Coal mine risk searching and evaluating method based on natural language processing

Abstract

The invention relates to the technical field of coal mine risk analysis, in particular to a coal mine risk searching and evaluating method based on natural language processing. Firstly, a coal mine risk knowledge graph is established through a natural language processing method, key risk factors and mutual relations thereof are extracted from original production data by utilizing entity recognition, relation extraction and event extraction technologies, and a knowledge graph comprehensively reflecting coal mine safety risks is constructed; secondly, the real-time production data and the coal mine risk knowledge graph are subjected to first similarity and second similarity comparison, and potential risks in the current production environment are rapidly recognized by calculating the similarity between the real-time data and entities and relationships in the graph; and finally, the causal relationship in the coal mine risk knowledge graph is extracted, an event tree is established, and a coal mine risk index is calculated through a probability value of an event tree path by analyzing a causal chain between risk events and constructing branches and probability distribution of the event tree.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine risk analysis, and specifically provides a coal mine risk search and evaluation method based on natural language processing. Background Art

[0002] As an important part of the energy industry, the coal mine production process involves a large number of risk factors and complex safety issues. Although modern technical means have continuously improved the safety of coal mine production, coal mine accidents still occur from time to time, causing a large number of casualties and property losses. Coal mine accidents not only pose a threat to the lives and safety of miners, but also have a serious impact on the country's energy supply and social stability. Therefore, timely discovery of potential risks and effective risk assessment are of crucial significance for reducing coal mine accidents and ensuring the safety of miners.

[0003] Currently, the search and evaluation of coal mine risks based on statistical methods are widely used in coal mine safety management. These methods analyze historical data to calculate the probability of accidents and risk levels, so as to help decision-makers identify potential safety hazards. However, traditional statistical methods mainly rely on historical accident data and ignore the complex factors behind accidents and potential hidden risks. These methods can only evaluate the risks that have occurred and lack the ability to effectively predict and quantitatively evaluate unknown risks. With the complexity of the coal mine production environment, traditional risk assessment methods based on statistical methods are difficult to comprehensively capture multi-dimensional information and have certain limitations in terms of real-time performance, accuracy, and operability. Therefore, it is urgent to introduce new technical means to improve the accuracy and forward-looking of coal mine risk search and assessment.

[0004] Therefore, a coal mine risk search and evaluation method based on natural language processing is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a coal mine risk search and evaluation method based on natural language processing. First, a coal mine risk knowledge graph is established through natural language processing methods to extract key risk factors and their relationships in texts such as accident reports and equipment failures. Secondly, the real-time production data is compared with the coal mine risk knowledge graph for similarity to identify potential risks, and it is determined whether they are unknown risks through a threshold. Finally, the causal relationships in the knowledge graph are extracted and an event tree is established to quantify potential risks and calculate the coal mine risk index, providing decision-making support for safety management.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A coal mine risk search and evaluation method based on natural language processing, comprising:

[0008] Obtain the original production data related to coal mine production in the Internet; the original production data includes accident reports, equipment failure logs, operation records, safety monitoring data, and safety inspection reports;

[0009] Perform text preprocessing on the original production data to obtain preprocessed data;

[0010] Through natural language processing methods, perform entity recognition, relationship extraction, and event extraction on the preprocessed data, map the obtained entities, semantic relationships, and event information to a graph database, and establish a coal mine risk knowledge graph through the graph database;

[0011] Collect real-time production data in actual coal mine production, extract the risk information of the real-time production data, convert the coal mine risk knowledge graph and the risk information into embedding vectors and calculate the maximum first similarity; if the maximum first similarity is less than the first threshold, mark the real-time production data as unknown coal mine risk, store the risk information in the graph database, and update the coal mine risk knowledge graph at the same time; otherwise, calculate the second similarity;

[0012] If the second similarity is less than the second threshold, determine that the real-time production data is in a low-risk state; otherwise, determine that the real-time production data has potential coal mine risks;

[0013] Extract the causal relationships of the coal mine risk knowledge graph and establish an event tree, determine the branches of the event tree according to the potential coal mine risks, and quantify the potential coal mine risks through the distribution probabilities of the branches to obtain a coal mine risk index.

