observation evidence; and dimensional unification is achieved by scaling the original data of various types of evidence to the [0,1] interval, forming a type-labeled evidence set. A three-dimensional Cartesian coordinate system evidence space geometry is constructed: the X-axis is mapped to the time freshness dimension, and the coordinate value is converted by calculating the evidence timestamp using an inverse proportional function (example: evidence from 30 days ago → X = 0.2); the Y-axis is mapped to the data completeness dimension, and the coordinate value is converted based on the data point completeness rate (example: 7 / 10 data complete → Y = 0.7); the Z-axis is mapped to the confidence strength dimension, and the rule confidence value is directly mapped to the coordinate value (example: 88% → Z = 0.88). Perform evidence space coordinate localization, output evidence space geometric structure dataset with coordinate index, and finally construct evidence space geometric structure.

[0102] Specifically, in this embodiment, the real-time data of transmission lines includes meteorological parameters, geological activity monitoring values, equipment mechanical status parameters, and power grid operation characteristic parameters. Meteorological parameters in the real-time data refer to atmospheric environmental monitoring values ​​such as wind speed, temperature, and humidity, used to quantify the physical damage risk to the lines caused by meteorological disasters such as typhoons and thunderstorms. Geological activity monitoring values ​​include geological disaster indicators such as mountain displacement and ground subsidence, functioning to predict tower structure failures caused by landslides and collapses. Equipment mechanical status parameters cover real-time mechanical characteristic parameters such as conductor icing thickness and insulator salt density, enabling early warning of abnormal equipment operating conditions. Power grid operation characteristic parameters include system operating boundary data such as line load rate and tripping frequency, with the core function of ensuring that power grid stability risks remain within a controllable threshold. These four types of data form a dynamic risk assessment basis through multi-source fusion, supporting the decision-making upgrade from single disaster early warning to complex disaster chain prevention and control.

[0103] S502. Calculate the distance between evidence points in the spatial geometric structure to form evidence clusters, and generate a risk probability surface by extracting the core evidence clusters.

[0104] Specifically, in this embodiment, the Euclidean distance between evidence points in the evidence space geometric structure dataset is calculated, the distance between two points is calculated, and evidence points are connected to form a connected subgraph. The connected subgraph is marked as an independent evidence cluster data structure. The core evidence cluster is extracted by statistically analyzing the evidence cluster with the most evidence points within each evidence cluster. A 0.1 step size three-dimensional grid is established in the evidence space coordinate system (X / Y / Z axes 0.0→1.0, divided into 10×10×10 grid cells). The number of core evidence cluster points in each grid cell is counted as the density value (example: grid (0.8-0.9, 0.7-0.8, 0.8-0.9) contains 3 points → density 3). The cubic spline interpolation algorithm is applied to transform the discrete density points into a continuous surface, and finally, a risk probability surface is generated.

[0105] S503. Calculate the risk probability by integrating the center trajectory based on the risk probability surface, and obtain the source address by marking the coordinate index of the original evidence in space.

[0106] Specifically, in this embodiment, the core high-probability region in the risk probability surface is identified; the vertex coordinate sequence of the high-probability region is extracted to form the central trajectory path, line integrals are calculated along the central trajectory path, and the cumulative probability is normalized to the [0,1] interval to output the final risk probability value; the coordinate points contained in the central trajectory are retrieved in the evidence space geometric structure dataset; the original evidence index identifier corresponding to the coordinate point is located (example: coordinates (0.85,0.75,0.85) are associated with evidence ID_E207), and the core evidence cluster identifier to which evidence ID_E207 belongs is queried through the evidence cluster data structure; the original storage address of all evidence within the core evidence cluster is extracted, and finally the source address is generated. The expression for calculating the risk probability is:

[0107]

[0108] Where R represents the probability of risk. This indicates that the risk probability surface lies at the center point of its trajectory. To represent the three-dimensional coordinates of discrete points on the central trajectory, L represents the total length of the central trajectory in the evidence space coordinate system, N represents the total number of segments after discretization of the central trajectory, k represents the index of the discrete points of the central trajectory, and Δs represents the arc length of the infinitesimal element of the central trajectory segment.

[0109] S504. Write the risk probability, traceability address, and evidence cluster into a dynamic data structure to generate a risk assessment report.

[0110] Specifically, in this embodiment, a risk assessment report is dynamically constructed, comprising five key elements: decision probability value, source tracing path record, core evidence cluster information, assessment time stamp, and line location identifier. The risk intensity value calculated from the spatial probability surface is written into the probability value field, quantifying the environmental risk level. The source tracing path record structure is integrated to store key evidence indexes such as expert knowledge paths, historical case paths, and real-time monitoring paths. Core evidence cluster information is written, including cluster number, evidence point coordinates, and evidence type classification. A time stamp is recorded at the time of the assessment, using an international standard time format. The line location identifier clearly defines the assessment area. The decision probability value reflects the severity of the risk, the source tracing path supports tracing back to original evidence layer by layer, the core evidence cluster displays the spatial distribution pattern, and the time stamp and line identifier record the assessment spatiotemporal benchmark. This report structure achieves a unified system of numerical conclusions, data tracing, and spatial distribution, ultimately generating a risk assessment report that meets the power system's requirements for transparency, verifiability, and traceability in risk assessment.

