Civil aviation control operation safety risk assessment method based on text mining and network analysis

Abstract

The invention discloses a civil aviation control operation safety risk assessment method based on text mining and network analysis, and relates to the field of civil aviation control operation safety risk detection, and the method comprises the following steps: S1, collecting domestic and overseas air traffic control operation unsafe event data, and carrying out the preprocessing, including the processing of irrelevant and duplicated data, and the processing of the unsafe event data; constructing an air traffic control operation safety risk management stop word and segmented word library; s2, based on the processed data, a BERTopic topic model is used for identifying civil aviation control operation safety risk topics, model parameters of the BERTopic are modified through topic coherence (TC) and topic diversity (TD) among the topics, and the risk topics are optimized; s3, setting causal relationship network nodes based on the civil aviation control operation safety risk theme, and constructing a causal relationship network to determine the mutual relationship between civil aviation control operation safety risk factors; and S4, identifying high-risk nodes in the civil aviation control operation safety causal relationship network based on topology analysis, determining key cause factors in civil aviation control operation safety events, and proposing a risk control strategy. By adopting the steps, the civil aviation control operation safety risk theme is identified, the causal relationship network is constructed, the risk cause factors are determined, and finally the targeted risk control strategy is put forward, so that stable and safe operation of the air transportation system is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of civil aviation control operation safety risk detection, and in particular to a civil aviation control operation safety risk assessment method based on text mining and network analysis. Background Art

[0002] Safety is the lifeline of the civil aviation industry and the eternal theme of civil aviation. With the rapid growth of aviation demand, airports and airspace are becoming more busy, and the pressure on air traffic control is gradually increasing. Therefore, in-depth research on the safety risks of civil aviation control operations has important scientific significance and application value for ensuring air traffic safety and promoting the efficient development of civil aviation.

[0003] Aviation safety data is diverse and multi-source, among which text data is one of the most important formats, providing detailed information about accidents, real-time data analysis, conclusions, etc. Therefore, text mining is needed to obtain potential information in text data. Although structural subject modeling (STM) and latent Dirichlet allocation (LDA) have begun to be applied in the field of transportation safety, the BERTopic model that can capture the meaning between words is rarely used, especially in the field of civil aviation control and operation safety. At the same time, the current research on aviation accidents mainly focuses on aviation accident report analysis, accident data mining, accident trend prediction, accident cause analysis and risk control, system safety analysis, etc. However, as an aviation transportation system is a complex system, few scholars analyze the risks of civil aviation control and operation from the perspective of complex system theory, and there is a lack of in-depth research on civil aviation control and operation risk assessment methods and emergency management of emergencies.

[0004] Therefore, it is necessary to provide a civil aviation control operation safety risk assessment method based on text mining and network analysis to solve the above problems. Summary of the invention

[0005] The purpose of this invention is to provide a civil aviation control operation safety risk assessment method based on text mining and network analysis, and to construct a new framework combining text mining and complex network analysis to accurately identify the main risk themes and key causal factors in civil aviation control operations, so as to facilitate the smooth and safe operation of the air transportation system.

[0006] To achieve the above object, the present invention provides a method for assessing the safety risk of civil aviation control operations based on text mining and network analysis, comprising the following steps:

[0007] S1: Collect data on unsafe air traffic control operations at home and abroad, and pre-process them, including processing irrelevant and repeated data, and building a stop word and word segmentation vocabulary for air traffic control operation safety risk management;

[0008] S2: Based on the preprocessed data, the BERTopic topic model is used to identify the civil aviation control operation safety risk topics, and the model parameters of BERTopic are modified through the topic coherence (TC) and topic diversity (TD) between each topic to optimize the risk topics;

[0009] S3: Set causal network nodes based on the civil aviation control operation safety risk theme and construct a causal network to determine the interrelationships between civil aviation control operation safety risk factors;

[0010] S4: Identify high-risk nodes in the causal network of civil aviation control operation safety based on topological analysis, determine the key causal factors in civil aviation control operation safety incidents, and propose risk control strategies.

[0011] Preferably, in step S1, the following steps are included:

[0012] S1A: Follow the following principles to collect data on unsafe incidents in civil aviation control operations in 2021: the accident description is detailed and clear; the airline should be a public air transport company, not general aviation or military aviation; aviation accidents caused by special reasons are not considered; there are no repeated incidents.

[0013] S1B: Follow the steps below to transform and clean the data and build a stop word and word segmentation vocabulary for air traffic control operation safety risk management: delete semantically repeated text and numbers and punctuation marks that are not relevant to the study; restore commonly used aviation abbreviations; and delete stop words.