[0014] Further, the text preprocessing includes:

[0015] Remove the noise information in the original production data to obtain denoised text data;

[0016] Perform data cleaning on the denoised text data to obtain cleaned text data;

[0017] Split the cleaned text data into independent words and phrases, and perform part-of-speech tagging to obtain preprocessed data.

[0018] Further, establishing the coal mine risk knowledge graph includes:

[0019] Identify the entities in the preprocessed data through NER technology;

[0020] Extract the semantic relationships in the preprocessed data through a relationship extraction model;

[0021] Extract the event information of the preprocessed data through an event extraction model, where the event information includes event type, event participants, and event occurrence time;

[0022] Map the entities, semantic relationships, and event information to the graph database to obtain the coal mine risk knowledge graph.

[0023] Furthermore, extracting the risk information from the real-time production data includes:

[0024] Preprocess the real-time production data to obtain preprocessed real-time data;

[0025] Perform entity recognition, relationship extraction, and event extraction on the preprocessed real-time data to obtain the risk information.

[0026] Furthermore, converting the coal mine risk knowledge graph and the risk information into embedding vectors specifically includes:

[0027] Convert the coal mine risk knowledge graph into a first embedding vector, where the first embedding vector includes a first entity embedding vector and a first relationship embedding vector;

[0028] Convert the risk information into a second embedding vector, where the second embedding vector includes a second entity embedding vector and a second relationship embedding vector.

[0029] Furthermore, the formula for calculating the maximum first similarity is:

[0030]

[0031] where FC represents the maximum first similarity, max represents the maximum value function, Cosine Similarity represents the cosine similarity calculation function, represents the i-th component of the first entity embedding vector, E real represents the second entity embedding vector.

[0032] Furthermore, the formula for calculating the second similarity is:

[0033]

[0034] where FC represents the second similarity, represents the i-th component of the first relationship embedding vector, R real represents the second relationship embedding vector, α represents the entity vector weight, and β represents the relationship vector weight.

[0035] Furthermore, obtaining the coal mine risk index includes:

[0036] Determining the initial risk event of the event tree based on the entity and event information in the coal mine risk knowledge graph;

[0037] Extracting causal relationships between event information from the coal mine risk knowledge graph, constructing branches of the event tree, and assigning probabilities to the branches;

[0038] Matching the event tree according to the potential coal mine risk to determine the event tree path corresponding to the potential coal mine risk;

[0039] Based on the probability value of the event tree path, a coal mine risk index is obtained.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. By building a coal mine risk knowledge graph using natural language processing methods, we can automatically extract and identify key entities, semantic relationships, and event information from massive amounts of coal mine production data. This process effectively transforms text data into structured information, which is stored and managed in a graph database, facilitating comprehensive analysis of coal mine safety risks. Compared to statistical analysis, this approach can rapidly process massive amounts of data, uncover potential risk factors, and reveal complex relationships between entities.

[0042] 2. Through the calculation of the first similarity, we can quickly conduct preliminary risk matching on real-time production data, determining whether it matches the information in the existing coal mine risk knowledge graph, thereby ensuring the dynamic expansion and improvement of the knowledge graph. The second similarity calculation further refines the risk assessment, focusing on the matching of entities and relationships. This multi-level similarity calculation method achieves more accurate search and location of coal mine safety risks.

[0043] 3. Event tree analysis can systematically identify and quantify potential risks in coal mines. This method extracts key entities and event information from the coal mine risk knowledge graph, constructs branches of the event tree, assigns probabilities to each branch, and ultimately calculates a coal mine risk index. This method can quantitatively assess coal mine risks, helping managers identify potential hazards and ultimately optimize safety warnings, decision support, and risk control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of a method for searching and evaluating coal mine risks based on natural language processing provided by an embodiment of the present invention;

[0045] Figure 2 Partial visualization data of the coal mine risk knowledge graph provided by the embodiment of the present invention;

[0046] Figure 3 A flow chart for obtaining a coal mine risk index is provided in an embodiment of the present invention. Specific Embodiments

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to 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 of 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.

[0048] Embodiment 1

[0049] A method for finding and evaluating coal mine risks based on natural language processing, as Figure 1 shown, includes.

