Coal mine accident cause analysis method based on knowledge graph

Abstract

The invention relates to the technical field of coal mine accident cause analysis, and provides a coal mine accident cause analysis method based on a knowledge graph. The method comprises the following specific steps: collecting a coal mine roof accident report and cleaning a text to obtain a roof accident word set; secondly, building a top plate accident cause knowledge graph by extracting accident cause nodes, accident attributes and relations of the accident cause nodes and the accident attributes of the word set; thirdly, calculating the similarity of accident cause nodes, and performing node merging based on the similarity; then, acquiring cause attribute codes and accident attribute codes, calculating cause association indexes of accident cause nodes, and combining the accident attribute codes to obtain single-cause risk indexes and multi-cause risk indexes; and finally, according to the single-cause risk and the multi-cause risk, determining a comprehensive risk index, and carrying out danger degree sorting on accident cause nodes. The method can effectively analyze the risk factors of the coal mine roof accident, thereby reducing the potential risk of the roof accident.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine accident cause analysis, and specifically to a method for analyzing coal mine accident causes based on a knowledge graph. Background Art

[0002] Coal mine roof accidents are common and serious safety accidents during mine operations, usually manifested as phenomena such as the collapse, sliding, and caving of the roofs of coal mine roadways or working faces. The occurrence of such accidents is closely related to multiple factors such as the geological conditions of the mine, the mining method, the support technology, and the roof management. During the operation of underground mines, due to problems such as insufficient understanding of geological conditions, lagging support technology, and imperfect mine management and safety production measures, the damage and collapse of rock strata or roofs may occur, resulting in serious property losses and environmental pollution, and even a large number of casualties. Therefore, introducing coal mine accident cause analysis technology can analyze the causes of roof collapse accidents, help find the root causes and key factors of accident occurrence, provide a basis for formulating more scientific and reasonable safety production measures, and thus reduce the incidence of coal mine accidents.

[0003] At present, the existing methods for analyzing coal mine accident causes only focus on environmental factors or geological factors, ignoring both equipment failure factors and human error factors, and not considering the correlation between these factors, resulting in the inability to effectively analyze roof accidents under the influence of multiple factors, thus reducing the accuracy of coal mine accident cause analysis, further affecting the formulation of accident prevention and safety management measures, and increasing the potential risk of accident occurrence.

[0004] Therefore, a method for analyzing coal mine accident causes based on a knowledge graph is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for analyzing coal mine accident causes based on a knowledge graph. By collecting coal mine roof accident reports and cleaning the text, a set of roof accident words is obtained; secondly, by extracting the accident cause nodes, accident attributes, and their relationships from this set of words, a roof accident cause knowledge graph is constructed; then, the similarity of accident cause nodes is calculated, and node merging is performed based on the similarity; then, the cause attribute codes and accident attribute codes are obtained, the cause correlation index of accident cause nodes is calculated, and the single-factor risk index and multi-factor risk index are obtained by combining the accident attribute codes; finally, according to the single-factor risk and multi-factor risk, the comprehensive risk index is determined, and the accident cause nodes are ranked according to the degree of danger. The present invention can effectively analyze the risk factors of coal mine roof accidents, thereby reducing the potential risk of roof accident occurrence.

[0006] A method for analyzing coal mine accident causes based on a knowledge graph includes:

[0007] Collect the accident reports of coal mine roof accidents, and through text cleaning operations, obtain the set of roof accident words;

[0008] According to the set of roof accident words, construct a roof accident entity extraction model, an entity relationship classification model, and an entity attribute extraction model, and combine the set of roof accident words for model training; use the trained models to perform entity extraction, entity relationship extraction, and attribute extraction on the set of roof accident words to obtain accident cause types, accident cause nodes, roof accident nodes, cause attributes, accident attributes, and accident cause association edges;

[0009] Combine the accident cause nodes, the roof accident nodes, the cause attributes, the accident attributes, and the accident cause association edges to construct a roof accident cause knowledge graph;

[0010] Calculate the cause text similarity of the accident cause nodes and the attribute similarity of the corresponding cause attributes to obtain the accident cause similarity; combine the accident attributes of the accident cause nodes to perform a merging operation on the accident cause nodes of the roof accident cause knowledge graph;

[0011] Obtain cause attribute codes and accident attribute codes, and calculate the cause association index of different accident cause nodes; combine the accident attribute codes and the roof accident risk assessment function to calculate the single - cause risk index; obtain the multi - cause risk index of the accident cause nodes, combine the single - cause risk index to obtain the comprehensive risk index, and rank the accident cause nodes according to the degree of danger.

[0012] Preferably, the specific implementation process of obtaining the accident cause nodes, the roof accident nodes, the cause attributes, the accident attributes, and the accident cause association edges includes:

[0013] Perform label annotation on several groups of the set of roof accident words to obtain a roof accident entity training set, an entity relationship training set, and an entity attribute training set; among them, the labels of the roof accident entity training set include accident cause node labels and roof accident node labels; the label of the entity relationship training set is the accident cause association edge label; the labels of the entity attribute training set include cause attribute labels and accident attribute labels;

[0014] Input the roof accident entity training set into the BERT pre - training model for coal mine roof accident entity training to obtain the roof accident entity extraction model; construct the entity relationship classification model based on the BERT pre - training model and a classifier, and perform coal mine roof entity relationship training according to the entity relationship training set; perform coal mine roof accident entity attribute training on the BERT pre - training model through the entity attribute training set to obtain the entity attribute extraction model;

[0015] Extract entities in the roof accident word set through the roof accident entity extraction model, and determine the accident cause nodes and the roof accident nodes; determine the relationships between different accident cause nodes and / or roof accident nodes according to the entity relationship classification model to obtain the accident cause association edges; combine with the entity attribute extraction model to obtain the cause attributes of the accident cause nodes and the accident attributes of the roof accident nodes.

