Semantic network generation method based on airport flight area operation data under large flight delay

Abstract

The invention provides a semantic network generation method based on airport flight area operation data under large flight delay, and belongs to the technical field of artificial intelligence. The technical problems that in existing airport operation data processing, information extraction is inaccurate, entity ambiguity resolution is difficult, and a structured semantic reasoning system is lacked are solved, and the technical scheme includes the steps that firstly, an airport operation data text is extracted and partitioned, and a large language model is combined with a few field samples to extract related examples; secondly, vectorizing an entity through an embedded model, disambiguating the entity ambiguity by adopting a fuzzy membership matrix, and constructing a graph structure taking the entity and the claim as nodes and taking the relationship frequency as an edge weight; then, a community discovery algorithm is applied to divide the sub-graphs, and a structured abstract containing a core entity, a relational network and a claim is generated; and finally, constructing a multi-level semantic abstract to support situation analysis and decision. According to the method, the intelligent level of decision-making such as flight area situation analysis under large-area flight delay can be improved.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph construction technology, and in particular to a method for generating semantic networks based on airport flight area operation data under conditions of long flight delays, which is applicable to aviation operation situation analysis and decision support systems. Background Technology

[0002] With the continuous development of the air transport industry and the increasing number of flights, the complexity of airport operations has also increased. Frequent large-scale flight delays have brought numerous adverse effects to passengers, airlines, and airports. The causes of flight delays are varied, including severe weather, air traffic control, air traffic management, aircraft malfunctions, airport facility problems, passenger strikes, or unforeseen events. These factors intertwine, making the operational situation of the airport's flight zone extremely complex.

[0003] Against this backdrop, accurately analyzing the operational status of airport flight zones is crucial for addressing large-scale flight delays. However, traditional airport operational status analysis methods have significant limitations. Many methods rely on a single data source or simple statistical models, failing to comprehensively reflect the complex operational conditions of the flight zone. For example, some methods analyze only historical flight data, ignoring dynamic factors such as real-time weather conditions and airspace traffic, leading to biased judgments of the current operational status and failing to provide timely and effective support for decision-making. Furthermore, rule-based reasoning methods lack flexibility in the face of complex and ever-changing realities, making it difficult to quickly adapt to new scenarios.

[0004] The paper "Construction Method of Flight Situation Knowledge Graph and its Question Answering System" points out that existing flight situation analysis technologies also reveal many problems when dealing with complex data and multiple factors in scenarios of large-scale flight delays. On the one hand, data fusion capabilities are insufficient. Flight area operations involve multi-source heterogeneous data such as flight plans, meteorological information, equipment status, and personnel deployment. Existing technologies struggle to efficiently fuse this data, thus failing to form a comprehensive and accurate description of the operational situation. On the other hand, the ability to model complex relationships is limited. Intricate relationships exist between flights and between flights and airport resources. Traditional methods cannot deeply explore these relationships, resulting in a lack of depth and accuracy in the analysis of the operational situation. For example, when predicting the propagation of flight delays, the temporal and spatial coupling relationships between different flights are not fully considered, leading to significant deviations between the predicted results and the actual situation. Furthermore, existing analysis methods also lack real-time performance and scalability, making it difficult to meet the needs for rapid decision support in the event of large-scale flight delays. Summary of the Invention

[0005] The purpose of this invention is to provide a semantic network generation method based on airport flight area operation data under long flight delays, which solves the problems of inaccurate information extraction, difficulty in resolving entity ambiguity, and lack of structured semantic reasoning system in existing airport operation data processing, and effectively improves the intelligence level of decision-making such as flight delay analysis and resource scheduling.

[0006] To achieve the above-mentioned objectives, this invention is implemented through the following measures: a semantic network generation method based on airport flight area operation data under conditions of significant flight delays, comprising the following steps:

[0007] S1: Extract text content C from the original airport operation data document D and divide it into text blocks B of length l. The size of text block B is balanced between processing efficiency and information retention.

[0008] S2: For each text block B, extract named entities N, the relationships R between them, and related factual claims F using a large language model, and generate a brief description S for each element;

[0009] S3: Use an embedding model to vectorize all extracted entities and construct a fuzzy membership matrix M. f Entities with similar semantics are clustered and merged based on a similarity threshold;

[0010] S4: Using the standardized entities S after clustering in step S3 N The factual claims extracted in step S2 F and relation set S R Constructing a semantic graph G sem The elements of the entity set and the fact claim set are used as nodes in graph G. Then, the relation set is traversed to connect each entity n. i and n j Relationships, create node n in the graph i Pointing to n j The directed edges are finally used to reflect the frequency of the relation being mentioned in the corpus through edge weights;

[0011] S5: The Louvain community detection algorithm is used to semantically divide the graph, generating multiple interconnected but semantically focused subgraphs. Based on each subgraph, a structured summary is generated, covering its core entities, relationship networks, and factual claims.