[0050] S100: Obtain the original production data related to coal mine production on the Internet; the original production data includes accident reports, equipment failure logs, operation records, safety monitoring data, and safety inspection reports;

[0051] S200: Perform text preprocessing on the original production data to obtain preprocessed data;

[0052] Further, the text preprocessing includes:

[0053] Remove the noise information in the original production data to obtain denoised text data;

[0054] Perform data cleaning on the denoised text data to obtain cleaned text data;

[0055] Split the cleaned text data into independent words and phrases, and perform part-of-speech tagging to obtain preprocessed data.

[0056] Specifically, use regular expressions to remove meaningless characters, use a stop word list to remove common useless words, delete HTML tags or other formatting symbols, use text processing tools (such as NLTK, spaCy, etc.) for data denoising to obtain denoised text data; remove blank lines or irrelevant text (such as advertisements, page information, etc.) from the denoised text data, use a spelling check tool to correct spelling mistakes, and standardize the text format at the same time to obtain cleaned text data; split the cleaned text data into independent words or phrases. For Chinese, a word segmentation tool such as jieba can be used for word segmentation, and for English, tools such as word_tokenize can be used, and then use natural language processing tools (NLTK, spaCy, etc.) for part-of-speech tagging to obtain preprocessed data.

[0057] Through text preprocessing, including removing meaningless characters, removing stop words, cleaning irrelevant text, correcting spelling mistakes, and standardizing formats, the quality and consistency of the data can be effectively improved. Subsequently, through word segmentation and part-of-speech tagging, the text is converted into structured data, providing an accurate basis for subsequent natural language processing tasks and knowledge graph construction. This process helps reduce noise and improve the usability and analysis accuracy of the data.

[0058] S300: Perform entity recognition, relation extraction, and event extraction on the preprocessed data through natural language processing methods, map the obtained entity, semantic relation, and event information to a graph database, and establish a coal mine risk knowledge graph through the graph database;

[0059] Furthermore, establishing the coal mine risk knowledge graph includes:

[0060] Identify the entities in the preprocessed data through NER technology;

[0061] Extract the semantic relations of the preprocessed data through a relation extraction model;

[0062] Extract the event information of the preprocessed data through an event extraction model, where the event information includes event type, event participants, and event occurrence time;

[0063] Map the entities, the semantic relations, and the event information to the graph database to obtain the coal mine risk knowledge graph.

[0064] Specifically, NER technology is an entity recognition technology. Commonly used NER models include spaCy, BERT, and StanfordNER, etc. Input the preprocessed data into the NER model, and the model will return the entities and corresponding categories in the text; relation extraction is to identify the semantic relations between different entities from the text (such as the relation between equipment and faults, the relation between operators and operations, etc.). The relation extraction model can be a dependency syntax tree or a deep learning model such as CNN, RNN, or BERT. Through rules or pattern matching, semantic relations are extracted from the preprocessed data; event extraction is to identify the events described in the text, including at least the type, participants, and time of the events. Commonly used event extraction models are LSTM and BERT. The event information of the preprocessed data is extracted through the event extraction model; the knowledge graph organizes and stores knowledge by mapping the extracted entities, relations, and events to a graph structure. The nodes in the graph represent entities, the edges represent relations, and event information (such as time, location, etc.) can be stored on the edges. An example of the coal mine risk knowledge graph construction process is shown in Table 1, and some visualized data of the coal mine risk knowledge graph are as Figure 2 shown.

[0065] Table 1 Example of the construction process of the coal mine risk knowledge graph

[0066]

[0067] Through natural language processing methods, entity recognition, relation extraction, and event extraction are performed on the preprocessed data, enabling the extraction of key entities, semantic relations, and event information during the coal mine production process and effectively mapping them into the graph database. Using NER technology to identify entities, a relation extraction model to extract association information, and an event extraction model to determine the type, participants, and time of events enhances the comprehensiveness and accuracy of the knowledge graph, thus providing strong data support for coal mine safety management and risk assessment.

[0068] S400: Collect real-time production data in actual coal mine production, extract the risk information of the real-time production data, convert the coal mine risk knowledge graph and the risk information into embedding vectors and calculate the maximum first similarity; if the maximum first similarity is less than the first threshold, mark the real-time production data as unknown coal mine risk, store the risk information in the graph database, and update the coal mine risk knowledge graph simultaneously; otherwise, calculate the second similarity;

[0069] Further, the extraction of the risk information from the real-time production data includes:

[0070] Preprocess the real-time production data to obtain preprocessed real-time data;

[0071] Perform entity recognition, relation extraction, and event extraction on the preprocessed real-time data to obtain the risk information.