[0016] Preferably, the accident cause types include geological factors, human error factors, equipment failure factors, and working environment factors; the accident cause nodes of geological factors include geological structure nodes, rock stratum nodes, strata nodes, and surface water nodes; the accident cause nodes of human error factors include mining method nodes, ventilation measure nodes, support method nodes, and overloading operation nodes; the accident cause nodes of equipment failure factors include support equipment nodes, ventilation equipment nodes, mechanical equipment nodes, and power equipment nodes; the accident cause nodes of working environment factors include mine type nodes, mine depth nodes, in-mine temperature nodes, in-mine gas concentration nodes, and in-mine humidity nodes; the accident attributes include location, time, accident scope, casualties, and economic losses.

[0017] Preferably, calculate the cause text similarity and the attribute similarity, and perform a merging operation on the accident cause nodes of the roof accident cause knowledge graph; the specific implementation process includes:

[0018] According to two different accident cause nodes and respectively obtain and the number of characters in the node names and through calculate the cause text similarity between the accident cause nodes and where min() represents the minimum value function; max() represents the maximum value function; used to calculate and and the number of identical characters in the node names;

[0019] respectively obtain the cause attribute sets {A1,..., A and corresponding to the two accident cause nodes m} and {B1,..., B n}, perform one-to-one matching on the cause attributes in the two different cause attribute sets to obtain k groups of cause attribute matching pairs; for any cause attribute matching pair Calculate the similarity of the attribute values of the two cause attributes; synthesize the similarity of the attribute values of k groups of cause attribute matching pairs to obtain the attribute similarity The specific calculation formula is:

[0020]

[0021] where m and n respectively represent and the number of the cause attributes of; A i and B j are two cause attributes in the cause attribute matching pair ; A i represents the i-th cause attribute of and i ∈ {1,..., m}, B j represents the j-th cause attribute of and j ∈ {1,..., n}; α i represents the influence factor of the cause attribute matching pair .

[0022] Combine the cause text similarity and of the accident cause nodes and the attribute similarity through calculate the comprehensive similarity of accident causes; where β represents the weight of.

[0023] If the comprehensive similarity of accident causes is greater than the predetermined similarity threshold, combine the accident cause nodes and into one node.

[0024] Preferably, the specific implementation process of calculating the cause association index of different accident cause nodes:

[0025] Perform data encoding on the cause attributes and the accident attributes to obtain the cause attribute encoding and the accident attribute encoding; according to the cause attribute encoding and the accident attribute encoding, calculate the cause association index between the roof accident node and different accident cause nodes, and the specific calculation formula is:

[0026]

[0027] where represents the cause association index between the roof accident node and the u-th accident cause node ; N roof represents the total number of coal mine roof accidents; represents The total number of coal mine roof accidents; COV() represents the covariance function; R represents the numerical matrix of the accident attribute codes of the u,l represents the numerical matrix of the causal attribute codes of the l-th causal attribute; σ R represents the standard deviation of R; represents C u,l 's standard deviation; γ l represents the influence degree of the l-th causal attribute on coal mine roof accidents; M represents the total number of the causal attributes of the

[0028] Preferably, combining the accident attribute codes and the causal association indices of different accident causal nodes, through the roof accident risk assessment function, the single-factor risk index of the accident causal nodes causing coal mine roof accidents is quantified. The specific formula is:

[0029]

[0030] Among them, represents the roof accident node and the u-th accident causal node the causal association index between them; RA u represents the single-factor risk index of the v represents the v-th accident attribute of the represents R v the roof accident risk assessment function of the v represents R v the influence coefficient of the consequences of coal mine roof accidents; K represents the total number of the accident attributes.

[0031] Preferably, calculate the multi-factor risk indices of different accident causal nodes, and combine the single-factor risk indices to obtain the comprehensive risk index of each accident causal node; the specific calculation formula is:

[0032]

[0033] Among them, CR u represents the comprehensive risk index of the u-th accident causal node ; ∈0 represents the weight coefficient of the single-factor risk index RA u ; represents the q-th multi-factor risk index of the q represents The corresponding weight coefficient; Q represents The total number of the multi-factor risk indexes of

[0034] Rank the risk levels of the accident causation nodes according to the comprehensive risk indexes of different accident causation nodes.

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

[0036] 1. The roof accident causation knowledge graph proposed by the present invention can effectively extract accident causation nodes, roof accident nodes, causation attributes, accident attributes, and accident causation association edges from the roof accident word set by combining entity extraction, entity relationship classification, and entity attribute extraction models, improving the processing and analysis efficiency of accident data. Through training the BERT pre-trained model for different training tasks, accurate entity recognition, relationship extraction, and attribute recognition are achieved, effectively improving the data processing efficiency. In addition, constructing the accident causal relationship in the form of a knowledge graph helps to analyze the root cause of the accident as a whole, promoting the improvement of accident prevention and control measures and decision-making support.