[0014] Preferably, in step S2, the following steps are included:

[0015] S2A: The bert-base-chinese model is used to encode the preprocessed text to achieve text embedding; the newly generated text embedding vector is compared and labeled with the known document embedding vector using the cosine similarity algorithm;

[0016] S2B: Cluster the word embedding vectors based on semantic similarity. First, use UMAP to reduce the dimensionality of BERT embeddings. Second, use HDBSCAN to cluster the embeddings and identify outlier documents.

[0017] S2C: Class-based term frequency-inverse document frequency (C-TF-IDF) and maximum marginal relevance (MMR) are used to extract and merge keywords based on the importance scores of the words and preliminarily determine the text topic. The importance score formula is as follows:

[0018]

[0019] Where T i,crepresents the importance score of word i in semantic cluster c, tf i,c represents the frequency of word i in semantic cluster c, f x represents the frequency of word i in all semantic clusters, and A represents the mean number of words in each semantic cluster.

[0020] S2D: Use visualize_topics to visualize the clustering results and keywords in the form of charts to understand the topic structure more intuitively;

[0021] S2E: We further study the potential hierarchical structure between topics and use cosine similarity to perform correlation analysis on each risk topic. The formula for calculating cosine similarity is as follows:

[0022]

[0023] Among them, A i and B i Represent the embedding vectors of the two topics respectively; the value range of cosθ is (-1,1). The closer the absolute value of cosθ is to 1, the more similar the two topics are; the closer the similarity is to 0, the weaker the relationship between the two topics.

[0024] S2F: Uses dynamic topic model to analyze topic trends;

[0025] S2G: Evaluate and optimize the BERTopic model based on topic coherence (TC) and topic diversity (TD). The calculation formulas for TC and TD are as follows:

[0026]

[0027]

[0028] Among them, NPMI is the normalized mutual information, which represents the co-occurrence between two words; w i 、w j Respectively represent the i-th and j-th keywords in a topic; P(w i ,w j ) indicates w i and w j The probability of co-occurrence in the same document; P(w i ) and P(w j ) represents the probability of a single word appearing in a document; W represents the total number of the first N topic words of all topics extracted; W unique Indicates the number of unique subject terms.

[0029] Preferably, in step S3, the following steps are included:

[0030] S3A: Based on the civil aviation control operation safety risk themes generated in S2, determine the cause nodes and accident nodes of the causal relationship network.

[0031] S3B: Mining association rules between civil aviation control and operation safety risk topics, obtaining the minimum support and minimum confidence thresholds, eliminating weaker relationships between civil aviation control and operation safety risk topics, and finally establishing a causal relationship network connection edge of civil aviation control and operation safety risk based on association relationships. The relevant formulas for association rule mining are as follows:

[0032]

[0033] Among them, Support is the support, which can measure the frequency of a topic appearing in all topics; Confidence is the confidence, which can measure the accuracy of the rule; Lift is the lift, which can measure the strength of the management rule.

[0034] S3C: Draw the causal relationship network of civil aviation control operation safety risk. Based on the accident causality theory and the priority of events, the accident chain is formed, and the average degree and confidence are used as the weights of nodes and edges respectively to establish the causal relationship network of each accident and the total causal relationship network.

[0035] Preferably, in step S4, the following steps are included:

[0036] S4A: Using topological analysis, based on the network characteristics, parameters, and centrality index of the civil aviation control operation safety risk network, high-risk nodes are identified. Among them, the important nodes in the causal network are determined based on the centrality index. The specific index and calculation formula are as follows:

[0037]

[0038] Among them, C D represents degree centrality, which is used to measure the degree of connectivity between a node and all other nodes in the network; g is the number of edges connecting node i; C B represents the betweenness centrality, which is an indicator used to measure the importance of a node as a bridge or key path in the network; sd(i,j,k) is the number of paths passing through i in the shortest path from node j to node k; C E represents the eigenvector centrality, which is used to identify the most influential nodes in the network; c is the proportionality constant; a ij =1, node i is connected to node j, a ij =0, node i is not connected to node j.

[0039] S4B: Comprehensively consider the four aspects of people, machines, environment and management, deeply explore the causes of high-risk nodes in civil aviation control operations, propose safety risk control strategies in technology, processes, training, etc., improve the resilience of the control operation safety system, and ensure the safety and reliability of air transportation.