[0072] Specifically, the real-time production data includes accident reports, equipment failure logs, operation records, safety monitoring data, and safety inspection reports. The preprocessing of the real-time production data follows the same process as the preprocessing of the original production data in S200, which will not be elaborated here; the entity recognition, relation extraction, and event extraction of the preprocessed real-time data follow the same process as the processing of the preprocessed data in S300, and finally the risk information is obtained.

[0073] By preprocessing, entity recognition, relation extraction, and event extraction of the real-time production data, key information related to coal mine risks can be effectively extracted from the real-time data, providing a data basis for subsequent risk search and risk evaluation.

[0074] Further, the conversion of the coal mine risk knowledge graph and the risk information into embedding vectors is specifically as follows:

[0075] Convert the coal mine risk knowledge graph into a first embedding vector, and the first embedding vector includes a first entity embedding vector and a first relation embedding vector;

[0076] Convert the risk information into a second embedding vector, where the second embedding vector includes a second entity embedding vector and a second relationship embedding vector.

[0077] Specifically, the coal mine risk knowledge graph is converted into a first embedding vector through the graph embedding algorithm Node2Vec, where the first embedding vector includes a first entity embedding vector and a first relationship embedding vector; the risk information is converted into a second embedding vector through the text embedding algorithm Word2Vec, where the second embedding vector includes a second entity embedding vector and a second relationship embedding vector.

[0078] Converting the coal mine risk knowledge graph and risk information into embedding vectors helps to convert complex text data and graph information into a numerical form that can be used in machine learning models, facilitating subsequent similarity calculation and risk identification.

[0079] Furthermore, the formula for calculating the maximum first similarity is:

[0080]

[0081] where FC represents the maximum first similarity, max represents the maximum value function, Cosine Similarity represents the cosine similarity calculation function, represents the i-th component of the first entity embedding vector, and E real represents the second entity embedding vector.

[0082] In a feasible implementation, an example of calculating the maximum first similarity is shown in Table 2. The entity of risk information 1 is "coal mine crusher", and the corresponding entity embedding vector is [0.12, -0.25, 0.57,...]. After comparing all entity embedding vectors in the knowledge graph and performing the first similarity calculation, the maximum first similarity obtained is 0.87, and the corresponding entity is coal mine equipment; the entity of risk information 2 is "lighting equipment", and the corresponding entity embedding vector is [0.32, -0.45, 0.67,...]. After comparing all entities in the knowledge graph and performing the first similarity calculation, the maximum first similarity obtained is 0.21, which is less than the first threshold. It is considered that there is an unknown coal mine risk in the current real-time production data. The risk information is stored in the graph database, and the coal mine risk knowledge graph is updated simultaneously.

[0083] Table 2 Example of Calculating the Maximum First Similarity

[0084]

[0085] By converting the coal mine risk knowledge graph and real-time risk information into embedding vectors and calculating the maximum first similarity, the matching degree between real-time production data and existing risk knowledge can be efficiently evaluated. If the similarity is less than the preset threshold, the risk can be marked as "unknown" and stored in the graph database for subsequent analysis and update. This process not only ensures the dynamic update of the knowledge graph, enhances its ability to handle new risks, but also can capture unknown risks in a timely manner, improving the breadth of risk detection.

[0086] S500: If the second similarity is less than the second threshold, it is determined that the real-time production data is in a low-risk state; otherwise, it is determined that the real-time production data has potential coal mine risks.

[0087] Furthermore, the formula for calculating the second similarity is:

[0088]

[0089] where FC represents the second similarity, represents the i-th component of the first relationship embedding vector, R real represents the second relationship embedding vector, α represents the entity vector weight, and β represents the relationship vector weight.