[0012] S6: Apply a community detection algorithm to the constructed knowledge graph to divide it into multiple semantically related subgraphs G. i Based on the community structure, it generates a summary covering nodes, relationships, and claims. Utilizing the multi-level semantic summary derived from the community structure, it supports the system in performing layer-by-layer question-and-answer and semantic reasoning on the airport flight operation status, thereby achieving structured and interpretable operational status analysis capabilities.

[0013] Furthermore, the specific steps of step S3 are as follows:

[0014] Step S3.1: Extract the entity set S from the flight area corpus. N ={n1,n2,…,n k}, where k is the number of entities, is input into a pre-trained text vectorization model. Using the encoding module in the text vectorization model, contextual semantic analysis and feature extraction are performed on the entity text, and a mapping operation n is executed. i =f emb (v i This generates each entity n. i semantic vector v i This forms a semantic vector set V. sem ={v1,v2,…,v k}, wherein the text vectorization model is a pre-trained language model based on the Transformer architecture, and the mapping function f emb Using the parameters learned by the pre-trained model, entity text is converted into semantic embedding vectors in a high-dimensional space;

[0015] Step S3.2: Based on the semantic vector set V generated in step S3.1 sem For all entity pairs (n) i ,n j ), by calculating the semantic vector v i With v j cosine similarity The cosine similarity matrix between entities is obtained, and then transformed using a Gaussian kernel function to construct the fuzzy membership matrix M. f =[μ ij The conversion formula is:

[0016]

[0017] Where μ ij Represents entity n i For the degree of membership belonging to semantic cluster j, the temperature parameter σ is used to adjust the fuzziness of the membership degree. The larger the value of σ, the smoother the distribution of membership degree.

[0018] Step S3.3: Pre-set the membership threshold τ, and traverse the fuzzy membership matrix M. f For those satisfying μ ij >τ entity pairs (n i ,n j ), and merge them into a unified node C. N , forming entity cluster C N ={n i |ni ∈S N ,μ ij >τfor n j ∈C N For each entity cluster C N The summarization aggregation function Agg(·) based on a large language model is used to generate the original description set of all entities within the cluster. The process involves semantic extraction and fusion operations to generate a unified description L. N The calculation formula is as follows:

[0019]

[0020] Furthermore, the specific steps of step S5 are as follows:

[0021] Step S5.1: Denote the semantic graph constructed in step S4 as G = (V, E), where the node set V contains entities and factual claims, the edge set E represents the semantic relationships between entities, and the edge e represents the semantic relationships between entities. ij The weight of ∈E is w ij This reflects the frequency of relational mentions and sets key parameters for the community detection algorithm, including the similarity threshold t. sim Used to determine the degree of node association, and to limit the number of iterations N. max This prevents the algorithm from getting stuck in an infinite loop;

[0022] Step S5.2: The Louvain community detection algorithm is used to perform semantic partitioning on the graph G. The algorithm continuously optimizes the modular function f. mod To identify the community structure, the modular function Q is calculated using the following formula:

[0023]

[0024] Where m is the total number of edges in the graph, A ij It is an adjacency matrix element, u i and u j These are the degrees of nodes i and j, respectively, δ(c i ,c j The function Q is an indicator function. It is 1 when nodes i and j belong to the same community c, and 0 otherwise. The algorithm forms communities by merging nodes locally and iteratively updates the community structure until the modular function Q no longer significantly improves or reaches the iteration limit N. max Ultimately, the graph is divided into multiple interconnected subgraphs G1 = (V1, E1), G2 = (V2, E2), ..., G n =(V n E n ),in and

[0025] Step S5.3: For each subgraph G i =(V i E i Extract the core entity set. These core entities are nodes with high connectivity and important semantic roles in the subgraph; extract the core relation set. The core relationship connects core entities and has a weight w. ij Edges exceeding the threshold ω are used to extract the main claim set.

[0026] Step S5.4: Based on each subgraph G i Extracting core elements to generate structured summaries S str The abstract is organized according to the logical structure of "core entity - core relationship - main claim". Natural language generation technology is used to generate the core entity set. Core Relationship Set and set of key factual claims The key information is transformed into a concise and coherent text description, ensuring that the summary covers the complete semantic network and core information within the subgraph.