[0040] Compared with the prior art, the beneficial effect of the present invention is that the present invention constructs a new framework combining text mining and complex network analysis to accurately identify the main risk topics and key causal factors in civil aviation control operations, thereby facilitating the smooth and safe operation of the aviation transportation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a method for assessing the safety risk of civil aviation control operations based on text mining and network analysis of the present invention;

[0042] Figure 2 It is a risk theme visualization diagram of a civil aviation control operation safety risk assessment method based on text mining and network analysis of the present invention;

[0043] Figure 3 It is a risk part theme clustering diagram of a civil aviation control operation safety risk assessment method based on text mining and network analysis of the present invention;

[0044] Figure 4 It is a partial causal relationship network diagram of a civil aviation control operation safety risk assessment method based on text mining and network analysis of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0046] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0047] like Figure 1 As shown, the present invention provides a civil aviation control operation safety risk assessment method based on text mining and network analysis, comprising the following steps:

[0048] S1: Collect domestic and foreign ATC unsafe event data and pre-process them, including processing irrelevant and duplicate data, and constructing stop words and word segmentation word libraries for ATC operation safety risk management; specifically, the following steps are included:

[0049] S1A: Follow the following principles to collect data on unsafe incidents in civil aviation control operations in 2021: the accident description is detailed and clear; the airline should be a public air transport company, not general aviation or military aviation; aviation accidents caused by special reasons are not considered; there are no repeated incidents.

[0050] S1B: Follow the steps below to transform and clean the data and build a stop word and word segmentation vocabulary for air traffic control operation safety risk management: delete semantically repeated text and numbers and punctuation marks that are not relevant to the study; restore commonly used aviation abbreviations; and delete stop words.

[0051] S2: Based on the preprocessed data, the BERTopic topic model is used to identify the civil aviation control operation safety risk topics, and the model parameters of BERTopic are modified through the topic coherence (TC) and topic diversity (TD) between each topic to optimize the risk topics; specifically, the following steps are included:

[0052] S2A: The bert-base-chinese model is used to encode the preprocessed text to achieve text embedding; the newly generated text embedding vector is compared and labeled with the known document embedding vector using the cosine similarity algorithm;

[0053] S2B: Cluster the word embedding vectors based on semantic similarity. First, use UMAP to reduce the dimensionality of BERT embeddings. Second, use HDBSCAN to cluster the embeddings and identify outlier documents.

[0054] S2C: Class-based term frequency-inverse document frequency (C-TF-IDF) and maximum marginal relevance (MMR) are used to extract and merge keywords based on the importance scores of the words and preliminarily determine the text topic. The importance score formula is as follows:

[0055]

[0056] Where T i,c represents the importance score of word i in semantic cluster c, tf i,c represents the frequency of word i in semantic cluster c, f x represents the frequency of word i in all semantic clusters, and A represents the mean number of words in each semantic cluster.

[0057] S2D: Use visualize_topics to visualize the clustering results and keywords in the form of graphs to more intuitively understand the topic structure. Some topics such as Figure 2 As shown;

[0058] S2E: In-depth study of the potential hierarchical structure between topics, and the cosine similarity is used to perform correlation analysis on each risk topic. The hierarchical clustering method is to aggregate multiple similar topics into one topic in the form of Figure 3 As shown; the cosine similarity calculation formula is as follows:

[0059]

[0060] Among them, A i and B i Represent the embedding vectors of the two topics respectively; the value range of cosθ is (-1,1). The closer the absolute value of cosθ is to 1, the more similar the two topics are; the closer the similarity is to 0, the weaker the relationship between the two topics.

[0061] S2F: Use dynamic topic model to analyze topic trends. Divide the data into months, count the number of documents contained in each topic in each time period, and draw a trend chart of topic changes over time based on the statistical results.

[0062] S2G: Evaluate and optimize the BERTopic model based on topic coherence (TC) and topic diversity (TD). The calculation formulas for TC and TD are as follows:

[0063]

[0064] Among them, NPMI is the normalized mutual information, which represents the co-occurrence between two words; w i 、w j Respectively represent the i-th and j-th keywords in a topic; P(w i ,w j ) indicates w i and w j The probability of co-occurrence in the same document; P(w i ) and P(w j ) represents the probability of a single word appearing in a document; W represents the total number of the first N topic words of all topics extracted; W unique Indicates the number of unique subject terms.

[0065] Based on the initial TC and TD results, the model was adjusted as follows:

[0066] (1) Optimize text preprocessing steps, such as better word segmentation and removal of stop words;

[0067] (2) Adjust the clustering parameters of BERTopic, such as n_neighbors and min_cluster_size;

[0068] (3) Increase the number of topics or refine the topics to capture more topic details;

[0069] (4) Adjust the word frequency threshold to filter out high-frequency but meaningless words;

[0070] (5) Repeatedly adjust and evaluate the model until a good balance between TC and TD is achieved.