[0037] 2. The present invention realizes the automatic merging operation of accident causation nodes through the causation text similarity and attribute similarity of accident causation nodes, thus effectively improving the construction efficiency of the roof accident causation knowledge graph. This method not only considers the character similarity of node names, but also deeply compares the attribute value similarity of causation attributes, and comprehensively evaluates using influence factors to ensure the accuracy and rationality of the merging operation. By merging highly similar accident causation nodes, the structure of the knowledge graph is effectively simplified, redundant information is reduced, the accuracy and utilization efficiency of the knowledge graph are improved, and thus the accuracy of coal mine accident causation analysis is improved.

[0038] 3. The present invention can accurately evaluate the influence degree of various causation nodes on coal mine roof accidents through the quantitative calculation of the causation association index, single-factor risk index, and multi-factor risk index of accident causation nodes, and rank the risk levels of accident causation nodes according to the comprehensive risk index. This method combines causation attribute coding and accident attribute coding to calculate the causation association index, comprehensively reflecting the relationship between each accident causation node and coal mine roof accidents, effectively improving the accuracy of coal mine accident risk prediction. At the same time, the comprehensive risk index can quantitatively measure the risk of accident causation nodes in multiple dimensions, ensuring that roof accidents under the influence of multiple factors can be effectively analyzed, thus providing a scientific basis for coal mine safety production, reducing the risk of coal mine roof accidents, and improving the safety management level of coal mines. Brief Description of the Drawings

[0039] Figure 1Flowchart of a method for analyzing causes of coal mine accidents based on a knowledge graph provided by an embodiment of the present invention application;

[0040] Figure 2 Schematic diagram of nodes of a knowledge graph of causes of roof accidents provided by an embodiment of the present invention application;

[0041] Figure 3 Flowchart of comprehensive risk index assessment provided by an embodiment of the present invention application. Detailed implementation manners

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] The occurrence of coal mine roof accidents is usually closely related to factors such as the geological conditions of the mine, the mining method, the support technology, and the roof management. With the increase in the depth of coal mine exploitation, the geological conditions become more complex, and the occurrence frequency and danger of roof accidents also gradually increase. This will not only cause casualties to miners, but also lead to the stagnation of mine production and serious economic losses. The previous methods for analyzing the causes of coal mine accidents only focused on environmental factors or geological factors, without considering the influence of equipment failure factors and human error factors and the correlation between multiple factors, thus unable to effectively analyze the risk factors for the occurrence of roof accidents.

[0044] The present invention proposes a method for analyzing the causes of coal mine accidents based on a knowledge graph to analyze the causes of coal mine roof accidents and determine the high-risk factors that trigger coal mine roof accidents. To illustrate the effectiveness of the method for analyzing the causes of coal mine accidents of the present invention, it will be specifically described in conjunction with the accompanying drawings of this embodiment and the following two embodiments.

[0045] Embodiment 1

[0046] This embodiment of the present application discloses a method for analyzing the causes of coal mine accidents based on a knowledge graph to effectively analyze the high-risk factors that trigger coal mine roof accidents. Refer to Figure 1, which is a flowchart of a coal mine accident causation analysis method based on a knowledge graph; the specific implementation steps include: S1. Collect accident reports of coal mine roof accidents, and obtain a set of roof accident words through text cleaning operations; S2. Construct a roof accident causation knowledge graph based on the set of roof accident words; S3. Calculate the similarity of causation texts and the similarity of attributes, determine the similarity of accident causations, and merge accident causation nodes; S4. Obtain causation attribute codes and accident attribute codes, and calculate the causation correlation index of different accident causation nodes; S5. Combine the accident attribute codes and the roof accident risk assessment function to calculate the single-cause risk index; S6. Obtain the multi-cause risk index of accident causation nodes, combine it with the single-cause risk index to obtain the comprehensive risk index, and rank the accident causation nodes according to the degree of danger.

[0047] Furthermore, collect accident reports of coal mine roof accidents, and obtain a set of roof accident words through text cleaning operations, corresponding to the above S1 step; among them, the text cleaning operations include removing stop words, removing noise information, synonym normalization, part-of-speech tagging, and word segmentation.

[0048] Specifically, removing stop words is used to remove stop words in accident reports, such as words unrelated to analysis like "of", "and", and "in". Removing noise information is used to remove tables, pictures, and special symbols in accident reports. Synonym normalization is used to convert similar expression forms in the same accident report into a unified expression, such as "roof cave-in", "roof collapse", and "roof instability" are all uniformly expressed as "roof collapse". Part-of-speech tagging and word segmentation are used to segment and tag the parts of speech of the text in accident reports to ensure that useful keywords such as correct nouns and verbs can be extracted.