[0027] Furthermore, the specific steps of step S6 are as follows:

[0028] Step S6.1: Divide the multiple subgraphs obtained in step S5 into G1 = (V1, E1), G2 = (V2, E2), ..., G n =(V n E n The set of subgraphs G = {G1, G2, ..., G} constitutes the set of subgraphs G = {G1, G2, ..., G}. n Define the inter-subgraph affinity matrix T, where the matrix elements T ij Subgraph G i and G j The strength of the correlation between them is calculated as follows:

[0029]

[0030] Where E ij For subgraph G i With G j The set of edges connecting w pq For edge weights, |V i ∩V j | indicates the number of overlapping nodes in the two subgraphs;

[0031] Step S6.2: Based on the flight area situation analysis results, process the situation analysis data of key areas and operational elements of the flight area to form a structured summary S of a single sub-map. str ;

[0032] Step S6.3: Set the abstract hierarchy parameter λ, starting from the bottom layer, with the abstract S of a single subgraph. str As a base layer summary, at the intermediate level, the correlation strength T is selected based on the correlation matrix T between subgraphs. ij Higher than the preset threshold t pre subgraph pairs (G) i G j Based on the operational logic and situational change patterns of the flight area, an information fusion algorithm is used to integrate the core information related to the flight area situational analysis in the two sub-graphs, generating a comprehensive summary containing key situational information from both sub-graphs. and At the top level, all closely related subgraph summaries are deeply integrated to form a global summary covering the core content of the entire flight area situation;

[0033] Step S6.4: Construct the question-answering reasoning model M qa Supported by multi-layer semantic summarization, for the situation query question q input by the user, the semantics of the question are analyzed by natural language processing technology, and key entities and relationships related to the situation of the flight area are extracted from it;

[0034] Step S6.5: Perform matching retrieval on the extracted key entities and relationships within the multi-layered summarization system. If matching information is found in the lower-level summary, return detailed situation analysis results and related analyses directly. If no matching information is found, extend the search to the upper-level summary, combining inter-subgraph correlation information and flight zone operation rules to perform situational reasoning analysis, ultimately generating a structured and interpretable response to the flight zone situational analysis. a =M qa (q).

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] 1. This invention significantly improves the accuracy of airport terminology understanding by combining a large language model with an information extraction framework based on few domain examples. It solves the problems of fuzzy entity recognition and fragmented relation extraction in traditional methods. At the same time, it introduces a fuzzy membership matrix to cluster and disambiguate vectorized entities, effectively addressing the challenge of diverse entity representations in airport corpora. It achieves accurate merging of synonyms and near-synonyms, laying a standardized data foundation for subsequent semantic reasoning. Compared with traditional rule matching methods, the entity disambiguation accuracy is improved by more than 30%.

[0037] 2. This invention constructs a semantic graph with entities and factual claims as nodes and relationship frequency as weights, transforming airport operation data into a computable graph structure model. Through graph representation, it breaks down the barriers of traditional data silos, enabling semantic association modeling of multi-source information such as flight dynamics, resource scheduling, and weather conditions. This graph not only supports the visual tracing of entity relationships but also quantifies relationship strength through edge weights, providing interpretable support for causal analysis of complex operational situations, thus upgrading airport operation logic from "data stacking" to "semantic network reasoning".

[0038] 3. This study employs the Louvain community detection algorithm for semantic subgraph partitioning, deconstructing the overall airport operation status into analytical units with clear semantic orientations, such as flight delay analysis, gate resource conflicts, and runway capacity bottlenecks. Each semantic subgraph constructs a structured summary containing core entities, relationship networks, and core claims, achieving hierarchical knowledge representation from macro-operation status to micro-business scenarios. This multi-level semantic abstraction mechanism supports the rapid location of key contradictions. For example, by using association analysis to identify the chain-like influence relationship between weather conditions, flight scheduling, and gate allocation within a specific time period, combined with a historical case knowledge base, the system can predict the development of the operation status and demonstrate significant advantages in multi-factor scheduling scheme analysis.

[0039] 4. Based on semantic summaries generated from a knowledge graph community structure, a closed-loop framework of "data-model-reasoning" is constructed for airport operation and management. In the field of flight delay management, the similar case retrieval mechanism of the knowledge graph can automatically associate the responsible parties, scope of impact, and handling plans of historical delay events, and generate multi-dimensional handling suggestions using a probabilistic reasoning model. In the airport resource scheduling scenario, the spatiotemporal coupling relationship of resources such as gates, jet bridges, and ground staff is quantitatively evaluated using a subgraph relationship network analysis model, and then a dynamic optimization configuration plan is output. This method effectively overcomes the local optimization bottleneck of traditional statistical models, realizes semantic-level collaborative decision-making across departments and businesses, and promotes the transformation of airport operation management paradigm from experience-driven to data-intelligent decision-making. Attached Figure Description

[0040] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used to explain the invention but do not constitute a limitation thereof.