[0071] Among them, topic coherence is used to measure the semantic consistency between words in each topic. The quantification process mainly relies on statistical methods such as point mutual information and topic consistency index to ensure the clarity and interpretability of the topic. Topic diversity is used to measure the differences between topics generated by the model. It uses methods such as topic overlap measurement and topic uniqueness index to ensure the breadth of topic coverage and the comprehensiveness of information, avoid repeated analysis, and improve the practical value of the model.

[0072] S3: Set causal network nodes based on the civil aviation control operation safety risk theme and construct a causal network to determine the relationship between civil aviation control operation safety risk factors; specifically include the following steps:

[0073] S3A: Based on the civil aviation control operation safety risk themes generated in S2, determine the cause nodes and accident nodes of the causal relationship network.

[0074] S3B: Mining association rules between civil aviation control and operation safety risk topics, obtaining the minimum support and minimum confidence thresholds, eliminating weaker relationships between civil aviation control and operation safety risk topics, and finally establishing a causal relationship network connection edge of civil aviation control and operation safety risk based on association relationships. The relevant formulas for association rule mining are as follows:

[0075]

[0076] Among them, Support is the support, which can measure the frequency of a topic appearing in all topics; Confidence is the confidence, which can measure the accuracy of the rule; Lift is the lift, which can measure the strength of the management rule.

[0077] S3C: Draw the causal relationship network of civil aviation control operation safety risk. Based on the accident causality theory and the priority of events, the accident chain is formed, and the average degree and confidence are used as the weights of nodes and edges respectively to establish the causal relationship network of each accident and the total causal relationship network. Figure 4 Shown is a network diagram of the causal relationship of some safety risks. It can be seen from the diagram that bird strikes and inadequate bird removal work are key causal factors, which can easily lead to risks such as damage to aircraft and waiting in the air and on the ground.

[0078] S4: Based on topological analysis, identify high-risk nodes in the causal relationship network of civil aviation control operation safety, determine the key causal factors in civil aviation control operation safety incidents, and propose risk control strategies; specifically, the following steps are included:

[0079] S4A: Use topological analysis to identify high-risk nodes. Determine the important nodes in the causal network based on the centrality index. The specific index and calculation formula are as follows:

[0080]

[0081] Among them, C D represents degree centrality, which is used to measure the degree of connectivity between a node and all other nodes in the network; g is the number of edges connecting node i; C B represents the betweenness centrality, which is an indicator used to measure the importance of a node as a bridge or key path in the network; sd(i,j,k) is the number of paths passing through i in the shortest path from node j to node k; C E represents the eigenvector centrality, which is used to identify the most influential nodes in the network; c is the proportionality constant; a ij =1, node i is connected to node j, a ij =0, node i is not connected to node j.

[0082] S4B: Comprehensively consider the four aspects of people, machines, environment and management, deeply explore the causes of high-risk nodes in civil aviation control operations, propose safety risk control strategies in technology, processes, training, etc., improve the resilience of the control operation safety system, and ensure the safety and reliability of air transportation.

[0083] Therefore, the present invention adopts the above-mentioned civil aviation control operation safety risk assessment method based on text mining and network analysis to accurately identify the main risk themes and key causal factors in civil aviation control operations, so as to facilitate the smooth and safe operation of the aviation transportation system.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A civil aviation control operation safety risk assessment method based on text mining and network analysis, characterized by: The following steps are involved: S1: Collect data on unsafe air traffic control operations at home and abroad, and pre-process them, including processing irrelevant and repeated data, and building a stop word and word segmentation vocabulary for air traffic control operation safety risk management; S2: Based on the preprocessed data, the BERTopic topic model is used to identify the civil aviation control operation safety risk topics, and the model parameters of BERTopic are modified through the topic coherence TC and topic diversity TD among the topics to optimize the risk topics; S3: Set causal network nodes based on the civil aviation control operation safety risk theme and construct a causal network to determine the interrelationships between civil aviation control operation safety risk factors; S4: Based on topological analysis, identify high-risk nodes in the causal network of civil aviation control operation safety, determine the key causal factors in civil aviation control operation safety incidents, and propose risk control strategies.