[0049] Furthermore, construct a roof accident causation knowledge graph based on the set of roof accident words; corresponding to the above S2 step, the specific implementation process includes:

[0050] Perform label annotation on several groups of roof accident word sets to obtain a roof accident entity training set, an entity relationship training set, and an entity attribute training set; among them, the labels of the roof accident entity training set include accident causation node labels and roof accident node labels; the labels of the entity relationship training set are accident causation association edge labels; the labels of the entity attribute training set include causation attribute labels and accident attribute labels;

[0051] Input the roof accident entity training set into the BERT pre-trained model for coal mine roof accident entity training to obtain a roof accident entity extraction model; construct an entity relationship classification model based on the BERT pre-trained model and a classifier, and perform coal mine roof entity relationship training according to the entity relationship training set; perform coal mine roof accident entity attribute training on the BERT pre-trained model through the entity attribute training set to obtain an entity attribute extraction model;

[0052] Extract entities in the roof accident word set through the roof accident entity extraction model to determine the accident cause nodes and roof accident nodes; determine the relationships between different accident cause nodes and / or roof accident nodes according to the entity relationship classification model to obtain accident cause association edges; combine the entity attribute extraction model to obtain the cause attributes of the accident cause nodes and the accident attributes of the roof accident nodes; construct a roof accident cause knowledge graph by combining the accident cause nodes, roof accident nodes, cause attributes, accident attributes, and accident cause association edges.

[0053] Optionally, during the training process of the model, for the roof accident entity extraction model, the model is trained and optimized through the Dice loss; for the entity relationship classification model and the entity attribute extraction model, the model is optimized through the cross-entropy loss function.

[0054] By constructing a roof accident cause knowledge graph and combining the BERT pre-trained model and related training sets for entity extraction, entity relationship classification, and entity attribute extraction, it is possible to efficiently and accurately extract accident cause nodes and roof accident nodes from the roof accident word set, providing a reliable basis for subsequent cause analysis.

[0055] Furthermore, the accident cause types include geological factors, human error factors, equipment failure factors, and working environment factors; refer to Figure 2 , which is a schematic diagram of the nodes of the roof accident cause knowledge graph; the accident cause nodes of geological factors include geological structure nodes, rock stratum nodes, stratum nodes, and surface water nodes; the accident cause nodes of human error factors include mining method nodes, ventilation measure nodes, support method nodes, and overloading operation nodes; the accident cause nodes of equipment failure factors include support equipment nodes, ventilation equipment nodes, mechanical equipment nodes, and power equipment nodes; the accident cause nodes of the working environment factors include mine type nodes, mine depth nodes, mine internal temperature nodes, mine internal gas concentration nodes, and mine internal humidity nodes; the accident attributes include location, time, accident scope, casualties, and economic losses.

[0056] Specifically, the causative attributes of geological structure nodes include geological structure types (such as faults, folds, and dikes, etc.), structure distribution, rock mass stability, and rock mass fragmentation degree; the causative attributes of rock layer nodes include rock layer thickness, rock layer hardness, rock layer brittleness, rock type, rock layer distribution, and rock layer friction coefficient; the causative attributes of stratum nodes include stratum categories (such as coal seams, shale layers, and sandstone layers, etc.), stratum thickness, and stratum gap; the causative attributes of surface water nodes include groundwater level, groundwater flow velocity, permeability, and the pressure of groundwater on the roof; the causative attributes of mining methods nodes include mining methods (such as gateway mining and reverse operation mining, etc.), mining sequence, mining depth, and mining speed; the causative attributes of ventilation measures nodes include ventilation methods (such as mechanical ventilation and natural ventilation), air volume, air velocity, and ventilation duct layout; the causative attributes of support methods nodes include support types (such as hydraulic supports, wooden supports, and steel supports, etc.), support density, and the rationality of support design; the causative attributes of overloading operations nodes include operation load, equipment bearing capacity, and operation intensity; the causative attributes of support equipment nodes include support equipment types (such as hydraulic supports and bolt support, etc.), equipment wear, equipment maintenance frequency, and equipment load; the causative attributes of ventilation equipment nodes include ventilation equipment types (such as fans and air ducts, etc.), ventilation capacity, equipment failure frequency, and equipment operation time; the causative attributes of mechanical equipment nodes include mechanical equipment types (such as mining equipment and transportation equipment, etc.), equipment performance, and failure rate; the causative attributes of electrical equipment nodes include electrical equipment types (such as generators and electrical control equipment, etc.), equipment failure rate, and power supply stability; the causative attributes of mine types nodes include mine types (such as underground coal mines, metal mines, and non-metal mines, etc.) and mine structure; the causative attributes of mine depth nodes include mine depth and the influence of depth on stratum pressure; the causative attributes of in-mine temperature nodes include mine temperature and temperature change range; the causative attributes of in-mine gas concentration nodes include the gas concentration in the mine (such as methane, carbon dioxide, and hydrogen sulfide, etc.) and gas concentration change; the causative attributes of in-mine humidity nodes include mine humidity and humidity fluctuation.

[0057] By refining the types of accident causative factors and clarifying the classification of various causative nodes, the specific causes of coal mine roof accidents can be analyzed more comprehensively and accurately. Dividing the accident causative factors into geological factors, human error factors, equipment failure factors, and working environment factors, and further refining them into multiple specific nodes, helps to reveal the deep-seated reasons for accidents. In addition, by comprehensively considering accident attributes (such as location, time, accident level, accident scope, etc.), more operable decision-making basis can be provided, which helps to accurately evaluate the influence scope and risk level of accidents.