[0090] Specifically, the first relationship embedding vector corresponds one-to-one with the first entity embedding vector; in another feasible implementation, an example of calculating the second similarity is shown in Table 3. The semantic relationship of Risk Information 1 is "damage", and the corresponding entity embedding vector is [0.07, -0.20, 0.51,...]. Comparing with the semantic relationship "fault" corresponding to the entity "coal mine equipment" in the knowledge graph, the corresponding relationship embedding vector is [0.09, -0.20, 0.47,...]. After calculating the second similarity, the second similarity obtained is 0.93, so it is considered that there are potential coal mine risks; the entity of Risk Information 2 is "operation according to regulations", and the corresponding relationship embedding vector is [0.22, -0.15, 0.40,...]. Comparing with the semantic relationship "fault" corresponding to the entity "coal mine equipment" in the knowledge graph, after calculating the second similarity, the second similarity obtained is 0.71, which is less than the second threshold, so it is considered that the real-time production data this time is low-risk data.

[0091] By calculating the second similarity, the similarity between the potential risk information in the real-time production data and the existing coal mine risk knowledge graph can be effectively measured. If the second similarity is less than the threshold, it is judged as low-risk, indicating that the potential risk is relatively low; if the second similarity is higher than the threshold, it indicates that the risk is more obvious and there may be potential coal mine risks. This method can automatically and real-time detect risks in coal mine production, helping to take preventive and emergency measures in a timely manner.

[0092] Table 3 Example of Calculating the Second Similarity

[0093]

[0094] S600: Extract the causal relationships of the coal mine risk knowledge graph and establish an event tree. Determine the branches of the event tree according to the potential coal mine risks, and quantify the potential coal mine risks through the distribution probabilities of the branches to obtain a coal mine risk index.

[0095] Furthermore, the obtained coal mine risk index is as Figure 3 shown, including:

[0096] Determine the initial risk events of the event tree according to the entity and event information in the coal mine risk knowledge graph;

[0097] Extract the causal relationships between event information from the coal mine risk knowledge graph, construct the branches of the event tree, and assign probabilities to the branches;

[0098] Match the event tree according to the potential coal mine risks to determine the event tree path corresponding to the potential coal mine risks;

[0099] Obtain a coal mine risk index based on the probability values of the event tree path.

[0100] Specifically, according to the entities (such as equipment, working environment, operation tasks, etc.) and event information (such as equipment failures, mine disasters, etc.) in the coal mine risk knowledge graph, select the root event of the event tree. The root event is usually the initial risk event that may cause greater disasters in coal mine operations; extract the causal relationships between event information from the coal mine risk knowledge graph, such as equipment failures may lead to work stoppages or accidents, and poor mine ventilation may lead to poisoning of personnel, etc. According to these causal relationships, construct the branches of the event tree, and each branch represents the transition path from an initial event to subsequent potential events; each event branch is assigned a probability of occurrence based on historical data or expert evaluation. The probability can be estimated through statistical analysis, expert judgment, or machine learning models; identify the potential coal mine risks in the current coal mine based on real-time production data and event information in the coal mine risk knowledge graph, match these potential risks with the root events in the event tree, and determine the event tree path corresponding to the potential risks; calculate the coal mine risk index based on the probability values of the event tree path. The calculation of the risk index can be through a probability-weighted method, multiplying the respective probability values of the event tree path to obtain the final coal mine risk index.

[0101] By establishing the causal relationships and event trees based on the coal mine risk knowledge graph, the potential coal mine risks can be quantitatively evaluated. First, by extracting the entity and event information in the knowledge graph, the initial risk events of the event tree are determined, and further, the causal relationships between events are analyzed from the graph to construct the branches of the event tree. After assigning appropriate probabilities to each branch, the event tree is matched according to the actual potential risk data to trace the paths of the risk events. Finally, the coal mine risk index is calculated based on the probability values of the paths, providing a scientific and systematic quantitative evaluation index. This method can achieve the precise quantification of coal mine risks, thus helping managers objectively identify, analyze, and evaluate the potential safety hazards in coal mines.