2. The method for assessing civil aviation control operation safety risk based on text mining and network analysis according to claim 1 is characterized by: In step S1, the following steps are included: S1A: Follow the following principles to collect data on unsafe events in civil aviation control operations in relevant years: the accident description is detailed and clear; the airline is a public air transport company, not general aviation or military aviation; aviation accidents caused by special reasons are not considered; there are no repeated events; S1B: Process irrelevant and duplicate data and build a stop word and word segmentation vocabulary for air traffic control operation safety risk management.

3. The method for assessing civil aviation control operation safety risk based on text mining and network analysis according to claim 1 is characterized by: In step S2, the following steps are included: S2A: uses the bert-base-chinese model to encode the preprocessed text and implement text embedding; S2B: Cluster the obtained word embedding vectors according to semantic similarity; S2C: It uses the class-based term frequency-inverse document frequency C-TF-IDF and the maximum marginal relevance algorithm MMR to extract and merge keywords according to the importance scores of the vocabulary and preliminarily determine the text topic; S2D: Use visualize_topics to visualize the clustering results and keywords in the form of charts to understand the topic structure more intuitively; S2E: Deeply study the potential hierarchical structure between topics and use cosine similarity to perform correlation analysis on each risk topic; S2F: Uses dynamic topic model to analyze topic trends; S2G: Evaluate and optimize the BERTopic model based on topic coherence TC and topic diversity TD.

4. The method for assessing civil aviation control operation safety risk based on text mining and network analysis according to claim 3 is characterized by: In step S2C, the importance score formula is as follows: Where T i,c represents the importance score of word i in semantic cluster c, tf i,c represents the frequency of word i in semantic cluster c, f x represents the frequency of word i in all semantic clusters, and A represents the mean number of words in each semantic cluster; In step S2E: the cosine similarity calculation formula is as follows: Among them, A i and B i Represent the embedding vectors of the two topics respectively; the value range of cosθ is (-1,1); the closer the absolute value of cosθ is to 1, the more similar the two topics are; the closer the similarity is to 0, the weaker the relationship between the two topics; In step S2G, the calculation formulas of TC and TD are as follows: Among them, NPMI is the normalized mutual information, which represents the co-occurrence between two words; w i 、w j Respectively represent the i-th and j-th keywords in a topic; P(w i ,w j ) indicates w i and w j The probability of co-occurrence in the same document; P(w i ) and P(w j ) represents the probability of a single word appearing in a document; W represents the total number of the first N topic words of all extracted topics; W unique Indicates the number of unique subject terms.

5. The method for assessing civil aviation control operation safety risk based on text mining and network analysis according to claim 1 is characterized by: In step S3, the following steps are included: S3A: Based on the civil aviation control operation safety risk themes generated in S2, determine the risk nodes of the causal relationship network; S3B: Mining the association rules between civil aviation control and operation safety risk topics, obtaining the minimum support and minimum confidence thresholds, eliminating the weaker mutual relationships between civil aviation control and operation safety risk topics, and finally establishing the civil aviation control and operation safety risk causal relationship network connection edges based on the association relationship; S3C: Draw a causal network of safety risks in civil aviation control operations.

6. The method for assessing civil aviation control operation safety risk based on text mining and network analysis according to claim 5 is characterized by: In step S3B, the relevant formula for association rule mining is as follows: Among them, Support is the support, which is used to measure the frequency of a topic appearing in all topics; Confidence is the confidence, which is used to measure the accuracy of the rule; Lift is the lift, which is used to measure the strength of the management rule.

7. The method for assessing civil aviation control operation safety risk based on text mining and network analysis according to claim 1 is characterized by: In step S4, the following steps are included: S4A: Using topological analysis, we can identify high-risk nodes based on the network characteristics, parameters, and centrality index of the civil aviation control operation safety risk network; S4B: Comprehensively consider the four aspects of people, machines, environment and management, deeply explore the causes of high-risk nodes in civil aviation control operations, propose safety risk control strategies in technology, processes, training, etc., improve the resilience of the control operation safety system, and ensure the safety and reliability of air transportation.

8. The method for assessing civil aviation control operation safety risk based on text mining and network analysis according to claim 7 is characterized by: In step S4A, important nodes in the causal network are determined based on the centrality index. The specific index and calculation formula are as follows: Among them, C D represents degree centrality, which is used to measure the degree of connectivity between a node and all other nodes in the network; g is the number of edges connecting node i; C B represents the betweenness centrality, which is an indicator used to measure the importance of a node as a bridge or key path in the network; sd(i,j,k) is the number of paths passing through i in the shortest path from node j to node k; C E represents the eigenvector centrality, which is used to identify the most influential nodes in the network; c is the proportionality constant; a ij =1, node i is connected to node j, a ij =0, node i is not connected to node j.

Images