[0058] Furthermore, the cause text similarity of the accident cause node and the attribute similarity of the corresponding cause attribute are calculated to obtain the accident cause similarity; the accident cause nodes of the roof accident cause knowledge graph are merged in combination with the accident attributes of the accident cause node; corresponding to the above step S3, the specific implementation process includes:

[0059] According to two different accident cause nodes and Get separately and Number of characters in node names and pass Calculate the accident cause node and The similarity of the causal text between Among them, min() represents the minimum function; max() represents the maximum function; Used for calculation and The number of identical characters in the node names;

[0060] Get two accident cause nodes respectively and The corresponding causal attribute set {A1,…,A m} and {B1,…,B n}, perform one-to-one matching on the causal attributes in two different causal attribute sets to obtain k sets of causal attribute matching pairs; for any causal attribute matching pair Calculate the similarity of the attribute values ​​of two causal attributes; combine the attribute value similarities of k groups of causal attribute matching pairs to obtain the attribute similarity The specific calculation formula is:

[0061]

[0062] Among them, m and n represent and The number of causal attributes of A i and B j Matching pairs for causal attributes The two causal properties in i express The i-th causal attribute of B j represents the jth causal attribute of B and j∈{1,…,n}; α i Represents the causal attribute matching pair Impact factor;

[0063] Combined with the accident cause node and The similarity of the causal text and attribute similarity Through calculate the comprehensive similarity of accident causes; where, β represents the weight of and 0 < β < 1;

[0064] If the comprehensive similarity of accident causes is greater than the predetermined similarity threshold, the accident cause nodes and are merged into one node.

[0065] To illustrate the effectiveness of the accident cause node merging operation proposed in the embodiments of the present application, Table 1 exemplarily lists the merging processes of different accident cause nodes; where, the predetermined similarity threshold is 0.85.

[0066] Table 1. Merging processes of different accident cause nodes

[0067] Node 1 Node 2 Text similarity Attribute similarity Comprehensive similarity Predetermined similarity threshold Whether to merge JD01 JD02 0.875 0.8 0.838 0.85 No JD02 JD03 0.857 0.9 0.878 0.85 Yes JD03 JD04 1 0.7 0.86 0.85 Yes JD04 JD05 0.555 0.6 0.578 0.85 No

[0068] By calculating the similarity of the cause text of accident cause nodes and the attribute similarity of cause attributes, the similarity between different accident cause nodes can be accurately evaluated, and then similar cause nodes can be effectively merged to reduce redundant information. This process ensures the accuracy of node merging through an accurate text and attribute matching algorithm in combination with a similarity threshold. By comprehensively considering the similarity of text and attributes, the relevance between accident cause nodes can be more comprehensively reflected, improving the accuracy and consistency of the knowledge graph and reducing the complexity of the graph caused by node redundancy.

[0069] Furthermore, obtain the cause attribute code and accident attribute code, and calculate the cause correlation index of different accident cause nodes, corresponding to Figure 1 step S4; refer to Figure 3 , which is the flowchart of comprehensive risk index evaluation, and the specific implementation process:

[0070] Perform data encoding on the cause attribute and accident attribute to obtain the cause attribute code and accident attribute code; according to the cause attribute code and accident attribute code, calculate the cause correlation index between the roof accident node and different accident cause nodes, and the specific calculation formula is:

[0071]

[0072] where, represents the cause correlation index between the roof accident node and the u-th accident cause node ; N roof represents the total number of coal mine roof accidents; represents The total number of coal mine roof accidents; COV() represents the covariance function; R represents the numerical matrix of the accident attribute codes; C u,l represents the numerical matrix of the causal attribute codes of the l-th causal attribute; σ R represents the standard deviation of R; represents C u,l 's standard deviation; γ l represents the influence degree of the l-th causal attribute on coal mine roof accidents; M represents the total number of causal attributes.

[0073] In the embodiments of the present application, by combining the causal attribute codes and the accident attribute codes to calculate the causal association index of accident causal nodes, the relationship between different causal factors and coal mine roof accidents can be quantified more accurately. By introducing the calculation methods of covariance function and standard deviation, not only can the influence degree of each accident causal node be accurately identified, but also the specific influence of causal attributes on coal mine roof accidents can be quantified, thereby providing a more accurate decision-making basis for coal mine safety management.

[0074] Further, by combining the accident attribute codes and the roof accident risk assessment function, the single-factor risk index is calculated; corresponding to Figure 1 the S5 step, the specific implementation process is as follows:

[0075] By combining the accident attribute codes and the causal association index of different accident causal nodes, through the roof accident risk assessment function, the single-factor risk index of accident causal nodes causing coal mine roof accidents is quantified. The specific formula is:

[0076]

[0077] Wherein, represents the roof accident node and the u-th accident causal node between the causal association index; RA u represents 's single-factor risk index; R v represents the v-th accident attribute of; represents R v 's roof accident risk assessment function, is the exponential coefficient of R v ; ω v represents R v 's influence coefficient on the consequences of coal mine roof accidents; K represents the total number of accident attributes.

[0078] By combining the accident attribute coding and the roof accident risk assessment function, the single - cause risk index can be accurately calculated, further quantifying the risk of coal mine roof accidents caused by each accident - causing node, reasonably evaluating the potential impact of different causations on coal mine roof accidents, improving the accuracy of accident risk identification, and thus reducing the accident rate.