[0102] Example Two

[0103] A coal mine company updated its original coal mine risk management system to strengthen the management of coal mine risks. The updated coal mine risk management system includes a coal mine risk search and evaluation method based on natural language processing, including:

[0104] Obtain the original production data related to coal mine production on the Internet; the original production data includes accident reports, equipment failure logs, operation records, safety monitoring data, and safety inspection reports;

[0105] Perform text preprocessing on the original production data to obtain preprocessed data;

[0106] Through natural language processing methods, entity recognition, relationship extraction, and event extraction are performed on the preprocessed data, and the obtained entities, semantic relationships, and event information are mapped to a graph database. A coal mine risk knowledge graph is established through the graph database;

[0107] Collect the real-time production data in actual coal mine production, extract the risk information of the real-time production data, convert the coal mine risk knowledge graph and the risk information into embedding vectors and calculate the maximum first similarity; if the maximum first similarity is less than the first threshold, mark the real-time production data as unknown coal mine risk, store the risk information in the graph database, and update the coal mine risk knowledge graph at the same time; otherwise, calculate the second similarity;

[0108] If the second similarity is less than the second threshold, determine that the real-time production data is in a low-risk state; otherwise, determine that the real-time production data has potential coal mine risks;

[0109] Extract the causal relationships of the coal mine risk knowledge graph and establish an event tree. Determine the branches of the event tree according to the potential coal mine risks, and quantitatively evaluate the potential coal mine risks through the assigned probabilities of the branches to obtain the coal mine risk index.

[0110] Further, the text preprocessing includes:

[0111] Removing the noise information in the original production data to obtain denoised text data;

[0112] Performing data cleaning on the denoised text data to obtain cleaned text data;

[0113] Splitting the cleaned text data into independent words and phrases, and performing part-of-speech tagging to obtain preprocessed data.

[0114] Further, building the coal mine risk knowledge graph includes:

[0115] Identifying the entities in the preprocessed data through NER technology;

[0116] Extracting the semantic relationships in the preprocessed data through a relationship extraction model;

[0117] Extracting the event information in the preprocessed data through an event extraction model, where the event information includes event type, event participants, and event occurrence time;

[0118] Mapping the entities, the semantic relationships, and the event information to the graph database to obtain the coal mine risk knowledge graph.

[0119] Further, extracting the risk information from the real-time production data includes:

[0120] Performing preprocessing on the real-time production data to obtain preprocessed real-time data;

[0121] Performing entity recognition, relationship extraction, and event extraction on the preprocessed real-time data to obtain the risk information.

[0122] Further, converting the coal mine risk knowledge graph and the risk information into embedding vectors specifically:

[0123] Converting the coal mine risk knowledge graph into a first embedding vector, where the first embedding vector includes a first entity embedding vector and a first relationship embedding vector;

[0124] Converting the risk information into a second embedding vector, where the second embedding vector includes a second entity embedding vector and a second relationship embedding vector.

[0125] Further, the formula for calculating the maximum first similarity is:

[0126]

[0127] Wherein, FC represents the maximum first similarity, max represents the maximum value function, and Cosine Similarity represents the cosine similarity calculation function. represents the i-th component of the first entity embedding vector, E real Represents the second entity embedding vector.

[0128] Furthermore, the formula for calculating the second similarity is:

[0129]

[0130] Wherein, FC represents the second similarity, represents the i-th component of the first relation embedding vector, R real represents the second relation embedding vector, α represents the entity vector weight, and β represents the relation vector weight.

[0131] Furthermore, obtaining the coal mine risk index includes:

[0132] Determining the initial risk event of the event tree based on the entity and event information in the coal mine risk knowledge graph;

[0133] Extracting causal relationships between event information from the coal mine risk knowledge graph, constructing branches of the event tree, and assigning probabilities to the branches;

[0134] Matching the event tree according to the potential coal mine risk to determine the event tree path corresponding to the potential coal mine risk;

[0135] Based on the probability value of the event tree path, a coal mine risk index is obtained.

[0136] Compared to the original coal mine risk management system based on traditional statistical methods, the updated system possesses the following features, as shown in Table 4. The new system, based on natural language processing, offers significant advantages over the traditional statistical-based coal mine risk management system. The new system processes unstructured text data from accident reports, work records, equipment failure logs, and other sources in real time, dynamically identifying potential risks. Traditional systems rely on historical data, which is subject to latency and information lag. By constructing a coal mine risk knowledge graph and event tree, the new system not only deeply explores complex causal relationships but also automatically assigns probabilities to risk events, enabling more accurate and comprehensive risk assessments. Furthermore, the new system offers greater flexibility and scalability, enabling it to quickly adapt to new risk scenarios and data sources. This provides coal mine managers with more intelligent and accurate decision support, effectively improving safety management.