[0041] Figure 1 This is a diagram illustrating the overall framework of the semantic network generation method based on airport flight area operation data under long flight delays as described in this invention.

[0042] Figure 2 This is an example diagram of the semantic network generation method based on airport flight area operation data under long flight delays according to the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0044] Example 1

[0045] See Figures 1 to 2 This embodiment provides a technical solution: a semantic network generation method based on airport flight area operation data under long flight delays. Taking airport flight area data as an example, the method includes the following steps from data loading to obtaining analysis results:

[0046] S1: Extract text content C from the original airport operation data document D and divide it into text blocks B of length l. The size of the text blocks needs to be balanced between processing efficiency and information preservation: longer blocks can reduce the number of large model calls and are suitable for macro-level understanding, while shorter blocks help to capture key details and facilitate high-precision information extraction.

[0047] S2: For each text block, extract named entities N, the relationships between them R, and related factual claims F using a large language model, and generate a brief description S for each element. By constructing a large number of small sample examples specific to the flight delay domain, the model's ability to understand terms is improved.

[0048] N = {Flight CA123, Runway 18L} (9)

[0049] R = {Flight CA123, dependent on runway 18L} (10)

[0050] F = {Due to visibility being below standard, runway 18L was closed, causing MU5289 to be delayed by approximately 35 minutes}(11)

[0051] S = {Flight CA123, one Air China flight}(12)

[0052] S3: Considering the ambiguity and diversity of entity representations in airport flight area corpora, we first use an embedding model to vectorize all extracted entities, and then construct a fuzzy membership matrix M. f Entities with similar semantics are clustered and merged based on a similarity threshold;

[0053] S4: Using the standardized entities S after clustering in S3 N Factual claims extracted by S2 F and relation set S R Constructing a semantic graph G sem Next, the elements of the entity set and the fact claim set are used as nodes in graph G. Then, the relation set is traversed to connect each entity n. i and nj Relationships, create node n in the graph i Pointing to n j The directed edges are used to reflect the frequency of the relation being mentioned in the corpus through edge weights. This graph, as a semantic network, provides structural support for subsequent information organization and reasoning.

[0054] S5: The Louvain community detection algorithm is used to semantically divide the graph, generating multiple interconnected but semantically focused subgraphs. Based on each subgraph, structured summary content is further generated, covering its core entities, relationship networks, and main claims.

[0055] S6: Apply a community detection algorithm to the constructed knowledge graph to divide it into multiple semantically related subgraphs G. i Based on the community structure, a summary covering nodes, relationships, and claims is generated. Finally, by utilizing the multi-level semantic summary derived from the community structure, the system supports layer-by-layer question answering and semantic reasoning of airport flight operation status, thereby achieving structured and interpretable operation status analysis capabilities.

[0056] Specifically, the steps of step S3 are as follows:

[0057] Step S3.1: Extract the entity set S from the airport flight area data. N ={n1,n2,…,n k}, where k is the number of entities, is input into a pre-trained text vectorization model. Using the encoding module in the text vectorization model, contextual semantic analysis and feature extraction are performed on the entity text, and a mapping operation n is executed. i =f emb (v i This generates each entity n. i semantic vector v i This forms a semantic vector set V. sem ={v1,v2,…,v k}, wherein the text vectorization model is a pre-trained language model based on the Transformer architecture, and the mapping function f emb Using the parameters learned by the pre-trained model, entity text is converted into semantic embedding vectors in a high-dimensional space;

[0058] S N ={Flight CA123, Flight MU5289,...,Taxiway C1,Taxiing Conflict} k (13)

[0059]

[0060] Step S3.2: Based on the semantic vector set V generated in S3.1sem For all entity pairs (n) i ,n j ), by calculating the semantic vector v i With v j cosine similarity The cosine similarity matrix between entities is obtained, and then transformed using a Gaussian kernel function to construct the fuzzy membership matrix M. f =[μ ij The conversion formula is:

[0061]

[0062] Where μ ij Represents entity n i For the degree of membership belonging to semantic cluster j, the temperature parameter σ is used to adjust the fuzziness of the membership degree. The larger the value of σ, the smoother the distribution of membership degree. ij As shown in formula (16):

[0063]

[0064] Step S3.3: Pre-set the membership threshold τ, and traverse the fuzzy membership matrix M. f For those satisfying μ ij >τ entity pairs (n i ,n j ), and merge them into a unified node C. N , forming entity cluster C N ={n i |n i ∈S N ,μ ij >τfor n j ∈C N For each entity cluster C N The summarization aggregation function Agg(·) based on a large language model is used to generate the original description set of all entities within the cluster. The process involves semantic extraction and fusion operations to generate a unified description L. N The calculation formula is as follows:

[0065]

[0066] L N ={The flights involved (CA123, MU5289) are the key targets in the current flight conflict} (18)

[0067] Specifically, the steps of step S5 are as follows:

[0068] Step S5.1: Denote the semantic graph constructed in step S4 as G = (V, E), where the node set V contains entities and factual claims, the edge set E represents the semantic relationships between entities, and the edge e represents the semantic relationships between entities. ij The weight of ∈E is w ij This reflects the frequency of relational mentions and sets key parameters for the community detection algorithm, including the similarity threshold t. sim Used to determine the degree of node association, and to limit the number of iterations N. max To prevent the algorithm from getting stuck in an infinite loop, the node set V and the edge set E are shown in formulas (19) and (20):

[0069] V = {Entity: Flight CA123, Runway 09L; Factual claim: CA123 waiting to taxi, Runway 09L closed}(19)

[0070] E = {(A123, Runway 09L): Waiting to be used, Weight = 5} (20)

[0071] Step S5.2: The Louvain community detection algorithm is used to perform semantic partitioning on the graph G. The algorithm continuously optimizes the modular function f. mod To identify the community structure, the modular function Q is calculated using the following formula:

[0072]

[0073] Where m is the total number of edges in the graph, A ij It is an adjacency matrix element, u i and u j These are the degrees of nodes i and j, respectively, δ(c i ,c j The function Q is an indicator function. It is 1 when nodes i and j belong to the same community c, and 0 otherwise. The algorithm forms communities by merging nodes locally and iteratively updates the community structure until the modular function Q no longer significantly improves or reaches the iteration limit N. max Ultimately, the graph is divided into multiple interconnected subgraphs G1 = (V1, E1), G2 = (V2, E2), ..., G n =(V n E n ),in and The subgraph is shown in formula (22):

[0074]

[0075] Step S5.3: For each subgraph G i =(V i E i Extract the core entity set. These core entities are nodes with high connectivity and important semantic roles in the subgraph; extract the core relation set. The core relationship connects core entities and has a weight w. ij Edges exceeding the threshold ω are used to extract the main claim set. Key factual claims are those closely related to core entities and relationships and have a significant impact on the operational situation.

[0076] Step S5.4: Based on each subgraph G i Extracting core elements to generate structured summaries S str The abstract is organized according to the logical structure of "core entity - core relationship - main claim". Natural language generation technology is used to generate the core entity set. Core Relationship Set and set of key factual claims The key information is transformed into concise and coherent text descriptions, ensuring that the summary covers the complete semantic network and core information within the subgraph, providing clear and accurate semantic support for subsequent operational situation reasoning. Structured summary S str As shown in formula (23):

[0077]

[0078] Specifically, the steps of step S6 are as follows:

[0079] Step S6.1: Divide the multiple subgraphs obtained in step S5 into G1 = (V1, E1), G2 = (V2, E2), ..., G n =(V n E n The set of subgraphs G = {G1, G2, ..., G} constitutes the set of subgraphs G = {G1, G2, ..., G}. n Define the inter-subgraph affinity matrix T, where the matrix elements T ij Subgraph G i and G j The strength of the correlation between them is calculated as follows:

[0080]

[0081] Where E ij For subgraph G i With G j The set of edges connecting w pq For edge weights, |V i ∩V j | indicates the number of overlapping nodes in the two subgraphs;

[0082] Step S6.2: Based on the flight area situation analysis results, process the situation analysis data of key areas and operational elements of the flight area to form a structured summary S of a single sub-map.str Specifically, the situational analysis data of each key area and operational element of the flight zone are structured to give them a clear format and standardized expression, thereby constructing a basic layer summary.

[0083] Step S6.3: Set the abstract hierarchy parameter λ, starting from the bottom layer, with the abstract S of a single subgraph. str As a base layer summary, at the intermediate level, the correlation strength T is selected based on the correlation matrix T between subgraphs. ij Higher than the preset threshold t pre subgraph pairs (G) i G j Based on the operational logic and situational change patterns of the flight area, an information fusion algorithm is used to integrate the core information related to the flight area situational analysis in the two sub-graphs, generating a comprehensive summary containing key situational information from both sub-graphs. and At the top level, all closely related subgraph summaries are deeply integrated to form a global summary covering the core content of the entire flight area situation, ultimately achieving a multi-level, structured expression of the complex situation of the flight area.