[0079] Furthermore, obtain the multi - cause risk index of the accident - causing node, combine it with the single - cause risk index to get the comprehensive risk index, and rank the accident - causing nodes according to the degree of danger; corresponding to Figure 1 Step S6, where calculate the multi - cause risk index of different accident - causing nodes, combine it with the single - cause risk index to get the comprehensive risk index of each accident - causing node; the specific calculation formula is:

[0080]

[0081] Among them, CR u represents the comprehensive risk index of the \(u\) - th accident - causing node ; \(\in_0\) represents the weight coefficient of the single - cause risk index \(RA\) of u ; represents the \(q\) - th multi - cause risk index of q ; \(\in\) represents the corresponding weight coefficient; \(Q\) represents

[0082]

[0083] Among them, represents the total number of coal mine roof accidents caused by the \(q\) - th multi - accident - causing node containing ; \(\delta\) q represents the influence weight of on coal mine roof accidents in the \(q\) - th multi - accident - causing node;

[0084] According to the comprehensive risk index of different accident - causing nodes, rank the accident - causing nodes according to the degree of danger. As shown in Table 2, Table 2 exemplarily lists the comprehensive risk indexes of some accident - causing nodes.

[0085] Table 2. Comprehensive risk indexes of some accident - causing nodes

[0086]

[0087] By combining the single - factor risk index and the multi - factor risk index, calculating the comprehensive risk index of each accident - causing node, the potential danger of coal mine roof accidents can be accurately quantified. By ranking the accident - causing nodes according to their risk levels, it helps to clarify the priority of various risks. This method not only improves the accuracy and comprehensiveness of accident - cause analysis, but also can effectively identify the mutual influence among multiple causal factors, promote the pertinence and effectiveness of accident prevention measures, and thus reduce the safety risks in the process of coal mine production.

[0088] In the embodiment of the present application, by combining the entity extraction, entity - relationship classification, and entity - attribute extraction models, the construction of accident causal relationships is optimized, significantly improving the processing and analysis efficiency of coal mine roof accident data. By automatically merging similar accident - causing nodes, redundant information is effectively reduced, the knowledge graph structure is simplified, the accuracy and utilization efficiency of the graph are improved, and thus the accuracy of coal mine accident - cause analysis is enhanced. At the same time, by quantitatively calculating the causal association index, single - factor risk index, and multi - factor risk index of accident - causing nodes, the accurate assessment of coal mine roof accident risks is realized, which can comprehensively reflect the influence degree of various causal nodes, and rank the accident - causing nodes according to the comprehensive risk index, providing a scientific basis for coal mine safety management and reducing the risk of roof accidents.

[0089] Embodiment Two

[0090] In the embodiment of the present application, a method for analyzing coal mine accident causes based on a knowledge graph is applied to the area to be mined in Coal Mine A to predict the risk factors in the coal mine mining management in this area, so as to prevent coal mine roof accidents in advance.

[0091] Specifically, collect the accident reports of coal mine roof accidents, and through text cleaning operations, obtain the set of roof accident words; among them, the text cleaning operations include removing stop words, removing noise information, synonym normalization, part - of - speech tagging, and word segmentation.

[0092] Furthermore, according to the set of roof accident words, construct a knowledge graph of roof accident causes; the specific implementation process includes:

[0093] Label several groups of roof accident word sets to obtain a roof accident entity training set, an entity - relationship training set, and an entity - attribute training set; among them, the labels of the roof accident entity training set include accident - causing node labels and roof accident node labels; the label of the entity - relationship training set is the accident - causing association edge label; the labels of the entity - attribute training set include causal - attribute labels and accident - attribute labels;

[0094] Input the roof accident entity training set into the BERT pre-trained model for coal mine roof accident entity training to obtain a roof accident entity extraction model; construct an entity relationship classification model based on the BERT pre-trained model and a classifier, and conduct coal mine roof entity relationship training according to the entity relationship training set; perform coal mine roof accident entity attribute training on the BERT pre-trained model through the entity attribute training set to obtain an entity attribute extraction model;

[0095] Extract entities in the roof accident word set through the roof accident entity extraction model to determine accident cause nodes and roof accident nodes; determine the relationships between different accident cause nodes and / or roof accident nodes according to the entity relationship classification model to obtain accident cause association edges; combine the entity attribute extraction model to obtain the cause attributes of accident cause nodes and the accident attributes of roof accident nodes; combine accident cause nodes, roof accident nodes, cause attributes, accident attributes, and accident cause association edges to construct a roof accident cause knowledge graph.

[0096] Furthermore, accident cause types include geological factors, human error factors, equipment failure factors, and working environment factors; accident cause nodes of geological factors include geological structure nodes, rock formation nodes, strata nodes, and surface water nodes; accident cause nodes of human error factors include mining method nodes, ventilation measure nodes, support method nodes, and overloading operation nodes; accident cause nodes of equipment failure factors include support equipment nodes, ventilation equipment nodes, mechanical equipment nodes, and power equipment nodes; accident cause nodes of working environment factors include mine type nodes, mine depth nodes, in-mine temperature nodes, in-mine gas concentration nodes, and in-mine humidity nodes; accident attributes include location, time, accident scope, casualties, and economic losses.