[0111] This embodiment also provides a computer device applicable to a transmission line environmental risk assessment method that integrates expert knowledge, such as... Figure 5 As shown, computer device 200 includes at least one memory 210 (e.g., non-volatile memory such as flash memory, ROM, hard disk drive, magnetic disk, optical disk), at least one processor 220, and computer program 230. Memory 210 stores the computer program 230, which can be executed by processor 220. Processor 230 is configured to run the computer program 230 stored on memory 210. The above-described method can be achieved by running the computer program stored on one or more memories on one or more processors (e.g., in a manner where multiple processors cooperate to run the computer program or a single processor runs the computer program independently). Figure 1 The method includes one or more steps or operations. The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the computer device is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be an LCD screen or an e-ink display screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0112] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a transmission line environmental risk assessment method incorporating expert knowledge as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0113] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0114] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for environmental risk assessment of transmission lines that integrates expert knowledge, characterized in that, Includes the following steps: S1. Collect multi-source disaster-causing rules and obtain rule symbol encoding through preprocessing; S2. Input rule symbol encoding, embed the output rule vector through knowledge graph, and construct a rule factor topology graph; S3. Input the rule factor topology graph into the graph neural network to extract high-order interaction features and generate composite rules; S4. Load the expert knowledge base dataset based on composite rules to obtain the initial confidence level, and generate the rule confidence level through the time-series decay factor algorithm; S5. Integrate rule confidence with real-time data of transmission lines, calculate risk probability through multi-source evidence, and generate risk assessment report through decision tracing.

2. The method for environmental risk assessment of transmission lines integrating expert knowledge as described in claim 1, characterized in that, The multi-source disaster-causing rules include case library evidence, meteorological disaster rules, geological disaster rules, power grid equipment status rules, and system operation characteristic rules.

3. The method for environmental risk assessment of transmission lines integrating expert knowledge as described in claim 2, characterized in that, The preprocessing in S1 includes data cleaning, outlier handling, time series alignment, and format normalization; based on the preprocessed multi-source disaster-causing rules, rule symbol encoding is obtained through feature symbol mapping; the rule symbol encoding... This includes characteristic state symbols, logical operators, and disaster conclusion symbols.

4. The method for environmental risk assessment of transmission lines integrating expert knowledge as described in claim 1, characterized in that, S2 specifically includes: S201. Based on rule symbol encoding, rule triples are constructed by decomposition, and an entity vector mapping dictionary is generated by assigning random initial vectors. S202. Extract the conditional entity vectors from the entity vector mapping dictionary, perform an arithmetic mean operation to obtain the fusion vector, and generate regular vectors through principal component analysis for dimensionality reduction. S203. Load all rule vectors into memory to construct a rule vector space, and generate the degree of association using the cosine similarity algorithm to obtain the rule similarity matrix; S204. Obtain the set of directed edges based on the core connections in the rule similarity matrix, and construct the rule factor topology graph by creating a set of nodes to generate a topology graph node.

5. The method for environmental risk assessment of transmission lines integrating expert knowledge as described in claim 1, characterized in that, S3 specifically includes: S301. Load the topology graph data in the rule factor topology graph and construct a graph neural network through the node feature update function; S302. Extract the regular nodes in the graph neural network and perform neighbor feature aggregation to generate a high-order interaction feature vector; S303. The high-order interaction feature vector is averaged and pooled to generate community feature vector. The high-order vector of the bridge node is obtained by identifying cross-community connection nodes, and composite rules are generated by fusing community feature vectors.

6. The method for environmental risk assessment of transmission lines integrating expert knowledge as described in claim 1, characterized in that, S4 specifically includes: S401. Read the expert knowledge base dataset to obtain the stored expert review records, and use the edit distance algorithm to perform similarity matching based on the composite rules to obtain the matching benchmark value; S402. Calculate the rule time decay difference based on expert review records, and generate rule confidence through an exponential decay function.

7. The method for environmental risk assessment of transmission lines integrating expert knowledge as described in claim 6, characterized in that, The expert review record in S4 includes the regular expression, initial confidence level, and last update time fields.

8. The method for environmental risk assessment of transmission lines integrating expert knowledge as described in claim 1, characterized in that, S5 specifically includes: S501. Combine the rule confidence level with the evidence from the case library to form a multi-source evidence set, and construct the geometric structure of the evidence space according to the distribution of evidence types. S502. Calculate the distance between evidence points in the spatial geometric structure to form evidence clusters, and generate a risk probability surface by extracting the core evidence clusters. S503. Calculate the risk probability by integrating the center trajectory based on the risk probability surface, and obtain the source address by marking the coordinate index of the original evidence in space. S504. Write the risk probability, traceability address, and evidence cluster into a dynamic data structure to generate a risk assessment report.

9. A computer device comprising a processor and a memory, the memory storing a computer program, wherein the processor, when executing the program, implements the steps of a transmission line environmental risk assessment method incorporating expert knowledge as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, the program, when executed by a processor, implementing the steps of a transmission line environmental risk assessment method incorporating expert knowledge as described in any one of claims 1 to 8.

Images