[0084] Step S6.4: Construct the question-answering reasoning model M qa Supported by multi-layer semantic summarization, it realizes layer-by-layer question answering and semantic reasoning of the airport flight area situation analysis results. For the situation query question q input by the user, the semantics of the question are parsed through natural language processing technology, and key entities and relationships related to the flight area situation are extracted from it.

[0085] Step S6.5: Perform matching retrieval on the extracted key entities and relationships within the multi-layered summarization system. If matching information is found in the lower-level summary, return detailed situation analysis results and related analyses directly. If no matching information is found, extend the search to the upper-level summary, combining inter-subgraph correlation information and flight zone operation rules to perform situational reasoning analysis, ultimately generating a structured and interpretable response to the flight zone situational analysis. a =M qa (q) to meet users' requirements for in-depth query of the airport flight area's operational status;

[0086] Why is q=A123 delayed? (25)

[0087] f a =Due to the T3 taxiway being occupied by ground support vehicles, A123 was unable to taxi normally, causing a delay. (26)

[0088] Example 2

[0089] See Figures 1 to 2This embodiment provides a technical solution: a semantic network generation method based on airport flight area operation data under long flight delays. Taking airport flight area data as an example, the method includes the following steps from data loading to obtaining analysis results:

[0090] S1: Extract text content C from the original airport operation data document D and divide it into text blocks B of length l. The size of the text blocks is balanced between processing efficiency and information retention.

[0091] S2: For each text block, extract named entities N, the relationships between them R, and the related factual claims F using a large language model, and generate a brief description S for each element;

[0092] N = {Flight CZ3101, Gate B23, Jet Bridge L05}(27)

[0093] R = {Flight CZ3101, requires parking at gate B23; Flight CZ3101, requires use of jet bridge L05}(28)

[0094]

[0095] S = {Flight CZ3101, a China Southern Airlines flight}(30)

[0096] S3: The entity set S extracted from the airport flight area data. N ={n1,n2,…,n k The embedding model is used to vectorize all extracted entities to obtain a semantic vector set V. sem ={v1,v2,…,v k Then construct a fuzzy membership matrix M. f By using a similarity threshold, semantically similar entities are clustered and merged to generate a unified description L. N ;

[0097] S N ={Flight CZ3101, Flight HU7890, ..., Ground support equipment, Equipment failure} k (31)

[0098]

[0099] S4: Using the standardized entities S after clustering in step S3 N The factual claims extracted in step S2 F and relation set S R Constructing a semantic graph G sem Next, the elements of the entity set and the fact claim set are used as nodes in graph G. Then, the relation set is traversed to connect each entity n. i and nj Relationships, create node n in the graph i Pointing to n j The directed edges are finally used to reflect the frequency of the relation being mentioned in the corpus through edge weights. The node set V and the edge set E are shown in formulas (34) and (35):

[0100]

[0101] S5: The Louvain community detection algorithm is used to semantically partition the graph, generating multiple interconnected but semantically focused subgraphs, as shown in formula (36). Based on each subgraph, a structured summary content S is generated. str As shown in formula (37), it encompasses its core entities, relational networks, and factual claims;

[0102]

[0103] S6: Apply a community detection algorithm to the constructed knowledge graph to divide it into multiple semantically related subgraphs G. i Based on the community structure, it generates a summary covering nodes, relationships, and claims. Utilizing the multi-level semantic summary derived from the community structure, it supports the system in performing layer-by-layer question-and-answer and semantic reasoning on the airport flight operation status, thereby achieving structured and interpretable operational status analysis capabilities.

[0104] q = What impact does the weather have on the airport? (38)

[0105]

[0106] Example 3

[0107] See Figures 1 to 2 This embodiment provides a technical solution: a semantic network generation method based on airport flight area operation data under long flight delays. Taking airport flight area data as an example, the method includes the following steps from data loading to obtaining analysis results:

[0108] S1: Extract text content C from the original airport operation data document D and divide it into text blocks B of length l. The size of the text blocks is balanced between processing efficiency and information retention.