[0097] Furthermore, calculate the cause text similarity of accident cause nodes and the attribute similarity of corresponding cause attributes to obtain accident cause similarity; combine the accident attributes of accident cause nodes to perform a merging operation on the accident cause nodes of the roof accident cause knowledge graph.

[0098] Furthermore, obtain cause attribute codes and accident attribute codes, and calculate the cause association index of different accident cause nodes; specific implementation process:

[0099] Perform data encoding on cause attributes and accident attributes to obtain cause attribute codes and accident attribute codes; according to cause attribute codes and accident attribute codes, calculate the cause association index of roof accident nodes and different accident cause nodes, and the specific calculation formula is:

[0100]

[0101] Among them, represents the roof accident node and the causal association index between the u-th accident-causing node ; N roof represents the total number of coal mine roof accidents that occurred; represents the total number of coal mine roof accidents caused; COV() represents the covariance function; R represents the numerical matrix of the accident attribute codes of; C u,l represents the numerical matrix of the causal attribute codes of the l-th causal attribute; σ R represents the standard deviation of R; represents C u,l 's standard deviation; γ l represents the influence degree of the l-th causal attribute on coal mine roof accidents; M represents the total number of causal attributes of.

[0102] Furthermore, combining the accident attribute codes and the roof accident risk assessment function, calculate the single-cause risk index; the specific implementation process is as follows:

[0103] Combining the accident attribute codes and the causal association index of different accident-causing nodes, through the roof accident risk assessment function, quantify the single-cause risk index of the accident-causing node causing coal mine roof accidents. The specific formula is:

[0104]

[0105] Among them, represents the roof accident node and the u-th accident-causing node the causal association index between; RA u represents 's single-cause risk index; R v represents the v-th accident attribute of; represents R v the roof accident risk assessment function of; ω v represents R v the influence coefficient of the consequences on coal mine roof accidents; K represents the total number of accident attributes.

[0106] Furthermore, obtain the multi-cause risk index of the accident-causing node, combine the single-cause risk index to obtain the comprehensive risk index, and rank the accident-causing nodes according to the degree of danger; among them, calculate the multi-cause risk index of different accident-causing nodes, combine the single-cause risk index to obtain the comprehensive risk index of each accident-causing node; the specific calculation formula is:

[0107]

[0108] Among them, CRu Represents the u-th accident-causing node The comprehensive risk index of; ∈0 represents The single-factor risk index RA of u The weight coefficient of; Represents The q-th multi-factor risk index of; ∈ q Represents The corresponding weight coefficient; Q represents The total number of multi-factor risk indices;

[0109] According to the comprehensive risk indices of different accident-causing nodes, rank the accident-causing nodes according to their danger levels.

[0110] Optionally, analyze the comprehensive risk indices of different accident-causing nodes, identify high-risk factors such as roof rock structure, unstable working face, improper mining method, and insufficient support measures, and rank these risk factors according to their priorities. Develop specific preventive measures for high-risk factors, such as strengthening roof stability detection, conducting geological exploration, optimizing working face design, using support systems such as hydraulic supports and support nets, and ensuring the effectiveness of support equipment. In addition, obtain real-time information such as roof deformation data, stress distribution, and geological changes in the area to be mined in Coal Mine A, and establish a dynamic early warning system in combination with the results of cause analysis to promptly detect potential risks and take corresponding measures. Establish a continuous optimization and improvement mechanism, collect accident and monitoring data through a feedback mechanism, analyze the effectiveness of preventive measures, and adjust the plan according to new data. At the same time, continuously update the knowledge graph to ensure the accuracy of cause analysis and optimize the implementation of preventive measures.

[0111] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing the causes of coal mine accidents based on knowledge graph, characterized in that: include: Collect accident reports of coal mine roof accidents, and obtain a set of roof accident words through text cleaning operations; According to the roof accident word set, a roof accident entity extraction model, an entity relationship classification model and an entity attribute extraction model are constructed, and the model training is performed in combination with the roof accident word set; entity extraction, entity relationship extraction and attribute extraction are performed on the roof accident word set through the trained model to obtain the accident cause type, accident cause node, roof accident node, cause attribute, accident attribute and accident cause association edge; Combining the accident cause node, the roof accident node, the cause attribute, the accident attribute and the accident cause associated edge, a roof accident cause knowledge graph is constructed; Calculating the cause text similarity of the accident cause node and the attribute similarity of the corresponding cause attribute to obtain the accident cause similarity; Combined with the accident attribute of the accident cause node, a merging operation is performed on the accident cause node of the roof accident cause knowledge graph; Obtaining the cause attribute code and the accident attribute code, and calculating the cause correlation index of different accident cause nodes; combining the accident attribute code and the roof accident risk assessment function to calculate the single cause risk index; The multi-cause risk index of the accident causation node is obtained, combined with the single-cause risk index to obtain a comprehensive risk index, and the accident causation nodes are ranked according to their degree of danger.