[0137] Table 4 Comparison of characteristics of the original system and the updated system

[0138]

[0139] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will appreciate 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 finding and evaluating coal mine risks based on natural language processing, characterized in that, Including: Obtain the original production data related to coal mine production in the Internet; The original production data includes accident reports, equipment failure logs, operation records, safety monitoring data, and safety inspection reports; Perform text preprocessing on the original production data to obtain preprocessed data; Through natural language processing methods, perform entity recognition, relationship extraction, and event extraction on the preprocessed data, map the obtained entities, semantic relationships, and event information to a graph database, and establish a coal mine risk knowledge graph through the graph database; Collect real-time production data in actual coal mine production, extract the risk information of the real-time production data, convert the coal mine risk knowledge graph and the risk information into embedding vectors and calculate the maximum first similarity; If the maximum first similarity is less than the first threshold, mark the real-time production data as unknown coal mine risk, store the risk information in the graph database, and update the coal mine risk knowledge graph at the same time; Otherwise, calculate the second similarity; If the second similarity is less than the second threshold, determine that the real-time production data is in a low-risk state; Otherwise, determine that the real-time production data has potential coal mine risks; Extract the causal relationships of the coal mine risk knowledge graph and establish an event tree, determine the branches of the event tree according to the potential coal mine risks, and quantify the potential coal mine risks through the distribution probabilities of the branches to obtain a coal mine risk index.

2. The coal mine risk search and evaluation method based on natural language processing according to claim 1, characterized in that, The text preprocessing includes: Remove the noise information in the original production data to obtain denoised text data; Perform data cleaning on the denoised text data to obtain cleaned text data; Split the cleaned text data into independent words and phrases, and perform part-of-speech tagging to obtain preprocessed data.

3. A coal mine risk search and evaluation method based on natural language processing according to claim 1, characterized in that Establishing the coal mine risk knowledge graph includes: Identify the entities in the preprocessed data through NER technology; Extract the semantic relationships of the preprocessed data through a relationship extraction model; Extract the event information of the preprocessed data through an event extraction model, and the event information includes event type, event participants, and event occurrence time; Map the entities, the semantic relationships, and the event information to the graph database to obtain the coal mine risk knowledge graph.

4. A method for finding and evaluating coal mine risks based on natural language processing according to claim 1, characterized in that, Extracting the risk information of the real-time production data includes: Perform preprocessing on the real-time production data to obtain preprocessed real-time data; Perform entity recognition, relationship extraction, and event extraction on the preprocessed real-time data to obtain the risk information.

5. A method for coal mine risk search and evaluation based on natural language processing according to claim 1, characterized in that, Specifically converting the coal mine risk knowledge graph and the risk information into embedding vectors is: Convert the coal mine risk knowledge graph into a first embedding vector, and the first embedding vector includes a first entity embedding vector and a first relationship embedding vector; Convert the risk information into a second embedding vector, and the second embedding vector includes a second entity embedding vector and a second relationship embedding vector.

6. The coal mine risk search and evaluation method based on natural language processing according to claim 1, characterized in that The formula for calculating the maximum first similarity is: Among them, FC represents the maximum first similarity, max represents the maximum value function, and Cosine Similarity represents the cosine similarity calculation function. represents the i-th component of the first entity embedding vector, E real represents the second entity embedding vector.

7. A method for finding and evaluating coal mine risks based on natural language processing according to claim 1, characterized in that, The formula for calculating the second similarity is: where, FC represents the second similarity degree, represents the i-th component of the first relation embedding vector, R real represents the second relation embedding vector, α represents the entity vector weight, and β represents the relation vector weight.

8. A method for finding and evaluating coal mine risks based on natural language processing according to claim 1, characterized in that, Obtaining the coal mine risk index includes: According to the entities and event information in the coal mine risk knowledge graph, determine the initial risk events of the event tree; Extract the causal relationships between event information from the coal mine risk knowledge graph, construct the branches of the event tree, and assign probabilities to the branches; Match the event tree according to the potential coal mine risks to determine the event tree path corresponding to the potential coal mine risks; Obtain the coal mine risk index based on the probability values of the event tree paths.

Images