[0109] S2: For each text block, extract named entities N, the relationships between them R, and the related factual claims F using a large language model, and generate a brief description S for each element;

[0110] N = {Flight MF8123, Approach Control Area Z, Tower Control System}(40)

[0111] R = {Flight MF8123, located in Approach Control Area Z; Flight MF8123, receiving command from the Tower Control System}(41)

[0112]

[0113] S = {Flight MF8123, a Xiamen Airlines flight}(43)

[0114] S3: The entity set S extracted from the airport flight area data. N ={n1,n2,…,n k The embedding model is used to vectorize all extracted entities to obtain a semantic vector set V. sem ={v1,v2,…,v k Then construct a fuzzy membership matrix M. f By using a similarity threshold, semantically similar entities are clustered and merged to generate a unified description L. N ;

[0115] S N ={Flight MF8123, Flight SC4567,..., Approach Control Area Z, Software Failure} k (44)

[0116]

[0117] S4: Using the standardized entities S after clustering in step S3 N The factual claims extracted in step S2 F and relation set S R Constructing a semantic graph G sem Next, the elements of the entity set and the fact claim set are used as nodes in graph G. Then, the relation set is traversed to connect each entity n. i and n j Relationships, create node n in the graph i Pointing to n j The directed edges are finally weighted to reflect the frequency of the relation being mentioned in the corpus. The node set V and the edge set E are shown in formulas (47) and (48):

[0118]

[0119] S5: The Louvain community detection algorithm is used to semantically partition the graph, generating multiple interconnected but semantically focused subgraphs, as shown in formula (49). Based on each subgraph, a structured summary content S is generated. str As shown in formula (50), it encompasses its core entities, relational networks, and factual claims;

[0120]

[0121] S6: Apply a community detection algorithm to the constructed knowledge graph to divide it into multiple semantically related subgraphs G. i Based on the community structure, it generates a summary covering nodes, relationships, and claims. Utilizing the multi-level semantic summary derived from the community structure, it supports the system in performing layer-by-layer question-and-answer and semantic reasoning on the airport flight operation status, thereby achieving structured and interpretable operational status analysis capabilities.

[0122] q = What is the current operational status of the entire flight area? (51)

[0123]

[0124] All the analysis questions in the dataset were put into the model to obtain the analysis results, as shown in Table 1 below. The performance of the situation analysis results was evaluated using real results, and the accuracy rate of the analysis was 85.5%, which is significantly better than traditional situation analysis methods.

[0125] Table 1

[0126]

[0127] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for generating semantic networks based on airport flight area operation data under conditions of significant flight delays, characterized in that, Includes the following steps: S1: Extract text content C from the original airport operation data document D and divide it into text blocks B of length l. The size of text block B is balanced between processing efficiency and information retention. S2: For each text block B, extract named entities N, the relationships R between them, and related factual claims F using a large language model, and generate a brief description S for each element; S3: Use an embedding model to vectorize all extracted entities and construct a fuzzy membership matrix M. f Entities with similar semantics are clustered and merged based on a similarity threshold; S4: Using the standardized entities S after clustering in step S3 N The factual claims extracted in step S2 F and relation set S R Constructing a semantic graph G sem The elements of the entity set and the fact claim set are used as nodes in graph G. Then, the relation set is traversed to connect each entity n. i and n j Relationships, create node n in the graph i Pointing to n j The directed edges are finally used to reflect the frequency of the relation being mentioned in the corpus through edge weights; S5: The Louvain community detection algorithm is used to semantically divide the graph, generating multiple interconnected but semantically focused subgraphs. Based on each subgraph, a structured summary is generated, covering its core entities, relationship networks, and factual claims. S6: Apply a community detection algorithm to the constructed knowledge graph to divide it into multiple semantically related subgraphs G. i Based on the community structure, it generates a summary covering nodes, relationships, and claims. Utilizing the multi-level semantic summary derived from the community structure, it supports the system in performing layer-by-layer question-and-answer and semantic reasoning on the airport flight operation status, thereby achieving structured and interpretable operational status analysis capabilities.

2. The semantic network generation method based on airport flight area operation data under long flight delays as described in claim 1, characterized in that, Step S3 includes the following steps: Step S3.1: Extract the entity set S from the flight area corpus. N ={n1,n2,…,n k }, where k is the number of entities, is input into a pre-trained text vectorization model. Using the encoding module in the text vectorization model, contextual semantic analysis and feature extraction are performed on the entity text, and a mapping operation n is executed. i =f emb (v i This generates each entity n. i semantic vector v i This forms a semantic vector set V. sem ={v1,v2,…,v k }, wherein the text vectorization model is a pre-trained language model based on the Transformer architecture, and the mapping function f emb Using the parameters learned by the pre-trained model, entity text is converted into semantic embedding vectors in a high-dimensional space; Step S3.2: Based on the semantic vector set V generated in step S3.1 sem For all entity pairs (n) i ,n j ), by calculating the semantic vector v i With v j cosine similarity The cosine similarity matrix between entities is obtained, and then transformed using a Gaussian kernel function to construct the fuzzy membership matrix M. f =[μ ij The conversion formula is: Where μ ij Represents entity n i For the degree of membership belonging to semantic cluster j, the temperature parameter σ is used to adjust the fuzziness of the membership degree. The larger the value of σ, the smoother the distribution of membership degree. Step S3.3: Pre-set the membership threshold τ, and traverse the fuzzy membership matrix M. f For those satisfying μ ij >τ entity pairs (n i ,n j ), and merge them into a unified node C. N , forming entity cluster C N ={n i |n i ∈S N ,μ ij >τfor n j ∈C N For each entity cluster C N The summarization aggregation function Agg(·) based on a large language model is used to generate the original description set of all entities within the cluster. The process involves semantic extraction and fusion operations to generate a unified description L. N The calculation formula is as follows:

3. The semantic network generation method based on airport flight area operation data under long flight delays as described in claim 1, characterized in that, Step S5 includes the following steps: Step S5.1: Denote the semantic graph constructed in step S4 as G = (V, E), where the node set V contains entities and factual claims, the edge set E represents the semantic relationships between entities, and the edge e represents the semantic relationships between entities. ij The weight of ∈E is w ij This reflects the frequency of relational mentions and sets key parameters for the community detection algorithm, including the similarity threshold t. sim Used to determine the degree of node association, and to limit the number of iterations N. max This prevents the algorithm from getting stuck in an infinite loop; Step S5.2: The Louvain community detection algorithm is used to perform semantic partitioning on the graph G. The algorithm continuously optimizes the modular function f. mod To identify the community structure, the modular function Q is calculated using the following formula: Where m is the total number of edges in the graph, A ij It is an adjacency matrix element, u i and u j These are the degrees of nodes i and j, respectively, δ(c i ,c j The function Q is an indicator function. It is 1 when nodes i and j belong to the same community c, and 0 otherwise. The algorithm forms communities by merging nodes locally and iteratively updates the community structure until the modular function Q no longer significantly improves or reaches the iteration limit N. max Ultimately, the graph is divided into multiple interconnected subgraphs G1 = (V1, E1), G2 = (V2, E2), ..., G n =(V n E n ),in and Step S5.3: For each subgraph G i =(V i E i Extract the core entity set. Extract the core relationship set The core relationship connects core entities and has a weight w. ij Edges exceeding the threshold ω are used to extract the main claim set. Step S5.4: Based on each subgraph G i Extracting core elements to generate structured summaries S str The abstract is organized according to the logical structure of "core entity - core relationship - main claim". The core entity set is generated using natural language processing technology. Core Relationship Set and set of key factual claims The key information is transformed into a concise and coherent text description, ensuring that the summary covers the complete semantic network and core information within the subgraph.

4. The semantic network generation method based on airport flight area operation data under long flight delays as described in claim 1, characterized in that, Step S6 includes the following steps: Step S6.1: Divide the multiple subgraphs obtained in step S5 into G1 = (V1, E1), G2 = (V2, E2), ..., G... n =(V n E n The set of subgraphs G = {G1, G2, ..., G} constitutes the set of subgraphs G = {G1, G2, ..., G}. n Define the inter-subgraph affinity matrix T, where the matrix elements T ij Subgraph G i and G j The strength of the correlation between them is calculated as follows: Where E ij For subgraph G i With G j The set of edges connecting w pq For edge weights, |V i ∩V j | indicates the number of overlapping nodes in the two subgraphs; Step S6.2: Based on the flight area situation analysis results, process the situation analysis data of key areas and operational elements of the flight area to form a structured summary S of a single sub-map. str ; Step S6.3: Set the abstract hierarchy parameter λ, starting from the bottom layer, with the abstract S of a single subgraph. str As a base layer summary, at the intermediate level, the correlation strength T is selected based on the correlation matrix T between subgraphs. ij Higher than the preset threshold t pre subgraph pairs (G) i G j Based on the operational logic and situational change patterns of the flight area, an information fusion algorithm is used to integrate the core information related to the flight area situational analysis in the two sub-graphs, generating a comprehensive summary containing key situational information from both sub-graphs. and At the top level, all closely related subgraph summaries are deeply integrated to form a global summary covering the core content of the entire flight area situation; Step S6.4: Construct the question-answering reasoning model M qa Supported by multi-layer semantic summarization, for the situation query question q input by the user, the semantics of the question are analyzed by natural language processing technology, and key entities and relationships related to the situation of the flight area are extracted from it; Step S6.5: Perform matching retrieval on the extracted key entities and relationships within the multi-layered summarization system. If matching information is found in the lower-level summary, return detailed situation analysis results and related analyses directly. If no matching information is found, extend the search to the upper-level summary, combining inter-subgraph correlation information and flight zone operation rules to perform situational reasoning analysis, ultimately generating a structured and interpretable response to the flight zone situational analysis. a =M qa (q).

Images