2. A method for analyzing the causes of coal mine accidents based on knowledge graph according to claim 1, characterized in that: The specific implementation process of obtaining the accident cause node, the roof accident node, the cause attribute, the accident attribute and the accident cause association edge includes: Labeling several groups of roof accident word sets to obtain roof accident entity training sets, entity relationship training sets and entity attribute training sets; wherein the labels of the roof accident entity training sets include accident cause node labels and roof accident node labels; the labels of the entity relationship training sets are accident cause associated edge labels; the labels of the entity attribute training sets include cause attribute labels and accident attribute labels; Input the roof accident entity training set into the BERT pre-training model, perform coal mine roof accident entity training, and obtain the roof accident entity extraction model; construct the entity relationship classification model based on the BERT pre-training model and the classifier, and perform coal mine roof entity relationship training according to the entity relationship training set; perform coal mine roof accident entity attribute training on the BERT pre-training model through the entity attribute training set to obtain the entity attribute extraction model; Entities in the roof accident word set are extracted through the roof accident entity extraction model to determine the accident cause node and the roof accident node; the relationship between different accident cause nodes and / or the roof accident nodes is determined according to the entity relationship classification model to obtain the accident cause association edge; combined with the entity attribute extraction model, the cause attribute of the accident cause node and the accident attribute of the roof accident node are obtained.

3. A method for analyzing the causes of coal mine accidents based on knowledge graph according to claim 1, characterized in that: The accident cause types include geological factors, human error factors, equipment failure factors and operating environment factors; the accident cause nodes of geological factors include geological structure nodes, rock formation nodes, stratum nodes and surface water nodes; the accident cause nodes of human error factors include mining method nodes, ventilation measures nodes, support method nodes and overload operation nodes; the accident cause nodes of equipment failure factors include support equipment nodes, ventilation equipment nodes, mechanical equipment nodes and power equipment nodes; the accident cause nodes of operating environment factors include mine type nodes, mine depth nodes, mine temperature nodes, mine gas concentration nodes and mine humidity nodes; the accident attributes include location, time, accident scope, casualties and economic losses.

4. A method for analyzing the causes of coal mine accidents based on knowledge graph according to claim 1, characterized in that: Calculating the cause text similarity and the attribute similarity, and merging the accident cause nodes of the roof accident cause knowledge graph; the specific implementation process includes: According to two different accident cause nodes and Get separately and Number of characters in node names and pass Calculate the accident cause node and The similarity between the causal texts Among them, min() represents the minimum function; max() represents the maximum function; Used for calculation and The number of identical characters in the node names; Get the two accident cause nodes respectively and The corresponding set of causal attributes {A1,…,A m } and {B1,…,B n }, perform one-to-one matching on the causal attributes in two different causal attribute sets to obtain k sets of causal attribute matching pairs; for any causal attribute matching pair Calculate the attribute value similarity of the two causal attributes; synthesize the attribute value similarities of k groups of causal attribute matching pairs to obtain the attribute similarity The specific calculation formula is: Among them, m and n represent and The number of causal attributes of A i and B j Matching pairs for causal attributes The two causal properties, A i express The i-th causal attribute and i∈{1,…,m}, B j express The jth causal attribute of and j∈{1,…,n}; α i Represents the causal attribute matching pair Impact factor; Combined with the accident cause node and The similarity of the causal text Similarity to the attribute pass Calculate the comprehensive similarity of accident causes; where β represents The weight of If the comprehensive similarity of the accident cause is greater than the predetermined similarity threshold, the accident cause node and Merge into one node.

5. The method for analyzing the causes of coal mine accidents based on knowledge graph according to claim 1 is characterized in that: The specific implementation process of calculating the cause correlation index of different accident cause nodes: The causal attribute and the accident attribute are data-encoded to obtain the causal attribute code and the accident attribute code; according to the causal attribute code and the accident attribute code, the causal correlation index of the roof accident node and different accident causal nodes is calculated, and the specific calculation formula is: in, Indicates the roof accident node and the uth accident-causing node The causal association index between roof It indicates the total number of coal mine roof accidents; express The total number of coal mine roof accidents caused by COV() represents the covariance function; R represents The numerical matrix of the accident attribute encoding; C u,l express The numerical matrix of the causal attribute encoding of the lth causal attribute; σ R represents the standard deviation of R; Represents C u,l The standard deviation of l represents the influence of the lth causal attribute on the coal mine roof accident; M represents The total number of causal attributes.

6. A method for analyzing the causes of coal mine accidents based on knowledge graph according to claim 1, characterized in that: Combining the accident attribute code and the cause association index of different accident cause nodes, the single cause risk index of coal mine roof accidents caused by the accident cause nodes is quantified through the roof accident risk assessment function. The specific formula is: in, Indicates the roof accident node and the uth accident-causing node The causal association index between u express The single-cause risk index of v express The vth accident attribute; Represents R v The roof accident risk assessment function of v Represents R v The influence coefficient on the consequences of coal mine roof accidents; K represents the total number of accident attributes.

7. A method for analyzing the causes of coal mine accidents based on knowledge graph according to claim 1, characterized in that: Calculating the multi-cause risk index of different accident cause nodes, and combining the single-cause risk index to obtain the comprehensive risk index of each accident cause node; The specific calculation formula is: Among them, CR u Represents the u-th accident cause node The comprehensive risk index; ∈0 means The single cause risk index RA u The weight coefficient of express The qth multi-causal risk index; ∈ q express The corresponding weight coefficient; Q represents The total number of the multi-factor risk indices; The accident-causing nodes are ranked according to their comprehensive risk indexes.

Images